From Department of Clinical Neuroscience - CNS

Karolinska Institutet, Stockholm, Sweden

PREDICTORS AND TREATMENT OF FATIGUE

IN MULTIPLE SCLEROSIS

Simon Englund

Stockholm 2025
All previously published papers were reproduced with permission from the publisher.
Published by Karolinska Institutet.
Printed by Universitetsservice US-AB, 2025
© Simon Englund
The comprehensive summary chapter of this thesis is licensed under CC BY 4.0. To view a copy
of this license, visit https://creativecommons.org/licenses/by/4.0/ Other licences or copyright
may apply to illustrations and attached articles.
ISBN 978-91-8017-551-7
DOI https://doi.org/10.69622/28563902
Cover illustration: by Camilla Doms
Predictors and treatment of fatigue in multiple
sclerosis
Thesis for Doctoral Degree (Ph.D.)

By

Simon Englund

The thesis will be defended in public at CMM Lecture Hall, Visionsgatan 18, Karolinska
University Hospital, Solna, L8:00 - June 12, 2025 at 09:00

Principal Supervisor: Opponent:

Professor Fredrik Piehl Professor emeritus Ann-Marie Landtblom
Karolinska Institutet Uppsala University
Department of Clinical Neuroscience Department of Medical Sciences
Division of Neuro Division of Neurology

Co-supervisors: Examination Board:

Docent Thomas Frisell Docent Giorgio Tettamanti
Karolinska Institutet Karolinska Institutet
Department of Medicine, Solna Institute of Environmental Medicine
Division of Clinical Epidemiology Division of Epidemiology

Docent Marie Kierkegaard Docent Anna Andreasson

Karolinska Institutet Stockholm University
Department of Neurobiology, Department of Psychology
Care Sciences and Society Division of Psychoneuroimmunology
Division of Physiotherapy
Docent Helena Igelström
PhD Elisa Longinetti Uppsala University
IQVIA Department of Women's and Children's Health
Physiotherapy, Occupational Therapy and
Behavioral Medicine Unit
Till Camilla, Herta och Lillebror.
Popular science summary of the thesis
Multiple sclerosis (MS) is a chronic inflammatory and neurodegenerative disease
of the central nervous system, that is, the brain and spinal cord, where the
resulting damage leads to disability that varies in both nature and extent. One of
the more common and most functionally impairing symptoms is fatigue, a kind of
energy depletion often described as overwhelming, which negatively affects
quality of life, work capacity, family life, leisure, and social relationships. There is
no universally accepted definition of fatigue in MS, but patients often describe it
in terms of mental fog, low energy levels, and a sense of exhaustion affecting
both mental and physical capacities. Fatigue is thus fundamentally a subjective
experience that is typically evaluated using self-reported questionnaires, which
assess severity based on its impact on various functional domains of daily life.
Because definitions of fatigue vary, as do the designs of different questionnaires,
comparing results across studies is challenging. The Fatigue Scale for Motor and
Cognitive Functions (FSMC) has, however, been developed specifically for MS
and has in recent years become one of the most widely used.

A range of factors have been proposed as causes of fatigue in MS, some of which
are directly linked to MS-specific physiological processes. Other potential
explanations include comorbidities, particularly psychiatric disorders such as
depression and anxiety. These are common in MS and are independently
associated with the presence of fatigue. Previous studies in MS have indicated
factors associated with fatigue, but interpretation is complicated by the fact
that many studies have been small and rarely followed individuals over longer
periods. Nonetheless, it appears that fatigue has multiple underlying causes that
vary between individuals.

There is still no cure for MS, but in the past decade, increasingly effective
disease-modifying therapies have been developed that protect against acute
relapse episodes. To a lesser extent, these therapies also help prevent the
development of permanent disability, especially in patients with relapsing form
of MS, but the extent to which they affect fatigue remains unclear. Another
category of treatment targets MS symptoms. Both pharmacological and non-
pharmacological treatment options for fatigue in MS are used to varying degrees,
but the lack of knowledge about the benefits and risks of these treatments
hinders the development of clear guidelines. Physical exercise and other non-
pharmacological methods are recommended as first-line treatments in the
Swedish national guidelines, particularly since the risks associated with such
treatments are considered low. Traditionally, aerobic exercise has been
especially recommended in MS, but there is a growing interest in resistance
training in general, including high-intensity resistance training. At the time this
doctoral project was planned, no published studies had investigated high-
intensity resistance training specifically in people with high levels of fatigue.
Additionally, off-label use of medications was occurring, that is, the use of
medications outside of their approved indication, where Swedish national
guidelines, based on limited evidence, recommended such use only in
exceptional cases. This includes the drug amantadine, as well as wakefulness-
promoting medications such as modafinil and ADHD medications. However,
more detailed knowledge about how these medications were used in Sweden,
and to what extent they had an effect on fatigue, was lacking.

Randomized controlled trials (RCTs) are widely accepted as the best study
design for examining the effects of treatments and other interventions. This
design involves randomly assigning study participants to one of two or more
treatment arms before the intervention begins, which makes it easier to attribute
any differences in outcomes to the intervention rather than to other factors.
However, this study design is often not feasible for practical or ethical reasons. In
such cases, one can instead use data collected in the context of routine clinical
care or through structured studies where participants have not been
randomized to an intervention, so-called non-interventional studies. However,
this makes the results more difficult to interpret, as treatment choices in clinical
practice are influenced by patient and physician preferences as well as local
treatment traditions. By statistically adjusting for differences in characteristics
between comparison groups, one can reduce, but not entirely eliminate the risk
that differences in outcomes are due to factors other than the intervention itself.

Sweden has a large number of nationwide registries that enable population-level

research. These registries contain extensive health data, such as the National
Patient Register (with diagnoses from healthcare) and the Prescribed Drug
Register (with information on all dispensed prescription medications). Other key
registries contain sociodemographic information such as place of residence, age,
sex, educational level, and details on sick leave, disability pension, and activity
compensation. In addition, there are more than one hundred so-called quality
registries. The Swedish MS Registry is a high-coverage, MS-specific quality
registry that includes approximately 80% of all individuals with MS in Sweden.
The registry contains a variety of disease-related measures continuously
recorded by physicians, nurses, and other healthcare professionals, as well as
several outcome measures reported directly by patients, including fatigue
measures. Thanks to Sweden’s personal identity number, data from different
registries can be linked at the individual level, allowing for research studies of
internationally high quality.

The aim of this thesis was to examine the relationships between fatigue and
other MS-specific factors, and to explore the potential effects of both
symptomatic drug treatments and non-pharmacological interventions. To
achieve these goals, we conducted three nationwide studies using data from the
Swedish MS Registry and other national registers, as well as a smaller randomized
controlled trial that evaluated the effect of high-intensity resistance training in
individuals with MS experiencing fatigue. The following specific research
questions formed the basis of the individual sub-studies:

Study I: To what extent is fatigue explained by MS-specific factors, and how

much by other potential factors such as comorbidity?
In this registry-based study, all fatigue scores assessed using the FSMC scale
were included for individuals with relapsing or secondary progressive MS
registered in the Swedish MS Registry between 2012 and 2018. This included a
total of 3,179 measurements from 2,165 individuals. Based on established cutoffs
in the scale scores, 41% reported none or mild fatigue, 13% mild, 14% moderate,
and 32% severe fatigue. The only MS-specific factors associated with clinically
relevant differences in fatigue scores were the degree of disability and cognitive
processing speed. Fatigue was also strongly associated with self-reported
impact of MS, reduced quality of life, reduced work capacity, psychiatric
comorbidities, and treatments for depression, anxiety, and fatigue-related
symptoms.

Study II: How does fatigue develop over time from the start of first disease-
modifying treatment and from the time of first switch of disease-modifying
treatment in people with relapsing MS?

Using an overlapping national population of MS patients, we identified the

progression of fatigue from the start of a first disease-modifying treatment
(1,587 individuals) or a first switch in treatment (1,818 individuals) in the
subsequent years. We identified five fatigue trajectories in the group starting
their first treatment and six in the group switching treatment. All trajectories
were stable over time, except for one clinically relevant increase in fatigue in one
of five individuals at treatment initiation, who also had a moderate level of fatigue
at the start.

Study III: Can group-based high-intensity resistance training reduce fatigue and
other related symptoms in people with MS who experience fatigue, and is there
a difference in effect between training twice per week versus once per week?

This was a two-armed, randomized controlled trial conducted over 12 weeks, in

which participants with high levels of fatigue were randomly assigned to group-
based training with a physiotherapist either twice weekly or once weekly. A non-
randomized control group with similar characteristics served as a comparison
group outside of the main study. Both intervention groups showed a clinically
relevant reduction in fatigue after the study, with no difference between the
groups. In contrast, the level of fatigue remained unchanged in the group that did
not participate in the training program.

Study IV: What is the pattern of use for medications aimed at alleviating fatigue
in MS in Sweden, and can their use be linked to a positive effect on the ability to
work, measured by sickness absence, disability pension and activity
compensation?

In this national register-based study, we identified individuals with MS from the

Swedish MS Registry and the National Patient Register. We investigated the
annual prevalence of prescriptions for fatigue-related medications between
2006 and 2023 among patients residing in Sweden with a diagnosis of MS either
prior to or during the respective year. Furthermore, we identified new users of
each fatigue medication and compared them to a matched control population
with MS who had not received any of these medications. The wakefulness-
promoting drug modafinil was the most commonly prescribed treatment, with an
annual prevalence of 8.5% in 2006 and 7% in 2023. All medication groups, as well
as the untreated group, exhibited increasing levels of work loss in the year
preceding treatment initiation, followed by a stable trend without further
increases in work loss during the subsequent 24 months. Users of modafinil
showed a significantly reduced rate of increase in work loss compared to the
untreated group. However, no significant differences were observed between the
various medication groups included in the study.
Based on these results, we drew the following conclusions:

I. With regard to MS-specific factors, fatigue in MS is primarily associated

with the degree of neurological disability and impairment in cognitive
processing speed, while other analyzed factors showed weak or no
associations.
II. The level of fatigue shows a stable trend in the years following the
initiation of a first disease-modifying therapy or treatment switch, except
for individuals with moderate fatigue at the start of initial therapy, who
reported increasing fatigue over time.
III. High-intensity resistance training is associated with a clinically relevant
reduction in fatigue among individuals with MS who experience fatigue,
but without a clear effect of training frequency.
IV. Modafinil has been the most commonly prescribed medication for
treating fatigue in individuals with MS in Sweden over the past two
decades. Initiation of modafinil and other fatigue-targeting drugs
coincides with a slowing in the increase of work loss, though no clear
differences in effect were seen between the medication groups.
Populärvetenskaplig sammanfattning
Multipel skleros (MS) är en kronisk inflammatorisk och nervskadande sjukdom i
det centrala nervsystemet, det vill säga hjärnan och ryggmärgen, där uppkomna
skador leder till funktionsnedsättning som varierar i karaktär och omfattning. Ett
av de vanligare och mest funktionspåverkande symtomen är fatigue, ett slags
energilöshet som ofta beskrivs som överväldigande, viket har en negativ
påverkan på livskvalitet, arbetsförmåga, familjeliv, fritid och sociala relationer. Det
finns ingen allmänt accepterad definition av fatigue vid MS, men patienter
beskriver det ofta i termer av mental dimma, låg energinivå och känsla av
utmattning avseende både mentala och fysiska förmågor. Fatigue är således i
grunden en subjektiv upplevelse som vanligen utvärderas med självrapporterade
frågeformulär, vilka uppskattar svårighetsgraden genom dess påverkan på olika
funktionsdomäner i det dagliga livet. I och med att definitionen av fatigue
varierar, liksom utformningen av olika frågeformulär, försvåras möjligheterna att
jämföra resultat mellan olika studier. Skalan Fatigue Scale for Motor and
Cognitive Functions (FSMC) har dock utvecklats specifikt för MS och har under
senare år blivit ett av de mest använda.

En mängd olika faktorer har föreslagits som orsaker till fatigue vid MS, varav vissa
är direkt kopplade till MS-specifika fysiologiska processer. Andra potentiella
förklaringar är samsjuklighet och då särskilt med psykiatriska tillstånd så som
depression och ångest. Dessa är vanligt förekommande vid MS och är på egen
hand kopplade till förekomst av fatigue. Tidigare studier inom MS har gett
indikationer på faktorer som kopplar till fatigue, men tolkningen försvåras av att
många studier har varit små och sällan följt personer över en längre tidsperiod.
Det förefaller dock som att fatigue har flera bakomliggande orsaker, som varierar
mellan individer.

Det finns ännu inget botemedel för MS, men under det senaste decenniet har
alltmer effektiva bromsmediciner som skyddar mot akuta försämringsepisoder
utvecklats. I mindre grad skyddar dessa även mot utveckling av permanent
funktionsnedsättning, särskilt för patienter med skovvis MS, men i vilken
utsträckning de påverkar fatigue vid MS är mer oklart. En annan kategori av
behandlingar är de som riktar sig mot symtom av MS. I varierande utsträckning
används både farmakologiska och icke-farmakologiska behandlingsalternativ för
fatigue vid MS, men bristande kunskap om fördelar och risker med dessa
behandlingar försvårar framtagande av tydliga riktlinjer. Fysisk träning och andra
icke-farmakologiska metoder rekommenderas som förstahandsbehandling i
svenska nationella riktlinjer, inte minst som behandlingsrelaterade risker anses
vara små. Traditionellt har särskilt konditionsträning rekommenderats vid MS,
men det finns ett ökande intresse för styrketräning, inklusive högintensiv
styrketräning, rent allmänt. Vid planeringen av detta doktorandprojekt fanns ännu
inga publicerade studier med högintensiv styrketräning inriktat specifikt mot
personer med hög grad av fatigue. Vidare förekom användning av läkemedel
utanför sin egentliga indikation (så kallad off-label), där svenska nationella
riktlinjer baserat på bristande kunskapsunderlag endast rekommenderade sådan
användning i undantagsfall. Detta rör läkemedlet amantadin, samt
vakenhetshöjande läkemedel så såsom modafinil och ADHD-läkemedel. Mer
detaljerad kunskap om hur dessa läkemedel användes i Sverige, samt i vilken
grad de har effekt på fatigue, var dock bristfällig.

Randomiserade kontrollerade studier har bred acceptans som den bästa

studiedesignen för att undersöka effekter av behandlingar och andra
interventioner. Detta innebär att studiedeltagarna lottas till en av två eller flera
behandlingsarmar innan interventionen startar, vilket gör att en eventuell skillnad
i utfall lättare kan härledas till interventionen snarare än till andra faktorer. Denna
studiedesign är dock ofta inte möjlig att genomföra på grund av praktiska eller
etiska skäl. I dessa situationer kan man i stället använda data som insamlas inom
ramen för vanlig klinisk vård eller via strukturerade studier där deltagarna inte har
lottats till en intervention, så kallade icke-interventionsstudier. Detta gör dock att
resultaten blir mer svårtolkade, eftersom behandlingsval i klinisk praxis påverkas
av patientens och läkarens preferenser samt lokala behandlingstraditioner.
Genom att statistiskt korrigera för skillnader i karaktäristik mellan grupperna som
jämförs kan man minska men inte helt ta bort risken för att skillnader i utfall beror
av annat än själva interventionen i sig.

I Sverige finns ett stort antal rikstäckande register som möjliggör forskning på
populationsnivå. Dessa register innehåller omfattande hälsodata, såsom
patientregistret (med diagnoser från sjukvården) och läkemedelsregistret (med
uppgifter om alla uthämtade receptbelagda läkemedel). Andra centrala register
innehåller sociodemografisk information såsom bostadsort, ålder, kön,
utbildningsnivå samt uppgifter om sjukskrivning, sjukersättning och
aktivitetsersättning. Utöver detta finns över hundra så kallade kvalitetsregister.
Det svenska MS-registret är ett MS-specifikt kvalitetsregister med hög
täckningsgrad, som omfattar omkring 80 % av alla som har MS i Sverige. Registret
innehåller en rad olika sjukdomsrelaterade mått som löpande registreras av
läkare, sjuksköterskor och annan vårdpersonal, men även flera utfallsmått som
rapporteras direkt av patienterna, inkluderande förekomst av fatigue. Genom det
svenska personnumret kan data från olika register länkas på individnivå, vilket
medger forskningsstudier som ur ett internationellt perspektiv håller mycket hög
kvalitet.

Syftet med denna avhandling var att undersöka sambanden mellan fatigue och
andra MS-specifika faktorer, samt att utforska de potentiella effekterna av både
symtomlindrande läkemedelsbehandlingar och icke-farmakologiska insatser. För
att uppnå dessa mål genomförde vi tre nationsövergripande studier med data
från det svenska MS-registret och andra nationella register, samt en mindre
randomiserad kontrollerad studie som utvärderade effekten av högintensiv
styrketräning hos personer med MS som upplever fatigue. Följande specifika
forskningsfrågor låg till grund för de separata understudierna:

Studie I: I vilken utsträckning förklaras fatigue av MS-specifika faktorer, och hur

mycket av andra potentiella faktorer såsom samsjuklighet?

I denna registerbaserade studie inkluderades alla mätvärden på fatigue skattade

med FSMC skalan för personer med skovvis eller sekundärprogressiv MS
registrerade i det svenska MS-registret mellan 2012 och 2018, vilket totalt
omfattade 3 179 mätpunkter från 2 165 individer. Baserat på etablerade
avgränsningar i skalpoäng, rapporterade 41% ingen eller låg, 13% mild, 14% måttlig
och 32% svår fatigue. De enda MS-specifika faktorer som var associerade med
kliniskt relevanta skillnader i fatigue poäng var grad av funktionsnedsättning och
kognitiv processhastighet. Fatigue var också starkt associerad med
självrapporterad påverkan av MS, nedsatt livskvalitet, nedsatt arbetsförmåga,
psykiatriska samsjukligheter samt behandlingar för depression, ångest och
fatigue-relaterade symtom.

Studie II: Hur utvecklas fatigue över tid från start av första bromsmedicin samt
från tidpunkten för första byte av bromsmedicin hos personer med skovvis MS?

Utgående från en överlappande nationell population av MS patienter

identifierade vi utveckling av fatigue från start av en första bromsbehandling
(1 587 personer) eller ett första läkemedelsbyte (1 818 personer) under de
följande åren. Vi identifierade fem trendlinjer i gruppen med start av första
bromsmedicin och sex i gruppen med ett första byte av bromsmedicin. Samtliga
trendlinjer var stabila över tid, med undantag för en kliniskt relevant ökning i
fatigue hos en av fem vid behandlingsstart av en första bromsmedicin, vilka
också kännetecknades av måttlig svårighetsgrad av fatigue vid behandlingsstart.

Studie III: Kan gruppbaserad högintensiv styrketräning minska fatigue och andra
relaterade symtom hos personer med MS som upplever fatigue, och finns det en
skillnad i effekt mellan att träna två gånger i veckan jämfört med en gång i
veckan?

Detta var en två-armad, randomiserad kontrollerad studie som pågick under 12

veckor där deltagare med hög nivå av fatigue lottades till en gruppbaserad
träning med fysioterapeut antingen två gånger i veckan eller en gång i veckan. En
icke-randomiserad kontrollgrupp med personer med liknande karaktäristik
utgjorde en kontrollgrupp utanför den egentliga studien. Båda
interventionsgrupper uppvisade en kliniskt relevant minskad nivå av fatigue efter
studien, men utan skillnad mellan grupperna. Däremot var graden av fatigue
oförändrad i gruppen som inte deltagit i träningsprogrammet.

Studie IV: Hur ser användningen ut av läkemedel för att lindra fatigue vid MS i
Sverige och kan användningen kopplas till en positiv effekt på arbetsförmågan,
mätt i form av sjukfrånvaro samt sjuk- eller aktivitetsersättning?

I denna nationella registerbaserade studie identifierade vi personer med MS från

det svenska MS-registret samt patientregistret. Vi studerade förekomsten av
förskrivningar av läkemedel mot fatigue varje år mellan 2006 och 2023 bland
patienter som var bosatta i Sverige och hade en MS-diagnos innan eller under
det aktuella året. Vidare identifierade vi nya användare av respektive fatigue-
läkemedel och jämförde dessa med en matchad kontrollpopulation med MS,
som inte behandlats med något av dessa läkemedel. Det vakenhetshöjande
läkemedlet modafinil var den mest förskrivna behandlingen, med en årlig
förekomst på 8,5 % år 2006 och 7 % år 2023. Alla läkemedelsgrupper samt den
obehandlade gruppen uppvisade en ökande sjukfrånvaro under året före
behandlingsstart, följt av en stabil trendlinje utan ökad sjukfrånvaro under de
följande 24 månaderna. Användare av modafinil uppvisade en signifikant minskad
grad i ökningstakten av arbetsoförmåga jämfört med den obehandlade gruppen.
Däremot noterades inga signifikanta skillnader mellan de olika
läkemedelsgrupperna som ingick i studien.
Baserat på dessa resultat drog vi följande slutsatser:

I. Avseende MS-specifika faktorer, är fatigue vid MS primärt associerat med

grad av neurologisk funktionsnedsättning och påverkan på kognitiv
processhastighet, med svaga eller avsaknad av samband för övriga
analyserade faktorer.
II. Grad av fatigue uppvisar en stabil trend åren efter start av en första
bromsmedicin eller ett första läkemedelsbyte, med undantag för att
personer med måttlig fatigue vid start av en första bromsbehandling
rapporterar ökande fatigue.
III. Högintensiv styrketräning är associerat med kliniskt relevant minskning av
fatigue hos personer med MS som upplever fatigue, men utan tydlig
påverkan av träningsfrekvens.
IV. Modafinil är det mest förskrivna läkemedlet för behandling av fatigue hos
personer med MS i Sverige under de senaste två decennierna. Start av
modafinil och andra läkemedel mot fatigue kopplar tidsmässigt till en
uppbromsning i ökningstakten för sjukfrånvaro, utan tydlig skillnad i effekt
mellan de ingående läkemedelsgrupperna.
Abstract
Fatigue is a self-reported symptom, often described as an extreme lack of
energy, which is common across many medical conditions. In multiple sclerosis
(MS), it is frequently reported as the most prevalent and debilitating symptom,
with significant impact on quality of life and daily functioning. Underlying causes
of fatigue in MS are still elusive, which impedes development of effective
therapeutic strategies. The objectives of my thesis were to investigate
associations between fatigue among people with MS and a range of
demographic and disease-related measures, as well as to explore potential
effects of symptomatic pharmaceutical and non-pharmaceutical interventions.
To this end, we conducted three nationwide cohort studies using data from the
Swedish MS Register and other national registers, as well as a smaller randomized
controlled trial (RCT) evaluating the effect of resistance training.

Study I was a register-based, cross-sectional study to assess the association

between fatigue and a range of demographic and disease-related measures. We
included 3,179 Fatigue Scale for Motor and Cognitive Functions (FSMC) scores
obtained from 2,165 individuals with relapsing-remitting MS (RRMS) or secondary
progressive MS (SPMS) between 2012 and 2018, all of which were or had been
undergoing disease-modifying therapy (DMT) for MS. Of these, 40.7% reported
no or low fatigue, 12.7% mild, 14.0% moderate, and 32% severe fatigue. Expanded
Disability Status Scale (EDSS) score and information processing speed,
measured with Symbol Digit Modalities Test (SDMT), were the only objective MS-
related measures displaying a clinically meaningful differences across fatigue
severity levels; FSMC scores were 17.6 points higher with severe (EDSS ≥6) vs
mild disability (EDSS 0-2.5; 95% CI: 13.1-22.2), and 10.7 points higher in the lowest
vs highest SDMT quartiles (95% CI: 8.0-13.4), respectively. However, self-
reported impact of MS and quality of life showed even greater differences
across fatigue severity levels. Additionally, fatigue severity was associated with
work loss, psychiatric comorbidities, and treatments for depression, anxiety, and
fatigue-related symptoms.

Study II was a register-based cohort study in which we identified and compared

fatigue and disability trajectories, measured with FSMS and EDSS, respectively, in
individuals with RRMS. Additionally, we investigated associations between
potential predictors of fatigue and the identified fatigue trajectories. We
included 1,587 individuals followed for a mean (SD) of 7.1 (2.2) years from
initiation of a first DMT (first DMT cohort), and 1,818 people with MS followed for a
mean (SD) of 7.3 (1.9) years from initiation of a second DMT (DMT switch cohort).
Using group-based trajectory modeling, we identified five fatigue trajectories in
the first DMT cohort, and six in the DMT switch cohort. All fatigue trajectories
displayed stable trends over time, except for one trajectory (20.1% of
participants, moderate fatigue at treatment initiation) of the first DMT cohort,
who reported an 11.5-point increase in FSMC during follow-up. A strong
association was observed between fatigue and disability trajectories in both
cohorts. Apart from fatigue severity at DMT initiation, few other covariates
independently predicted membership in the higher stable or increasing fatigue
trajectories.

Study III was a two-armed RCT exploring the effect of group-based high-
intensity resistance training (HIRT) for 12 weeks, supervised by a physiotherapist,
in people with MS reporting at least moderate fatigue. Seventy-one participants
were randomized 1:1 to HIRT once or twice a week, with an additional 69
individuals with similar characteristics serving as an uncontrolled, non-
intervention control group. Based on intention-to-treat, mean changes in FSMC
scores were -9.8 (95% CI: -13.2 to -6.3) and -10.9 (95% CI: -14.8 to -6.9) in the
once and twice weekly HIRT groups, respectively, indicating clinically relevant
improvements, but with no significant difference between the two groups.
Corresponding values for the combined HIRT group and the non-intervention
control group were -10.3 (95% CI: -12.9 to -7.7) and 1.5 (95% CI: -0.6 to 3.6),
respectively.

Study IV was a register-based cohort study to explore the use of central

stimulants and amantadine as symptomatic off-label treatments for fatigue in
MS, and to evaluate their impact on work loss. We identified 29,257 people with
MS through the Swedish MS and National Patient Registers. To assess annual
prescription rates, we created yearly cohorts of individuals alive and residing in
Sweden from January to December of each year between 2006 and 2023. For
additional analyses, we included 2,162 new modafinil users, 462 new amantadine
users, and 424 new users of attention deficit hyperactivity disorder (ADHD)
drugs, who were followed for 12 months before and 24 months after their first
filled prescription (index date). We also identified a cohort of 9,762 individuals
with no prior fatigue treatment, using risk-set sampling, matched on first
modafinil start. Modafinil was the most prescribed treatment, with an annual
prevalence of 8.5% in 2006 and 7% in 2023. All cohorts showed increasing work
loss in the year preceding the index date, followed by no change in work loss
rates during the subsequent 24 months. Modafinil users demonstrated a
significantly greater reduction in the trajectory of average monthly work loss
compared to the untreated cohort (–0.17 days; 95% CI: –0.22 to –0.12). No
significant differences were observed between the modafinil group and the
other treated cohorts.

In conclusion, fatigue in MS is associated with greater disability and slower

cognitive processing speed, while other MS-related characteristics show weak or
no associations. Fatigue severity remains relatively stable in the years following
both the first and second initiation of DMT, except for a clinically meaningful
increase among a minority with moderate fatigue at first DMT start. Moreover,
HIRT is associated with clinically meaningful reductions in fatigue among fatigued
individuals with MS, although a definitive causal relationship could not be
established due to the lack of a randomized non-intervention control group.
Over the past two decades in Sweden, modafinil has been the most commonly
prescribed off-label pharmacological treatment for fatigue in MS. Modafinil and
other stimulants may help mitigate the progression of work loss in people with
MS, although no single treatment appears to be superior in effectiveness.
List of scientific papers
I. Predictors of patient-reported fatigue symptom severity in a
nationwide multiple sclerosis cohort
Simon Englund, Marie Kierkegaard, Joachim Burman, Katharina Fink,
Anna Fogdell-Hahn, Martin Gunnarsson, Jan Hillert, Annette Langer-
Gould, Jan Lycke, Petra Nilsson, Jonatan Salzer, Anders
Svenningsson, Johan Mellergård, Tomas Olsson, Elisa Longinetti,
Thomas Frisell, Fredrik Piehl
Mult Scler Relat Disord. 2023;70:104481.

II. Trajectories of self-reported fatigue following initiation of multiple

sclerosis disease-modifying therapy
Simon Englund, Thomas Frisell, Ying Qu, Kavita Gandhi, Annika
Hultén, Marie Kierkegaard, Fredrik Piehl, Elisa Longinetti
J Neurol Neurosurg Psychiatry. 2024;95:1012-1020.

III. High-intensity resistance training in people with multiple sclerosis

experiencing fatigue: A randomised controlled trial
Simon Englund, Fredrik Piehl, Marie Kierkegaard
Mult Scler Relat Disord. 2022;68:104106.

IV. Use of symptomatic drug treatment for fatigue in multiple sclerosis

and its impact on work loss: a register-based Swedish cohort study
Simon Englund, Johan Reutfors, Thomas Frisell, Fredrik Piehl
Manuscript
Contents
1 Background .......................................................................................................................................................... 1
1.1 Multiple Sclerosis ............................................................................................................................... 1
1.1.1 Epidemiology ........................................................................................................................ 1
1.1.2 Pathophysiology................................................................................................................. 1
1.1.3 Diagnosis and symptoms ........................................................................................... 1
1.1.4 Morbidity and mortality ..............................................................................................2
1.1.5 Treatment and other interventions ...................................................................3
1.2 Fatigue in Multiple Sclerosis ................................................................................................... 4
1.2.1 Epidemiology of fatigue ............................................................................................. 4
1.2.2 Pathophysiology of fatigue ..................................................................................... 4
1.2.3 Impact of fatigue on daily life................................................................................ 5
1.2.4 Symptomatic treatment for fatigue ................................................................ 5
1.3 Measuring disease activity, progression, and patient-reported
impact of MS........................................................................................................................................ 8
1.3.1 Traditional measures of disease activity and progression
in MS ........................................................................................................................................... 8
1.3.2 Patient-reported outcome measures (PROMs).................................... 9
1.4 Previous research on predictors and treatment for fatigue ........................11
1.4.1 Predictors of fatigue in MS .......................................................................................11
1.4.2 Evolution of fatigue in MS ........................................................................................ 13
1.4.3 Exercise and resistance training for fatigue in MS ............................. 13
1.4.4 Pharmacological treatment for fatigue in MS ......................................... 14
2 Research aims................................................................................................................................................. 15
3 Materials and methods ........................................................................................................................... 17
3.1 Study design ....................................................................................................................................... 17
3.1.1 Randomized controlled trial (RCT)................................................................... 17
3.1.2 Cohort study ...................................................................................................................... 17
3.2 Data sources ....................................................................................................................................... 18
3.2.1 MS specific data sources ........................................................................................18
3.2.2 Health registers ................................................................................................................19
3.2.3 Demographic Registers ........................................................................................... 20
3.3 General introduction to methodology .......................................................................... 21
3.3.1 Statistical methods....................................................................................................... 21
3.3.2 Epidemiological considerations ........................................................................22
3.3.3 Missing data ...................................................................................................................... 32
 3.4 Study design and analysis..................................................................................................... 34
3.4.1 Study I - Predictors of fatigue in MS ............................................................ 34
3.4.2 Study II - Trajectories of fatigue after initiation of MS
DMTs ........................................................................................................................................ 37
3.4.3 Study III - HIRT in people with MS experiencing fatigue .............. 39
3.4.4 Study IV - Symptomatic drug treatment for fatigue in MS ...... 42
3.5 Ethical considerations .............................................................................................................. 46
4 Results ................................................................................................................................................................. 49
4.1 Study I - Predictors of fatigue in MS ............................................................................ 49
4.1.1 Baseline characteristics .......................................................................................... 49
4.1.2 Potential predictors for fatigue ........................................................................ 49
4.2 Study II - Trajectories of fatigue after initiation of MS DMTs .................... 51
4.2.1 Trajectories of fatigue and disability ............................................................. 51
4.2.2 Membership across fatigue and disability trajectories. .............. 53
4.2.3 Baseline characteristics .......................................................................................... 53
4.2.4 Predictors of fatigue trajectories .................................................................... 53
4.2.5 Linking fatigue trajectories to incidence of psychiatric
comorbidities and work loss............................................................................... 53
4.3 Study III - HIRT in people with MS experiencing fatigue .............................. 55
4.3.1 Baseline characteristics .......................................................................................... 55
4.3.2 Effects of HIRT................................................................................................................. 56
4.4 Study IV - Symptomatic drug treatment for fatigue in MS....................... 58
4.4.1 Annual fatigue treatment prescription rate ........................................... 58
4.4.2 Characteristics at index and rate of re-filled prescriptions .... 58
4.4.3 Effectiveness of fatigue treatment ............................................................... 59
5 Discussion ........................................................................................................................................................ 63
5.1 Findings ................................................................................................................................................. 63
5.1.1 Study I - Predictors of fatigue in MS ............................................................ 63
5.1.2 Study II - Trajectories of fatigue after initiation of MS
DMTs ....................................................................................................................................... 65
5.1.3 Study III - HIRT in people with MS experiencing fatigue .............. 68
5.1.4 Study IV - Symptomatic drug treatment for fatigue in MS ...... 70
5.1.5 Final thoughts ...................................................................................................................74
5.2 Methodological considerations ......................................................................................... 77
5.2.1 Precision................................................................................................................................ 77
5.2.2 External validity (generalizability) ....................................................................78
5.2.3 Causal inference.............................................................................................................79
 5.2.4 Internal validity ...............................................................................................................80
6 Conclusions .................................................................................................................................................... 89
7 Points of perspective ...............................................................................................................................91
8 Acknowledgements ................................................................................................................................. 93
9 References ....................................................................................................................................................... 97
List of abbreviations
ADHD Attention deficit hyperactivity disorder

ATC Anatomical Therapeutic Chemical

CI Confidence interval

CNS Central nervous system

COMBAT-MS Comparison Between All Immuno-Therapies for

Multiple Sclerosis

COSMIN COnsensus-based Standards for the selection of

health Measurement INstruments

DAG Directed acyclic graph

DMT Disease-modifying therapy

EDSS Expanded Disability Status Scale

EQ-VAS EQ visual analogue scale

FSS Fatigue Severity Scale

FSMC Fatigue Scale for Motor and Cognitive Functions

HADS Hospital Anxiety and Depression Scale

HIRT High-intensity resistance training

HRQoL Health-related quality of life

ICD International Classification of Diseases

LISA Longitudinal integrated database for health insurance

and labour market studies

MAR Missing at Random

MCAR Missing Completely at Random

MD Mean difference

MFIS Modified Fatigue Impact Scale

MiDAS Mikrodatabas för analys av sjukfrånvaro

MNAR Missing Not at Random

MRI Magnetic resonance imaging

MS Multiple sclerosis
MSIS-29 Multiple Sclerosis Impact Scale-29

NPR National Patient Register

OGQ Occupational Gaps Questionnaire

PDR Prescribed Drug Register

PPMS Primary-progressive MS

PROM Patient-reported outcome measure

PwMS People with multiple sclerosis

RCT Randomized controlled trial

RM Repetition maximum

RRMS Relapsing-remitting MS

SD Standard deviation

SDMT Symbol Digit Modalities Test

SMSreg Swedish Multiple Sclerosis Register

SPMS Secondary-progressive MS
Introduction
The outlook for people diagnosed with Multiple Sclerosis (MS) have improved
significantly since the introduction of disease-modifying therapies in the 1990s,
particularly with regard to reduced disability rates and increased life
expectancy. As these advances have reshaped the important aspects of the
disease, less visible yet common and impactful symptoms, such as fatigue, have
received increasing scientific and clinical attention. Despite numerous efforts to
define and understand fatigue in MS, there is still no universally accepted
definition, and its underlying mechanisms remain elusive. MS stands out among
many conditions in that both pharmacological and non-pharmacological
treatments are available for the symptomatic management of fatigue, although
the supporting evidence for these interventions remains incomplete.

Studying fatigue presents with many challenges, including the absence of a clear
definition and objective measures. Further, its complexity requires large study
populations to be meaningful. In consequence, research on fatigue in MS often
fails to yield clear or immediately clinically applicable findings. These challenges
can be discouraging and may deter researchers from studying this prevalent and
debilitating symptom. We had the opportunity to study fatigue and related
factors in large population-based cohorts of MS patients, using partially
prospectively collected data linked to nationwide registers, as well as
conducting a randomized controlled trial to evaluate the effect of resistance
training on fatigue. My aim with this thesis was to contribute new insights into
the predictors and treatment options for fatigue in MS, a challenge I couldn’t
resist taking on.
1 Background
1.1 Multiple Sclerosis
Multiple sclerosis (MS) is a chronic autoimmune disorder of the central nervous
system (CNS). It is characterized by episodes of exacerbation (relapses) or
gradual progression of symptoms, which may include motor impairments,
fatigue, and cognitive difficulties.1 Although the cause of MS remains uncertain,
it's believed that a combination of environmental, lifestyle, and genetic factors
contributes to its onset and progression.2

1.1.1 Epidemiology

MS typically emerges between the ages of 20 and 30, with a female-to-male

ratio of 3:1.2 Globally, around 2 million people are affected by MS,3 with the
highest prevalence in Europe, North America, Australia, and New Zealand.3 In
Sweden, the prevalence of MS ranges from 190 to 215 per 100,000,4,5
representing between 20,140 and 22,790 people among Sweden’s 10.6 million
inhabitants. The incidence rate is approximately 10 cases per 100,000 person-
years,6 meaning that around 1,000 individuals are diagnosed with MS each year.

1.1.2 Pathophysiology

The pathophysiology of MS is not fully known, but it is believed that the immune
system mistakenly targets the myelin, the protective covering around the nerve
fibers (axons) in the CNS. This immune response triggers inflammation, leading to
demyelination and axonal degeneration, which result in the formation of lesions.
These lesions disrupt the normal flow of electrical impulses along the nerves (i.e.,
dysfunction in neuronal circuits). Over time, ongoing inflammation and the
accumulation of lesions contribute to neurodegeneration and brain atrophy.2

1.1.3 Diagnosis and symptoms

MS is typically classified into three primary types based on its disease course:
relapsing-remitting MS (RRMS), secondary-progressive MS (SPMS), and primary-
progressive MS (PPMS).1 The most common form at diagnosis is RRMS, which is
characterized by episodes of exacerbations, known as relapses, with varying
degrees of recovery. To be classified as a relapse, symptoms must persist for at
least 24 hours. Symptoms typically worsen over days to weeks, peak within 2-3
weeks, and recover within 2-4 weeks. Most individuals with RRMS eventually
transition to SPMS, marked by progressive and often irreversible worsening of

1
disability, regardless of relapses. Approximately 10-15% of people with MS
(PwMS) have continuous disease progression already from the disease onset, so
called as PPMS. Although this classification is still relevant, it is important to note
that disability accumulation in MS today primarily occurs independently of
relapses.7

The clinical presentation of MS varies depending on the location of

demyelinating lesions within the CNS and can include a range of symptoms, such
as sensory disturbances, motor impairments, brainstem and cerebellar
dysfunction, cognitive impairments, fatigue, mood changes, sexual dysfunction,
and sphincter dysfunction.1 Sensory disturbances are the first manifestation in
40-50% of PwMS and include symptoms such as paresthesia, numbness and
tingling, impaired vibration and joint sense, and reduced pain and touch
perception. Motor impairments affect approximately 30-40% of individuals early
in the disease and nearly all PwMS over time. These impairments include
pyramidal signs such as Babinski reflex, exaggerated reflexes, clonus, paresis, and
spasticity. Brainstem and cerebellar dysfunction occur in up to 70% of PwMS
and manifests as nystagmus, diplopia, ataxia, gait imbalance, poor coordination,
slurred speech, and swallowing difficulties. Cognitive impairments are observed
in 40-70% of PwMS, primarily affecting information processing speed and
episodic memory.8

1.1.4 Morbidity and mortality

While survival rates for PwMS have increased over the years, life expectancy
remains 5-7 years shorter than that of the general population.9 Common
comorbidities in MS include depression, anxiety, hypertension, hyperlipidemia,
diabetes, heart disease, and chronic lung disease.10,11 Studies have shown that
these comorbidities are more prevalent among PwMS compared to the general
population, interestingly even several years before an MS diagnosis.11 Moreover,
evidence suggests that certain comorbidities may affect traditional MS-related
outcomes such as disability, magnetic resonance imaging (MRI) findings, and
cognitive impairment.12

Psychiatric comorbidities, particularly depression and anxiety, are often reported

as the most prevalent comorbidities in MS.10,11 A systematic review of population-
based studies estimated the prevalence of depression at 24% and anxiety at
22% in PwMS.13 Depression and anxiety are common even at the time of MS
diagnosis and tend to become more prevalent as the disease progresses.13 The

2
reasons for the strong correlation between depression and MS is likely
multifactorial.13 Immunological and inflammatory changes, along with brain
atrophy and lesion accumulation, may contribute to the development of
depression in MS. Additionally, both depression and anxiety can arise as
psychological responses to chronic illness, with psychosocial factors such as
social support, coping mechanisms, and stress playing a significant role.13

The prevalence of sleep disorders in MS is less studied, but research suggests

that certain sleep disorders are more common in PwMS than in the general
population.14,15 A systematic review of non-population-based studies reported
prevalence estimates ranging from 0-2% for narcolepsy, 14-58% for restless legs
syndrome, 2-3% for rapid eye movement behavior disorder, and 7-58% for
obstructive sleep apnea.16 In addition, several MS-related symptoms and
common comorbidities in MS can affect sleep quality including pain, bladder
dysfunction, spasticity, muscle cramps, depression and medication side
effects.14

1.1.5 Treatment and other interventions

Currently, there is no definitive cure for MS. Due to the heterogeneity of its
clinical presentation, interventions and treatments vary widely, ranging from
symptom management to therapies aimed at preventing relapses and reducing
the rate of disease progression. Disease-modifying therapies (DMTs) are the
cornerstone of treatment for most of PwMS.17 DMTs are particularly effective for
RRMS and aim to prevent relapses and improve long-term outcomes. Since their
introduction in the 1990s, DMTs have likely contributed to reducing overall
disease severity in MS.17

Symptomatic treatments focus on alleviating MS-related symptoms without

directly modifying the disease course. Their goal is to relieve symptoms, reduce
complications, improve functional ability, and enhance quality of life.18 Common
symptoms that may require targeted interventions include fatigue, cognitive
impairment, mobility impairment, swallowing and speech difficulties, spasticity,
pain, and bladder and bowel dysfunction.2 Treatment options typically involve a
combination of pharmacological and non-pharmacological approaches.18
National guidelines recommend non-pharmacological interventions such as
symptom management courses, strength training, or endurance training as first-
line treatments for fatigue, with pharmacological options considered only in
exceptional cases.18 In contrast, for conditions like spasticity, pharmacological

3
treatment is often the first-line approach, typically in combination with non-
medical interventions.18

1.2 Fatigue in Multiple Sclerosis

Fatigue is often described as an extreme lack of energy, and is a common
symptom across multiple psychological and physical conditions, such as
depression, cancer, neurological and inflammatory diseases.19,20 It is one of the
most prevalent symptoms of MS and has a great impact on quality of life, as well
as other aspects of daily living, such as maintaining employment.21 There is no
universally accepted definition of fatigue in MS. However, one of the most
frequently cited definitions originates from the 1998 Multiple Sclerosis Clinical
Practice Guidelines, which characterize fatigue in MS as “A subjective lack of
physical and/or mental energy that is perceived by the individual or caregiver to
interfere with usual and desired activities”.22 PwMS often describe fatigue as a
mental fog, a low battery, or feelings of depletion, exhaustion, or tiredness,
affecting both cognitive and motor functions.23

1.2.1 Epidemiology of fatigue

Fatigue is often reported as the most prevalent symptom of MS throughout the

disease course and appears to be common even in its early stages.24 According
to a systematic review, the prevalence of fatigue in MS ranges from 37% to 78%,
with most studies reporting figures between 60% and 70%.21 A Norwegian
population-based study found a fatigue prevalence of 81% using the Fatigue
Scale for Motor and Cognitive Functions (FSMC), with 15% experiencing mild, 19%
moderate, and 67% severe fatigue.25 Fatigue in other neurological conditions has
been studied less extensively than in MS, however, it also appears to be highly
prevalent. Prevalence rates include approximately 32-50% in Parkinson’s
disease, around 75% in post-stroke fatigue, and 80-90% in amyotrophic lateral
sclerosis.20 Attempts to estimate the prevalence of fatigue in the general
population have found figures around 16%.26

1.2.2 Pathophysiology of fatigue

The specific cause(s) of fatigue in MS remain unknown. However, more than 30

suggested pathophysiological factors have been identified.27 Some of these
factors are considered part of the underlying MS disease, while others are
attributed to common comorbidities in MS.22 The cause of fatigue likely varies
between individuals, with comorbidities playing a significant role for many.20

4
Inflammation, demyelination, axonal loss, brain atrophy, and dysfunction of
neuronal circuits are considered central to the pathophysiology of MS and have
also been proposed as potential contributors to fatigue in PwMS.27 Inflammation,
particularly elevated levels of pro-inflammatory cytokines, is one of the most
studied contributor to fatigue.27-29 The observation that relapses can intensify
fatigue symptoms, along with the frequent reporting of fatigue by newly
diagnosed PwMS, supports this hypothesis.24,27,30 Moreover, evolving brain
atrophy, has been suggested as a potential predictor of fatigue in MS. More
pronounced fatigue is associated with a higher degree of atrophy in both the
white and gray matter of the brain.27 Reduced or blocked nerve conduction
velocity and neuroaxonal degeneration result in decreased CNS activation and
can cause various neurological symptoms in MS, potentially including fatigue.27
This is especially evident when body temperature is elevated, such as during
infection or physical activity, and can sometimes be misinterpreted as a relapse,
a so called pseudo-relapse.31 Moreover, PwMS often report that fatigue
symptoms are triggered or exacerbated by stress and environments with
excessive stimuli.20

1.2.3 Impact of fatigue on daily life

Fatigue has a significant impact on people, affecting their daily lives, work,
finances, and overall well-being.21 A qualitative interview study of PwMS reported
that leisure, social, and family activities were often limited due to a lack of energy
after completing essential responsibilities such as work and daily tasks.23
Participants noted that engaging in too many activities could result in
exhaustion, sometimes requiring rest for several hours or even days. Some
reported having to modify their job responsibilities, reduce working hours, or
stop working altogether.

1.2.4 Symptomatic treatment for fatigue

Population-based studies have shown that only one-third of PwMS who

experience fatigue receive any form of treatment for their symptoms.24,32
Managing fatigue likely requires individualized treatments or interventions, as its
causes are probably heterogeneous.33 Multiple meta-analyses and systematic
reviews have compiled evidence on a range of fatigue interventions, including
both pharmacological and non-pharmacological strategies.34-38 However, the
ability to compare outcomes across studies is hindered by substantial variations
in interventions and fatigue measures used. Additionally, many of these studies

5
exhibit methodological limitations, including short follow-up periods and not
assessing fatigue as a primary outcome. As a result, the conclusions and
recommendations drawn from these studies provide limited guidance for
clinicians and patients.

One of the most comprehensive reviews of available treatment options is the

2022 clinical guidelines by the National Institute for Health and Care Excellence
in the United Kingdom. These guidelines recommend several non-
pharmacological approaches, including exercise interventions such as aerobic
exercise, resistance training, balance exercises, and yoga, along with behavioral
interventions like cognitive-behavioral therapy, mindfulness-based training,
fatigue management techniques, and energy conservation therapy.35 Although
evidence for these interventions is still limited, the minimal risks associated with
exercise and behavioral therapies support their recommendation for addressing
fatigue in MS. Pharmacological treatments were found to be less effective and
are associated with the risk of side effects. Therefore, they are recommended
only in special circumstances.35 Similarly, the National Board of Health and
Welfare’s national guidelines for MS in Sweden also recommend non-
pharmacological treatments as the first choice. On a 1-10 priority scale, fatigue
management courses, resistance training, and aerobic training received a rating
of 4, while pharmacological treatments, including amantadine and modafinil,
were rated 10.18

1.2.4.1 Exercise

Physical activity is essential for overall health, not only by reducing mortality and
morbidity but also by enhancing quality of life and mood.39 Over the past few
decades, a rapidly growing body of evidence has highlighted the benefits of
exercise as a treatment for various chronic diseases, including MS.40 Traditionally,
PwMS were advised to preserve reserves, however, they are now generally
encouraged to exercise.18 While some PwMS may experience discomfort and
temporary symptoms during exercise, leading to concerns about risk of relapses,
these symptoms are typically short-lived and have been shown not to trigger
relapses.36,40

The understanding of how exercise affects fatigue in MS revolves around several

hypotheses.27 Some suggest that exercise may influence fatigue by modulating
peripheral proinflammatory factors, and by stimulating the production of
neuroendocrine growth factors. These growth factors may help protect nerve

6
cells and promote neuroplasticity, the ability of the brain to reorganize and form
new connections.41 Additionally, exercise is believed to alleviate fatigue by
addressing conditions such as depression, sleep disorders, and physical
deconditioning.27 Traditionally, PwMS have been encouraged to participate in
aerobic training.42 However, there is growing interest and accumulating evidence
supporting the benefits of resistance training, including high-intensity
protocols.43,44 Resistance training may directly impact MS by influencing cytokine
levels,27 aligning with research suggesting that skeletal muscles function as a
secretory organ, generating and releasing cytokines in response to contraction,
thereby contributing to anti-inflammatory effects.45

1.2.4.2 Pharmacological treatment

There is no pharmacological treatment officially approved for fatigue in MS.18

However, in clinical practice, modafinil, amantadine, and amphetamine-like CNS
stimulants are used off-label to promote wakefulness and help PwMS manage
fatigue in daily life.18

Modafinil, a non-amphetamine CNS stimulant with wakefulness-promoting

properties, was initially introduced in Sweden in 2001 to treat narcolepsy and
soon began to be used off-label to manage fatigue in MS. Since 2010, the
European Medicines Agency has restricted its use exclusively to adults
experiencing excessive sleepiness associated with narcolepsy, with or without
cataplexy.46 In the Unites States, however, modafinil still has broader indications,
including use in adults with narcolepsy, obstructive sleep apnea, shift work sleep
disorder, and jet lag.

Amantadine, an older drug originally developed for influenza A, was later

approved for treating Parkinson’s disease as well.47 The neurochemical effects of
amantadine are not fully understood, but it is believed to mitigate fatigue via a
dopaminergic mechanism, supporting the dopamine imbalance hypothesis.48
This hypothesis links fatigue to CNS dopamine imbalance, further supported by
the effectiveness of dopaminergic psychostimulants in conditions like chronic
fatigue syndrome and cancer.48 An individual license from the Swedish Medical
Products Agency is required to prescribe amantadine to PwMS.

In Sweden, amphetamine and amphetamine-like CNS stimulants, including

dexamphetamine, methylphenidate, atomoxetine, and lisdexamphetamine, are
approved for attention deficit hyperactivity disorder (ADHD) treatment.49 Only
psychiatrists, and as of 2018, neurologists, are authorized to prescribe these

7
drugs.49 Previously, an individual license from the Swedish Medical Products
Agency was required for these treatments, and for amphetamine, it is still
needed.

1.3 Measuring disease activity, progression, and patient-reported

impact of MS
Several measures are available to quantify and grade the clinical manifestations
of MS, including disability, relapse, lesion burden, cognitive impairment, health-
related quality of life (HRQoL), and fatigue.

1.3.1 Traditional measures of disease activity and progression in MS

Disease activity and progression are commonly evaluated based on the

occurrence of relapses, MRI findings, and changes in disability.2 The Expanded
Disability Status Scale (EDSS) is a widely accepted measure for quantifying
disability in MS.1 It ranges from 0, indicating a normal neurological examination
(no disability), to 10, representing death due to MS. The EDSS includes 8
subscales, representing functional systems that assess the major neurological
domains affected by MS. These domains encompass the pyramidal (motor),
cerebellar, brainstem, sensory, bowel and bladder, visual, mental, and other
related functions.50 MRI is essential for assessing disease activity and
progression in MS by detecting new or enlarging T2 lesions, which reflect ongoing
disease activity. It helps identify subclinical disease activity that may not be
apparent through clinical symptoms and is used to monitor the effectiveness of
DMTs.2

Cognitive impairment is partially assessed as a self-reported component of the

EDSS, but this provides only a rough measure of cognitive function.8 Several
validated tools are available for evaluating cognitive impairment in MS, varying in
complexity and administration time.51 The Minimal Assessment of Cognitive
Function in Multiple Sclerosis52 is a comprehensive test battery that takes
around 90 minutes to complete and requires trained personnel, limiting its use in
clinical practice. Instead, more time-efficient tools are often used. One of the
most widely used tools is the Symbol Digit Modalities Test (SDMT), which is part
of many test batteries, including the Minimal Assessment of Cognitive Function
in Multiple Sclerosis.51 It takes only 5-10 minutes to complete and assesses
information processing speed, showing high sensitivity to cognitive impairment.51

8
1.3.2 Patient-reported outcome measures (PROMs)

Patient-reported outcome measures (PROMs) are reports of a patient's health

status provided directly by the patient.53

1.3.2.1 The Multiple Sclerosis Impact Scale-29 (MSIS-29)

The Multiple Sclerosis Impact Scale-29 (MSIS-29) is a PROM used to assess the
impact of MS on a person’s physical and psychological well-being.54 The scale
consists of 29 items, which are divided into two subscales: a physical impact
subscale (20 items) and a psychological impact subscale (9 items). Participants
respond to each item based on how much they have been affected by MS over
the past two weeks, addressing aspects such as physical limitations, fatigue,
emotional well-being, common MS symptoms, and the overall impact of the
disease on their life. Higher scores indicate worse health.

1.3.2.2 EQ-5D

The EQ-5D is a self-reported measure used to assess a person’s overall health

status.55 It consists of five dimensions: mobility, self-care, usual activities,
pain/discomfort, and anxiety/depression. A Visual Analogue Scale is also
included to capture the person’s self-rated overall health, with higher scores
indicating better HRQoL.

1.3.2.3 Fatigue measures

The gold standard for assessing fatigue in MS relies on the patient’s subjective
experience. In both research and clinical practice, PROMs are used to capture
recalled summaries of fatigue severity and its impact on daily life.56 There is a
variety of generic and MS-specific PROMs available, each capturing different
aspects of fatigue, as defined in various ways.56 Among the most widely used
MS-specific PROMs are the Fatigue Severity Scale (FSS),57 and the Modified
Fatigue Impact Scale (MFIS).22 The FSS, introduced in 1989, is a unidimensional
tool that assesses fatigue severity and its overall impact on daily activities, with
an emphasis on motor fatigue. It consists of nine items rated on a 7-point scale,
without an explicit reference to a specific time frame. Higher scores indicate a
greater impact of fatigue.57 The MFIS, introduced in 1997, examine the impact of
fatigue on physical, cognitive, and psychosocial functioning over the past month.
As such, the MFIS provides a more comprehensive view of how fatigue affects
daily life compared to the FSS. It includes 21 items and generates both a total

9
score and three subscale scores (physical, cognitive, and psychosocial), with
higher scores indicating a greater impact of fatigue on daily functioning.22

In the studies of my doctoral thesis, we have primarily used the Fatigue Scale for
Motor and Cognitive Functions (FSMC)58 to assess fatigue, due to its well-
established use in the Swedish MS community. Introduced in 2009, the FSMC
evaluates the impact of fatigue on both motor and cognitive functions in daily
life. A summary of the included items is presented in Table 1.1. The FSMC does
not specify a particular time frame; instead, it asks respondents to reflect on
their usual experience of fatigue in day-to-day life. It consists of 20 items and
generates both a total score and two subscale scores (motor and cognitive),
with higher scores indicating a greater impact of fatigue on these functions.

Table 1.1. Summary of the items in the Fatigue Scale for Motor and Cognitive Functions
(Adapted from Penner et al., 2009)
Item key in the cognitive subscale Item key in the motor subscale
Concentration Skillfulness
Decision making/executive functions Stamina/resting periods
Learning Stress and physical power
Occupational demands Social environment
Stress and concentration Muscles/strength
Heat and thinking Physical stamina
Thinking/motivation/drive Drive/motivation
Verbal fluency Speed reduction
Attention/stamina Reactivity
Memory Heat and physical energy

1.3.2.4 Psychometric properties of fatigue measures

Several reviews have systematically assessed the measurement properties of

PROMs used to evaluate fatigue in MS, employing the COSMIN checklist.56,59-61
Psychometric properties, including reliability, construct validity, and
responsiveness, serve as measurable indicators of a scale’s overall quality.53
These properties are evaluated through statistical tests, providing a clear and
direct assessment of a scale’s strengths and are commonly reported.53

Reliability is often assessed through internal consistency and test-retest

reliability. Internal consistency is high for the FSS (Cronbach’s alpha α = 0.93),62
MFIS (Cronbach’s alpha α = 0.96)62 and FSMC (Cronbach’s alpha α = 0.91).58 Test-
retest reliability, which reflects the stability of scores over time when no change

10
is expected, is also high for FSS (intraclass correlation coefficient = 0.75), MFIS
(intraclass correlation coefficient = 0.86), and FSMC (Pearson correlation =
0.87).53

Construct validity often involves evaluating the sensitivity and specificity of a

scale in detecting fatigue in PwMS, typically through convergent validity
(correlation with similar constructs) and discriminant validity (correlation with
unrelated constructs).53 This involves administering the PROM alongside related
and unrelated psychometric instruments.53 High convergent and discriminant
validity have been reported for FSMC,58 moderate for MFIS,60 and conflicting
results for FSS.60 One exception is depression, where all three scales are highly
correlated with self-reported depression.58

Additionally, assessing clinically relevant within-individual thresholds is crucial

for interpretability in both clinical practice and research settings. While well-
established thresholds exist for clinically relevant differences in FSS and MFIS,
evidence for FSMC is weak.60,63 However, FSMC has an advantage over other MS-
specific fatigue scales due to its predefined cut-off scores.56

1.4 Previous research on predictors and treatment for fatigue

1.4.1 Predictors of fatigue in MS

The association between fatigue and other MS-related symptoms and

comorbidities remains poorly understood. However, some associations are more
established than others. It is well known that fatigue negatively affects HRQoL in
PwMS,21 and a strong association exists between fatigue and psychiatric
comorbidities, such as depression and anxiety, although the causal relationship
remains unclear.64 In contrast, there is limited consensus regarding the
association between fatigue and MS-related characteristics, including MS type,
disease duration, disability, and relapse.65

A longitudinal study found that higher EDSS scores were associated with greater
fatigue severity, even after adjusting for factors such as age, sex, disease
duration, age of symptom onset, and MS type.66 On the other hand, a Swedish
longitudinal study found no evidence of EDSS as a predictor of increasing fatigue
(measured by the FSS) after adjusting for a wide range of factors, including age,
sex, sense of coherence, living situation, work status, DMT, depressive symptoms,
MS type and time since diagnosis.67 Moreover, findings from previous cross-
sectional studies on the association between EDSS and fatigue have been

11
inconsistent. A Norwegian study found an association between EDSS and fatigue
(measured by the FSMC) in multivariable analyses, although the specific
covariates included were not clearly reported.25 Another cross-sectional study
observed an association between fatigue and EDSS, but this association
disappeared after adjusting for depression.68

Studies examining MS type as a predictor of longitudinal fatigue have also

reported mixed results66,67,69. One longitudinal study found that PwMS with a
progressive disease course were more likely to experience increasing fatigue
compared to those with RRMS.66 However, after adjusting for EDSS and MS
duration, MS type was no longer a significant predictor of fatigue.66 A similar
pattern was observed in a cross-sectional study, where an association between
fatigue and MS type was evident in unadjusted analyses but disappeared after
adjusting for age and EDSS.25 In contrast, another cross-sectional study reported
that MS type remained associated with fatigue even after adjusting for other
MS-related variables.69 The association between fatigue and time since MS
diagnosis remains unclear.65,69,70

Studies examining the association between fatigue in MS and cognitive

impairment have also reported conflicting results. One study found no
association between fatigue (measured by the FSS) and cognitive impairment, as
assessed using the Minimal Assessment of Cognitive Function in Multiple
Sclerosis, in either cross-sectional or longitudinal analyses.71 In contrast, other
studies have reported associations between higher fatigue levels and slower
information processing speed 58,72 as well as impaired working memory.58

Similarly, studies exploring the relationship between fatigue and demographic

factors, such as age and sex, have yielded inconsistent findings,65 although some
studies suggest a higher prevalence of fatigue among women.25 Additionally,
fatigue severity has been associated with level of education, employment status,
and work loss, including sick leave and disability pension.21,65

Several common comorbidities and symptoms of MS have been linked to, and
could potentially cause fatigue.27 These include psychiatric disorders, sleep
disorders, physical deconditioning, and medication side effects.27 Psychological
disorders such as depression and anxiety are frequently observed in PwMS, and
there is a well-established association between these psychiatric comorbidities
and fatigue severity, as well as their impact on HRQoL.13,64 However, the causality
of this relationship remains uncertain, partly due to potentially shared

12
pathophysiological mechanisms and overlapping features.73 Common physical
comorbidities of MS, such as hypertension, lung disease, and thyroid
dysfunction, may also contribute to fatigue, although they are less studied.10
Furthermore, there is potential overlap between primary sleep disorders and
fatigue in MS.14,33

1.4.2 Evolution of fatigue in MS

The long-term evolution of fatigue in PwMS remains another poorly studied area.
Most existing studies indicate persistent or fluctuating fatigue patterns over
time.67,70,74-77 However, the follow-up periods in these studies were in general
short, ranging from 1 to 2.5 years,67,70,74-76 with the exception of one small study
that followed 96 participants over a 10-year period, assessing fatigue at two
time points.77 Additionally, the disease duration of study participants varied, with
means/medians ranging from 7 to 16.5 years,67,74-77 except for one study that
followed 230 participants from MS diagnosis.70

In a recent study from the United States, 944 PwMS were followed with a median
follow-up of 8 years, and a maximum of 5 years since MS diagnosis at inclusion.78
This study found significant worsening of fatigue in 52% of the participants.78 In
contrast, several observational studies have investigated the evolution of fatigue
following the initiation of specific DMTs. While some studies suggest no change
in fatigue levels79-82 or fluctuating fatigue patterns,83 a considerable number
report a decrease in fatigue following DMT initiation.84-90 However, these studies
generally had short follow-up periods, ranging from 12 weeks to 3 years.79-81,83-90
One observational study, which followed 1,128 participants for a mean of 16.3
years after teriflunomide initiation, demonstrated no change in fatigue.82

1.4.3 Exercise and resistance training for fatigue in MS

Several meta-analyses and systematic reviews have highlighted the positive

effects of specific exercise modalities on fatigue, including resistance
training,35,38,91 aerobic exercise,35,36,91 balance exercises,35,91 yoga,35 and combined
exercise programs that integrate at least two exercise components.36,38

Few studies have specifically assessed resistance training as an intervention

among PwMS who experience fatigue. A recent meta-analysis assessing the
effects of resistance training on fatigue in PwMS found an overall decrease in
fatigue, regardless of whether participants belonged to the intervention or
control group.44 However, interventions varied across the included studies,

13
particularly in the control group, which included components such as stretching,
usual care, and instructions to either refrain from engaging in exercise or not to
alter existing exercise habits.44

1.4.4 Pharmacological treatment for fatigue in MS

As previously discussed, evidence supporting the effect of pharmacological

treatments for fatigue in MS is generally considered poor.34,35,92 Multiple trials
have investigated the efficacy of modafinil in MS, yielding conflicting results.34,35,93
Studies that reported improvements in fatigue scores were predominantly
single-arm trials.94-96 However, in the majority of trials with more complex
designs, modafinil did not demonstrate superiority over placebo.97-101 Various
randomized controlled trials (RCTs) comparing amantadine with placebo have
shown significant fatigue reduction with amantadine.100,102-106 However, meta-
analyses and systematic reviews have considered the evidence from most
amantadine studies to be limited by small sample sizes and short follow-up
periods.34,35,92 Apart from a few clinical trials involving pemoline,106,107 there have
been limited RCTs evaluating amphetamine-like CNS stimulants for managing
fatigue in MS. These studies did not demonstrate superiority over placebo in
alleviating fatigue.101,108

14
2 Research aims
The overall aim of this thesis was to investigate the association between fatigue
in MS and other disease measures, as well as to explore the potential effects of
both symptomatic pharmaceutical and non-pharmaceutical interventions for
fatigue in MS.

Specific aims:

I) To evaluate the extent to which MS-related characteristics and

cognitive processing speed are independently associated with fatigue
in MS and to assess whether these associations are influenced by
sociodemographic characteristics, work loss, comorbidity, HRQoL, and
self-reported physical and psychological impact of MS.
II) To identify fatigue trajectories and their association with disability
trajectories, and to explore predictors of fatigue trajectory groups.
III) To evaluate the effects of high-intensity resistance training (HIRT) on
fatigue, mood, and HRQoL in fatigued PwMS by comparing HIRT twice a
week to once a week, and by comparing the entire HIRT group to a
matched, non-randomized cohort as a non-intervention control group.
Additionally, to explore associations between changes in fatigue and
plasma inflammatory protein levels.
IV) To describe the utilization of central stimulants and amantadine as
symptomatic treatments for fatigue in MS, and to evaluate their
potential comparative effectiveness in reducing work loss in PwMS.

15
3 Materials and methods
3.1 Study design
Three of the studies in my doctoral thesis were cohort studies, and one was a
randomized controlled trial (RCT).

3.1.1 Randomized controlled trial (RCT)

In Study II we conducted a RCT. An RCT is an experimental study (i.e., with

researcher intervention) and is often considered the gold standard for assessing
the efficacy of an intervention, a treatment, or varying doses of an intervention or
treatment. Participants are followed prospectively from time zero to evaluate
outcomes.

RCTs have some important key features. Randomization ensures that

participants are randomly assigned to one of two or more groups. If sample size
is sufficiently large, this process helps create comparable groups at time zero,
isolating the effect of the intervention from other factors. Controlled refers to
the comparison between the intervention group and a control group, which
typically receives a placebo, usual care, another intervention, or no treatment.
Trial refers to the structured testing of the intervention following a predefined
study protocol. Blinding is another important feature that helps reduce bias,
though it is not always feasible. In a single-blind study, participants typically do
not know which treatment they are receiving, while in a double-blind study, both
participants and researchers are unaware of the assigned treatment.

When conducted properly, RCTs provide strong evidence of causality. However,

they are often time-consuming, expensive, and may have low generalizability.
Additionally, some research questions cannot be addressed using an RCT due to
ethical constraints.

3.1.2 Cohort study

A cohort study is an observational study (i.e., without researcher intervention) in

which a group of individuals with shared characteristics (e.g., exposed) is
compared to another group (e.g., unexposed) by following them over time to
assess health outcomes (e.g., disease or death). Cohort studies are either
prospective or retrospective, depending on the data collection process. In a
prospective study, a cohort is identified at time zero and followed forward in
time to observe outcomes as they occur. In contrast, retrospective studies use

17
existing records to define the cohort and analyze past exposure and outcome
data. While prospective studies allow for better control over data collection, they
are often more expensive and time-consuming. Retrospective studies, on the
other hand, are more efficient but may carry a higher risk of incomplete data in
relation to the specific research question.

In Study IV, we applied a retrospective study design based on register data.

Similarly, in Study II, we used a retrospective design, but the data came from the
COMBAT-MS study, which employed a mixed prospective and retrospective
approach. In Study I, we conducted a cross-sectional analysis using register
data. Instead of following individuals over time, we examined associations
between characteristics at a single time point. However, the use of register data
allowed us to organize some characteristics in a temporal order, which was
partially done.

3.2 Data sources

Swedish and other Nordic administrative data are widely recognized for their
usefulness for population-based studies due to their high quality as well as the
ability to link individual data across national registers using a unique personal
identification number.109 This enables longitudinal assessment of individual-level
data, allowing researchers to track health outcomes and other important
characteristics over time. Additionally, Sweden has over 150 disease-specific
quality registers, primarily used to improve healthcare for patients with specific
diseases. These registers are also extensively used in research, as they collect
individual-level data on clinical and patient-reported outcomes that are often
unavailable in national administrative registers. Participation in these registers is
voluntary for both patients and treating clinicians, leading to variations in
coverage.

All data sources used in Studies I, II, and IV are described below. The
specification of study design for each study are detailed in Section 3.4: Study
Design and Analysis.

3.2.1 MS specific data sources

3.2.1.1 The Swedish Multiple Sclerosis Register (SMSreg)

The Swedish MS Register (SMSreg) is an MS-specific quality register introduced

in 2001. It has good coverage, encompassing nearly 80% of Sweden’s estimated
prevalent MS population.110 It is a valuable tool for epidemiological research and

18
an important clinical resource, facilitating the monitoring and management of MS
patients. MS-related characteristics are collected during routine clinical visits,
including EDSS scores, information on conversion to SPMS, DMT use, and
patient-reported outcome measures. Additionally, relapses and MRI findings are
recorded as they occur.

3.2.1.2 The Comparison Between All Immuno-Therapies for Multiple Sclerosis

(COMBAT-MS) study

The COMBAT-MS study was a population-based cohort study in RRMS that

assessed the benefit-risk balance of MS therapies compared with rituximab.111
Among 9,979 eligible individuals, 3,905 who initiated their first or second DMT
between 2011 and 2018 were included in the study. Prospective data from annual
assessments conducted between 2017 and 2022 were recorded in the SMSreg.
Additionally, retrospective data from 2011 to 2017 were retrieved from the
SMSreg and supplemented by a medical chart review to ensure validity of the
data.112 Of the included individuals, 2,580 contributed both prospective and
retrospective data, 506 contributed retrospective data only, and 817
contributed prospective data only.

3.2.2 Health registers

3.2.2.1 The National Patient Register (NPR)

The National Patient Register (NPR) is maintained by the National Board of Health
and Welfare and covers all inpatient visits in Sweden since 1987, including
admissions to somatic, geriatric, and psychiatric hospital care.113 Since 2001, it
also covers outpatient visits from specialist care provided by physicians.
Additionally, it has recorded data on day surgery since 1997 and compulsory
psychiatric care since 2010. The NPR does not include information on primary
care or specialized outpatient visits in somatic care provided by healthcare
professionals other than physicians. The register includes details such as visit
dates, admission and discharge dates, and main and contributory International
Classification of Diseases (ICD) codes.

3.2.2.2 The Prescribed Drug Register (PDR)

The Prescribed Drug Register (PDR) is also a national register maintained by the
National Board of Health and Welfare, launched in 2005.114 It records Anatomical
Therapeutic Chemical (ATC) codes for all prescription drugs dispensed from
pharmacies in Sweden. The register includes information such as the prescribing

19
date, dispensing date, drug amount and dosage, as well as the prescriber's
profession. However, it does not include information on over-the-counter
medications or drugs used in hospitals. Another limitation is that it is partially
incomplete for ambulatory care and nursing homes. Importantly, the register
does not capture the indication for treatment, although this information may be
partially inferred from free-text fields related to the prescription provided by
prescribing physicians.

3.2.3 Demographic Registers

3.2.3.1 The Total Population Register

The Total Population Register is an administrative database maintained by

Statistics Sweden.115 It contains information on all individuals residing in Sweden,
including basic demographic details such as age, sex, civil status, migration
history, and region of residence.

3.2.3.2 The longitudinal integrated database for health insurance and labour market
studies (LISA)

LISA is also a national register maintained by Statistics Sweden, launched in

1990.116 It contains information on various sociodemographic factors aggregated
on a calendar year level, including sick leave, disability pension/activity
compensation, unemployment benefits, disposable income, education, civil
status, and migration. LISA covers nearly all individuals aged 15 years and older in
Sweden (16 years and older before 2010).

3.2.3.3 MiDAS

MiDAS (Mikrodatabas för analys av sjukfrånvaro) is an administrative database

maintained by the Swedish Social Insurance Agency, launched in 2004 to
facilitate the analysis of social insurance benefit data. It contains information on
sickness absence, disability pension, and activity compensation for all individuals
in Sweden. MiDAS provides individual-level data on the duration and reasons for
absence, as well as the associated costs. This database serves as a valuable
resource for studying the impact of work loss, both at the individual level and
within a broader societal context.

20
3.3 General introduction to methodology

3.3.1 Statistical methods

3.3.1.1 Regression analysis

Regression analysis is a statistical method used to assess the relationship

between a dependent variable (outcome) and one or several independent
variables (exposures). While both types of variables can follow various
probability distributions, only a few regression models are commonly applied in
epidemiological research and clinical trials. Frequently used models include
generalized linear models, for example linear and logistic regression.

Generalized Linear Models: Generalized linear models consist of three main

components. The first component is the distribution of the residuals (or error
terms), such as the normal distribution for continuous data or the binomial
distribution for binary data. In practice, this often corresponds to the distribution
of the dependent variable. The second is the summation of the independent
variables, each multiplied by a coefficient. This combination forms the linear
predictor. The third component is the link function, which transforms the
expected value (mean) of the dependent variable to align with the linear
predictor.

In linear regression, the dependent variable is assumed to follow a normal

distribution, and the link function is the identity function. This means that the
expected value of the dependent variable (i.e., the mean outcome) is directly
modeled as a linear combination of the independent variables.

In logistic regression, the dependent variable is categorical. If the dependent

variable is binary and follows a binomial distribution, binary logistic regression is
used. The logit function is used as the link function and models the log-odds of
the dependent variable occurring as a linear combination of the independent
variables. The exponentiated parameters are odds ratios. Multinomial logistic
regression extends binary logistic regression to multiple categories. In this case,
the dependent variable follows a multinomial distribution, and the link function
compares the log-odds of each category relative to a reference category.

3.3.1.2 Regression analysis for repeated measurements

Longitudinal data include repeated measurements on the same subject.

Regression models to account for the correlation between repeated
measurements on the same subject are available in common statistical

21
softwares. Common regression models include generalized estimating equations
for population-averaged effects and mixed-effects models commonly said to
model subject-specific effects. The choice of model depends on the research
question and the assumptions about variability within and between individuals.

Generalized Estimating Equations extend generalized linear models by

accounting for the correlation between repeated measurements on the same
subjects. Generalized estimating equations provide population-averaged
estimates by using a working correlation structure to model the relationships
between repeated measurements within an individual. This approach employs
robust standard errors, which are robust to misspecifications of the correlation
structure, making it especially useful when the exact correlation between
measurements is uncertain.

3.3.1.3 Group-based trajectory modeling

Group-based trajectory modeling is a method to cluster individuals based on

similar outcome trajectories over time.117 These trajectories reflect the average
pattern of change within each group, but do not account for individual-level
variability. Various software programs support group-based trajectory modeling,
enabling the estimation of different distributions.117

3.3.2 Epidemiological considerations

3.3.2.1 Causal inference

In RCTs, the aim is to establish a causal relationship between an intervention and

one or more outcomes. In contrast, epidemiological research often measures
statistical associations. However, in many cases, we also seek to draw causal
conclusions from observational data. In the following section introducing causal
inference and directed acyclic graphs (DAGs), I was inspired by the KI doctoral
course Causal inference for epidemiological research by Arvid Sjölander and the
introduction to causal inference section in the thesis by Peter Alping.118 For
further accessible reading on causal inference, I recommend the short
introduction by Romio et al.119

A key difference between association and causation involves the concepts of

independence and association. Two random variables, X and Y, are
independent when the distribution of one remains unchanged across levels of
the other. For example, this would be the case if disability levels are the same
among women as they are among men. In contrast, two variables are said to be

22
associated if the distribution of one change across more than one level of the
other. For example, if disability levels differ between men and women.
Conditional independence occurs when X and Y are independent given a third
variable, Z. For example, the association between sex and disability may exist
without adjusting for age, but once we adjust for age, sex and disability may
become conditionally independent, indicating that age “explains” the association
between sex and disability. While association between variables is a necessary
condition for causality, it does not by itself imply a causal relationship.

3.3.2.2 The potential outcome framework of causal inference

The potential outcome model is a framework for defining causality by comparing

two hypothetical scenarios, known as potential outcomes: 1) The outcome if a
subject had been exposed to the intervention (the factual outcome), 2) the
outcome if the same subject had not been exposed (the counterfactual
outcome). If these two potential outcomes are different, we can say that the
exposure have a causal effect on the outcome. However, as we cannot observe
the counterfactual outcome at the same time as the factual outcome, the
counterfactual outcome is unobserved. This fundamental limitation makes it
unrealistic to examine individual causal effects, but we can estimate population-
level causal effects under specific assumptions.

Exchangeability is a key concept in the potential outcome model, meaning that

the potential outcomes are independent of exposure. In other words, if the
distribution of potential outcomes is the same across different levels of
exposure, then the exposed and unexposed groups are comparable. In RCTs,
exchangeability is guaranteed because treatment assignment is random
(unconditional exchangeability). In observational studies, exchangeability can be
assumed if confounders are sufficiently controlled for (conditional
exchangeability).

3.3.2.3 Directed acyclic graphs (DAGs)

Directed acyclic graphs (DAGs) are graphical representations of underlying

causal structures in the data. They allow us to visualize our prior causal
understanding or hypotheses into a structured diagram. DAGs are valuable tools
for covariate selection in regression models and can help visualize how study
design and missing data influence causal relationships. I will begin by explaining
the basics of DAGs, as I will later use them to illustrate and discuss various
epidemiological concepts.

23
Figure 3.1a illustrates a simple DAG, where each arrow represents a causal effect.
X, Y, and Z are called nodes and represent events occurring at specific points in
time. In this section, I will refer to them as variables for simplicity. A DAG is said
to be directed because connections between variables consists of arrows,
indicating the direction of causality. It is said to be acyclic because the graph
contains no directed cycles, meaning a variable cannot cause itself, either
directly or indirectly.

DAGs rely on several key assumptions:

 The presence of an arrow illustrate potential causality: An arrow from X to Y
implies that X could potentially impact Y, but Y cannot impact X (Figure 3.1a).
 The absence of an arrow illustrate non-causality: If there is no arrow from X to
Y, this indicates that X does not impact Y (Figure 3.1b).
 The existence of a variable Z illustrate that X and Y could potentially share a
common cause (Figure 3.1a), whereas the lack of Z illustaret that X and Y do
not share a common cause (Figure 3.1c)

(a) (b) (c)

Figure 3.1. Three simple DAGs illustrating key assumptions, including the presence of an
arrow from X to Y, the absence of an arrow from X to Y, and the role of common causes.

Figure 3.2 illustrates a slightly more complex DAG to explain the concept of
paths. A path can be explained as a connection between two variables, which
doesn’t have to follow the direction of the arrows. In this example, the paths
between X and Y include:
 Direct path: X -> Y
 Chain (mediator): X -> V -> Y
 Fork (common cause/confounder): X <- Z -> Y
 Inverted fork (collider): X -> W <- Y

24
A causal path is also a connection between two variables, however, a causal
path always follows the direction of the arrows. In Figure 3.2, the causal paths
between X and Y are:
 X -> Y
 X -> V -> Y

Figure 3.2. DAG to explain the concept of paths.

Causal and non-causal paths are either open or blocked, depending on what
variables we have controlled for, or by study design. The two key rules for
blocking paths are:
 If it contains a variable Z that appears as part of a chain (mediator: X->Z->Y)
or a fork (confounder: X <- Z -> Y), and we have conditioned on Z.
 If it contains a variable Z that appears as an inverted fork (collider: X->Z<-Y),
and we have not conditioned on Z or any variable affected by Z.

Using these blocking rules, we can determine whether two variables are
independent or associated. X and Y are independent if all paths between X and Y
are blocked. However, X and Y are associated if one or more path remains open.

3.3.2.4 Bias

Bias, or systematic error, refers to a systematic difference between an estimated

parameter and the true population parameter due to limitations in study design,
data collection, or analysis. It is important to distinguish bias from generalizability
and precision. While bias affects internal validity, generalizability refers to
external validity and indicates the extent to which study findings apply to other
populations beyond the study sample. Precision, on the other hand, reflects the
impact of random error on an estimate. It is influenced by sample size and
variability, meaning that a precise estimate can still be biased.

Bias is commonly categorized into confounding, selection bias (collider

stratification bias), and information bias (misclassification bias) in observational

25
studies. While these types of bias can also occur in RCTs, they are generally less
frequent due to the use of randomization and blinding.

Confounding: To best understand the concept of confounding, I will discuss it in

relation to a simple DAG (Figure 3.3). Confounding emerges when the exposure
(X) and the outcome (Y) have a common cause (Z). This structure creates a path
between X and Y (X <- Z -> Y) that is non-causal. As a result, X will be associated
with Y, but X may or may not cause Y. This open path can be blocked by
conditioning on Z, by adjusting for Z in a regression model, or by study design. If
we fail to condition on Z and observe an association between X and Y, this
association is either explained by a causal path (X -> Y) or through the path that
is not causal (X <- Z -> Y). By properly conditioning on Z, we can assert that we
have conditional exchangeability given Z.

Figure 3.3. DAG to explain the concept of confounding.

One intuitive example of confounding can be formulated like this: Assume X =

use of a drug (e.g., Drug A), Y = health outcome (e.g., symptom improvement), Z =
age of the patient. Assume that younger patients are more likely to being
prescribed Drug A than older patients, but they also tend to experience better
outcomes from Drug A. Hence, patients using Drug A are more likely to be
younger than those not using it, which could lead to better outcomes in the drug
group purely or partly due to age differences. If we fail to control for age, we
might find an exaggerated positive association between drug use and symptom
improvement, even though a true causal effect of the drug does not exist. This
type of confounding is often called confounding by indication.

It is important to note that the DAG must be correctly specified to draw valid
causal conclusions, and in practice, some degree of uncertainty always remains.
For example, residual confounding is always present to some degree due to
factors such as unmeasured variables or model misspecification.

Selection bias (collider stratification bias): Selection bias, also known as

collider stratification bias can arise due to non-random selection of study

26
participants, loss to follow-up, missing data, or incorrectly adjusting for colliders.
In DAG terms, it occurs when there is a variable W that acts as a collider in the
causal structure, as illustrated in Figure 3.4. Both the exposure (X) and the
outcome (Y) influence W, creating an inverted fork (X -> W <- Y). If we condition
on W, for example, by adjusting for it in a regression model or by restricting the
study sample to individuals who meet a certain criterion (such as being
hospitalized), we open up a non-causal path between X and Y going through W.
Hence, by not conditioning on W, we have unconditional exchangeability, given
W. It is important to consider the risk of selection bias when designing a study
and to avoid adjusting for colliders in analysis, such as through regression
modeling or stratification.

Figure 3.4. DAG to explain the concept of selection bias.

A classic example of selection bias is Berkson’s bias. This bias occurs when the
study sample is not drawn from the general population, but instead from a
hospitalized population. Imagine you want to assess whether diabetes (X) causes
cholecystitis (Y), and you only study hospitalized individuals. In this case, we
introduce a collider variable, W (Figure 3.4), representing the probability of being
included in the study. By conditioning on W (i.e., restricting the analysis to
hospitalized patients), we open a non-causal path: X -> W <- Y. Within this
selected group, individuals are hospitalized either due to diabetes or
cholecystitis, which creates a negative spurious association between the two
conditions.

Information bias (misclassification bias): Information bias, also called

misclassification bias, occurs when there is an error in measuring either the
exposure (X) or the outcome (Y), leading to misclassification. This can distort the
true association between X and Y. A key distinction is between non-differential
and differential misclassification:
 Non-differential misclassification occurs when the misclassification of X or Y
is unrelated to the other variable. This generally biases results toward the null,
meaning it underestimates the true association.

27
 Differential misclassification occurs when the misclassification of X is related
to Y (or vice versa). This can bias results in any direction, either exaggerating
or attenuating the true effect.

3.3.2.5 Specific study design to strengthen causal inference

Different study designs can be used when the ultimate goal is to establish a
causal relationship between an exposure and an outcome. In practice, this is
achieved by reducing various types of bias. As previously discussed, RCTs are
the gold standard for causal inference, but they are not always feasible or the
most suitable design for answering a research question. However, several study
designs can also enhance causal inference when using observational data.

RCT: In RCTs, unconditional exchangeability is guaranteed, if done properly,

because treatment assignment is random. Due to the randomization,
confounding by indication is not introduced. Hence, ensuring that both known
and unknown confounders are equally distributed across treatment and control
groups. This aligns with the potential outcome framework, which aims to
estimate the counterfactual outcome for each individual, what would have
happened in an alternative scenario. The active treatment group ideally
represents the evaluation of the outcome if a subject had been exposed to the
intervention, while the control group represents the outcome if the same subject
had not been exposed. However, as noted previously, we are not assessing
individual causal effects, but instead estimating population-level causal effects.

However, RCTs may suffer from low generalizability, as they often include a
selected group of participants. While random assignment reduces selection bias
at baseline (before the intervention is known), it does not remove the risk of
selection bias because of attrition (loss to follow-up). An intention-to-treat
analysis is commonly applied to mitigate selection bias. This analysis includes all
participants as originally assigned, regardless of whether they completed the
treatment. While it mitigates selection bias, it may introduce information bias if
participants stop or switch treatments during follow-up, as their outcomes
might not reflect the true effect of the initial treatment.

Active comparator, new user design: The active comparator, new user design is
often considered the gold standard in pharmacoepidemiological studies due to
its aim to approximate an RCT. This study design consists of two components:
the active comparator and the new user design.120 The active comparator
design involves comparing an active drug to another drug with the same

28
indication, rather than comparing it to a non-treatment group (or to another
drug with different indication). This helps reduce both measured and
unmeasured confounding. The new user design means that study participants
are followed from the start of treatment, enabling the assessment of pre-
treatment characteristics as well as assessment of the outcome from treatment
initiation. Confounding by indication is a particular problem in clinical practice, as
clinicians typically choose treatments based on patient characteristics, which
can cause imbalances across treatment and untreated groups, potentially
affecting the outcome. The active comparator design aims to reduce the
problem of confounding by indication. In Study IV, we partly used an active
comparator, new user design. In study II, our analyses were based on the
COMBAT-MS study, which applied an active comparator, new user design.

Interrupted time series analysis: Natural experiments are commonly used in

economics and epidemiology to infer causal relationships between interventions
and outcomes using observational data.121 These natural experiments attempt to
mimic RCTs by applying quasi-randomization, an allocation method that is not
truly random and is not assigned by a researcher. Similar to the active
comparator design, the goal is to reduce both measured and unmeasured
confounding at the study design stage by comparing exposure groups that are
comparable. In Study IV, we did not strictly conduct a natural experiment.
However, we applied interrupted time series analysis, a common method used in
natural experiments. Interrupted time series analysis compares the rate of an
outcome across multiple pre- and post-intervention time points, allowing for a
discontinuity (an “interruption”) in average rates of the outcome.122 This can be
done either within the intervention group alone (single-group interrupted time
series) or in comparison to a control group. In a single-group interrupted time
series, the slopes of best-fit lines before and after the intervention are
compared, with the option to also measure the immediate change in the
outcome due to the intervention. To compare differences in these changes
between the intervention and the control group, interaction terms are included
in the analysis.

Single-group interrupted time series are sometimes used to draw causal

conclusions if an adequate number of measurements are available before and
after the intervention, and if no other interruptions affect the outcome during the
study period. If the study population is static, the same individuals can serve as
their own control, which strengthens the causal interpretation of the results.

29
Moreover, the risk of regression to the mean is often an issue in longitudinal
analyses with only one pre-intervention measurement and one post-
intervention measurement, since extreme initial values (either high or low) are
likely by chance, to be followed by values closer to the average. By assessing
multiple measures over time, both pre- and post-intervention, the issue of
regression to the mean is reduced.122

However, the best approach is to compare the intervention group to a control

group, similar to the use of a control group in RCTs, to determine what would
have happened in a hypothetical scenario where the same subjects were not
affected by the intervention (the counterfactual). The causal interpretation of
the results is strengthened if the exposure groups are comparable in terms of
outcomes and covariates before the intervention, as well as in the trajectories of
the outcome variable. To further enhance comparisons between the groups,
adjustment methods such as using weighting methods can be used (see next
section).122

3.3.2.6 Controlling for variables to strengthen causal inference

Various analytical methods can be used to address confounding when

estimating a causal relationship between an exposure and an outcome.
Traditional approaches include conditioning on confounders through
stratification or covariate adjustment in regression models. Both methods
estimate conditional effects. Stratification involves dividing the sample into
subgroups based on a confounding variable (e.g., sex) and analyzing each
subgroup separately. Covariate adjustment in regression models accounts for
confounding by including potential confounders as covariates alongside the
exposure variable. For example, adjusting for sex in a regression model helps
estimate the exposure effect while holding sex constant.

An alternative approach is the use of weighting methods, which create a

pseudo-population where confounders are balanced across exposure groups.
This allows for the estimation of marginal effects, representing the overall
population-level impact of exposure rather than effects conditional on specific
confounder levels. Two commonly used weighting methods are risk set sampling
and inverse probability of treatment weighting. We used both of these methods
in Study IV. A simpler matching approach was applied in study III.

Matching and risk set sampling: Matching is a method used in epidemiological

studies to control for confounding by pairing or grouping participants with

30
similar characteristics (e.g., age, sex) while differing in the exposure of interest.
This helps ensure that differences in outcomes are more likely attributable to the
exposure rather than other factors. Risk set sampling with replacement is a
specific matching method commonly used in cohort studies to further minimize
bias.123 In this approach, participants are matched at the time of an exposure or
outcome event, ensuring that comparisons are made within the same risk set.
The replacement aspect means that an individual selected as a control can be
chosen again for another match.

Inverse Probability of Treatment Weighting: The aim of inverse probability of

treatment weighting is to generate a pseudo-population where treatment
exposure is independent of baseline covariates.124 Inverse probability of
treatment weighting generally involves the following steps:

 Define treatment groups and covariates: Define one or more treatment

groups and potentially a control group. Identify baseline covariates measured
before any intervention occurs. These covariates should be potential
confounders that influence both treatment assignment and the outcome.
 Estimate propensity scores: The propensity scores represents the probability
of receiving a specific treatment, conditional on baseline covariates. Usually it
is estimated using regression models, such as binary logistic regression for a
binary exposure or multinomial logistic regression when there are more than
two exposure levels.
 Calculate the inverse probability of treatment weights: For individuals in each
exposure group, the weight is calculated as the inverse of the estimated
propensity score. This means that individuals with a lower probability of
receiving the treatment (based on their covariates) receive a higher weight,
as they are underrepresented in the treatment group. Stabilized weights can
be applied by multiplying each weight by the marginal probability of
treatment in the overall sample. This helps reduce variance and improves the
stability of the estimates.
 Check balance of covariates: After weighting, one typically checks balance
across exposure groups using standardized mean differences. The
standardized mean difference is calculated by subtracting the mean of one
group from the mean of another and dividing the result by the pooled
standard deviation. A standardized mean difference of less than 0.1 between
exposure levels is generally considered well-balanced.

31
 Apply the weights in analysis: In the final step, the computed weights are
incorporated into the analysis. This is typically done by including them as
weights in regression models.

3.3.3 Missing data

3.3.3.1 Missing data mechanisms

Missing data is a significant issue in research, potentially affecting precision,

generalizability, and introducing bias. It is typically classified into three types
based on the mechanism causing the missingness:

 Missing Completely at Random (MCAR): Missingness is completely at

random, meaning that the probability of missing data is the same for all
participants and is unrelated to both observed and unobserved data. MCAR
reduces precision but does not introduce bias. While this scenario is rare in
real-world data, it can occasionally occur in RCTs and prospective cohort
studies, for example, due to accidental data entry errors.
 Missing at Random (MAR): Missingness depends on observed data but not
on unobserved data. This means that missingness can differ across groups
but remains predictable based on available information. While MAR can
introduce bias, it can be addressed using statistical techniques such as
multiple imputation. This scenario is realistic in real-world data. In RCTs, MAR
is common when dropout rates vary by treatment group or patient
characteristics.
 Missing Not at Random (MNAR): Missingness is related to unobserved data,
meaning that the reason data is missing is itself unknown or depends on the
missing values. MNAR cannot be addressed using standard statistical
methods based on observed data.

3.3.3.2 Dealing with missing data

Several methods are available to handle missing data. Complete case analysis
is a common and straightforward method, but it requires the missing data to be
MCAR. In this approach, any observation with missing data on at least one
variable is excluded from the analysis. Although complete case analysis is
unbiased under the MCAR assumption, it can lead to a reduction in precision. In
cases where data is MAR or MNAR, generalizability may be reduced, and
selection bias could be a problem. In Study I, we used complete case analysis to
handle missing data.

32
Multiple imputation involves generating multiple datasets in which missing
values are replaced with plausible estimates based on observed data and the
conditional distributions of relevant covariates.125 By performing this process
multiple times, the method accounts for uncertainty in the imputed values,
preventing underestimation of variability. The general approach involves creating
several versions of the incomplete dataset, where missing values are imputed
using probabilistic models. Each imputed dataset contains slightly different
estimates for the missing data, reflecting the inherent uncertainty. These
datasets are then analyzed separately, and their results are combined (pooled)
using Rubin’s rules, which appropriately account for both within-imputation and
between-imputation variability.125 In Study II and IV, we used multiple imputation
to handle missing data.

33
3.4 Study design and analysis

3.4.1 Study I - Predictors of fatigue in MS

3.4.1.1 Study design and study population

This was a nationwide, register-based, cross-sectional study in which we

assessed the association between fatigue in MS and other MS-related
characteristics. Additionally, we examined associations with sociodemographic
factors, work loss, comorbidities, HRQoL, and self-reported physical and
psychological impact of MS.

We identified all individuals with RRMS and SPMS registered in SMSreg who had
at least one complete fatigue measurement between 2012 and 2018 and were
either currently or previously undergoing DMT treatment. Individuals could
contribute one or multiple fatigue measurements, with the date of fatigue
questionnaire completion defined as baseline. For a sub-analysis of work loss, we
restricted the sample to individuals aged 18-64 years.

3.4.1.2 Assessment of fatigue

Fatigue was assessed using the FSMC total score and its cognitive and motor
subscales. For descriptive purposes, FSMC scores were categorized as no/low,
mild, moderate, or severe fatigue based on the predefined scale thresholds.58 In
the main analyses, the FSMC total score and subscales were treated as
continuous variables. A mean total FSMC score difference of ≥10 was considered
clinically meaningful.126

3.4.1.3 Potential predictors of fatigue

Potential predictors of fatigue were defined in relation to each FSMC

measurement, meaning that each FSMC measurement had its own set of
assessment covariates. These covariates, along with their data sources, are listed
in Table 3.1, and the corresponding time frames are illustrated in Figure 3.5. To
minimize obvious overlap between the FSMC and the MSIS-29, we excluded item
23 from the psychological component of the MSIS-29, "Feeling mentally
fatigued?".

34
Table 3.1. Covariates in statistical models and descriptive analyses in Studies I-II and IV,
with data sources and missing data proportions at baseline.
Covariates Data source Study Missing % (Study)
Socio-demographic
Age Population Register I-II, IV 0%
Sex Population Register I-II, IV 0%
Country of birth Population Register I-II, IV 0%
Region of residence Population Register I-II, IV 0%
Highest education achieved LISA I-II, IV 0%
Employment status LISA IV 0%
Work loss
Sick leave MiDAS I-II, IV 0%
Disability pension or activity MiDAS I-II, IV 0%
compensation
Comorbidities and health care
consumption
Days hospitalized last 5 years Inpatient NPR I, IV 0%
Depression In- or outpatient NPR I-II, IV 0%
Anxiety In- or outpatient NPR I-II, IV 0%
Other psychiatric comorbidities In- or outpatient NPR I-II, IV 0%
Sleep disorders In- or outpatient NPR IV 0%
MACE In- or outpatient NPR I, IV 0%
Arrhythmia In- or outpatient NPR I, IV 0%
Invasive cancer The Cancer Registry I, IV 0%
Hospitalized infection Inpatient NPR I, IV 0%
Charlson Comorbidity Index In- or outpatient NPR, II 0%
PDR
Treatment dispensations
Antidepressant use PDR I-II, IV 0%
Anxiolytics treatment use PDR I-II, IV 0%
Antipsychotic treatment use PDR I 0%
Sleeping aids treatment use PDR I-II, IV 0%
Pain treatment use PDR II, IV 0%
Antidiabetic use PDR I, IV 0%
Central stimulant use PDR I-II, IV 0%
Amantadine use PDR I-II, IV 0%
Fatigue treatment order PDR IV 0%
MS-related
MS type SMSreg I, IV 0% (I), 2-11% (IV)
Years since MS diagnosis SMSreg I-II, IV 3% (I), 0% (II, IV)
Any relapse last year SMSreg I-II, IV 0% (I-II), 5-23% (IV)
Cerebral lesions SMSreg I-II 0% (I), 12-16% (II)
DMT SMSreg II, IV 0% (II), 5-23% (IV)
DMT order SMSreg I 0%
EDSS SMSreg I-II 36% (I), 19-22% (II)
SDMT SMSreg I-II 26% (I), 25-47% (II)
PROMS assessing the impact of
MS and HRQoL
MSIS-29 physical SMSreg I-II 11% (I), 24-44% (II)
MSIS-29 psychological SMSreg I-II 11% (I), 24-44% (II)
FSMC SMSreg I-II 0% (I), 83-87% (II)
MS symptoms inventory score SMSreg II 83-87%
EQ-VAS SMSreg I-II 28% (I), 39-51% (II)

35
Figure 3.5. General time frames for Studies I-II and Study IV. Time zero was defined
differently across the studies: as the date of fatigue questionnaire completion in Study I,
DMT start in Study II, and the index date in Study IV. The follow-up period varied, with no
follow-up in Study I due to its cross-sectional design. Covariates such as comorbidities
and healthcare consumption were collected from five years preceding time zero, while
other covariates were recorded at time zero or within the previous year.

3.4.1.4 Statistical analyses

For the main analysis, we used linear regression to identify potential predictors
of fatigue, modeling FSMC total, cognitive, and motor subscale scores as
functions of different covariates. Since participants could contribute multiple
measurements, we calculated robust standard errors (Huber-White) and 95%
confidence intervals (CIs). At the request of peer reviewers to limit the risk of
type I errors, we also evaluated Bonferroni-corrected significance levels of
0.0001, specifically for associations where we found clinically meaningful
differences across predictor levels.

To explore associations between fatigue and multiple covariates simultaneously,

we first performed univariate regression to assess crude associations between
FSMC scores and each covariate. In subsequent steps, we added groups of
covariates and conducted multivariable regression while adjusting for other
covariates. The groups of covariates included in each model are listed in Table
3.2. Since a sub-analysis of work loss was performed separately, work loss was
not included as a covariate in the main models.

We handled missing data using complete case analysis. For respondents with
missing data in the MSIS-29 psychological and physical scales, respondent-
specific mean imputation was applied if at least 50% of the items had been
completed.

36
Table 3.2. In Study I, linear regression was used to explore predictors of fatigue by
sequentially adding groups of covariates to the statistical models (Model 1 - Model 6).
Models Group of covariates
Model 1 Univariate
Model 2 Socio-demographic
Model 3 MS-related
Model 4 MSIS-29 psychological, MSIS-29 physical
and EQ-VAS
Model 5 Comorbidities, health care consumption
and treatment dispensations
Model 6 Fatigue treatment

3.4.2 Study II - Trajectories of fatigue after initiation of MS DMTs

3.4.2.1 Study design and study population

This was a nationwide, register-based cohort study in which we identified fatigue

trajectories and their associations with trajectories of disability in individuals
with RRMS. Additionally, we examined the associations between potential
predictors of fatigue and the identified fatigue trajectories. In addition to these
main analyses, we applied a sub-study, where we assessed the incidence of
depression, anxiety, and disability pension/activity compensation across the
identified fatigue trajectories. These findings were presented as a poster at an
international MS conference.127

We included all participants from the COMBAT-MS study who had received
either dimethyl fumarate, fingolimod, glatiramer acetate, interferons (interferon
beta-1a, peginterferon beta-1a, and interferon beta-1b), natalizumab, rituximab, or
teriflunomide as their first DMT, or dimethyl fumarate, fingolimod, natalizumab,
rituximab, or teriflunomide as their second DMT. Participants were analyzed as
two separate cohorts: the first DMT cohort and the DMT switch cohort.
Additional inclusion criteria included residing in Sweden for ≥5 years prior to MS
diagnosis and having at least three FSMC measurements recorded after DMT
start. Participants were followed from DMT start until emigration, death,
withdrawal of consent, or the end of follow-up (12 May 2022), whichever came
first.

In the sub-study, we created sub-cohorts within the first DMT cohort, including
only individuals at risk of the outcomes studied. Specifically, we included: 1)
individuals without a depression diagnosis or antidepressant use in the five years
prior to first DMT start (depression sub-cohort); 2) individuals without an anxiety

37
disorder diagnosis or anxiolytic use in the five years prior to first DMT start
(anxiety sub-cohort); and 3) individuals without disability pension or activity
compensation in the five years prior to first DMT start (work loss sub-cohort).

3.4.2.2 Outcome

The main outcome of the main study was the fatigue trajectories, with disability
trajectories as a secondary outcome. Fatigue was assessed using the FSMC total
score, along with its contributing cognitive and motor subscales. Disability was
assessed using the EDSS. In the sub-study, depression, anxiety, and disability
pension/activity compensation were considered as outcomes (and fatigue
trajectories considered as exposure). Depression and anxiety were defined
through diagnoses recorded in the in- or outpatient components of the NPR or
through dispensations of antidepressants or anxiolytics in the PDR. Disability
pension/activity compensation was defined using the MiDAS.

3.4.2.3 Potential predictors of fatigue trajectories

Potential predictors of fatigue trajectories were assessed at the start of the first
DMT or the second DMT start. These potential predictors, along with their
respective data sources, are listed in Table 3.1. The corresponding time frames
for each predictor are illustrated in Figure 3.5.

3.4.2.4 Statistical analyses

To identify individual trajectories of fatigue and disability in PwMS, group-based

trajectory modeling was employed, using censored normal models to link time
since first DMT start or second DMT start with the fatigue and disability
measures through polynomial relationships.117 The analysis followed a data-driven
approach as outlined by Song et al.,128 and model selection was based on the
bayesian information criterion, where the lowest bayesian information criterion
indicated the best model fit.128

Key steps included:

 Identifying the optimal number of trajectory groups using a quadratic form,

with the requirement that each trajectory group contained at least 10% of
participants.
 Determining the appropriate polynomial function order for each trajectory
group, allowing shapes up to a cubic form.

38
 Fitting the model using Maximum Likelihood Estimation to estimate the
model parameters and determine the most likely trajectory for each group.
 Assigning individuals to trajectory groups based on their posterior
membership probabilities, with the highest probability determining group
assignment.
 Evaluating model adequacy using the average posterior probability of group
membership. Models were considered adequate if average posterior
probability values exceeded the recommended threshold of ≥0.70.128

Moreover, we used joint trajectory modeling within the group-based trajectory

modeling method to analyze connections between trajectories of FSMC and
EDSS, and reported conditional probabilities across the outcomes. We used
multinomial logistic regression to identify potential predictors of fatigue
trajectories, modeling FSMC total, cognitive, and motor trajectories as a function
of covariates. Each predictor was first analyzed separately and then
progressively included in a full model. We addressed missing baseline covariate
values with multiple imputation using chained equations. In the sub-study,
across fatigue trajectories, we assessed Kaplan-Meier survival estimates and
95% confidence intervals for depression, anxiety and work loss.

3.4.3 Study III - HIRT in people with MS experiencing fatigue

3.4.3.1 Study design and study population

This was a two-armed RCT evaluating the effects of high-intensity resistance

training (HIRT) on fatigue, mood, participation in everyday occupations, and
HRQoL in fatigued PwMS by comparing HIRT twice a week to once a week. The
choice of control group was based on a proof-of-concept approach,
hypothesizing that HIRT twice a week is superior to HIRT once a week in reducing
fatigue and related factors in fatigued PwMS. Additionally, as an exploratory
analysis, we compared changes in fatigue in the entire HIRT group to a
contemporary, non-randomized cohort serving as a non-intervention control
group.

Study participants were recruited from an MS clinic in Stockholm, Sweden

(Academic Specialist Center, the Center of Neurology) in August 2020 and
January 2021. The non-intervention control group consisted of participants from
the Combat-MS study followed at the same clinic. See Table 3.3 for eligibility
criteria.

39
Table 3.3. Eligibility criteria for Study III.
Inclusion criteria Exclusion criteria
HIRT groups  Age ≥ 18 years  Practicing high-intensity
 Diagnosis of MS training within the past 6
 FSMC score ≥ 53 (moderate months
fatigue)  Comorbidity interfering
 Ability to understand and with the ability to engage in
communicate in Swedish HIRT or evaluate the
endpoints
 Pregnancy or breastfeeding
Non-intervention  Study participant of the  Exclusion criteria’s
control group COMBAT-MS study according to the COMBAT-
 FSMC score ≥ 53 (moderate MS study
fatigue), assessed in Sept-
Oct 2020 and re-assessed
with the FSMC within 8-16
weeks

3.4.3.2 Intervention

The HIRT program followed current guidelines for resistance training.129

Participants underwent 60-minute HIRT sessions twice a week or once a week
for 12 weeks at Karolinska University Hospital, Stockholm, Sweden, under the
supervision of a physiotherapist. Each session began with a 5-10 minute warm-
up on a stationary bicycle, followed by four upper body exercises: lat pulldown,
push-up, chest press, and latissimus pull. Three lower body exercises were
included: leg extension, leg curl, and leg press. Additionally, one whole-body
exercise, the plank position, was performed. The session concluded with a 5-10
minute cool-down incorporating stretching exercises.

A progressive resistance training model was applied. At the start of the

intervention, 1 repetition maximum (1RM) for each machine exercise was
estimated using a 10RM test, which determined the maximum load a participant
could lift for 10 consecutive repetitions but not 11, corresponding to 70% of their
1RM. For the plank position, the maximum time held was considered the 1RM for
that exercise.

40
3.4.3.3 Endpoints

The primary endpoint was the change in fatigue, measured using the FSMC total,
with a 10-point change considered clinically meaningful.126 Secondary endpoints
were assessed for the HIRT groups only and included changes in fatigue
measured with the FSS, as well as assessments using the Hospital Anxiety and
Depression Scale (HADS),130 the Occupational Gaps Questionnaire,131 the MSIS-29,
and the 5-level EQ-5D version, including the EQ-VAS.132 As an exploratory
endpoint, changes in inflammatory protein markers in plasma were measured
using a highly sensitive multiplex technique.133

3.4.3.4 Procedures

The timeline of procedures for the HIRT groups and the non-intervention control
group is illustrated in Figure 3.6. Potentially eligible individuals for HIRT first
underwent a preliminary screening for eligibility. Those who met the initial criteria
were invited for a detailed screening, followed by baseline assessments if
eligible. After baseline assessments, an independent nurse performed the
randomization using a computer-generated method (Sealed Envelope Ltd).
Following the 12-week intervention, study participants returned for a follow-up
assessment. Participants in the non-intervention control group, who were
assessed with FSMC in September–October 2020, were re-assessed with the
FSMC within 8 to 16 weeks.

Figure 3.6. Timeline of procedures for the HIRT groups and the non-intervention control
group. Participants were assessed at baseline, with follow-up assessments conducted
after the 12-week intervention period for the HIRT groups and between 8 to 16 weeks
after baseline for the non-intervention control group.

41
3.4.3.5 Statistical analyses

An intention-to-treat approach was used for data analysis. The endpoints were
assessed after the 12-week intervention period for the HIRT groups and after 8 to
16 weeks for the non-intervention control group. Generalized linear models with
repeated measures were used to evaluate within-group and between-group
effects for primary and secondary endpoints. Changes in immune-related
protein levels were analyzed using paired samples t-tests. Associations between
changes in FSMC and immune-related protein levels were analyzed with
multivariable linear regression models, adjusting for age, sex, and the number of
days between the last HIRT session and follow-up sampling.

3.4.4 Study IV - Symptomatic drug treatment for fatigue in MS

3.4.4.1 Study design and study population

This was a nationwide, register-based cohort study in which we described the

utilization of central stimulants and amantadine as symptomatic treatments for
fatigue ("fatigue treatments") and evaluated their comparative effectiveness.
Specifically, we addressed the following research questions: i) the annual
prescription rate of fatigue treatments, ii) prescription refills for fatigue
treatments within two years of initiation, iii) characteristics of new users and
non-users of fatigue treatments, and iv) the impact of initiating fatigue
treatments on monthly work loss rates, both across different treatments and
when comparing modafinil to untreated PwMS.

Figure 3.7 presents the study flowchart and delineated cohorts. We included
PwMS identified through SMSreg or by at least three separate MS diagnoses
(ICD-10 G35.9) recorded in the NPR. To assess the annual prescription rate of
fatigue treatments between 2006 and 2023 in this population, annual cohorts
consisted of individuals who were alive and residing in Sweden between January
1 and December 31 of each study year, with an MS diagnosis recorded during or
before that year. For all additional analyses, we included PwMS diagnosed on or
after January 1, 2007, ensuring at least one year of follow-up since the launch of
the PDR in July 2005.

42
Figure 3.7. Study flowchart illustrating the inclusion, follow-up and analysis of study IV.

3.4.4.2 Exposure

Fatigue treatment served as the exposure, with an untreated MS cohort as a

comparator group. Fatigue treatments, including modafinil, amantadine, and
ADHD drugs of the central stimulant type, were identified in the PDR based on
filled prescriptions, as listed in Table 3.4. To enhance the specificity of the
exposure, prescriptions issued by physicians specializing in child and adolescent
psychiatry, general psychiatry, or forensic psychiatry were excluded. Individuals
could contribute to multiple treatment cohorts if they switched to another
fatigue treatment. In the analysis of the annual prescription rate of fatigue
treatments, exposure was defined as having at least one filled prescription for a
specific fatigue treatment within each study year. For all additional analyses, new
users of fatigue treatments were identified based on the first filled prescription
for each fatigue treatment (i.e., index date) between January 1, 2007, and
November 23, 2021. Risk-set sampling was used to match first-time modafinil
users with up to five untreated controls at the index date.

43
Table 3.4. Fatigue treatment cohorts and included treatments
Fatigue treatment Drugs (ATC code)
Modafinil Modafinil (N06BA07)
Amantadine Amantadine (N04BB01)
ADHD drugs Amfetamine (N06BA01), dexamfetamine (N06BA02),
methylphenidate (N06BA04), and lisdexamfetamine (N06BA12)

Figure 3.8 illustrates the identification of fatigue treatment cohorts and the risk-
set sampling process. New treatment use was determined by the first filled
prescription for each fatigue treatment, defining the index date (e.g., PwMS 1, who
contributed to both the modafinil and ADHD drug cohorts). Risk-set sampling
was used to match first-time modafinil users with up to five untreated controls
based on sex, age (±1 year), years since MS diagnosis (±1 year), and prior-year
work loss (yes/no). Matched controls had no prior use of modafinil, amantadine,
or ADHD drugs (e.g., PwMS 8 and 11). Individuals could serve as controls for
multiple modafinil users (illustrated by PwMS 2) and could not initiate fatigue
treatments within the 24-month follow-up period (illustrated by PwMS 6 and 8).
Observations were censored at death, emigration, or the end of follow-up
(November 23, 2023), whichever occurred first (e.g., PwMS 3, 4, and 9, with PwMS
4 and 9 not included in the analysis due to lost to follow-up before 24 months).
An ever-treated approach was applied to the treatment cohorts, following
individuals from treatment initiation for 24 months, regardless of discontinuation
or switching.

Figure 3.8. Timelines for identifying treated and untreated cohorts with illustrative
examples.

44
3.4.4.3 Outcome

The primary outcome of this study was work loss, measured using data from
MiDAS. Work loss was defined as a combined measure encompassing days of
sick leave, disability pension, and activity compensation.

3.4.4.4 Covariates

Covariates were assessed at the index date. A detailed list of covariates and
their respective data sources is provided in Table 3.1, while the corresponding
time frames are illustrated in Figure 3.5.

3.4.4.5 Statistical analyses

The annual prescription rate of modafinil, ADHD drugs, and amantadine (2006-
2023) was calculated as the number of PwMS who filled at least one prescription
divided by the at-risk MS population each year.

An interrupted time series analysis with generalized estimating equations was

used to assess the impact of fatigue treatments on work loss, using data from 12
months before to 24 months after treatment initiation. Changes in monthly work
loss rates were compared within and between cohorts, with modafinil as the
reference for comparisons between fatigue treatments, and the untreated
cohort as the reference for comparison between modafinil users and untreated
individuals. In interrupted time series analysis, it is standard to assess the
immediate change in the outcome due to the intervention (a step function).
However, this was not done in our study because we hypothesized that the
impact of fatigue treatments on work loss is not immediate. Robust standard
errors (Huber-White) were applied to account for the weighting and the fact that
individuals could contribute to more than one cohort. Missing baseline
covariates were handled with multiple imputation using chained equations.
Stabilized inverse probability of treatment weighting was used to adjust for
cohort imbalances, and standardized mean differences were assessed for
balance. Region of residence, days of work loss, years since MS diagnosis, relapse
history, and use of DMTs, specifically interferons, showed standardized mean
differences outside the range of -0.1 to 0.1.

45
3.5 Ethical considerations
In this doctoral thesis, we conducted three register-based studies (Studies I-II,
IV) and one RCT (Study III), each with distinct ethical considerations. A
fundamental aspect of ethical considerations is weighing the potential risks to
study participants against the expected and potential benefits for those to
whom the results may apply.

The register-based studies are based on data linkage from various national
health and population registers, as well as the SMSreg, to identify a cohort of
PwMS. Many of these national registers contain data on all individuals living in
Sweden, including all PwMS in Sweden within our MS cohort. However, many
individuals may be unaware that their data is registered in these databases. In
contrast, the SMSreg, as a quality register, allows patients to opt out, ensuring
that all their data is removed. Given the high prevalence of fatigue among PwMS,
findings from these register-based studies have the potential to benefit a high
proportion of PwMS in Sweden.

Compared to clinical trials, register-based research involves minimal physical

risk, making the need for obtaining informed consent less necessary. The
Swedish Ethical Review Authority has the authority to waive the demand of
informed consent, particularly when obtaining consent would be extremely
difficult or not possible in practice. In our register-based studies, which were
based on our MS cohort linking data from over a hundred thousand individuals,
both living and deceased, obtaining consent from each person would not only be
nearly impossible but could also introduce bias due to potential non-responses.

Although participants are unaware of their involvement in register-based

research, there is a potential risk of harm in the event of a security breach. To
protect participant privacy, several safeguards are in place. Data transmitted
from registry holders to the university undergoes pseudonymization, where
personal identification numbers are replaced with serial numbers to prevent
individual identification. Additionally, data is stored on secure, encrypted servers
with restricted access. Findings are reported exclusively in aggregated form to
further minimize the risk of identifying individuals.

For Study III, informed consent was required and obtained from all participants
due to the trial design. The study protocol concluded that the training
intervention and other study components posed no more than minimal risk.
These minimal risks included potential discomfort from training, such as muscle

46
soreness or joint pain. Some health-related questions could have been
perceived as intrusive, and blood sample collection might have caused minor
pain. Another potential burden was the time commitment required for
participation. To mitigate the risk of severe side effects related to medical
conditions that could pose a danger during training, a medical assessment was
conducted before randomization. Furthermore, all training sessions took place at
Karolinska University Hospital under the supervision of certified physiotherapist,
ensuring participant safety. To minimize the time commitment for participants,
data collection was often scheduled alongside neurology appointments or
regular DMT infusions. Finally, the potential benefits of the study extended to all
participants, as both the treatment and comparator groups received training,
though at different frequencies (twice per week versus once per week).

Ethical approval for Study I (2017/700-31/4, with amendment 2020-02892),

Studies II and IV (2021-02384, with amendment 2023-07906-02), and Study III
(2019-05105, with amendment 2020-05614) has been granted by the Swedish
Ethical Review Authority.

47
4 Results
4.1 Study I - Predictors of fatigue in MS

4.1.1 Baseline characteristics

We identified 3,179 FSMC measurements in 2,165 PwMS, with a mean age (SD) of
41.6 (10.2) years, a mean time since MS diagnosis (SD) of 10.6 (7.8) years, a mean
EDSS (SD) of 1.9 (1.5), and a mean FSMC total score (SD) of 50.7 (22.2). 70.2% of
the FSMC measurements belonged to females. Of the FSMC total
measurements, 1,309 (40.7%) were categorized as no/low, 407 (12.7%) as mild,
449 (14.0%) as moderate, and 1,048 (32%) as severe fatigue. A gradual trend was
observed for most characteristics across the fatigue severity categories.
Specifically, being in the severe fatigue group was associated with older age,
female sex, lower education, increased sick leave, disability pension/activity
compensation, a higher prevalence of psychiatric comorbidities, and the use of
antidepressants, anxiolytics, sleeping pills, and symptomatic drug treatments for
fatigue. Furthermore, higher levels of fatigue were associated with having SPMS,
higher EDSS, higher MSIS-29 scores, lower SDMT scores, and lower EQ-VAS
scores. The proportion of observations with missing baseline data for each
variable is presented in Table 3.1.

4.1.2 Potential predictors for fatigue

After adjusting for age, sex, region of residence, educational level, country of
birth, MS type, DMT order, years since MS diagnosis, relapse history, and history
of cerebral lesions, the only MS-related characteristics associated with clinically
meaningful differences across the different predictor levels (according to the 10-
point threshold) were EDSS and SDMT. Self-reported impact of MS and HRQoL,
assessed using MSIS-29 and EQ-VAS, demonstrated even greater differences
across the levels of these predictors (Figure 4.1).

We observed similar trends for the cognitive and motor subscales of FSMC.
However, one exception was EDSS. Patients with severe MS disability (EDSS ≥ 6)
reported a mean difference (MD) of 6.9 (95% CI: 4.3, 9.4) in FSMC cognitive
scores compared to those with mild disability (EDSS 0-2.5), while the
corresponding MD for FSMC motor scores was 10.7 (95% CI: 8.5, 12.9).

In addition to MS-related characteristics, self-reported impact of MS, and

HRQoL, several other covariates were found to predict fatigue. Characteristics

49
associated with clinically meaningful differences across predictor levels included
sick leave, disability pension/activity compensation, psychiatric comorbidities,
and treatment dispensations for depression, anxiety disorders, and other
psychiatric conditions. The use of sleeping aids and symptomatic drug
treatments for fatigue was also linked to higher fatigue severity.

Figure 4.1. Associations with FSMC total.

By sequentially adding groups of covariates to the linear regression model, we

observed significant differences in how these covariates influenced each other’s
associations with fatigue. When we further adjusted for MSIS-29 and EQ-VAS, all
previously significant covariates fell below the threshold for a clinically
meaningful difference. The only covariates that remained significantly associated
with clinically meaningful differences across predictor levels were MSIS-29 and
EQ-VAS. Adding comorbidities, healthcare consumption, treatment

50
dispensations, and fatigue treatments did not meaningfully alter these
associations.

4.2 Study II - Trajectories of fatigue after initiation of MS DMTs

We identified 1,587 PwMS who were followed for a mean (SD) of 7.1 (2.2) years
from the initiation of their first DMT (first DMT cohort) and 1,818 PwMS who were
followed for a mean (SD) of 7.3 (1.9) years from the initiation of their second DMT
(DMT switch cohort). The mean (SD) time from MS diagnosis to the start of the
first DMT was 1.0 (3.3) years, while the mean time from MS diagnosis to the
initiation of the second DMT was 6.3 (5.8) years. The proportion of observations
with missing baseline data for each variable is presented in Table 3.1.

4.2.1 Trajectories of fatigue and disability

Using the group-based trajectory modeling algorithm, we identified five fatigue

trajectories in the first DMT cohort and six fatigue trajectories in the DMT switch
cohort (Figure 4.2). Additionally, three physical disability trajectories were
identified in both cohorts. All fatigue trajectories showed little to no change in
fatigue severity over time, except for 20.1% of the first DMT cohort with
moderate fatigue at the start of DMT. This group exhibited an 11.5-point increase
in the FSMC total score from DMT initiation to the end of the follow-up period.

Regarding disability, 26.5% of PwMS in the first DMT cohort maintained no

disability, 57.0% maintained minimal disability, and 16.5% progressed from
moderate to severe disability after 11 years of follow-up. In the DMT switch
cohort, similar disability trajectories were observed, though individuals in this
cohort had overall higher EDSS values at baseline. In this cohort, 21.3%
maintained no disability, 59.0% maintained minimal disability, and 19.7%
progressed from moderate to severe disability over the 11-year follow-up.
According to the standard definition of confirmed disability worsening, defined
as a ≥1.5-point increase in EDSS from a baseline EDSS of 0, a ≥1.0-point increase
from a baseline EDSS of 1.0-5.0, or a ≥0.5-point increase from a baseline EDSS of
>5.5, the increase in the moderate disability trajectory indicated a clinically
significant change from four years after the first DMT start and from eight years
after the second DMT start.111

51
Figure 4.2. Proportion of study participants (%), mean FSMC total scores at DMT
initiation (MSV), and change in mean scores from DMT initiation to the end of follow-up
(MD) in the first DMT cohort and the DMT switch cohort.

52
4.2.2 Membership across fatigue and disability trajectories.

A strong association was observed between fatigue and disability trajectories in

both the first DMT and DMT switch cohorts. In the first DMT cohort, individuals in
the severe fatigue trajectory group had a 48.5% probability of being classified in
the moderate disability trajectory. This probability decreased stepwise across
the fatigue severity levels, with participants in the moderate (23.8%), mild
(12.5%), low (3.2%), and no (1.2%) fatigue trajectories being less likely to belong to
the moderate physical disability trajectory.

4.2.3 Baseline characteristics

Similar to the baseline characteristics observed in Study I, a gradual trend was

noted for most potential predictors across the fatigue trajectory groups.

4.2.4 Predictors of fatigue trajectories

Aside from baseline fatigue, only a few covariates independently predicted

membership in the higher stable or increasing fatigue trajectories in the full
model. These covariates included female sex, sick leave, depression diagnosis,
pain treatment use, teriflunomide, MSIS-29 psychological score, and MS
symptoms inventory scores.

4.2.5 Linking fatigue trajectories to incidence of psychiatric comorbidities and

work loss

Kaplan-Meier survival curves demonstrate differences in the incidence of

depression (Figure 4.3a) and anxiety (Figure 4.3b) in the years following DMT
initiation across the fatigue trajectories. Depression was observed in 6.6% of
individuals in the no fatigue trajectory group, with the incidence increasing
stepwise to 13.1% in the low, 18.1% in the mild, 31.0% in the moderate, and 41.6% in
the severe fatigue group. The corresponding incidence of anxiety was 9.8% in the
no, 18.5% in the low, 25.7% in the mild, 40.0% in the moderate, and 48.4% in the
severe fatigue group.

Kaplan-Meier survival curves in Figure 4.4 illustrate significant differences in the

incidence of disability pension/activity compensation in the years following DMT
initiation across the fatigue trajectories. A notable proportion of individuals in the
moderate and severe fatigue trajectory groups (15.6% and 24.2%, respectively)
received a disability pension/activity compensation, compared to only 0.4% to
4.3% in the no, low, and mild fatigue groups.

53
 (a) Depression

(b) Anxiety

Figure 4.3. Kaplan–Meier survival curves with 95% pointwise confidence intervals and
numbers at risk for depression (top) and anxiety (bottom) in the first DMT cohort,
stratified by FSMC total score trajectories.

54
 Proportion with no work loss

Figure 4.4. Kaplan–Meier survival curves, including 95% pointwise confidence intervals
and number at risk, for disability pension/activity compensation (work loss) in the first
DMT cohort, stratified by FSMC total score trajectories.

4.3 Study III - HIRT in people with MS experiencing fatigue

We screened 144 PwMS for eligibility, of whom 71 were enrolled in the study.
Thirty-six participants were allocated to HIRT once a week, and 35 to HIRT twice
a week. Additionally, 69 PwMS were assigned to the non-intervention control
group. A total of 11 participants (15.5%) in the HIRT groups did not complete the
intervention, either before it began or during the 12-week intervention period.
However, following the intention-to-treat approach, all participants were
analyzed according to their allocated HIRT group, except for three individuals
with missing follow-up data, resulting in 34 participants in each HIRT group
included in the analysis. Session adherence, defined as completing at least 75%
of the stipulated HIRT sessions, was achieved by 26 participants (74%) in the
once-a-week group and 24 participants (67%) in the twice-a-week group.
Content adherence, defined as a mean percentage increase in muscle strength
of ≥10%, was observed in 52 participants (73%).

4.3.1 Baseline characteristics

There were some differences in baseline characteristics between the two HIRT
groups. Participants allocated to HIRT once a week were older (mean age 44 vs.
41), had a higher proportion of women (94% vs. 80%), had a longer time since

55
diagnosis (mean 9.4 vs. 7.7 years), had a higher median EDSS (2.3 vs. 2.0), and
reported more physical activity in daily life (64% achieving ≥150 activity minutes
per week vs. 51%). Baseline values of primary and secondary outcomes also
differed between the groups in HADS anxiety, Occupational Gaps Questionnaire,
and MSIS-29 physical and psychological scores (Table 4.1). When comparing the
overall HIRT groups to the non-intervention control group, the control group had
a lower proportion of women (77% vs. 87%) and a higher median EDSS (2.5 vs.
2.0). Additionally, baseline FSMC total scores were lower in the non-intervention
group (Table 4.1).

4.3.2 Effects of HIRT

The effects of HIRT are summarized in Table 4.1. We found no significant

difference between the two HIRT groups in the primary endpoint, as the mean
FSMC change score decreased by around 10 points in both groups, indicating a
clinically relevant improvement. At the individual level, 28 participants (15 from
the HIRT twice a week group and 13 from HIRT once a week group) experienced a
reduction of ≥10 points in their FSMC score. For the secondary endpoints, HADS
anxiety and MSIS-29 psychological scores demonstrated a significant between-
group difference, favoring HIRT twice a week. A significant time effect was
observed for all measures, except for Occupational Gaps Questionnaire. The
reduction in FSMC scores in the combined HIRT groups was greater compared to
the non-intervention group, which showed little to no change between baseline
and follow-up.

Results from the exploratory endpoint, changes in plasma inflammatory protein

levels, are reported for the combined HIRT groups. Baseline and follow-up
samples were obtained from 63 out of 68 participants, as five were unable to
attend the follow-up in person. The median time between the last HIRT session
and follow-up blood sampling was 5 days (interquartile range: 2-10 days). Using
a false discovery rate cut-off of 5%, a significant upregulation was observed in 18
out of 51 proteins. Surprisingly, multivariable linear regression analyses showed
that increased levels of tumour necrosis factor and interleukin-17A were
significantly associated with improvements in FSMC.

56
57
4.4 Study IV - Symptomatic drug treatment for fatigue in MS

4.4.1 Annual fatigue treatment prescription rate

We identified 29,257 individuals diagnosed with MS between 2001 and 2023 who
were eligible for analyses of the yearly prevalence of fatigue treatments from
2006 to 2023, as illustrated in Figure 4.5. Modafinil was the most commonly
prescribed treatment throughout the study period, while amantadine and ADHD
drugs were used less frequently. Over time, the use of modafinil and amantadine
declined, whereas prescriptions for ADHD drugs increased. Overall, the
proportion of PwMS receiving any fatigue treatment remained relatively stable,
decreasing from 11.3% in 2006 to 10.2% in 2023.

Figure 4.5. Annual period prevalence of modafinil, amantadine, and ADHD drug use
among PwMS from 2006 to 2023.

4.4.2 Characteristics at index and rate of re-filled prescriptions

Among the 29,257 PwMS diagnosed between 2001 and 2023, we included 14,464
individuals with an MS diagnosis as of January 1, 2007, who had no prior fatigue
treatments. These individuals were analyzed to assess prescription refills,
characteristics at index, and the impact of initiating fatigue treatments on
monthly work loss rates. After further exclusions (Figure 3.7), the final analysis

58
included 2,162 new modafinil users, 462 new amantadine users, 424 new ADHD
drug users, and 9,762 untreated individuals. At index, the mean (SD) age was 41.3
(10.0) years, 72.0% were female, 86.7% had RRMS, and the mean (SD) time since
MS diagnosis was 2.8 (2.6) years. Age, MS type, and sex distribution were similar
across cohorts. Unemployment rates were evenly distributed, but the mean
number of work loss days varied from 109 to 158 in the run-in year to index.
Depression, anxiety, and related treatments were most prevalent in the ADHD
drug cohort and least common in the untreated cohort. Pain treatments were
frequent across all cohorts but less common in the untreated group. Notably,
95.1% of new modafinil users had no prior fatigue treatment, compared to 48.1%
of new amantadine users and 11.1% of new ADHD drug users. ADHD drug users
had the longest MS duration (4.4 years) and the fewest relapses (7.7%). Among
new users of modafinil and ADHD drugs, 27.8% and 15.1%, respectively, did not
refill a prescription within the first two years, while the corresponding proportion
for amantadine was 49.6%. The proportion of observations with missing data for
each variable is presented in Table 3.1.

4.4.3 Effectiveness of fatigue treatment

Figure 4.6 illustrates the weighted mean monthly work loss from 12 months
before to 24 months after the index date, while Table 4.2 presents the
corresponding estimated mean monthly work loss from the generalized
estimating equations model. All fatigue treatment cohorts showed similar
trajectories, with a gradual increase in mean monthly work loss during the year
before the index date (ranging from 0.18 to 0.29 days per month), followed by
average work loss levels remaining unchanged over the next two years. This
trend was also observed, though less pronounced, in the untreated reference
cohort.

The change in the trajectory of mean monthly work loss from the pre- to post-
index period was significantly greater in the modafinil cohort compared to the
untreated cohort, with a mean difference of -0.17 days (95% CI: -0.22 to -0.12).
Among the treated cohorts, no statistically significant differences were observed
in the change in work loss trajectories between the amantadine and ADHD drug
cohorts relative to the modafinil cohort. However, the point estimate of a 0.09
days per month difference in change favored modafinil over ADHD drugs.

59
Figure 4.6. Weighted mean monthly days of work loss from 12 months before to 24
months after the index date are shown for the fatigue treatment cohorts and the
untreated modafinil-matched MS cohort. Weights from stabilized inverse probability of
treatment weighting were applied to adjust for imbalances between cohorts at the
index date.

60
Table 4.2. Estimated weighted monthly days of work loss from 12 months before to 24
months after the index for the fatigue treatment cohorts and the untreated modafinil-
matched MS cohort. A positive number correspond to a net loss of work capacity.
Mean Monthly Slope of Monthly Work Loss,
Work Loss Mean (95% CI)
(95% CI)
Pre-index Pre-index Pre- to post-index
change
Cohort
Untreated 9.76 (9.51; 10.01) 0.07 (0.05; 0.09) -0.10 (-0.12; -0.08)
Modafinil 11.15 (10.59; 11.70) 0.29 (0.24; 0.33) -0.27 (-0.32; -0.22)
Amantadine 12.33 (10.95; 13.71) 0.26 (0.15; 0.36) -0.23 (-0.36; -0.10)
ADHD drugs 11.78 (10.32; 13.23) 0.18 (0.08; 0.29) -0.18 (-0.31; -0.05)

Comparison
between cohorts
Modafinil vs. 1.38 (0.78-1.99) 0.21 (0.17; 0.26) -0.17 (-0.22; -0.12)
Untreated
Amantadine vs. 1.18 (-0.31; 2.67) -0.03 (-0.14; 0.09) 0.04 (-0.10; 0.18)
Modafinil
ADHD drugs vs. 0.63 (-0.93; 2.19) -0.10 (-0.22; 0.01) 0.09 (-0.05; 0.24)
Modafinil

61
5 Discussion
5.1 Findings

5.1.1 Study I - Predictors of fatigue in MS

In Study I, we found that PwMS with moderate and severe disability, measured
with EDSS, had statistically significant and clinically relevant higher fatigue scores
compared to those with mild disability. We also confirmed the well-established
link between fatigue and depression in MS. However, due to the cross-sectional
design, no conclusions can be drawn regarding causality. Only the lowest quartile
of SDMT scores (slowest information processing speed), compared to the
highest quartile (fastest information processing speed), showed a clinically and
statistically significant difference in fatigue. Even larger differences were
observed for MSIS-29 and EQ-VAS. However, other MS-related characteristics,
including MS type (RRMS, SPMS), disease duration, and history of relapses, did
not show clinically relevant associations with fatigue. Additionally, fatigue
severity was associated with sick leave, disability pension/activity
compensation, psychiatric comorbidities, and treatments for depression,
anxiety, and fatigue-related symptoms.

While various studies have identified associations between the potential

underlying pathophysiology of MS and fatigue,27 there remains inconsistency in
the literature regarding the relationship between fatigue and MS-related
variables, including MS type, disease duration, and disability measures such as
the EDSS.

Previous studies examining the association between disability and fatigue in MS

have shown discordant results.25,65-69 Comparing findings across studies is
challenging due to variations in fatigue and disability assessments. Furthermore,
large population-based studies that include both disability and fatigue data are
rare. Many studies have had to exclude a significant portion of the source
population due to missing fatigue measures,25,66,69 increasing the risk of selection
bias and the risk of comparing populations that are not representative of the
broader MS population. A Swedish longitudinal study found an association
between baseline EDSS and increasing fatigue only among PwMS who had ≥10
years since diagnosis or a progressive disease course.67 In contrast, we found a
significant association between EDSS and fatigue even after adjusting for other
MS-related characteristics, such as disease duration and MS type. This

63
discrepancy may stem from differences in study population characteristics or
variations in the fatigue measures used.

Consistent with several other studies, we found a progressive disease course to

be associated with worse fatigue in unadjusted analyses, but not after adjusting
for other MS-related characteristics, such as disability and disease duration.25,66
However, it is important to note that only 5% of our study population in Study I
had SPMS, and individuals with PPMS were not included, meaning that the
findings are primarily generalizable to those with RRMS. In contrast, a MS
register-based study from United Kingdom69 found an association between
fatigue and MS type even after adjusting for other MS-related variables. This
study included a much larger proportion of individuals with a progressive
disease course (37.4%). However, it also had limitations, as it only included 775
PwMS out of 15,985 eligible patients due to missing fatigue data. This
underscores the potential impact of selection bias in research on fatigue in MS.

When comparing the highest (fastest) and lowest (slowest) quartiles of SDMT,
we found a clinically relevant difference in fatigue. However, this difference was
not observed across the other quartiles of SDMT, suggesting that fatigue may be
more strongly correlated with cognitive impairment than with general variability
in processing speed. PwMS often report that fatigue impacts their cognitive
performance, but the evidence linking fatigue with cognitive impairment remains
limited.8 This relationship is further complicated by the overlap in measurement
tools, as most fatigue scales include a cognitive fatigue component that
addresses the impact of fatigue on cognition.73 However, this issue did not seem
to affect Study I. When analyzing the two FSMC subscales separately, the
associations with SDMT were similar for both the motor and cognitive subscales.
This suggests that both motor and cognitive fatigue may impact processing
speed (or vice versa), rather than the observed relationship being an artifact of
measurement overlap between FSMC and SDMT. This finding is supported by
Penner and colleagues, who reported similar correlation coefficients for cognitive
and motor FSMC with SDMT (Pearson correlation: -0.33 and -0.34,
respectively).58 Furthermore, they found a stronger correlation between the self-
reported component of the Multiple Sclerosis Neuropsychological Questionnaire
and cognitive FSMC (Pearson correlation: 0.61) compared to motor FSMC
(Pearson correlation: 0.39), potentially suggesting that the self-reported
components of both scales may be an important factor contributing to the
overlap.58

64
The strong association between fatigue and MSIS-29 led us to speculate
whether these two self-reported scales actually measure similar concepts.
Although we excluded item 23 from the psychological component of MSIS-29
(“Feeling mentally fatigued?”), the measures remained highly correlated in our
analysis. Upon examining the actual items in FSMC and MSIS-29, several items
overlap. MSIS-29 addresses issues that could be attributed to fatigue, including
more general concerns like “Having to depend on others to do things for you?”
and “Feeling unwell?”, as well as questions more directly linked to fatigue, such as
“Heavy arms and/or legs?” and “Problems concentrating?”. A study from the
United Kingdom using an online survey examined the relationship between
fatigue severity (FSS), fatigue impact (MFIS), and MS-related characteristics.134
Interestingly, they found stronger correlations between fatigue impact and both
the physical (Pearson correlation = 0.660) and psychological (Pearson
correlation = 0.721) components of MSIS-29 compared to the correlations
between fatigue severity and MSIS-29 physical (Pearson correlation = 0.547) and
psychological (Pearson correlation = 0.466). This suggests that the impact
component of MSIS-29 may explain part of the high association observed
between MSIS-29 and fatigue in our study as well, as FSMC focuses on the
impact of fatigue on PwMS.

Regarding the temporal relationship between MSIS-29, EQ-VAS, and FSMC in the
full model, we included MSIS-29 and EQ-VAS to test whether they confound the
relationship between MS-related characteristics and fatigue. However, according
to the literature, it is more likely that both MS-related characteristics and fatigue
reduce HRQoL and lead to higher scores on MSIS-29 (i.e., colliders).64

5.1.2 Study II - Trajectories of fatigue after initiation of MS DMTs

In Study II, we found that a high proportion of PwMS already experienced fatigue
at their first DMT start, and that fatigue levels remained largely unchanged during
follow-up. However, individuals with moderate fatigue at first DMT start
experienced a clinically relevant increase of 11.5 points in FSMC total. In contrast,
disability trajectories showed greater variability in EDSS over time. PwMS with
moderate disability at their first and second DMT start experienced a clinically
significant increase in EDSS after four and eight years, respectively. Fatigue and
disability trajectories were strongly associated. Furthermore, few variables
predicted fatigue trajectories. Lastly, we found that greater fatigue severity in
the years following the initiation of the first DMT was associated with a higher
incidence of psychiatric comorbidities and work loss.

65
5.1.2.1 Fatigue and EDSS trajectories

We found that fatigue levels remained relatively unchanged over time, consistent
with several previous studies with follow-up periods of 1-3 years, which reported
no significant changes in fatigue.70,74,76,79,80 However, a three-year follow-up may
be insufficient to detect meaningful changes. An American study of 944 RRMS
patients, with a median MS duration of two years at baseline, followed
participants for a median of eight years.78 Despite using different analytical
approaches and therefore not being directly comparable, their findings suggest a
higher proportion of individuals with a clinically relevant increase in fatigue
(measured with the fatigue performance scale) over time compared to our
results. Moreover, while our results indicated that fatigue worsening primarily
occurred in individuals with moderate fatigue at baseline, the American study
found that lower baseline fatigue levels were associated with increasing fatigue.
Several factors could explain these differences, but an important distinction lies
in how fatigue was conceptualized. The FSMC assesses multiple dimensions of
fatigue, whereas the fatigue performance scale is a single-item measure
capturing fatigue severity and impact over the past month. Fatigue performance
scale may be more sensitive to detecting immediate changes in fatigue due to
its shorter reference period, whereas FSMC instructs individuals to assess their
fatigue symptoms more generally.

Although we observed some variations in fatigue over time across fatigue

trajectory groups, visually, they ran relatively parallel over time. This phenomenon
has previously been discussed by Vachon et al.135 as the “Rainbow effect”. This
effect occurs when the analysis of repeated measures data shows several
parallel trajectories that may appear to represent distinct subgroups, but in
reality, these are not true subgroups with different trajectories. Instead, the
trajectories represent gradations along a continuum of values, suggesting that
differences in the development of fatigue severity in MS is continuous rather
than categorical.

Previous studies examining the association between longitudinal changes in

fatigue and disability are scarce.76,78 Consistent with the cross-sectional findings
from Study I, where we observed a clinically relevant association between
disability and fatigue, we found a strong association between fatigue and
disability trajectories also here. Conditional probabilities from joint trajectory
modeling showed that among individuals with moderate disability at first DMT
start, approximately 24% belonged to the moderate fatigue trajectory, while 49%

66
belonged to the severe fatigue trajectory group. In other words, among PwMS
with moderate disability at DMT start and clinically increasing disability over
time, about 24% transitioned from moderate to severe fatigue, while 49%
maintained stable severe fatigue levels.

5.1.2.2 Predictors of fatigue trajectories

In Study II, we opted for a more complex model as our main analysis compared
to Study I, which was more similar to the most complex model in Study I (Model
6). This model included MSIS-29 psychological, MSIS-29 physical, and EQ-VAS,
all of which were highly correlated with FSMC. Consequently, only a few variables
emerged as significant predictors of higher-stable or moderately increasing
fatigue trajectories in the first DMT cohort, including female sex, sick leave,
depression diagnosis, pain treatment use, teriflunomide, MSIS-29 psychological,
and MS-symptoms inventory scores. Given the more complex regression model
and the overall stability of fatigue trajectories, we found little evidence
supporting independent predictors of increasing or decreasing fatigue. Notably,
baseline EDSS was not associated with fatigue trajectories, which was likely
explained by the adjustment for MSIS-29. This raises a methodological
consideration regarding the role of MSIS-29 as a confounder in fatigue research.
If MSIS-29 acts as a mediator rather than a confounder, adjusting for it will
remove the impact of EDSS on fatigue trajectories that goes through MSIS-29,
capturing only the direct effect.

An interesting finding was that individuals prescribed teriflunomide as their first

DMT were more likely to belong to the higher-stable or moderately increasing
fatigue trajectories compared to those treated with rituximab. These results may
indicate a small true negative effect of teriflunomide but could also be due to
residual confounding. Future studies are needed to confirm these findings. There
is limited evidence from previous comparative effectiveness studies of DMTs on
fatigue.136-138 A recent systematic review of 130 RCTs on DMTs for MS identified
MS-related fatigue as an outcome in only seven studies, with only two
demonstrating statistically significant results.139 Additionally, fatigue is a well-
documented side effect of interferon-beta.140 However, we found no such
indication in our study.

5.1.2.3 Linking fatigue trajectories to incidence of psychiatric comorbidities

Few studies have examined the incidence of depression in MS. A German study
found that nearly 35% of newly diagnosed PwMS were diagnosed with

67
depression within five years of their MS diagnosis, with almost half of them were
diagnosed within the first year.141 This aligns with our findings of a high incidence
of depression after DMT initiation. Additionally, we observed that a higher
incidence of depression was associated with higher fatigue severity following
first DMT start, emphasizing the strong association between fatigue and
depression rather than suggesting a causal relationship.

5.1.2.4 Linking fatigue trajectories to incidence of disability pension/activity

compensation

Consistent with previous research showing that newly diagnosed PwMS have an
increased risk of work loss,142,143 we also observed a high incidence of disability
pension/activity compensation following first DMT start. Previous studies
emphasize the impact of fatigue on the ability to sustain employment in PwMS.21
Our findings agree with this, showing that the incidence of work loss was
particularly pronounced among individuals with high fatigue levels.

5.1.3 Study III - HIRT in people with MS experiencing fatigue

In Study III, we found a similar and clinically relevant reduction in fatigue, as

measured by the FSMC, in both intervention groups, HIRT once a week and HIRT
twice a week. A clinically relevant reduction in fatigue was also observed when
compared to a contemporary non-intervention control group. Additionally, we
observed significant reductions in secondary outcomes, including fatigue
measured by the FSS, depression and anxiety assessed by HADS, the impact of
MS on physical and psychological well-being measured by MSIS-29, and HRQoL
assessed by EQ-VAS. However, we did not find any relevant evidence of
associations between the reduction in fatigue and changes in plasma
inflammatory protein levels previously linked to fatigue in MS.

Among intervention studies on resistance training in PwMS evaluating fatigue as

an outcome,44 only a few specifically applied progressive resistance training to
fatigued PwMS, and these studies involved small sample sizes.144,145 Akbar et al.144
compared 5 fatigued PwMS who participated in progressive resistance training
for 16 weeks, three times a week, to 5 fatigued PwMS who did stretching during
the same period with the same frequency. Both training programs combined
group training (supervised by exercise leaders) and home-based exercise, with
the majority of sessions being unsupervised at home. Changes in fatigue,
assessed by the MFIS, revealed a small time effect in both groups, but no time x
group interaction effect was observed. However, due to the small sample size, it

68
is not possible to draw any definitive conclusions from these results. A study by
Gomez-Illan et al.145 compared 13 fatigued PwMS undergoing a supervised
maximal strength training protocol for 12 weeks with 13 fatigued PwMS who were
instructed to refrain from regular physical exercise during the study period.
Changes in fatigue, assessed by the FSS, showed a significant time x group
interaction effect, suggesting a reduction in fatigue in favor of the intervention
group. However, this effect diminished at the follow-up conducted 10 weeks
after the intervention ended. Again, due to the small sample size, these findings
may not be a reliable indicator of true efficacy. Another progressive resistance
training intervention study by Dodd et al.146 on PwMS with RRMS, not specifically
including fatigued individuals, compared 36 PwMS who participated in
supervised progressive resistance training twice a week for 10 weeks with 35
PwMS who received usual care and participated in a social program once a week
during the same period. Changes in fatigue, assessed by the MFIS, showed
improvements in physical fatigue favoring the training group. However, 12 weeks
after the intervention ended, nearly no between-group differences remained.

As shown in exercise studies on fatigue in general, it is difficult to make direct

comparisons between studies on progressive resistance training due to the
significant variation in methodology.35,139 However, there are of course also
similarities. Most studies have used validated fatigue measures, including the
FSMC, MFIS, and FSS, which have been shown to be highly correlated.58
Additionally, most of these studies have based their interventions on the 2009
guidelines from the American College of Sports Medicine.129 On the other hand,
there is considerable variability in the control groups used. For example, Akbar et
al. compared progressive resistance training to stretching,144 Dodd et al.
compared progressive resistance training to a social program,146 and Gomez-Illan
et al. instructed their control group to refrain from regular physical exercise
during the study period.145 In our study, we decided to compare HIRT once a
week to HIRT twice a week, and further compared the combined HIRT groups to a
non-intervention control group. For chronic diseases, the recommendation is
muscle-strengthening exercise twice a week. However, most primary care
facilities can only offer supervised exercise once a week for 8-12 weeks.
Therefore, we aimed to investigate whether there is a difference in effect
between exercising twice a week versus once a week. The primary reason for
selecting a non-experimental control group was ethical considerations. It would

69
not be ethical to instruct someone to refrain from physical exercise when there
is strong evidence supporting its benefits.

Unexpectedly, we found associations between reduction in fatigue and increase

in immune-related proteins, including tumour necrosis factor and interleukin-17A,
measured before and after 12 weeks of HIRT. This contradicts previous evidence
suggesting a link between increased tumour necrosis factor and interleukin-17A
levels and greater fatigue severity.28 It also differs from findings in a pilot study
by Kierkegaard and colleagues,147 which reported a statistically significant
reduction in tumour necrosis factor after HIRT. These discrepancies may be due
to differences in the timing of the last exercise session and the amount of rest
before sampling, both of which are key factors influencing plasma inflammatory
protein measurements. Similar inconsistencies have been observed in other
studies examining tumour necrosis factor and interleukin-17A changes after
exercise, both in PwMS148 and in healthy adults.149

One might speculate about which components of the intervention were actually
beneficial, as we found no significant differences in most outcomes when
comparing HIRT once a week to twice a week, nor evidence of an association
between reduced fatigue and a decrease in inflammatory protein levels. For
instance, the social aspect of group training may also have played a role.
However, Dodd et al.146 demonstrated some superiority of progressive resistance
training compared to a social program, suggesting that the social component
alone does not explain the benefits.

5.1.4 Study IV - Symptomatic drug treatment for fatigue in MS

In Study IV, we found that modafinil was the most frequently prescribed
treatment for fatigue. Over the study period, the number of PwMS who
dispensed at least one prescription of modafinil or amantadine declined, while
prescriptions for ADHD drugs increased. All treatment groups demonstrated a
statistically significant and similar shift in the trajectory of average monthly work
loss, transitioning from an increasing trend before treatment initiation to no
further increase in the number of lost workdays over the following two years.
Although the untreated group also showed a small change in trajectory, the
change in the trajectory was significantly greater in the modafinil group.

70
5.1.4.1 Annual fatigue treatment prescription rate

An interesting finding was that the yearly prevalence of modafinil increased

between 2006 and 2009 but then gradually declined after 2009. This decline
was likely due to the European Medicines Agency revising its recommendations
in 2010, restricting modafinil's use to narcolepsy.46 Previously, modafinil had been
prescribed for various conditions associated with excessive sleepiness, such as
obstructive sleep apnea and shift work sleep disorder. The European Medicines
Agency's decision was based on clinical trial data indicating that the drug's
benefit–risk profile was favorable only for adults with narcolepsy, with or without
cataplexy. This restriction was primarily driven by concerns about serious
psychiatric disorders and severe skin reactions, particularly in children.

5.1.4.2 Characteristics at index

We observed several key differences between PwMS in the fatigue treatment

cohorts and those in the untreated cohort. While these differences provide
valuable clinical insights, they may also present methodological challenges for
comparative effectiveness analyses. Treated individuals had higher rates of
depression, anxiety, sleep disturbances, and pain, suggesting a greater burden of
psychiatric comorbidities, and polypharmacy. These findings align with Study I,
where we observed that PwMS with the highest reported fatigue had more
psychiatric comorbidities than those with lower fatigue levels. Given the well-
documented association between depression and fatigue,13 our results suggest
that individuals receiving fatigue treatment likely experience more severe fatigue
than those in the untreated cohort. However, due to the low proportion of
individuals with fatigue measures at the index date, we did not include fatigue in
the analyses. Furthermore, findings from Study II, where we found an association
between higher fatigue and a higher incidence of disability pension/activity
compensation, suggest that work loss may serve as a reasonable proxy for
fatigue severity when evaluating the effectiveness of fatigue treatments.

5.1.4.3 Comparative effectiveness analyses on work loss

To our knowledge, this is the first population-based study using an objective

outcome measure to evaluate the comparative effectiveness of different fatigue
treatments in PwMS. Previous studies have primarily been clinical trials, yielding
inconsistent results. Notably, several placebo-controlled studies on amantadine
have shown some efficacy,100,102-105 although some of them reported only small
effects.102,103,105 Additionally, several single-arm trials have indicated an effect of

71
modafinil.94-96 Rammohan and colleagues150 conducted a double-blinded,
placebo-controlled crossover trial and demonstrated a significant effect of 200
mg/day modafinil in reducing fatigue, as assessed with the FSS, MFIS, and a visual
analogue scale for fatigue. However, no significant effect was observed for 400
mg/day on any of the fatigue scales. Other well-designed trials provide little
evidence for an effect of amantadine and modafinil when compared to
placebo.97,98,101 The largest trial in this field, conducted by Nourbakhsh and
colleagues,101 applied a crossover design with four treatment sequences to
increase sample size and minimize unmasking due to possible off-target effects
(e.g., jitteriness). Clinically relevant but similar decreases in fatigue levels were
reported across modafinil, amantadine, methylphenidate, and placebo,
suggesting that the observed effects were primarily driven by placebo
responses. The potential influence of placebo effects has been previously
discussed in fatigue trials in MS.98,102,106 Differences in how treatment allocation
was masked across studies may have led to inconsistent findings.101,151 Studies
with simpler designs, such as parallel-group trials, carry a higher risk of
unmasking due to the noticeable side effects of these medications which could
lead to a false attribution of efficacy to the active drug.101

We observed a potential effect of modafinil when compared to untreated PwMS.

However, due to differences in methodology and outcome measures, direct
comparisons with the above-discussed trials are not possible. It is more
appropriate to view these different results as complementary findings.
Nevertheless, the extent to which the effect we observed reflects a true causal
effect of modafinil remains unclear. Key concerns include the risk of confounding
by indication, non-parallel trajectories of work loss before the index date, and
regression to the mean due to the matching process, which I discuss further in
the methodological considerations section.

As far as we know, only two trials, in addition to the study by Nourbakhsh and
colleagues, have compared more than one fatigue treatment to placebo within
the same study population.100,106 Krupp and colleagues106 conducted a double-
blind, randomized, parallel-group trial of amantadine, pemoline (a discontinued
CNS stimulant previously used for ADHD and narcolepsy), and placebo over 6
weeks in 93 PwMS. They found that amantadine was superior to both placebo
and pemoline in reducing fatigue. Ledinek et al.100 conducted a pilot study using
a single-blind, randomized, parallel-group design with amantadine, modafinil,
acetyl-L-carnitine (an over-the-counter dietary supplement), and placebo in 69

72
PwMS. They found that amantadine was superior to placebo after one month of
treatment, while modafinil and acetyl-L-carnitine showed no statistically
significant difference compared to placebo, although acetyl-L-carnitine showed
a tendency toward efficacy. These studies suggest that amantadine may be
more effective compared to other common drugs used for fatigue in MS.
However, countering the suggestion that amantadine is superior are the results
from the Nourbakhsh study,101 as well as the 2022 National Institute for Health
and Care Excellence guidelines,35 which could not recommend any
pharmacological treatment over another. This aligns with our findings, which did
not show any significant difference in the effect on work loss across fatigue
treatments. Furthermore, we observed that approximately half of the study
participants did not collect a new prescription within two years after their first
dispensed prescription of amantadine, compared to 27.8% for modafinil and
15.1% for ADHD drugs, suggesting that many were not satisfied with amantadine.
This further counter the suggestion that amantadine is superior to other fatigue
treatments.

Several trials evaluating the effects of amantadine, modafinil, and other

amphetamine-like CNS stimulants on fatigue have also assessed their impact on
excessive daytime sleepiness as secondary endpoints or post hoc analyses.
Rammohan et al. found that modafinil (200 mg/day and 400 mg/day) improved
excessive daytime sleepiness compared to placebo.150 Nourbakhsh et al. further
explored this and found that modafinil and methylphenidate had a clinically
relevant effect on fatigue in participants with excessive daytime sleepiness, but
not in those without it. In contrast, amantadine did not show such an effect.101
Sleep disorders are common in MS, and the association between fatigue and
sleep disorders in MS has been repeatedly reported, although not all PwMS with
fatigue have a sleep disorder.16,33 In study IV, we observed only a small proportion
of individuals with diagnosed sleep disorders at index across cohorts (1.5%-
3.3%). However, a relatively high proportion of participants had filled at least one
prescription for a sleeping aid treatment in the year before the index: 22.3% for
modafinil, 22.9% for amantadine, and 23.8% for ADHD drugs, compared to 16.8%
in the untreated cohort. This suggests that PwMS using fatigue treatments may
experience more sleep difficulties than those not using a fatigue treatment.

73
5.1.5 Final thoughts

5.1.5.1 Fatigue as a theoretical construct

I think it’s a good idea to take a step back and think about fatigue as a
theoretical construct. Although the FSMC demonstrates a strong correlation with
the MFIS (Pearson correlation = 0.83) and the FSS (Pearson correlation = 0.80),58
the difference in how fatigue is conceptualized in the scales remains an issue in
terms of comparability across studies.35 Content validity evaluate how
accurately an instrument captures the relevant aspects of a theoretical
construct, such as fatigue, HRQoL, or depression.53 Typically, this evaluation
involves input from the target population and experts in the field to ensure that
the instrument covers all important dimensions of the theoretical construct.
Strong content validity requires a clearly defined theoretical construct and a
well-described conceptualization process, explaining how the theoretical
construct was translated into items and corresponding scores.53 Establishing
content validity is challenging due to the lack of a universal method.53 A well-
defined theoretical construct is particularly problematic in the case of fatigue,
where no unified definition exists. In 2009, the US Food and Drug Administration
emphasized the significance of content validity in their guidelines for PROMs,53
which, in turn, influenced the COSMIN checklist.61 Consequently, PROMs
developed since 2009, such as the FSMC, have more clearly described their
conceptualization processes.61 However, even after 2009, the conceptualization
process is not clearly described for most fatigue PROMs in MS.61

5.1.5.2 Potential underlying factors contributing to fatigue in MS

There are several hypotheses regarding the underlying factors that contribute to
fatigue in MS. Is it primarily a consequence of the disease itself, or does it stem
from common comorbidities in MS.20,27 The short answer is that we don’t know.
However, the likely answer lies somewhere in between, with multiple factors
interacting to drive fatigue in MS. While scales such as FSMC and MFIS aim to
capture MS-specific fatigue, I think it is a lot to ask of patients to distinguish
between fatigue caused by MS and fatigue caused by comorbid conditions like
psychiatric or sleep disorders.

In study I, and to some extent in Study II, we observed a clinically relevant

association between fatigue and disability, as measured by EDSS, as well as
between fatigue and information processing speed, with a clinically relevant
difference between the lowest and highest quartiles of SDMT scores. This

74
provides some evidence that MS itself may contribute to fatigue through
mechanisms such as inflammation, demyelination, axonal loss, and progressive
brain atrophy.27 However, we found little evidence linking fatigue to disease
duration, MS type, or previous relapses/lesions. Although inflammation has often
been proposed as a key factor of fatigue in MS, our studies did not support this
hypothesis. Fatigue was not associated with relapses or lesions in Studies I-II, nor
did we observe a decrease in plasma inflammatory protein levels in relation to
the reduced fatigue levels in Study III. One possible explanation is that most of
our participants were actively treated with a DMT, minimizing inflammation-
related variability. However, previous studies on the impact of DMTs on fatigue
have yielded limited evidence.139 Another possibility is that the methods used to
measure inflammation were not sensitive enough.

Some studies suggest that high fatigue severity can be present early in the
disease course, even among individuals with low disability, while others highlight
the role of neurodegeneration later in disease progression.152,153 However, we
found no clear evidence that fatigue is more prevalent during any specific phase
of the disease. In Study I, we assessed whether time since MS diagnosis
predicted fatigue severity, and in Study II, we examined fatigue trajectories over
time in RRMS, neither analysis suggested that disease duration is a strong
predictor of fatigue. In summary, our findings suggest that fatigue severity is
influenced by multiple factors rather than a single MS-specific predictor.

As previously discussed, fatigue is often reported as the most common

symptom of MS, and depression and anxiety as the two most common
comorbidities.13,21 It is very likely that depression contributes to fatigue
symptoms in MS to some extent. In terms of symptom overlap, fatigue is one of
the nine primary diagnostic criteria for major depression according to the DSM-
5.154 This overlap was also evident in Penner’s validation study of the FSMC, which
found poor discriminant validity between fatigue (as measured by FSMC, MFIS,
and FSS) and depression (as measured by the Beck Depression Inventory), with
Pearson correlation coefficients of 0.49 for FSMC, 0.56 for MFIS, and 0.45 for
FSS.58 Even in studies that exclude overlapping items from depression
questionnaires, an association between fatigue and depression remains.65
Adding further complexity, previous studies have reported an overlap between
fatigue, depression, and cognitive impairment.8,155 However, findings from Study I
did not support this overlap. The association observed between processing
speed and fatigue remained nearly unchanged after adjusting for psychiatric

75
comorbidities. This is consistent with results from Penner et al., who found that
controlling for depression did not alter the correlation coefficient between
different fatigue measures and SDMT.58

Previous research on fatigue, depression, and cognitive impairment in MS

suggests a significant discrepancy between how patients and neurologists
assess symptoms.58 We observed a much stronger association between the
self-reported impact of MS, measured by the MSIS-29, and fatigue compared to
neurologist-reported MS disability, measured by the EDSS. In self-reported
measures, the patients have the primary interpretative authority, meaning
personality traits and psychological factors may influence how they perceive
and report their disease severity and its impact on daily life. There is some
evidence linking the personality trait of negative affectivity to both MS-related
fatigue and depression.73 Additionally, it has been hypothesized that the
motivational system plays a role in both conditions, albeit through different
mechanisms.156 Furthermore, similar to depression, several studies suggest that
reduced control over a person's symptoms or environment may predict fatigue
in MS.67,157

5.1.5.3 HIRT

The findings from Study III, showing relevant improvement in fatigue, mood, and
well-being in both HIRT groups, suggest that fatigued PwMS who have the time
and willingness to engage in such training are very likely to benefit from HIRT.
However, the lack of between HIRT group effect and the high risk of confounding
in the comparison between the combined HIRT group and the non-intervention
control group, keep us from drawing any causal conclusions. The overall session
adherence rate (70%) and content adherence rate (73%), along with
improvements in most outcomes, suggest good feasibility. Given that fatigue can
be a major barrier to physical activity, high-intensity training may be a viable
alternative, as it enables effective training within a short period.158 However,
previous studies indicate that long-term follow-up (10-12 weeks post-
intervention) suggests that training must be maintained to sustain its
effects.145,146

More broadly, achieving a high level of confidence in exercise studies on fatigue

is inherently challenging. The methodological limitations, particularly the difficulty
of blinding participants and personnel in physical activity trials, make causal
conclusions difficult to establish.36 Consequently, some observed benefits of

76
exercise on fatigue may, to some extent, be influenced by the placebo effect.151
Nevertheless, given the minimal risk of adverse effects and the numerous health
benefits of exercise, it is reasonable to apply a lower threshold for
recommending such interventions compared to those interventions with higher
risks of harm.

5.1.5.4 Symptomatic drug treatments for fatigue in MS

The overall evidence for symptomatic drug treatments for fatigue in MS remains
limited, raising an important question: should clinicians prescribe these
treatments at all? While both clinicians and patients report that these
medications may help some PwMS, their direct effectiveness in reducing fatigue
remains uncertain. In Study IV, we did not evaluate fatigue as an outcome, so we
cannot determine the impact of fatigue treatments on fatigue. However, our
results suggest that these treatments could prevent further increase in work
loss. The 2022 National Institute for Health and Care Excellence guidelines
suggest that pharmacological treatments may be considered before non-
pharmacological approaches when rapid symptom relief is needed.35 Coping
with work responsibilities could be one such indication. However, the guidelines
also emphasize that non-pharmacological interventions should typically be
considered first. They also highlight the importance of discussing potential risks
with patients and ruling out other possible causes of fatigue (e.g. infection)
before initiating treatment.

5.2 Methodological considerations

5.2.1 Precision

Precision reflects the impact of random error on an estimate, which depends on

sample size and variability within the sample. High precision is generally
desirable, as it is necessary to observe associations between variables. However,
high precision can also be misleading, as studies with large sample sizes have a
higher chance of finding statistically significant results that may not be clinically
relevant. Therefore, when possible, we always considered both clinical relevance
and statistical significance in the studies of this doctoral thesis.

A major challenge in Studies I and II was the risk of low precision due to the small
proportion of PwMS with FSMC measurements in SMSreg. We accounted for this
risk when planning these studies. In Study I, we applied a cross-sectional design,
using nearly all FSMC measurements in SMSreg to increase the sample size.

77
However, applying a complete case analysis to handle missing data led to the
exclusion of 5,236 out of 8,449 FSMC measurements. Another issue with FSMC
measurements in SMSreg is that they are often not tied to specific events, such
as DMT initiation or MS diagnosis. Nevertheless, by using a group-based
trajectory modeling design in Study II, we were able to assess fatigue trajectories
over time without requiring a specific baseline value.

In Study III, we were unable to include the number of participants specified in the
sample size calculations to detect a 10-point difference between groups in
average change of FSMC. However, this is unlikely to explain why we failed to
demonstrate a superiority of HIRT twice a week compared to once a week. In
Study IV, precision was high, as we were able to include nearly all PwMS in
Sweden who had filled a prescription for the fatigue treatments under study.

5.2.2 External validity (generalizability)

External validity refers to generalizability, which indicates the extent to which

study findings can be applied to other populations, settings, or time periods
beyond the study sample. A key aspect of external validity is determining
whether the study population is representative of the population of interest, the
target population.

In Studies I and II, we used SMSreg to identify PwMS, which covers around 80% of
the prevalent MS population in Sweden. Therefore, the overall generalizability to
the Swedish MS population is likely high. However, the small proportion of
individuals with available FSMC measurements in SMSreg not only reduced
precision but also limited the generalizability of the findings, as we may have
studied a selected subgroup. For example, missing values in the outcome used
to create trajectory groups cannot be accounted for in group-based trajectory
modeling.117 This means that the fatigue trajectories were based solely on the
available FSMC measurements and may not be fully representative of the
broader source population. In Study II, due to the prospective data collection
from 2017 to 2022 as part of the COMBAT-MS study, the fatigue scores
collected during that period are likely more representative of the Swedish MS
population.

To assess the generalizability of findings from Study I, I will compare participant

characteristics in more detail with a Norwegian population-based study that
assessed fatigue prevalence in 1,454 PwMS using the FSMC.25 In Study I, we found
that 60% of the population was categorized as having fatigue based on the

78
FSMC total score, with 32% experiencing severe fatigue, which aligns with fatigue
prevalence reported in previous studies on fatigue in MS.21 The Norwegian study
reported a higher prevalence of fatigue (81%) and severe fatigue (67%). Their
cohort differed also in other several aspects: participants were older on average
(mean age 52 years vs. 41.4 years), a higher proportion had progressive MS (14%
vs. 5%), the median EDSS score was higher (2.5 vs. mean 1.9), and a larger
proportion of patients were not receiving DMT treatment (59%). These
differences are not surprising, as our study population primarily consisted of
individuals from COMBAT-MS, which includes actively DMT-treated RRMS
patients. Similar to our study, the Norwegian study also had a high proportion of
individuals with missing FSMC measures, with only 37% of those mailed the
questionnaire returning a complete FSMC assessment. This suggests that the
responders may not have been fully representative of the source population. In
summary, I would argue that Studies I-II, are predominantly representative of
actively DMT-treated individuals with RRMS.

As is common in RCTs, the generalizability of findings in Study III is limited, as

participants were not a random sample of PwMS with fatigue. Instead,
recruitment was influenced by participants' interest and their willingness to
invest time in a training intervention to reduce fatigue. On the other hand, a key
strength of the study was that only fatigued PwMS were included, which is
relatively rare in exercise studies, as fatigue is not often the primary endpoint.

Similar to precision, the generalizability of Study IV to Sweden is likely high.

However, some points are worth noting. The prevalence of specific fatigue
treatments varied significantly across regions in Sweden, suggesting that
fluctuations in yearly prevalence should be interpreted at a national level, as
regional differences in treatment patterns exist. On the other hand, there is little
reason to believe that living in a specific region would influence the effect of the
drug on work loss, as the social benefit system is the same throughout Sweden.
However, since social benefit systems differ between countries, the results from
the comparative effectiveness analyses are most likely relevant to Sweden and
possibly other Nordic countries.

5.2.3 Causal inference

Before discussing methodological considerations in relation to internal validity, I

first want to share some general thoughts about causal inference in relation to
the studies of my thesis. Both study I-II were explorative in its study design and

79
rather descriptive than inferential. Instead of aiming of assessing causal
relationships between characteristics and fatigue, we were interested in
associations and to describe fatigue evolution over time.

The random allocation in Study III strengthens the causal interpretation of the
findings. However, since blinding was not possible for study participants in this
study, we cannot rule out the possibility of a placebo effect of HIRT. Moreover,
the comparison between HIRT and the matched non-randomized cohort does
not demonstrate a causal effect of HIRT, but rather serves as a descriptive
observation that fatigue appears to remain unchanged in PwMS in the short
term.

The interrupted time series design in Study IV strengthens the causal

interpretation of the findings in the comparison across fatigue treatments.
However, the comparison between modafinil and the untreated cohort provides
weaker causal evidence. In interrupted time series analysis, causal interpretation
is strengthened when the outcome level (intercept) and trajectory (slope) are
similar before the intervention (i.e., parallel pre-index trajectories). If the
trajectories then diverge post-index, this supports a causal effect. In our
analysis, the level of monthly work loss before treatment initiation was not
significantly different between amantadine, ADHD drugs, and modafinil. Likewise,
the pre-index trajectories of monthly work loss ran in parallel. However, these
trajectories continued to run in parallel even after treatment initiation, suggesting
no difference in the effect of the fatigue treatments on work loss. When
comparing modafinil to the untreated cohort, the pre-index level of monthly
work loss was significantly different, and the trajectories were not parallel.
Although the trajectories diverged after index, the better attenuation of monthly
work loss in the modafinil group should be interpreted with caution due to these
pre-index differences. Moreover, treatment status was defined based on the
filling of a single prescription. Many individuals, particularly new amantadine
users, did not refill their prescriptions during the follow-up period. This raises
uncertainty about whether the observed changes in monthly work loss rates can
be attributed to the treatment under which participants were categorized.

5.2.4 Internal validity

I will here go through some of the most relevant considerations in relation to

each bias introduced in the method section.

80
5.2.4.1 Selection bias

Selection bias, also known as collider stratification bias, occurs when study
participants are selected or lost to follow-up in a non-random manner, leading to
systematic differences between those included and those excluded. By
definition, selection bias arises when both the exposure and the outcome
influence the probability of being selected or lost to follow-up. Selection bias
also arises when incorrectly adjusting for colliders.

As with precision and generalizability, the risk of selection bias in Study IV is low
due to the use of nationwide data, which allowed us to include nearly all PwMS in
Sweden. We followed individuals from their first fatigue treatment start, including
those with an MS diagnosis recorded in SMSreg or in the NPR (defined as ≥3
separate MS diagnoses). The sensitivity and specificity of this definition have
been previously validated.159 However, there is still a risk that some PwMS were
not captured by our inclusion criteria (or that we misclassified individuals as
having MS when they do not have MS). However, any systematic differences in
the risk of missing PwMS across fatigue treatment cohorts are likely small. In
contrast, selection bias could potentially be a problem when comparing
modafinil users to the untreated cohort. As illustrated in the simplified DAG in
Figure 5.1a, untreated PwMS might have a lower probability of meeting our
inclusion criteria of MS diagnosis, potentially due to a more stable disease
course leading to fewer healthcare visits. Additionally, work loss may increase
the probability of receiving an MS diagnosis according to our criteria, as
healthcare contact is required for extended sick leave or disability
pension/activity compensation. As the probability of having an MS diagnosis (i.e.
being part of our study population) is a collider in the non-causal path between
treatment exposure and work loss, and as we don’t include all PwMS in Sweden,
we open up the non-causal path: treatment exposure -> MS diagnosis <- work
loss. As a result, some untreated PwMS with a lower risk of work loss may be
underrepresented in our study population. This could make the untreated group
appear worse than it actually is, as those untreated with a higher risk of work loss
are more likely to be included.

81
 (a) (b)

Figure 5.1. DAG to explain potential collider stratification bias in Study IV (left) and Study
III (right).

As previously mentioned, the risk of selection bias is much higher in Studies I-II,
where fatigue was assessed as an outcome. This is due to the low proportion of
individuals in SMSreg with an FSMC measurement and the high rate of missing
data in other SMSreg variables. For example, when examining the association
between disability and fatigue, individuals with higher EDSS scores may be more
overrepresented in SMSreg compared to those with lower EDSS. This is because
patients with greater disability may require more frequent monitoring in the
clinic. However, the opposite could also be true: PwMS with progressive disease
(with higher EDSS) may be underrepresented, as they are less frequently treated
with DMT and therefore not monitored as regularly as those with RRMS. The
absence of FSMC measurements in SMSreg could result from patient-related
factors, such as unwillingness to fill in fatigue questionnaires due to poor health
(i.e., healthy adherer bias), or clinician-related factors, such as a lack of interest
in research. However, FSMC is more likely to be collected in clinical practice
when fatigue is suspected or confirmed. In summary, it is difficult to determine
whether this potential selection bias would underestimate or overestimate the
association between disability and fatigue in MS. In Study II, we attempted to
mitigate selection bias due to missing data by using multiple imputation and in
Study I, a comparison of characteristics between the included cohort and the full
cohort with missing data revealed similar profiles. While this approach may have
reduced selection bias within our study population, it did not address the
exclusion of individuals without any FSMC measurement in Study I or fewer than
three FSMC measurements in Study II.

As previously discussed, selection bias due to attrition is a common concern in

RCTs, particularly when blinding is not feasible. This bias occurs when dropout is
related to both the intervention and the outcome. However, this potential bias is
often mitigated by using an intention-to-treat analysis. Study III is no exception

82
to the risk of selection bias. As illustrated in the DAG in Figure 5.1b, it is plausible
that being allocated to one of the intervention groups influenced the probability
of continuing the intervention. Furthermore, improvements in fatigue or other
outcomes could also impact the probability of continuing the intervention. We
observed a difference in discontinuing the intervention between the groups: the
HIRT once-a-week group had a higher proportion of participants discontinuing
the intervention, either before or after the first HIRT session. While we cannot be
certain whether those who discontinued the intervention would have
experienced better or worse improvements in fatigue had they completed the
full program, the fact that discontinuation was more common in the once-a-
week group raises the possibility that those who dropped out had poorer fatigue
improvement from the intervention. If this was the case, the treatment effect for
the once-a-week group would have been overestimated, as the group’s average
improvement would appear better than it truly is. To address this potential bias,
we applied an intention-to-treat analysis, as is standard practice in RCTs.

5.2.4.2 Information bias

Information bias arises when errors in measuring either the exposure or the
outcome distort their relationship. Non-differential misclassification occurs
when such errors are unrelated to the other variable, typically biasing estimates
toward the null. In contrast, differential misclassification arises when errors in the
exposure is related to the outcome or vice versa, which may bias the results in
both directions. I will begin by discussing the risk of information bias in Study IV,
which is somewhat more straightforward. Then I will move on to Studies I–III,
where information bias is a more theoretical concern due to the subjective
nature of fatigue as an outcome.

In Study IV, there is a potential risk of misclassification in both the exposure and
the outcome. I will first briefly address the outcome, work loss, before turning to
exposure misclassification. Employers typically cover day 2 to day 14 in Sweden,
with the Swedish Social Insurance Agency covering longer absences, while
unemployed are entitled to sick leave benefits from day 2. To ensure
comparability across employment statuses, only sick leave episodes exceeding
14 days were included in the analysis. Instead it may have introduced non-
differential misclassification, likely diluting any true differences between
exposure groups. Moreover, the exclusion of shorter sick leave periods likely
enhanced the specificity for identifying sick leave due to fatigue, as periods

83
shorter than 14 days are less likely to be related to fatigue. Future studies could
examine sick leave certificates to determine the underlying reasons for absence.

Misclassification of the exposure is a more substantial concern in Study IV,

particularly because we investigated off-label use of treatments and the PDR
does not contain easily accessible information on the indication for which a
treatment was prescribed. While amantadine and modafinil are more commonly
prescribed with MS as the likely indication, there is greater uncertainty
surrounding ADHD drugs. Some individuals with MS may have been prescribed
these drugs primarily for ADHD rather than fatigue. Although we applied several
methodological strategies to increase the likelihood that prescriptions reflected
treatment for fatigue, such as excluding individuals with a prior ADHD diagnosis,
misclassification of the exposure cannot be ruled out. If these individuals
experienced greater improvements in work loss due to better-managed ADHD
rather than alleviated fatigue, this could lead to an overestimation of the
effectiveness of ADHD drugs for treating fatigue-related work loss.

Additionally, we used an ever-treated approach in Study IV to reduce selection

bias, assuming that individuals remained on their assigned treatment throughout
the follow-up period. While this method helps avoid bias from treatment
switching or discontinuation, it may have introduced information bias,
specifically, a potential underestimation of the true effect of ADHD drugs. For
instance, many individuals did not refill their amantadine prescription and
instead filled a prescription for an ADHD drug. As a result, any beneficial effect of
the ADHD drug might have been incorrectly attributed to amantadine. An on-
treatment approach could potentially address this issue; however, it is generally
discouraged in effectiveness studies because it risks overestimating treatment
effects by excluding those who discontinue treatment.

The potential for information bias in Studies I-III is more theoretical, primarily due
to the subjective nature of fatigue as an outcome. There is no gold standard for
measuring MS-related fatigue. Instruments like the MFIS and FSMC rely on
patients’ subjective assessments of severity and impact on fatigue, which may
be influenced by various psychological or situational factors. If such subjective
judgments differ from a theoretical construct of fatigue (i.e., the “true” fatigue
level), and this error is unrelated to treatment or predictor variables, it
represents non-differential misclassification, potentially biasing results toward
the null. However, if the error is differential, for instance if fatigue ratings vary

84
systematically with levels of disability or depression, this may underestimate or
overestimate associations. A relevant example is the overlap between fatigue
and depression. As discussed earlier, the FSMC shows generally good validity,
but its discriminant validity relative to self-reported depression is limited. 58
Penner et al. showed high sensitivity and specificity for FSMC, MFIS, and FSS
when comparing individuals with and without MS, but these measures seem to
lack specificity in differentiating between MS-related fatigue and fatigue due to
depression. This uncertainty is one reason we chose to use a more objective
outcome measure (i.e., work loss) in Study IV when evaluating drug effects.

In Study III, information bias may also have influenced the results. Since blinding
was not feasible, participants might have overestimated improvements in fatigue
due to expectations about the effects of HIRT. However, as both groups received
the same intervention, differing only in frequency, any misclassification of
improvement is likely to have been similar across groups. This would constitute
non-differential misclassification, which typically biases estimates toward the
null. In other words, a true effect of exercise frequency on fatigue reduction may
have been more difficult to detect.

5.2.4.3 Confounding

Confounding emerges when the exposure and the outcome have a common
cause. As previously discussed, confounding is closely related to the concept of
causal inference. As also mentioned earlier, Studies I-II were exploratory in their
design, were we treated potential confounders differently compared to Studies
III-IV, where our ultimate aim was to assess the causal effects of treatments for
fatigue. Study III was an RCT, which by design do not introduce confounding.
Therefore, this section will concern the other studies included in my doctoral
thesis.

In general, the extensive register linkage data allowed us to include a wide range
of covariates in our regression models, including socio-demographic factors,
comorbidities, health care utilization, concomitant treatments, MS-related
characteristics, and various PROMs. However, a high proportion of missing data
in several SMSreg variables was one of the main reasons for risk of residual
confounding. Moreover, we lacked information on lifestyle factors such as
smoking, physical activity, and alcohol consumption, likely important
confounders in the context of fatigue.

85
In Study I, in line with our exploratory approach, we sequentially added groups of
covariates to the statistical model to examine how associations with fatigue
changed. As previously discussed, we concluded that a model including MSIS-29
and EQ-VAS would not be optimal. This highlights the importance of carefully
considering which covariates to include in future etiological and treatment
studies where fatigue is the outcome. DAGs can serve as useful tools in this
context. Figure 5.2 presents three simplified DAGs illustrating possible
relationships between disability, fatigue, and HRQoL, where HRQoL is used to
represent both MSIS-29 and EQ-VAS for simplicity. Figure 5.2a illustrates HRQoL
as a confounder, Figure 5.2b as a collider, and Figure 5.2c as a mediator. As
previously discussed, the most realistic scenarios are likely alternatives (b) and
(c), or a combination of the two. In scenario (b), adjusting for HRQoL would
introduce collider stratification bias, while in scenario (c), adjusting for HRQoL
would imply estimating only the direct path between disability and fatigue,
excluding the causal path through HRQoL. In reality, this direct path is probably
small and of limited clinical relevance. Importantly, in Study I, FSMC, MSIS-29,
EQ-VAS, and EDSS were measured simultaneously due to the cross-sectional
design, making these causal paths speculative in relation to this study.

In Study II, when identifying potential predictors of fatigue trajectories, MSIS-29

and EQ-VAS were included in the final model. Therefore, interpretation of the
results should consider the previous discussion regarding their role as potential
confounders, colliders, or mediators. A key difference between Study II and
Study I was that Study II allowed for the assessment of temporal relationships
between predictors and fatigue. This temporal ordering makes scenario (b),
where HRQoL acts as a collider, less likely. Scenario (c), where HRQoL functions
as a mediator, appears more plausible. In contrast, scenario (a), where HRQoL is a
confounder, seems unlikely (Figure 5.2).

(a) (b) (c)

Figure 5.2. Simplified DAGs illustrating how MSIS-29 and EQ-VAS (represented here as
HRQoL) may function as a (a) confounder, (b) collider, or (c) mediator in the relationship
between disability and fatigue.

86
In Study IV, we addressed confounding by indication by applying an active
comparator, new user design when comparing fatigue treatments. However,
residual confounding may still be present, particularly in the comparison
between ADHD drugs and modafinil. ADHD drugs often appear to be used as a
second-line or third-line treatment after modafinil, which was partly confirmed
by our data: 95% of new modafinil users received it as their first fatigue
treatment, whereas 89% of ADHD drug users had previously been treated with
modafinil or amantadine. This treatment sequencing may introduce unmeasured
confounding. Restricting the comparison to only completely new users of ADHD
drugs and modafinil would reduce this bias, but such an approach would likely
suffer from low statistical power and limit generalizability. On the other hand, the
use of an interrupted time series design strengthens the causal interpretation, as
pre-treatment levels of monthly work loss as well as trajectories in monthly work
loss were similar across treatment groups.

As discussed earlier, the interrupted time series design did not support a causal
interpretation of the difference in work loss trajectories between the modafinil
and untreated cohort. It is well known that comparing a treated group to an
untreated one in observational studies is problematic, as confounding by
indication can be substantial. For this reason, active comparator designs are
generally recommended, especially given the difficulty of measuring frailty and
other unmeasured confounders.120 It is likely that frail individuals are not being
prescribed these fatigue treatments due to the risk of side effects. At the same
time, frail individuals are also more likely to experience greater work loss. Hence,
causing confounding by indication. On the other hand, from descriptive statistics
in study IV, where we compared the untreated cohort to the fatigue-treated
cohorts, suggests that the untreated individuals may have milder MS and fewer
psychiatric comorbidities. Additionally, it’s uncertain to what extent individuals
across the cohorts receive other types of fatigue interventions. So, why did we
even attempt to compare to an untreated group? Our reasoning was based on
the hypothesis that fatigue treatments may not have any effect at all. Although
the potential for causal interpretation is limited due to the aforementioned
issues, the comparison may still offer informative insights.

87
6 Conclusions
 Fatigue in MS is associated with a higher disability level (measured by
EDSS) and slower cognitive processing speed (measured by SDMT), with
clinically relevant differences in average fatigue scores across levels of
EDSS and SDMT. In contrast, type of MS (RRMS or SPMS), time since MS
diagnosis, line of DMT, and history of relapses show no or weak
associations. Moreover, patient-reported outcome measures evaluating
HRQoL (measured by EQ-VAS) and physical and psychological impact of
MS (measured by MSIS-29) are highly correlated with fatigue.
 Fatigue severity remains relatively stable in the years following the first
DMT initiation and the second DMT start in actively treated people with
RRMS, regardless of fatigue level at DMT start. However, there is some
evidence of an increased risk of clinically relevant worsening of fatigue
among individuals with moderate fatigue at DMT initiation.
 High-intensity resistance training is associated with a clinically meaningful
reduction in fatigue scores among fatigued people with MS, though a
causal relationship could not be established due to the lack of a
randomized non-intervention control group.
 Off-label use of modafinil has predominantly been the most common
symptomatic drug treatment for fatigue in Sweden over the last two
decades, followed by amantadine and central stimulants approved for
ADHD. There are potential treatment benefits of modafinil and other
stimulants used for symptomatic fatigue treatment in MS, in terms of
mitigating the worsening of work loss. However, there is no evidence that
one treatment is superior to another.

89
7 Points of perspective
The conclusions drawn from this doctoral thesis give rise to important questions
that warrant further investigation in future research:

 There is a need for further prospective, population-based studies with

fatigue as the primary outcome to better understand its predictors and
consequences in MS. Ideally, such studies should incorporate a range of
validated fatigue questionnaires to capture different dimensions of
fatigue, along with assessments of related symptoms and comorbidities
relevant to both MS and fatigue. Linking this information to national
registers would improve the efficiency and completeness of data
collection, while also reducing the burden on study participants.
 Additional population-based cohort studies are needed to further
evaluate the comparative effectiveness of commonly used disease-
modifying therapies on fatigue in people with MS.
 Future randomized controlled trials are needed to investigate how people
with MS experiencing fatigue respond to various types of exercise
interventions, including comparisons with different high-intensity
resistance training protocols, but also against or intergrated with aerobic
exercise interventions. In the longer term, larger pragmatic randomized
controlled trials are warranted to examine the real-world effect of these
interventions in routine clinical practice.
 To enhance specificity in future cohort studies on the comparative
effectiveness of symptomatic drug treatments for fatigue, efforts should
be made to better capture MS-specific work loss and more accurately
determine treatment indications, potentially through the analysis of free-
text entries in the Prescribed Drug Register and sick leave cerificates. To
improve causal interpretation when comparing symptomatic drug
treatment for fatigue to untreated groups, further methodological
consideration is needed to ensure greater pre-intervention comparability
between groups.

91
8 Acknowledgements
First, I would like to thank some people at the Department of Clinical
Neuroscience (CNS) and the Division of Clinical Epidemiology (KEP), as well
as former colleagues and classmates.

Fredrik Piehl, my main supervisor. Always busy, yet always available, empathetic
and intellectually present. It's hard to find someone more kind-hearted and
supportive , no wonder you're the patients’ favorite! I'm so glad you tricked me
into applying for this PhD position! Thank you for all the support and for believing
in me.

Thomas Frisell, my co-supervisor. Thank you for teaching me the fine craft of
working with Swedish registers and various fun epidemiological and statistical
methods. Your thoughtful and concrete responses have been incredibly valuable
whenever I’ve gotten lost in my SAS code or methodological considerations.
Thank you for letting me be part of your team at KEP.

Elisa Longinetti, my co-supervisor, it was such a pleasure working closely with

you on the fatigue trajectory study and getting to know you. We made a truly
great team, and you were an incredible support, and the best lunch companion!

Marie Kierkegaard, my co-supervisor, thank you for playing such an important

role in connecting my thesis to clinical reality, not only in the exercise study but
also throughout my other studies.

Suvi Virtanen, my COMBAT-MS coordinator colleague and friend. The best

problem solver in the world, and you really taught me how to monitor a
coordinate a cohort study, pragmatically and with finesse. Thank you for all the
stimulating and valuable discussions.

Jette, thank you for being part of creating and running the excellent Public Health
program at KI, and for being such a great support during my doctoral studies.

Peter Alping, thank you for all the enjoyable and insightful conversations. You’re
an excellent office neighbor and a true role model for how to think and work
systematically.

To the rest of the MS team and Team Frisell at KEP: Kyla, Huiling, Stefanie and
Jacob. It is such a pleasure sharing an office with you.

93
Current and former doctoral students at KEP: Ngoc, Viktor M, Matilda, Andrei, Ida,
Anton, Viking, Karin K, Marina, Kelsi, Simon S, Younes, Felix, Joshua, and others.
Thank you for the great discussions during seminars and lunches.

To everyone in the Askling research group: Johan, Bénédicte, Viktor W, Elizabeth,

Daniela, Marie H, Björn, Martin, Maxime, Helga, Mehdi, Gustaf, Karin H and others.
Thank you for welcoming me into your group and for the valuable feedback on
my research.

Johan Reutfors, thank you for being an engaging co-author, and for your
intellectual insights and reflections on how psychiatric diagnoses are formed.

Current and former IMSE and COMBAT team: Linda, Veronica, Victoria, Stina, Suvi,
and Emily. It was such a pleasure working with you. A special thanks to Emily for
being such a great office neighbor and for all the invaluable help!

All patients and clinicians at the Centre of Neurology and other MS clinics in
Sweden, thank you for teaching me about MS and fatigue, and for contributing by
entering data into the Swedish MS Register.

Thank you to all the participants in the training study who made it possible to
carry out the study. Also thanks to Mohsen and Ann-Maria for their help with
sample collection and handling of blood samples

A big thank you to the helpful and friendly staff at CNS and KEP, especially
Helena and Alexandre.

Ellinor Berglund, thank you for recognizing my somewhat undeveloped interest in

research at the emergency department at SöS, and introducing me to the world
of clinical research and quality registers.

All my friends from the Master's program in Public Health at KI: Diego, Vicky, Hilda,
Jonathan, Gustavo, Alexandro, Ning, Kelsi, Huiling, and everyone else. Thank you
for being such amazing classmates, for all the fun times, and for being a part of
my journey to become a researcher.

Finally my friends and family.

Marcus, Jonas, Hagge, Toffe, Kim och Jujje från bokklubben: tack för att ni är så
fina vänner och ser till att jag inte bara läser litteratur om MS och
epidemiologiska metoder.

94
Linus, John, David och Mange, mina gymnasievänner som nog alltid kommer ha
en särskild plats i mitt hjärta. Tack för alla fina samtal om livets små och stora
frågor.

John och Emma, mina syskon som alltid finns där, och förstår att den verkliga
världen, där man egentligen borde befinna sig, är på Brändö, i omfamning av den
åländska skärgården. Tur att vi kan träffas rätt ofta där i alla fall, och på andra
ställen för den delen.

Mamma och pappa, tack för att ni gav mig en så trygg uppväxt på den finaste
platsen på jordklotet. Och tack för att ni alltid har ställt upp i alla väder, oavsett
vad det har gällt. Tack för all kärlek och uppmuntran.

Camilla, mitt livs kärlek och livskamrat. Du är den som ser till att livet verkligen
händer, och fyller det med allt viktigt som ett liv ska innehålla: värme, trygghet,
smågalna äventyr och spännande utmaningar. Utan dig hade jag nog aldrig tagit
steget att bli doktorand, och utan dig hade jag nog gått in i studierna lite för
mycket. Tack för att du balanserar upp mitt liv och gör det så fint!

Herta, min älskade dotter. Det finns inget viktigare eller mer glädjefyllt än att få
vara med dig och se dig växa upp.

95
9 References

1. Filippi M, Bar-Or A, Piehl F, et al. Multiple sclerosis. Nat Rev Dis Primers.
2018;4(43):43.
2. McGinley MP, Goldschmidt CH, Rae-Grant AD. Diagnosis and Treatment of
Multiple Sclerosis: A Review. JAMA. 2021;325(8):765-779.
3. GBD 2015 Neurological Disorders Collaborator Group. Global, regional, and
national burden of neurological disorders during 1990-2015: a systematic
analysis for the Global Burden of Disease Study 2015. Lancet Neurol.
2017;16(11):877-897.
4. Ahlgren C, Odén A, Lycke J. High nationwide prevalence of multiple
sclerosis in Sweden. Mult Scler. 2011;17(8):901-908.
5. The Multiple Sclerosis International Federation. Atlas of MS, 3rd Edition
2020; https://www.atlasofms.org/map/sweden/epidemiology/number-of-
people-with-ms#about.
6. Ahlgren C, Odén A, Lycke J. High nationwide incidence of multiple
sclerosis in Sweden. PLoS One. 2014;9(9):e108599.
7. Kappos L, Wolinsky JS, Giovannoni G, et al. Contribution of Relapse-
Independent Progression vs Relapse-Associated Worsening to Overall
Confirmed Disability Accumulation in Typical Relapsing Multiple Sclerosis
in a Pooled Analysis of 2 Randomized Clinical Trials. JAMA Neurol.
2020;77(9):1132-1140.
8. Rocca MA, Amato MP, De Stefano N, et al. Clinical and imaging assessment
of cognitive dysfunction in multiple sclerosis. Lancet Neurol.
2015;14(3):302-317.
9. Kingwell E, van der Kop M, Zhao Y, et al. Relative mortality and survival in
multiple sclerosis: findings from British Columbia, Canada. J Neurol
Neurosurg Psychiatry. 2012;83(1):61-66.
10. Marrie RA, Cohen J, Stuve O, et al. A systematic review of the incidence
and prevalence of comorbidity in multiple sclerosis: overview. Mult Scler.
2015;21(3):263-281.
11. Marrie RA, Patten SB, Tremlett H, et al. Sex differences in comorbidity at
diagnosis of multiple sclerosis: A population-based study. Neurology.
2016;86(14):1279-1286.
12. Marrie RA. Comorbidity in multiple sclerosis: implications for patient care.
Nat Rev Neurol. 2017;13(6):375-382.

97
13. Marrie RA, Reingold S, Cohen J, et al. The incidence and prevalence of
psychiatric disorders in multiple sclerosis: a systematic review. Mult Scler.
2015;21(3):305-317.
14. Kaminska M, Kimoff RJ, Schwartzman K, Trojan DA. Sleep disorders and
fatigue in multiple sclerosis: evidence for association and interaction. J
Neurol Sci. 2011;302(1-2):7-13.
15. Brass SD, Li CS, Auerbach S. The underdiagnosis of sleep disorders in
patients with multiple sclerosis. J Clin Sleep Med. 2014;10(9):1025-1031.
16. Marrie RA, Reider N, Cohen J, et al. A systematic review of the incidence
and prevalence of sleep disorders and seizure disorders in multiple
sclerosis. Mult Scler. 2015;21(3):342-349.
17. Piehl F. Current and emerging disease-modulatory therapies and
treatment targets for multiple sclerosis. J Intern Med. 2021;289(6):771-791.
18. Socialstyrelsen. Nationella riktlinjer för vård vid multipel skleros (MS) -
Prioriteringsstöd till beslutsfattare och chefer 2022. Socialstyrelsen 2022.
19. Rosenthal TC, Majeroni BA, Pretorius R, Malik K. Fatigue: an overview. Am
Fam Physician. 2008;78(10):1173-1179.
20. Penner IK, Paul F. Fatigue as a symptom or comorbidity of neurological
diseases. Nat Rev Neurol. 2017;13(11):662-675.
21. Oliva Ramirez A, Keenan A, Kalau O, Worthington E, Cohen L, Singh S.
Prevalence and burden of multiple sclerosis-related fatigue: a systematic
literature review. BMC Neurol. 2021;21(1):468.
22. Multiple Sclerosis Council Paralyzed Veterans of America. Fatigue and
Multiple Sclerosis: Evidence-based Management Strategies for Fatigue in
Multiple Sclerosis. Multiple Sclerosis Council for Clinical Practice
Guidelines; 1998.
23. Van Denend T, Gecht-Silver M, Kish J, Plow M, Preissner K. Appreciating the
experience of multiple sclerosis fatigue. Br J Occup Ther. 2023;86(6):441-
450.
24. Rommer PS, Eichstädt K, Ellenberger D, et al. Symptomatology and
symptomatic treatment in multiple sclerosis: Results from a nationwide
MS registry. Mult Scler. 2019;25(12):1641-1652.
25. Broch L, Simonsen CS, Flemmen H, et al. High prevalence of fatigue in
contemporary patients with multiple sclerosis. Mult Scler J Exp Transl Clin.
2021;7(1):2055217321999826.
26. Yoon JH, Park NH, Kang YE, Ahn YC, Lee EJ, Son CG. The demographic
features of fatigue in the general population worldwide: a systematic
review and meta-analysis. Front Public Health. 2023;11:1192121.

98
27. Langeskov-Christensen M, Bisson EJ, Finlayson ML, Dalgas U. Potential
pathophysiological pathways that can explain the positive effects of
exercise on fatigue in multiple sclerosis: A scoping review. J Neurol Sci.
2017;373:307-320.
28. Heesen C, Nawrath L, Reich C, Bauer N, Schulz KH, Gold SM. Fatigue in
multiple sclerosis: an example of cytokine mediated sickness behaviour? J
Neurol Neurosurg Psychiatry. 2006;77(1):34-39.
29. Flachenecker P, Bihler I, Weber F, Gottschalk M, Toyka KV, Rieckmann P.
Cytokine mRNA expression in patients with multiple sclerosis and fatigue.
Mult Scler. 2004;10(2):165-169.
30. Mäurer M, Comi G, Freedman MS, et al. Multiple sclerosis relapses are
associated with increased fatigue and reduced health-related quality of
life - A post hoc analysis of the TEMSO and TOWER studies. Mult Scler
Relat Disord. 2016;7:33-40.
31. Marino FE. Heat reactions in multiple sclerosis: an overlooked paradigm in
the study of comparative fatigue. Int J Hyperthermia. 2009;25(1):34-40.
32. Picariello F, Freeman J, Moss-Morris R. Defining routine fatigue care in
Multiple Sclerosis in the United Kingdom: What treatments are offered and
who gets them? Mult Scler J Exp Transl Clin. 2022;8(1):20552173211072274.
33. Veauthier C, Hasselmann H, Gold SM, Paul F. The Berlin Treatment
Algorithm: recommendations for tailored innovative therapeutic strategies
for multiple sclerosis-related fatigue. EPMA J. 2016;7(1):25.
34. Asano M, Finlayson ML. Meta-analysis of three different types of fatigue
management interventions for people with multiple sclerosis: exercise,
education, and medication. Mult Scler Int. 2014;2014:798285.
35. National Institute for Health and Care Excellence. Multiple sclerosis in
adults: management. London: National Institute for Health and Care
Excellence (UK);2022.
36. Heine M, van de Port I, Rietberg MB, van Wegen EE, Kwakkel G. Exercise
therapy for fatigue in multiple sclerosis. The Cochrane database of
systematic reviews. 2015;15(9):Cd009956.
37. Pilutti LA, Greenlee TA, Motl RW, Nickrent MS, Petruzzello SJ. Effects of
exercise training on fatigue in multiple sclerosis: a meta-analysis.
Psychosom Med. 2013;75(6):575-580.
38. Torres-Costoso A, Martínez-Vizcaíno V, Reina-Gutiérrez S, et al. Effect of
Exercise on Fatigue in Multiple Sclerosis: A Network Meta-analysis
Comparing Different Types of Exercise. Arch Phys Med Rehabil.
2022;103(5):970-987.e918.
39. WHO. Global health risks: mortality and burden of disease attributable to
selected major risks. Geneva World Health Organization;2009.

99
40. Pedersen BK, Saltin B. Exercise as medicine - evidence for prescribing
exercise as therapy in 26 different chronic diseases. Scand J Med Sci
Sports. 2015;25 Suppl 3:1-72.
41. Andreasen AK, Stenager E, Dalgas U. The effect of exercise therapy on
fatigue in multiple sclerosis. Mult Scler. 2011;17(9):1041-1054.
42. Motl RW, Gosney JL. Effect of exercise training on quality of life in multiple
sclerosis: a meta-analysis. Mult Scler. 2008;14(1):129-135.
43. Bae M, Kasser SL. High intensity exercise training on functional outcomes
in persons with multiple sclerosis: A systematic review. Mult Scler Relat
Disord. 2023;75:104748.
44. Andreu-Caravaca L, Ramos-Campo DJ, Chung LH, Martínez-Rodríguez A,
Rubio-Arias J. Effects and optimal dosage of resistance training on
strength, functional capacity, balance, general health perception, and
fatigue in people with multiple sclerosis: a systematic review and meta-
analysis. Disabil Rehabil. 2023;45(10):1595-1607.
45. Pedersen BK. Muscle as a secretory organ. Compr Physiol. 2013;3(3):1337-
1362.
46. The European Medicines Agency. Modafinil - referral. 2011;
https://www.ema.europa.eu/en/medicines/human/referrals/modafinil.
47. Tur C. Fatigue Management in Multiple Sclerosis. Curr Treat Options
Neurol. 2016;18(6):26.
48. Dobryakova E, Genova HM, DeLuca J, Wylie GR. The dopamine imbalance
hypothesis of fatigue in multiple sclerosis and other neurological
disorders. Front Neurol. 2015;6:52.
49. Du Rietz E, Brikell I, Butwicka A, et al. Mapping phenotypic and aetiological
associations between ADHD and physical conditions in adulthood in
Sweden: a genetically informed register study. Lancet Psychiatry.
2021;8(9):774-783.
50. Kurtzke JF. Rating neurologic impairment in multiple sclerosis: an
expanded disability status scale (EDSS). Neurology. 1983;33(11):1444-1452.
51. Benedict RH, DeLuca J, Phillips G, LaRocca N, Hudson LD, Rudick R. Validity
of the Symbol Digit Modalities Test as a cognition performance outcome
measure for multiple sclerosis. Mult Scler. 2017;23(5):721-733.
52. Benedict RH, Fischer JS, Archibald CJ, et al. Minimal neuropsychological
assessment of MS patients: a consensus approach. Clin Neuropsychol.
2002;16(3):381-397.
53. Food and Drug Administration. Guidance for industry: patient-reported
outcome measures: use in medical product development to support
labeling claims. Food and Drug Administration;2009.

100
54. Hobart J, Lamping D, Fitzpatrick R, Riazi A, Thompson A. The Multiple
Sclerosis Impact Scale (MSIS-29): a new patient-based outcome measure.
Brain. 2001;124(Pt 5):962-973.
55. Rabin R, de Charro F. EQ-5D: a measure of health status from the EuroQol
Group. Ann Med. 2001;33(5):337-343.
56. Cohen ET, Matsuda PN, Fritz NE, et al. Self-Report Measures of Fatigue for
People With Multiple Sclerosis: A Systematic Review. J Neurol Phys Ther.
2023;00:1-9.
57. Krupp LB, LaRocca NG, Muir-Nash J, Steinberg AD. The fatigue severity
scale. Application to patients with multiple sclerosis and systemic lupus
erythematosus. Arch Neurol. 1989;46(10):1121-1123.
58. Penner IK, Raselli C, Stocklin M, Opwis K, Kappos L, Calabrese P. The Fatigue
Scale for Motor and Cognitive Functions (FSMC): validation of a new
instrument to assess multiple sclerosis-related fatigue. Mult Scler.
2009;15(12):1509-1517.
59. Mokkink LB, Terwee CB, Patrick DL, et al. The COSMIN checklist for
assessing the methodological quality of studies on measurement
properties of health status measurement instruments: an international
Delphi study. Qual Life Res. 2010;19(4):539-549.
60. Elbers RG, Rietberg MB, van Wegen EE, et al. Self-report fatigue
questionnaires in multiple sclerosis, Parkinson's disease and stroke: a
systematic review of measurement properties. Qual Life Res.
2012;21(6):925-944.
61. Close J, Vandercappellen J, King M, Hobart J. Measuring Fatigue in Multiple
Sclerosis: There may be Trouble Ahead. Neurol Ther. 2023;12(5):1649-1668.
62. Amtmann D, Bamer AM, Noonan V, Lang N, Kim J, Cook KF. Comparison of
the psychometric properties of two fatigue scales in multiple sclerosis.
Rehabil Psychol. 2012;57(2):159-166.
63. Greene N, Quéré S, Bury DP, et al. Establishing clinically meaningful within-
individual improvement thresholds for eight patient-reported outcome
measures in people with relapsing-remitting multiple sclerosis. J Patient
Rep Outcomes. 2023;7(1):61.
64. Berrigan LI, Fisk JD, Patten SB, et al. Health-related quality of life in multiple
sclerosis: Direct and indirect effects of comorbidity. Neurology.
2016;86(15):1417-1424.
65. Krupp LB, Serafin DJ, Christodoulou C. Multiple sclerosis-associated
fatigue. Expert Rev Neurother. 2010;10(9):1437-1447.
66. Patrick E, Christodoulou C, Krupp LB. Longitudinal correlates of fatigue in
multiple sclerosis. Mult Scler. 2009;15(2):258-261.

101
67. Johansson S, Ytterberg C, Hillert J, Widén Holmqvist L, von Koch L. A
longitudinal study of variations in and predictors of fatigue in multiple
sclerosis. J Neurol Neurosurg Psychiatry. 2008;79(4):454-457.
68. Bakshi R, Shaikh ZA, Miletich RS, et al. Fatigue in multiple sclerosis and its
relationship to depression and neurologic disability. Mult Scler.
2000;6(3):181-185.
69. Moore H, Nair KPS, Baster K, Middleton R, Paling D, Sharrack B. Fatigue in
multiple sclerosis: A UK MS-register based study. Mult Scler Relat Disord.
2022;64:103954.
70. Valentine TR, Alschuler KN, Ehde DM, Kratz AL. Prevalence, co-occurrence,
and trajectories of pain, fatigue, depression, and anxiety in the year
following multiple sclerosis diagnosis. Mult Scler. 2022;28(4):620-631.
71. Morrow SA, Weinstock-Guttman B, Munschauer FE, Hojnacki D, Benedict
RH. Subjective fatigue is not associated with cognitive impairment in
multiple sclerosis: cross-sectional and longitudinal analysis. Mult Scler.
2009;15(8):998-1005.
72. Andreasen AK, Spliid PE, Andersen H, Jakobsen J. Fatigue and processing
speed are related in multiple sclerosis. Eur J Neurol. 2010;17(2):212-218.
73. Bol Y, Duits AA, Hupperts RM, Vlaeyen JW, Verhey FR. The psychology of
fatigue in patients with multiple sclerosis: a review. J Psychosom Res.
2009;66(1):3-11.
74. Lerdal A, Celius EG, Krupp L, Dahl AA. A prospective study of patterns of
fatigue in multiple sclerosis. Eur J Neurol. 2007;14(12):1338-1343.
75. Wood B, van der Mei IA, Ponsonby AL, et al. Prevalence and concurrence of
anxiety, depression and fatigue over time in multiple sclerosis. Mult Scler.
2013;19(2):217-224.
76. Téllez N, Río J, Tintoré M, Nos C, Galán I, Montalban X. Fatigue in multiple
sclerosis persists over time: a longitudinal study. J Neurol.
2006;253(11):1466-1470.
77. Koch M, Uyttenboogaart M, van Harten A, Heerings M, De Keyser J. Fatigue,
depression and progression in multiple sclerosis. Mult Scler.
2008;14(6):815-822.
78. Salter A, Keenan A, Le HH, et al. Severity and worsening of fatigue among
individuals with multiple sclerosis. Mult Scler J Exp Transl Clin.
2023;9(2):20552173231167079.
79. Patti F, Amato MP, Trojano M, et al. Quality of life, depression and fatigue in
mildly disabled patients with relapsing-remitting multiple sclerosis
receiving subcutaneous interferon beta-1a: 3-year results from the
COGIMUS (COGnitive Impairment in MUltiple Sclerosis) study. Mult Scler.
2011;17(8):991-1001.

102
80. Kunkel A, Fischer M, Faiss J, Dahne D, Kohler W, Faiss JH. Impact of
natalizumab treatment on fatigue, mood, and aspects of cognition in
relapsing-remitting multiple sclerosis. Front Neurol. 2015;6:97.
81. Masingue M, Debs R, Maillart E, et al. Fatigue evaluation in fingolimod
treated patients: An observational study. Mult Scler Relat Disord.
2017;14:8-11.
82. Kallmann BA, Tiel-Wilck K, Kullmann JS, Engelmann U, Chan A. Real-life
outcomes of teriflunomide treatment in patients with relapsing multiple
sclerosis: TAURUS-MS observational study. Ther Adv Neurol Disord.
2019;12:1756286419835077.
83. Melanson M, Grossberndt A, Klowak M, et al. Fatigue and cognition in
patients with relapsing multiple sclerosis treated with interferon β. Int J
Neurosci. 2010;120(10):631-640.
84. Putzki N, Yaldizli O, Tettenborn B, Diener HC. Multiple sclerosis associated
fatigue during natalizumab treatment. J Neurol Sci. 2009;285(1-2):109-113.
85. van Ballegooijen H, van der Hiele K, Enzinger C, de Voer G, Visser LH. The
longitudinal relationship between fatigue, depression, anxiety, disability,
and adherence with cognitive status in patients with early multiple
sclerosis treated with interferon beta-1a. eNeurologicalSci.
2022;28:100409.
86. Jongen PJ, Lehnick D, Koeman J, et al. Fatigue and health-related quality of
life in relapsing-remitting multiple sclerosis after 2 years glatiramer
acetate treatment are predicted by changes at 6 months: an
observational multi-center study. J Neurol. 2014;261(8):1469-1476.
87. Ziemssen T, Hoffman J, Apfel R, Kern S. Effects of glatiramer acetate on
fatigue and days of absence from work in first-time treated relapsing-
remitting multiple sclerosis. Health Qual Life Outcomes. 2008;6:67.
88. Iaffaldano P, Viterbo RG, Paolicelli D, et al. Impact of natalizumab on
cognitive performances and fatigue in relapsing multiple sclerosis: a
prospective, open-label, two years observational study. PLoS One.
2012;7(4):e35843.
89. Svenningsson A, Falk E, Celius EG, et al. Natalizumab treatment reduces
fatigue in multiple sclerosis. Results from the TYNERGY trial; a study in the
real life setting. PLoS One. 2013;8(3):e58643.
90. Wilken J, Kane RL, Sullivan CL, et al. Changes in Fatigue and Cognition in
Patients with Relapsing Forms of Multiple Sclerosis Treated with
Natalizumab: The ENER-G Study. Int J MS Care. 2013;15(3):120-128.
91. Harrison AM, Safari R, Mercer T, et al. Which exercise and behavioural
interventions show most promise for treating fatigue in multiple sclerosis?
A network meta-analysis. Mult Scler. 2021;27(11):1657-1678.

103
92. Pucci E, Branãs P, D'Amico R, Giuliani G, Solari A, Taus C. Amantadine for
fatigue in multiple sclerosis. Cochrane Database Syst Rev.
2007;2007(1):Cd002818.
93. Sheng P, Hou L, Wang X, et al. Efficacy of modafinil on fatigue and
excessive daytime sleepiness associated with neurological disorders: a
systematic review and meta-analysis. PLoS One. 2013;8(12):e81802.
94. Zifko UA, Rupp M, Schwarz S, Zipko HT, Maida EM. Modafinil in treatment of
fatigue in multiple sclerosis. Results of an open-label study. J Neurol.
2002;249(8):983-987.
95. Nagels G, D'Hooghe M B, Vleugels L, Kos D, Despontin M, De Deyn PP. P300
and treatment effect of modafinil on fatigue in multiple sclerosis. J Clin
Neurosci. 2007;14(1):33-40.
96. Brioschi A, Gramigna S, Werth E, et al. Effect of modafinil on subjective
fatigue in multiple sclerosis and stroke patients. Eur Neurol.
2009;62(4):243-249.
97. Stankoff B, Waubant E, Confavreux C, et al. Modafinil for fatigue in MS: a
randomized placebo-controlled double-blind study. Neurology.
2005;64(7):1139-1143.
98. Möller F, Poettgen J, Broemel F, Neuhaus A, Daumer M, Heesen C. HAGIL
(Hamburg Vigil Study): a randomized placebo-controlled double-blind
study with modafinil for treatment of fatigue in patients with multiple
sclerosis. Mult Scler. 2011;17(8):1002-1009.
99. Ford-Johnson L, DeLuca J, Zhang J, Elovic E, Lengenfelder J, Chiaravalloti
ND. Cognitive effects of modafinil in patients with multiple sclerosis: A
clinical trial. Rehabil Psychol. 2016;61(1):82-91.
100. Ledinek AH, Sajko MC, Rot U. Evaluating the effects of amantadin, modafinil
and acetyl-L-carnitine on fatigue in multiple sclerosis--result of a pilot
randomized, blind study. Clin Neurol Neurosurg. 2013;115 Suppl 1:S86-89.
101. Nourbakhsh B, Revirajan N, Morris B, et al. Safety and efficacy of
amantadine, modafinil, and methylphenidate for fatigue in multiple
sclerosis: a randomised, placebo-controlled, crossover, double-blind trial.
Lancet Neurol. 2021;20(1):38-48.
102. The Canadian MS Research Group. A randomized controlled trial of
amantadine in fatigue associated with multiple sclerosis. The Canadian MS
Research Group. Can J Neurol Sci. 1987;14(3):273-278.
103. Cohen RA, Fisher M. Amantadine treatment of fatigue associated with
multiple sclerosis. Arch Neurol. 1989;46(6):676-680.
104. Murray TJ. Amantadine therapy for fatigue in multiple sclerosis. Can J
Neurol Sci. 1985;12(3):251-254.

104
105. Ashtari F, Fatehi F, Shaygannejad V, Chitsaz A. Does amantadine have
favourable effects on fatigue in Persian patients suffering from multiple
sclerosis? Neurol Neurochir Pol. 2009;43(5):428-432.
106. Krupp LB, Coyle PK, Doscher C, et al. Fatigue therapy in multiple sclerosis:
results of a double-blind, randomized, parallel trial of amantadine,
pemoline, and placebo. Neurology. 1995;45(11):1956-1961.
107. Weinshenker BG, Penman M, Bass B, Ebers GC, Rice GP. A double-blind,
randomized, crossover trial of pemoline in fatigue associated with multiple
sclerosis. Neurology. 1992;42(8):1468-1471.
108. Natsheh JY, DeLuca J, Costa SL, Chiaravalloti ND, Dobryakova E.
Methylphenidate may improve mental fatigue in individuals with multiple
sclerosis: A pilot clinical trial. Mult Scler Relat Disord. 2021;56:103273.
109. Ludvigsson JF, Otterblad-Olausson P, Pettersson BU, Ekbom A. The
Swedish personal identity number: possibilities and pitfalls in healthcare
and medical research. Eur J Epidemiol. 2009;24(11):659-667.
110. Hillert J, Stawiarz L. The Swedish MS registry - clinical support tool and
scientific resource. Acta Neurol Scand. 2015;132(199):11-19.
111. Piehl F, Alping P, Virtanen S, et al. COMBAT-MS: A Population-Based
Observational Cohort Study Addressing the Benefit-Risk Balance of
Multiple Sclerosis Therapies Compared with Rituximab. Ann Neurol. 2024.
112. Alping P, Piehl F, Langer-Gould A, Frisell T, COMBAT-MS Study Group.
Validation of the Swedish Multiple Sclerosis Register: Further Improving a
Resource for Pharmacoepidemiologic Evaluations. Epidemiology.
2019;30(2):230-233.
113. Everhov Å H, Frisell T, Osooli M, et al. Diagnostic accuracy in the Swedish
national patient register: a review including diagnoses in the outpatient
register. Eur J Epidemiol. 2025.
114. Wettermark B, Hammar N, Fored CM, et al. The new Swedish Prescribed
Drug Register--opportunities for pharmacoepidemiological research and
experience from the first six months. Pharmacoepidemiol Drug Saf.
2007;16(7):726-735.
115. Ludvigsson JF, Almqvist C, Edstedt Bonamy A-K, et al. Registers of the
Swedish total population and their use in medical research. Eur J
Epidemiol. 2016;31(2):125-136.
116. Ludvigsson JF, Svedberg P, Olén O, Bruze G, Neovius M. The longitudinal
integrated database for health insurance and labour market studies (LISA)
and its use in medical research. Eur J Epidemiol. 2019;34(4):423-437.
117. Nagin DS, Odgers CL. Group-based trajectory modeling in clinical
research. Annu Rev Clin Psychol. 2010;6:109-138.

105
118. Alping P. Pharmacoepidemiological studies of rituximab and other recent
therapies in multiple sclerosis, Karolinska Institutet; 2022.
119. Romio SA, Bellocco R, Corrao G. Introductive remarks on causal inference.
Statistica. 2010;70(3):353-362.
120. Yoshida K, Solomon DH, Kim SC. Active-comparator design and new-user
design in observational studies. Nat Rev Rheumatol. 2015;11(7):437-441.
121. Khullar D, Jena AB. "Natural Experiments" in Health Care Research. JAMA
Health Forum. 2021;2(6):e210290.
122. Linden A, Adams JL. Applying a propensity score-based weighting model
to interrupted time series data: improving causal inference in programme
evaluation. J Eval Clin Pract. 2011;17(6):1231-1238.
123. Langholz B, Goldstein L. Risk set sampling in epidemiologic cohort studies.
Statistical Science. 1996;11:35-53.
124. Austin PC, Stuart EA. Moving towards best practice when using inverse
probability of treatment weighting (IPTW) using the propensity score to
estimate causal treatment effects in observational studies. Stat Med.
2015;34(28):3661-3679.
125. White IR, Royston P, Wood AM. Multiple imputation using chained
equations: Issues and guidance for practice. Stat Med. 2011;30(4):377-399.
126. D'Hooghe M, Van Gassen G, Kos D, et al. Improving fatigue in multiple
sclerosis by smartphone-supported energy management: The MS
TeleCoach feasibility study. Mult Scler Relat Disord. 2018;22:90-96.
127. Englund S, Frisell T, Qu Y, et al. Linking fatigue trajectories in RRMS to
incidence of psychiatric comorbidities and productivity loss in the years
following DMT initiation [Poster presentation]. 9th Joint ECTRIMS-ACTRIMS
Meeting; 2023; Milan.
128. Song M. Trajectory analysis in obesity epidemiology: a promising life
course approach. Curr Opin Endocr Metab Res. 2019;4:37-41.
129. American College of Sports Medicine. Progression models in resistance
training for healthy adults. Med Sci Sports Exerc. 2009;41(3):687-708.
130. Zigmond AS, Snaith RP. The hospital anxiety and depression scale. Acta
Psychiatr Scand. 1983;67(6):361-370.
131. Eriksson G, Tham K, Kottorp A. A cross-diagnostic validation of an
instrument measuring participation in everyday occupations: the
Occupational Gaps Questionnaire (OGQ). Scand J Occup Ther.
2013;20(2):152-160.
132. Herdman M, Gudex C, Lloyd A, et al. Development and preliminary testing
of the new five-level version of EQ-5D (EQ-5D-5L). Qual Life Res.
2011;20(10):1727-1736.

106
133. Assarsson E, Lundberg M, Holmquist G, et al. Homogenous 96-plex PEA
immunoassay exhibiting high sensitivity, specificity, and excellent
scalability. PLoS One. 2014;9(4):e95192.
134. Rooney S, Wood L, Moffat F, Paul L. Prevalence of fatigue and its
association with clinical features in progressive and non-progressive
forms of Multiple Sclerosis. Mult Scler Relat Disord. 2019;28:276-282.
135. Vachon DD, Krueger RF, Irons DE, Iacono WG, McGue M. Are Alcohol
Trajectories a Useful Way of Identifying At-Risk Youth? A Multiwave
Longitudinal-Epidemiologic Study. J Am Acad Child Adolesc Psychiatry.
2017;56(6):498-505.
136. Metz LM, Patten SB, Archibald CJ, et al. The effect of immunomodulatory
treatment on multiple sclerosis fatigue. J Neurol Neurosurg Psychiatry.
2004;75(7):1045-1047.
137. Yildiz M, Tettenborn B, Putzki N. Multiple sclerosis-associated fatigue
during disease-modifying treatment with natalizumab, interferon-beta
and glatiramer acetate. Eur Neurol. 2011;65(4):231-232.
138. Broch L, Flemmen H, Simonsen CS, et al. "No association between disease
modifying treatment and fatigue in multiple sclerosis". Mult Scler Relat
Disord. 2023;79:104993.
139. Cruz Rivera S, Aiyegbusi OL, Piani Meier D, et al. The effect of disease
modifying therapies on fatigue in multiple sclerosis. Mult Scler Relat
Disord. 2023;79:105065.
140. Comi G, Leocani L, Rossi P, Colombo B. Physiopathology and treatment of
fatigue in multiple sclerosis. J Neurol. 2001;248(3):174-179.
141. Thielscher C, Thielscher S, Kostev K. The risk of developing depression
when suffering from neurological diseases. Ger Med Sci. 2013;11:Doc02.
142. Landfeldt E, Castelo-Branco A, Svedbom A, Löfroth E, Kavaliunas A, Hillert
J. Sick leave and disability pension before and after diagnosis of multiple
sclerosis. Mult Scler. 2016;22(14):1859-1866.
143. Murley C, Karampampa K, Alexanderson K, Hillert J, Friberg E. Diagnosis-
specific sickness absence and disability pension before and after multiple
sclerosis diagnosis: An 8-year nationwide longitudinal cohort study with
matched references. Mult Scler Relat Disord. 2020;42:102077.
144. Akbar N, Sandroff BM, Wylie GR, et al. Progressive resistance exercise
training and changes in resting-state functional connectivity of the
caudate in persons with multiple sclerosis and severe fatigue: A proof-of-
concept study. Neuropsychol Rehabil. 2020;30(1):54-66.
145. Gomez-Illan R, Reina R, Barbado D, Sabido R, Moreno-Navarro P, Roldan A.
Effects of Maximal Strength Training on Perceived-Fatigue and Functional

107
 Mobility in Persons with Relapsing-Remitting Multiple Sclerosis. Medicina
(Kaunas). 2020;56(12).
146. Dodd KJ, Taylor NF, Shields N, Prasad D, McDonald E, Gillon A. Progressive
resistance training did not improve walking but can improve muscle
performance, quality of life and fatigue in adults with multiple sclerosis: a
randomized controlled trial. Mult Scler. 2011;17(11):1362-1374.
147. Kierkegaard M, Lundberg IE, Olsson T, et al. High-intensity resistance
training in multiple sclerosis - An exploratory study of effects on immune
markers in blood and cerebrospinal fluid, and on mood, fatigue, health-
related quality of life, muscle strength, walking and cognition. J Neurol Sci.
2016;362:251-257.
148. Wong VL, Holahan MR. A systematic review of aerobic and resistance
exercise and inflammatory markers in people with multiple sclerosis.
Behav Pharmacol. 2019;30(8):653-660.
149. Calle MC, Fernandez ML. Effects of resistance training on the inflammatory
response. Nutr Res Pract. 2010;4(4):259-269.
150. Rammohan KW, Rosenberg JH, Lynn DJ, Blumenfeld AM, Pollak CP,
Nagaraja HN. Efficacy and safety of modafinil (Provigil) for the treatment of
fatigue in multiple sclerosis: a two centre phase 2 study. J Neurol
Neurosurg Psychiatry. 2002;72(2):179-183.
151. Bourdette D. Are drugs for multiple sclerosis fatigue just placebos? Lancet
Neurol. 2021;20(1):20-21.
152. Runia TF, Jafari N, Siepman DA, Hintzen RQ. Fatigue at time of CIS is an
independent predictor of a subsequent diagnosis of multiple sclerosis. J
Neurol Neurosurg Psychiatry. 2015;86(5):543-546.
153. Hanken K, Eling P, Hildebrandt H. Is there a cognitive signature for MS-
related fatigue? Mult Scler. 2015;21(4):376-381.
154. American Psychiatric Association. Diagnostic and statistical manual of
mental disorders: DSM-5™, 5th ed. Arlington, VA, US: American Psychiatric
Publishing, Inc.; 2022.
155. Brenner P, Piehl F. Fatigue and depression in multiple sclerosis:
pharmacological and non-pharmacological interventions. Acta Neurol
Scand. 2016;134 (Suppl. 200):47-54.
156. Kuppuswamy A. The fatigue conundrum. Brain. 2017;140(8):2240-2245.
157. van der Werf SP, Evers A, Jongen PJ, Bleijenberg G. The role of
helplessness as mediator between neurological disability, emotional
instability, experienced fatigue and depression in patients with multiple
sclerosis. Mult Scler. 2003;9(1):89-94.

108
158. Smith C, Olson K, Hale LA, Baxter D, Schneiders AG. How does fatigue
influence community-based exercise participation in people with multiple
sclerosis? Disabil Rehabil. 2011;33(23-24):2362-2371.
159. Teljas C, Boström I, Marrie RA, et al. Validating the diagnosis of multiple
sclerosis using Swedish administrative data in Värmland County. Acta
Neurol Scand. 2021;144(6):680-686.

109