R e s e a rc h M e t h o d o l o g y i n

R e c u r re n t Preg n a n c y L o s s
a,b,
Ole B. Christiansen, MD, DMSc *

KEYWORDS
 Recurrent pregnancy loss  Recurrent miscarriage  Medical research
 Research studies  Methodological flaws

KEY POINTS
 There is currently substantial disagreement concerning the diagnostic criteria for recurrent
pregnancy loss (RPL), which renders comparisons between research studies in the area
difficult.
 There are numerous methodological pitfalls that threaten the validity of research studies in
the field of RPL and it is necessary for scientists and clinicians to be aware of them.
 Some of the methodological pitfalls are common for medical research in general, whereas
some are specific for RPL research.
 Frequently seen methodological flaws in case-control studies are comparisons of bio-
markers between patients with RPL and controls that differ from patients with regard to
previous ongoing pregnancies, relevant endocrine factors, and viability of the fetal tissue
at the time of sampling.
 In cohort studies, incomplete follow-up of patients in many studies has resulted in huge
variations in estimates of the prognosis after RPL.
 Only a few small and heterogeneous double-blinded placebo-controlled trials of treat-
ments of RPL have been carried out with very heterogeneous results.
 Proposals are given for improvements in the design of research studies in RPL that hope-
fully can improve the quality of studies in the future.

INTRODUCTION

Compared with the situation in other reproductive medicine disorders, such as tubal or
male factor infertility and in other areas of medicine, there is very little consensus
about which investigations are useful for identifying causes or estimating the prog-
nosis and which treatments are effective in recurrent pregnancy loss (RPL). It is gener-
ally agreed that when the tubes are occluded, as diagnosed by laparoscopy or

a
Fertility Clinic 4071, Copenhagen University Hospital Rigshospitalet, Blegdamsvej 9, DK-2100,
Copenhagen, Denmark; b Department of Obstetrics and Gynecology, Aalborg University Hospi-
tal, Reberbansgade, DK-9000, Aalborg, Denmark
* Fertility Clinic 4071, Copenhagen University Hospital Rigshospitalet, Blegdamsvej 9, DK-2100,
Copenhagen, Denmark.
E-mail address: [email protected]

Obstet Gynecol Clin N Am 41 (2014) 19–39

http://dx.doi.org/10.1016/j.ogc.2013.10.001 obgyn.theclinics.com
0889-8545/14/$ – see front matter Ó 2014 Elsevier Inc. All rights reserved.
20 Christiansen

hysterosalpingography (HSG), pregnancy can happen only after in vitro fertilization

(IVF), and when the number of viable spermatozoa is very low, pregnancy can happen
only after intracytoplasmatic sperm injection (ICSI). It is also generally agreed that
IVF or ICSI are very efficient treatment methods for the 2 reproductive disorders. In
recurrent early pregnancy loss there is much more disagreement about diagnosis,
cause, and treatments. Although most guidelines from specialist societies do not sup-
port the screening of RPL women for hereditary thrombophilia factors or peripheral
blood or endometrial natural killer (NK) cell numbers and function,1–3 many clinics
are still doing this, and whereas the Cochrane review4 or national guidelines do not
recommend immunotherapy1,2 or preimplantation genetic screening (PGS)5 for RPL,
immunotherapy and PGS are still widely used in many clinics.
There could be many reasons why doctors very often do not adhere to the clinical
guidelines regarding RPL:
 Pressure from desperate patients to do something although very few proven
therapies really exist
 The doctors’ economical motives, as many patients are desperate and willing to
pay a lot of money for treatments that may provide them with some hope for a
solution to their problem
 Current guidelines are based on few, small, and often poor-quality studies that
cannot support strong, evidence-based recommendations
In this author’s view, the third statement is the most important cause for this poor
adherence to RPL guidelines. With regard to almost every diagnostic test or treatment
for RPL, it is possible to find studies presenting data strongly in favor of this test or
treatment and other studies strongly against. It is therefore often up to clinicians them-
selves to decide which studies they find trustworthy.
The aim of this article is to highlight pitfalls in research methodology that may
explain why studies in RPL often provide very divergent results, and it is hoped that
insight in this issue may help clinicians to decide which published studies are most
valid. It may help researchers to eliminate methodological flaws in future studies,
which may hopefully come to some kind of agreement about the usefulness of diag-
nostic tests and treatments in RPL.

CONTROVERSIES OF DEFINITION

It is disputed how to define RPL. It is important to realize that RPL is defined quite
differently from most other diseases. Most diseases are defined by some unique path-
oanatomical, clinical, or paraclinical findings being permanently present, whereas RPL
is defined by a series of transient events in the past that may have be poorly registered.
The controversies concerning how to define RPL deal with
 The number of miscarriages needed for the diagnosis
 The role of nonconsecutive miscarriages
 The role of preclinical losses
Until 10 years ago, the definition of RPL was undisputedly 3 or more consecutive
miscarriages, because it was commonly agreed that after 3 miscarriages the chance
of live birth the next pregnancy without treatment was substantially decreased.6 How-
ever, during the recent years, some national guidelines have adopted an RPL definition
of only 2 clinically recognized miscarriages1 or 2 not necessarily consecutive miscar-
riages.7 This redefinition is based on finding similar frequencies of selected factors
suggested to cause RPL: uterine abnormalities; antiphosphospholipid antibodies
 Research Methodology in Recurrent Pregnancy Loss 21

(APL); parental chromosome aberrations; or the factor V Leiden mutation in women

with 2, women with 3, and women with more miscarriages.7,8 It is argued that when
such risk factors already recommended in the screening of couples with 3 miscar-
riages can be found with similar prevalence in those with 2 miscarriages, they should
also be examined in the latter. If the same tests are recommended in couples with 2 as
well as those with 3 or more miscarriages, it is a short step to redefine RPL as 2 or
more miscarriages. There is also now disagreement regarding which kind of preg-
nancy losses should be included in the criteria for RPL. Thirty years ago, because
of the nonexistence of ultrasonic examinations and high sensitive pregnancy tests,
pregnancies could not be diagnosed before gestational week (GW) 6 to 7; therefore,
the pregnancy losses considered in the RPL diagnosis were miscarriages, which
had normally been confirmed by curettage and histology. Pregnancies can now, owing
to highly sensitive and specific b-human chorionic gonadotropin (hCG) tests, be diag-
nosed a few days after the due menstrual period, and many of these (biochemical
pregnancies) will fail before it is possible to do transvaginal ultrasound. There is
thus an urgent need to find a place for these kinds of losses in the RPL diagnosis.
Because transiently positive pregnancy tests at the time of the due period are a
frequent finding in women not using anticonception,9 many gynecologists have
been reluctant to include biochemical pregnancies in the RPL diagnosis, and the
American Society for Reproductive Medicine (ASRM) definition of RPL (2 or more clin-
ical miscarriages) completely disregards them.1 A recent study from the European So-
ciety for Human Reproduction and Embryology (ESHRE) early pregnancy special
interest group on the other side found that in patients with RPL, each early pregnancy
loss confirmed only by a b-hCG test displays a negative prognostic impact equal to
that of a clinical miscarriage, supporting the view that biochemical pregnancies should
be included in the RPL diagnosis.10 The different diagnostic criteria recommended in
the national guidelines or by leading RPL clinics are1,2,7,10 as follows:
 3 consecutive pregnancy losses before 24 weeks of gestation
 2 consecutive clinical pregnancy losses
 3 consecutive clinical miscarriages and biochemical pregnancies
 2 not necessarily consecutive pregnancy losses before 24 weeks of gestation
In numerous studies, it was found that the strongest predictive factor for new
miscarriage in patients with RPL is the number of previous losses.6,11 Because the
same diagnosis, RPL, for the time being covers patients with a wide range of previ-
ous losses and therefore very different pregnancy prognoses, it will in the future be
increasingly difficult to compare and combine different studies of outcome in pa-
tients with RPL (eg, results of randomized controlled trials [RCTs]). The only way
to overcome this obstacle is that the investigators in such studies stratify the results
according to the number of previous clinical miscarriages, as well as biochemical
pregnancies or pregnancies of unknown location (PULs).

TYPES OF RESEARCH STUDIES

Different pitfalls characterize the 3 main types of research studies done in RPL:
 Case-control studies
 Cohort studies
 Intervention and treatment studies
For each category of studies, I provide an overview of the pitfalls that threaten the
validity of the studies and the flaws often seen in publications: some of them can be
22 Christiansen

seen in other areas of medical research and they are discussed superficially, whereas
others that are specific for RPL research are discussed in more detail.

CASE-CONTROL STUDIES

Case-control studies always have a retrospective design. The frequency of a poten-

tial risk factor is investigated in a group of patients who have been sampled during a
defined period (typically in a single clinic) with a specific disease diagnosis and
compared with the corresponding frequency in a group of randomly selected individ-
uals either without the disease or (if the disease is rare) selected from the back-
ground population. An estimate of the potential risk factors’ association with the
disease is typically given by the odds ratio (OR) with 95% confidence limits indicating
the ratio between the frequency of the risk factor in diseased individuals and in
controls.
Methodological errors in case-control studies can occur during the sampling of both
patients and control subjects and errors can occur in the testing for potential risk
factors.
Flaws in Sampling of Patients and Controls
Inconsistent diagnosis
The different definitions of RPL have already been discussed. Different RPL definitions
will make comparisons between case-control studies originating from regions with
different definitions increasingly difficult in the future.
Misclassification of disease/outcome status
Even if identical RPL definitions are used, patients in different studies can differ
regarding the severity of the disease: the number of previous miscarriages. Because
the RPL diagnosis is based on a series of past events, the validity of the diagnosis is
dependent on the quality of the information that is available about these events. In
many clinics, information about previous pregnancy losses comes primarily from inter-
viewing the patients. Misclassification of disease/outcome status (the number of pre-
vious miscarriages) can be random or nonrandom. There are many examples of
random outcome misclassification when dealing with miscarriages. Only 71% of mis-
carriages reported by women without RPL who have been treated at hospital could
be verified in hospital records,12 and in a retrospective study, 348 women recalled
30 (6%) of 507 miscarriages that were not reported in a prospective study several
years before.13 Random outcome misclassification results in an underestimate of
the hypothesized association between exposure and disease/outcome in case-
control studies (Table 1).
Retrospective information about previous pregnancy losses can also be subjected
to nonrandom misclassification, also called recall or information bias. This misclassi-
fication is a difference in the ability or inclination to remember or report events or ex-
posures in the past in individuals with or without particular characteristics. Women
who had given birth to a child with a congenital malformation will search their memory
extensively for any potentially teratogenic exposures during pregnancy and therefore
retrospectively report more exposures than women who had delivered a healthy child,
although there is no real difference in the frequency of harmful exposures in the
2 groups. Some women with 1 or 2 confirmed miscarriages will be more prone to inter-
pret delayed menstruations or recall previous terminations as miscarriages than
women without recent pregnancy losses, and thus erroneously get a diagnosis of
RPL. Such nonrandom misclassification of disease/outcome status will result in either
an overestimation or underestimation (see Table 1) of the true size of the hypothesized
 Research Methodology in Recurrent Pregnancy Loss 23

Table 1
Important methodological pitfalls typical for research studies in RPL

Factor to Evaluate Effect on Study Outcome

Definition of RPL 2 miscarriages Decreases difference between risk variables in
CCS and treatment effect in RCTs
Random misclassification Decreases difference between risk variables in CCS
Nonrandom misclassification Decreases or increases difference between risk
variables in CCS or outcome variables in CS
Ascertainment bias Increases difference between risk variables in CCS
Relevant mismatches between Increases difference between risk variables in CCS
patients and controls
Lack of protocol details/multiple Overestimates significance of chance findings
testing
Historical controls Increases effects in treatment studies
Nonblinding Increases or decreases treatment effects in RCTs
Premature termination after interim Decreases treatment effect in RCTs
analysis
Inclusion after detection of fetal heart Decreases treatment effect in RCTs
action
Poor characterization of RPL and Renders comparisons between CCS and RCTs
subgroups of RPL difficult and makes meta-analysis difficult
Unfounded exclusions of RCTs in Bias combined risk estimates in meta-analyses
systematic reviews

Abbreviations: CCS, case-control study; RCT, randomized controlled trial; RPL, recurrent pregnancy
loss.

relationship between exposure and disease/outcome (RPL). To avoid random and

nonrandom misclassification in RPL case-control studies, information about previous
pregnancies should as much as possible be confirmed from external sources: records
from hospitals, fertility clinics, and practitioners and serum hCG measurements should
be documented from laboratory reports.

Ascertainment bias
In case-control studies, the frequency of a potential risk factor for the disorder under
study (eg, RPL) is compared between patients and controls. Ascertainment or selec-
tion bias happens when patients with some clinical or paraclinical risk factor are pref-
erentially referred to a specific clinic because of knowledge of the clinic’s expertise
or interest, and a study focusing on this particular risk factor is undertaken in the
clinic. An example of ascertainment bias in RPL has been reported by Out and col-
leagues.14 In this study, the frequency of the APLs anticardiolipin and lupus antico-
agulant was higher in patients with RPL referred to a Dutch center for APL research
than in controls. However, when patients with RPL with a history of thromboembolic
or lupuslike symptoms were excluded, the prevalence of APL in the remaining pa-
tients with RPL did not differ from that of controls. The high prevalence of APL in
the total group of patients resulted from the preferential referral of patients with
RPL with APL-associated symptoms to the clinic due to its special expertise (see
Table 1).
Ascertainment bias can also work in controls. In studies of risk factors for adverse
pregnancy outcome, information is often obtained or blood samples drawn from
women with healthy pregnancies coming for routine pregnancy control. Women giving
24 Christiansen

informed consent to become controls are often better educated than average women,
thereby decreasing the occurrence of lifestyle factors and exposures that may be
harmful in pregnancy. When staff members or healthy blood donors are used as con-
trols, these also undergo positive selection for being healthier than the average
population.

Flaws in Estimating Risk Factors

Misclassification of exposure status
If the risk factor(s) under study is a potentially harmful exposure (eg, infection, smok-
ing, medicamentation) in which the estimate of exposure is based on information
retrospectively obtained from the patients and controls themselves; as previously dis-
cussed, this information can be subject to both random and nonrandom misclassifica-
tion. If exposures have the same probability of being overreported/underreported in
patients and controls, we are dealing with random misclassification and this will
lead to an underestimate of the hypothesized relationship between exposure and dis-
ease/outcome. When the probability of exposure misclassification differs between pa-
tients and controls, nonrandom misclassification occurs. An example of nonrandom
misclassification is when patients with RPL often recall an episode of fever during a
pregnancy that subsequently failed, whereas controls with successful pregnancies
report such episodes less often, although the incidence of fever in the 2 groups
may be similar. Nonrandom misclassification of exposures can lead to an underesti-
mate or overestimate of the true size of association between exposure and disease/
outcome. To avoid misclassification in RPL case-control studies, data about expo-
sures should be collected as much as possible from external sources: records from
hospitals or practitioners or be collected before the outcome is known: in early preg-
nancy before miscarriage is diagnosed.

Confounding
In case-control studies, a confounding factor is a clinical or paraclinical factor that is
associated with both the risk factor and the disease/outcome under study. If adequate
measures are not taken, a confounding factor can be mistaken as a causal factor or
diminish the estimate of the impact (OR) of the risk factor under study. In RPL studies,
age is an important and common confounding factor. In studies of the prevalence of
autoantibodies in patients with RPL and controls, age is associated with both the risk
of miscarriage and RPL and the occurrence of autoantibodies. Elimination of the con-
founding effect of age during the inclusion phase of such studies can be undertaken by
age-matching patients and controls and in the analysis phase by reporting autoanti-
body frequencies in different age strata of patients and controls and subsequently
do adjustment by multivariate statistical methods.

Mismatch between patient and controls group

In case-control studies, the aim is to compare diseased and healthy individuals, which
is quite straightforward regarding most diseases, but in RPL, research finding a suit-
able control group is a much more complex task (see Table 1; Table 2). Case-control
studies in RPL research typically compare the following:
 Biomarkers in the blood or endometrium of nonpregnant women or in the blood
of pregnant patients with RPL with the same biomarkers in women who do not
have RPL
 Biomarkers in decidual or trophoblast tissue from women with RPL who have
miscarried with the same biomarkers in tissue from non-RPL women who had
an induced abortion.
 Table 2
Suggested optimal control groups for the investigation of biomarkers in patients with RPL or their products of conception

Genetic
Nongenetic Biomarkers Biomarkers
Peripheral Blood/
Peripheral Blood Luteal Phase Endometrium Decidual/Trophoblast Tissue Tissue
RPLa Controlsa RPLb Controlsb RPLc Controlsc RPL Controls

Research Methodology in Recurrent Pregnancy Loss

Primary/NP Nulligravida/NP Primary Nulligravida Primary, Primary RPL All Multipara
Euploid male embryo Aneuploid embryo
Secondary/NP Previous birth/NP Secondary Previous birth Secondary, Secondary RPL
Euploid male embryo Aneuploid embryo
Primary GW5 Primigravida GW5
Secondary GW5 Previous birth GW5
Primary > GW6 Primary RPL > GW6
Euploid male embryo Aneuploid embryo
Secondary > GW6 Secondary RPL > GW6
Euploid male embryo Aneuploid embryo

In each double column, the suggested optimal controls are found to the right of the RPL column.
Abbreviations: GW, gestational week calculated from the first day of last menstrual period; NP, not pregnant; RPL, recurrent pregnancy loss.
a,b,c
Women in each comparable pair should ideally have similar estrogen and progesterone levels in the nonpregnant state and similar hCG, estrogen, and pro-
gesterone levels in the pregnant state.

25
26 Christiansen

In studies of external exposures or genetic biomarkers, the ideal control group for
patients with RPL is women with proven fertility: typically women with 2 or more un-
complicated births and no miscarriages, as genetic polymorphisms will not change
according to reproductive history or be affected by endocrine factors or inflammation
(see Table 2).
However, it is much more complex to identify the ideal controls for the investigation
of nongenetically determined biomarkers in the blood, the endometrium, or tropho-
blast tissue. Some biomarkers, although exhibiting no impact on pregnancy outcome,
may be different in patients with RPL and women with previous births merely because
of their different reproductive histories.
An illustrative example is anti-HLA antibodies and studies of their role in RPL.
Several years ago many articles reported that these antibodies could be detected
much less frequently in the blood of patients with primary RPL (no previous ongoing
pregnancies) than in control women who had previously given birth. It was postulated
that lack of these “blocking antibodies” was a cause of the RPL and that deliberate
immunization of the patients with paternal lymphocytes with the aim to stimulate anti-
body production would improve the pregnancy prognosis. It is now generally recog-
nized that these antibodies are a common feature of normal ongoing pregnancy
due to passage of fetal cells through the placenta into the maternal circulation during
late pregnancy and they often persist for many years.15 Therefore, it is not surprising
that patients with primary RPL normally lack these antibodies, whereas multipara are
often positive.
It has also been shown that cellular immunity is being permanently changed after a
birth. Clones of cytotoxic lymphocytes with specificity for male-specific minor histocom-
patibility (HY) antigens develop in half of the women pregnant with a male fetus and the
anti-HY cytotoxicity remains unchanged up to 18 years after delivery.16 Long-term
persistence of regulatory T cells17 or/and persistence of fetal cells in the maternal circu-
lation (fetal microchimerism) after a first ongoing pregnancy18,19 may be the reason that
most women remain immunologically tolerant to the fetus in spite of production of anti-
bodies and lymphocytes with reactivity toward fetal antigens. The maternal immune sys-
tem therefore recognizes many paternal/fetal alloantigens during an ongoing pregnancy
and this very often induces permanent changes in immune reactivity to the fetus or
trophoblast that may reside in lymphocytes carrying immunologic memory (memory
T cells) in the peripheral blood, endometrium, or regional uterine lymph nodes. It is
also possible, although much less studied, that transcription of messenger RNA
(mRNA) and expression or production of proteins that are not related to the immune
function (eg, receptors for hormones in the uterus, production of coagulation factors
in the liver) can be permanently altered subsequent to the extensive physiologic changes
taking place during a prior ongoing pregnancy (a pregnancy passing GW 22).
Because of these considerations, in RPL case-control studies controls should be
matched to patients with RPL regarding a history of previous ongoing pregnancies
(see Table 2). The ideal control group for patients with RPL who had never had an
ongoing pregnancy would be women with repeated first trimester terminations due to
social reasons; however, because this group is fortunately small, an alternative suitable
control group would be women with no previous pregnancy. It may be argued that 1%
of nulligravida would later experience RPL, but this small error is insignificant compared
with the error associated with comparing nulliparous patients with RPL with multiparous
controls. Patients with secondary RPL who had previously had a birth should of course
be compared with controls with 1 previous live birth and no miscarriages.
Another factor that can confound case-control studies in RPL research and should
be adjusted for as much as possible is differences in hormones relating to
 Research Methodology in Recurrent Pregnancy Loss 27

reproduction and pregnancy. Many factors relating to immune function (eg, cytokines)
and the coagulation system (eg, proteins C and S) are influenced by estrogen and pro-
gesterone levels.20 The level of expression of a series of cytokine mRNA in endometrial
cells increases markedly from the follicular to the late secretory phase, probably influ-
enced by cyclic changes of estrogen and progesterone.21 It has been shown that lym-
phocytes that can induce immunologic tolerance (regulatory T cells) are attracted to
the feto-maternal interface by hCG produced by the trophoblast,22 which may have
profound importance for measurements of immune biomarkers in the uterus during
pregnancy. Because concentrations of hCG, estrogens, progesterone, cortisone,
and pregnancy-associated placental protein A (PAPP-A) change markedly according
to menstrual cycle phase or progressing gestation, the level of biomarkers affected by
hormones is dependent on phase of menstrual cycle or time of gestation. Therefore,
as a main rule, nongenetic biomarkers should be investigated in patients with RPL
and controls matched by menstrual cycle phase in nonpregnant women and length
of gestation during pregnancy (see Table 2).
A further problem arises when biomarkers in the blood or uterus, which are affected
by hormones, are investigated in patients with RPL just before or at the time of miscar-
riage and compared with similar measurements in controls with a healthy ongoing
pregnancy. Most of the pregnancy-related hormones, hCG, estrogen, and PAPP-A,
decrease in threatened miscarriage and finding differences in a specific biomarker be-
tween women with a miscarriage or a healthy pregnancy, respectively, may be
severely confounded by the fact that hormones in the former group are lower than
in the control group.
A last but important factor that may confound case-control studies in RPL research
is differences in the viability and inflammatory status of the uterine content before or at
the time of miscarriage in patients and at the time of induced abortion in controls. At
the time of embryonic death, the trophoblast will undergo necrosis and intrauterine
hemorrhage will often induce inflammation in the decidual tissue, which can be re-
flected in measurements of immunologic biomarkers in the blood or decidual/tropho-
blast tissue. In contrast, the same biomarkers in the blood of controls with an ongoing
normal pregnancy or in the blood or decidual/trophoblast tissue from controls with an
induced abortion on social indication will not be influenced by inflammation. Unfortu-
nately, in many publications, levels of biomarkers associated with immune function or
apoptosis (programed cell death) in the blood or decidual/trophoblast tissue are
compared between women with a missed abortion and women with a normal ongoing
pregnancy or women undergoing induced abortion. In many of these publications,
finding differences in levels of such biomarkers are interpreted as proof that changes
in maternal immune reactions or apoptosis are causing miscarriage or RPL. To avoid
detecting biomarkers in patients with RPL that differ from those in controls merely due
to processes being a result of rather than a cause of miscarriage, measurements in pe-
ripheral blood should be undertaken in very early pregnancy (eg, GW 5) at a time when
the feto-placental unit is very tiny and not expected to affect systemic inflammatory
responses and before the confounding effect of declining hormones in patients take
place (see Table 2).23
Another approach to counteract the methodological problem associated with the
comparison of necrotic and vital tissue is to do karyotyping of embryos from missed
abortions in women with or without RPL. Biomarkers in the blood or decidual/tropho-
blast tissue can then be compared between patients with euploid male embryos (to
avoid erroneous karyotyping of maternal tissue) and patients with embryos with a
chromosome abnormality that definitively will cause early embryonic death. Levels
of biomarkers that are influenced by embryonic and trophoblast necrosis and
28 Christiansen

inflammation may be equally affected in women with euploid and aneuploid miscar-
riages. Therefore, differences in expression of biomarkers between the women/em-
bryos from the 2 groups can probably be attributed to factors that may have
caused euploid miscarriage. Examples of such studies that point to a causal role of
inflammatory cytokines in RPL is the study by Calleja-Agius and colleagues,24 finding
significantly higher plasma tumor necrosis factor-a, interferon-g, interleukin (IL)-6, and
IL-10 levels in women with euploid miscarriages than in healthy pregnancy, whereas
these cytokines were not increased in women with aneuploid miscarriages. Another
study similarly found increased numbers of activated leucocytes in RPL women
with a miscarriage with an euploid compared an aneuploid embryo.25
As illustrated previously, finding a suitable control group for measurement of nonge-
netically determined biomarkers is a difficult task. Adherence to the recommendations
given previously (see Table 2) may diminish the risk of conducting a case-control
study that is methodologically flawed but offers no guarantee. The factors that should
not differ between cases and controls in RPL research are the following:
 Number of previous ongoing pregnancies
 Levels of estrogen and progesterone in nonpregnant women
 Levels of estrogen, progesterone, and hCG in pregnant women
 Degree of viability of decidual or trophoblast tissue
I realize that very few case-control studies in RPL meet these criteria, especially the
criteria of similar hormonal levels. More research should be done regarding genetically
determined biomarkers because these are robust to the confounding effects dis-
cussed in this section. It is possible that some nongenetic biomarkers may also be
robust to the effects of the mentioned confounding factors (eg, reproductive hor-
mones). However, only when it has been proven in separate studies that a specific fac-
tor (eg, hormone) is not affecting the biomarker under study, a confounding effect of
the factor can be ignored in subsequent case-control studies.

Lack of Protocol Data or a Priory Hypothesis

As a referee, I have often discovered that some research groups have published a
series of case-control studies in different articles often in different journals, each
case-control study dealing with only one specific biomarker, which was found to be
significantly increased in patients with RPL. When the articles are carefully read and
the data compared it seems that all the biomarkers have been investigated in the
same patient and control groups in the same period. In the articles, no information
was given about how many biomarkers were investigated in each study and what
was the a priori hypothesis. The suspicion is that in many of these studies, the inves-
tigators have tested a huge series of biomarkers using a multiplex testing panel that
can test for hundreds of (often genetic) biomarkers in 1 day and they report only
data of those biomarkers that (by chance) are found significantly increased in RPL. Af-
ter this “fishing expedition,” the biomarkers being significantly associated with RPL
are published one by one in separate articles according to the “salami method”
without providing any information about how many biomarkers were investigated. In
this way, the investigators can expand their publication list in an easy way. This
“disguised multiple testing” is a maligned design, because the referees of the articles
have no chance to know that multiple testing was done and to ask the investigators to
do the appropriate statistical adjustment (see later in this article) and modify the con-
clusions. In the more benign cases, the investigators openly provide details of all
tested biomarkers/variables in tables and the referees can then ask for adjustment if
not already done.
 Research Methodology in Recurrent Pregnancy Loss 29

This illustrates the problem that in many studies (especially case-control studies) deal-
ing with RPL and other disorders, it is often unclear whether the result being reported as
the main finding really was part of the a priori hypothesis from the start or whether it was a
finding discovered during the conduct of the study: a post hoc finding (see Table 1).
As a preventive measure, high-ranking journals request the investigators to give
more details of the study protocols or the protocols should be available on an online
public registry (www.ClinicalTrials.Gov) before the beginning of the study. All journals
should, as a minimum, request the investigators in the “materials and methods” sec-
tion to provide information from the study protocol about the a priori hypothesis, main
outcome measures, and sample size calculations, and to list all risk factors and bio-
markers that were planned to test and actually tested.

Multiple Testing
Most case-control studies in RPL can be categorized as discovery studies in which
there is no prior hypothesis about which associations are probable and therefore a
series of biomarkers are investigated openly or disguised. Testing of multiple bio-
markers is facilitated by the introduction of multiplex testing panels that can test hun-
dreds of biomarkers at a time, thereby reducing the costs immensely. If, for example,
40 different biomarkers are tested in patients and controls and a P value for signifi-
cance of less than .05 was chosen, it is expected that 2 (40  0.05 5 2) biomarkers
will be significantly increased or decreased in patients merely by chance. If no a priori
hypothesis about which biomarker was the primary focus of the study was made (dis-
covery study) the P values for all tested biomarkers should be subject to the Bonferroni
adjustment by multiplying the P value with the number of comparisons made. If 1 or
2 biomarkers are still significantly associated with RPL after this adjustment, the asso-
ciation must be confirmed in at least 1 independent study (replication study) with the a
priori hypothesis that the variables are associated with RPL.

Interpretation
Causality can rarely, if ever, be stated after finding a potential risk factor statistically
significantly associated with a disease. According to Bradford-Hill criteria,26 there
are several demands for stating causality, the most important being the following:
 The criterion of temporality: that the risk factor occurs before the occurrence of
the disease.
 The criterion of a biologic gradient: there should be a positive correlation be-
tween the severity of a putative risk factor and the severity of the disorder.
In studies of RPL, the first criterion can be documented only in prospective studies
finding that the presence of a potential risk factor increases the risk of a new miscar-
riage compared with its absence. In RPL, the other criterion can be documented if
there is a correlation between the severity of a potential risk factor, for example, con-
centration of APL antibody, and the severity of RPL 5 number of miscarriages in the
patients. Findings in some previously quoted studies may indeed be interpreted as ev-
idence against the investigated variables being causative for RPL because no biologic
gradient was discovered.7,8

COHORT STUDIES

As stated in the previous section: to document that an exposure found to be associ-

ated with RPL in a case-control study is indeed a causal factor for RPL, a prospective
study is needed.
30 Christiansen

A prospective study is normally designed as a cohort study. A group of individuals

positive and another group of individuals negative for the exposure/potential risk fac-
tors for the disease (outcome) are identified. These 2 groups are now called cohorts
and they are observed during a defined time period in which the occurrence of the
outcome under study is monitored in the cohorts.

Concurrent Versus Nonconcurrernt Cohorts

In concurrent cohort studies, individuals assigned to the 2 cohorts are followed
prospectively, whereas in nonconcurrent (or retrospective) cohort studies, the assign-
ment to the cohorts is done on the basis of the detection of an exposure at a time in the
past when the outcome under study had not yet happened. If the disease under study
is RPL, such a concurrent cohort study could ideally assign 20-year-old women who
had not yet been pregnant and who have been tested for relevant biomarkers into a
hereditary thrombophilia positive and negative cohort. These 2 cohorts are then fol-
lowed until the women had aged 45 years and the occurrence of RPL in the 2 cohorts
is registered and compared between the cohorts. Adjustment for other risk factors for
RPL, such as age at the first pregnancy attempt, body mass index (BMI), and so forth,
should be undertaken in a multivariate analysis. It is clear that such a cohort study will
probably never be undertaken; because of the low frequency of RPL, tens of thou-
sands of women must be included and followed for 25 years: nobody would have
the resources for that. Instead cohort studies focusing on pregnancy outcome after
a diagnosis of RPL will be a easier task to perform. Still, concurrent cohort studies
can take many years and therefore almost all cohort studies of patients with RPL
have been nonconcurrent. To give an example, my group investigated hereditary
thrombophilia factors (factor II and factor V Leiden mutations) in the patients referred
to the RPL clinic in Denmark between 1986 and 2008, in 62% of the cases using DNA,
which was stored but not tested for the thrombophilia-associated mutations before
after the outcome of the first pregnancy after referral had been registered.27 In most
cases, the patients and their doctors were therefore unaware of the exposure status
at the time of the first pregnancy after referral and none of the patients received anti-
coagulation treatment. In a good nonconcurrent cohort study, information about both
exposure and outcome status should not be available at the time the cohorts are
formed. The patients were thus in a nonconcurrent (retrospective) cohort study
assigned to a hereditary thrombophilia positive and negative cohort, and outcome
of the first pregnancy after referral to the clinic was “outcome.” In a logistic regression
analysis, adjusting for the impact of number of previous pregnancy losses, maternal
age, and smoking, it was found that the presence of the hereditary thrombophilia fac-
tors significantly decreased the OR for birth in the first subsequent pregnancy (OR 5
0.48, P 5 .05) compared with their absence. Thus, a much easier approach than doing
a large concurrent cohort study can provide results documenting the importance of
potential causal factors in RPL.

Nonrandom Misclassification
Cohort studies are prone to methodological errors, which are important to recognize
when conducting or assessing them. Nonrandom misclassification is a significant
threat to the validity of cohort studies; however, unlike case-control studies in which
the patients or controls are the main source of the erroneous information leading to
misclassification, in cohort studies it is the researchers who are responsible. In occu-
pational medicine, workers are exposed to substances thought to increase the risk for
some disease. Because of that knowledge, these workers are often monitored more
closely than nonexposed workers and symptoms of disease (outcome) may be
 Research Methodology in Recurrent Pregnancy Loss 31

registered more often in the exposed than in the nonexposed cohort. Biased intensity
of monitoring is also a problem in reproductive medicine research. In a cohort study of
women with obesity (the exposure) or no obesity, to investigate whether obesity
increases the risk of miscarriage, many obese women will have polycystic ovary syn-
drome (PCOS) and anovulation and therefore undergo assisted reproductive technol-
ogy (ART) treatment, whereas most nonobese women will be able to conceive without
the need of ART. In ART cycles, a b-hCG test is normally done 14 days after ovulation
or embryo transfer, and all biochemical pregnancies will be detected, whereas among
non-ART patients, they will often remain undetected. Therefore such a cohort study
may show that obese women have an increased risk of pregnancy loss but this may
be a result of nonrandom misclassification of the outcome.
In RPL cohort studies, misclassification is also a significant problem. The prognosis
after a diagnosis of RPL is heavily disputed: various prospective studies have
reported the chance of live birth in the first pregnancy after referral in patients with
3 miscarriages to be between 63% and 87%, with 4 miscarriages between
44% and 73%, and with 5 miscarriages between 25% and 52%, respectively.11 These
very different estimates are frustrating: for a patient with RPL, it is of utmost im-
portance to know whether her chance for live birth is 44% or 73%. This variation
may be caused by random misclassification within individual studies. Data from
2 placebo-controlled trials of intravenous immunoglobulin (IvIg) conducted in Swe-
den28 and Denmark,29 respectively, may clarify the issue. Patients with 3 or more mis-
carriages who met the inclusion criteria for participation in the trials were encouraged
to conceive and call the clinics as soon as they got pregnant. When the Swedish trial
(cohort 1) was concluded after 4 years, 50.6% of the patients were classified as not
having achieved pregnancy, only 3.4% had pre-embryonic losses, and 10.1% had
embryonic losses (Fig. 1). When the Danish trial (cohort 2) was concluded after
6 years, only 14.7% did not report pregnancy, 22.4% had pre-embryonic losses,
and 27.9% had embryonic losses. When live birth rates in cohorts 1 and 2 are calcu-
lated with the number of recognized pregnancies in the denominator, they become
72.7% and 44.8%, respectively (P<.005). The most striking difference between the
2 cohorts is the very much higher frequency of nonconception in cohort 1 than in
cohort 2. This may be because of the simple fact that the Danish patients were told
that they could not be included in the trial if they contacted the clinic more than

60

50

40

Not pregnant
% 30 Pre-embryonic losses
Embryonic losses
20 Live births

10

0
Cohort 1 Cohort 2
N = 89 N = 68

Fig. 1. Classification of pregnancy outcome in 4 categories of patients with RPL eligible for
participation in a Swedish (cohort 1) and a Danish (cohort 2) placebo-controlled trial of IvIg.
In each cohort, patients receiving IvIg and placebo are combined.
32 Christiansen

5 days after the missed menstrual period, whereas in the Swedish trial, patients were
included only when fetal heart action could be demonstrated in GW 6 to 7. In cohort 2,
every conception was probably registered, whereas in cohort 1, many pre-embryonic
losses may have been misclassified as nonconception. Another published cohort of
patients with RPL that estimated the chance of live birth in the first subsequent preg-
nancy, also found that 30% of the patients reported no further pregnancies after
referral to the clinic or were lost to follow-up.30 This is again a much higher frequency
that the 14.7% in the Danish cohort. As in the Swedish cohort, this difference is prob-
ably caused by incomplete follow-up resulting in misclassification of preclinical preg-
nancy losses as nonconception. When results from different cohort studies are
compared, this misclassification becomes nonrandom (see Table 1).

Cohorts Estimating Live Births Per Time Unit

The aforementioned examples illustrate the unreliability of studies of RPL cohorts
based on the patients’ self-reporting of outcome of the first subsequent pregnancy.
My group has therefore proposed a more robust method of estimating the prognosis
for the patients. We obtained information about subsequent live birth from the national
birth registry (which registers 100% of all births in Denmark) for all patients with RPL
referred to our clinic from 1986 to 2008. We could then calculate age-specific and prior
number of miscarriages–specific proportions of patients who had achieved a live birth
within 5 and 15 years after the first consultation (66.7% and 71.1%, respectively).11
Such an estimate with the rate of live birth in the numerator and the observation
time in the denominator is a robust and clinical relevant way to estimate the prognosis,
and it is not sensitive to the intensity of monitoring as in the previously discussed
methods. However, it can, of course, be used only in countries with a complete and
valid national birth registration system. Our estimates of the long-term live birth rates
in subsets of patients with RPL, most of whom had attempted pregnancy several
times after referral, are generally lower than those reported in the first pregnancy in
a frequently quoted cohort,30 and because our cohort is not sensitive to misclassifica-
tion, the other cohort probably provides too optimistic rates.

TREATMENT TRIALS

In a time of embracement of evidence-based medicine, all treatments, new as well

as established, should undergo testing in clinical trials. With regard to RPL, clinical
treatment trials compare miscarriage rates in patients with RPL given one specific
treatment in the present pregnancy compared with outcome in a control group not
receiving the treatment.

Historical Controls
In some treatment trials, pregnancy outcome in a group of patients treated in the pre-
sent was compared with outcome in nontreated controls from the past, so-called his-
torical controls.
In a small study of PGS in the treatment of RPL in women aged 35 years,31 the live
birth rate after PGS was for each patient compared with the expected rate in women
with similar age and number of miscarriages calculated from a historical cohort.30 The
investigators found that the live birth rate after PGS was higher than the expected rate
using the data from the historical cohort. Comparing outcome in contemporary RPL
patient groups with historical control groups carries a substantial risk of producing un-
reliable results (see Table 1). It is extremely difficult to be sure that a group of patients
with RPL identified in the past is comparable with a present group, as ascertainment of
 Research Methodology in Recurrent Pregnancy Loss 33

patients, screening methods, and pregnancy-monitoring procedures have changed

substantially with time.
A variant of the use of historical controls is the use of the treated patients as their
own controls. In RPL research, the live birth rate in patients with RPL before a specific
intervention is compared with the birth rate after the intervention. The use of this
method is still seen in publications especially concerning surgical treatment of RPL,
but also regarding other treatment modalities, such as metformin.32 Although using
the patients as their own controls at a first glance would appear ideal, this method
is severely flawed due to the phenomenon regression to the mean.33 The population
of patients with RPL may comprise the following:
 A subset of women who have experienced the miscarriages merely because of
chance (repeated embryonic aneuploidies) with a low intrinsic risk of miscarriage
in each pregnancy but the women have been sampled asymmetrically from a
binominal distribution.
 A subset of women with maternal risk factors with a higher intrinsic risk of
miscarriage.
In the former subset, on any subsequent measure (new pregnancy) the mean risk of
miscarriage will be closer to the (low) mean of the original population in spite of no inter-
vention due to regression to the mean. If they are selected to a treatment trial, the
chance of success in the next pregnancy will be excellent in spite of no intervention.
In a trial of cerclage, the posttreatment birth rate was statistically significantly higher
than before cerclage and it was concluded that cerclage increases the live birth rate.
However, using this method, all types of interventions can be proved to be efficient
in the treatment of RPL, even placebo.34 In an RCT of infusion of allogeneic lymphocytes
versus placebo in RPL, the patients who received placebo had a live birth rate of 12.5%
before they entered the trial and subsequently 47.7% gave birth after placebo infusions.
By comparing birth rates before and after placebo by c2 test, the statistics claim that
placebo is highly efficient in treating RPL. This apparent, but false, improvement of
outcome can be completely attributed to the effect of regression to the mean.
Randomized Controlled Trials
Proper evaluation of treatments in RPL can be done only in RCTs. In RCTs, patients
are prospectively allocated to treatment or no-treatment or placebo by some random-
ization procedure that cannot be influenced by the researcher. This will increase the
chance that confounding variables, which can influence the prognosis, are equally
distributed in the groups.
Blinding
In the ideal RCT, allocation to active treatment or a placebo that cannot be distin-
guished from each other is undertaken according to a randomization list so that the
trial is blinded (masked) for both the researchers and patients, a so-called double-
blind design. The effects of blinding in RCTs are as follows:
 Blinding of the doctors will ensure the same monitoring and concomitant therapy
in both groups
 Blinding of the patients will provide them with the full placebo effect that due to
neuroendocrine pathways may exhibit a positive effect on pregnancy outcome in
RPL35
 Blinding of the patients will diminish the dropout rate and associated poor regis-
tration of subsequent outcome in those allocated to the expected less-efficient
treatment
34 Christiansen

Several RCTs in the RPL area are unfortunately not blinded. None of the trials testing
the effect of heparin plus low-dose aspirin in patients with RPL with or without APL has
compared this intervention against a placebo for heparin but only against low-dose
aspirin alone or a peroral placebo for aspirin.36–38 This is probably due to the reluc-
tance to use a placebo drug for daily subcutaneous injections during the whole period
of pregnancy. Although the outcome in RPL, a new miscarriage, can be determined by
objective methods that are not impacted by blinding/nonblinding, nonblinding can
negatively or positively influence outcome (see Table 1) in the groups receiving
nontreatment or expected less-efficient treatment compared with the expected
more-efficient treatment (tablets vs injections plus tablets).
Although a double-blinded randomized placebo-controlled trial is the best design to
obtain valid data concerning treatment effect, this method does not guarantee that the
results are trustworthy, as other circumstances can confound or weaken the results.

Confounding in RCTs
Most RCTs in RPL have included a small number of patients and this increases the risk
that a confounding variable, such as number of prior miscarriages and age, by chance
is unevenly distributed between the 2 allocation groups. In a small RCT, it is not suf-
ficient to show that the mean number of previous miscarriages or age is not statisti-
cally significantly different in the 2 allocation groups; these variables can still
significantly confound the results. The best way to counteract uneven distribution of
prognostic variables in RCTs is by doing block-randomization, whereby separate allo-
cation of patients is done according to groups with comparable number of miscar-
riages and age.

Inclusion after GW 6
In numerous RCTs, patients with RPL have been allocated and included in the trial only
when an ultrasonographic examination has demonstrated a viable intrauterine preg-
nancy in GW 6 to 7.28,38,39 This design can unfortunately substantially diminish any ef-
fect of the active intervention (see Table 1). Twenty-two percent of closely monitored
patients with RPL have pre-embryonic losses (see Fig. 1), which comprise almost half
of all their pregnancy losses.29 When treatment in RCTs starts only after the demon-
stration of fetal heart action, almost half of the risk period for miscarriage has thus
passed and the spontaneous prognosis at that time is good.40 Therefore, very large
numbers of participants are needed to show any treatment effect. Furthermore,
when treatment is started only in GW 6 to 7, the pregnancy has been exposed to
the full harmful effect of thrombophilic or immunologic factors for 2 to 3 weeks before
initiation of the potential beneficial therapy. The trophoblast or fetus, although viable at
that time, may have suffered irreversible damage that cannot be counteracted by initi-
ating active therapy.

Premature discontinuation of an RCT

In many RCTs, at least 1 interim analysis is performed during the conduct of the trial
and it is prematurely stopped if the results at this analysis show that a statistically sig-
nificant effect of the intervention could not be obtained even if the trial was continued
until the originally planned number of participants was reached. Many RCTs in the area
of RPL have been stopped prematurely after performance of an interim anal-
ysis,28,41,42 and this poses a substantial problem in interpreting the results, especially
when they are included in systematic reviews and meta-analyses. There is a great risk
that the difference between the outcome in the intervention and nonintervention/pla-
cebo groups at the stopping point after the interim analysis due to a random fluctua-
tion is smaller than if the trial had been continued to the planned number of
 Research Methodology in Recurrent Pregnancy Loss 35

participants; if the difference had been larger the trial had been allowed to continue.
RCTs stopped after interim analyses will therefore be expected to report intervention
effects that are smaller than the true effects (see Table 1) and, if included in systematic
reviews, their scientific quality should accordingly be down-graded.

Inclusion of several outcomes from the same patient

In case series and RCTs in RPL research, several pregnancy outcomes from the same
patients are sometimes included.43,44 This is a methodological flaw, as the outcomes
of pregnancies in individual patients with RPL due to the importance of maternal risk
factors (eg, age, number of prior losses) are linked variables. The commonly used sta-
tistical methods, such as c2 test and Fisher test, require that the tested variables are
independent and cannot be used. Therefore, only one pregnancy outcome from each
patient should be included in trials testing interventions, whether randomized or not.

Systematic Reviews
A systematic review is a review that uses systematic and explicit methods to critically
appraise a research topic; statistical methods such as meta-analyses may be used to
analyze and summarize the results of the included studies. Both case-control studies
and RCTs can be subject to such reviews and their results are considered the highest
level of evidence. In RPL research, systematic reviews have in particular focused
on intervention/treatment studies. The pooled OR calculated from the combination
of results in the meta-analysis is considered a good measure for the overall effect of
the intervention/treatment under study. However, because most treatments exhibit
different effects in different subsets of patients (eg, men and women or patients
with severe vs less severe disease), an important use of meta-analyses is to identify
subgroups of patients who have the largest benefit of treatment.45 I focus my discus-
sion regarding systematic reviews in RPL on those dealing with immunotherapy with
IvIg, because these reviews have included the largest numbers of studies and patients
and the discussion illustrates the pitfalls and methodological flaws that characterize
systematic reviews in the area of RPL treatment. Four different systematic reviews
have been published concerning IvIg treatment in RPL,4,42,46,47 which have provided
very different results and conclusions. This variability is in my view caused by either
(1) a failure to recognize the need of doing separate analysis in relevant subgroups
of patients or (2) by unfounded exclusions of RCTs in the systematic reviews. A
Cochrane review4 did not find any difference in live birth rate between IvIg and placebo
in all included patients with RPL but made no distinction between patients with pri-
mary and secondary RPL, which would be relevant, as 2 published trials had found
the IvIg efficient exclusively in the latter group.29,48 In contrast, another systematic re-
view based on almost the same patients recognized the relevance of doing this sub-
group analysis and found the live birth rate significantly higher in IvIg-treated than in
placebo-treated patients with secondary RPL.46
The 2 most recent systematic reviews on IvIg in RPL included only patients with “un-
explained” RPL, which meant that trials that had included patients with RPL positive
for APL according to the assays used in the individual clinics and according to the
local cutoff values were excluded.42,47 The investigators considered that 2 RCTs
included APL-positive patients and all 92 patients in these RCTs were completely
excluded from the systematic reviews comprising 25.3% of all 364 patients with
RPL participating in RCTs of IvIg. Exclusion of this substantial proportion of all ran-
domized patients with RPL appears to be an enormous waste of valid information.
In one of the excluded RCTs, none of the patients in fact had APL and in the other
almost all patients who tested APL-positive had very low titers.49 Only 3% would be
36 Christiansen

considered APL-positive according to the current criteria.50 On the other hand, an RCT
that was not excluded did in fact include APL-positive patients51 and several of the
other RCTs did not report details of APL assays or cutoff levels. Because both
excluded RCTs had found a substantial effect of IvIg in patients with secondary
RPL, doing the meta-analyses without these resulted in the conclusion in both system-
atic reviews that there is no benefit of IvIg either in primary or secondary RPL.42,47 As
stated by my group and others, we find the methodology of these systematic reviews
questionable and without the exclusions it can be proved that IvIg is indeed efficient in
secondary RPL.49,52 As previously mentioned, it is legal and desirable to do meta-
analyses in subgroups of patients to identify subgroups that benefit best from treat-
ment, and such a subgroup could be patients with RPL without APL. However, a
systematic review excluding these patients should of course exclude only those
patients who are positive for APL and not complete RCTs with only a tiny minority
of participants being APL-positive. Furthermore, in most published relevant RCTs,
insufficient information is given about the APL assays and cutoff values, which may
vary substantially between the trials. To carry out a meta-analysis of therapies in
APL-negative patients with RPL based on good research methodology, it is therefore
necessary to collect data on APL levels for each included patient. In most cases, these
data on individual patients can be obtained only by contacting the investigators of the
RCTs. If such a collection of raw data is not done, the resulting systematic review and
meta-analysis will produce biased results.
This discussion illustrates that even conclusions from systematic reviews should be
evaluated critically and a systematic review is not per se the highest level of evidence.
If they are not conducted according to rigorous and systematic rules, they will end up
being no more evidence-based than the narrative reviews of the old days (see
Table 1).

SUMMARY

Most causes of RPL are poorly elucidated and may have a more multifactorial etiology
than infertility and in many instances the cause-effect relationship is unclear. Further-
more, the RPL population is much smaller that the infertile population, thereby making
it difficult to conduct studies with adequate statistical power.
These problems render research in RPL inherently difficult and only methodologi-
cally high-quality studies are expected to produce useful results. Unfortunately,
such high-quality research studies are rare in the area of RPL.
All types of studies in RPL research are characterized by methodological pitfalls,
which are often not recognized by the researchers or readers. Failure to recognize
these can completely invalidate studies in the area. It is hoped that this review, by
setting focus on the problem, can help improve studies in the area of RPL in the future.

REFERENCES

1. The Practice Committee of the American Society for Reproductive Medicine.

Evaluation and treatment of recurrent pregnancy loss: a committee opinion.
Fertil Steril 2012;98:1103–11.
2. Royal College of Obstetricians and Gynaecologists. The investigation and treat-
ment of couples with recurrent first-trimester and second-trimester miscarriage.
RCOG Green-top Guideline No. 17. 2011.
3. Jauniaux E, Farquharson RG, Christiansen OB, et al. Evidence-based guide-
lines for the investigation and medical treatment of recurrent miscarriage.
Hum Reprod 2006;21:2216–22.
 Research Methodology in Recurrent Pregnancy Loss 37

4. Porter TF, LaCoursiere Y, Scott JR. Immunotherapy for recurrent miscarriage.

Cochrane Database Syst Rev 2006;(2):CD000112. http://dx.doi.org/10.1002/
14651858.
5. The Practice Committee of the Society for Assisted Reproductive Technology,
Practice committee of the American Society for Reproductive Medicine. Preim-
plantation genetic testing: a Practice Committee opinion. Fertil Steril 2008;90:
S136–46.
6. Nybo Andersen AM, Wohlfart J, Christens P, et al. Maternal age and fetal loss:
population based register linked study. BMJ 2000;320:1708–12.
7. van den Boogaard E, Cohn DM, Korevaard JC, et al. Number and sequence of
preceding miscarriages and maternal age for the prediction of antiphospholipid
syndrome in women with recurrent miscarriage. Fertil Steril 2013;99:188–92.
8. Jaslow CR, Careney JL, Kutteh WH. Diagnostic factors identified in 1020 women
with two versus three or more recurrent pregnancy losses. Fertil Steril 2010;193:
1234–43.
9. Wilcox AJ, Weinberg CR, O’Connor JF, et al. Incidence of early loss of preg-
nancy. N Engl J Med 1988;319:189–94.
10. Kolte AM, van Oppenraaij RH, Quenby S, et al. Biochemical pregnancy loss and
pregnancy of unknown location are prognostically important for unexplained
recurrent miscarriage. Under review.
11. Lund M, Kamper-Jørgensen M, Nielsen HS, et al. Prognosis for live birth in
women with recurrent miscarriage. What is the best measure of success?
Obstet Gynecol 2012;119:37–43.
12. Axelsson G, Rylander R. Validation of questionnaire reported miscarriage,
malformation and birth weight. Int J Epidemiol 1984;13:94–8.
13. Wilcox AJ, Horney LF. Accuracy of spontaneous abortion recall. Am J Epidemiol
1984;120:727–33.
14. Out HJ, Bruinse HW, Christiaens GC, et al. A prospective, controlled multicenter
study on the obstetric risks of pregnant women with antiphospholipid anti-
bodies. Am J Obstet Gynecol 1992;167:26–32.
15. Regan L, Braude PR, Hill DP. A prospective study of the incidence, time of
appearance and significance of anti-paternal lymphocytotoxic antibodies in
human pregnancy. Hum Reprod 1991;6:294–8.
16. Lissauer D, Piper K, Goodyear O, et al. Fetal-specific CD81 cytotoxic T cell re-
sponses develop during normal human pregnancy and exhibit broad functional
capacity. J Immunol 2012;189:1072–80.
17. Wegienka G, Havstad S, Bobbitt KR, et al. Within-woman change in regulatory T
cells from pregnancy to the postpartum period. J Reprod Immunol 2011;88:
58–65.
18. Gamill HS, Guthrie KA, Aydelotte TM, et al. Effect of parity on fetal and
maternal microchimerism: interaction of grafts within a host? Blood 2010;
116:2706–12.
19. O’Donoghue K, Chan J, de la Fuente J, et al. Microchimerism in female bone
marrow and bone decades after fetal mesenchymal stem-cell trafficking in preg-
nancy. Lancet 2004;364:179–82.
20. Raghupathy R, Al Mutawa E, Makhseed M, et al. Modulation of cytokine produc-
tion by dydrogesterone in lymphocytes from women with recurrent miscarriage.
BJOG 2005;112:1096–101.
21. von Wolff M, Thaler CJ, Strowitzki T, et al. Regulated expression of cytokines in
human endometrium throughout the menstrual cycle: dysregulation in habitual
abortion. Mol Hum Reprod 2000;6:627–34.
38 Christiansen

22. Schumacher A, Brachwitz N, Sohr S, et al. Human chorionic gonadotrophin

attracts regulatory T cells into the fetal-maternal interface during human preg-
nancy. J Immunol 2009;182:5488–97.
23. Piosek ZM, Goegebeur Y, Klitlou L, et al. Plasma TNF-a levels are higher in early
pregnancy in patients with secondary compared with primary recurrent miscar-
riage. Am J Reprod Immunol 2013;70:347–58.
24. Calleja-Agius J, Jauniaux E, Pizzey AR, et al. Investigation of systemic inflam-
matory response in first trimester pregnancy failure. Hum Reprod 2012;27:
349–57.
25. Quack KC, Vassiliadou N, Pudney J, et al. Leukocyte activation in the decidua
of chromosomally normal and abnormal fetuses from women with recurrent
abortion. Hum Reprod 2001;16:949–55.
26. Bradford-Hill A. The environment and disease: association or causation? Proc R
Soc Med 1965;9:295–300.
27. Lund M, Nielsen HS, Hviid TV, et al. Hereditary thrombophilia and recurrent
pregnancy loss: a cohort study of pregnancy outcome and obstetric complica-
tions. Hum Reprod 2010;25:2978–84.
28. Jablonowska B, Selbing A, Palfi M, et al. Prevention of recurrent spontaneous
abortion by intravenous immunoglobulin: a double-blind placebo-controlled
trial. Hum Reprod 1999;14:838–41.
29. Christiansen OB, Pedersen B, Rosgaard A, et al. A randomized, double-blind,
placebo-controlled trial of intravenous immunoglobulin in the prevention of
recurrent miscarriage: evidence for a therapeutic effect in women with second-
ary recurrent miscarriage. Hum Reprod 2002;17:809–16.
30. Brigham SA, Conlon C, Farquharson RG. A longitudinal study of pregnancy
outcome following idiopathic recurrent miscarriage. Hum Reprod 1999;14:
2868–71.
31. Munné S, Chen S, Fischer J, et al. Preimplantation genetic diagnosis reduces
pregnancy loss in women aged 35 years and older with a history of recurrent
miscarriage. Fertil Steril 2005;84:331–5.
32. Glueck CJ, Wang P, Goldenberg N, et al. Pregnancy outcomes among women
with polycystic ovary syndrome treated with metformin. Hum Reprod 2002;17:
2858–64.
33. Yudkin PL, Stratton IM. How to deal with regression to the mean in intervention
studies. Lancet 1996;347:241–3.
34. Christiansen OB. Transabdominal cervicoisthmic cerclage in the management
of recurrent second trimester miscarriage and preterm delivery. BJOG 1996;
103:595–6.
35. Stray-Pedersen B, Stray-Pedersen S. Etiological factors and subsequent repro-
ductive performance in 195 couples with a history of habitual abortion. Am J
Obstet Gynecol 1984;148:140–6.
36. Kutteh WH. Antiphospholipid antibody-associated recurrent pregnancy loss:
treatment with heparin and low-dose aspirin is superior to low-dose aspirin
alone. Am J Obstet Gynecol 1996;174:1584–9.
37. Rai R, Cohen H, Dave M, et al. Randomised controlled trial of aspirin and aspirin
plus heparin in pregnant women with recurrent miscarriage associated with
phospholipid antibodies (or antiphospholipid antibodies). BMJ 1997;314:253–7.
38. Kaandorp SP, Goddijn M, van der Post JAM. Aspirin plus heparin or aspirin
alone in women with recurrent miscarriage. N Engl J Med 2010;362:1586–96.
39. The German RSA/IVIV Group. Intravenous immunoglobulin in the prevention of
recurrent miscarriage. BJOG 1994;101:1072–7.
 Research Methodology in Recurrent Pregnancy Loss 39

40. Stern J, Coulam CB. Mechanism of recurrent spontaneous abortion. I. Ultraso-

nographic findings. Am J Obstet Gynecol 1992;166:1844–52.
41. Ober C, Karrison T, Odem RR, et al. Mononuclear-cell immunisation in preven-
tion of recurrent miscarriages: a randomised trial. Lancet 1999;354:365–9.
42. Stephenson MD, Kutteh WH, Purkiss S, et al. Intravenous immunoglobulin and
idiopathic secondary recurrent miscarriage: a multicentered randomized
placebo-controlled trial. Hum Reprod 2010;25:2203–9.
43. Brenner B, Hoffman R, Blumenfeld Z, et al. Gestational outcome in thrombo-
philic women with recurrent pregnancy loss treated with enoxaparin. Thromb
Haemost 2000;83:693–7.
44. Grandone E, Brancaccio V, Colaizzo D, et al. Preventing adverse obstetric
outcomes in women with genetic thrombophilia. Fertil Steril 2002;78:371–5.
45. Thompson SG, Higgins JPT. Can meta-analysis help target interventions at indi-
viduals most likely to benefit? Lancet 2005;365:341–6.
46. Hutton B, Sharma R, Fergusson D, et al. Use of intravenous immunoglobulin for
treatment of recurrent miscarriage: a systematic review. BJOG 2007;114:
134–42.
47. Ata B, Tan SL, Shehata F, et al. A systematic review of intravenous immunoglob-
ulin for treatment of unexplained recurrent miscarriage. Fertil Steril 2011;95:
1080–5.
48. Christiansen OB, Mathiesen O, Husth M. Placebo-controlled trial of treatment of
unexplained secondary recurrent spontaneous abortions and recurrent late
spontaneous abortions with i.v. immunoglobulin. Hum Reprod 1995;10:2690–5.
49. Christiansen OB, Larsen EC, Husth M, et al. Intravenous immunoglobulin and
recurrent miscarriage. Fertil Steril 2011;7:e35.
50. Myakis S, Lockshin MD, Atsuma T, et al. International consensus statement on
an update on the update of the classification criteria for the definite antiphos-
pholipid syndrome (APS). J Thromb Haemost 2006;4:295–306.
51. Mueller-Eckhardt G, Mallmann P, Neppert J, et al. Immunogenetic and serolog-
ical investigations in nonpregnant and in pregnant women with a history of recur-
rent spontaneous abortions. J Reprod Immunol 1994;27:95–109.
52. Clark DA. Intravenous immunoglobulin and idiopathic secondary recurrent
miscarriage: methodological problems. Hum Reprod 2011;26:2586–91.