See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/317500153

Suicidal Behavior During Lithium and Valproate Treatment: A Within-

Individual 8-Year Prospective Study of 50,000 Patients With Bipolar Disorder

Article in American Journal of Psychiatry · June 2017

DOI: 10.1176/appi.ajp.2017.16050542

CITATIONS READS

98 261

8 authors, including:

Jie Song Erik Joas

Karolinska Institutet University of Gothenburg
50 PUBLICATIONS 941 CITATIONS 42 PUBLICATIONS 1,194 CITATIONS

SEE PROFILE SEE PROFILE

Sarah E Bergen Bo Runeson

Karolinska Institutet Karolinska Institutet
191 PUBLICATIONS 32,882 CITATIONS 189 PUBLICATIONS 10,035 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Jie Song on 30 March 2022.

The user has requested enhancement of the downloaded file.

 ARTICLES

Suicidal Behavior During Lithium and Valproate

Treatment: A Within-Individual 8-Year Prospective
Study of 50,000 Patients With Bipolar Disorder
Jie Song, Ph.D., Arvid Sjölander, Ph.D., Erik Joas, M.S., Sarah E. Bergen, Ph.D., Bo Runeson, M.D., Ph.D.,
Henrik Larsson, Ph.D., Mikael Landén, M.D., Ph.D., Paul Lichtenstein, Ph.D.

Objective: Conclusions regarding lithium’s antisuicidal effect Results: During follow-up, 10,648 suicide-related events
for bipolar disorder have been limited due to nonrepresentative occurred. The incidence rate was significantly decreased
subjects and potential confounding factors, including varying by 14% during lithium treatment (hazard ratio 0.86, 95%
severity of illness. Findings regarding the effect of valproate, the confidence interval [CI] 0.78–0.95) but not during val-
most common alternative to lithium, are inconsistent for sui- proate treatment (hazard ratio 1.02, 95% CI 0.89–1.15). The
cidal behavior. This study investigated the associations of these difference in hazard ratios of suicide-related events be-
two drugs with the risk of suicide-related events, and possible tween lithium and valproate was statistically significant.
differences between drugs, by using within-individual designs Estimates of the population attributable fraction suggested
in a register-based longitudinal cohort. that 12% (95% CI 4%220%) of suicide-related events could
have been avoided if patients had taken lithium during the
Method: Through linkage of multiple Swedish national reg- entire follow-up.
isters, 51,535 individuals with bipolar disorder were followed
from 2005 to 2013 for treatment with lithium and valproate. Conclusions: The results suggest that lithium should be
Stratified Cox regression was used to estimate the hazard considered for patients with bipolar disorder with suspected
ratios of suicide-related events during treated periods com- suicidal intentions, although risk for suicide is only one of the
pared with untreated periods. For significant associations considerations when providing clinical care.
between medication and suicide-related events, the pop-
ulation attributable fraction was estimated to assess the public
health impact for patients with bipolar disorder. AJP in Advance (doi: 10.1176/appi.ajp.2017.16050542)

Patients with bipolar disorder have a high risk of suicidal but was inconclusive about its ability to reduce the risk of
behavior compared with both the general population and suicide attempts more than either placebo or valproate. A
patients with other psychiatric disorders (1–3). Among the landmark long-term randomized controlled trial was con-
options of maintenance treatments for bipolar disorder, lith- ducted among high-risk patients with bipolar disorder, and it
ium is the first-line treatment and is suggested to have an detected no significant difference in the rate of suicide at-
antisuicidal effect (4). Valproate is a widely used alternative tempts between lithium and valproate, but the study was only
to lithium in preventing manic episodes and relapses (5), but powered to detect high relative risks (18). More studies on
its effect on suicidal behavior is unclear (6). valproate were conducted following an alert of increased
Despite accumulated reports of a protective effect for risk of suicidal behavior related to anticonvulsants reported
lithium (7–9), no consistent evidence indicates whether it is by the U.S. Food and Drug Administration, but the results
superior to valproate in reducing suicide risk (10–14). Three gathered were contradictory (4).
meta-analytic reviews from Baldessarini’s group suggested The strength and generalizability of existing random-
significantly lower risks of attempted and completed sui- ized controlled trials that examine the antisuicidal effect of
cides among lithium-treated patients than in either patients lithium (17, 19) are limited due to rare events and selected
not treated with lithium (8, 15) or patients treated with populations. Among the 48 trials included in the updated
anticonvulsants (including valproate) (16). A recent meta- meta-analysis of randomized controlled trials for all types
analysis of randomized controlled trials for patients with of mood disorders (17), only 19 focused on patients solely
mood disorders (17) showed that lithium is more effective with bipolar disorder and only 10 of these had more than
than placebo in reducing the number of completed suicides 100 participants with follow-up time exceeding 1 year.

ajp in Advance ajp.psychiatryonline.org 1

SUICIDAL BEHAVIOR DURING LITHIUM AND VALPROATE TREATMENT

Moreover, randomized controlled trials typically exclude specified, and schizoaffective disorder bipolar type and is
actively suicidal individuals and thus may not be represen- sufficiently sensitive and specific to be used in Swedish
tative of the population of patients for whom the medication register-based studies (24).
is actually prescribed.
Observational studies avoid the ethical and logical prob- Medications. The main exposure was defined as medica-
lems encountered by randomized controlled trials and, tion with lithium sulfate (Anatomical Therapeutic Chemical
additionally, have the advantage of large sample size with [ATC] classification code: N05AN01) or sodium valproate/
long-term follow-up, offering adequate numbers of rare valproic acid (ATC code: N03AG01) in the Prescribed Drug
suicide-related events. Nevertheless, observational phar- Register. In routine psychiatric practice in Sweden, oral
macoepidemiological studies are highly susceptible to con- medications are unlikely to be dispensed for longer than
founding by indication, that is, patients are selected for a 3 months at a time. Therefore, as was done in previous studies
medication based on their risk for the outcome. For example, (21, 25), we defined a medication period as a sequence of at
in Sweden, patients at high risk for suicide might be more likely least two prescriptions, with no more than 3 months (92 days)
to be given prescriptions for lithium because of previous re- between any two consecutive prescriptions. Thus, for both
ports of its antisuicidal properties. Valproate is believed to be lithium and valproate, individuals were defined as on med-
more effective in rapid cycling and mixed states (20), which ication during the time interval between two dispensed
can steer valproate prescriptions to these patients. Many ap- prescriptions, unless the dispensed prescriptions occurred
proaches, including comparison of patients with different more than 3 months apart. To determine whether an indi-
numbers of medication purchases (7, 11, 12) and propensity vidual was on or off medication initially, the follow-up start
score models accounting for exposure-related covariates (14), was set to Oct. 1, 2005, because the coverage of the Prescribed
are applied to address confounding, but these approaches Drug Register started on July 1, 2005.
cannot account for potential unmeasured confounding.
Alternatively, designs using within-individual compari- Outcomes. The main outcome was suicide-related events,
sons are promising approaches to address the above issues defined as attempted or completed suicide (ICD-10:
(21). With each person serving as his or her own matched X60–X84, Y10–Y34), which included those with un-
control, the individual-level comparison implicitly controls determined intent. Dates and diagnoses were retrieved from
for all time-invariant confounders (e.g., baseline severity of the National Patient Register and Cause of Death Register.
the illness, genetic predisposition to mental disorder/suicidal
behavior, childhood environment). By using a large Swedish Statistical Analyses
cohort of 50,000 patients with bipolar disorder over 8 years of To control for time-invariant covariates, we performed within-
follow-up, we aimed to use within-individual designs to in- individual analyses using stratified Cox regression. This
vestigate the associations of lithium and valproate treatment method, by design, exclusively draws information from in-
with suicidal behavior, and their possible differences. dividuals who ever attempted suicide during follow-up (i.e.,
individuals without suicide-related events or varying covar-
iates during follow-up were noninformative). More metho-
METHOD
dological details are given in a related thorough review (26).
Subjects We split the follow-up time into consecutive periods. A
We linked longitudinal Swedish population-based registers, new period started after a medication switch (i.e., from off
enabled by unique personal identification numbers (22). By medication to on medication or vice versa, for either lithium
linking the Total Population Register, Migration Register, or valproate) or a suicide attempt. For the latter, we restarted
Cause of Death Register, National Patient Register, and the following period at baseline (i.e., to set the underlying
Prescribed Drug Register (23), we identified 51,535 individ- time scale as the time since the last suicide attempt). Lithium
uals with bipolar disorder followed from Oct. 1, 2005, or age and valproate treatments were defined as time-varying di-
15, or date of first diagnosis if later than Oct. 1, 2005, until chotomous exposures, respectively. Age range (15–25, 26–38,
emigration, death, or Dec. 31, 2013, whichever occurred first. 39–50, 51–100 years, grouped by the quartiles of the baseline
This study was approved by the Ethics Committee at age distribution among those who attempted suicide during
Karolinska Institutet. the follow-up) and previous number of suicide attempts were
adjusted for as time-varying categorical covariates. We es-
Measures timated the hazard ratios and 95% confidence intervals (CIs)
Bipolar disorder. We applied a modified validated algorithm for differences in the rate of suicide-related events between
(24), which defined bipolar disorder as at least two inpatient periods. This method can be found in detail in related pub-
or outpatient visits for a core discharge diagnosis of bipolar lications (21, 25, 27, 28).
disorder (for International Classification of Disease [ICD] For significant associations between medication and
codes see Table S1 in the data supplement accompanying the suicide-related events, we estimated the population attrib-
online version of this article). This algorithm includes bipolar utable fraction to assess the public health impact for patients
I disorder, bipolar II disorder, bipolar disorder not otherwise with bipolar disorder. It measures the proportion of events

2 ajp.psychiatryonline.org ajp in Advance

 SONG ET AL.

that would be eliminated if the whole cohort would be individuals with bipolar I and bipolar II disorder identified
medicated during the entire follow-up. The approaches for from the Swedish Bipolar Quality Register.
estimation and interpretation of population attributable To assess potential misclassification of exposure periods
fraction are described in the online data supplement and or delayed onset of drug action, we 1) defined the end of each
elsewhere (29). The CIs were estimated by nonparametric medication period as 14 days after the last dispensing date, 2)
bootstrap methods. defined the medication period as a sequence of dispensed
prescriptions with less than 4 months between them, and 3)
Sensitivity Analyses set the start of a medicated period to 7 days after the first
We also performed standard between-individual Cox re- dispensing date.
gression for comparison. Period splitting and covariate ad- To examine the robustness of our definition of suicide-
justment were the same as for the stratified Cox regression, related events, we repeated our analyses for events with 1)
with additional adjustments for sex, baseline severity of ill- determined intent, 2) undetermined intent, and 3) exclusion
ness (measured by previous hospitalizations), and baseline of completed suicides.
history of suicide attempts. We estimated hazard ratios and To test whether the association was biased due to the
95% CIs with cluster-robust standard errors accounting for assumption that a patient may be given a lithium prescription
within-individual correlations. Furthermore, we used a dif- after attempting suicide, we repeated our main analysis ex-
ferent method, a propensity score model, for comparison. cluding periods containing a switch to lithium medication
Since this method is commonly used for time-invariant ex- within 7, 14, and 30 days after a suicide attempt, respectively.
posures, we studied a shorter period without switch of It has been suggested that combination therapy with
medication status (the methods are described in the online lithium plus valproate is more effective in preventing relapse
data supplement). than valproate alone (31). To test the potential benefit of
For the significant association between lithium and combination therapy for antisuicidal effect, we repeated our
suicide-related events, we conducted additional within- analysis by defining medication periods with lithium alone,
individual analyses to examine to what extent the asso- valproate alone, and lithium plus valproate. Moreover, con-
ciation was affected by follow-up, inclusion criteria, and sidering that patients with lithium monotherapy might be
definitions of exposure and outcome. different from patients who have switched between lithium
First, we set the start of follow-up to October 2005 re- and valproate, we repeated our analysis for the subgroup with
gardless of the date of bipolar disorder diagnosis. This was to lithium monotherapy.
take into consideration the possibility that early symptoms A recent study showed an increased risk of suicide after
of adverse outcomes of illness might occur before diagnosis. discontinuation of lithium (versus valproate) (32). We also
To examine confounding by other concomitant psycho- explored the risks after initiation or discontinuation of
tropic treatment, we adjusted for concurrent medications, medication by further examining the time-varying effect
including lamotrigine, antipsychotics, antidepressants, ben- (periods were categorized as on medication for less than
zodiazepines, and other anticonvulsants. The definitions of 30 days/more than 30 days and off medication for less than
medication periods were the same as for lithium and val- 30 days/more than 30 days).
proate (for ATC codes, see the online data supplement). Finally, we used two negative controls to evaluate potential
To test for confounding by disease misclassification and confounding. First, to test whether the association could be
comorbid conditions, we evaluated different inclusion and due to decreased likelihood of suicide during periods of active
exclusion criteria. Instead of two admissions for bipolar treatment, we examined risks of suicidal-related events
disorder we 1) allowed for only one, 2) restricted analyses to during use of thyroid medications. Second, to test whether
patients with bipolar disorder without lifetime diagnoses of the inverse association is due to individuals receiving lith-
comorbid psychiatric conditions (for ICD codes, see Table S1 ium prescriptions during less chaotic phases of illness, we
in the online data supplement), and 3) used all individuals to examined the rate of any bone fracture during lithium
whom lithium was ever dispensed irrespective of diagnosis. medication.
To better guide clinicians, additional research on the time All analyses were performed with Stata 13.0 (33).
and symptoms for which lithium versus another mood sta-
bilizer should be employed as an antisuicidal treatment
RESULTS
intervention is recommended (19, 30). Therefore, we 1) con-
ducted an analysis by restricting the cohort to individuals Among the 51,535 patients with bipolar disorder, a total of
who were given prescriptions for lithium within 1 year after 10,648 suicide-related events occurred in 4,643 individuals
first diagnosis of bipolar disorder, to address the clinical (9.0%) during 273,140 person years of follow-up (Table 1).
question of whether lithium shortly after diagnosis is ben- Lithium treatment was most prevalent (41.0%), followed
eficial in preventing suicidal behavior, 2) analyzed a cohort of by valproate (16.3%). About 50% of the patients were never
individuals whose records indicated they ever had a mixed exposed to lithium or valproate during the study period. The
episode, to investigate lithium’s effect for patients with potential low percentage of patients on medications may partially be
high risks, and 3) performed analyses on two subcohorts of because a considerable proportion of patients identified from

ajp in Advance ajp.psychiatryonline.org 3

SUICIDAL BEHAVIOR DURING LITHIUM AND VALPROATE TREATMENT

TABLE 1. Characteristics at Baseline and During Follow-Up of Patients With Bipolar Disorder in Sweden (2005–2013), by Lithium and
Valproate Treatment Status
Individuals With Bipolar Disorder (for between-individual analysis)
Never Treated Individuals
With Lithium Eligible for Within-
Lithiuma Valproatea or Valproate Total Individual Analysis
(N=21,129) (N=8,411) (N=25,780) (N=51,535) (N=4,405)
Variable N % N % N % N % N %
Male 8,471 40.1 3,610 42.9 9,008 34.9 19,485 37.8 1,419 32.2
Age at baseline (years)b
15–25 2,525 12.0 1,224 14.6 4,596 17.8 7,822 15.2 1,172 26.6
26–38 4,346 20.6 1,991 23.7 5,956 23.1 11,335 22.0 1,137 25.8
39–50 4,825 22.8 2,116 25.2 5,732 22.2 11,674 22.7 1,051 23.9
51–100 9,433 44.6 3,080 36.6 9,496 36.8 20,704 40.2 1,045 23.7
Other psychiatric medications
prescribed during follow-up
Antipsychotics 12,967 61.4 6,117 72.7 11,736 45.5 27,728 53.8 3,277 74.4
Antidepressants 14,954 70.8 6,379 75.8 18,872 73.2 37,327 72.4 3,973 90.2
Anticonvulsants other than valproate 4,353 20.6 2,345 27.9 4,873 18.9 10,358 20.1 1,854 42.1
Lamotrigine 7,045 33.3 2,972 35.3 10,439 40.5 18,902 36.7 2,399 54.5
Benzodiazepines 9,104 43.1 4,145 49.3 9,623 37.3 20,793 40.3 2,877 65.3
Comorbid psychiatric conditions
Conduct disorder 125 0.6 97 1.2 244 0.9 435 0.8 98 2.2
ADHD 1,985 9.4 1,266 15.1 3,546 13.8 6,253 12.1 957 21.7
Substance use disorder 5,298 25.1 2,833 33.7 7,391 28.7 14,241 27.6 2,853 64.8
Personality disorder 3,497 16.6 1,880 22.4 5,301 20.6 9,834 19.1 1,945 44.2
Acute stress, adjustment disorders or 4,533 21.5 2,220 26.4 6,632 25.7 12,364 24.0 1,928 43.8
PTSD
At least one suicide-related event during 2,142 10.1 1,105 13.1 2,018 7.8 4,643 9.0 4,405 100.0
follow-up
Completed suicide during follow-up 230 1.1 99 1.2 308 1.2 590 1.1 359 8.1
Proportion of time exposed to lithium or 39.1 36.6 21.3 21.2
valproate
a
3,785 individuals were ever treated with both lithium and valproate, though not necessarily during overlapping periods.
b
Age range was categorized according to the quartiles of age distribution of individuals who had suicide-related events at any time during follow-up.

TABLE 2. Risk of Suicide-Related Events During Lithium and Valproate Treatment for Patients With prescribed other drugs, and
Bipolar Disorder in Sweden (2005–2013) had more comorbid psychi-
Within-Individual Analysisa Between-Individual Analysisb atric conditions.
Medication or Test Hazard Ratioc 95% CI p Hazard Ratioc 95% CI p We found a 14% reduced
Lithium 0.86 0.78–0.95 0.86 0.79–0.94 rate of suicide-related events
Valproate 1.02 0.89–1.15 1.11 0.99–1.24 for periods on compared
Test of difference between 0.038 0.001 with off lithium treatment
hazard ratios (hazard ratio 0.86, 95% CI
a
Stratified Cox regression was applied with adjustment for time-varying covariates including categorical age and previous 0.78–0.95; Table 2), but this
number of suicide attempts. was not the situation for
b
Ordinary Cox regression was applied with adjustment for the same covariates as in the stratified Cox regression and,
additionally, with adjustment for time-invariant covariates including sex, length of baseline hospitalization periods due to valproate (hazard ratio 1.02,
psychiatric admissions (a measure of illness severity), and history of suicide-related events before entering follow-up. 95% CI 0.89–1.15). The test
c
Hazard ratio is the ratio of the hazard rates during medication periods compared with nonmedication periods, for for the difference in hazard
lithium and valproate separately.
ratios for suicide-related events
between lithium and val-
the register were older and had stopped taking medication. proate had a x2 of 4.29 (p=0.038). As in the main analyses, the
Between-individual analyses were performed among all the between-individual analyses showed a lower rate of suicide-
patients with bipolar disorder. related events during lithium medication but not during
After excluding individuals without suicide-related events valproate medication (Table 2), with a significant difference
or varying covariates during follow-up, 4,405 individuals between them (p=0.001). Periods on lithium medication
were eligible for the within-individual analysis. These pa- constituted 17.9% of the total person-time for all patients.
tients had 10,403 suicide-related events, were more often Based on the exposure rate and hazard ratios, the population

4 ajp.psychiatryonline.org ajp in Advance

 SONG ET AL.

FIGURE 1. Sensitivity Within-Individual Analyses for Association Between Lithium Treatment and Suicide-Related Events in Patients With
Bipolar Disorder in Sweden (2005–2013)a

Sensitivity Analysis Hazard Ratio (95% CI)

Set start of follow-up at Oct. 1, 2005, regardless of diagnosis date 0.92 (0.85−1.01)
Adjustment for concurrent medication 0.81 (0.73−0.90)
Redefinition of subjects
Individuals with at least one bipolar diagnosis 0.86 (0.78−0.94)
Individuals without comorbid psychiatric conditions 0.68 (0.45−1.02)
Individuals with comorbid substance use disorder 0.84 (0.75−0.94)
Individuals ever using lithium 0.92 (0.86−0.99)
Individuals with lithium monotherapy 0.80 (0.71−0.90)
Lithium prescribed within 1 year after bipolar diagnosis 0.84 (0.73−0.96)
Individuals ever diagnosed with mixed episode 0.87 (0.74−1.03)
Individuals with bipolar I disorder 0.85 (0.67−1.09)
Individuals with bipolar II disorder 0.62 (0.46−0.82)
Redefinition of exposure time
Treatment ended 14 days after last dispensing 0.90 (0.82−0.99)
Treatment status defined by 4-month cut-off 0.84 (0.76−0.92)
Treatment started 7 days after first dispensing 0.88 (0.80−0.97)
Excluding periods when patient received lithium within certain
days after event
Received lithium within 7 days after attempted suicide 0.87 (0.79−0.96)
Received lithium within 14 days after attempted suicide 0.87 (0.78−0.96)
Received lithium within 30 days after attempted suicide 0.88 (0.79−0.97)
Redefinition of suicide-related events
Events of determined intent 0.86 (0.77−0.95)
Events with undetermined intent 0.78 (0.53−1.15)
Events with suicide attempt only 0.89 (0.80−0.98)
Medication with thyroid hormones 1.01 (0.86−1.19)
Occurrence of any bone fracture 0.95 (0.86−1.04)

0.45 0.60 0.80 1.0 1.2

Hazard Ratio (95% CI)
a
Stratified Cox regression was applied for all analyses with adjustment for valproate medication (for medication with thyroid hormones, both lithium
and valproate medication were adjusted), categorical age, and previous number of suicide attempts. Concurrent treatment with other drugs
(lamotrigine, antipsychotics, antidepressants, anticonvulsants, and benzodiazepines) was defined by using the same method used for lithium and
valproate. Patients with bipolar I disorder and bipolar II disorder were identified by linking to the Swedish Bipolar Quality Register. The International
Classification of Diseases (ICD) discharge diagnoses (and codes) were as follows: mixed episode (ICD-10: F316), suicide-related events with
determined intent (ICD-10: X60–X84), suicide-related events with undetermined intent (ICD-10: Y10–Y34), and any fracture (ICD-10: S02, S07, S12,
S22, S32, S42, S52, S62, S72, S82, S92). The Anatomical Therapeutic Chemical (ATC) classification code for medication with thyroid hormones is H03.

attributable fraction was estimated as 12% (95% CI 4%2 although not all significant, for patients with bipolar I dis-
20%), which suggests that 12% of the suicide-related events order (hazard ratio 0.85, 95% CI 0.67–1.09), bipolar II dis-
could have been avoided if the patients would have been order (hazard ratio 0.62, 95% CI 0.46–0.82), and mixed
treated with lithium during the entire follow-up. The results episodes (hazard ratio 0.87, 95% CI 0.74–1.03). Notably,
remained consistent when we followed a shorter period with suicide-related events were less common (980 events in
time-invariant medication by using 1:1 propensity score 24,631 patients) when we restricted the subjects to patients
matching (hazard ratio 0.45, 95% CI 0.29–0.68 for lithium without comorbid psychiatric conditions; further investiga-
monotherapy; hazard ratio 0.93, 95% CI 0.50–1.73 for tion showed that the majority of suicide-related events oc-
valproate monotherapy; see Table S4 in the online data curred among patients with comorbid substance use (7,976
supplement). events in 15,927 patients) and that lithium remained asso-
Results for additional analyses to test the robustness of the ciated with reduced suicide-related events in this group
association between lithium treatment and reduced suicide- (hazard ratio 0.84, 95% CI 0.75–0.94). In contrast to the
related events are shown in Figure 1. None of these analyses results for lithium, there was no evidence of an association
showed any substantive difference from our main results. between suicide-related events and thyroid therapy among
The inverse association between suicide-related events patients with bipolar disorder (hazard ratio 1.01, 95% CI
and lithium remained after adjustment for other concur- 0.86–1.19).
rent psychiatric medication (for hazard ratios for other Analyses for lithium plus valproate are presented in Table
psychiatric medications, see Table S2 in the online data S3 in the data supplement and yielded no substantial dif-
supplement). Subgroup analyses also showed reduced rates, ference from lithium alone. Results of medication periods

ajp in Advance ajp.psychiatryonline.org 5

SUICIDAL BEHAVIOR DURING LITHIUM AND VALPROATE TREATMENT

TABLE 3. Risk of Suicide-Related Events During Lithium and Valproate Treatment for Patients With (14). It is worth noting that
Bipolar Disorder in Sweden (2005–2013), by Length of Treatment observational studies and
Within-Individual Analysisa Between-Individual Analysisb meta-analyses constitute the
Medication Status Hazard Ratioc 95%CI Hazard Ratioc 95%CI main sources of evidence for
Lithium
lithium’s antisuicidal effects
On for less than 30 days 0.71 0.58–0.88 0.70 0.60–0.82 (17, 19) and that adequately
On for more than 30 days 0.84 0.75–0.94 0.78 0.70–0.87 powered randomized con-
Off for less than 30 days 1.33 1.09–1.61 1.37 1.21–1.56 trolled trials are lacking.
Valproate Our results, with emphasis
On for less than 30 days 1.00 0.78–1.29 1.07 0.88–1.30 on high-risk patients, aug-
On for more than 30 days 0.97 0.84–1.12 0.97 0.86–1.09
ment existing evidence and
Off for less than 30 days 1.26 0.95–1.68 1.34 1.11–1.62
cumulatively support the
a
Stratified Cox regression was applied with adjustment for time-varying covariates including categorical age and previous
hypothesis that lithium is
number of suicide attempts
b
Ordinary Cox regression was applied with adjustment for the same covariates as in the stratified Cox regression and, protective against suicidal
additionally, with adjustment for time-invariant covariates including sex, length of baseline hospitalization periods due behavior. Additionally, our
to psychiatric admissions, and history of suicide-related events before entering follow-up.
c separate analyses on definite
Hazard ratio is the ratio of the hazard rates during periods of different medication status (i.e., on for less than 30 days, on
for more than 30 days, off for less than 30 days) compared with periods off medication for more than 30 days, for and uncertain events showed
lithium and valproate separately. no material difference. Fu-
ture research on the mecha-
nisms behind the association
defined by time-varying cutoffs are shown in Table 3. It is between lithium and suicidal behavior is warranted and could
noteworthy that patients had an increased rate of suicidal inform the neurobiology of suicidal behavior.
behavior within 30 days of lithium discontinuation (hazard In the analyses of people treated with lithium within 1
ratio 1.33, 95% CI 1.09–1.61). year after the first diagnosis of bipolar disorder, we observed
similar reductions in suicide-related events, corroborating
the positive effects of lithium prescription in this group (34).
DISCUSSION
Similar conclusions can be drawn for the high-risk group of
Using a long follow-up period and what we believe is the those with mixed episodes. We observed nonsignificantly
largest sample ever reported, we found that rates of suicide- reduced suicide-related events during lithium treatment for
related events were significantly decreased during lithium bipolar I disorder and a significantly reduced rate for bipolar
treatment but not valproate treatment, with a possible II disorder. This suggests that further exploration of sub-
difference between them. Since the within-individual groups within subtypes could be worthwhile. A history of
analyses drew information exclusively from people who substance use disorder predicts an increased rate of suicidal
attempted suicide during follow-up, our results demon- behavior in bipolar disorder (3), and here we provide further
strated that the association between lithium and reduced evidence of lithium’s protective effect in this high-risk sub-
suicide-related events existed even among a high-risk group. Additionally, we observed a higher rate of suicide-
population, which is unlikely to be studied in randomized related events shortly after lithium discontinuation. Although
controlled trials. Moreover, our suggestive between-drug misclassification of medication periods cannot be ruled out,
differences supported evidence of unequal antisuicidal ef- this finding, along with previous reports (32), indicates a need
fects for lithium and valproate. Finally, we estimated that, in for close monitoring after discontinuation.
the absence of potential confounding, more than 10% of For valproate, we found no protective effect for suicide-
suicide-related events could have been prevented if all related events, with a significant difference between lithium
patients had been treated with lithium during the entire and valproate. Moreover, the combination of lithium plus
follow-up. By comparing treatment and nontreatment pe- valproate showed no improvement. One study of U.S. Med-
riods within the same individual, our approach automati- icaid patients revealed a greater risk of suicide attempts
cally controlled for all time-stationary confounders and thus among valproate users compared with lithium users, but no
reduces the likelihood of confounding by indication in an significant difference for completed suicide (35). Two Danish
observational study. Therefore, the observed inverse asso- studies found similarly reduced risks of completed suicide
ciation supports the hypothesis of an antisuicidal effect of among consistent lithium and anticonvulsant purchasers,
lithium. whereas for patients beginning anticonvulsant treatment,
Our finding of reduced suicide-related events during only a switch to or addition of lithium yielded reduced suicide
lithium treatment is in line with findings in many previous rates (11, 12). Compared with findings in the previous liter-
studies (7, 8), including a recent U.K. study that found de- ature, our conclusion, by design, emphasizes the concomitant
creased rates of self-harm and unintentional injury after effect of drugs in patients with histories of medication
treatment with lithium and suggested that the mecha- switches and suicidal behavior. While current randomized
nism could be a lithium-induced reduction in impulsivity controlled trials lack the power to detect a difference of effect

6 ajp.psychiatryonline.org ajp in Advance

 SONG ET AL.

between lithium and valproate (18), our study suggests that lithium treatment and the rate of bone fractures, which
there might be a distinct association between lithium and suggests that lithium’s effect could be suicide-specific. Fi-
antisuicidal behavior, as compared with valproate (36). This nally, we tested the findings with follow-up of periods
finding calls into question whether valproate can be said to without medication changes and used propensity score
have the same antisuicidal effect as lithium. It is therefore analyses to limit potential confounding measurable at base-
worrying that lithium use has decreased steadily during re- line; the conclusions remained consistent.
cent years in Sweden (37), probably due to easier dose op- Despite these efforts, observational studies like ours
timization and greater safety in case of overdose for valproate cannot exclude a potential lithium-specific effect due to the
in clinical practice. It remains to be seen whether these required routine blood tests during lithium treatment to
changes alter the rates of suicidal behavior in the patient monitor for rare side effects. It is also possible that clinicians
population. hesitate to prescribe lithium to high-risk patients because of
Compared with clinical studies and randomized con- the potential for lethal toxicity in case of overdose. Altogether,
trolled trials, our large population-based longitudinal sample more evidence from randomized controlled trials with large
is representative of the population, thereby avoiding selective samples and long follow-up are warranted to address these
participation, which threatens validity and generalizability. questions.
The information on medication is complete and free from Three additional issues should be considered when inter-
recall bias. To reduce selection effects and confounding by preting the results. First, we cannot evaluate the associations
indication, we used a within-individual design that controls with completed suicide directly, as stratified Cox regression is
for unobservable time-invariant confounding for each per- not applicable to nonrepeated events. However, the percentage
son during the follow-up, as well as measured time-varying of completed suicides was small (590 in 10,648 suicide-related
confounding. events, 5.5%), and analyses excluding these events yielded
However, within-individual designs cannot exclude the similar results. Second, the estimate of population attributable
potential existence of unmeasured time-varying confounding fraction applies to our study population only during the fol-
within each person, including varying severity of symptoms low-up period, since the estimation is based on exposure
of illness (e.g., rapid cycling) (38), frequency of hospital visits, proportion and thus will vary depending on prescription rates
treatment site, and other types of concomitant treatment. and medication adherence. Third, it would have been ideal to
We tried to address many possible alternative explana- have information on suicidal behavior for patients who had
tions for the inverse association between lithium and suicide- switched between lithium and valproate; however, we lacked
related events. Consistently, no material differences were power to perform this critical comparison.
found in a series of sensitivity analyses (Figure 1), indicating In summary, we demonstrated an association between
that misclassifications of disease, exposure, and outcome are lithium and reduced suicide-related events, whereas the
unlikely to invalidate the results. To account for the effect on results provided no equivalent effect of valproate. Our results,
prescription by previous suicide history, our main models set in conjunction with existing literature, indicate that in pa-
the underlying time scale as the time since the last suicide tients with bipolar disorder and suspected suicidal intentions,
attempt. Unavoidable limitations include the lack of in- lithium should be considered as a suicide preventive strategy,
formation regarding adherence, similar to the situation in with a balance between efficacy and tolerability.
intention-to-treat analyses in randomized controlled trials,
but if anything, this would result in underestimation of our AUTHOR AND ARTICLE INFORMATION
reported associations. Our conservative way of defining the From the Department of Medical Epidemiology and Biostatistics and
end of the medication period is another possible source of the Department of Clinical Neuroscience, Centre for Psychiatry Re-
underestimation (i.e., individuals classified as off medica- search, Karolinska Institutet, Stockholm; the Institute of Neuroscience
and Physiology, The Sahlgrenska Academy at Gothenburg University,
tion at the date of last dispensed prescription were probably
Gothenburg, Sweden; the Stanley Center for Psychiatric Research, Broad
truly medicated). Lithium’s antisuicidal impact might have Institute of the Massachusetts Institute of Technology and Harvard Uni-
a delayed effect after the beginning of treatment, but our versity, Cambridge, Mass.; and the Department of Medical Sciences,
sensitivity analysis using different measures of start, end, and Örebro University, Örebro, Sweden.
length of medication period resulted in consistent estimates. Address correspondence to Dr. Song ([email protected]).
One hypothesis posits that the reduced rate of suicidal be- Supported in part by the Swedish Research Council, the Swedish Council
havior is due to active treatment rather than the effect of the for Health, Working Life and Welfare, the Swedish Foundation for Strategic
drugs themselves. Another source of confounding could be Research (KF10-0039), and the China Scholarship Council.
that factors impacting a change in medication (e.g., stressful Dr. Larsson has served as a speaker for Eli Lilly and Shire and has received
life events) may also impact the risk of suicidal behavior. research grants from Shire. Over the past 36 months, Dr. Landén has
received lecture honoraria from Lundbeck and AstraZeneca Sweden and
To test for these hypotheses, we estimated rates of suicide-
has served as scientific consultant for EPID Research Oy. Dr. Lichtenstein
related events among patients who received thyroid hor- has served as a speaker for Medice. The other authors report no financial
mone medication (which is believed to have no effect on relationships with commercial interests.
bipolar disorder or suicidal behavior) and found no asso- Received May 11, 2016; revisions received Nov. 7, 2016, Jan. 13 and March
ciations. Furthermore, we found no association between 7, 2017; accepted March 27, 2017.

ajp in Advance ajp.psychiatryonline.org 7

SUICIDAL BEHAVIOR DURING LITHIUM AND VALPROATE TREATMENT

REFERENCES 20. Freeman TW, Clothier JL, Pazzaglia P, et al: A double-blind com-
1. Nordentoft M, Mortensen PB, Pedersen CB: Absolute risk of suicide parison of valproate and lithium in the treatment of acute mania. Am
after first hospital contact in mental disorder. Arch Gen Psychiatry J Psychiatry 1992; 149:108–111
2011; 68:1058–1064 21. Lichtenstein P, Halldner L, Zetterqvist J, et al: Medication for at-
2. Chen YW, Dilsaver SC: Lifetime rates of suicide attempts among tention deficit-hyperactivity disorder and criminality. N Engl J Med
subjects with bipolar and unipolar disorders relative to subjects with 2012; 367:2006–2014
other axis I disorders. Biol Psychiatry 1996; 39:896–899 22. Ludvigsson JF, Otterblad-Olausson P, Pettersson BU, et al: The
3. Webb RT, Lichtenstein P, Larsson H, et al: Suicide, hospital- Swedish personal identity number: possibilities and pitfalls in
presenting suicide attempts, and criminality in bipolar disorder: healthcare and medical research. Eur J Epidemiol 2009; 24:659–667
examination of risk for multiple adverse outcomes. J Clin Psychiatry 23. Wettermark B, Hammar N, Fored CM, et al: The new Swedish
2014; 75:e809–e816 Prescribed Drug Register—opportunities for pharmacoepidemio-
4. Latalova K, Kamaradova D, Prasko J: Suicide in bipolar disorder: a logical research and experience from the first six months. Phar-
review. Psychiatr Danub 2014; 26:108–114 macoepidemiol Drug Saf 2007; 16:726–735; correction, 2008; 17:533
5. Cipriani A, Reid K, Young AH, et al: Valproic acid, valproate and 24. Sellgren C, Landén M, Lichtenstein P, et al: Validity of bipolar
divalproex in the maintenance treatment of bipolar disorder. Cochrane disorder hospital discharge diagnoses: file review and multiple
Database Syst Rev 2013; 10:CD003196 register linkage in Sweden. Acta Psychiatr Scand 2011; 124:447–453
6. Tondo L, Baldessarini RJ: Suicidal behavior in mood disorders: 25. Molero Y, Lichtenstein P, Zetterqvist J, et al: Selective serotonin
response to pharmacological treatment. Curr Psychiatry Rep 2016; reuptake inhibitors and violent crime: a cohort study. PLoS Med
18:88 2015; 12:e1001875
7. Kessing LV, Søndergård L, Kvist K, et al: Suicide risk in patients 26. Allison PD: Fixed Effects Regression Models. Thousand Oaks, Calif,
treated with lithium. Arch Gen Psychiatry 2005; 62:860–866 SAGE Publications, 2009
8. Baldessarini RJ, Tondo L, Davis P, et al: Decreased risk of suicides 27. Chang Z, Lichtenstein P, D’Onofrio BM, et al: Serious transport
and attempts during long-term lithium treatment: a meta-analytic accidents in adults with attention-deficit/hyperactivity disorder and
review. Bipolar Disord 2006; 8:625–639 the effect of medication: a population-based study. JAMA Psychiatry
9. Tondo L, Hennen J, Baldessarini RJ: Lower suicide risk with long- 2014; 71:319–325
term lithium treatment in major affective illness: a meta-analysis. 28. Chen Q, Sjölander A, Runeson B, et al: Drug treatment for attention-
Acta Psychiatr Scand 2001; 104:163–172 deficit/hyperactivity disorder and suicidal behaviour: register based
10. Goodwin FK, Fireman B, Simon GE, et al: Suicide risk in bipolar study. BMJ 2014; 348:g3769
disorder during treatment with lithium and divalproex. JAMA 2003; 29. Bruzzi P, Green SB, Byar DP, et al: Estimating the population at-
290:1467–1473 tributable risk for multiple risk factors using case-control data. Am J
11. Smith EG, Søndergård L, Lopez AG, et al: Association between Epidemiol 1985; 122:904–914
consistent purchase of anticonvulsants or lithium and suicide risk: a 30. Scheen L, Brandt L, Bodén R, et al: Predictors for initiation of
longitudinal cohort study from Denmark, 1995-2001. J Affect Disord pharmacological prophylaxis in patients with newly diagnosed bi-
2009; 117:162–167 polar disorder—a nationwide cohort study. J Affect Disord 2015; 172:
12. Søndergård L, Lopez AG, Andersen PK, et al: Mood-stabilizing 204–210
pharmacological treatment in bipolar disorders and risk of sui- 31. BALANCE investigators and collaborators, Geddes JR, Goodwin GM,
cide. Bipolar Disord 2008; 10:87–94 et al: Lithium plus valproate combination therapy versus monotherapy
13. Ahearn EP, Chen P, Hertzberg M, et al: Suicide attempts in veterans for relapse prevention in bipolar I disorder (BALANCE): a randomised
with bipolar disorder during treatment with lithium, divalproex, and open-label trial. Lancet 2010; 375: 385–395
atypical antipsychotics. J Affect Disord 2013; 145:77–82 32. Smith EG, Austin KL, Kim HM, et al: Suicide risk in Veterans Health
14. Hayes JF, Pitman A, Marston L, et al: Self-harm, unintentional injury, Administration patients with mental health diagnoses initiating
and suicide in bipolar disorder during maintenance mood stabilizer lithium or valproate: a historical prospective cohort study. BMC
treatment: a UK population-based electronic health records study. Psychiatry 2014; 14:357
JAMA Psychiatry 2016; 73:630–637 33. StataCorp: Stata: Release 13. College Station, Tex, StataCorp, 2013
15. Baldessarini RJ, Tondo L: Lithium and suicidal risk. Bipolar Disord 34. Nery-Fernandes F, Quarantini LC, Guimarães JL, et al: Is there an
2008; 10:114–115 association between suicide attempt and delay of initiation of mood
16. Baldessarini RJ, Tondo L: Suicidal risks during treatment of bipolar stabilizers in bipolar I disorder? J Affect Disord 2012; 136:1082–1087
disorder patients with lithium versus anticonvulsants. Pharma- 35. Collins JC, McFarland BH: Divalproex, lithium and suicide among
copsychiatry 2009; 42:72–75 Medicaid patients with bipolar disorder. J Affect Disord 2008; 107:
17. Cipriani A, Hawton K, Stockton S, et al: Lithium in the prevention of 23–28
suicide in mood disorders: updated systematic review and meta- 36. Wasserman D, Rihmer Z, Rujescu D, et al: The European Psychiatric
analysis. BMJ 2013; 346:f3646 Association (EPA) guidance on suicide treatment and prevention.
18. Oquendo MA, Galfalvy HC, Currier D, et al: Treatment of suicide Eur Psychiatry 2012; 27:129–141
attempters with bipolar disorder: a randomized clinical trial com- 37. Karanti A, Kardell M, Lundberg U, et al: Changes in mood stabilizer
paring lithium and valproate in the prevention of suicidal behavior. prescription patterns in bipolar disorder. J Affect Disord 2016; 195:
Am J Psychiatry 2011; 168:1050–1056 50–56
19. Schaffer A, Isometsä ET, Tondo L, et al: Epidemiology, neurobiology 38. Schaffer A, Isometsä ET, Azorin JM, et al: A review of factors as-
and pharmacological interventions related to suicide deaths and sociated with greater likelihood of suicide attempts and suicide
suicide attempts in bipolar disorder: part I of a report of the In- deaths in bipolar disorder: part II of a report of the International
ternational Society for Bipolar Disorders Task Force on Suicide in Society for Bipolar Disorders Task Force on Suicide in Bipolar
Bipolar Disorder. Aust N Z J Psychiatry 2015; 49:785–802 Disorder. Aust N Z J Psychiatry 2015; 49:1006–1020

8 ajp.psychiatryonline.org ajp in Advance

View publication stats