Research

JAMA Psychiatry | Original Investigation

Real-world Effectiveness of Pharmacologic Treatments

for the Prevention of Rehospitalization in a Finnish
Nationwide Cohort of Patients With Bipolar Disorder
Markku Lähteenvuo, MD, PhD; Antti Tanskanen, PhLic; Heidi Taipale, PhD; Fabian Hoti, PhD; Pia Vattulainen, MSc;
Eduard Vieta, MD, PhD; Jari Tiihonen, MD, PhD

Editorial page 314

IMPORTANCE Mood stabilizers and antipsychotics are the main maintenance treatments for Supplemental content
bipolar disorder. Lithium is considered to be the most effective mood stabilizer, but very little
is known about overall health outcomes associated with specific treatments and the
comparative long-term effectiveness of specific psychotropics or routes of administration in
the prevention of rehospitalizations.

OBJECTIVE To study the comparative effectiveness of pharmacologic treatments in the

prevention of rehospitalization in a nationwide cohort of patients with bipolar disorder.

DESIGN, SETTING, AND PARTICIPANTS This cohort study examined the risk of psychiatric,
cardiovascular, and all-cause hospitalization from January 1, 1987, to December 31, 2012,
among all patients in Finland who had been hospitalized for bipolar disorder (N = 18 018;
mean follow-up time, 7.2 years) using prospectively gathered nationwide databases for
hospitalization and dispensed medications. The primary analysis was within-individual
analysis, in which each individual was used as his or her own control to eliminate selection
bias. The study adjusted for the effect of concomitant psychotropic medications, duration of
illness, and the temporal orders of exposure and nonexposure periods. Statistical analysis was
conducted from January 1, 1996, to December 31, 2012.

MAIN OUTCOMES AND MEASURES Adjusted hazard ratios (HRs) for rehospitalization were
calculated.

RESULTS Among the cohort (9558 women and 8460 men; mean [SD] age, 46.6 [17.0] years),
9721 patients (54.0%) had at least 1 psychiatric rehospitalization. In comparison between use
and no use among specific agents reaching nominal statistical significance, risperidone
long-acting injection (HR, 0.58 [95% CI, 0.34-1.00]), gabapentin (HR, 0.58 [95% CI,
0.44-0.77]), perphenazine long-acting injection (HR, 0.60 [95% CI, 0.41-0.88]), and lithium
carbonate (HR, 0.67 [95% CI, 0.60-0.73]) were associated with the lowest risk of psychiatric
rehospitalization. Concerning all-cause hospitalization, lithium (HR, 0.71 [95% CI, 0.66-0.76])
was associated with the lowest risk. The most frequently used antipsychotic treatment,
quetiapine fumarate, showed only modest effectiveness (risk of psychiatric rehospitalization:
HR, 0.92 [95% CI, 0.85-0.98]; risk of all-cause hospitalization: HR, 0.93 [95% CI,
0.88-0.98]). Long-acting injections were associated with substantially better outcomes
compared with identical oral antipsychotics (risk of psychiatric rehospitalization: HR, 0.70
[95% CI, 0.55-0.90]; risk of all-cause hospitalization: HR, 0.70 [95% CI, 0.57-0.86]). Results
from sensitivity analyses showed consistent beneficial effects only for lithium and for
long-acting injections compared with their oral counterparts.

CONCLUSIONS AND RELEVANCE Lithium was the most effective mood stabilizer, and
long-acting injections the most effective antipsychotics, in preventing hospitalization due to Author Affiliations: Author
affiliations are listed at the end of this
mental or physical illness.
article.
Corresponding Author: Jari
Tiihonen, MD, PhD, Department of
Clinical Neuroscience, Karolinska
Institutet, Byggnad R5, S-17176
JAMA Psychiatry. 2018;75(4):347-355. doi:10.1001/jamapsychiatry.2017.4711 Stockholm, Sweden
Published online February 28, 2018. (jari.tiihonen@ki.se).

(Reprinted) 347
© 2018 American Medical Association. All rights reserved.
 Research Original Investigation
B
ipolar disorder is a serious, often debilitating, and re-
curring chronic psychiatric disorder,1 although indi-
vidual disease courses may vary. Bipolar disorder is rated
as the sixth most common cause of disability in the world by
the World Health Organization, accounting for more loss of
disability-adjusted life years than all cancers combined.2 Long-
term medication is often required to attain remission and pre-
vent relapses, although, even with advanced treatment proto-
cols, rates of remission remain low.3 Many randomized clinical
trials and their meta-analyses have been performed to dis-
cover the most efficacious treatment forms for prevention of re-
lapse; some more recent trials also included antipsychotics in
long-acting injectable (LAI) form.4,5 However, these studies did
not include oral antipsychotics, and the comparative effective-
ness of LAI vs oral antipsychotics has remained unknown.
Some meta-analyses have shown clinically meaningful
differences between pharmacotherapies in their compara-
tive effectiveness of acute treatment,6-8 but treatment guide-
lines until recently did not recognize such differences.9,10 How-
ever, as has been noted in the field of depression research,
randomized clinical trials often have stringent inclusion and
exclusion criteria and limited follow-up times, which may im-
pair their ability to reflect the real-world effectiveness of thera-
pies in their actual clinical application.11 Randomized clinical
trials usually exclude patients with the most comorbidities and
the highest severity of illness (eg, propensity to suicide). Fur-
thermore, many efficient medications might cause adverse ef-
fects on a very long time scale. Thus, the most efficient way
to obtain a good estimate on the overall real-world effective-
ness of therapies for bipolar disorder may be through obser-
vational studies.12 One such recent study by Hayes et al13
indicated that lithium appears to be more successful as mono-
therapy maintenance treatment for bipolar disorder than val-
proate sodium, olanzapine, or quetiapine fumarate. Similar
findings on the superiority of lithium compared with other
pharmacologic treatments in bipolar disorder have been
reported by Kessing et al 14 in a 2011 registry study from
Denmark and by Simhandl et al15 in a 4-year naturalistic
follow-up study from Austria in 2014. Joas et al16 suggested that
lithium is more effective than both quetiapine and olanza-
pine in prevention of relapse in bipolar disorder, and a study
by Song et al17 suggested that lithium is superior to valproate
sodium in preventing suicidal behavior in bipolar patients.
However, these studies were limited to only a few phar-
macologic agents. To our knowledge, no studies have inves-
tigated the comparative effectiveness of long-acting antipsy-
chotic injections vs identical oral agents. This investigation
aimed to overcome these shortcomings in the largest and most
complete register-based, real-world effectiveness study of
pharmacotherapies for bipolar disorder to date.
Methods
Study Design and Data Acquisition
Finnish nationwide databases were used to combine prospec-
tively collected registry data to conduct a population-based co-
hort study of patients hospitalized for bipolar disorder. The reg-
348 JAMA Psychiatry April 2018 Volume 75, Number 4 (Reprinted)
© 2018 American Medical Association. All rights reserved.
Real-world Effectiveness of Pharmacologic Treatments in Bipolar Disorder
Key Points
Question What is the comparative effectiveness of
pharmacologic treatments in the prevention of rehospitalization in
bipolar disorder?
Findings In this Finnish nationwide cohort study of 18 018
patients, lithium use was associated with the lowest risk of
rehospitalization because of mental or somatic disorder. The risk
of rehospitalization was about 30% lower during treatment with
long-acting injections compared with treatment with their oral
counterparts.
Meaning In bipolar disorder, lithium should remain the first line of
treatment, and long-acting injections might offer a safe, effective
option for patients in whom lithium is not suitable.
isters were used to identify the study cohort (patients
hospitalized for bipolar disorder between January 1, 1987, and
December 31, 2012); to determine the incidences, durations,
and reasons for rehospitalizations; to obtain information on
reimbursed medications dispensed from pharmacies (all psy-
chotropic medications except small packages of benzodiaz-
epines were reimbursed in this indication); and to retrieve in-
formation on deaths. The databases and their use have been
described in more detail in previous pharmacoepidemiologic
studies.18-21 For details, see the eAppendix in the Supple-
ment. The cohorts are described in detail in the Table. This re-
search project was approved by the Ethics Committee of the
Finnish National Institute for Health and Welfare. Further per-
missions were granted by pertinent institutional authorities
at the Finnish National Institute for Health and Welfare (per-
mission THL/1466/6.02.00/2013), the Social Insurance Insti-
tution of Finland (34/522/2013), and Statistics Finland (TK53-
305-13). Informed consent is not required for register-based
studies using anonymized data.
Exposure
The PRE2DUP (Prescription drug purchases to Drug Use Peri-
ods) method was used to define exposure and nonexposure
periods for medications.18-21 The PRE2DUP method calcu-
lates the current dose with a sliding mean, uses package in-
formation (eg, number of tablets and administration inter-
vals for injections), and takes into account stockpiling when
constructing time periods of continuous use. Previous publi-
cations on the validation of the method indicate that PRE2DUP
is the most precise method currently available to estimate drug
use, and it gives highly accurate drug use periods for most drug
classes, especially those meant for long-term use.19,20 As varia-
tion in dose is allowed within the method, no artificial grace
periods are used. Thus, the risk is attributed to the ongoing
treatment(s) according to the PRE2DUP method for each day.
Concerning switching and cross-titration of treatments, the ex-
posure of the first drug is defined to last until the whole amount
of the purchased drug would be used as calculated with
PRE2DUP, usually overlapping the use of the second drug.
Therefore, if the hospitalization takes place within a few weeks
after a change in medication status, the rehospitalization is typi-
cally attributed to both treatments. Antipsychotics were de-
fined as Anatomical Therapeutic Chemical code N05A except
jamapsychiatry.com
Real-world Effectiveness of Pharmacologic Treatments in Bipolar Disorder Original Investigation Research

Table. Characteristics of Study Cohorts Table. Characteristics of Study Cohorts (continued)

Patients, No. (%) Patients, No. (%)

Total Incident Total Incident
Cohort Cohort Cohort Cohort
Covariate (N = 18 018) (n = 2074) Covariate (N = 18 018) (n = 2074)
Sex Use of mood stabilizers during follow-up
Male 8460 (47.0) 1111 (53.6) No 4749 (26.4) 827 (39.9)
Female 9558 (53.0) 963 (46.4) Yes 13 269 (73.6) 1247 (60.1)
Age at cohort entry date, y Use of antidepressants during follow-up
<30 3345 (18.6) 588 (28.4) No 4832 (26.8) 1182 (57.0)
30-49 7121 (39.5) 689 (33.2) Yes 13 186 (73.2) 892 (43.0)
50-69 5577 (31.0) 534 (25.8) Use of antipsychotics during follow-up
≥70 1975 (11.0) 263 (12.7) No 3413 (18.9) 603 (29.1)
Calendar year of cohort entry date Yes 14 605 (81.1) 1471 (70.9)
1996-1999 5107 (28.3) 343 (16.5)
2000-2003 3102 (17.2) 501 (24.2)
for lithium carbonate (N05AN01); antidepressants as N06A,
2004-2007 4280 (23.8) 573 (27.6)
mood stabilizers as N03AF, N03AG, N03AX, and N05AN01
2008-2012 5529 (30.7) 657 (31.7)
(lithium); benzodiazepines as N05BA; and sedatives as N05C.
Patients remaining in cohort after censoring
hospitalizations longer than 30.5 d Antipsychotics were modeled according to drug form (ie, oral
No 105 (0.6) 13 (0.6) and LAIs separately) based on package information recorded
Yes 17 913 (99.4) 2061 (99.4) for each filling of a prescription.
Patients remaining in cohort after censoring
hospitalizations longer than 0 d Statistical Analysis
No 141 (0.8) 22 (1.1)
Statistical analysis was conducted from January 1, 1996, to
Yes 17 877 (99.2) 2052 (98.9)
December 31, 2012. Three analyses were performed on the
Time since diagnoses at cohort entry date, y
cohorts using within-individual Cox proportional hazards re-
0-5 16 531 (91.8) 2074 (100.0)
gression analysis. First, rehospitalization owing to any
>5-10 1472 (8.2) 0
mental disorder (International Statistical Classification of
>10 15 (0.1) 0
Diseases and Related Health Problems, Tenth Revision
Time since diagnoses at end of follow-up, y
[ICD-10], diagnosis codes beginning with the letter F) was in-
0-5 7120 (39.5) 870 (42.0)
vestigated as a proxy marker for treatment failure (effective-
>5-10 5357 (29.7) 669 (32.3)
ness; primary analysis). Second, all-cause hospitalization
>10 5541 (30.8) 535 (25.8)
(including for somatic reasons) was used as a proxy marker for
Two-year history of all-cause
hospitalizations at cohort entry date effectiveness vs tolerability of treatments (ie, taking into
0 1364 (7.6) 0 account the hospitalizations owing to severe adverse effects;
1-2 12 129 (67.3) 1872 (90.3) secondary analysis). Third, hospitalization owing to cardio-
>2 4525 (25.1) 202 (9.7) vascular diseases (ICD-10 diagnosis codes beginning with the
Two-year history of all-cause letter I) was used as a proxy for cardiovascular tolerability (sec-
hospitalizations at end of follow-up
ondary analysis). In addition, an analysis for rehospitalization
0 7321 (40.6) 1055 (50.9)
for somatic reasons was performed for lithium, valproic acid,
1-2 6756 (37.5) 765 (36.9)
and quetiapine, the 3 most frequently used medications.
>2 3941 (21.9) 254 (12.3)
The analyses used the within-individual Cox propor-
Two-year history of any psychiatric
hospitalizations before cohort entry date tional hazards regression model in which each individual is
0 1777 (9.9) 2074 (100.0) assigned his or her own stratum and the follow-up time is
1-2 14 097 (78.2) 0 reset at the initiation of a new treatment, thus eliminating
>2 2144 (11.9) 0 selection bias (eFigure in the Supplement).21 Other time-
Two-year history of any psychiatric dependent adjusting variables considered to increase the re-
hospitalizations at end of follow-up
liability of the analysis are detailed in eTables 1 and 2 in the
0 10 997 (61.0) 1435 (69.2)
Supplement (for all analyses: the effect of time since diagno-
1-2 5312 (29.5) 574 (27.7)
sis, order of treatments, current use of other treatments, and
>2 1709 (9.5) 65 (3.1)
polypharmacy; for the traditional analysis models: number of
Use of benzodiazepines during follow-up
No 8580 (47.6) 1466 (70.7)
rehospitalizations within 2 years prior to index date [indica-
Yes 9438 (52.4) 608 (29.3)
tor of inherent risk of relapse], age at index date, sex, and cal-
Use of sedatives during follow-up
endar year of index date). P < .05 was considered significant.
No 8069 (44.8) 1306 (63.0) The P values shown in the figures were corrected for multiple
Yes 9949 (55.2) 768 (37.0) comparisons using the Benjamini-Hochberg false discovery rate
method. The P values shown in the eTables have not been cor-
(continued)
rected for multiple comparisons. The antipsychotic thiorida-

jamapsychiatry.com (Reprinted) JAMA Psychiatry April 2018 Volume 75, Number 4 349

© 2018 American Medical Association. All rights reserved.

 Research Original Investigation
zine hydrochloride was included in the original analyses. How-
ever, since it was withdrawn from the market in the middle of
the observation period in 2006, no data for thioridazine are
shown because they might be skewed. The analyses for thio-
ridazine were incorporated into the multiple comparisons
corrections, so as to not underestimate type I error.
As observational studies are prone to bias and confound-
ing, a number of sensitivity analyses were performed. As a gen-
eral sensitivity analysis, the within-individual Cox propor-
tional hazards regression analyses for risk of psychiatric and
all-cause rehospitalization were also performed by doing time
resets at outcome (hospitalization) rather than at initiation of
a new treatment (eTables 3 and 4 in the Supplement). To ac-
count for putative survival bias (selective mortality owing to
previous treatments) for the primary analysis, the primary
analysis for psychiatric rehospitalization was also performed
in the incident cohort (eTable 5 in the Supplement). A total of
8714 patients did not have psychiatric rehospitalizations, 4609
did not have all-cause hospitalizations, and 15 662 did not have
cardiovascular hospitalizations or variation in the exposure
and, therefore, did not contribute to the corresponding within-
individual analysis. To test for generalizability in the total co-
hort population (also including patients without any out-
come incidents or any changes in their medication during
follow-up), traditional between-groups Cox proportional haz-
ards regression analyses were also performed for all primary
and secondary outcomes (psychiatric hospitalization, all-
cause hospitalization, and cardiovascular hospitalization)
(eTables 6-8 in the Supplement).
Results
The mean (SD) age of the cohort was 46.6 (17.0) years, and the
cohort included 9558 women and 8460 men. A total of 9721 pa-
tients (54.0%) had at least 1 psychiatric rehospitalization. The so-
ciodemographic, clinical, and treatment characteristics of the
total cohort and the incident cohorts are shown in the Table. The
study cohort consisted of 18 018 individuals with a total obser-
vation time of 128 353 person-years, with a mean observation
(follow-up) time of 7.2 years (range, 1 day-17.0 years). Since 141
patients were hospitalized during their entire follow-up, 17 877
patients were at risk for rehospitalization. During this period, we
observed a total of 82 858 hospitalizations (for any cause; approxi-
mately 4.6 per individual), of which 36 131 were psychiatric
hospitalizations (approximately 2.0 per individual) and 4862
were for cardiovascular reasons (approximately 0.3 per indi-
vidual). The person-years and events of rehospitalization are
shown in eTables 3, 4, and 8 in the Supplement.
Figure 1, Figure 2, and Figure 3 display hazard ratios (HRs)
for hospitalization during use vs no use of specific psychophar-
macologic treatments. The results for psychiatric rehospital-
ization are shown in Figure 1. As a therapeutic group, only
mood stabilizers were associated with a decreased risk of
relapse leading to psychiatric hospitalization, and the risk re-
duction remains modest (HR, 0.91 [95% CI, 0.86-0.97]).
However, when comparing individual medications instead of
therapeutic groups, more pronounced differences and risk re-
350 JAMA Psychiatry April 2018 Volume 75, Number 4 (Reprinted)
© 2018 American Medical Association. All rights reserved.
Real-world Effectiveness of Pharmacologic Treatments in Bipolar Disorder
ductions emerge. Of all the medications analyzed reaching nomi-
nal statistical significance, risperidone LAI (HR, 0.58 [95% CI,
0.34-1.00]; P = .049) was associated with the lowest risk of
psychiatric rehospitalization, followed by gabapentin (HR, 0.58
[95% CI, 0.44-0.77]), perphenazine LAI (HR, 0.60 [95% CI, 0.41-
0.88]), and lithium (HR, 0.67 [95% CI, 0.60-0.73]). Owing to a
relatively low number of person-years, the result for risperi-
done LAI did not reach significance when correction for mul-
tiple comparisons was applied. The statistically significant re-
sults for gabapentin and sulpiride did not survive sensitivity
analyses with time resets at outcome (hospitalization) (eTable
3 in the Supplement). In general, LAIs were associated with a
significantly lower risk of psychiatric rehospitalizations than
their oral counterparts (HR, 0.70 [95% CI, 0.55-0.90]), al-
though the number of LAIs administered was low and the HRs
for many individual LAI formulations did not reach statistical
significance. In the sensitivity analysis with time reset at out-
come, neither risperidone LAI nor perphenazine LAI re-
mained statistically significant, but in the overall compari-
sons, LAIs were superior to identical oral formulations. The most
frequently used antipsychotic treatment, quetiapine fuma-
rate, showed only modest effectiveness (risk of psychiatric re-
hospitalization: HR, 0.92 [95% CI, 0.85-0.98]). Benzodiaz-
epine use was associated with an increased risk of psychiatric
rehospitalization (HR, 1.19 [95% CI, 1.12-1.26]).
The results for all-cause hospitalizations are shown in
Figure 2. Of all the medications studied, lithium (HR, 0.71 [95%
CI, 0.66-0.76]) and sulpiride (HR, 0.73 [95% CI, 0.59-0.90]) were
associated with the best outcomes. The lowered risk for sulpi-
ride did not survive sensitivity analysis with time reset at out-
come (hospitalization) (eTable 4 in the Supplement). Long-
acting injections were associated with substantially better
outcomes compared with identical oral antipsychotics (HR, 0.70
[95% CI, 0.57-0.86]). Quetiapine again showed only modest ef-
fectiveness (risk of all-cause hospitalization: HR, 0.93; 0.88-
0.98). Benzodiazepine use was associated with an increased risk
of hospitalization for any cause (HR, 1.15 [95% CI, 1.11-1.20]).
The results for cardiovascular hospitalizations are shown
in Figure 3. Mood stabilizers as a therapeutic group were as-
sociated with an increased risk of cardiovascular hospitaliza-
tion (HR, 1.32 [95% CI, 1.10-1.58]). However, of the individual
mood stabilizers, only valproic acid (HR, 1.53 [95% CI, 1.16-
2.01]) and carbamazepine (HR, 1.95 [95% CI, 1.30-2.93]) were
associated with a significantly increased risk of cardiovascu-
lar hospitalization. Apart from the mentioned mood stabiliz-
ers, no other medications studied were associated with a sig-
nificantly altered risk of cardiovascular hospitalization when
false discovery rate correction was applied, although sulpi-
ride (HR, 0.36 [95% CI, 0.16-0.80]) and chlorpromazine (HR,
0.54 [95% CI, 0.31-0.92]) had 95% CIs suggesting an associa-
tion with a lowered risk. In the comparison between the 3 most
frequently used medications, lithium was associated with the
lowest incidence of hospitalization owing to physical illness
(6635 hospitalizations per 24 815 person-years [0.27 hospital-
izations per person-year]), followed by valproate (9849 hos-
pitalizations per 26 091 person-years [0.38 hospitalizations per
person-year]) and quetiapine (8642 hospitalizations per 22 092
person-years [0.39 hospitalizations per person-year]).
jamapsychiatry.com
Real-world Effectiveness of Pharmacologic Treatments in Bipolar Disorder Original Investigation Research

Figure 1. Within-Individual Analysis of Psychiatric Hospitalization

FDR
Medication HR (95% CI) Corrected Favors Use Favors Nonuse P Value
Benzodiazepines 1.19 (1.12-1.26) .001a <.001
Sedatives 1.06 (1.01-1.12) .07 .03
Antidepressants 1.07 (1.02-1.13) .04 .01
Antipsychotics (all) 1.01 (0.95-1.07) .79 .75
Mood stabilizers (all) 0.91 (0.86-0.97) .02 .004
Gabapentin 0.58 (0.44-0.77) .001a <.001
Lithium 0.67 (0.60-0.73) .001a <.001
Carbamazepine 0.74 (0.63-0.87) .001a <.001
Oxcarbazepine 0.76 (0.56-1.03) .15 .08
Lamotrigine 0.78 (0.71-0.87) .001a <.001
Valproic acid 0.88 (0.81-0.95) .005a <.001
Topiramate 1.32 (0.96-1.81) .16 .09
Risperidone (LAI) 0.58 (0.34-1.00) .10 .049
Perphenazine (LAI) 0.60 (0.41-0.88) .03a .009
Periciazine 0.61 (0.32-1.17) .22 .14
Olanzapine (LAI) 0.68 (0.16-2.92) .75 .60
Sulpiride 0.72 (0.56-0.93) .04a .01
Haloperidol (LAI) 0.74 (0.46-1.19) .33 .22
Dixyrazine 0.78 (0.59-1.04) .16 .09
Risperidone 0.86 (0.77-0.96) .03a .007
Levomepromazine 0.89 (0.80-0.98) .04a .02
Haloperidol 0.89 (0.74-1.08) .37 .25
Aripiprazole 0.89 (0.78-1.02) .16 .09
Clozapine 0.90 (0.54-1.50) .78 .69
Quetiapine 0.92 (0.85-0.98) .04a .02
Hazard ratios (HRs) for the risk of
Zuclopenthixol (LAI) 0.93 (0.53-1.63) .84 .81 psychiatric hospitalization for
Flupentixol 0.94 (0.76-1.17) .75 .59 different medications. Mood
Chlorpromazine 0.96 (0.77-1.18) .78 .68 stabilizers are indicated with orange
Melperone 0.96 (0.78-1.18) .78 .69 squares, antipsychotics with blue
Chlorprothixene 0.98 (0.87-1.10) .79 .75 squares, and other psychotropics
Perphenazine 1.00 (0.85-1.18) .97 .97 with gray squares. The last line
Olanzapine
describing pooled HRs for long-acting
1.03 (0.93-1.13) .75 .58
injections (LAIs) vs equivalent oral
Zuclopenthixol 1.18 (0.88-1.57) .38 .27
antipsychotic treatments was carried
Prochlorperazine 1.84 (1.00-3.37) .10 .049
out in a separate analysis (eg,
LAI vs oral antipsychotics 0.70 (0.55-0.90) .005 risperidone LAI vs risperidone oral).
a
0.1 1.0 10 Significant P value after
HR (95% CI) Benjamini-Hochberg correction for
5% false discovery rate (FDR).

Results from sensitivity analyses are shown in eTables 3 to
8 in the Supplement (showing also raw data on incidents and
person-years). The main results on the comparative effectiveness
of lithium and LAIs were in line with the previous analyses.
Discussion
To our knowledge, this is the first study on the comparative
real-world effectiveness of all widely used psychopharmaco-
logic agents and routes of administration in bipolar disorder.
The main results of our study indicate that lithium is superior
to other mood stabilizers and that LAIs are markedly better than
identical oral formulations of antipsychotics.
Our most important finding is that when a patient with
bipolar disorder uses an LAI, the patient’s risk of relapse lead-
ing to psychiatric hospitalization as well as all-cause hospital-
ization owing to mental or somatic illness is about 30% lower
than during time periods when the same patient uses an iden-
tical oral antipsychotic. It is questionable if this result can be
generalized with certainty to patients who have never used
LAIs. However, in the traditional between-individual analy-
sis including all patients, all LAIs were associated with a sub-
stantially lower risk of rehospitalizations than their identical
oral formulations. This result is similar to results from a Swed-
ish nationwide cohort of patients with schizophrenia, which
indicate that insufficient treatment adherence is a similar prob-
lem in bipolar disorder and schizophrenia.22 Most treatment
guidelines10 do not encourage the use of LAIs instead of oral
formulations, which is owing to lack of any studies on head-
to-head comparison between LAIs and oral formulations.
Our results indicate that, as a therapeutic group, mood sta-
bilizers are associated with the lowest risk of psychiatric hos-
pitalization in bipolar patients. Of the specific medications
studied in order of risk reduction, risperidone LAI, gabapen-
tin, perphenazine LAI, lithium, sulpiride, carbamazepine,
lamotrigine, risperidone, valproic acid, levomepromazine
maleate, and quetiapine were associated with a lowered risk
of psychiatric hospitalization. However, from the medica-
tions listed, the results with gabapentin and sulpiride did not

jamapsychiatry.com (Reprinted) JAMA Psychiatry April 2018 Volume 75, Number 4 351

© 2018 American Medical Association. All rights reserved.

 Research Original Investigation Real-world Effectiveness of Pharmacologic Treatments in Bipolar Disorder

Figure 2. Within-Individual Analysis of All-Cause Hospitalization

FDR
Medication HR (95% CI) Corrected Favors Use Favors Nonuse P Value
Benzodiazepines 1.15 (1.11-1.20) .001a <.001
Sedatives 1.10 (1.06-1.14) .001a <.001
Antidepressants 1.05 (1.01-1.09) .001a <.001
Antipsychotics (all) 0.99 (0.95-1.03) .06 .02
Mood stabilizers (all) 0.92 (0.88-0.96) .74 .65
Lithium 0.71 (0.66-0.76) .001a <.001
Lamotrigine 0.86 (0.80-0.93) .001a <.001
Carbamazepine 0.87 (0.77-0.98) .06 .02
Oxcarbazepine 0.90 (0.75-1.08) .34 .25
Valproic acid 0.95 (0.90-1.01) .24 .12
Gabapentin 0.97 (0.81-1.16) .76 .72
Topiramate 1.12 (0.88-1.41) .47 .36
Olanzapine (LAI) 0.64 (0.20-2.01) .55 .44
Perphenazine (LAI) 0.68 (0.45-1.02) .16 .07
Haloperidol (LAI) 0.71 (0.47-1.06) .21 .10
Sulpiride 0.73 (0.59-0.90) .01a .003
Periciazine 0.73 (0.48-1.12) .26 .15
Risperidone (LAI) 0.74 (0.47-1.18) .33 .20
Zuclopenthixol (LAI) 0.75 (0.48-1.18) .33 .21
Dixyrazine 0.76 (0.60-0.95) .06 .02
Aripiprazole 0.84 (0.75-0.93) .006a .001
Clozapine 0.87 (0.58-1.29) .58 .48
Levomepromazine 0.88 (0.81-0.95) .004a <.001
Chlorpromazine 0.90 (0.76-1.07) .34 .22
Haloperidol 0.91 (0.81-1.03) .26 .14 Hazard ratios (HRs) for the risk of
Flupentixol 0.91 (0.77-1.06) .34 .23 all-cause hospitalization for different
Risperidone 0.91 (0.84-0.99) .06 .02 medications. Mood stabilizers are
Melperone 0.93 (0.81-1.08) .47 .35 indicated with orange squares,
Quetiapine 0.93 (0.88-0.98) .04 .01 antipsychotics with blue squares, and
Chlorprothixene 0.98 (0.89-1.07) .74 .67 other psychotropics with gray
Perphenazine 0.99 (0.86-1.14) .88 .88
squares. The last line describing
pooled HRs for long-acting injections
Olanzapine 1.02 (0.95-1.09) .74 .65
(LAIs) vs similar oral antipsychotic
Zuclopenthixol 1.04 (0.79-1.36) .82 .80
treatments was carried out in a
Prochlorperazine 1.36 (0.91-2.03) .26 .14 separate analysis (eg, risperidone LAI
LAI vs oral antipsychotics 0.70 (0.57-0.86) <.001 vs risperidone oral).
a
0.1 1.0 10 Significant P value after
HR (95% CI) Benjamini-Hochberg correction for
5% false discovery rate (FDR).

survive sensitivity analyses with time reset at outcome rather
than treatment initiation (eFigure in the Supplement). In ad-
dition, the result for risperidone LAI did not remain statisti-
cally significant when corrected for multiple comparisons.
Therefore, the results for these medications should be inter-
preted with caution. The Finnish national treatment guide-
lines recommend avoiding (“Do not do”) the use of gabapen-
tin as monotherapy for either bipolar mania or bipolar
depression in (“Käypä hoito” [current care guidelines]).
Gabapentin is thus rarely used as a first-line medication for
bipolar disorder. However, some studies have suggested that
gabapentin might be effective, especially as an adjunct treat-
ment to prevent relapses in bipolar disorder.23
As previous studies have reported,13-17 we also observed
a marked association for reduced risk of psychiatric hospital-
ization with lithium. Surprisingly, lithium use was also asso-
ciated with the lowest risk of hospitalization for any cause of
all the compounds studied. The incidence of hospitalization
owing to any physical illness was substantially lower during
lithium treatment when compared with 2 of the other most
frequently used medications, valproate and quetiapine. Mood
stabilizers as a group were associated with an increased risk
of cardiovascular hospitalization, and, of the individual mood
stabilizers, only valproic acid and carbamazepine were asso-
ciated with a significant increase in risk. High serum carba-
mazepine levels have been associated with coma, seizures, re-
spiratory failure, sinus tachycardia, and cardiac conduction
defects in previous studies,24,25 and valproic acid use has been
associated with increased weight gain,26 although studies that
have not found such associations exist as well.27 The ob-
served potential risk with these medications should, how-
ever, be considered when prescribing these medications in
clinical practice.
Benzodiazepine use was associated with an increased
risk for both psychiatric hospitalization and hospitalization
for any cause, as reported previously. 28-30 Although the
within-individual model does control for confounding
related to anxiety, we cannot rule out that some of the
observed association with an increased risk for hospitaliza-
tion with benzodiazepines comes from patients with an

352 JAMA Psychiatry April 2018 Volume 75, Number 4 (Reprinted) jamapsychiatry.com

© 2018 American Medical Association. All rights reserved.

Real-world Effectiveness of Pharmacologic Treatments in Bipolar Disorder Original Investigation Research

Figure 3. Within-Individual Analysis of Cardiovascular Hospitalization

FDR
Medication HR (95% CI) Corrected Favors Use Favors Nonuse P Value
Benzodiazepines 1.10 (0.92-1.31) .47 .30
Sedatives 1.17 (0.99-1.38) .24 .07
Antidepressants 0.89 (0.75-1.06) .43 .20
Antipsychotics (all) 0.96 (0.81-1.15) .78 .67
Mood stabilizers (all) 1.32 (1.10-1.58) .03a .003
Topiramate 0.99 (0.36-2.73) .98 .98
Lithium 1.03 (0.76-1.38) .89 .86
Lamotrigine 1.35 (0.90-2.01) .38 .15
Valproic acid 1.53 (1.16-2.01) .03a .003
Gabapentin 1.70 (0.68-4.27) .47 .26
Oxcarbazepine 1.81 (0.97-3.38) .24 .06
Carbamazepine 1.95 (1.30-2.93) .03a .001
Sulpiride 0.36 (0.16-0.80) .11 .01
Chlorpromazine 0.54 (0.31-0.92) .17 .03
Zuclopenthixol 0.58 (0.16-2.07) .59 .40
Melperone 0.64 (0.39-1.06) .24 .08
Aripiprazole 0.66 (0.35-1.23) .43 .19
Haloperidol 0.69 (0.47-1.03) .24 .07
Flupentixol 0.69 (0.27-1.76) .60 .44
Dixyrazine 0.73 (0.19-2.75) .78 .64
Quetiapine 0.77 (0.61-0.98) .19 .03
Levomepromazine 0.78 (0.55-1.11) .41 .17
Chlorprothixene 0.78 (0.49-1.25) .47 .31
Risperidone (LAI) 0.78 (0.30-2.01) .78 .61
Risperidone 0.93 (0.66-1.31) .78 .69
Olanzapine 0.94 (0.69-1.27) .78 .67
Periciazine 1.14 (0.31-4.13) .89 .85 Hazard ratios (HRs) for risk of
Clozapine 1.26 (0.19-8.22) .89 .81 cardiovascular hospitalization for
Perphenazine 1.67 (0.66-4.18) .47 .28 different medications. Mood
Haloperidol (LAI) stabilizers are indicated with orange
2.24 (0.60-8.34) .49 .23
squares, antipsychotics with blue
Zuclopenthixol (LAI) 2.98 (1.04-8.59) .21 .04
squares, and other psychotropics
Prochlorperazine 3.31 (0.87-12.64) .24 .08
with gray squares. LAI indicates
Perphenazine (LAI) 3.50 (0.17-70.50) .59 .41 long-acting injection.
a
0.1 1.0 10 Significant P value after
HR (95% CI) Benjamini-Hochberg correction for
5% false discovery rate (FDR).

acutely worsened state of comorbid anxiety or a substance
abuse disorder. As data on the harmful effects of long-term
benzodiazepine use are becoming more prominent, care
should be taken when prescribing these medications to
ensure they remain in use for only as long as necessary.
Limitations
Although our study was comprehensive, any results should
be interpreted with caution. The study population included
only Finns, and as such, all patients were diagnosed with
Finnish International Classification of Diseases, Ninth
Revision and ICD-10 criteria according to Finnish practice.
Although Finnish registries are comprehensive and of high
quality,31 they still do not contain all the information that
would be needed to perform optimal observational studies.
Sources of bias are thus unavoidable. The most prominent
of these biases is confounding by indication, which rises
from the fact that treatments for individual patients are (it is
hoped) not selected at random but are rather products of
comprehensive clinical decision making, the reasons for
which are often not stored in basic registries. However, the
within-individual approach used here eliminates bias from
permanent individual characteristics since the individual is
used as his or her own control. In between-individual analy-
sis, the order of comparative effectiveness was in line with
the results from the within-individual analysis, but the HRs
were higher than in the primary within-individual analysis,
indicating residual confounding due to selection bias (ie,
the most mildly ill patients did not use any medication dur-
ing the follow-up). The results of the study also must be
framed in the context of the primary outcome (hospitaliza-
tion). Although this is a measurable, reliable, and clinically
relevant outcome, bipolar disorder morbidity is not limited
to full episodes or suicidality requiring hospitalization.
Hence, the findings on drug effectiveness do not necessarily
apply to alleviating other symptoms that do not often lead
to hospitalization, such as subthreshold depressive
symptoms,32 which have a strong effect on psychosocial
functioning. 33 Moreover, since mania is a more common
cause of hospitalization than is depression, the results may
be favoring compounds with a high polarity index34 (more
antimanic than antidepressant preventive efficacy) or
with antisuicidal properties, 35 as suicidal patients are
more often hospitalized.

jamapsychiatry.com (Reprinted) JAMA Psychiatry April 2018 Volume 75, Number 4 353

© 2018 American Medical Association. All rights reserved.

 Research Original Investigation Real-world Effectiveness of Pharmacologic Treatments in Bipolar Disorder

Conclusions
According to our data, lithium is associated with substan-
tially reduced risks of psychiatric and all-cause hospitaliza-
tion and should remain as the first line of treatment for bipo-
lar disorder, after decades of underprescription.35 The most
widely used agent, quetiapine, showed only modest effective-
ness (8% risk reduction), which does not support its use for this
indication. Long-acting injectable formulations of antipsy-
chotic medications were associated with approximately a 30%
lower risk than identical oral formulations of the same medi-
cation. Although more research is needed to support the
notion, LAIs might offer a safe and effective option for
relapse prevention in bipolar disorder for patients for whom
lithium is not suitable.

ARTICLE INFORMATION
Accepted for Publication: December 17, 2018.
Published Online: February 28, 2018.
doi:10.1001/jamapsychiatry.2017.4711
Author Affiliations: Department of Forensic
Psychiatry, University of Eastern Finland,
Niuvanniemi Hospital, Kuopio, Finland
(Lähteenvuo, Tanskanen, Tiihonen); Department of
Clinical Neuroscience, Karolinska Institutet,
Stockholm, Sweden (Tanskanen, Taipale, Tiihonen);
Impact Assessment Unit, National Institute for
Health and Welfare, Helsinki, Finland (Tanskanen);
Kuopio Research Centre of Geriatric Care,
University of Eastern Finland, Kuopio, Finland
(Taipale); School of Pharmacy, University of Eastern
Finland, Kuopio, Finland (Taipale); EPID Research
Oy, Espoo, Finland (Hoti, Vattulainen); Hospital
Clinic, Institute of Neuroscience, University of
Barcelona, Institut d’Investigacions Biomèdiques
August Pi i Sunyer, Centro de Investigación
Biomédica en Red de Salud Mental, Barcelona,
Catalonia, Spain (Vieta).
Author Contributions: Dr Hoti and Ms Vattulainen
had full access to all the data in the study and take
responsibility for the integrity of the data and the
accuracy of the data analysis.
Study concept and design: Lähteenvuo, Tanskanen,
Taipale, Hoti, Vieta, Tiihonen.
Acquisition, analysis, or interpretation of data: All
authors.
Drafting of the manuscript: Lähteenvuo, Tanskanen,
Vieta, Tiihonen.
Critical revision of the manuscript for important
intellectual content: Lähteenvuo, Taipale, Hoti,
Vattulainen, Vieta.
Statistical analysis: Taipale, Hoti, Vattulainen, Vieta.
Obtained funding: Tiihonen.
Administrative, technical, or material support:
Tanskanen, Vieta, Tiihonen.
Study supervision: Hoti, Vieta, Tiihonen.
Conflict of Interest Disclosures: Dr Lähteenvuo
reported being a major shareholder and board
member at Genomi Solutions Ltd, a Finnish based
bioinformatics company; receiving research grants
or awards from Boehringer-Ingelheim; receiving
travel grants from Sunovion Ltd; and working as a
coordinator for a research project funded by the
Stanley Foundation. Dr Tanskanen reported
participating in research projects funded by
Janssen Cilag and Eli Lilly with grants paid to the
Karolinska Institutet and serving as a member of
the advisory board for Janssen-Cilag. Dr Taipale
reported participating in research projects funded
by Janssen-Cilag and Eli Lilly, with grants paid to the
Karolinska Institutet. Dr Hoti and Ms Vattulainen
are employed by EPID Research, which is a contract
research organization that performs commissioned
pharmacoepidemiologic studies; thus, its
employees have been and currently are working in
collaboration with several pharmaceutical
companies. Dr Vieta reported receiving grants and
serving as consultant, advisor, or continuing
medical education speaker for AB-Biotics, Aequus,
Adamed, Alexza, Allergan, Almirall, AstraZeneca,
Bial, Bristol-Myers Squibb, Dainippon Sumitomo
Pharma, Elan, Eli Lilly, Esteve, Ferrer, Forest
Research Institute, Gedeon Richter,
GlaxoSmithKline, Janssen-Cilag, Jazz, Johnson &
Johnson, Lundbeck, Merck, Novartis, Organon,
Otsuka, Pfizer, Pierre-Fabre, Rovi, Qualigen, Roche,
Sanofi, Servier, Schering-Plough, Shire, Solvay,
Sunovion, Takeda, Telefónica, Teva, the Spanish
Ministry of Science and Innovation, the Seventh
European Framework Programme, the Stanley
Medical Research Institute, United Biosource
Corporation, and Wyeth. Dr Tiihonen reported
serving as a consultant to AstraZeneca,
Bristol-Myers Squibb, Eli Lilly, F. Hoffman–La Roche,
Janssen-Cilag, Lundbeck, Organon, and Finnish
Medicines Agency; receiving fees for giving expert
testimony to AstraZeneca, Bristol-Myers Squibb,
Eli Lilly, GlaxoSmithKline, Janssen-Cilag, Lundbeck,
Otsuka, and Pfizer; receiving lecture fees from
AstraZeneca, Bristol-Myers Squibb, Eli Lilly,
GlaxoSmithKline, Janssen-Cilag, Lundbeck,
Novartis, Otsuka, and Pfizer; receiving grants from
Stanley Foundation and Sigrid Jusélius Foundation;
serving as a member of advisory boards for
AstraZeneca, Eli Lilly, Janssen-Cilag, and Otsuka;
and having research collaboration with Lilly and
Janssen-Cilag.
Funding/Support: This study was funded by the
Finnish Ministry of Social Affairs and Health through
the developmental fund for Niuvanniemi Hospital.
Role of the Funder/Sponsor: The funding source
had no role in the design and conduct of the study;
collection, management, analysis, and
interpretation of the data; preparation, review, or
approval of the manuscript; and decision to submit
the manuscript for publication.
Additional Contributions: Aija Räsänen and Tarja
Koskela, Niuvanniemi Hospital, provided secretarial
assistance. They were not compensated for their
contribution other than their monthly salary.
REFERENCES
1. Saunders KE, Goodwin GM. The course of bipolar
disorder. Adv Psychiatr Treat. 2010;16(5):318-328.
doi:10.1192/apt.bp.107.004903
2. Global Burden of Disease Study 2013
Collaborators. Global, regional, and national
incidence, prevalence, and years lived with
disability for 301 acute and chronic diseases and
injuries in 188 countries, 1990-2013: a systematic
analysis for the Global Burden of Disease Study
2013. Lancet. 2015;386(9995):743-800.
3. Perlis RH, Ostacher MJ, Patel JK, et al. Predictors
of recurrence in bipolar disorder: primary outcomes
from the Systematic Treatment Enhancement
Program for Bipolar Disorder (STEP-BD). Am J
Psychiatry. 2006;163(2):217-224.
4. Calabrese JR, Sanchez R, Jin N, et al. Efficacy
and safety of aripiprazole once-monthly in the
maintenance treatment of bipolar I disorder:
a double-blind, placebo-controlled, 52-week
randomized withdrawal study. J Clin Psychiatry.
2017;78(3):324-331.
5. Vieta E, Montgomery S, Sulaiman AH, et al.
A randomized, double-blind, placebo-controlled
trial to assess prevention of mood episodes with
risperidone long-acting injectable in patients with
bipolar I disorder. Eur Neuropsychopharmacol.
2012;22(11):825-835.
6. Miura T, Noma H, Furukawa TA, et al.
Comparative efficacy and tolerability of
pharmacological treatments in the maintenance
treatment of bipolar disorder: a systematic review
and network meta-analysis. Lancet Psychiatry.
2014;1(5):351-359.
7. Severus E, Taylor MJ, Sauer C, et al. Lithium for
prevention of mood episodes in bipolar disorders:
systematic review and meta-analysis. Int J Bipolar
Disord. 2014;2:15.
8. Gigante AD, Lafer B, Yatham LN. Long-acting
injectable antipsychotics for the maintenance
treatment of bipolar disorder. CNS Drugs. 2012;26
(5):403-420.
9. National Institute for Health and Care
Excellence. Bipolar Disorder: The Management of
Bipolar Disorder in Adults, Children and Adolescents,
in Primary and Secondary Care. GC38. London,
England: National Institute for Health and Care
Excellence; 2006.
10. Fountoulakis KN, Grunze H, Vieta E, et al. The
International College of Neuro-Psychopharmacology
(CINP) treatment guidelines for bipolar disorder in
adults (CINP-BD-2017), part 3: the clinical guidelines.
Int J Neuropsychopharmacol. 2017;20(2):180-195.
11. Benraad CE, Kamerman-Celie F, van Munster
BC, Oude Voshaar RC, Spijker J, Olde Rikkert MG.
Geriatric characteristics in randomised controlled
trials on antidepressant drugs for older adults:
a systematic review. Int J Geriatr Psychiatry. 2016;31
(9):990-1003.
12. Vieta E. Observational, pragmatic, and clinical
trials in bipolar disorder. J Clin Psychiatry. 2008;69
(9):e27.
13. Hayes JF, Marston L, Walters K, Geddes JR, King
M, Osborn DPJ. Lithium vs valproate vs olanzapine
vs quetiapine as maintenance monotherapy for
bipolar disorder: a population-based UK cohort
study using electronic health records. World
Psychiatry. 2016;15(1):53-58.
14. Kessing LV, Hellmund G, Geddes JR, Goodwin
GM, Andersen PK. Valproate v lithium in the
treatment of bipolar disorder in clinical practice:

354 JAMA Psychiatry April 2018 Volume 75, Number 4 (Reprinted) jamapsychiatry.com

© 2018 American Medical Association. All rights reserved.

Real-world Effectiveness of Pharmacologic Treatments in Bipolar Disorder
observational nationwide register-based cohort
study. Br J Psychiatry. 2011;199(1):57-63.
15. Simhandl C, König B, Amann BL. A prospective
4-year naturalistic follow-up of treatment and
outcome of 300 bipolar I and II patients. J Clin
Psychiatry. 2014;75(3):254-262.
16. Joas E, Karanti A, Song J, Goodwin GM,
Lichtenstein P, Landén M. Pharmacological
treatment and risk of psychiatric hospital admission
in bipolar disorder. Br J Psychiatry. 2017;210(3):
197-202.
17. Song J, Sjölander A, Joas E, et al. Suicidal
behavior during lithium and valproate treatment:
a within-individual 8-year prospective study of
50,000 patients with bipolar disorder. Am J
Psychiatry. 2017;174(8):795-802.
18. Tanskanen A, Taipale H, Koponen M, et al. Drug
exposure in register-based research—an
expert-opinion based evaluation of methods. PLoS
One. 2017;12(9):e0184070.
19. Tanskanen A, Taipale H, Koponen M, et al. From
prescription drug purchases to drug use periods—
a second generation method (PRE2DUP). BMC Med
Inform Decis Mak. 2015;15:21.
20. Taipale H, Tanskanen A, Koponen M,
Tolppanen AM, Tiihonen J, Hartikainen S.
Agreement between PRE2DUP register data
modeling method and comprehensive drug use
interview among older persons. Clin Epidemiol.
2016;8:363-371.
21. Tiihonen J, Tanskanen A, Hoti F, et al.
Pharmacological treatments and risk of readmission
to hospital for unipolar depression in Finland:
jamapsychiatry.com
© 2018 American Medical Association. All rights reserved.
a nationwide cohort study. Lancet Psychiatry. 2017;
4(7):547-553.
22. Tiihonen J, Mittendorfer-Rutz E, Majak M, et al.
Real-world effectiveness of antipsychotic
treatments in a nationwide cohort of 29 823
patients with schizophrenia. JAMA Psychiatry. 2017;
74(7):686-693.
23. Vieta E, Manuel Goikolea J, Martínez-Arán A,
et al. A double-blind, randomized,
placebo-controlled, prophylaxis study of adjunctive
gabapentin for bipolar disorder. J Clin Psychiatry.
2006;67(3):473-477.
24. Hojer J, Malmlund HO, Berg A. Clinical features
in 28 consecutive cases of laboratory confirmed
massive poisoning with carbamazepine alone.
J Toxicol Clin Toxicol. 1993;31(3):449-458.
25. Spiller HA, Krenzelok EP, Cookson E.
Carbamazepine overdose: a prospective study of
serum levels and toxicity. J Toxicol Clin Toxicol.
1990;28(4):445-458.
26. Hayes JF, Marston L, Walters K, Geddes JR,
King M, Osborn DP. Adverse renal, endocrine,
hepatic, and metabolic events during maintenance
mood stabilizer treatment for bipolar disorder:
a population-based cohort study. PLoS Med. 2016;
13(8):e1002058.
27. Harit D, Aggarwal A, Kalra S, Chhillar N. Effect
of carbamazepine and valproate monotherapy on
cardiovascular risks in epileptic children. Pediatr
Neurol. 2015;53(1):88-90.
28. Taipale H, Tolppanen AM, Koponen M, et al.
Risk of pneumonia associated with incident
(Reprinted) JAMA Psychiatry April 2018 Volume 75, Number 4
Original Investigation Research
benzodiazepine use among community-dwelling
adults with Alzheimer disease. CMAJ. 2017;189(14):
E519-E529.
29. Dodds TJ. Prescribed benzodiazepines and
suicide risk: a review of the literature. Prim Care
Companion CNS Disord. 2017;19(2):16r02037.
30. Markota M, Rummans TA, Bostwick JM, Lapid
MI. Benzodiazepine use in older adults: dangers,
management, and alternative therapies. Mayo Clin
Proc. 2016;91(11):1632-1639.
31. Sund R. Quality of the Finnish hospital
discharge register: a systematic review. Scand J
Public Health. 2012;40(6):505-515.
32. Sanchez-Moreno J, Bonnín C, González-Pinto
A, et al; CIBERSAM Functional Remediation Group.
Do patients with bipolar disorder and
subsyndromal symptoms benefit from functional
remediation? a 12-month follow-up study. Eur
Neuropsychopharmacol. 2017;27(4):350-359.
33. Bonnín C del M, González-Pinto A, Solé B, et al;
CIBERSAM Functional Remediation Group. Verbal
memory as a mediator in the relationship between
subthreshold depressive symptoms and functional
outcome in bipolar disorder. J Affect Disord. 2014;
160:50-54.
34. Popovic D, Reinares M, Goikolea JM, Bonnin
CM, Gonzalez-Pinto A, Vieta E. Polarity index of
pharmacological agents used for maintenance
treatment of bipolar disorder. Eur
Neuropsychopharmacol. 2012;22(5):339-346.
35. Grande I, Berk M, Birmaher B, Vieta E. Bipolar
disorder. Lancet. 2016;387(10027):1561-1572.
355
 Supplementary Online Content

Lähteenvuo M, Tanskanen A, Taipale H, et al. Real-world effectiveness of

pharmacologic treatments for the prevention of rehospitalization in a Finnish
nationwide cohort of patients with bipolar disorder. JAMA Psychiatry. Published
online February 28, 2018. doi:10.1001/jamapsychiatry.2017.4711

eAppendix. Methods
eTable 1. Definition of Adjusting Variables in the Within-Individual and Traditional Cox
Proportional Hazards by Treatment Effect Studied
eTable 2. Definition of Additional Adjusting Variables in the Traditional Cox
Proportional Hazards Models by Study Outcome
eTable 3. The Association Between Use vs No Use of Medications and Risk of
Rehospitalization Due to Psychiatric Disorder in Within-Individual Analysis in the Total
Cohort (N = 18,018)
eTable 4. The Association Between Use vs No Use of Medications and Risk of
Rehospitalization Due to Any Cause in Within-Individual Analysis in the Total Cohort
(N = 18,018)
eTable 5. The Association Between Use vs No Use of Medications and Risk of
Rehospitalization Due to Psychiatric Disorder in Within-Individual Analysis in the
Incident Cohort (N = 2,074)
eTable 6. The Association Between Use vs No Use of Medications and Risk of
Rehospitalization Due to Psychiatric Disorder in Traditional Analysis in the Total Cohort
(N = 18,018)
eTable 7. The Association Between Use vs No Use of Medications and Risk of
Rehospitalization Due to Any Cause in Traditional Analysis in the Total Cohort (N =
18,018)
eTable 8. The Association Between Use vs No Use of Medications and Risk of
Rehospitalization Due to Cardiovascular Reasons in Traditional Analysis in the Total
Cohort (N = 18,018)
eFigure. Exposure to Psychotropics During Follow-up and Time Resetting in Within-
Individual Cox Proportional Hazards Model

This supplementary material has been provided by the authors to give readers additional
information about their work.

American Medical Association All rights reserved

eAppendix. Methods

Study design and data acquisition

In Finland, every individual has a unique identification code, which makes it possible to track them
even if they would change their name or location of residence. All hospital treatments, deaths, and
prescription fillings are documented in the nationwide databases. The current study included every
subject (N = 18,018) hospitalized at least once with a bipolar disorder diagnosis (ICD 10
diagnoses F30-31, Finnish ICD 9 diagnoses 2962-2964 and 2967A) between January 1:st, 1987
and December 31:st, 2012, who had not been diagnosed with broadly defined schizophrenia
during this time period (ICD 10: F20–29, ICD 9: 295, 2971A, 2973A, 2988A, 2989X, 3012C) and
who were still alive at the start of the observation period. A total of 25,860 subjects were initially
identified, but 7,758 were excluded due to having schizophrenia as described above, 77 excluded
due to death before start of observation period, and 7 excluded due to not having any follow-up
time (cohort entry on the same day as the end of the observation period). In order to account for
survival bias (selective mortality due to treatment history), we also formed and analyzed a sub-
cohort of the study subjects who had not ever been hospitalized with any mental disorder or used
psychotropic medication one year prior to the study observation period, namely incident cohort (N
= 2,074). Cohort entry date was set as January 1:st, 1996 for subjects hospitalized due to bipolar
disorder between January 1st 1987 to December 31st 1995, and as the first hospital discharge
date for subjects hospitalized for the first time on January 1st 1996 or later (the incident cohort).

American Medical Association All rights reserved

eTable 1. Definition of Adjusting Variables in the Within-Individual and Traditional Cox
Proportional Hazards by Treatment Effect Studied

Treatment studied
Drug classes:
Benzodiazepines,
Sedatives, Individual Individual
Antidepressants,
Antipsychotics, Mood stabilizers: Antipsychotics:
Mood stabilizers

Time since Time since Time since Time since

diagnosis diagnosis (0-5 ,5- diagnosis (0-5 ,5- diagnosis (0-5 ,5-
10,>10; years) 10,>10; years) 10,>10; years)

Order of Order of drug Order of mood Order of

treatment classes (0-1,2,>2; stabilizers (0- antipsychotic (0-
cumulative number 1,2,>2; cumulative 1,2,>2; cumulative
of different drug number of number of
classes) different mood different
stabilizer drugs) antipsychotic
drugs)
Adjusting
Current use of Current use of Current use of Current use of
covariates
other other drug classes other mood other
treatments (yes/no; for each stabilizers antipsychotic
drug class (yes/no; for each (yes/no; for each
separately) individual mood individual
stabilizer antipsychotic
separately) separately)

Polypharmacy Polypharmacy Polypharmacy Polypharmacy

(yes/no; concurrent (yes/no; (yes/no;
use of more than concurrent use of concurrent use of
one drug class) more than one more than one
mood stabilizer) antipsychotic)

American Medical Association All rights reserved

eTable 2. Definition of Additional Adjusting Variables in the Traditional Cox Proportional Hazards
Models by Study Outcome

Hospitalization due to Hospitalizations due to Hospitalizations due to

Study outcome psychiatic disorders all causes cardiovascular causes

Number of re- Number of re- Number of re- Number of re-

hospitalizations hospitalizations due to hospitalizations due to hospitalizations due to
within 2 years psychiatric disorders all causes within 2 cardiovascular causes
prior to index date within 2 years prior years prior to index within 2 years prior
toindex date (0,1-2,>2) date (0,1-2,>2) toindex date (0,1-2,>2)

Age at index date Age at index date Age at index date Age at index date
Adjusting (<30,30-49,50-69,>70; (<30,30-49,50-69,>70; (<30,30-49,50-69,>70;
variables years) years) years)

Gender Gender (Female/male) Gender (Female/male) Gender (Female/male)

Calendar year of Calendar year of index Calendar year of index Calendar year of index
index date date (1996-1999,2000- date (1996-1999,2000- date (1996-1999,2000-
2003,2004-2007,2008- 2003,2004-2007,2008- 2003,2004-2007,2008-
2012) 2012) 2012)

American Medical Association All rights reserved

eTable 3. The Association Between Use vs No Use of Medications and Risk of Rehospitalization
Due to Psychiatric Disorder in Within-Individual Analysis in the Total Cohort (N = 18,018)

      Fully 
Treatment  Events  Person years  Drug adjusted model  adjusted 
         HR                     CI  HR  CI  P Value 
mood stabilizers  21054  63139  0.89  (0.85–0.93)  0.90  (0.86–0.95)  <.001 
antipsychotics  22008  54547  0.93  (0.90–0.97)  0.97  (0.93–1.02)  .28 
antidepressants  18930  47147  1.08  (1.04–1.13)  1.08  (1.03–1.13)  <.001 
sedatives  11902  27448  1.08  (1.03–1.13)  1.10  (1.05–1.16)  <.001 
benzodiazepines  12056  24598  1.16  (1.11–1.21)  1.18  (1.13–1.24)  <.001 
lithium  6946  24815  0.80  (0.75–0.85)  0.78  (0.73–0.84)  <.001 
carbamazepine  1657  5409  0.94  (0.84–1.04)  0.87  (0.77–0.98)  .02 
lamotrigine  4970  12641  0.87  (0.82–0.93)  0.96  (0.89–1.04)  .34 
gabapentin  176  541  0.90  (0.70–1.16)  0.96  (0.75–1.24)  .76 
valproic acid  9998  26091  0.98  (0.93–1.03)  0.99  (0.94–1.05)  .80 
oxcarbazepine  331  881  1.05  (0.86–1.29)  1.06  (0.84–1.33)  .62 
topiramate  406  506  1.41  (1.12–1.77)  1.56  (1.21–2.00)  <.001 
olanzapine (LAI)  6  11  0.56  (0.18–1.75)  0.61  (0.19–1.91)  .40 
clozapine  320  446  0.64  (0.47–0.88)  0.73  (0.54–0.99)  .04 
periciazine  140  351  0.82  (0.58–1.18)  0.79  (0.54–1.16)  .22 
haloperidol (LAI)  68  210  0.84  (0.49–1.42)  0.83  (0.50–1.39)  .48 
perphenazine (LAI)  40  89  0.87  (0.64–1.19)  0.86  (0.66–1.12)  .28 
dixyrazine  194  222  0.91  (0.69–1.20)  0.89  (0.67–1.18)  .42 
sulpiride  204  604  0.91  (0.72–1.14)  0.91  (0.72–1.15)  .43 
risperidone (LAI)  49  98  0.83  (0.58–1.19)  0.91  (0.63–1.32)  .62 
levomepromazine  3583  7295  0.95  (0.89–1.02)  0.96  (0.89–1.04)  .32 
chlorpromazine  646  1613  1.02  (0.87–1.19)  0.98  (0.83–1.15)  .81 
quetiapine  9326  22092  0.90  (0.86–0.94)  0.98  (0.93–1.04)  .54 
oral risperidone  2144  5441  0.97  (0.90–1.05)  0.99  (0.91–1.07)  .74 
zuclopenthixol (LAI)  171  358  1.02  (0.68–1.54)  1.04  (0.69–1.56)  .86 
oral haloperidol  971  1917  1.08  (0.91–1.28)  1.05  (0.87–1.27)  .62 
aripiprazole  1150  1979  0.98  (0.88–1.10)  1.06  (0.94–1.19)  .37 
chlorprothixene  1980  4787  1.07  (0.98–1.18)  1.08  (0.99–1.19)  .10 
oral olanzapine  4505  9783  1.02  (0.95–1.10)  1.08  (1.00–1.17)  .05 
melperone  435  821  1.10  (0.95–1.27)  1.09  (0.94–1.27)  .25 
flupentixol  245  495  1.12  (0.92–1.37)  1.12  (0.91–1.37)  .28 
oral perphenazine  906  2349  1.18  (1.02–1.36)  1.15  (1.00–1.33)  .05 
other AP  361  513  1.12  (0.93–1.35)  1.20  (0.99–1.45)  .06 
oral zuclopenthixol  213  503  1.33  (1.09–1.62)  1.31  (1.07–1.60)  <.01 
prochlorperazine  25  36  2.27  (0.98–5.26)  2.42  (1.05–5.62)  .04 
LAI vs.   289 vs. 8445  713 vs. 19596 
oral AP  0.81  (0.67–0.99)  0.81  (0.66–0.99)  .04 
Results are based on the within-individual Cox Proportional Hazards model, and are adjusted for
time since diagnoses, temporal order of treatment, current use of other treatments, and
polypharmacy. Drug adjusted model is adjusted for drug related variables, fully adjusted for all
variables. Time reset was performed at outcome (psychiatric hospitalization).

American Medical Association All rights reserved

eTable 4. The Association Between Use vs No Use of Medications and Risk of Rehospitalization
Due to Any Cause in Within-Individual Analysis in the Total Cohort (N = 18,018)
      Fully 
Treatment  Events  Person  Drug adjusted model  adjusted 
    years       HR                   CI  HR  CI  P Value 
mood stabilizers  42907  63139  0.92  (0.89–0.95)  0.92  (0.89–0.96)  <.001 
antipsychotics  43491  54547  0.97  (0.94–1.00)  0.98  (0.95–1.01)  .20 
antidepressants  39813  47147  1.06  (1.03–1.09)  1.07  (1.03–1.11)  <.001 
sedatives  27307  27448  1.10  (1.07–1.14)  1.11  (1.07–1.14)  <.001 
benzodiazepines  25522  24598  1.14  (1.11–1.18)  1.15  (1.11–1.19)  <.001 
lithium  13581  24815  0.80  (0.77–0.84)  0.78  (0.74–0.82)  <.001 
carbamazepine  4087  5409  0.93  (0.86–1.01)  0.90  (0.83–0.98)  .02 
lamotrigine  9263  12641  0.95  (0.91–1.00)  0.98  (0.93–1.04)  .48 
valproic acid  19847  26091  1.03  (0.99–1.06)  1.01  (0.97–1.05)  .57 
oxcarbazepine  836  881  1.04  (0.91–1.20)  1.03  (0.89–1.20)  .70 
topiramate  592  506  1.20  (1.02–1.41)  1.22  (1.03–1.45)  .02 
gabapentin  863  541  1.24  (1.06–1.44)  1.26  (1.08–1.46)  <.01 
olanzapine (LAI)  7  11  0.64  (0.23–1.77)  0.64  (0.24–1.74)  .38 
zuclopenthixol (LAI)  306  358  0.73  (0.53–1.00)  0.74  (0.54–1.01)  .06 
clozapine  464  446  0.79  (0.60–1.02)  0.81  (0.63–1.04)  .10 
periciazine  324  351  0.83  (0.62–1.12)  0.82  (0.61–1.11)  .20 
haloperidol (LAI)  113  210  0.84  (0.55–1.30)  0.85  (0.55–1.32)  .48 
sulpiride  438  604  0.88  (0.75–1.04)  0.89  (0.75–1.05)  .16 
dixyrazine  285  222  0.93  (0.75–1.15)  0.92  (0.74–1.14)  .46 
perphenazine (LAI)  61  89  0.96  (0.72–1.27)  0.93  (0.71–1.23)  .63 
chlorpromazine  1208  1613  0.95  (0.83–1.07)  0.95  (0.83–1.08)  .39 
levomepromazine  6713  7295  0.94  (0.89–0.99)  0.95  (0.89–1.00)  .06 
aripiprazole  1662  1979  0.94  (0.86–1.03)  0.96  (0.87–1.05)  .35 
oral haloperidol  2118  1917  1.00  (0.90–1.11)  1.00  (0.90–1.11)  .96 
oral risperidone  4448  5441  1.00  (0.95–1.06)  1.00  (0.94–1.06)  .97 
risperidone (LAI)  91  98  1.00  (0.77–1.30)  1.00  (0.77–1.31)  .98 
quetiapine  17968  22092  0.99  (0.96–1.03)  1.01  (0.97–1.05)  .74 
oral perphenazine  1626  2349  1.05  (0.94–1.17)  1.04  (0.94–1.16)  .46 
melperone  1032  821  1.05  (0.93–1.17)  1.05  (0.93–1.18)  .45 
flupentixol  469  495  1.05  (0.93–1.19)  1.06  (0.93–1.21)  .35 
chlorprothixene  3798  4787  1.05  (0.98–1.13)  1.06  (0.98–1.14)  .13 
oral zuclopenthixol  391  503  1.06  (0.87–1.31)  1.07  (0.86–1.31)  .55 
oral olanzapine  8178  9783  1.08  (1.03–1.14)  1.08  (1.02–1.14)  <.01 
other AP  490  513  1.10  (0.94–1.28)  1.13  (0.97–1.33)  .12 
prochlorperazine  70  36  1.25  (0.87–1.80)  1.27  (0.88–1.83)  .21 
LAI vs.  511 vs.  713 vs. 
oral AP  16305  19596  0.78  (0.66–0.92)  0.77  (0.65–0.91)  <.01 
Results are based on the within-individual Cox Proportional Hazards model, and are adjusted for
time since diagnoses, temporal order of treatment, current use of other treatments, and
polypharmacy. Drug adjusted model is adjusted for drug related variables, fully adjusted for all
variables. Time reset was performed at outcome (any cause hospitalization).

American Medical Association All rights reserved

eTable 5. The Association Between Use vs No Use of Medications and Risk of Rehospitalization
Due to Psychiatric Disorder in Within-Individual Analysis in the Incident Cohort (N = 2,074)

     
Treatment  Events  Person  Drug adjusted model  Fully adjusted 
  years          HR                       CI  HR  CI  P Value 
mood stabilizers  976  4960  0.70  (0.55–0.88)  0.90  (0.69–1.17)  .41 
benzodiazepines  263  987  0.93  (0.73–1.18)  1.08  (0.85–1.38)  .51 
antidepressants  506  2107  0.91  (0.70–1.18)  1.12  (0.86–1.46)  .40 
antipsychotics  991  4141  0.88  (0.72–1.09)  1.22  (0.95–1.55)  .11 
sedatives  374  1451  1.03  (0.80–1.32)  1.28  (0.98–1.66)  .07 
Results are based on the within-individual Cox Proportional Hazards model, and are adjusted for
time since diagnoses, temporal order of treatment, current use of other treatments, and
polypharmacy. Drug adjusted model is adjusted for drug related variables, fully adjusted for all
variables. Time reset was performed at change in medication status. Due to low number of
subjects, the results are shown only for the main medication categories.

American Medical Association All rights reserved

eTable 6. The Association Between Use vs No Use of Medications and Risk of Rehospitalization
Due to Psychiatric Disorder in Traditional Analysis in the Total Cohort (N = 18,018)

 
Treatment  Drug adjusted model  Fully adjusted 
           HR                         CI        HR                       CI                    P Value 
mood stabilizer  1.14  (1.08–1.20)       0.95  (0.91–0.99)  .03 
antidepressants  1.38  (1.31–1.45)  1.12  (1.08–1.17)  <.001 
sedatives  1.33  (1.26–1.40)  1.13  (1.08–1.18)  <.001 
benzodiazepines  1.55  (1.46–1.64)  1.22  (1.17–1.27)  <.001 
antipsychotics  1.64  (1.56–1.73)  1.22  (1.17–1.27)  <.001 
gabapentin  1.17  (0.92–1.48)  0.84  (0.70–1.01)  .06 
lithium  1.14  (1.06–1.22)  0.98  (0.93–1.03)  .38 
lamotrigine  1.42  (1.33–1.53)  1.01  (0.95–1.07)  .87 
carbamazepine  1.27  (1.13–1.42)  1.03  (0.95–1.11)  .50 
oxcarbazepine  1.45  (1.14–1.85)  1.03  (0.87–1.21)  .75 
valproic acid  1.50  (1.42–1.59)  1.04  (1.00–1.09)  .06 
topiramate  2.63  (1.90–3.62)  1.52  (1.19–1.95)  <.001 
olanzapine (LAI)  2.86  (0.69–11.87)  0.58  (0.14–2.43)  .45 
haloperidol (LAI)  1.25  (0.73–2.15)  0.81  (0.56–1.19)  .28 
risperidone (LAI)  1.98  (1.37–2.86)  0.96  (0.74–1.26)  .78 
sulpiride  1.13  (0.86–1.50)  1.01  (0.82–1.23)  .95 
oral risperidone  1.59  (1.45–1.75)  1.09  (1.01–1.19)  .03 
perphenazine (LAI)  1.67  (0.66–4.23)  1.09  (0.63–1.88)  .76 
zuclopenthixol (LAI)  1.95  (1.36–2.78)  1.15  (0.90–1.48)  .27 
levomepromazine  1.76  (1.60–1.93)  1.17  (1.08–1.26)  <.001 
chlorprothixene  1.62  (1.42–1.85)  1.17  (1.08–1.27)  <.001 
quetiapine  1.73  (1.64–1.83)  1.19  (1.13–1.24)  <.001 
chlorpromazine  1.52  (1.07–2.14)  1.21  (0.94–1.56)  .13 
periciazine  1.58  (1.08–2.32)  1.22  (0.92–1.60)  .16 
oral olanzapine  1.90  (1.75–2.05)  1.23  (1.16–1.31)  <.001 
oral perphenazine  1.43  (1.18–1.73)  1.25  (1.12–1.40)  <.001 
oral zuclopenthixol  1.65  (1.20–2.27)  1.25  (1.00–1.56)  .05 
melperone  1.90  (1.53–2.36)  1.29  (1.10–1.50)  <.01 
aripiprazole  1.80  (1.55–2.09)  1.30  (1.15–1.48)  <.001 
dixyrazine  2.78  (1.94–3.97)  1.33  (1.09–1.62)  <.01 
oral haloperidol  1.76  (1.45–2.13)  1.33  (1.14–1.56)  <.001 
flupentixol  1.69  (1.27–2.24)  1.36  (1.12–1.64)  <.01 
other AP  2.07  (1.75–2.46)  1.40  (1.23–1.61)  <.001 
prochlorperazine  2.50  (1.49–4.19)  1.46  (1.03–2.06)  .03 
clozapine  3.53  (2.09–5.96)  1.56  (1.06–2.28)  .02 
LAI vs. oral  1.07  (0.82–1.40)  0.77  (0.65–0.92)  <.01 
Results are based on the traditional Cox Proportional Hazards model, and are adjusted for age at
index date, gender, number of psychiatric hospitalizations within 2 years prior to index date,
calendar year of index date, time since diagnoses, temporal order of treatment, current use of
other treatments, and polypharmacy. Drug adjusted model is adjusted for drug related variables,
fully adjusted for all variables.

American Medical Association All rights reserved

eTable 7. The Association Between Use vs No Use of Medications and Risk of Rehospitalization
Due to Any Cause in Traditional Analysis in the Total Cohort (N = 18,018)

 
Treatment  Drug adjusted model  Fully adjusted 
         HR                        CI         HR                         CI                       P Value 
mood stabilizers  0.94  (0.91–0.97)  0.93  (0.90–0.96)  <.001 
antidepressants  1.27  (1.23–1.31)  1.09  (1.06–1.12)  <.001 
antipsychotics  1.25  (1.21–1.30)  1.09  (1.05–1.12)  <.001 
sedatives  1.47  (1.42–1.53)  1.17  (1.14–1.20)  <.001 
benzodiazepines  1.47  (1.41–1.53)  1.22  (1.18–1.25)  <.001 
lithium  0.87  (0.83–0.91)  0.89  (0.86–0.92)  <.001 
lamotrigine  1.12  (1.07–1.18)  0.96  (0.92–1.00)  .06 
valproic acid  1.20  (1.15–1.25)  0.99  (0.96–1.02)  .51 
carbamazepine  1.25  (1.14–1.37)  1.03  (0.97–1.10)  .27 
oxcarbazepine  1.53  (1.30–1.80)  1.16  (1.04–1.29)  <.01 
topiramate  1.72  (1.36–2.17)  1.30  (1.08–1.55)  <.01 
gabapentin  2.48  (1.95–3.15)  1.37  (1.12–1.68)  <.01 
olanzapine (LAI)  1.14  (0.34–3.80)  0.45  (0.15–1.37)  .16 
haloperidol (LAI)  0.85  (0.53–1.35)  0.79  (0.62–1.00)  .05 
risperidone (LAI)  1.51  (1.11–2.06)  0.91  (0.71–1.15)  .42 
perphenazine (LAI)  1.06  (0.55–2.03)  0.94  (0.60–1.48)  .80 
sulpiride  1.08  (0.89–1.32)  1.00  (0.87–1.14)  .96 
oral risperidone  1.35  (1.26–1.44)  1.00  (0.95–1.05)  .97 
aripiprazole  1.21  (1.07–1.36)  1.05  (0.95–1.15)  .36 
zuclopenthixol (LAI)  1.39  (1.01–1.90)  1.05  (0.84–1.32)  .67 
oral perphenazine  1.09  (0.93–1.26)  1.07  (0.97–1.18)  .17 
oral olanzapine  1.39  (1.31–1.47)  1.07  (1.02–1.11)  <.01 
quetiapine  1.36  (1.31–1.42)  1.07  (1.04–1.11)  <.001 
chlorprothixene  1.28  (1.17–1.40)  1.09  (1.03–1.15)  <.01 
periciazine  1.52  (1.19–1.93)  1.10  (0.95–1.28)  .22 
oral zuclopenthixol  1.26  (0.97–1.64)  1.10  (0.92–1.32)  .31 
chlorpromazine  1.19  (0.96–1.47)  1.11  (0.95–1.29)  .20 
levomepromazine  1.43  (1.33–1.53)  1.12  (1.06–1.19)  <.001 
dixyrazine  1.85  (1.35–2.53)  1.12  (0.91–1.38)  .28 
other AP  1.35  (1.16–1.56)  1.13  (1.01–1.26)  .04 
melperone  1.93  (1.64–2.28)  1.20  (1.07–1.35)  <.01 
flupentixol  1.41  (1.11–1.79)  1.22  (1.03–1.44)  .02 
prochlorperazine  2.96  (2.08–4.21)  1.25  (0.97–1.63)  .09 
oral haloperidol  1.67  (1.47–1.89)  1.26  (1.15–1.38)  <.001 
clozapine  1.89  (1.24–2.89)  1.35  (1.00–1.81)  .05 
LAI vs. oral  0.92  (0.73–1.16)  0.86  (0.73–1.01)  .07 
Results are based on the traditional Cox Proportional Hazards model, and are adjusted for age at
index date, gender, number of all-cause hospitalizations within 2 years prior to index date,
calendar year of index date, time since diagnoses, temporal order of treatment, current use of
other treatments, and polypharmacy. Drug adjusted model is adjusted for drug related variables,
fully adjusted for all variables.

American Medical Association All rights reserved

eTable 8. The Association Between Use vs No Use of Medications and Risk of Rehospitalization
Due to Cardiovascular Reasons in Traditional Analysis in the Total Cohort (N = 18,018)

     
Treatment  Events  Person  Drug adjusted model  Fully adjusted 
    years     HR                     CI     HR                     CI              P Value 
antipsychotics  2002  54547  0.86  (0.77–0.97)  0.92  (0.82–1.02)  .11 
mood stabilizer  2158  63139  0.77  (0.69–0.86)  0.97  (0.88–1.07)  .51 
antidepressants  2063  47147  1.16  (1.03–1.31)  1.07  (0.97–1.19)  .18 
benzodiazepines  1239  24598  1.22  (1.05–1.40)  1.09  (0.98–1.21)  .12 
sedatives  1605  27448  1.77  (1.57–1.99)  1.17  (1.06–1.31)  <.01 
lithium  675  24815  0.65  (0.54–0.79)  0.81  (0.72–0.91)  <.001 
lamotrigine  273  12641  0.55  (0.46–0.66)  0.84  (0.68–1.03)  .10 
gabapentin  33  541  1.58  (0.93–2.70)  0.90  (0.61–1.35)  .62 
topiramate  8  506  0.43  (0.20–0.91)  0.95  (0.47–1.91)  .88 
valproic acid  966  26091  0.94  (0.83–1.07)  0.97  (0.87–1.09)  .65 
oxcarbazepine  52  881  1.55  (1.05–2.28)  1.28  (0.95–1.74)  .11 
carbamazepine  387  5409  1.87  (1.29–2.70)  1.33  (1.04–1.71)  .02 
clozapine  4  446  0.23  (0.05–1.10)  0.38  (0.08–1.88)  .24 
haloperidol (LAI)  4  210  0.52  (0.16–1.74)  0.60  (0.27–1.34)  .21 
oral zuclopenthixol  16  503  0.82  (0.42–1.62)  0.66  (0.36–1.21)  .18 
other  6  513  0.34  (0.12–0.95)  0.66  (0.25–1.74)  .40 
zuclopenthixol (LAI)  10  358  0.70  (0.35–1.38)  0.69  (0.37–1.29)  .25 
levomepromazine  230  7295  0.79  (0.65–0.95)  0.78  (0.67–0.92)  <.01 
flupentixol  15  495  0.72  (0.41–1.27)  0.80  (0.47–1.35)  .40 
aripiprazole  35  1979  0.52  (0.35–0.78)  0.86  (0.57–1.31)  .49 
chlorprothixene  150  4787  0.78  (0.60–1.00)  0.88  (0.71–1.08)  .21 
quetiapine  604  22092  0.70  (0.62–0.80)  0.88  (0.78–1.00)  .05 
periciazine  22  351  1.61  (0.81–3.20)  0.89  (0.52–1.55)  .69 
sulpiride  26  604  1.08  (0.67–1.76)  0.89  (0.59–1.34)  .59 
dixyrazine  9  222  1.05  (0.45–2.47)  0.90  (0.46–1.79)  .77 
oral haloperidol  131  1917  1.79  (1.20–2.68)  0.94  (0.70–1.26)  .68 
oral olanzapine  335  9783  0.89  (0.72–1.09)  0.94  (0.79–1.12)  .50 
chlorpromazine  73  1613  1.13  (0.67–1.90)  0.98  (0.65–1.46)  .91 
oral risperidone  298  5441  1.43  (1.19–1.71)  1.01  (0.86–1.18)  .91 
risperidone (LAI)  4  98  1.18  (0.36–3.81)  1.07  (0.41–2.82)  .88 
oral perphenazine  119  2349  1.35  (0.60–3.02)  1.18  (0.68–2.06)  .56 
prochlorperazine  9  36  6.36  (2.79–14.51)  1.31  (0.61–2.82)  .48 
perphenazine (LAI)  4  89  1.20  (0.37–3.87)  1.32  (0.42–4.10)  .64 
melperone  112  821  3.50  (1.75–6.98)  1.41  (0.78–2.56)  .25 
Results are based on the traditional Cox Proportional Hazards model, and are adjusted for age at
index date, gender, number of cardiovascular hospitalizations within 2 years prior to index date,
calendar year of index date, time since diagnoses, temporal order of treatment, current use of
other treatments, and polypharmacy. Drug adjusted model is adjusted for drug related variables,
fully adjusted for all variables.
Note: There were no events recorded for olanzapine LAI, results are thus not shown.

American Medical Association All rights reserved

eFigure. Exposure to Psychotropics During Follow-up and Time Resetting in Within-Individual Cox
Proportional Hazards Model

Supplementary Figure 1a Exposure to psychotropics during follow up.

Supplementary Figure 1b. Time resetting in within individual analyses.

a: Illustration of drug exposure periods for two patients. Follow-up time is divided into 3 periods for
patient 1 (periods A, B and C) and into 4 periods for patient 2. Follow-up time is reset at the start
of antipsychotics or mood stabilizers (also lithium for patient 2). Follow-up time is reset to zero at
the beginning of each period.

b: Illustration of periods after time resetting for patient 1. Periods A, B and C each start from time
zero. In the within-individual Cox Proportional Hazards model each individual forms his/her own
stratum. Thus, for the hospitalization that occurred at the end of period A, both periods A and C
contribute to the effect estimation (through the likelihood) at the time of the hospitalization.

This procedure has been previously used in studies on psychotropic drug use versus criminality,
and re-hospitalization due to mental disorders.1-4

American Medical Association All rights reserved

eReferences

1. Lichtenstein P, Halldner L, Zetterqvist J, et al. Medication for attention deficit-hyperactivity

disorder and criminality. N Engl J Med 2012;367:2006-2014.

2. Chang Z, Lichtenstein P, Långström N, Larsson H, Fazel S. Association between prescription

of major psychotropic medications and violent reoffending after prison release. JAMA
2016;316:1798-1807.

3. Tiihonen J, Tanskanen A, Hoti F, Vattulainen P, Taipale H, Mehtälä J, Lähteenvuo M.

Pharmacological treatments and risk of readmission tohospital for unipolar depression in
Finland: a nationwide cohort study. Lancet Psychiatry. 2017;4(7):547-553.

4. Tiihonen J, Mittendorfer-Rutz E, Majak M, et al. Real-world effectiveness of antipsychotic

treatments in a nationwide cohort of 29 823 patients with schizophrenia. JAMA Psychiatry.
201;74(7):686-693.

American Medical Association All rights reserved