Introduction
Suicide has a strong association with psychiatric disorders, particularly the major
affective illnesses, bipolar disorder (BD) and major depressive disorder (MDD). A
relatively recent discovery is that suicidal risk might be reduced by medicinal treatment
[1]
[2]
[3]. However, scientifically sound therapeutic investigations of suicide prevention
remain uncommon and very challenging. Indeed, only 1 treatment—the highly effective
antipsychotic drug clozapine—has regulatory recognition for ability to reduce suicidal
risk, and only for patients diagnosed with schizophrenia [4]
[5]. Here, we review findings pertaining to suicidal risks associated with long-term
treatment of mood disorder patients with various psychotropic drugs aimed at preventing
suicidal behavior. For BD, long-term use of lithium has particularly substantial evidence
of association with reduced risk of suicide and attempts. This evidence is reviewed
here, focusing on prevention of completed suicide and comparing these effects of lithium
to placebo and to other psychotropic drug treatments.
The average international (183 countries), adult, general population, age-standardized
suicide rate in 2015 was 10.5 (95% confidence interval [CI]: 9.56–11.3) per 100,000
per year (0.0105% per year) overall, 16.1 (14.6–17.5) for men and 5.24 (4.68–5.80)
for women (male/female ratio=3.07) [6]. The standardized mortality ratio for suicide in clinical groups versus the comparable
general population is highest in mood disorders, usually at 10–20-fold [7]. In BD, the reported suicide rate (per 100,000 per year) in 26 studies averaged
164 (CI: 5.0–324), and in 7 studies the rate was 366 in men and 217 in women [8]. These rates are, respectively, 20- and 41-times higher than among men and women
in the general population. The rate of suicide attempts among BD patients in 101 studies averaged 4240 per 100,000 per year (CI: 3780–4700)
(25.9-times above the suicide rate) and was 1.40-fold higher among women than men
[9]. However, these comparisons should be considered with caution as they derive from
different geographic populations. Depressive phases of BD, and especially mixed (agitated-dysphoric)
states, are far more likely to be associated with suicidal behaviors than manic or
hypomanic periods [10]
[11]. Moreover, rates of suicides and attempts are at least as high among type II as
in type I BD patients [9]
[12].
Among BD patients, suicide risk remains high despite the growing variety of treatments
with effective mood-stabilizing effects [13]. This disparity almost certainly reflects the great difficulty of effectively treating
depressive and mixed manic-depressive states of BD [14]
[15]
[16]
[17]
[18]
[19]
[20]
[21], which represent the majority of residual morbidity with clinically applied long-term
treatments [22]
[23]
[24]. Modern psychiatric treatments, rapid hospitalization, and even electroconvulsive
treatment (ECT) may be useful as short-term interventions but lack evidence of long-term suicide preventive effects [25]
[26]
[27]
[28]
[29].
Assessment of treatments for suicide prevention
Difficulties in conducting therapeutic studies to prevent suicide include clinical
and ethical risks involved in withholding treatment, such as in a placebo condition,
and seeking outcomes that may include life-threatening or lethal events, as well as
difficulties in identifying, recruiting, and retaining subjects; and the rarity of
suicide or even attempts as outcome measures [25]
[26]
[30]
[31]. An additional potential limitation of all studies of any therapeutic effect is
that patients who accept, tolerate, and sustain any type of long-term treatment may
be favorably self-selected and not entirely representative of all clinically encountered
patients. Many studies rely on surrogate outcomes such as self-injurious acts, communicated
suicidal plans or ideation, or interventions to avoid suicide—all of which may or
may not precede a suicide attempt. In addition, definitions and prevalence of nonfatal
suicide-related behaviors, and their quantitative, predictive association with suicide
itself, are matters of discussion related to distinctions among ideation, plans, and
attempts, including their intent and potential lethality [31]. Definitions and quantification of suicidal ideation are especially problematic.
Notably, the predictive value for suicidal behavior of passive ideation, such as thoughts
of weariness of life, probably differs from that of active ideation with specific
planning and preparing for a suicide attempt. Moreover, in research on suicide ideation
and behavior, crucial assessment of intent to die often is neglected [32]
[33]
[34].
The relative rarity of suicide requires assessment of large subject-samples for extended
times to detect a signal in studies of treatment effects or involves pooling data
across multiple studies. In addition, even randomized-controlled treatment trials
(RCTs) have shortcomings for the study of prevention of suicidal behavior. They include
potential unreliability of essentially incidental and passive ascertainment of suicidal
thoughts or behaviors based on typical “adverse event reporting” procedures under
conditions not designed to detect and assess suicidal events actively and explicitly.
However, efforts are being made to include regular, standardized assessments of suicidal
behaviors and other specified “adverse events” in trials of new, centrally active
drugs [33]. In addition, the relatively short duration of most treatment trials is unlikely
to yield statistically adequate numbers of suicidal behaviors. Another technical limitation
to assessing suicidal risks in treatment trials is that observed rates of events presumed to be related to suicide rarely are corrected for actual and matched
exposure times for treatments given, let alone for individual subjects. Such matching
for exposure can matter. For example, earlier dropping out of a trial arm involving
placebo treatment can artifactually make active drug treatment seem “riskier” than placebo in association with longer exposure and observation. Despite
all these limitations, RCTs are the most reliable sources of information about the
effect of treatment on even uncommon outcomes including suicidal behaviors. Accordingly,
this review emphasizes assessment of long-term clinical trials involving randomization to lithium compared to placebo or to other active agents for the treatment of mood
disorder patients.
Methods
Initially, we summarized findings from previous reviews on the topic that could include
fatal or nonfatal suicidal behavior and both BD and MDD subjects. In addition we considered
studies involving randomized treatments that compared lithium to placebo or other
active medicinal treatments and involved suicide as the specific outcome. We included
all identified studies reporting on suicides in RCTs with at least 1 suicide or attempt
in either arm of the study; studies with no suicidal acts in either trial arm were
considered uninformative and not included. Candidate studies for analysis were identified
from a recent systematic review and update [35]
[36], our own earlier reviews [37]
[38], and by an additional, systematic, computerized literature search of the PubMed
database for the past 5 years.
We identified 12 trials comparing risks of suicide with lithium versus placebo or
other treatments in 10 reports [39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]. Alternative treatments included 4 studies of lithium versus placebo, 3 versus an
antidepressant (amitriptyline), 4 versus 1 of 2 anticonvulsants (carbamazepine, lamotrigine),
and 1 versus an antipsychotic (olanzapine). For meta-analysis we used the Peto method
to deal with typically small numbers of suicides (numerators) as well as fixed-effects
modeling as interstudy heterogeneity was not significant. We pooled all diagnoses
and comparison treatments as well as meta-analyzed BD and MDD patients separately
and compared lithium versus placebo or versus other agents. Finally, we used meta-regression
analysis to address the possible influence of various factors of interest in the outcome
of the primary, pooled meta-analysis. Analyses used commercial software (Statview.5,
SAS Institute, Cary, NC) for spreadsheets and STATA.13 (StataCorp, College Station,
TX) for computations. Data are presented as means±standard deviation (SD) or with
95% CI.
In addition, we summarized findings from correlational studies that compared lithium
concentrations in local drinking water with publicly reported suicide (and homicide)
rates in the same regions, based on a recent systematic review [49] as well as literature searching of PubMed for the last 5 years. Findings were summarized
descriptively owing to the variety of reported outcomes. Finally, we summarized recent
progress with the use of treatments other than lithium aimed at reducing suicidal
risk, concentrating on mood-stabilizing anticonvulsants, antidepressants, and antipsychotics.
Results
Previous findings about lithium
We summarized reported findings regarding risks of suicides or attempts among patients
given randomly assigned, long-term treatment with lithium or alternative treatments
and meeting modern diagnostic criteria for BD or any recurrent major affective disorder
or only MDD [37]
[50]
[51]
[52]. These studies found highly significantly lower rates of suicidal behavior during
treatment with lithium, and similarly against both suicides and attempts, in BD, broad
samples of mood-disorder patients, and MDD patients specifically, and in comparison
with mood-stabilizing anticonvulsants ([Table 1]). These effects were sustained when corrected for reported exposure times in studies
of MDD patients and in comparisons of lithium to carbamazepine, lamotrigine, or valproate
([Table 2]).
Table 1 Risk of suicides or attempts with versus without long-term lithium treatment.
|
Outcome
|
Studies
|
RR (95%CI)
|
z-score
|
p-value
|
|
All suicidal acts
a
|
34
|
0.241 (0.176–0.331)
|
8.82
|
<0.0001
|
|
Suicides
|
26
|
0.252 (0.169–0.377)
|
6.71
|
<0.0001
|
|
Attempts
|
19
|
0.211 (0.132–0.337)
|
6.52
|
<0.0001
|
|
BD only
|
14
|
0.187 (0.126–0.279)
|
8.28
|
<0.0001
|
|
MDD only
b
|
17
|
0.215 (0.158–0.292)
|
9.82
|
<0.0001
|
|
Acts: lithium vs. anticonvulsants
c
|
6
|
0.350 (0.254–0.437)
|
5.24
|
<0.0001
|
a Includes suicides and attempts among BD and MDD patients not considered separately.
Studies with no suicidal events in either treatment arm (both numerators=0) are excluded.
b Adjusted for exposure time (rates: lithium, 0.174% per year vs. no lithium, 1.48%
per year) incidence rate ratio (IRR)=8.71 (CI: 2.10–77.2); exact p=0.0005; MDD subjects
only.
c Adjusted for exposure time (rates: lithium, 0.278% per year vs. anticonvulsants [carbamazepine,
lamotrigine, or valproate], 0.884% per year) IRR=2.96 (CI: 2.32–3.79); exact p<0.0001;
BD subjects only.
Data are adapted from previous meta-analytic studies [37]
[50]
[51]
[52].
Suicide with lithium vs. other treatments in randomized trials
Among the 12 long-term randomized trials, 6 involved only BD subjects and another
6 had mainly MDD patients or both BD and MDD cases. Patients were followed for an
average of 92.2 (95% CI: 67.0–117) weeks, with an average daily trough serum lithium
concentration of 0.73 (0.64–0.81) mEq/L ([Table 2]). With lithium treatment, there were 3 suicides among 974 patients (0.31% [0.06–0.90]),
compared to 21 suicides among 1070 subjects treated with placebo or other active agents
(1.96% [1.22–2.98], Fisher’s exact p=0.01). The risk of suicide with lithium treatment
was lower among both BD (p=0.05) and MDD or MDD+BP cases (p=0.01), and compared only
to placebo (p=0.04) or to other active drugs (p=0.02) ([Table 2]).
Table 2 Suicide risk in randomized trials of lithium vs. other treatments for major affective
disorders.
|
Trial
|
Year
|
Lithium
|
Other Rx
|
Dx
|
Follow-up (wk)
|
Age
|
[Li+] (mEq/L)
|
Other Rx
|
|
Suicide
|
No Suicide
|
Risk (%)
|
Suicide
|
No Suicide
|
Risk (%)
|
|
Prien et al.
|
1973
|
0
|
45
|
0.00
|
1
|
38
|
2.56
|
MAD
|
104
|
39
|
0.80
|
PBO
|
|
Prien et al.
|
1973
|
0
|
101
|
0.00
|
1
|
103
|
0.96
|
BD
|
104
|
39
|
0.70
|
PBO
|
|
Glen et al.
|
1984
|
0
|
57
|
0.00
|
1
|
49
|
2.00
|
MDD
|
128
|
45
|
0.90
|
AMI
|
|
Greil et al.
|
1996
|
0
|
40
|
0.00
|
1
|
40
|
2.44
|
MDD
|
128
|
41.5
|
0.60
|
AMI
|
|
Thies-Flechtner et al.
|
1996
|
0
|
46
|
0.00
|
1
|
46
|
2.13
|
MDD
|
130
|
41.5
|
–––
|
AMI
|
|
Thies-Flechtner et al.
|
1996
|
0
|
87
|
0.00
|
6
|
82
|
6.82
|
BD
|
130
|
41.5
|
–––
|
CBZ
|
|
Greil et al.
|
1997
|
1
|
86
|
1.15
|
5
|
83
|
5.68
|
BD
|
128
|
41.5
|
0.60
|
CBZ
|
|
Bauer et al.
|
2000
|
0
|
14
|
0.00
|
1
|
14
|
6.67
|
MDD
|
20
|
≥18
|
0.75
|
PBO
|
|
Calabrese et al.
|
2003
|
0
|
121
|
0.00
|
1
|
220
|
0.45
|
BD
|
78
|
≥18
|
0.58
|
LTG
|
|
Tohen et al.
|
2005
|
1
|
213
|
0.47
|
0
|
217
|
0.00
|
BD
|
52
|
≥18
|
0.75
|
ONZ
|
|
Lauterbach et al.
|
2008
|
0
|
84
|
0.00
|
3
|
80
|
3.61
|
MDD
|
52
|
≥18
|
0.70
|
PBO
|
|
Licht et al.
|
2010
|
1
|
77
|
1.28
|
0
|
77
|
0.00
|
BD
|
52
|
≥18
|
0.90
|
LTG
|
|
Weighted means (95% CI)
|
n=12
|
3/974 0.308% (0.06–0.90)
|
|
|
21/1070 1.96% (1.22–2.98)
|
|
|
6 BD
6 MDD
or MAD
|
92.2 (67.0–117)
|
≥18
|
0.728 (0.645–0.811)
|
5 Rxs
|
Some MDD trials include a minority of BD. Studies with no suicides with lithium or
other treatments are not included.
Overall Fisher’s exact p=0.001; with active treatments only, p=0.02, and vs. placebo,
p=0.04; for BD subjects, Fisher’s exact p=0.05; for MDD subjects, p=0.01.
AMI: amitriptyline; CBZ: carbamazepine; LTG: lamotrigine; MAD: major affective disorders;
ONZ: olanzapine; PBO: placebo; Rx: treatment.
Data are derived from previous reviews [35]
[36].
These findings based on simple weighted averages also were sustained by meta-analyses.
That is, lithium treatment was associated highly significantly with lower suicide
risk than with placebo or with other active treatments, overall (z=3.65, p<0.0001),
as shown in [Fig. 1]. This trend was followed in 10/12 trials even though the numbers of suicides per
trial was small, as expected (0–1 with lithium, 0–6 with other treatments). Exceptions
(lithium apparently less effective) were 1 trial each of lithium versus olanzapine or lamotrigine, in which
there was only 1 suicide with lithium and none with the other agents. In addition,
suicide was significantly less frequent with lithium versus placebo (z=2.47, p=0.01)
and versus other drug treatments (z=2.79, p=0.005). Meta-regression analysis following
the overall meta-analysis indicated that both diagnosis (z=1.00, p=0.32) and comparison
treatment (placebo or other drugs; z=0.45, p=0.65) lacked significant influence on
the outcome. In addition, the duration of treatment (20–130 weeks; mean=1.77 [1.29–2.25]
years; z=1.07, p=0.28), mean serum lithium concentration (z=1.67, p=0.10), and year
of publication (z=0.12, p=0.90) also lacked influence on the outcome of the meta-analytic
comparison of lithium to all other treatments.
Fig. 1 Forest plot of comparative risk of suicide during long-term, randomized-controlled
treatment (mean: 92 [67]
[68]
[69]
[70]
[71]
[72]
[73]
[74]
[75]
[76]
[77]
[78]
[79]
[80]
[81]
[82]
[83]
[84]
[85]
[86]
[87]
[88]
[89]
[90]
[91]
[92]
[93]
[94]
[95]
[96]
[97]
[98]
[99]
[100]
[101]
[102]
[103]
[104]
[105]
[106]
[107]
[108]
[109]
[110]
[111]
[112]
[113]
[114]
[115]
[116]
[117] weeks) with lithium carbonate versus placebo (PBO) or a psychotropic drug (AMI:
amitritpyline; CBZ: carbamazepine; LTG: lamotrigine; or ONZ: olanzapine) among 1070
subjects with mood disorders (BD: bipolar; MAD: major affective; or MDD: major depressive
disorder). Studies with no suicides in any treatment-arm are omitted; due to low numerators,
Peto’s method for meta-analysis was employed. In 10/12 comparisons, lithium was associated
with fewer suicides (in the 2 exceptions with lithium vs. olanzapine [48] or lamotrigine [45], the numerators [suicides] differed by only 1 vs. 0). The overall, pooled odds ratio
(OR with 95% CI)=0.222 [0.099–0.497]; z=3.65, p<0.0001. Meta-regression found no effect
of the comparison treatment, diagnosis, weeks of treatment, or year of publication.
Separate meta-analyses found significant superiority of lithium versus placebo (pooled
OR=0.132 [CI: 0.026–0.656]; z=2.47, p=0.013) or other drugs (OR=0.264 [0.104–0.674];
z=2.79, p=0.005), and in both BD (OR=0.290 [0.108–0.784; z=2.44, p=0.015) and MDD
or MDD cases (OR=0.131 [0.033–0.524]; z=2.87, p=0.004).
Lithium in drinking water
Finally, for lithium, we carried out a systematic search and summarized findings of
a growing number of studies that evaluated associations of the low concentrations
of lithium in local drinking water with reported suicide or homicide rates in the
same communities or regions. We found 18 such reports [53]
[54]
[55]
[56]
[57]
[58]
[59]
[60]
[61]
[62]
[63]
[64]
[65]
[66]
[67]
[68]
[69]
[70]. Several of these reports involved an apparently single set of data from Austria,
and their findings are pooled [57]
[58]
[59]
[62]. Most of the studies (12/15) found a significant, inverse association between local
lithium levels and local rates of suicide or homicide, though sometimes (inconsistently)
only in men or women or seemingly modified by local climatic conditions or altitude.
One notable study involving a national Danish sample found no such association when
individual people were considered rather than relying on group (ecological) associations
[63]. Two studies found a protective relationship with homicide rates in addition to
suicide rates [54]
[55]. Although these findings are not entirely consistent, they are suggestive that a
favorable relationship might exist between higher local concentrations of lithium
in drinking water and less violent behavior (suicide or homicide). Notably, concentrations
of lithium in ground water were low (averaging 2–83 µg/L or 6–18 µEq/L) and far below
serum concentrations considered to be therapeutic in BD (600–1200 µEq/L) [13]. Nevertheless, it might be that years of exposure to even such low concentrations
of lithium can have biological effects ([Table 3]).
Table 3 Studies of lithium in drinking water and risk of suicide or homicide.
|
Reports
|
Location
|
Regions Sampled
|
[Li+] (µg/L)
|
Findings
|
|
Dawson et al. 1972
|
Texas
|
24
|
0–139
|
Less suicide & homicide with higher [Li];
urine [Li] correlated with water [Li]
|
|
Schrauzer & Shrestha 1990
|
Texas
|
27
|
0–160
|
Less suicide with higher [Li],
only ≥70 ug/L
|
|
Ohgami et al. 2009
|
Japan
|
18
|
0.7–59
|
Less suicide with higher [Li]
|
|
Kabacs et al. 2011
|
East England
|
47
|
1–21
|
No association
|
|
Kapusta et al. 2011; Helbich et al. 2012, 2013, 2015
|
Austria
|
99
|
11–27
|
Less suicide with higher [Li], only at lower altitudes, not in women
|
|
Blüml et al. 2013
|
Texas
|
226
|
2.8–219
|
Less suicide with higher [Li]
|
|
Sugawara et al. 2013
|
Japan
|
40
|
0–13
|
Nonsignificantly less suicide with higher [Li], especially in women
|
|
Giotakos et al. 2013
|
Greece
|
34
|
11–21
|
Less suicide with higher [Li]
|
|
Giotakos et al. 2015
|
Greece
|
34
|
11–21
|
Less homicide with higher [Li]
|
|
Ishii et al. 2015
|
Japan
|
274
|
0–130
|
Less suicide with higher [Li],
only in men
|
|
Pompili et al. 2015
|
Italy
|
145
|
0.1–61
|
Less suicide with higher [Li],
mainly in women in 1980 s
|
|
Shiotsuki et al. 2016
|
Japan
|
153
|
0.1–43
|
Less suicide with higher [Li], only in men, but also with more sunshine, higher ambient
temperature & less rain
|
|
Knudsen et al. 2017
|
Denmark
|
3.7 M individuals
|
0.6–31
|
No association
|
|
König et al. 2017
|
Chile
|
12
|
1.0–207
|
Less suicide in high-[Li] region (high Atacama Desert)
|
|
Liaugaudaite et al. 2017
|
Lithuania
|
22
|
0.5–36
|
Less suicide with higher [Li], only in men
|
|
|
|
|
|
|
Totals/averages
|
9 countries
|
1121 regions
|
2.06 (0–4.29) to 83.4 (41.5–125)
|
12/15 inverse correlation
4/12 in men only
2/12 in women mainly
|
Many of these studies were reviewed by Vita et al. (2015). Lithium concentrations
in drinking water (usually based on mass spectroscopy assays) ranged from 0 to 219
and averaged 2–83 µg/L or 0.30 [0–0.62] to 12.0 [6.0–18.1] µEq/L. Note that 2 studies
found less homicide with higher concentrations of lithium in drinking water [55]
[68]. Knudsen et al. [63] found no association of exposure to lithium in drinking water in Denmark and risk
of suicide in bipolar disorder.
Other treatments: anticonvulsants
There is little research that directly compares suicidal risks during treatment with
proved or putative mood-stabilizers other than lithium [47]
[71]
[72]
[73]
[74]. However, several studies found substantially lower average risks of suicidal behavior
with lithium than with carbamazepine or valproate among BD or schizoaffective patients
[47]
[71]
[75]
[76]
[77]
[78]
[79]. In a meta-analysis [75], we compared protective effects against suicidal behavior of lithium versus several
mood-stabilizing anticonvulsants (mainly valproate and some use of carbamazepine or
lamotrigine) in 6 direct comparisons (half involved randomized assignments to treatments)
including more than 30,000 patients. Subjects were at risk longer with lithium than
with an anticonvulsant (31 vs. 19 months), which may tend to increase risk of suicidal events with lithium. Nevertheless, the observed rate of suicidal
acts averaged 0.3% per year during treatment with lithium versus 0.9% per year with
anticonvulsants, yielding a highly significant meta-analytically pooled risk ratio
of 2.86 (CI: 2.29–3.57; p<0.0001). This finding indicates a nearly 3-fold superiority
favoring lithium over the few anticonvulsants tested in this way. However, addition
of valproate as well as lithium yielded lower suicidal risks compared to treatment
with only antipsychotics in a Danish study of over 16,600 persons sampled for 6 years
[80]. In other studies valproate and lithium had similar associations with suicidal behavior,
suggesting that both may reduce suicidal risk [73]
[81]
[82]
[83]. Taken together, these studies suggest that anticonvulsants may have some beneficial
effects on suicidal behavior, including patients with major affective disorders, though
possibly less than lithium.
A contrasting perspective arose from a U.S. Food and Drug Administration (FDA) meta-analysis
of placebo-controlled trials involving 11 anticonvulsants, submitted for regulatory
review [76]. This review found more prevalent, incidentally reported suicidal ideation and behavior with anticonvulsants
than with placebo, at least among patients with epilepsy, though not in psychiatric
patients [76]. The lack of such adverse effects among psychiatric patients was supported by several
other studies [80]
[82]
[84]
[85]
[86]. In conclusion, research on anticonvulsants and suicidal risk remains limited, inconsistent,
and inconclusive, although lithium appears to be superior in preventing suicidal behavior,
based on direct comparisons ([Table 1] and [Table 2]).
Other treatments: antidepressants
The strong association of depressive morbidity with suicide in mood-disorder patients
would suggest that treatment with antidepressants might reduce suicidal risk. Evidence
for short- and long-term efficacy of antidepressant treatment in unipolar MDD is substantial
[13]
[19]
[87]
[88]
[89]
[90]
[91]
[92]
[93] but not universally accepted [93]. However, antidepressant treatment is not explicitly approved for use in bipolar
depression and may not be effective or safe long-term in BD, in which its prophylactic
value versus destabilizing risks remains insufficiently studied [19]
[94]
[95]. Antidepressant response in BD may be better among patients without agitated-dysphoric
forms of depression or mixed features [95] and without previous rapid-cycling [97]. Conversely, suicidal risk is likely to increase with antidepressant treatment in
some cases of either BD or MDD involving agitation, anger, dysphoria, restlessness,
irritability, insomnia, or behavioral disinhibition, especially when complicated by
substance abuse, and in juvenile patients [28]
[98]
[99]
[100]
[101]
[102]
[103]
[104]
[105]
[106]
[107]
[108]
[109]. An emerging view is that agitated-dysphoric depression can include “mixed-features”
based on DSM-5 diagnostic criteria [11]
[110]. Studies of associations of antidepressant treatment and suicidal behavior vary
widely in design, are limited mainly to MDD, and provide inconsistent evidence concerning
suicides or attempts, which have usually been noted incidentally as adverse effects
rather than as an explicit outcome measure.
There is some evidence of lower suicidal risk during trials of treatment with an antidepressant
versus placebo, mostly among adult MDD patients. Most of this evidence is based on
questionable extraction of data from specific items in depression symptom-rating scales,
which are weighted toward suicidal ideation [104]
[105]
[106]
[109]
[111]
[112]
[113].
Lower rates of suicide with greater use of antidepressants have been reported in some
pharmacoepidemiological studies based on finding inverse correlations between use
of antidepressants and suicide rates in some regions or countries, particularly following
the introduction of modern antidepressants in the 1990 s [114]. However, such ecological-correlational studies have not yielded consistent findings
across countries or regions. Correlations of interest have largely been limited to
Nordic countries and the United States but appear to be randomly distributed across
the world [19]
[98]
[113]
[114]
[115]. Moreover, in the United States and Sweden, at least, declining suicide rates were
noted at least a decade before introduction of fluoxetine as the first clinically
successful modern antidepressant in the late 1980 s [19]
[113]
[114]. In general, such correlational analyses are fraught with potential confounding
by uncontrolled factors [114]
[116].
Additional studies involving mainly retrospective observations of large cohorts of
depressed patients and case-control comparisons have yielded inconsistent and inconclusive
findings [19]
[100]
[102]. In one clinical follow-up study, during treatment, monthly assessments indicated
that 81% of subjects considered suicidal at intake became nonsuicidal with treatment
and time, and only 0.5% of initially nonsuicidal subjects reported new suicidal thoughts,
with no new attempts [117]. Not all patients who had started treatment were followed-up, and details of suicidal
status during the initial days of treatment were limited. These observations underscore
the difficulty of evaluating interactions of treatment, time, and suicidal behavior,
long-term.
Randomized controlled trials should provide the best information on effects of antidepressant
treatment on suicidal risks, but individual trials are limited in numbers and exposure
times, reducing their ability to identify relatively rare suicide-related outcome
events. Moreover, as their identification usually has been based on incidental and
passively acquired nonexplicit assessments of suicidal outcomes, some events may be
missed, even after having made efforts to exclude potentially suicidal subjects. This
situation may change as the FDA makes efforts to require explicit rating of suicidal
status in trials of centrally active substances [34]. Despite efforts to limit suicidal risk in treatment trials, rates of suicidal behaviors
may be at least as high in antidepressant RCTs (usually involving acutely depressed
subjects) as in cohort studies of MDD patients in various clinical states [98]
[103]
[107]. For example, suicide rates pooled across a large number of trials of modern and
older antidepressants or placebo were similar with all treatments, averaging 0.862%
per year, or 57 times above the approximate average international general population
rate of 015% per year and 17 times above the reported rate of approximately 0.050%
per year in outpatients in various phases of MDD [89]
[98]
[108]
[111]
[113]. Another caveat is that the high observed rates from the cited meta-analyses of
controlled trials may be exaggerated by annualizing observed rates based on brief
exposure times (typically 6–12 weeks) in most trials in acute depression (for example,
the apparent, annualized rate of 1 suicidal event among 100 subjects in a 12 week-trial
[1.0%] would be 4.3% per year). An additional concern is that, short of well-matched
exposure times, it is possible that suicidal risk may vary artifactually between shorter
and longer exposure times, for example, obscuring potential benefits of active treatments
if early dropout rates are more likely with placebo. Most studies and meta-analyses
have found only minor differences or somewhat lower rates of suicidal behaviors in
depressed patients treated with antidepressants versus a placebo [118]
[119]
[120]. Others have detected indications of somewhat greater risks, mainly of suicidal
ideation, with antidepressants versus placebo controls, particularly in juveniles
and very young adults, but decreased risks in older adults, with an overall outcome
of no difference [101]
[105]
[112]
[121]
[122]. These analyses assume that the trials considered remained well randomized despite
potentially dissimilar dropout rates and that temporal exposures in both drug and
placebo arms remained well balanced over time. They also assume that such surrogate
measures as suicidal ideation or even minor self-injurious behaviors, identified as
“adverse effects,” are fairly and comparably ascertained in different treatment groups
and that they have important predictive value for suicide itself. All of these are
questionable assumptions.
To recapitulate, research on effects of antidepressants on suicide risk presents important
and difficult methodological problems. However, current data, though based on thousands
of subjects treated with antidepressants versus placebo, do not provide sufficiently
rigorous and consistent information to support either an increase or a decrease of
suicidal ideation or behavior in mood-disorder patients. They raise the possibility
that increased suicidal ideation and possibly also suicidal behaviors may be increased
with antidepressant treatment in young patients treated with antidepressants but decreased
in older adults. At this time, it is not possible to compare effects on suicidal behavior
with antidepressants versus lithium, in particular. Most of the evidence regarding
suicidal risk in association with antidepressant treatment pertains to MDD patients,
whereas long-term research on antidepressant treatment in BD remains far less well
developed, although innovative treatments for depressive phases of BD are emerging
and should be evaluated for potential antisuicidal effects [20]
[123]
[124]. Finally, it is our clinical impression that antidepressants should be avoided in
the presence of depressive states accompanied by “mixed” symptoms including agitation,
anger, or insomnia, in which suicidal risks are elevated.
Other treatments: antipsychotics
With the exception of clozapine for schizophrenia [4]
[5], no other treatment has regulatory approval of an indication for an antisuicidal
effect, including lithium. A common feature of patients who appear to benefit from
long-term treatment with lithium or clozapine is that they require and receive especially
close clinical monitoring that may provide added support and facilitate early identification
of emerging symptoms that might lead to suicidal behavior. This possibility was not
supported in the InterSePT trial for schizophrenia patients in which clinician contact
time was similar between treatment options [5]. Nevertheless, we reported previously that various measures that can be considered
indices of access to clinical care were closely correlated with state suicide rates in the United States [125].
In the past several years, several epidemiological studies have added to the conclusions
that long-term antipsychotic treatment is associated with reduced all-cause mortality
rates, and risk of suicide in particular, and that clozapine is probably more effective
against suicidal risk than other older or newer antipsychotic agents [126]
[127]
[128]
[129]. However, almost all of this work involves patients diagnosed with schizophrenia
rather than mood disorders.
Impressions arising from the preceding overview of effects of treatments other than
with lithium on risk of suicidal behavior are summarized in [Table 4].
Table 4 Summary of findings from studies of pharmacological treatments aimed at reducing
suicidal risks other than lithium.
|
Treatment
|
Timing
|
Findings
|
Limitations
|
|
Anticonvulsants
|
Short- and long-term effects are not established.
|
Valproate most studied. Anticonvulsants may be less effective vs. suicide than lithium.
|
Suicidal behaviors incidental, not explicit outcomes. FDA proposal that anticonvulsants
may increase suicidal risk is not supported for mood disorders.
|
|
Antidepressants
|
Short- and long-term effects on suicide risk not established.
|
Inconsistent findings in controlled & uncontrolled trials in MDD; little research
in BD; may increase risk of nonlethal suicidal behavior at ages<25 years but decrease
risk in older adults.
|
Studies lack of data for actual, matched exposure times. Suicidal status usually assessed
passively and incidentally rather than explicit outcome measure.
|
|
Antipsychotics
|
Short- and long-term effects on suicide risk are not adequately tested in mood disorders.
|
Clozapine: only FDA-recognized “antisuicidal” treatment (schizophrenia only). Modern
antipsychotics require further study, especially those with antidepressant effects
in BD.
|
Clozapine’s status based on one controlled trial with no effect on mortality (surrogate
outcome measures only).
|
Adapted from Tondo and Baldessarini [28].
Discussion
The findings reviewed here illustrate many difficulties for the design, conduct, and
interpretation of studies aimed at testing for antisuicidal effects of specific treatments.
The ethics of studies with suicide as a potential outcome are daunting and make use
of placebo-controlled conditions in randomized trials highly problematic. Also, the
infrequency of suicidal behaviors, even in high-risk samples, makes it difficult to
reach sound conclusions from samples of modest size followed for limited times, even
with well-matched exposures in parallel groups randomly assigned to alternative treatments.
In addition, suicidal risk appears to vary with age, the type, duration, and severity
of affective illnesses, and the timing of interventions in different phases of illness.
It is possible that patients who accept, tolerate, benefit from, and continue to take
a treatment for any purpose may differ in unknown but critical ways from those who
refuse or discontinue the treatment. Clearly, randomized and prospective trials involving
explicit outcome measures relevant to risk of suicidal behavior are required. Such
trials are rare, probably reflecting the ethical, clinical, and liability challenges
of efforts to test for reduction of suicidal risks, as well as the lack of clear commercial
advantages of such an achievement. For example, there is little commercial interest
in lithium as an unpatentable mineral, and having an antisuicide indication for clozapine
appears to have had little effect on the already small market of this important but
potentially toxic drug. Moreover, ethically feasible, head-to-head comparisons of
similarly plausible experimental treatments aimed at preventing suicide are not likely
to be favored by manufacturers of only one of the products. More generally, the low
frequency of suicide itself severely constrains market interest in a treatment aimed
at preventing it.
Mood disorders are associated with major increases of suicidal behavior in association
with depressed mood. Risks are especially high in BD and in mixed, dysphoric-agitated
states, and perhaps also with anger, aggression, or impulsivity and insomnia—all of
which are particularly prevalent in BD patients, but also occur in MDD, and contribute
to suicide risk. In such conditions, antidepressants can risk worsening arousal and
agitation, potentially even increasing suicidal risk, at least temporarily, especially
early in treatment of young patients and without close, initial clinical follow-up.
In general, and particularly during new use of antidepressants in BD or MDD patients
calls for thoughtful and responsive clinical monitoring, especially in the initial
days of treatment, seeking early detection of worsening or emerging agitation, dysphoria,
restlessness, insomnia, anger, and psychotic symptoms, including mixed manic-depressive
states. Use of mood-stabilizing or antipsychotic agents in depressed patients with
agitation is probably a safer and more rational option and may reduce conditions conducive
to suicide, based on clinical observations, although this clinical impression required
experimental testing.
The effectiveness of lithium treatment in preventing suicide is likely to be associated
with reduced impulsivity and aggressiveness associated with depression or dysphoric-agitated,
mixed states, which are particularly associated with suicidal acts [35]
[52]
[130]
[131]
[132]
[133]. Alternatively, lithium may have specific effects against suicide independent of
its mood-stabilizing actions, as suicidal risk has been found to be reduced even among
patients whose primary mood symptoms had not responded well to lithium [134]
[135]
[136]. The apparent, major beneficial effect of lithium treatment on risk of suicides
and attempts may be superior to any such effect of anticonvulsants proposed as mood-stabilizers,
and comparisons of lithium with modern antipsychotic drugs are needed.
Finally, the preceding overview underscores the conclusion that research support for
specific therapeutic interventions aimed at reducing suicidal risk in mood-disorder
patients remains limited. Treatments with evidence of value, including clozapine for
schizophrenia or lithium for major mood disorders, seem to be most useful for long-term
reduction of suicidal risk, whereas ECT and rapid hospitalization probably are effective
short-term in acute suicidal crises but are not known to have long-term preventive
effects. Nevertheless, the need for effective clinical management of suicidal patients
makes it essential to rely on clinical experience, with skillful and sensitive application
of direct and supportive personal interventions in an environment as protective as
possible.
