Drug and Alcohol Dependence 76S (2004) S11S19

Association of alcohol and drug use disorders and completed suicide:

an empirical review of cohort studies
Holly C. Wilcoxa,b, , Kenneth R. Connerc , Eric D. Cainec
a

Department of Psychiatry and Behavioral Sciences, Center for Family Research, George Washington University, Washington DC 20037, USA
b Department of Mental Health, Bloomberg School of Public Health, Johns Hopkins University, 624 N. Broadway Room 888,
8th Floor, Baltimore, MD 21205, USA
c Laboratory of Violence and Victimization, Department of Psychiatry, Center for the Study and Prevention of Suicide,
University of Rochester Medical Center, Rochester, New York, USA
Received 8 July 2004; accepted 4 August 2004

Abstract
Background: This study updated and expanded upon Harris and Barracloughs empirical review [Harris, E.C., Barraclough, B., 1997. Suicide
as an outcome for mental disorders. A meta-analysis, Br. J. Psychiatry 170, 205228] of retrospective and prospective cohort studies of alcohol
and drug use disorders and suicide.
Method: Studies presenting data on alcohol and drug use disorders and suicide originally identified by Harris and Barraclough were used in
this study. To find additional studies, (1) the location of English language reports on MEDLINE (19942002) were identified with the search
terms substance-disorders with mortality and follow-up, (2) read throughs were conducted of four prominent alcohol and drug specialty
journals from 1966 through 2002, and (3) the reference sections of studies that met criteria were searched for additional reports. This strategy
yielded 42 new studies meeting eligibility criteria.
Results: The estimated standardized mortality ratios (SMR; 95% confidence interval) for suicide were as follows: alcohol use disorder (979;
95% CI 8981065; p < 0.001), opioid use disorder (1351; 95% CI 10471715; p < 0.001), intravenous drug use (1373; 95% CI 10291796; p
< 0.001), mixed drug use (1685; 95% CI 14731920; p < 0.001), heavy drinking (351; 95% CI 251478; p < 0.001). SMR estimates stratified
by sex were also calculated.
Conclusions: Additional studies on the association of suicide and mixed drug use, heavy drinking, and alcohol use disorders in women
augmented the findings of Harris and Barraclough, along with a novel estimate for intravenous drug use, a byproduct of intensive research
on HIV in the past decade. There is a large empirical literature on alcohol use disorders and suicide and a moderate literature on suicide and
opioid use disorders and IV drug use. There remains limited prospective data on the association of suicide and other drug use disorders (e.g.,
cocaine, cannabis).
2004 Elsevier Ireland Ltd. All rights reserved.
Keywords: Suicide; Substance-related disorders; Epidemiology; Mortality

1. Introduction
Evidence for an association of suicide with alcohol and
drug use disorders is based largely upon retrospective and
prospective cohort studies and postmortem psychological
autopsy studies. Empirical reviews of cohort studies (Harris

Corresponding author. Tel.: +1 410 614 2852; fax: +1 410 955 9088.
E-mail address: hwilcox@jhsph.edu (H.C. Wilcox).

0376-8716/$ see front matter 2004 Elsevier Ireland Ltd. All rights reserved.
doi:10.1016/j.drugalcdep.2004.08.003

and Barraclough, 1997) and psychological autopsy studies

(Cavanagh et al., 2003) have demonstrated that alcohol and
drug use disorders are strongly associated with suicide. However, given limited sample sizes, psychological autopsy studies have typically combined drug use disorders into a single
broad category (e.g., Lesage et al., 1994). As a result, psychological autopsy studies have provided little data on the
association of specific drug use disorders and suicide, with
the exception of alcohol. Harris and Barracloughs empiri-

S12

H.C. Wilcox et al. / Drug and Alcohol Dependence 76S (2004) S11S19

cal review of cohort studies containing data on alcohol and

drug use disorders and suicide consisted primarily of reports
on alcohol use disorders and, to a lesser extent, opioid use
disorders. The magnitude of the association of other drug
use disorders (e.g., cannabis, cocaine) and suicide based on
cohort studies is unclear.
Suicide research has predominantly been conducted in
western, industrialized countries where women have lower
rates of suicide and a lower prevalence of alcohol and drug use
disorders, contributing to small subsamples of women with
drug use disorders in suicide studies. The tendency of many
earlier studies of alcohol and drug misuse to exclude women
also has contributed to a dearth of knowledge. Murphy (1998)
reviewed sex differences in alcoholism in American and European studies and found that sex differences were inconsistent. Some (Harris and Barraclough, 1997) but not all evidence (Conner et al., 2003a) suggests that sex moderates the
association of alcohol use disorders and suicide. However,
there are meager data on this issue with regard to other drug
use disorders.
This study updates and augments the empirical review
by Harris and Barraclough of retrospective and prospective cohort investigations that have provided estimates of
the suicide risk (standardized mortality ratios, SMRs) associated with specific alcohol and drug use disorders. Harris
and Barracloughs review covered studies identified through
1993 (using a MEDLINE search) and through mid-1995
(using read throughs of predominantly psychiatry journals).
There have been several reports published since that time,
necessitating an update. Moreover, Harris and Barraclough
covered a variety of mental disorders (e.g., major depression, schizophrenia) and did not exclusively focus on alcohol and drug-related disorders per se. Given our explicit
focus on alcohol and drug use disorders, we expanded
upon their approach in this domain by targeting alcohol
and drug use disorders in the MEDLINE search strategy
and by reading through four major alcohol and drug specialty journals. We also examined heterogeneity between
studies in each alcohol and drug use disorder strata, as
well as characteristics of studies (e.g., publication year, geographic region) that may account for heterogeneity in reports of alcohol and drug use disorders and risk for suicide. By expanding upon and updating their search, our goal
has been to identify additional investigations of the association of suicide and drug use disorders, including studies
that stratified by sex in order to examine the association of
drug use disorders and suicide individually in women and
men.

2. Methods
2.1. Search strategy
Our approach was to carry over Harris and Barracloughs
fundamental approach to maintain consistency as well as

to expand it in a targeted manner to provide better coverage of the alcohol and drug use disorders literature. Consistent with the review by Harris and Barraclough, the inclusion criterion were: (1) 2-plus years follow-up, (2) <10%
lost to follow-up, (3) the observed number of suicides are
specified, and (4) the authors provide the expected value
for suicides or provide the necessary data in order to obtain it from World Health Organization mortality statistics
(WHO, 1961; WHO, 19622002). Titles and abstracts were
reviewed and studies that appeared to be relevant were obtained and analyzed to determine if they met eligibility criteria. Studies were excluded if their sample was not independent of another study. To identify studies of alcohol
and drug misuse and suicide not previously identified by
Harris and Barraclough, we used the medical subject heading (MeSH) search string substance-related disorders with
mortality and follow-up using English language and human study limits in PubMed from 19661993, and then extended the search from 1994 through the end of December,
2002. Harris and Barraclough had conducted read throughs
up to mid-1995 of The Lancet, British Medical Journal, New
England Journal of Medicine, British Journal of Psychiatry, Psychological Medicine, Archives of General Psychiatry, and Acta Psychiatrica Scandinavica. We carried on and
read these journals through from 1995 to 2002. To increase
the coverage of the alcohol and drug use disorder literature, we also conducted read throughs from 1966 to 2002
of four prominent drug and alcohol specialty journals: Addiction, Alcoholism: Clinical and Experimental Research,
Drug and Alcohol Dependence, and Journal of Studies on
Alcohol. These journals were selected because they are longstanding journals with solid impact factors and broad readership. Consistent with Harris and Barraclough, reference
lists in papers that met eligibility criteria provided further
citations. We abstracted a total of 42 papers, 27 from the
MEDLINE search and 15 from read throughs and the review
of reference sections of abstracted papers. Twenty papers
had been published since Harris and Barracloughs study and
22 papers were identified that were published prior to their
investigation.
2.2. Data abstraction
We used the same methodology as Harris and Barraclough
to obtain observed and expected values for suicide for each
cohort study. Specifically, observed values were confined
to deaths reported as suicide, probable or possible suicides
were not included. Expected values for suicide were given in
three (7%) of the 42 papers. We used the expected values
provided by the study authors or estimated expected values using the procedure stated in Harris and Barraclough
for the remaining 39 reports by using WHO statistical reports (World Health Organization, 19622002) for the relevant country and years, combined with the age/sex composition of the sample and mean observation period for each
study.

H.C. Wilcox et al. / Drug and Alcohol Dependence 76S (2004) S11S19

2.3. Reliability analyses of the overall estimates of

expected suicides
We sought to determine if it were reasonable to carry over
the observed and expected number of suicides determined by
Harris and Barraclough for the previously identified studies,
rather than generate independent estimates. Therefore, an interrater reliability analysis was conducted in which one of the
authors independently estimated the observed and expected
suicides for the articles reviewed by Harris and Barraclough
concerning alcohol use disorders. The alcohol disorder category was chosen because it contained the most reports on
which to base the reliability analysis. The observed number
of suicides differed by 0.9%, the expected number differed by
1.6%, and the derived standardized mortality ratios differed
by 0.7%. The comparable results indicated that the data provided by Harris and Barraclough (1997) could be carried over
to the present study.
2.4. Statistical methods
The standardized mortality ratio (SMR) is a relative index
of mortality, expressing the mortality experience of a given
study population relative to that of a comparison (standard)
population. In this study, the SMRs were used to estimate
whether risk for suicide among those with specific alcohol
or drug use disorders were at greater risk than expected in
the general population. SMRs were calculated by dividing
the observed number of suicides by the expected number
of suicides and multiplying by 100, in order to yield results
without decimals (Gordis, 1996). For example, a SMR of 100
indicates that the observed number of suicides is the same
as the expected number of suicides whereas an SMR of 500
indicates a five-fold greater number of observed compared to
expected suicides. The confidence intervals were estimated
based upon the assumption that the observed variables are
Poisson counts and the expected values are known without
error (StataCorp., 2003). Similarly, Harris and Barraclough
used a Poisson approach to calculate their SMR confidence
intervals.
Aside from the basic SMR estimates, fixed and random effects SMR estimates and their 95% confidence intervals (CIs)
were computed using methods outlined by Sharp and Sterne
(1997) in the STATA statistical software package (version
8.0) (StataCorp., 2003) for each drug and alcohol use disorder strata. The meta command in STATA uses inversevariance weighting to calculate fixed and random effects
pooled summary estimates, confidence limits, a test for true
differences between study effects (as was presented with the
Chi-squared heterogeneity statistic, Q) and the Der Simonian
and Laird (1986) estimate of between studies variance (Sharp
and Sterne, 1997). Fixed effects SMRs are weighted averages across studies under the assumption of homogeneity,
which takes into account within study variance. Random effects SMRs allow for heterogeneity between studies as well
as within study variance. Empirical Bayes estimates were

S13

calculated for each study with nonzero variance. The contribution of larger studies to the summary estimate is greater,
although random effect methods give relatively more weight
to smaller studies, as compared to fixed effects models (Sterne
et al., 2001). Testing for heterogeneity will gauge whether the
differences in results across studies were greater than could
be expected by chance. Discrepancies between random and
fixed effects SMRs indicate between study heterogeneity.
When heterogeneity between studies was found, we used
meta-regression to analyze associations between the SMR
and two covariates, year of study publication and geographical region (coded 1: USA and Canada; 2: Scandinavia; 3:
Europe; 4: Middle East; 5: Africa; 6: Asia). This would help
to assess whether the publication year or geographical area
could have influencing the magnitude of effect across studies.
Random effects meta-regression was conducted in STATA using the metareg command (Sharp, 1998). The metareg command extends a random effects meta-analysis to estimate the
extent to which one or more covariates explain heterogeneity
in the SMRs (Thompson and Sharp, 1999). These two covariates were studied because earlier studies may have provided
different results than studies published later, possibly due to
treatment being more effective or available. Because suicide
rates vary by geographic region, we also thought that this
covariate may also account for heterogeneity in the SMRs.
Data were extracted from reports identified by Harris
and Barraclough to provide the observed number of suicides as well as the expected number of suicides associated
with a given alcohol or drug use disorder. Summary data
provided by Harris and Barraclough were combined with
the newly collected data. Because in some instances sexspecific SMRs were possible and Harris and Barraclough
calculated only overall SMRs, we recalculated the overall
SMRs (without the sex-specific studies) and separately calculated sex-specific SMRs. We extracted data from studies
of both men and women for the overall estimates provided
in the tables, and data from exclusively male or female samples and from studies providing data on suicide stratified by
sex for the sex-stratified estimates provided in the tables.
For example, for alcohol use disorders: 22 studies identified by Harris and Barraclough based on mixed sex samples
yielded 457 suicides and 49.92 expected suicides; 11 studies with data on men yielded 282 suicides and 72.4 expected
suicides; 5 studies with data on women yielded 36 suicides
and 1.95 expected suicides. Direct tests of the equivalence
of the SMRs across sex were conducted using a formula to
determine the statistical difference between rates (NCHS,
1977). The formula takes the difference between the two
SMRs the square root of the width of the females SMR
squared + the width of the males SMR. If the resulting interval
does not include zero, the difference is statistically significant
at the 95% level (NCHS, 1977). We obtained expected values for studies that did not provide estimates by using WHO
statistical reports (WHO, 1961; WHO, 19622002) for the
relevant country and years, sample age/sex composition, and
mean observation period of each report.

S14

H.C. Wilcox et al. / Drug and Alcohol Dependence 76S (2004) S11S19

The overall SMR for alcohol use disorders was 979 (95% CI
8981065; p < 0.001) while the SMR using data from Harris
and Barraclough was 916 (95% CI 8331003; p < 0.001). The
fixed effects SMR for alcohol use disorder combining our
data with data from Harris and Barraclough was 1158 (95%
CI 10591266; p < 0.001). The random effects SMR estimate,
which takes into account the additional variability between
studies, was 1142 (95% CI 8761504; p < 0.001), indicating
heterogeneity between studies (test for heterogeneity Q =
183.0, p < 0.001, moment-based estimate of between studies
variance = 0.364).

3. Results
The obtained SMRs and 95% confidence intervals are presented for alcohol use disorders, opioid use disorders, mixed
intravenous drug use, mixed (general) drug use, and heavy
drinking. Intravenous drug use is a novel category, whereas
the other alcohol and drug categories were previously presented by Harris and Barraclough.
Findings on alcohol use disorders and suicide are provided
in Table 1. Harris and Barraclough reported combined sex
data from 22 studies and we identified 11 additional studies.
Table 1
Alcohol use disorders
Study

Country

Suicides
Observed

Overall (male + female)

Extracted from Harris and Barraclough (1997)
Bellamy et al. (2001)
Bilal et al. (1989)
Combs-Orme et al. (1983)
Finney and Moos (1992)
Gerdner and Berglund (1997)
Goby et al. (1979)
Gual et al. (1999)
Hutchins et al. (2002)
McCabe (1986)
Nicholls et al. (1974)
Seelye (1979)

USA
Kuwait
USA
USA
Sweden
USA
Spain
UK
Scotland
England
USA

Total
Males only
Extracted from Harris and Barraclough (1997)
Allebeck and Allgulander(1990)
Berglund and Tunving (1985)
Berglund (1986)
Combs-Orme et al. (1983)
De Soto et al. (1989)
Gerdner and Berglund (1997)
Liskow et al. (2000)
Mackenzie et al. (1986)
McCabe (1986)
Nicholls et al. (1974)
Ojesjo et al. (1998)
Ornstein and Cherepon (1985)
Rossow and Amundsen (1995)
Rossow et al. (1999)
Tashiro and Lipscomb (1963)

Sweden
Sweden
Sweden
USA
USA
Sweden
USA
USA
Scotland
England
Sweden
USA
Norway
Sweden
USA

Total
Females only
Extracted from Harris and Barraclough (1997)
Combs-Orme et al. (1983)
De Soto et al. (1989)
Gerdner and Berglund (1997)
McCabe (1986)
Nicholls et al. (1974)
Smith and Cloninger (1981)
Tashiro and Lipscomb (1963)
Total

Expected value calculated by us.

Expected value provided in paper.

USA
USA
Sweden
Scotland
England
USA
USA

SMR

95% CI

Expected

457
1
6
10
3
2
1
5
1
0
46
5

49.92
0.20
0
1.88
0.15
0.34
0.03
0.33
0.02
0.08
1.85
0.06

916
500

532
2000
588
3333
1515
5000
0
2487
8333

8331003
132786 *

255978 *
4135844 *
712124 *
8418572 *
4923535 *
12727858 *
04610 *
18203317 *
270619445 *

537

54.86

979

8981065

282
46
16
68
7
1
2
6
4
0
39
3
3
38
43
6

72.4
10.70
2.10
5.60
1.68
0.14
0.29
0.98
0.12
0.07
1.47
0.25
0.62
8.90
10.26
1.31

390
430
762
1214
417
714
690
612
3333
0
2653
1200
484
427
419
458

345438
315573 *
4361237 *
9431539 *
168858 *
183980 *
842490 *
2251332 *
9088533 *
05268 *
18873627 *
2483506 *
1001414 *
302586 *
303565 ***
168997 *

564

116.89

483

444524

36
3
0
0
0
7
2
0

1.95
0.20
0.04
0.05
0.01
0.38
0.13
0.08

1846
1500
0
0
0
1842
1539
0

12932556
3094383 *
09219 *
07375 *
036876 *
7413795 *
1875555 *
04610 *

48

2.84

1690

12462241

H.C. Wilcox et al. / Drug and Alcohol Dependence 76S (2004) S11S19

Harris and Barraclough reported data from five studies of

alcohol use disorders among females and 11 among males.
We identified seven additional studies on alcohol use disorders among females and 15 among males. The SMR for
women using our data as well as data from Harris and Barraclough was 1690 (95% CI 12462241; p < 0.001) which
was higher than the SMR for males (438; 95% CI 444524;
p < 0.001). The corresponding sex-stratified SMRs calculated from data reported in Harris and Barraclough were 1846
(95% CI 12932556; p < 0.001) for females and 390 (95% CI
345438; p < 0.001) for males. The derived fixed effects SMR
associated with alcohol use disorders was also greater for
women (1874; 95% CI 13722559; p < 0.001) than for men
(590; 95% CI 541643; p < 0.001). The random effects SMR
for females was identical to the fixed effects SMR while the
random effects SMR for males differed from the fixed effects
SMR estimate, 665 (95% CI 480922; p < 0.001), indicating heterogeneity between studies for males (Q = 241.6, p <
0.001, between study variance = 0.467). Using meta-analysis
regression (Sharp, 1998), the influence of geographic region
and year of publication on the overall SMR was examined to
find that year was influential (p = 0.039) but region was not.
Neither covariate influenced the magnitude of SMRs across
studies for males (studies of females did not have significant
heterogeneity thus these additional analyses were not conducted).
Findings on opioid use disorders and suicide are presented
in Table 2. Harris and Barraclough identified nine reports in
this category. In the present investigation, three additional
papers were identified, all from England. The overall SMR
for opioid use was 1351 (95% CI 10471715; p < 0.001),
which was similar to that reported by Harris and Barraclough
(1400; 95% CI 10791788; p < 0.001). The corresponding
sex-specific SMR estimates were 357 (95% CI 91990; p >

S15

0.05) for females and 756 (95% CI 4401210; p < 0.001)

for males. The overall SMR was larger than the sex-specific
SMRs because some studies did not report sex-specific rates.
The fixed and random effects SMR for opioid use were approximately 14 times greater than expected (1375; 95% CI
10721763; p < 0.001), and sex-specific reports yielded a
SMR of 796 (95% CI 4701345; p < 0.001) for men. The
fixed and random effects SMR could not be conducted for
women based upon sparse data. The fixed and random effects SMR estimates were identical overall and for studies of
men.
Eight studies conducted in seven countries were identified that reported data on mixed intravenous (IV) drug use
and suicide (see Table 3). With the acceleration of research
on HIV and AIDS in the past decade, there have been several
published studies in this area since the review by Harris and
Barraclough who did not distinguish an IV drug use category,
but had indicated that those classified in their paper with opioid dependence and abuse were all IV drug users. Although
the majority of such individuals in the studies included in the
mixed IV drug use table were presumed or known to have
used opiates, most often heroin, these recent reports on IV
drug use seldom contained data about the nature of addiction (e.g., opioid, cocaine, amphetamine, etc.) among suicide
decedents, and so we have decided to distinguish mixed IV
drug use studies from those providing data on opioid use disorders. Moreover, subjects in studies of IV drug use were typically recruited for investigations of HIV and therefore, may
differ in other respects from subjects recruited for research
on opiates. The overall SMR for IV drug use was 1373 (95%
CI 10291796; p < 0.001). The estimated combined sex fixed
effects SMR for suicide associated with IV drug use was over
18 times greater than expected (1856; 95% CI 13802496; p
< 0.001). The overall random effects SMR was somewhat dif-

Table 2
Opioid drug use
Study

Overall (male + female)

Extracted from Harris and Barraclough (1997)
Oppenheimer et al. (1994)
Wille (1981)

Country

England
England

Total
Males only
Extracted from Harris and Barraclough (1997)
James (1967)
Oppenheimer et al. (1994)

England
England

Total
Females only
Extracted from Harris and Barraclough (1997)
James (1967)
Oppenheimer et al. (1994)
Total

Expected value calculated by us.

England
England

Suicides

SMR

95% CI

Observed

Expected

64
2
1

4.57
0.28
0.11

1400
714
909

10791788
872579*
235065*

67

4.96

1351

10471715

11
5
1

1.61
0.40
0.24

683
1250
417

3411222
4062917*
112322*

17

2.25

756

4401210

0
0
1

0.13
0.11
0.04

0
0
2500

02837
03352*
6313929*

0.28

357

91990

S16

H.C. Wilcox et al. / Drug and Alcohol Dependence 76S (2004) S11S19

Table 3
Mixed intravenous drug users
Study

Overall (male + female)

Eskild et al. (1993)
Eskild et al. (1994)
Frischer et al. (1997)
Fugelstad et al. (1995)
Galli and Musicco (1994)
Semba et al. (1995)
Tyndall et al. (2001)
van Haastrecht et al. (1996)

Country

Suicides

Norway
Norway
Scotland
Sweden
Italy
USA
Canada
Netherlands

Total

SMR

95% CI

Observed

Expected

9
5
2
9
10
1
7
10

0.63
0.11
0.40
0.46
1.41
0.36
0.22
0.27

1429
4546
500
1957
709
278
3182
3704

6532712*
147610606*
611805*
8953714*
3401304*
71548*
12796555*
17766811*

53

3.86

1373

10291796*

Expected value calculated by us.

Table 4
General (mixed) drug use
Study

Country

Overall (male + female)

Extracted from Harris and Barraclough (1997)
Barr et al. (1984)
Benson and Holmberg (1984)
Hurt et al. (1996)
Wahren et al. (1997)

Suicides

USA
Sweden
USA
Sweden

Total
Males only
Allebeck and Allgulander (1990)
Benson and Holmberg (1984)

Sweden
Sweden

Total
Females only
Benson and Holmberg (1984)

Sweden

SMR

95% CI

Observed

Expected

135
0
1
13
77

7.02
0.41
0.80
1.90
3.28

1923
0
125
684
2348

16122276
0899*
3696***
3641170***
18532934*

226

13.41

1685

14731920

27
0

7.50
0.70

360
0

237524*
0527***

27

8.20

329

217479

0.10

1000

255572***

Expected value calculated by us.

Expected value provided in paper.

ferent than the fixed effects SMR (1771; 95% CI 10133099;

p < 0.001), indicating significant heterogeneity between studies (Q = 21.5, p = 0.004, between studies variance = 0.396).
Using meta-analysis regression (Sharp, 1998), the influence
of geographic region and year of publication on the overall

SMR was examined. Neither region nor year of publication

was influential in accounting for variability in the overall
SMR.
Harris and Barraclough identified four studies on mixed
drug use and suicide and obtained an estimated SMR of 1923

Table 5
Heavy alcohol use
Study

Country

Suicides

SMR

95% CI

Observed

Expected

Overall (male + female)

Extracted from Harris and Barraclough (1997)
Klatsky et al. (1981)
USA
Skurtveit et al. (2002)
Norway

15
12
13

2.93
4.16
4.31

512
289
302

287844
149504*
161516*

Total

40

11.40

351

251478

1
1

0.71
4.46

141
22

4785*
1125*

5.17

39

5140

Males only
Kristenson et al. (1983)
Kristenson et al. (2002)
Total

Expected value calculated by us.

Sweden
Sweden

H.C. Wilcox et al. / Drug and Alcohol Dependence 76S (2004) S11S19

(95% CI 16122276; p < 0.001) (Table 4). Four additional

studies in this category conducted in the U.S. and Sweden
were identified. The SMR for mixed drug use was 1685 overall (95% CI 14731920; p < 0.001), 1000 for women (95%
CI 255572; p > 0.05), and 329 for men (95% CI 217479; p
< 0.001). Combining the study estimates yielded an overall
fixed effects SMR approximately 20 times greater than that
expected (2018; 95% CI 17622312; p < 0.001) and random
effects SMR approximately 17 times greater than expected
(1745; 95% CI 11402672; p < 0.001), indicating substantial
heterogeneity between studies (Q = 40.7, p < 0.001, between
study variance = 0.231). There were not adequate data for
meaningful fixed and random effects comparisons between
sexes. When the influence of geographic region and year of
publication on the overall SMR were examined, neither region nor year of publication was influential in accounting for
variability.
Harris and Barraclough identified one study of heavy
drinkers and four studies in three countries were newly identified that reported on heavy drinking, two of which were
for males only (Table 5). The overall SMR for heavy drinking was 351 (95% CI 251478; p < 0.001), while it was 39
(95% CI 5140; p > 0.05) for males. The overall fixed effects
estimate yielded a SMR approximately three to four times
greater than that expected (364; 95% CI 261507; p < 0.001)
while the fixed effects SMR for men should be interpreted
with caution in light of meager data (SMR = 52; 95% CI
9307). The overall and male random effect estimates were
not significantly different than the pooled fixed effects estimates.
One study from Sweden on amphetamine abusers
(Tunving, 1988) was obtained yielding a combined SMR of
1136 (95% CI 3692652). Additional studies that met eligibility criteria and reported on suicide and sedative hypnoticand cannabis-use disorders were not identified. No studies
on cocaine or hallucinogen use disorders and suicide were
identified that met inclusionary criteria.

4. Discussion
Twenty studies not previously identified by Harris and
Barraclough and 22 studies published since their review were
combined with their data. The additional studies contributed
to a more robust estimate of the association of suicide with
opioid use disorders overall, alcohol use disorders among
women, as well as novel summary data on the association
of IV drug use and suicide. Alcohol and drug use disorders were associated with suicide for all categories of disorders for which data were available. There remains meager
data using cohort designs on the association of many drug
use disorders and suicide including cocaine and cannabis
use disorders. Moreover, with the exception of alcohol use
disorders, there were meager data from cohort studies on
the association of suicide and drug use disorders among
women.

S17

Studies of alcohol use disorders (for both sexes and males

only), as well as studies of IV drug users and mixed drug
users showed substantial between study heterogeneity. Heterogeneity between studies could be due to variation in study
samples, variation in study methods or other known or unknown factors. We studied the influence of year of publication
and geographic region on variability in the SMRs. The year
of publication was influential on the variability of the overall SMRs for studies of alcohol use disorders, implying that
earlier studies produced different results than the later studies. However, due to limited data for many drug disorders,
such patterns warrant further examination. The reason(s) for
the slightly dissimilar findings between our estimates and
those of Harris and Barraclough are not clear and, with the
exception of studies on alcohol use disorders, there was limited available evidence from which to draw firm conclusions,
warranting caution in interpretation.
The major limitation of this study was the potential for systematic bias in the estimation of the SMR estimates. Three
aspects of the data may have biased the estimates of the SMR
estimates upward. First, studies were not included when the
number of suicides (or lack thereof) was not explicitly mentioned. It seems likely that some mortality studies without any
suicides would fail to note their absence, and the exclusion
of such reports may have resulted in an under-representation
of nil findings on drug use and suicide. Second, most reports were based on cohorts of treated samples which could
inflate the estimates given greater severity in clinical samples. Third, the analyses accounted for age and sex but did
not account for other potential confounding variables associated with both drug use disorders and suicide (e.g., depression, socioeconomic status). There was also the potential for
downward biasing given the likelihood of misclassifying suicides as non-intentional overdoses, particularly in studies of
IV drug users and opiate use disorders in which overdose
deaths are common. Indeed suicide intent is more difficult to
determine in overdose deaths compared to other methods of
suicide (e.g., hanging, firearm), and overdose was a leading
cause of death in studies of opioid use disorders and IV drug
use. The search strategy of Harris and Barraclough was followed and thus we did not extend our search to databases other
than MEDLINE. The use of other appropriate databases may
have resulted in the identification of more eligible studies.
Other limitations include limited data on women and that the
majority of investigations were from industrialized countries
in Europe and North America, with unclear generalizability
to non-western or developing countries.
The findings indicate that individuals with opioid use disorders and mixed intravenous drug use bear an elevated risk
for suicide and suggest that the risk is somewhat greater
than that associated with alcoholism. Suicide is a multidetermined outcome. It will be important to discern what role
opiates, and other drugs of abuse, play in the generation of
suicidal states. What are the direct effects of their neuropharmacological actions? How much risk is due to the lifestyles
of illicit drug users? What are the interactions between alco-

S18

H.C. Wilcox et al. / Drug and Alcohol Dependence 76S (2004) S11S19

hol and drug use disorders and other co-morbid psychiatric

conditions, such as depression or bipolar disorder? And to
what extent do pre-morbid characteristics (e.g., impulsivity) play a role in both alcohol and drug use and suicide?
The data from this review support Harris and Barracloughs conclusion that the association of alcohol use disorders and suicide was more pronounced in women than in
men. We are aware of only one psychological autopsy investigation that has addressed this issue directly (Conner et
al., 2003a,b). In analyses of case-control postmortem suicide
data gathered in New Zealand, the investigators concluded
that sex did not moderate the association between alcohol
dependence and suicide (Conner et al., 2003a) and indeed, as
in the general population, male sex conferred greater risk for
suicide in a sub-investigation limited to alcoholics (Conner
et al., 2003b). Conner et al. (2003a,b) controlled for depression, which is especially prevalent among female alcoholics,
potentially reducing the association of alcohol dependence
and suicide more so in women than in men. As well, in our
empirical review, observed suicides among women with alcoholism were compared with the population rates among
women, which were uniformly lower than men. Therefore,
the same number of observed suicides among women and
men would yield a greater SMR in women given their lower
base rate of suicide. Nonetheless, it is difficult to dismiss
the greatly elevated SMR associated with alcohol use disorders in women compared to men purely on methodological
grounds. There was not adequate data on women to discuss
potential sex differences in the association of other drug use
disorders and suicide. Clearly, more data on suicide and drug
use disorders in women is needed in order to make accurate estimates of the magnitude of risk among women, and
to prioritize prevention and intervention efforts accordingly.
It is important to consider these findings in the larger context of the epidemiology of suicide and approaches to suicide
prevention. Most prevention programs have relied upon detecting those at increased or imminent risk, for whom hospitalization, crisis intervention efforts or psychotherapeutic interventions are prescribed. While such approaches often are
required by the urgency of clinical situations that confront
care providers, they have not sufficed as effective prevention
efforts associated with lowering rates of suicide (Knox et al.,
2004). One possible explanation may arise from the focus of
such programs on tertiary care of mental disorders such as alcoholism or depression. A focus on the primary prevention of
alcohol and drug use disorders and other psychopathological
disorders associated with suicide, as well as intervention for
those showing early indications of such disorders, are needed
in order to have a meaningful impact on the population rate
of suicide.

Acknowledgments
We would like to acknowledge the landmark empirical review by Harris and Barraclough (1997) that provided the

foundation for this study and the assistance of Liz Schifano in gathering the data. This work was supported by NIH
grants T32 MH019833, F31 DA14454, R03 AA13300, K23
AA00318, and R13 MH62073.

References
Allebeck, P., Allgulander, C., 1990. Suicide among young men: psychiatric illness, deviant behavior and substance abuse. Acta Psychiatry
Scand. 81, 565570.
Barr, H.L., Antes, D., Ottenberg, D.J., Rosen, A., 1984. Mortality of
treated alcoholics and drug addicts: the benefits of abstinence. J. Stud.
Alcohol. 45 (5), 440452.
Bellamy, C.O., DiMartini, A.M., Ruppert, K., Jain, A., Dodson, F., Torbenson, M., Starzl, T.E., Fung, J.J., Demetris, A.J., 2001. Liver transplantation for alcoholic cirrhosis: long-term follow-up and impact of
disease recurrence. Transplantation 72 (4), 619626.
Benson, G., Holmberg, M.B., 1984. Drug-related mortality in young people. Acta Psychiatry Scand. 70 (6), 525534.
Berglund, M., 1986. Suicide in male alcoholics with peptic ulcers. Alcohol. Clin. Exp. Res. 10 (6), 631634.
Berglund, M., Tunving, K., 1985. Assaultive alcoholics 20 years later.
Acta Psychiatry Scand. 71 (2), 141147.
Bilal, A.M., Al-Ansari, E.A., Kristof, J., Shaltout, A., El-Islam, M.F.,
1989. Prospective study of alcoholism treatment in Kuwait: a 5-year
follow-up report. Drug Alcohol Depend. 23 (1), 8386.
Cavanagh, J.T., Carson, A.J., Sharpe, M., Lawrie, S.M., 2003. Psychological autopsy studies of suicide: a systematic review. Psychol. Med.
33 (3), 395405.
Combs-Orme, T., Taylor, J.R., Scott, E.B., Holmes, S.J., 1983. Violent
deaths among alcoholics. A descriptive study. J. Stud. Alcohol. 44
(6), 938949.
Conner, K.R., Beautrais, A.L., Conwell, Y., 2003a. Moderators of the
relationship between alcohol dependence and suicide and medically
serious suicide attempts: analyses of canterbury suicide project data.
Alcohol. Clin. Exp. Res. 27, 11561161.
Conner, K.R., Beautrais, A.L., Conwell, Y., 2003b. Risk factors for suicide and medically serious suicide attempts among alcoholics: analyses of canterbury suicide project data. J. Stud. Alcohol 64, 551
554.
Der Simonian, R., Laird, N.M., 1986. Meta-analysis in clinical trials.
Control. Clin. Trials 7, 177188.
De Soto, C.B., ODonnell, W.E., De Soto, J.L., 1989. Long-term recovery
in alcoholics. Alcohol Clin. Exp. Res. 13 (5), 693697.
Eskild, A., Magnus, P., Sohlberg, C., Kittelsen, P., Olving, J.H., Teige, B.,
Skullerud, K., 1994. Slow progression to AIDS in intravenous drug
users infected with HIV in Norway. J. Epidemiol. Commun. Health
48 (4), 383387.
Eskild, A., Magnus, P., Samuelsen, S.O., Sohlberg, C., Kittelsen, P., 1993.
Differences in mortality rates and causes of death between HIV positive and HIV negative intravenous drug users. Int. J. Epidemiol. 22
(2), 315320.
Finney, J.W., Moos, R.H., 1992. The long-term course of treated alcoholism. II. Predictors and correlates of 10-year functioning and mortality. J. Stud. Alcohol. 53 (2), 142153.
Frischer, M., Goldberg, D., Rahman, M., Berney, L., 1997. Mortality and
survival among a cohort of drug injectors in Glasgow, 19821994.
Addiction 92 (4), 419427.
Fugelstad, A., Rajs, J., Bottiger, M., Gerhardsson de Verdier, M., 1995.
Mortality among HIV-infected intravenous drug addicts in Stockholm
in relation to methadone treatment. Addiction 90 (5), 711716.
Galli, M., Musicco, M., 1994. Mortality of intravenous drug users living
in Milan. Italy: role of HIV-1 infection. COMCAT study group. AIDS
8 (10), 14571463.

H.C. Wilcox et al. / Drug and Alcohol Dependence 76S (2004) S11S19
Gerdner, A., Berglund, M., 1997. Mortality of treated alcoholics after
eight years in relation to short-term outcome. Alcohol Alcohol. 32
(5), 573579.
Goby, M.J., Bradley, N.J., Bespalec, D.A., 1979. Physicians treated for alcoholism: a follow-up study. Alcohol. Clin. Exp. Res. 3 (2), 121124.
Gordis, L., 1996. Epidemiology. W.B. Saunders Company, Philadelphia.
Gual, A., Lligona, A., Colom, J., 1999. Five-year outcome in alcohol
dependence. A naturalistic study of 850 patients in Catalonia. Alcohol
Alcohol. 34 (2), 183192.
Harris, E.C., Barraclough, B., 1997. Suicide as an outcome for mental
disorders. A meta-analysis. Br. J. Psychiatry 170, 205228.
Hurt, R.D., Offord, K.P., Croghan, I.T., Gomez-Dahl, L., Kottke, T.E.,
Morse, R.M., Melton 3rd., L.J., 1996. Mortality following inpatient
addictions treatment. Role of tobacco use in a community-based cohort. JAMA 275 (14), 10971103.
Hutchins, R.R., Hart, R.S., Pacifico, M., Bradley, N.J., Williamson, R.C.,
2002. Long-term results of distal pancreatectomy for chronic pancreatitis in 90 patients. Ann. Surg. 236 (5), 612618.
James, I.P., 1967. Suicide and mortality amongst heroin addicts in Britian.
Br. J. Addict. 62, 391398.
Klatsky, A.L., Friedman, G.D., Siegelaub, A.B., 1981. Alcohol and mortality. A 10-year KaiserPermanente experience. Ann. Intern. Med. 95
(2), 139145.
Knox, K.L., Conwell, Y., Caine, E.D., 2004. If suicide is a public health
problem, what are we doing to prevent it? Am. J. Public Health 94,
3745.
Kristenson, H., Osterling, A., Nilsson, J.A., Lindgarde, F., 2002. Prevention of alcohol-related deaths in middle-aged heavy drinkers. Alcohol.
Clin. Exp. Res. 26 (4), 478484.
Kristenson, H., Ohlin, H., Hulten-Nosslin, M.B., Trell, E., Hood, B.,
1983. Identification and intervention of heavy drinking in middleaged men: results and follow-up of 2460 months of long-term study
with randomized controls. Alcohol. Clin. Exp. Res. 7 (2), 203209.
Lesage, A.D., Boyer, R., Grunberg, F., Vanier, C., Morisette, R., MenardButeau, C., et al., 1994. Suicide and mental disorders: a case-control
study of young men. Am. J. Psychiatry 151, 10631068.
Liskow, B.I., Powell, B.J., Penick, E.C., Nickel, E.J., Wallace, D., Landon,
J.F., Campbell, J., Cantrell, P.J., 2000. Mortality in male alcoholics
after 1014 years. J. Stud. Alcohol. 61 (6), 853861.
Mackenzie, A., Allen, R.P., Funderburk, F.R., 1986. Mortality and illness
in male alcoholics: an 8-year follow-up. Int. J. Addict. 21 (8), 6582.
McCabe, R.J., 1986. Alcohol-dependent individuals 16 years on. Alcohol
Alcohol. 21 (1), 8591.
Murphy, G.E., 1998. Why women are less likely than men to commit
suicide. Compr. Psychiatry 39 (4), 165175.
National Center for Health Statistics, 1977. Mortality. In: Kleinman, J.C.
(Ed.), Statistical Notes for Health Planners, No. 3. Health Resources
Administration, Washington, DC.
Nicholls, P., Edwards, G., Kyle, E., 1974. Alcoholics admitted to four
hospitals in England. II. General and cause-specific mortality. Q. J.
Stud. Alcohol 35 (3), 841855.
Ojesjo, L., Hagnell, O., Otterbeck, L., 1998. Mortality in alcoholism
among men in the Lundby Community Cohort. Sweden: a 40-year
follow-up. J. Stud. Alcohol 59 (2), 140145.

S19

Oppenheimer, E., Tobutt, C., Taylor, C., Andrew, T., 1994. Death and
survival in a cohort of heroin addicts from London clinics: a 22-year
follow-up study. Addiction 89 (10), 12991308.
Ornstein, P., Cherepon, J.A., 1985. Demographic variables as predictors of alcoholism treatment outcome. J. Stud. Alcohol 46 (5), 425
432.
Rossow, I., Amundsen, A., 1995. Alcohol abuse and suicide: a 40year prospective study of Norwegian conscripts. Addiction 90, 685
691.
Rossow, I., Romelsjo, A., Leifman, H., 1999. Alcohol abuse and suicidal
behaviour in young and middle aged men: differentiating between
attempted and completed suicide. Addiction 94 (8), 11991207.
Seelye, E.E., 1979. Relationship of socioeconomic status, psychiatric diagnosis and sex to outcome of alcoholism treatment. J. Stud. Alcohol
40 (1), 5762.
Semba, R.D., Caiaffa, W.T., Graham, N.M., Cohn, S., Vlahov, D., 1995.
Vitamin A deficiency and wasting as predictors of mortality in human
immunodeficiency virus-infected injection drug users. J. Infect. Dis.
171 (5), 11961202.
Sharp, S., Sterne, J., 1997. Meta-analysis. Stata Tech. Bull. 38, 914.
Sharp, S., 1998. Meta-analysis regression. Stata Tech. Bull. 42, 1624.
Skurtveit, S., Christophersen, A.S., Grung, M., Morland, J., 2002. Increased mortality among previously apprehended drunken and drugged
drivers. Drug Alcohol Depend. 68 (2), 143150.
Smith, E.M., Cloninger, C.R., 1981. Alcoholic Females: mortality at 12year followup. Focus Women 2, 113.
StataCorp., 2003. Stata Statistical Software: release 8.0. StataCorp. LP,
College Station, TX.
Sterne, J., Bradburn, M.J., Egger, A., 2001. Meta-analysis in Stata. Syst.
Rev. 18, 347369.
Tashiro, M., Lipscomb, W.R., 1963. Mortality experience of alcoholics.
Q. J. Stud. Alcohol. 24, 203212.
Thompson, S.G., Sharp, S.J., 1999. Explaining heterogeneity in metaanalysis: a comparison of methods. Stat. Med. 18 (20), 26932708.
Tyndall, M.W., Craib, K.J., Currie, S., Li, K., OShaughnessy, M.V.,
Schechter, M.T., 2001. Impact of HIV infection on mortality in a
cohort of injection drug users. J. Acquir. Immune Defic. Syndr. 28
(4), 351357.
van Haastrecht, H.J., van Ameijden, E.J., van den Hoek, J.A., Mientjes, G.H., Bax, J.S., Coutinho, R.A., 1996. Predictors of mortality
in the Amsterdam cohort of human immunodeficiency virus (HIV)positive and HIV-negative drug users. Am. J. Epidemiol. 143 (4), 380
391.
Wahren, C.A., Allebeck, P., Rajs, J., 1997. Unnatural causes of death
among drug addicts in Stockholm: an analysis of health care and
autopsy records. Subst. Use Misuse 32 (14), 21632183.
Wille, R., 1981. Ten-year follow-up of a representative sample of London heroin addicts: clinic attendance, abstinence and mortality. Br. J.
Addict. 76 (3), 259266.
World Health Organization, years up to 1961. Annual Epidemiological
and Vital Statistics. WHO: Geneva.
WHO, 1962. World Health Organization, 1962 and years following. World
Health Statistics Annual. WHO: Geneva.