OUP user menu


Margda Waern
DOI: http://dx.doi.org/10.1093/alcalc/agg060 249-254 First published online: 1 May 2003


Aims: To assess suicide risk associated with alcohol use disorder in elderly men and women, and to examine the role of social stressors in elderly suicides with and without alcohol use disorders. Methods: This retrospective case–control study included 85 suicide cases aged 65 years and above (46 men, 39 women) and 153 randomly selected population controls (84 men, 69 women). Interviews were carried out with control persons and with informants for the suicide cases. Mental disorders were diagnosed in accordance to DSM-IV. Results: A history of alcohol dependence or misuse was observed in 35% of the elderly men who died by suicide and in 18% of the women. This disorder was uncommon among persons in the control group (2% of the men and 1% of the women). Alcohol use disorder remained an independent predictor of suicide risk in the regression models for both sexes. Among suicide cases, those with alcohol use disorders were younger and less likely to be suffering from severe physical illness (35 vs 63%) than those without this disorder. Conclusion: Alcohol use disorder is associated with suicide in elderly men and women. Prevention programmes need to target this important subgroup.


The lifetime risk of suicide in persons with alcoholism has recently been estimated at 7%, not unlike the risk associated with depression (Inskip et al., 1998). Retrospective studies of young (Runeson, 1990) and mixed age (Murphy, 2000) suicides demonstrated that 25–55% suffer from substance dependence or misuse. There is some evidence that the importance of alcohol use disorder as an antecedent to suicide diminishes with increasing age (Rich et al., 1986; Conwell and Brent, 1995). However, studies that focused specifically on the elderly report a wide range of figures. As shown in Table 1, proportions as low as 3% (Barraclough, 1971), and as high as 44% (Conwell et al., 1991) have been reported. Some of the discrepancy may be explained by methodological differences. While all of the studies cited in Table 1 were based on biographical data derived from informant interviews, a variety of diagnostic criteria were employed. Also, routines regarding forensic examination and cause of death classification varied in different settings. Studies that deal exclusively with certain suicides may systematically miss substance users, who tend to be over-represented among uncertain suicides. Also, the age and sex distribution of the cases and the prevalence rate of alcohol use disorder in the underlying population will affect results. Most psychological autopsy studies lack controls and cannot generate risk data.

View this table:
Table 1.

Retrospective studies of late life suicide: proportions with selected mental disorders

Authors (year)CountryAge group (years)nMenSubstance use disorderDepressionAny mental disorderDiagnostic system
aIncludes only cases with recent primary care contact.
Barraclough (1971) UK65+303% chronic alcoholism87% affective disorder87%Clinical practice
Conwell et al.(1991) USA50+1883%44% substance use disorder67% major depression89%DSM-III-R
Clark and Clark (1993) USA65+71%22% alcoholism83% depression‘all but 2’Research diagnostic criteria
Carney et al.(1994) USA60+4959%8% ‘pure substance use’, 14% substance use + depression37% affective disorder86%DSM-III
Henriksson et al.(1995) Finland60+4379%25% alcohol dependence/misuse, 5% anxiolytic dependence/misuse67% depressive disorder, 44% major depression91%DSM-III-R
Conwell et al.(1996) USA55+5074%42% alcohol use disorder, 4% drug use disorder66% mood disorder, 50% major depression86%DSM-III-R
Conwell et al.(2000) USA60+42a71%14% substance use disorder76% mood syndrome, 62% major depression79%DSM-III-R
Harwood et al.(2001) UK60+1005% alcohol misuse, 5% other substance misuse4% bipolar-depressive, 59% depressive episodes77%ICD-10
Waern et al.(2002) Sweden65+8554%27% substance use disorder82% mood disorder, 46% major depression96%DSM-IV

The aim of the current study was therefore to examine the suicide risk associated with alcohol use disorder in elderly men and women. We used a control group representative of the general population (65+ years). A second aim was to examine the role of social stressors in suicides with and without alcohol use disorders.


Ascertainment of suicide cases

One hundred consecutive cases of suicide in elderly (65+ years) Scandinavian-born persons who underwent necropsy at the Göteborg Institute of Forensic Medicine were examined. There were 82 cases of certain suicide [International Classification of Disease (ICD-9) E950–959] (World Health Organization, 1977). The circumstances of death strongly suggested suicide in the remaining 18 cases, but there had been no expression of suicide intent and the cause of death was classified as undetermined (ICD-9 E980–989). The catchment area included the city of Göteborg and two adjacent counties (elderly population 210 703 at the start of the study). The forensic cases comprised 90% of all suicides (n = 111 registered among Scandinavian-born elderly during the study period (January 1994 to May 1996).

The next-of-kin for 98 of the 100 potential study cases were identified and informed about the study. Informants for 85 suicides (46 men and 39 women) agreed to participate. As reported earlier (Waern et al., 2002), suicides with an informant interview did not differ from the total suicides in terms of mean age, proportion of women, proportion of certain suicides, and proportion with a positive screening for antidepressants and/or lithium at necropsy.

The comparison group

Two persons living in the same area of residence and with the same sex and birth year (±2 years) as the person who committed suicide were randomly chosen from the tax roster. Potential control persons received a letter of information about the study and were then contacted by telephone. When an individual declined participation, a new person was invited to take part in the study (a maximum of eight per case). In all, 240 persons were invited to take part in the study and 153 (64%, 84 men, 69 women) accepted. Reasons for declining included poor health (n = 13), social reasons (n = 8) and lack of interest (n = 60).

The interview

The semi-structured interview included questions about the subject’s social situation, life events, past and current mental and physical health, history of suicidal behaviour, use of alcohol and illicit drugs, contacts with in- and out-patient services, and use of prescription drugs. The interview included past month psychiatric signs and symptoms derived from the Comprehensive Psychiatric Rating Scale (Åsberg et al., 1978) and questions about dementia symptoms (Skoog et al., 1993). Past year life events were rated according to a revised version of the Recent Life Change Questionnaire (Paykel et al., 1969), modified by Heikkinen et al.(1994). All life events were self-reported for the controls and informant-reported for the suicides.

The author, a psychiatrist, performed all informant interviews for the suicide cases. Interviews with control persons were carried out by a geriatrician, a psychiatric nurse or a psychiatric occupational therapist, all with long clinical and interview experience. In order to reduce the risk of non-participation due to poor health, proxy interviews with informants were carried out for 11 control persons who could not participate actively due to serious illness.

Record reviews

Case notes from psychiatric in- and out-patient clinics and primary care facilities were reviewed for the suicide cases. Records from other disciplines (e.g. internal medicine, geriatrics, oncology) were obtained when deemed relevant on the basis of the informant interview. Individuals in the control group were asked about contacts with health care services and their case records were requested on the basis of this information. Seven control persons did not agree to the release of their records.

Forensic reports were scrutinized in order to assess alcohol use in connection with the suicide and type of suicide method. Post-mortem analysis of alcohol in blood and urine specimens was carried out at the Institute of Forensic Chemistry in Linköping for 82 of the 85 suicide cases. Suicide method was determined by the forensic examiner. For the purpose of this report, drug overdose, drowning and carbon monoxide poisoning were classified as non-violent methods (Conwell et al., 1998). Hanging, shooting, jumping and collision with a train were classified as violent methods.

Diagnosing mental disorders

The author used data from interviews, case records and forensic reports (suicide cases only) to make retrospective diagnoses according to the diagnostic algorithms of the DSM-IV Axis I (American Psychiatric Association, 1994). Cases with symptom constellations that did not fit the diagnostic algorithms were assigned ‘best estimate’ diagnoses after discussion with a senior psychiatrist. A more detailed description of diagnostic procedures is given in our recent report (Waern et al., 2002).

Statistical analysis

In the univariate analyses, odds ratios (ORs) for suicide among persons with a disorder compared to those without the disorder were calculated with logistic regression. Due to a lag between the suicide deaths and the control interviews, the controls were somewhat older than the suicide cases at the time of the interview (mean age ± SD in suicide victims 75.0 ± 7.7 years, in controls 78.8 ± 7.7 years). Therefore, all ORs were adjusted for age. Variables that were included in the multivariate regression models (enter) for men and women included major depression, alcohol dependence/misuse (ADM), family conflict and serious somatic illness. For the purpose of this study, serious somatic illness was defined as a rating of 3 or 4 in any somatic category in accordance with the Cumulative Illness Rating Scale for Geriatrics (CIRS-G) (Miller et al., 1992).

Fisher’s exact test was employed to examine differences in proportions between suicide subgroups with and without substance misuse disorder. The t-test was used to compare continuous variables. Two-sided P-values were applied (P < 0.05). All exploratory and formal statistical analyses were performed with SPSS version 10.1 for Windows.


Informed consent was obtained from the informants and the control persons after they had received oral and written information about the study, including an assurance that they could withdraw from the study at any time. A close relative gave proxy consent for control persons suffering from dementia. The Research Ethics Committee at Göteborg University approved the study.


The ORs for suicide for the five most common DSM-IV diagnoses are shown separately for men and women in Table 2. A lifetime history of alcohol use disorder was observed in 35% of the male suicide cases and in 2% of the individuals in the comparison group. Such a history was thus a strong predictor of suicide risk in men. A history of alcohol use disorder was observed in 18% of the female suicides. For comparison, only one of the 69 women in the control group had this disorder. The estimated risk of suicide in women with alcohol use disorder was only half that observed in men, but the association reached significance [OR = 9.5; 95% confidence interval (CI) 1.1–84.2]. When the sexes were combined, a lifetime history of alcohol use disorder was observed in 23 out of 85 of the suicide cases (27%) and in three of the 153 individuals in the control group (2%). Alcohol use disorder was thus a strong predictor of suicide risk (OR = 14.5; 95% CI 4.1–50.7; P = 0.000).

View this table:
Table 2.

Past month mental disorder in elderly suicide victims and controls:a age-adjusted odds ratios (ORs) and 95% confidence intervalsb

Men Women
Diagnosis (DSM-IV)Cases (n = 46) n (%)Controls (n = 84) n (%)ORPCases (n = 39) n (%)Controls (n = 69) n (%)ORP
aMultiple diagnoses allowed.
bORs for suicide in persons with a disorder compared with all others.
cORs could not be estimated. P-values in accordance with Fisher’s exact test.
dLifetime history.
Major depression21 (45.7)2 (2.4)30.2 (6.5–139.3)0.018 (46.2)2 (2.9)31.4 (6.1–160.9)0.0
Minor depression8 (17.4)3 (3.6)8.2 (1.8–36.0)0.0067 (17.9)0c0.001
Alcohol use disorderd16 (34.8)2 (2.4)18.4 (3.9–86.2)0.07 (17.9)1 (1.4)9.5 (1.1–84.2)0.04
Anxiety disorder4 (8.7)3 (3.6)1.9 (0.4–9.0)0.49 (23.1)3 (4.3)5.8 (1.4–24.2)0.02
Dementia3 (6.5)5 (6.0)1.5 (0.3–7.0)0.609 (13.0)c0.02

Suicides with and without alcohol use disorder

Cases with alcohol use disorder were younger at the time of suicide than those without this disorder (mean age 71.6 years as compared with 76.2 years in those without alcohol use disorder, mean difference 4.7; 95% CI of the difference, 1.1–8.3; P = 0.01). Table 3 shows the proportions with stressful life events during the past year for suicides with and without alcohol use disorder. There were no differences in proportions between the two subgroups. Separations were rare in both groups and financial difficulties were uncommon. Half of those with alcohol use disorder had experienced family conflict during the year that preceded the suicide. Such conflicts were common also among those without alcohol problems.

View this table:
Table 3.

Past year life events in elderly suicides with and without alcohol use disordera

Alcohol use disorder
ParameterYes (n = 23)n (%)No (n = 62)n (%)
aThere were no differences in proportions between subgroups with and without alcohol use disorder.
bInformation missing for one person.
cIncludes death of spouse, family member or close friend.
dInformation missing for two persons.
Illness in family7 (30.4)19 (31.1)b
Deathc1 (4.3)11 (18.3)d
Separation1 (4.3)2 (3.3)d
Family discord12 (52.2)26 (42.6)b
Residence change2 (8.7)5 (8.2)b
Financial problems3 (14.3)d8 (13.3)c
Criminal offence1 (4.3)0b
Somatic illness18 (78.3)49 (79.0)
Any event21 (91.3)56 (90.3)

As shown in Table 4, half of the cases with a lifetime history of alcohol use disorder had made at least one suicide attempt. Persons with alcohol use disorder were more likely to use non-violent methods in connection with the final act. As expected, they were more likely to have a positive test for alcohol at post-mortem analysis. Serious somatic illness, as defined by the CIRS-G, was less common among suicides with alcohol use disorder, which is probably a reflection of the age difference noted above. There were no significant differences between subgroups concerning DSM-IV Axis I mental disorders. The proportions with affective illness were strikingly similar in cases both with and without alcohol use disorder. Table 4 shows further that half of the suicides with comorbid alcohol use disorder and depression had received antidepressant treatment during their final year of life. Only one of the suicide cases had been treated with disulfuram during this time period and none had contact with a specialist clinic for alcohol dependence.

View this table:
Table 4.

Forensic and diagnostic characteristics of elderly suicides (65 years of age and over) with and without alcohol use disorder: past month diagnoses according to DSM-IV

Alcohol use disordera
ParameterYes (n = 23)n (%)No (n = 62)n (%)Significance(P)
aLifetime history of alcohol use disorder.
bInformation missing for one person with alcohol use disorder and two persons without this disorder.
cRating of 3 or 4 in any organ system in accordance with the Cumulative Illness Rating Scale (Miller et al., 1992).
dProportion of those with any affective disorder (alcohol use disorder, n = 19; all others, n = 51).
Forensic data
    History of suicide attempt12 (52.2)25 (40.3)0.3
    Suicide note5 (21.7)20 (32.3)0.4
    Non-violent suicide method16 (69.6)26 (41.9)0.03
    Positive test for alcohol post-mortemb14 (63.6)11 (18.3)0.0
Somatic health
    Serious somatic illness/disabilityc8 (34.8)39 (62.9)0.03
Mental health
    Any affective disorder19 (82.6)51 (82.3)1.0
    Bipolar disorder2 (8.7)2 (3.2)0.3
    Major depression, single episode2 (8.7)10 (16.1)0.5
    Major depression, recurrent7 (30.4)20 (32.3)1.0
    Minor depression4 (17.4)11 (17.7)1.0
    Any anxiety disorder5 (21.7)8 (12.9)0.3
    Any psychotic disorder3 (13.0)4 (6.5)0.4
    Dementia03 (4.8)0.6
Treatment for affective illness
    Past year10 (52.6)d31 (60.8)d0.6

Regression models

Multivariate analyses were carried out in order to determine whether alcohol use disorder independently predicted suicide in men and women. For men, major depression (OR = 15.8; 95% CI 2.7–93.4; P = 0.002), alcohol use disorder (OR = 8.2; 95% CI 1.01–65.7; P = 0.049), family discord (OR = 27.8; 95% CI 3.1–247.6; P = 0.003) and operationally defined serious physical illness (OR = 6.5; 95% CI 1.8–23.2; P = 0.004) all remained in the final model. The CI were large, however. Major depression (OR = 40.6; 95% CI 7.3–225.1; P = 0.000), alcohol use disorder (OR = 13.3; 95% CI 1.2–149.7; P = 0.04), and family discord (OR = 9.3; 95% CI 2.4–35.5; P = 0.001) remained significant factors in women.


Methodological considerations

To our knowledge, this is the first retrospective study of consecutive suicide victims to examine suicide risk in elderly men and women with alcohol problems. A major strength is the case–control design, which permits an estimation of the strength of the association between alcohol use disorder and suicide. However, some limitations need to be addressed. First, due to the proxy nature of the suicide interviews, we found it difficult to quantify current alcohol use. We did not attempt to distinguish between misuse and dependence.

Secondly, we have no data regarding alcohol use among potential control persons who declined participation. It is likely that persons with alcohol problems would decline participation more often than those without. We note, however, that the 1 month prevalence rate of alcohol use disorder in our control group (0.7%) was similar to that reported for the 65+ population in the Epidemiological Catchment Area Study (Regier et al., 1984). Still, the prevalence of alcohol use disorder may have been underestimated in our study. Seven control persons did not consent to the release of their case records, and one reason for this might be unwillingness to disclose a diagnosis of alcoholism.

Thirdly, the rater was not blind to case–control status. This problem was alleviated somewhat by the use of case notes for both suicides and controls. These notes were recorded by clinicians before the suicides occurred. Clinicians had identified alcohol use disorder and recorded this in the case notes of all eight women who received a study diagnosis of alcohol use disorder. This was also the case in 14 of the 18 men with such a study diagnosis.

Fourthly, while control persons were interviewed in person, data for the suicide cases were collected from informants. There is an inherent risk that informants may under-report psychiatric symptoms due to memory bias and guilt feelings. Recent research, however, speaks for the validity of proxy-based data in suicide research (Conner et al., 2001). Agreement was substantial for alcohol dependence in that study, but poorer for misuse. Several of the informants in the current study appeared uncomfortable with questions related to the decedents’ alcohol intake. This, taken together with the observation that the DSM-IV criteria for alcohol use disorder are not particularly suited to the elderly (Hocking et al., 1995), suggests that the proportion of suicides with a history of alcohol use disorder in the current study (27%) represents a minimum figure. A final limitation of import is the small size of some of the subgroups, which is reflected in the large CI.


The main finding was a strong association between alcohol use disorder and suicide in persons 65 years of age and above. The estimated risk of suicide was even higher than that (OR = 8.4) observed in a study of mixed-age suicides in Northern Ireland (Foster et al., 1997). One reason for the higher OR was the very low frequency of alcohol use disorder in the control group, which was expected, considering the advanced age of the participants. Alcohol use disorder was observed in 27% of the suicide victims in our study, a figure similar to those reported from previous uncontrolled studies of elderly suicides in the USA (Clark and Clark, 1993) and Finland (Henriksson et al., 1995), but considerably higher than the 3.8% observed in a recent British study (Harwood et al., 2001). The low rate of alcohol misuse among elderly suicides in this latter study was consistent with previous British findings (Barraclough, 1971). Thus, the disparate findings presented in Table 1 may reflect real international differences. There is evidence, on a population level, that the association between alcohol and suicide is more pronounced in ‘dry’ northern European cultures than in settings with higher per capita alcohol consumption (Ramstedt, 2001). Culturally determined attitudes to suicide, drinking behaviour, help-seeking behaviour, and the availability of health care services are some factors that will determine mortality (both natural and unnatural) in persons with alcohol use disorders.

The finding that alcohol use disorder is a significant predictor of suicide risk also in women was somewhat unexpected. Community studies of completed suicides consistently show lower rates of alcohol use disorders in women than in men. While a retrospective diagnosis of alcoholism was demonstrated in over half (56%) of mixed aged male suicides in Stockholm (Beskow, 1979), such a diagnosis was observed in only 15% of female suicides from the same setting (Åsgård, 1990). A recent register study failed to show an association between alcohol dependence and suicide in women of mixed ages (Baxter and Appleby, 1999). However, such studies will miss individuals who seek care outside the psychiatric sector. An advantage of the current approach is that persons with alcohol problems can be identified regardless of health care utilization.

The loss of a close interpersonal relationship has been shown to be more common among suicides with alcoholism than among those without (Berglund and Öjehagen, 1998; Murphy, 2000). Recent loss due to separation was uncommon in our study. One reason for this disparity might be that we are dealing specifically with the elderly. Many of the alcoholics in our study suffered separations earlier in life and now had ‘nothing left to lose’. This is in line with the Carney et al.(1994) finding that interpersonal loss was less common among elderly suicide victims than among their younger counterparts. Poor personal economy is another factor reported to be more common among suicides with alcohol use disorder, than in those without (Heikkinen et al., 1994). Problems with personal finances were uncommon in those both with and without alcohol use disorder in our study. We are dealing with a survival population and risk factors for suicide in younger age groups will not necessarily apply in the oldest segment of the population. Persons with both alcohol use disorder and financial difficulties would be expected to have higher mortality earlier in life.

The proportion with a positive post-mortem screening for ethanol in our study (29%) was not unlike that reported for elderly white suicides (21%) in a recent US study (Garlow, 2002). Persons with alcohol use disorder in our study were more likely to use non-violent suicide methods than those without this disorder. This finding, which replicates the work of others (Conwell et al., 1998), underlines the need for clinicians to be restrictive in prescribing sedatives and hypnotics to this patient group.

Implications for suicide prevention

In conclusion, results from the current study show a strong association between alcohol use disorder and late life suicide in both men and women in this northern European setting. While the number of persons with alcohol dependence and misuse who reach old age is limited, due to increased mortality by both natural and unnatural causes, those who do survive appear to remain at risk for suicide.


The author wishes to thank the informants and the control persons who made this study possible. Thanks to Christina Andersson, Eva Johansson and Katarina Wilhelmson who carried out interviews. Katarina Wilhelmson and Eva Rubenowitz reviewed records for the controls. We also thank Kerstin Boström and Adam Berkowicz at the Göteborg Institute of Forensic Medicine. Jan Beskow, Bo Runeson, Ingmar Skoog and Peter Allebeck contributed scientific advice. The study was supported by grants from the Swedish Council for Social Research (grant F00 42/1998), the Swedish Foundation for Health Care Science and Allergy Research (V98 226), Pharmacia-Upjohn, the Söderström König Foundation and Yhlen’s Foundation.


View Abstract