- Email: [email protected]
Aftercare and clinical characteristics of people with mental illness who commit suicide: a case-control study
ARTICLES
Aftercare and clinical characteristics of people with mental illness who commit suicide: a case-control study Louis Appleby, John A Dennehy, Christopher S Thomas, E Brian Faragher, Glyn Lewis
Summary Background Suicide prevention is now a health priority in many countries. In the UK, there are specific targets for reducing the suicide rate in the general population and in people with mental illness. However, there is almost no evidence for the effectiveness of health services in reducing suicide, and little evidence linking suicide to any aspect of health-service care. Method We conducted a case-control study of people who committed suicide after discharge from psychiatric inpatient care. Cases were a 30-month sample of 149 people who had received an inquest verdict of suicide or open verdict in Greater Manchester, and who had a history of psychiatric admission in the 5 years before death. Controls were surviving psychiatric patients individually matched for age, sex, diagnosis, and date of last admission. Cases and controls were compared on aspects of psychiatric care, and on clinical and social variables, information being obtained from case notes. Findings Those who took their own lives were more likely to have had their care reduced (odds ratio 3·7 [95% CI 1·8–7·6]) at the final appointment in the community before death. Suicide was also associated with a history of selfharm (3·1 [1·7–5·7]), suicidal thoughts during aftercare (1·9 [1·0–3·5]) and the most recent admission as the first illness (2·0 [1·1–3·6]). The associations reported above took account of a number of confounding factors, including the predictable risk of suicide judged from case notes. Only 34% of suicides had an identifiable key worker, the essence of the Care Programme Approach. This frequency was no higher than that for controls, reflecting the difficulty of identifying those likely to commit suicide. Interpretation Reductions in care are strongly associated with suicide by people with mental illness, and may be contributory. The implication is that maintaining care beyond the point of clinical recovery is important in protecting highrisk individuals. Several clinical variables indicate high risk but greater risk is not generally addressed in health service provisions. Lancet 1999; 353: 1397–1400
Introduction A reduction in the suicide rate in people with severe mental illness is a main target in The Health of the Nation,1 a document produced by the UK Department of Health. Suicide risk is high in severe mental illness2,3 and about half of those who commit suicide have previously been referred to psychiatric services. 2,4 However, there is to our School of Psychiatry and Behavioural Sciences, University Hospital of South Manchester, West Didsbury, Manchester, M20 8LR, UK (Prof Louis Appleby MD, J A Dennehy MRCPsych, C S Thomas MD); Research Support Unit, Medical School, Manchester (E B Faragher PhD); and University of Wales College of Medicine, Heath Park, Cardiff (Prof G Lewis PhD ) Correspondence to: Prof Louis Appleby
THE LANCET • Vol 353 • April 24, 1999
knowledge no evidence that the activities of mental-health services affect suicide risk. A randomised controlled trial of suicide prevention by such services would require several thousand patients. An observational study is therefore the only practicable way of addressing the relationship of suicide to service provision. Such a study is made difficult because one would expect patients at higher risk to receive more intensive treatment, so the analysis of data would have to be adjusted for this. Our study used a case-control design to assess the relationship of suicide in severe mental illness to aspects of mental-health care, to clinical features, and to historical variables, both clinical and social. All cases and controls had previously been admitted to hospital— suicide risk is especially high after hospital discharge.5 Risk of suicide was assessed without awareness of case histories to enable adjustment of the analysis. Preliminary results, on a 1-year sample, suggested that conventional population risk factors for suicide do not identify the potential for suicide in people with severe mental illness.4 In the full 3-year study, the larger sample allows a more detailed examination of the relationship between suicide and mental-health care; the main aim was to relate suicide to the components of community-based aftercare and to recent changes in the aftercare of individuals.
Methods Cases A 30-month consecutive sample of suicides and probable suicides in Greater Manchester was collated. All individuals had received an inquest verdict of suicide or open verdict between Oct 1, 1993, and March 31, 1996, at one of the four coroner’s courts serving the area. From demographic information in the coroner’s inquest file, the local psychiatric unit of each individual was found and those with a history of contact with mental-health services were identified. In some cases information on psychiatric contact was also obtained from inquest files or medical records held by the general practitioner or family-health-service authority. In addition, details were checked against the patient databases of all psychiatric hospitals in the eight health districts that make up Greater Manchester, to detect those who had been admitted outside the most recent district of residence. Individuals were included in the study if their home address was inside the Greater Manchester area, and if they had had an inpatient psychiatric admission in one of the psychiatric hospitals in the study area in the 5 years before death. We obtained and examined psychiatric case records and we recorded information on the last admission before death and care after discharge from hospital. Ancillary records such as those kept by general practitioners, social workers, and community psychiatric nurses were also examined and the information collated.
Controls Controls were identified following block randomisation of the hospitals in the study area. This process excluded from the randomisation the hospital in which a case was treated, and with each subsequent selection excluded any hospital from which controls had already been selected. When all possible pairs of hospitals for a given case hospital had been selected, the cycle began again. Block randomisation ensured maximum variation in the pairings of case and control hospitals, and therefore in the care provided to case and control patients (taking cases and
1397
ARTICLES
Unemployed Married Living alone Index admission = first illness Frequent relapse (>1 episode of illness per year) Previously sectioned under Mental Health Act History of substance misuse History of self-harm History of dangerous self-harm (violent methods) Suicidal ideas leading to last admission Self-harm leading to last admission Suicidal ideas during last admission Depression during aftercare Suicidal thoughts during aftercare Depression in 3 months before death Final admission lasting <1 week† Final admission lasting >3 months†
Suicides (n=149)
Controls (n=149)
103 (69%) 35 (23%) 60 (40%) 57 (38%) 41 (28%) 49 (33%) 74 (50%) 111 (74%) 64 (43%) 78 (52%) 59 (40%) 99 (66%) 48 (32%) 59 (40%) 44 (30%) 37 (25%) 15 (10%)
101 (68%) 29 (19%) 50 (34%) 45 (30%) 32 (21%) 37 (25%) 63 (42%) 73 (49%) 41 (28%) 59 (40%) 30 (20%) 67 (45%) 35 (23%) 31 (21%) 23 (15%) 23 (15%) 8 (5%)
Odds ratio (95% CI) 1·09 (0·62–1·89) 0·74 (0·40–1·38) 1·36 (0·83–2·21) 1·48 (0·89–2·46) 1·81 (1·01–3·25) 1·86 (0·97–3·56) 1·79 (0·93–3·44) 3·38 (1·93–5·90) 2·00 (1·21–3·30) 1·76 (1·08–2·88) 2·93 (1·63–5·27) 2·60 (1·55–4·35) 1·59 (0·93–2·71) 2·56 (1·48–4·41) 2·31 (1·29–4·16) 1·89 (1·04–3·42) 2·13 (0·86–5·32)
Adjusted* odds ratio (95% CI) 1·12 (0·60–2·11) 1·07 (0·51–2·26) 1·62 (0·94–2·80) 2·00 (1·12–3·60) 1·36 (0·70–2·66) 1·76 (0·81–3·83) 1·67 (0·80–3·52) 3·05 (1·65–5·65) 1·22 (0·64–2·31) 1·01 (0·56–1·84) 1·39 (0·68–2·82) 1·51 (0·82–2·78) 1·08 (0·54–2·17) 1·89 (1·02–3·51) 1·30 (0·68–2·59) 1·60 (0·82–3·12) 2·04 (0·74–5·68)
*Adjusted for history of self-harm, suicidal ideas during aftercare, index admission=first illness, and depression in 3 months before date of suicide. †Baseline=final admission lasting 1 week to 3 months.
Table 1: Suicide and demographic, historical, and clinical variables controls from the same hospital would have narrowed differences in care arising from local policies). For each case possible controls were identified with the same age (within 5 years), sex, clinical diagnosis, and date of admission (within 6 months); when more than one suitable control was available, the control with the closest date of admission was selected. Information on each control was then collected as above, up to the day on which the corresponding case committed suicide. Cases and controls were thus matched for duration of follow-up since admission.
compulsory hospital admission. For the purpose of the analysis, these numerical ratings were converted to three categories—low (0–1), moderate (2–3), and high risk (4–5)—and subsequently to two categories—low and moderate/high. On the six-point scale, the rate of initial agreement between the two raters was 81% (=0·72); on the condensed three-category and two-category scales, the figure was 89% (0·80) and 93% (0·85), respectively. Ethical approval was obtained from the research ethics committees of all health districts in the study area.
Data collection
Statistical analysis
Information was recorded under three headings: historical variables (clinical and demographic); clinical features since the final inpatient admission; and aspects of care after inpatient discharge. The care variables of interest were components of aftercare at the time of suicide (or index date in controls, defined as the date of suicide by the matched case), and increases or decreases in care at the final contact with services in the community before the suicide or index date. Decreases in care were defined as: decreases in frequency of follow-up; decreases in drug dosage; transfer to a treatment setting with a lower level of observation, such as day hospital to outpatient clinic; discharge from follow-up; combinations of the foregoing. Discharge from inpatient care itself was excluded (cases and controls had been matched for timing of inpatient admission). Increases in care corresponded to these decreases. In order to relate suicide to the appropriateness of care, as well as actual care, it was necessary to adjust the analysis for the confounding effect of predictable risk on which the provision of care at the time of suicide might have been based. Predictable risk was defined as the risk of suicide as it would have been perceived by a clinician at the last contact before the suicide or index date, namely the last point at which care could have been adjusted to reflect risk. Predictable risk was judged by two psychiatrists from our team of investigators from clinical histories of each patient prepared by a third psychiatrist on our team from case notes; the histories included reference to known risk factors for suicide and emphasised the details of the last admission and clinical progress after discharge. Risk was rated without knowledge of the status of cases and controls, and by each rater independently. Disagreements in ratings were discussed and an agreed rating was arrived at in each case. Ratings were initially made on a six point scale—0 indicated low risk (at a level that might lead to discharge from follow-up), 5 indicated risk high enough to justify urgent and
The proposed sample size was initially 300 cases and 300 controls, which would have detected as statistically significant (at the 5% level) an increased prevalence in a risk factor equivalent to an odds ratio of 1·85 with 80% power and 15% exposed in the controls. The achieved sample of 149 cases and 149 controls was sufficient to detect (at the 5% level) an odds ratio of 2·34 with 80% power. Conditional logistic regression7 was done using the clogit procedure of Stata (release 5, Statacorp; Texas) to calculate odds ratios and CIs for suicide. As suicide is uncommon, the odds ratios have to have the same value as rate ratios. Many of the variables listed in table 1 were highly correlated; we therefore adjusted for four variables that were most strongly associated with suicide and appeared to address different issues (history of self-harm, suicidal ideas during aftercare, index admission as first illness, and depression in the 3 months before suicide). As an additional check on our sampling of controls we investigated the proportion of patients from each hospital who had reductions in care. There was no evidence to support any differences between hospitals (2 13=9·6, p=0·72 in controls). We also adjusted the conditional logistic regression results with hospital of last admission as a categorical variable.
Low Moderate High Moderate ⫹ High
Results 708 individuals received an inquest verdict of suicide (411) or open verdict (297) in Greater Manchester during the study period. 12 open verdicts were excluded because a specific cause of death other than suicide was suspected (such as postoperative deaths, deaths of infants), leaving a total sample of 696 suicides and probable suicides. There were 518 males and 178 females, a male to female ratio of 2·9:1. 229 (33%, 95% CI 29–36) had a history of psychiatric inpatient
Suicides (n=149)
Controls (n=149)
Odds ratio (95% CI)
Adjusted* odds ratio (95% CI)
60 (40%) 74 (50%) 15 (10%) 89 (60%)
87 (58%) 54 (36%) 8 (5%) 62 (41%)
1·00† 2·47 (1·39–4·40) 3·33 (1·24–8·96) 2·59 (1·48–4·53)
1·00† 1·40 (0·71–2·75) 1·19 (0·37–3·80) 1·40 (0·71–2·69)
*Adjusted for history of self-harm, suicidal ideas during aftercare, index admission=first illness, and depression in 3 months before date of suicide. †Baseline.
Table 2: Predictable risk
1398
THE LANCET • Vol 353 • April 24, 1999
ARTICLES
admission, 86 (12%, 10–15) had been seen at a psychiatric outpatient clinic but not admitted, and a further 31 (4%, 3–6) had had contact with mental health services at an accident and emergency department or through a domiciliary visit. In addition, 68 people (10%, 8–12) were known to have been treated by a general practitioner for a psychiatric condition but not referred to mental-health services. 282 people (41%, 37–44) had no record of contact with psychiatric services or treatment by a general practitioner for a psychiatric condition. Of the 229 people who had a history of psychiatric inpatient admission, 27 (12%) had last been an inpatient more than five years before suicide, whereas 11 (5%) had been admitted outside the study area. 33 (5% of all suicides) were inpatients at the time of suicide. There were therefore 158 people who committed suicide in Greater Manchester within 5 years of discharge from inpatient psychiatric care; the psychiatric records on 149 (94%) were traced. The following analyses relate to these 149 cases and their matched controls. The median age of the sample was 38 years (range 16–93). Compared with suicides in the general population, the sex difference (male: female=1·8:1) was small. There was evidence of postdischarge clustering of suicides, 7 (5%) occuring within one week, 47 (32%) within 3 months, 86 (58%) within 12 months of discharge. 23 (15%) people committed suicide before their first follow-up appointment. The most common primary diagnoses were major affective disorder (60 cases, 40%), alcohol dependence (36 cases, 24%), schizophrenia (34 cases, 23%), personality disorder (8 cases, 5%), and drug dependence and misuse (7 cases, 5%). Table 1 shows the relationship between suicide and demographic, past clinical, and recent clinical variables. Many of these variables were significantly associated with suicide. However, they are also associated with each other so the adjustment results in the last column are of the greatest interest. The strongest associations were between suicidal thoughts and behaviour. A history of self-harm (lifetime) and suicidal ideas during aftercare were both independently associated with suicide. The risk was also increased if the index admission was the first admission. None of the other variables had a statistically significant association with suicide, though it should be noted that the CIs are broad and potentially important associations cannot be ruled out. Table 2 shows that suicides were more likely to be rated as moderate or high risk by the clinicians’ ratings of the case notes, although when this analysis was adjusted for some of the main clinical risk factors the association with predictable risk was no longer significant. There was
no significant difference in the likelihood that risk in the suicides would be rated as either moderate or high (the number of individuals rated as high was small), and the moderate and high risk groups were therefore combined and the analysis done again after adjustment for the new dichotomous risk variable. Table 3 shows that the components of aftercare received by suicides and controls were generally similar according to variables available to this study, both before and after adjusting for predictable risk and for the main variables associated with suicide. When all aftercare variables were taken together, there was no association with suicide (26=3·42, p=0·75). Around a third of patients in both groups had an identifiable key worker, a central component of the care programme approach. This initiative was introduced in 1991 and in some cases (and controls) the final contact with services could have occurred before its introduction. For this reason, the equivalent figures for those who committed suicide within a year of hospital discharge were calculated and were slightly higher: 42% (cases) and 38% (controls). However, suicides were more likely to have had their care reduced (likelihood ratio x2(1)=9·8, p=0·002 after adjustment) and less likely to have had their care increased at the final appointment before death. The odds ratios associated with decreases in care were greater after adjustment for predictable risk and for the main associations of suicide. Decreases in care occurred in 70% of suicides and 37% of controls at low predictable risk, and in 27% of suicides and 11% of controls at high predictable risk. The association of suicide with decreases in care was still present when discharges from psychiatric follow-up were excluded (there remained an association with reduced but continuing care). With adjustment for hospital of last admission, the findings were the same. In total, 66 (44%) suicides and 39 (26%) controls had at least one decrease in care at the final consultation before death; in 11 suicides and 4 controls, decreases occurred in two or more aspects of care. The decreases were: reduced appointment frequency (41 suicides, 19 controls); less supervised location (24 suicides, 12 controls); lower drug dose (13 suicides, 12 controls). The diagnostic profile of these 66 cases was similar to that of the total sample, the most common diagnoses being major affective disorder (33 cases, 50%), alcohol dependence (16 cases, 24%), and schizophrenia (13 cases, 20%). In the majority of these suicides (37 cases, 56%) and controls (30 cases, 77%) whose care was reduced this was because the clinician judged the patient to be well; 13 (20%) suicides and 4 (10%) controls refused treatment or requested a decrease. In the majority
Suicides (n=149)
Controls (n=149)
Odds ratio (95% CI)
Adjusted odds ratio* (95% CI)
Adjusted odds ratio† (95% CI)
Components of aftercare at time of suicide Key Worker Monthly appointments in the community with CPN/PSW Rehabilitation/occupational therapy Adequate housing Contact with psychiatrist within 3 months Compliance with treatment plan
51 (34%) 54 (36%) 40 (27%) 14 (77%) 92 (62%) 58 (39%)
45 (31%) 42 (28%) 33 (22%) 122 (82%) 85 (57%) 66 (44%)
1·21 (0·70–2·08) 1·63 (0·92–2·89) 1·35 (0·76–2·41) 0·71 (0·40–1·27) 1·32 (0·76–2·29) 0·76 (0·45–1·27)
1·02 (0·58–1·80) 1·36 (0·75–2·48) 1·42 (0·77–2·61) 0·68 (0·37–1·24) 1·28 (0·72–2·27) 0·95 (0·54–1·65)
0·71 (0·34–1·49) 0·97 (0·45–2·09) 1·04 (0·50–2·17) 0·48 (0·23–1·02) 1·25 (0·58–2·70) 0·76 (0·36–1·61)
Final contact before suicide Decrease in care (excluding inpatient discharge) Decrease in care (excluding any discharge) Increase in care
66 (44%) 45 (30%) 13 (9%)
39 (26%) 26 (17%) 22 (15%)
2·17 (1·33–3·56) 2·06 (1·17–3·61) 0·59 (0·30–1·17)
3·56 (1·93–6·59) 3·52 (1·72–7·01) 0·45 (0·22–0·94)
3·73 (1·82–7·65) 3·28 (1·47–7·32) 0·47 (0·20–1·10)
CPN=community psychiatric nurse; PSW=psychiatric social worker. *Adjusted for predictable risk of suicide. †Adjusted for history of self-harm, suicidal ideas during aftercare, index admission was first illness, depression in 3 months before date of suicide, predictable risk of suicide, hospital of index admission.
Table 3: Psychiatric aftercare
THE LANCET • Vol 353 • April 24, 1999
1399
ARTICLES
of suicides (37 cases, 56%) death occurred within 3 months of the reduction in care.
Discussion The main finding of this study is that suicides in people with mental illness were associated with reductions in care at the final service contact before death. In most cases, patients had been regarded as well at the time when care was reduced, yet suicide occurred within 3 months of final contact. Increases in care at final contact were more common in controls matched for age, sex, diagnosis, and duration of follow-up since admission; this was statistically significant after adjustment for predictable risk but not after all adjustments. Other findings of the study confirmed the difficulties facing clinical practice in identifying the patients who are at highest suicide risk. The majority of suicides were not subject to the care programme approach, a formal system of providing and documenting psychiatric aftercare aimed at the most vulnerable people with mental illness. The components of aftercare were similar in suicides and controls. However, there were clinical indications of risk in the suicides, the most important being a history of selfharm (usually overdose), the most recent admission being the first admission, and expressing suicidal thoughts during aftercare. In addition, the CIs for some other variables are wide and we cannot exclude the possibility that they had a significant association with suicide. Clinical ratings of suicide risk based upon the case notes did not significantly improve prediction of suicide once risk factors described above had been taken into account. Our findings are open to criticism, however, in that they are based on information recorded in psychiatric case notes. Case notes vary in quality, and in general do not record all aspects of assessment or clinical decision making. They do not allow judgments to be made of the quality of care, and our care variables were therefore relatively crude, being based, for example, on the involvement of particular professional groups or frequency of appointments. However, these are likely to be key elements of a care plan and the quality of notes was generally sufficient to allow these and the other variables of interest to be studied. The information in the case notes was also collected by clinicians who did not know which individuals would commit suicide, as such it is unbiased information; any random misclassification in the care variables would tend to bias an observed odds ratio towards 1·0. The majority of reductions in care before suicide occurred in people who were judged to be improving or clinically well; in the suicides, more of the reductions in care were initiated by the patient (treatment refusals or requests) but the numbers in both suicides and controls was small. Most reductions in care would have led to reduced supervision. There was little evidence in the case notes that clinicians were wrong to view their patients as well at the final contact and two alternative explanations are at least as likely. First, clinical deterioration may have occurred after care was reduced, when services were unable to detect it or respond sufficiently quickly. Second, the care provided at the time of final appointment may have been necessary to prevent relapse or protect the patient from his persistent suicide risk. The latter explanation is in keeping with a model of suicide risk based on risk factor balanced by protective factor;8 according to this model, suicide risk is dependent on the balance of risk factors (such as suicidal thoughts, 1400
depressed mood) and protective factors (aspects of psychiatric care), and a patient can be at higher risk either when risk factors increase (when mental state deteriorates) or when protective factors decrease (when care is reduced). The clinical significance of this model is that it links the prevention of suicide to specific clinical activities. Our findings provide evidence for this, by suggesting that some suicides may be prevented by maintaining the intensity of aftercare beyond the point of perceived clinical recovery or by appropriate increases in supervision or medication, or both. The population attributable fraction (PAF) for decreases in care can be calculated9 from the figures in the final column of table 3 to be 41·2% (decreases excluding inpatient discharge) and 28·1% (decreases excluding any discharge). In other words, this is the maximum reduction in suicides in this patient group that could be achieved by maintaining aftercare, assuming the relationship to be causal. In practice the decrease is likely to be smaller, and to be achieved at a cost of maintaining aftercare in a large number of patients who will not commit suicide in any case. If these results are to have clinical application, there needs to be a way of targeting this activity. For example, one possible policy would be to maintain aftercare at the same intensity for the first 12 months after discharge. About half of the suicides in this study occurred within the first year of discharge corresponding to a PAF of about 20%. A more precisely defined group of high-risk patients would make it easier to incorporate such a strategy into clinical practice. Our findings provide evidence for a possible causal link between maintaining aftercare of psychiatric patients and suicidal behaviour. We were unable to study the influence of the content or the continuity of the care provided by the psychiatric teams but these are also likely to be of some importance. Although it is often assumed that psychiatrists can reduce the risk of suicide in their patients this is difficult to support by empirical evidence. Our findings provide a possible route towards prevention of suicide in a particularly high-risk group, namely psychiatric patients who have been discharged from hospital. Contributors L Appleby initiated and supervised the study and with the other investigators designed the study protocol. J Dennehy collected the data E B Faragher and G Lewis did the statistical analysis. L Appleby and C S Thomas did assessments of predictable risk. L Appleby prepared the paper and all investigators contributed to the final manuscript.
Acknowledgments The study was funded by North West NHS Executive
References 1
Department of Health. The Health of the Nation. London: HMSO, 1992. 2 Barraclough BM, Bunch J, Nelson B, et al. A hundred cases of suicide. Br J Psychiatry 1974; 125: 355–73. 3 Allebeck P, Allgulander C, Fisher LD. Predictors of completed suicide in a cohort of 50,465 young men. BMJ 1988; 297: 176–78. 4 King E, Barraclough B. Violent death and mental illness: a study of a single catchment area over 8 years. Br J Psychiatry 1990; 156: 714–20. 5 Goldacre M, Seagroatt V, Hawton K. Suicide after discharge from psychiatric in-patient care. Lancet 1993; 342: 283–86. 6 Dennehy JE, Appleby L, Thomas CS, Faragher B. A case control study of suicide by discharged psychiatric patients. BMJ 1996; 312: 1580. 7 Breslow NE, Day NE. Statistical methods in cancer research. Vol 1: The analysis of case-control studies. Lyon: International Agency for Research on Cancer; 1980. 8 Appleby L. Suicide in psychiatric patients; risk and prevention. Br J Psychiatry 1992; 161: 749–58. 9 Last JM, ed. Dictionary of Epidemiology, 2nd ed. Oxford: Oxford University Press. 1988.
THE LANCET • Vol 353 • April 24, 1999
Aftercare and clinical characteristics of people with mental illness who commit suicide: a case-control study Louis Appleby, John A Dennehy, Christopher S Thomas, E Brian Faragher, Glyn Lewis
Summary Background Suicide prevention is now a health priority in many countries. In the UK, there are specific targets for reducing the suicide rate in the general population and in people with mental illness. However, there is almost no evidence for the effectiveness of health services in reducing suicide, and little evidence linking suicide to any aspect of health-service care. Method We conducted a case-control study of people who committed suicide after discharge from psychiatric inpatient care. Cases were a 30-month sample of 149 people who had received an inquest verdict of suicide or open verdict in Greater Manchester, and who had a history of psychiatric admission in the 5 years before death. Controls were surviving psychiatric patients individually matched for age, sex, diagnosis, and date of last admission. Cases and controls were compared on aspects of psychiatric care, and on clinical and social variables, information being obtained from case notes. Findings Those who took their own lives were more likely to have had their care reduced (odds ratio 3·7 [95% CI 1·8–7·6]) at the final appointment in the community before death. Suicide was also associated with a history of selfharm (3·1 [1·7–5·7]), suicidal thoughts during aftercare (1·9 [1·0–3·5]) and the most recent admission as the first illness (2·0 [1·1–3·6]). The associations reported above took account of a number of confounding factors, including the predictable risk of suicide judged from case notes. Only 34% of suicides had an identifiable key worker, the essence of the Care Programme Approach. This frequency was no higher than that for controls, reflecting the difficulty of identifying those likely to commit suicide. Interpretation Reductions in care are strongly associated with suicide by people with mental illness, and may be contributory. The implication is that maintaining care beyond the point of clinical recovery is important in protecting highrisk individuals. Several clinical variables indicate high risk but greater risk is not generally addressed in health service provisions. Lancet 1999; 353: 1397–1400
Introduction A reduction in the suicide rate in people with severe mental illness is a main target in The Health of the Nation,1 a document produced by the UK Department of Health. Suicide risk is high in severe mental illness2,3 and about half of those who commit suicide have previously been referred to psychiatric services. 2,4 However, there is to our School of Psychiatry and Behavioural Sciences, University Hospital of South Manchester, West Didsbury, Manchester, M20 8LR, UK (Prof Louis Appleby MD, J A Dennehy MRCPsych, C S Thomas MD); Research Support Unit, Medical School, Manchester (E B Faragher PhD); and University of Wales College of Medicine, Heath Park, Cardiff (Prof G Lewis PhD ) Correspondence to: Prof Louis Appleby
THE LANCET • Vol 353 • April 24, 1999
knowledge no evidence that the activities of mental-health services affect suicide risk. A randomised controlled trial of suicide prevention by such services would require several thousand patients. An observational study is therefore the only practicable way of addressing the relationship of suicide to service provision. Such a study is made difficult because one would expect patients at higher risk to receive more intensive treatment, so the analysis of data would have to be adjusted for this. Our study used a case-control design to assess the relationship of suicide in severe mental illness to aspects of mental-health care, to clinical features, and to historical variables, both clinical and social. All cases and controls had previously been admitted to hospital— suicide risk is especially high after hospital discharge.5 Risk of suicide was assessed without awareness of case histories to enable adjustment of the analysis. Preliminary results, on a 1-year sample, suggested that conventional population risk factors for suicide do not identify the potential for suicide in people with severe mental illness.4 In the full 3-year study, the larger sample allows a more detailed examination of the relationship between suicide and mental-health care; the main aim was to relate suicide to the components of community-based aftercare and to recent changes in the aftercare of individuals.
Methods Cases A 30-month consecutive sample of suicides and probable suicides in Greater Manchester was collated. All individuals had received an inquest verdict of suicide or open verdict between Oct 1, 1993, and March 31, 1996, at one of the four coroner’s courts serving the area. From demographic information in the coroner’s inquest file, the local psychiatric unit of each individual was found and those with a history of contact with mental-health services were identified. In some cases information on psychiatric contact was also obtained from inquest files or medical records held by the general practitioner or family-health-service authority. In addition, details were checked against the patient databases of all psychiatric hospitals in the eight health districts that make up Greater Manchester, to detect those who had been admitted outside the most recent district of residence. Individuals were included in the study if their home address was inside the Greater Manchester area, and if they had had an inpatient psychiatric admission in one of the psychiatric hospitals in the study area in the 5 years before death. We obtained and examined psychiatric case records and we recorded information on the last admission before death and care after discharge from hospital. Ancillary records such as those kept by general practitioners, social workers, and community psychiatric nurses were also examined and the information collated.
Controls Controls were identified following block randomisation of the hospitals in the study area. This process excluded from the randomisation the hospital in which a case was treated, and with each subsequent selection excluded any hospital from which controls had already been selected. When all possible pairs of hospitals for a given case hospital had been selected, the cycle began again. Block randomisation ensured maximum variation in the pairings of case and control hospitals, and therefore in the care provided to case and control patients (taking cases and
1397
ARTICLES
Unemployed Married Living alone Index admission = first illness Frequent relapse (>1 episode of illness per year) Previously sectioned under Mental Health Act History of substance misuse History of self-harm History of dangerous self-harm (violent methods) Suicidal ideas leading to last admission Self-harm leading to last admission Suicidal ideas during last admission Depression during aftercare Suicidal thoughts during aftercare Depression in 3 months before death Final admission lasting <1 week† Final admission lasting >3 months†
Suicides (n=149)
Controls (n=149)
103 (69%) 35 (23%) 60 (40%) 57 (38%) 41 (28%) 49 (33%) 74 (50%) 111 (74%) 64 (43%) 78 (52%) 59 (40%) 99 (66%) 48 (32%) 59 (40%) 44 (30%) 37 (25%) 15 (10%)
101 (68%) 29 (19%) 50 (34%) 45 (30%) 32 (21%) 37 (25%) 63 (42%) 73 (49%) 41 (28%) 59 (40%) 30 (20%) 67 (45%) 35 (23%) 31 (21%) 23 (15%) 23 (15%) 8 (5%)
Odds ratio (95% CI) 1·09 (0·62–1·89) 0·74 (0·40–1·38) 1·36 (0·83–2·21) 1·48 (0·89–2·46) 1·81 (1·01–3·25) 1·86 (0·97–3·56) 1·79 (0·93–3·44) 3·38 (1·93–5·90) 2·00 (1·21–3·30) 1·76 (1·08–2·88) 2·93 (1·63–5·27) 2·60 (1·55–4·35) 1·59 (0·93–2·71) 2·56 (1·48–4·41) 2·31 (1·29–4·16) 1·89 (1·04–3·42) 2·13 (0·86–5·32)
Adjusted* odds ratio (95% CI) 1·12 (0·60–2·11) 1·07 (0·51–2·26) 1·62 (0·94–2·80) 2·00 (1·12–3·60) 1·36 (0·70–2·66) 1·76 (0·81–3·83) 1·67 (0·80–3·52) 3·05 (1·65–5·65) 1·22 (0·64–2·31) 1·01 (0·56–1·84) 1·39 (0·68–2·82) 1·51 (0·82–2·78) 1·08 (0·54–2·17) 1·89 (1·02–3·51) 1·30 (0·68–2·59) 1·60 (0·82–3·12) 2·04 (0·74–5·68)
*Adjusted for history of self-harm, suicidal ideas during aftercare, index admission=first illness, and depression in 3 months before date of suicide. †Baseline=final admission lasting 1 week to 3 months.
Table 1: Suicide and demographic, historical, and clinical variables controls from the same hospital would have narrowed differences in care arising from local policies). For each case possible controls were identified with the same age (within 5 years), sex, clinical diagnosis, and date of admission (within 6 months); when more than one suitable control was available, the control with the closest date of admission was selected. Information on each control was then collected as above, up to the day on which the corresponding case committed suicide. Cases and controls were thus matched for duration of follow-up since admission.
compulsory hospital admission. For the purpose of the analysis, these numerical ratings were converted to three categories—low (0–1), moderate (2–3), and high risk (4–5)—and subsequently to two categories—low and moderate/high. On the six-point scale, the rate of initial agreement between the two raters was 81% (=0·72); on the condensed three-category and two-category scales, the figure was 89% (0·80) and 93% (0·85), respectively. Ethical approval was obtained from the research ethics committees of all health districts in the study area.
Data collection
Statistical analysis
Information was recorded under three headings: historical variables (clinical and demographic); clinical features since the final inpatient admission; and aspects of care after inpatient discharge. The care variables of interest were components of aftercare at the time of suicide (or index date in controls, defined as the date of suicide by the matched case), and increases or decreases in care at the final contact with services in the community before the suicide or index date. Decreases in care were defined as: decreases in frequency of follow-up; decreases in drug dosage; transfer to a treatment setting with a lower level of observation, such as day hospital to outpatient clinic; discharge from follow-up; combinations of the foregoing. Discharge from inpatient care itself was excluded (cases and controls had been matched for timing of inpatient admission). Increases in care corresponded to these decreases. In order to relate suicide to the appropriateness of care, as well as actual care, it was necessary to adjust the analysis for the confounding effect of predictable risk on which the provision of care at the time of suicide might have been based. Predictable risk was defined as the risk of suicide as it would have been perceived by a clinician at the last contact before the suicide or index date, namely the last point at which care could have been adjusted to reflect risk. Predictable risk was judged by two psychiatrists from our team of investigators from clinical histories of each patient prepared by a third psychiatrist on our team from case notes; the histories included reference to known risk factors for suicide and emphasised the details of the last admission and clinical progress after discharge. Risk was rated without knowledge of the status of cases and controls, and by each rater independently. Disagreements in ratings were discussed and an agreed rating was arrived at in each case. Ratings were initially made on a six point scale—0 indicated low risk (at a level that might lead to discharge from follow-up), 5 indicated risk high enough to justify urgent and
The proposed sample size was initially 300 cases and 300 controls, which would have detected as statistically significant (at the 5% level) an increased prevalence in a risk factor equivalent to an odds ratio of 1·85 with 80% power and 15% exposed in the controls. The achieved sample of 149 cases and 149 controls was sufficient to detect (at the 5% level) an odds ratio of 2·34 with 80% power. Conditional logistic regression7 was done using the clogit procedure of Stata (release 5, Statacorp; Texas) to calculate odds ratios and CIs for suicide. As suicide is uncommon, the odds ratios have to have the same value as rate ratios. Many of the variables listed in table 1 were highly correlated; we therefore adjusted for four variables that were most strongly associated with suicide and appeared to address different issues (history of self-harm, suicidal ideas during aftercare, index admission as first illness, and depression in the 3 months before suicide). As an additional check on our sampling of controls we investigated the proportion of patients from each hospital who had reductions in care. There was no evidence to support any differences between hospitals (2 13=9·6, p=0·72 in controls). We also adjusted the conditional logistic regression results with hospital of last admission as a categorical variable.
Low Moderate High Moderate ⫹ High
Results 708 individuals received an inquest verdict of suicide (411) or open verdict (297) in Greater Manchester during the study period. 12 open verdicts were excluded because a specific cause of death other than suicide was suspected (such as postoperative deaths, deaths of infants), leaving a total sample of 696 suicides and probable suicides. There were 518 males and 178 females, a male to female ratio of 2·9:1. 229 (33%, 95% CI 29–36) had a history of psychiatric inpatient
Suicides (n=149)
Controls (n=149)
Odds ratio (95% CI)
Adjusted* odds ratio (95% CI)
60 (40%) 74 (50%) 15 (10%) 89 (60%)
87 (58%) 54 (36%) 8 (5%) 62 (41%)
1·00† 2·47 (1·39–4·40) 3·33 (1·24–8·96) 2·59 (1·48–4·53)
1·00† 1·40 (0·71–2·75) 1·19 (0·37–3·80) 1·40 (0·71–2·69)
*Adjusted for history of self-harm, suicidal ideas during aftercare, index admission=first illness, and depression in 3 months before date of suicide. †Baseline.
Table 2: Predictable risk
1398
THE LANCET • Vol 353 • April 24, 1999
ARTICLES
admission, 86 (12%, 10–15) had been seen at a psychiatric outpatient clinic but not admitted, and a further 31 (4%, 3–6) had had contact with mental health services at an accident and emergency department or through a domiciliary visit. In addition, 68 people (10%, 8–12) were known to have been treated by a general practitioner for a psychiatric condition but not referred to mental-health services. 282 people (41%, 37–44) had no record of contact with psychiatric services or treatment by a general practitioner for a psychiatric condition. Of the 229 people who had a history of psychiatric inpatient admission, 27 (12%) had last been an inpatient more than five years before suicide, whereas 11 (5%) had been admitted outside the study area. 33 (5% of all suicides) were inpatients at the time of suicide. There were therefore 158 people who committed suicide in Greater Manchester within 5 years of discharge from inpatient psychiatric care; the psychiatric records on 149 (94%) were traced. The following analyses relate to these 149 cases and their matched controls. The median age of the sample was 38 years (range 16–93). Compared with suicides in the general population, the sex difference (male: female=1·8:1) was small. There was evidence of postdischarge clustering of suicides, 7 (5%) occuring within one week, 47 (32%) within 3 months, 86 (58%) within 12 months of discharge. 23 (15%) people committed suicide before their first follow-up appointment. The most common primary diagnoses were major affective disorder (60 cases, 40%), alcohol dependence (36 cases, 24%), schizophrenia (34 cases, 23%), personality disorder (8 cases, 5%), and drug dependence and misuse (7 cases, 5%). Table 1 shows the relationship between suicide and demographic, past clinical, and recent clinical variables. Many of these variables were significantly associated with suicide. However, they are also associated with each other so the adjustment results in the last column are of the greatest interest. The strongest associations were between suicidal thoughts and behaviour. A history of self-harm (lifetime) and suicidal ideas during aftercare were both independently associated with suicide. The risk was also increased if the index admission was the first admission. None of the other variables had a statistically significant association with suicide, though it should be noted that the CIs are broad and potentially important associations cannot be ruled out. Table 2 shows that suicides were more likely to be rated as moderate or high risk by the clinicians’ ratings of the case notes, although when this analysis was adjusted for some of the main clinical risk factors the association with predictable risk was no longer significant. There was
no significant difference in the likelihood that risk in the suicides would be rated as either moderate or high (the number of individuals rated as high was small), and the moderate and high risk groups were therefore combined and the analysis done again after adjustment for the new dichotomous risk variable. Table 3 shows that the components of aftercare received by suicides and controls were generally similar according to variables available to this study, both before and after adjusting for predictable risk and for the main variables associated with suicide. When all aftercare variables were taken together, there was no association with suicide (26=3·42, p=0·75). Around a third of patients in both groups had an identifiable key worker, a central component of the care programme approach. This initiative was introduced in 1991 and in some cases (and controls) the final contact with services could have occurred before its introduction. For this reason, the equivalent figures for those who committed suicide within a year of hospital discharge were calculated and were slightly higher: 42% (cases) and 38% (controls). However, suicides were more likely to have had their care reduced (likelihood ratio x2(1)=9·8, p=0·002 after adjustment) and less likely to have had their care increased at the final appointment before death. The odds ratios associated with decreases in care were greater after adjustment for predictable risk and for the main associations of suicide. Decreases in care occurred in 70% of suicides and 37% of controls at low predictable risk, and in 27% of suicides and 11% of controls at high predictable risk. The association of suicide with decreases in care was still present when discharges from psychiatric follow-up were excluded (there remained an association with reduced but continuing care). With adjustment for hospital of last admission, the findings were the same. In total, 66 (44%) suicides and 39 (26%) controls had at least one decrease in care at the final consultation before death; in 11 suicides and 4 controls, decreases occurred in two or more aspects of care. The decreases were: reduced appointment frequency (41 suicides, 19 controls); less supervised location (24 suicides, 12 controls); lower drug dose (13 suicides, 12 controls). The diagnostic profile of these 66 cases was similar to that of the total sample, the most common diagnoses being major affective disorder (33 cases, 50%), alcohol dependence (16 cases, 24%), and schizophrenia (13 cases, 20%). In the majority of these suicides (37 cases, 56%) and controls (30 cases, 77%) whose care was reduced this was because the clinician judged the patient to be well; 13 (20%) suicides and 4 (10%) controls refused treatment or requested a decrease. In the majority
Suicides (n=149)
Controls (n=149)
Odds ratio (95% CI)
Adjusted odds ratio* (95% CI)
Adjusted odds ratio† (95% CI)
Components of aftercare at time of suicide Key Worker Monthly appointments in the community with CPN/PSW Rehabilitation/occupational therapy Adequate housing Contact with psychiatrist within 3 months Compliance with treatment plan
51 (34%) 54 (36%) 40 (27%) 14 (77%) 92 (62%) 58 (39%)
45 (31%) 42 (28%) 33 (22%) 122 (82%) 85 (57%) 66 (44%)
1·21 (0·70–2·08) 1·63 (0·92–2·89) 1·35 (0·76–2·41) 0·71 (0·40–1·27) 1·32 (0·76–2·29) 0·76 (0·45–1·27)
1·02 (0·58–1·80) 1·36 (0·75–2·48) 1·42 (0·77–2·61) 0·68 (0·37–1·24) 1·28 (0·72–2·27) 0·95 (0·54–1·65)
0·71 (0·34–1·49) 0·97 (0·45–2·09) 1·04 (0·50–2·17) 0·48 (0·23–1·02) 1·25 (0·58–2·70) 0·76 (0·36–1·61)
Final contact before suicide Decrease in care (excluding inpatient discharge) Decrease in care (excluding any discharge) Increase in care
66 (44%) 45 (30%) 13 (9%)
39 (26%) 26 (17%) 22 (15%)
2·17 (1·33–3·56) 2·06 (1·17–3·61) 0·59 (0·30–1·17)
3·56 (1·93–6·59) 3·52 (1·72–7·01) 0·45 (0·22–0·94)
3·73 (1·82–7·65) 3·28 (1·47–7·32) 0·47 (0·20–1·10)
CPN=community psychiatric nurse; PSW=psychiatric social worker. *Adjusted for predictable risk of suicide. †Adjusted for history of self-harm, suicidal ideas during aftercare, index admission was first illness, depression in 3 months before date of suicide, predictable risk of suicide, hospital of index admission.
Table 3: Psychiatric aftercare
THE LANCET • Vol 353 • April 24, 1999
1399
ARTICLES
of suicides (37 cases, 56%) death occurred within 3 months of the reduction in care.
Discussion The main finding of this study is that suicides in people with mental illness were associated with reductions in care at the final service contact before death. In most cases, patients had been regarded as well at the time when care was reduced, yet suicide occurred within 3 months of final contact. Increases in care at final contact were more common in controls matched for age, sex, diagnosis, and duration of follow-up since admission; this was statistically significant after adjustment for predictable risk but not after all adjustments. Other findings of the study confirmed the difficulties facing clinical practice in identifying the patients who are at highest suicide risk. The majority of suicides were not subject to the care programme approach, a formal system of providing and documenting psychiatric aftercare aimed at the most vulnerable people with mental illness. The components of aftercare were similar in suicides and controls. However, there were clinical indications of risk in the suicides, the most important being a history of selfharm (usually overdose), the most recent admission being the first admission, and expressing suicidal thoughts during aftercare. In addition, the CIs for some other variables are wide and we cannot exclude the possibility that they had a significant association with suicide. Clinical ratings of suicide risk based upon the case notes did not significantly improve prediction of suicide once risk factors described above had been taken into account. Our findings are open to criticism, however, in that they are based on information recorded in psychiatric case notes. Case notes vary in quality, and in general do not record all aspects of assessment or clinical decision making. They do not allow judgments to be made of the quality of care, and our care variables were therefore relatively crude, being based, for example, on the involvement of particular professional groups or frequency of appointments. However, these are likely to be key elements of a care plan and the quality of notes was generally sufficient to allow these and the other variables of interest to be studied. The information in the case notes was also collected by clinicians who did not know which individuals would commit suicide, as such it is unbiased information; any random misclassification in the care variables would tend to bias an observed odds ratio towards 1·0. The majority of reductions in care before suicide occurred in people who were judged to be improving or clinically well; in the suicides, more of the reductions in care were initiated by the patient (treatment refusals or requests) but the numbers in both suicides and controls was small. Most reductions in care would have led to reduced supervision. There was little evidence in the case notes that clinicians were wrong to view their patients as well at the final contact and two alternative explanations are at least as likely. First, clinical deterioration may have occurred after care was reduced, when services were unable to detect it or respond sufficiently quickly. Second, the care provided at the time of final appointment may have been necessary to prevent relapse or protect the patient from his persistent suicide risk. The latter explanation is in keeping with a model of suicide risk based on risk factor balanced by protective factor;8 according to this model, suicide risk is dependent on the balance of risk factors (such as suicidal thoughts, 1400
depressed mood) and protective factors (aspects of psychiatric care), and a patient can be at higher risk either when risk factors increase (when mental state deteriorates) or when protective factors decrease (when care is reduced). The clinical significance of this model is that it links the prevention of suicide to specific clinical activities. Our findings provide evidence for this, by suggesting that some suicides may be prevented by maintaining the intensity of aftercare beyond the point of perceived clinical recovery or by appropriate increases in supervision or medication, or both. The population attributable fraction (PAF) for decreases in care can be calculated9 from the figures in the final column of table 3 to be 41·2% (decreases excluding inpatient discharge) and 28·1% (decreases excluding any discharge). In other words, this is the maximum reduction in suicides in this patient group that could be achieved by maintaining aftercare, assuming the relationship to be causal. In practice the decrease is likely to be smaller, and to be achieved at a cost of maintaining aftercare in a large number of patients who will not commit suicide in any case. If these results are to have clinical application, there needs to be a way of targeting this activity. For example, one possible policy would be to maintain aftercare at the same intensity for the first 12 months after discharge. About half of the suicides in this study occurred within the first year of discharge corresponding to a PAF of about 20%. A more precisely defined group of high-risk patients would make it easier to incorporate such a strategy into clinical practice. Our findings provide evidence for a possible causal link between maintaining aftercare of psychiatric patients and suicidal behaviour. We were unable to study the influence of the content or the continuity of the care provided by the psychiatric teams but these are also likely to be of some importance. Although it is often assumed that psychiatrists can reduce the risk of suicide in their patients this is difficult to support by empirical evidence. Our findings provide a possible route towards prevention of suicide in a particularly high-risk group, namely psychiatric patients who have been discharged from hospital. Contributors L Appleby initiated and supervised the study and with the other investigators designed the study protocol. J Dennehy collected the data E B Faragher and G Lewis did the statistical analysis. L Appleby and C S Thomas did assessments of predictable risk. L Appleby prepared the paper and all investigators contributed to the final manuscript.
Acknowledgments The study was funded by North West NHS Executive
References 1
Department of Health. The Health of the Nation. London: HMSO, 1992. 2 Barraclough BM, Bunch J, Nelson B, et al. A hundred cases of suicide. Br J Psychiatry 1974; 125: 355–73. 3 Allebeck P, Allgulander C, Fisher LD. Predictors of completed suicide in a cohort of 50,465 young men. BMJ 1988; 297: 176–78. 4 King E, Barraclough B. Violent death and mental illness: a study of a single catchment area over 8 years. Br J Psychiatry 1990; 156: 714–20. 5 Goldacre M, Seagroatt V, Hawton K. Suicide after discharge from psychiatric in-patient care. Lancet 1993; 342: 283–86. 6 Dennehy JE, Appleby L, Thomas CS, Faragher B. A case control study of suicide by discharged psychiatric patients. BMJ 1996; 312: 1580. 7 Breslow NE, Day NE. Statistical methods in cancer research. Vol 1: The analysis of case-control studies. Lyon: International Agency for Research on Cancer; 1980. 8 Appleby L. Suicide in psychiatric patients; risk and prevention. Br J Psychiatry 1992; 161: 749–58. 9 Last JM, ed. Dictionary of Epidemiology, 2nd ed. Oxford: Oxford University Press. 1988.
THE LANCET • Vol 353 • April 24, 1999