- Email: [email protected]
Suicide and severe mental illnesses. Cohort study within the UK general practice research database
Available online at www.sciencedirect.com
Schizophrenia Research 99 (2008) 134 – 138 www.elsevier.com/locate/schres
Suicide and severe mental illnesses. Cohort study within the UK general practice research database David Osborn a,⁎, Gus Levy a , Irwin Nazareth b , Michael King a a
b
Department of Mental Health Sciences, (Hampstead Campus), Royal Free and University College Medical School, Rowland Hill Street, London, NW3 2PF, UK Department of Primary Care and Population Sciences, (Hampstead Campus), Royal Free and University College Medical School, UCL, Rowland Hill Street, London, NW3 2PF, UK Received 19 September 2007; received in revised form 13 November 2007; accepted 18 November 2007 Available online 26 December 2007
Abstract We aimed to evaluate suicide risk across the life-course in severe mental illnesses (SMI) including schizophrenia. Using survival analysis, we compared suicide risk in cohorts of 46,136 people with SMI and 300,426 without. The overall unadjusted hazard ratio (HR) for suicide in SMI was 12.97 (95% CI: 9.75–17.25). The unadjusted HRs differed by age band: 18–30 years: 19.56 (9.76–39.17); 30–50 years: 13.14 (8.64–19.99); 50–70 years: 16.39 (9.15–29.37); 70+: 3.25 (1.33–7.94). In schizophrenia, risk was significantly higher when young but marked risk persisted until age 70. Greatest risk was associated with: increased consultation rates; antidepressant prescriptions and living in less deprived areas. © 2007 Elsevier B.V. All rights reserved. Keywords: Schizophrenia; Bipolar affective disorder; Suicide; Primary care; Risk factors
1. Introduction Suicide is a public health priority in people with severe mental illnesses (SMI) including schizophrenia and bipolar affective disorder (Department of Health, 1999; Hawton et al., 2005). Although it is believed that people with schizophrenia are at greatest risk of suicide early in the illness (Palmer et al., 2005; Limosin et al., 2007) recent reviews have questioned this evidence and have highlighted the lack of data from community ⁎ Corresponding author. Tel.: +44 20 77940500x33950; fax: +44 20 78302808. E-mail addresses: [email protected] (D. Osborn), [email protected] (G. Levy), [email protected] (I. Nazareth), [email protected] (M. King). 0920-9964/$ - see front matter © 2007 Elsevier B.V. All rights reserved. doi:10.1016/j.schres.2007.11.025
samples (Hawton et al., 2005; Palmer et al., 2005). Research on suicide in people with SMI has been limited by relatively small sample sizes; inadequate follow-up periods; few suicide events and sub-optimal statistical analyses. In the absence of an accurate estimation of the burden of suicide and the level of risk throughout the course of illness, we examined in a large community sample the risk of suicide in people with SMI and the demographic and family practice health service predictors of suicide. 2. Methods We undertook a cohort analysis of the risk of suicide in people on the UK general practice research database (GPRD) between 1987 and 2002, aged 18 years and over,
D. Osborn et al. / Schizophrenia Research 99 (2008) 134–138
with and without SMI. We included every patient with a clinical diagnosis of schizophrenia, bipolar affective disorders and other psychotic illnesses including delusional disorders and depression with psychosis. People with drug induced or organic psychoses were excluded. The comparison cohort, without such diagnoses, was randomly selected to achieve a ratio of 1:6, to maximise the statistical power to detect real differences. There were no other exclusion criteria to ensure representativeness. The GPRD contains routine clinical data from 755 general practices representative of those in the UK and has been used for pharmaco-epidemiological studies of suicide outcomes (Jick et al., 2004). In the UK, 95% of the population are registered with a general practice. If people with SMI are not registered with primary care practitioners this is usually organised by their community mental health team. The primary care practices also keep registers of those with SMI. The validity of a GPRD diagnosis of SMI has been confirmed (Nazareth et al., 1993a). Variables must be extracted a priori from the GPRD. We only chose variables which might be related to suicide and which are known to be valid and reliable in the GPRD, which is a working clinical database. For instance, whilst prescriptions and consultation episodes will be recorded accurately in the GPRD, the exact content of a consultation will be inconsistently entered by different practitioners. In deciding which variables to extract for this study we focused on socio-demographic predictors of suicide, consultation frequency prior to suicide and treatment of depression. We obtained data concerning age, sex, SMI diagnosis, General Practice (GP) consultation rate, antidepressant medication, smoking, and social deprivation using the Carstairs index for each general practice, categorised into quintiles. The GPRD did not include accurate data on deprivation or ethnicity at the level of the individual. The Carstairs index of the GP was the only available measure and is a standard index of deprivation that is calculated for each geographical area on four variables. These are: male unemployment; households with no car; overcrowding and head of households in unskilled manual occupation (Morris and Carstairs, 1991). To assess change in consultation behaviour we compared number of GP visits in the final 3 months on the GPRD database with the number of visits made in the same 3 months of the previous year. This comprised the last year of life in the case of suicide, whereas it was the last year on the data base for the remainder. We performed survival analysis using Cox regression techniques in STATA (version 8.2) (StataCorp, 2003). We estimated hazard ratios for suicide firstly comparing people with and without SMI and then restricting the
135
analysis to people with SMI. In a multivariate analysis we adjusted for major available explanatory variables including age, calendar period and geographical social deprivation. We also used smoking as a “proxy covariate” for the multivariate analysis since it is associated with individual deprivation and with poor physical and mental health. A test of proportional hazards was employed to determine whether hazard ratios were stable across ages. The cohort was stratified into age bands to estimate risk of suicide in different periods of life in the three main SMI diagnostic groups. Population attributable risk (PAR) was calculated using the hazard ratio for suicide and choosing the prevalence of SMI in the GPRD at the midpoint of the cohort (1994). 3. Results The cohorts comprised 46,136 people with SMI and 300,426 without SMI. The median follow-up time was 4.7 years in the SMI group, but 4.3 years in the comparison group. The majority of people attracted only one SMI diagnosis during their time in the GPRD (35,097/46,136 = 76.1%). The most common SMI diagnoses were schizophrenia (40.2%), bipolar affective disorder (23.3%), delusional disorder (19.2%), depressive psychosis (5.9%) and schizoaffective disorder (5.3%) (Osborn et al., 2007). This afforded statistical power to examine suicide rates separately in schizophrenia and bipolar affective disorder and the remaining SMI diagnosis was grouped together for further analyses. Where a person had attracted more than one diagnosis we applied two rules. Diagnoses of schizophrenia, schizoaffective disorder or bipolar disorder were given priority over delusional disorder, brief psychotic episodes, unipolar depressive psychoses or non-specified psychoses. Where a diagnosis had changed between schizophrenia, bipolar disorder or schizoaffective disorder over time we utilised the most recent diagnosis. In SMI, the crude rate for GP recorded suicide was far greater than in the comparison group (0.58 vs. 0.047 per 1000 person years), yielding an overall unadjusted hazard ratio of 12.97 (Table 1). This increased risk was observed irrespective of SMI diagnosis, sex or age in the under 70s. In the oldest group, the three-fold excess risk was still significant. After adjustment in the multivariate analysis, hazard ratios for suicide in SMI were smaller in magnitude. The individual variable responsible for this difference in magnitude was GP prescription of antidepressants.
136
D. Osborn et al. / Schizophrenia Research 99 (2008) 134–138
Table 1 Suicide mortality in people with and without severe mental illness Total
Suicides % suicide Crude hazard ratio
Adjusted hazard ratio a
SMI vs. comparison group
Diagnosis
Sex
Age bands
Social deprivation
Within SMI group only Sex
Comparison SMI Comparison Schizophrenia Bipolar disorder Other SMI Female comparison Female SMI Male comparison Male SMI Comparison 18–30 SMI 18–30 Comparison 30–50 SMI 30–50 Comparison 50–70 SMI 50–70 Comparison 70+ SMI 70+ Comparison 20% most deprived SMI 20% most deprived Comparison 20% least deprived SMI 20% least deprived
Female Male Age bands b 18–30 30–50 50–70 70+ Social deprivation quintiles 1 least deprived 20% 2 3 4 5 most deprived 20% Consultation change in final 3 months b2 more visits 2–4 more visits 4 or more visits Antidepressant in last 6 months No Yes
300,246 72 46,136 143 300,246 72 18,555 48 10,742 41 16,839 54 156,781 18 24,608 54 143,465 54 21,528 89 102,653 12 9600 24 134,455 33 20,223 65 88,937 15 17,265 46 50,152 12 11,121 8 63,312 16
0.02 0.31 0.02 0.26 0.38 0.32 0.01 0.22 0.04 0.41 0.01 0.25 0.02 0.32 0.02 0.27 0.02 0.07 0.03
1.00 12.97 1.00 10.99 15.20 13.67 1.00 18.05 1.00 10.96 1.00 19.56 1.00 13.14 1.00 16.39 1.00 3.25 1.00
(9.75, 17.25) (7.61, 15.85) (10.32, 22.37) (9.60, 19.48) (10.49, 31.06) (7.81, 15.37) (9.76, 39.17) (8.64, 19.99) (9.15, 29.37) (1.33, 7.94)
1.00 7.72 1.00 7.00 8.74 7.85 9.82 1.00 6.75 1.00 13.26 1.00 7.80 1.00 9.19 1.00 2.08 1.00
(5.66, 10.51) (4.78, 10.25) (5.80, 13.17) (5.38, 11.46) (5.55, 17.38) (4.65, 9.81) (6.24, 28.15) (4.92, 12.36) (4.93, 17.11) (0.81, 5.31)
12,469 31,825
32 9
0.26 0.03
10.26 (5.60, 18.79) 1.00
5.30 (2.77, 10.15) 1.00
3536
16
0.45
16.25 (7.13, 37.04)
10.52 (4.25, 26.05)
24,608 21,528 9600 20,223 17,265 11,121 3536 5526 8507 9542 12,469 35,162 6361 4613 33,757 12,379
54 89 24 65 46 8 16 24 28 29 32 70 38 35 62 81
0.22 0.41 0.25 0.32 0.27 0.07 0.45 0.43 0.33 0.30 0.26 0.20 0.60 0.76 0.18 0.65
1.00 1.70 1.00 1.05 0.85 0.25 1.00 0.87 0.72 0.63 0.53 1.00 3.53 5.45 1.00 3.27
(1.20, 2.40) (0.66, 1.68) (0.52, 1.39) (0.11, 0.55) (0.46, 1.65) (0.39, 1.33) (0.34, 1.16) (0.29, 0.96) (2.38, 5.24) (3.63, 8.20) (2.35, 4.55)
1.00 1.94 1.00 1.13 0.98 0.30 1.00 0.88 0.73 0.63 0.54 1.00 3.22 4.51 1.00 3.56
(1.37, 2.76) (0.71, 1.81) (0.59, 1.61) (0.14, 0.69) (0.47, 1.66) (0.40, 1.36) (0.34, 1.15) (0.30, 0.99) (2.16, 4.80) (2.98, 6.83) (2.55, 4.98)
a
Adusted for sex, age, calendar period, geographical social deprivation, antidepressant prescription in last 6 months and smoking. b Age bands were compared using a Poisson model as a Cox model could not be estimated due to co-linearity between the age bands. Bold text: 95% CI's do not overlap zero. SMI: Severe mental illness.
Hazard ratios for suicide in different age bands varied with SMI diagnosis (Table 2). In bipolar disorder and other SMI, highest rates occurred in the older age groups, but differences across age groups were not significant. Conversely, the hazard ratios for suicide in schizophrenia decreased significantly with age (test for proportional hazards p = 0.001). Stratifying by age band with the comparison group as the baseline, the suicide HR for
schizophrenia was 22.6 (95% CI 10.1–50.4) in 18–30 year olds; 11.2 (6.6–18.8) in 30–50 year olds and 8.77 (3.9– 19.5) in 50–70 year olds. Of the 48 suicides in people with schizophrenia, 13 (27.1%) occurred in the 18–30 year olds, 25 (52.1%) in the 30–50 year olds, 10 (20.8%) in the 50–70 year olds, and none in the over 70 year olds. In 1994 the overall GPRD prevalence of SMI was 0.87% yielding an unadjusted population attributable
D. Osborn et al. / Schizophrenia Research 99 (2008) 134–138 Table 2 Hazard ratios for suicide in SMI by diagnosis and age group, compared to controls 18–30 years
30–49 years
50–70 years
Schizophrenia 22.6 (10.1–50.4) 11.2 (6.6–8.8) 8.8 (3.9–19.5) Bipolar 14.2 (4.0–50.5) 14.8 (8.2–26.6) 24.8 (12.5–49.4) disorder Other SMI 13.2 (5.4–32.4) 13.5 (7.9–23.0) 21.4 (10.8–42.6) (psychosis) (95% confidence intervals).
risk percent of 9.4%. The PARs decreased over time, from 10.3% in 1988 to 7.7% in 2001. Within the SMI group, increasing age was associated with high relative rates of suicide until 70 after which there was a decreased rate (Table 1). Those completing suicide were more likely to be male and were over three times more likely to have received antidepressants in the last 6 months. In the 3 months prior to suicide, GP consultations increased by a mean of 1.90 (95% CI 1.34–2.47) visits, compared to the same 3 months of the previous year. Higher rates of suicide were associated with lower social deprivation scores. 4. Discussion We report a major excess of suicide deaths in schizophrenia, bipolar disorder and other SMI. Our data indicate that approximately one in ten of the suicides recorded in UK general practice may be attributable to SMI. Although the risk of suicide in schizophrenia relative to the comparison group was highest in people aged under 30, it was still raised eight-fold in 50 to 70 year olds. Seventy-three percent of suicides in schizophrenia occurred in people aged 30 and over, demonstrating that the risk must be assumed throughout the illness. This counters current wisdom that the majority of suicides occur early in schizophrenia and that suicide-prevention strategies are best delivered early in the course of the illness. The relative risk of suicide in other SMI was elevated throughout life, but rose sharply, after 50. While suicide rates are raised across all groups with SMI, those in the least socially deprived areas and those treated for depression are the most at risk. The importance and novelty of these findings lie in the large community nature of the sample, rather than relying on opportunistic follow-up of relatively small numbers of people from less generalisable settings such as inpatient psychiatric units. The GPRD also provided a
137
valid community comparison group and allowed us to control for other explanatory factors. Limitations include the fact that the GPRD lacks validated accurate data regarding social, demographic, economic or psychiatric status at the level of the individual. Data extraction is a major undertaking in the GPRD and we deliberately chose variables which are recorded reliably during clinical practice, to ensure the validity of our results. Furthermore, although recording of causes of death and diagnosis of SMI are validated in the GPRD, it is a working clinical database and some degree of mis-coding is unavoidable. Patients who are not in contact with general practitioners such as the homeless may be underrepresented in this sample. Our finding that SMI patients consult more often and are more likely to be depressed and prescribed antidepressants before their suicide may be a useful clinical warning sign. In the UK, family practitioners each have approximately 20 patients with SMI on their lists and are being urged to monitor them more vigilantly. Patients with SMI consult their doctors more often than average (Nazareth et al., 1993b) and are usually well known to them. Although suicide remains an unusual event in SMI, the presence of depression and increased consultations in primary care could alert GPs to monitor the extent of suicidal intent. The false negative rate would be high, but over half the suicides in SMI were associated with 2 or more GP consultations over 3 months, compared to the same period in the preceding year. Role of funding source The study was funded by a Trial Platform grant from the UK Medical Research Council. Reference: G0301032. The funder had no further involvement in the execution or writing of the study, or decision to submit for publication. Contributors DO, IN & MK designed the study, wrote the protocol, supervised the analysis, interpreted the results and contributed to the final manuscript. DO wrote the initial draft. GL designed the study, analysed the data, interpreted the findings and contributed to the final manuscript. GL had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Conflict of interest None for all contributing authors. Acknowledgements We are grateful to Dr Irene Petersen and Amir Islam for advice and assistance with data management.
138
D. Osborn et al. / Schizophrenia Research 99 (2008) 134–138
References Department of Health, 1999. A national Service Framework for Mental Health. Department of Health, London. Hawton, K., Sutton, L., Haw, C., et al., 2005. Schizophrenia and suicide: a systematic review of risk factors. British Journal of Psychiatry 187, 9–20. Jick, H., Kaye, J.A., Jick, S.S., 2004. Antidepressants and the risk of suicidal behaviors. JAMA 292, 338–343. Limosin, F., Loze, J., Philippe, A., Casadebaig, F., Rouillon, F., 2007. Ten-year prospective follow-up study of the mortality by suicide in schizophrenic patients. Schizophrenia Research 94 (1–3), 23–28. Morris, R., Carstairs, V., 1991. Which deprivation? A comparison of selected deprivation indices. Journal of Public Health Medicine 13, 318–326.
Nazareth, I., King, M., Haines, A., et al., 1993a. Accuracy of diagnosis of psychosis on general practice computer system. BMJ 307, 32–34. Nazareth, I., King, M., Haines, A., et al., 1993b. Care of schizophrenia in general practice. British Medical Journal 307, 910. Osborn, D.P.J., Levy, G., Nazareth, I., Petersen, I., Islam, A., King, M., 2007. Relative risk of cardiovascular and cancer mortality in people with severe mental illness from the United Kingdom's general practice research database. Archives of General Psychiatry 64, 242–249. Palmer, B.A., Pankratz, V.S., Bostwick, J.M., 2005. The lifetime risk of suicide in schizophrenia: a reexamination. Archives of General Psychiatry 62, 247–253. StataCorp, 2003. Stata Statistical Software: Release 8.2. Stata Corporation, College Station, TX.
Schizophrenia Research 99 (2008) 134 – 138 www.elsevier.com/locate/schres
Suicide and severe mental illnesses. Cohort study within the UK general practice research database David Osborn a,⁎, Gus Levy a , Irwin Nazareth b , Michael King a a
b
Department of Mental Health Sciences, (Hampstead Campus), Royal Free and University College Medical School, Rowland Hill Street, London, NW3 2PF, UK Department of Primary Care and Population Sciences, (Hampstead Campus), Royal Free and University College Medical School, UCL, Rowland Hill Street, London, NW3 2PF, UK Received 19 September 2007; received in revised form 13 November 2007; accepted 18 November 2007 Available online 26 December 2007
Abstract We aimed to evaluate suicide risk across the life-course in severe mental illnesses (SMI) including schizophrenia. Using survival analysis, we compared suicide risk in cohorts of 46,136 people with SMI and 300,426 without. The overall unadjusted hazard ratio (HR) for suicide in SMI was 12.97 (95% CI: 9.75–17.25). The unadjusted HRs differed by age band: 18–30 years: 19.56 (9.76–39.17); 30–50 years: 13.14 (8.64–19.99); 50–70 years: 16.39 (9.15–29.37); 70+: 3.25 (1.33–7.94). In schizophrenia, risk was significantly higher when young but marked risk persisted until age 70. Greatest risk was associated with: increased consultation rates; antidepressant prescriptions and living in less deprived areas. © 2007 Elsevier B.V. All rights reserved. Keywords: Schizophrenia; Bipolar affective disorder; Suicide; Primary care; Risk factors
1. Introduction Suicide is a public health priority in people with severe mental illnesses (SMI) including schizophrenia and bipolar affective disorder (Department of Health, 1999; Hawton et al., 2005). Although it is believed that people with schizophrenia are at greatest risk of suicide early in the illness (Palmer et al., 2005; Limosin et al., 2007) recent reviews have questioned this evidence and have highlighted the lack of data from community ⁎ Corresponding author. Tel.: +44 20 77940500x33950; fax: +44 20 78302808. E-mail addresses: [email protected] (D. Osborn), [email protected] (G. Levy), [email protected] (I. Nazareth), [email protected] (M. King). 0920-9964/$ - see front matter © 2007 Elsevier B.V. All rights reserved. doi:10.1016/j.schres.2007.11.025
samples (Hawton et al., 2005; Palmer et al., 2005). Research on suicide in people with SMI has been limited by relatively small sample sizes; inadequate follow-up periods; few suicide events and sub-optimal statistical analyses. In the absence of an accurate estimation of the burden of suicide and the level of risk throughout the course of illness, we examined in a large community sample the risk of suicide in people with SMI and the demographic and family practice health service predictors of suicide. 2. Methods We undertook a cohort analysis of the risk of suicide in people on the UK general practice research database (GPRD) between 1987 and 2002, aged 18 years and over,
D. Osborn et al. / Schizophrenia Research 99 (2008) 134–138
with and without SMI. We included every patient with a clinical diagnosis of schizophrenia, bipolar affective disorders and other psychotic illnesses including delusional disorders and depression with psychosis. People with drug induced or organic psychoses were excluded. The comparison cohort, without such diagnoses, was randomly selected to achieve a ratio of 1:6, to maximise the statistical power to detect real differences. There were no other exclusion criteria to ensure representativeness. The GPRD contains routine clinical data from 755 general practices representative of those in the UK and has been used for pharmaco-epidemiological studies of suicide outcomes (Jick et al., 2004). In the UK, 95% of the population are registered with a general practice. If people with SMI are not registered with primary care practitioners this is usually organised by their community mental health team. The primary care practices also keep registers of those with SMI. The validity of a GPRD diagnosis of SMI has been confirmed (Nazareth et al., 1993a). Variables must be extracted a priori from the GPRD. We only chose variables which might be related to suicide and which are known to be valid and reliable in the GPRD, which is a working clinical database. For instance, whilst prescriptions and consultation episodes will be recorded accurately in the GPRD, the exact content of a consultation will be inconsistently entered by different practitioners. In deciding which variables to extract for this study we focused on socio-demographic predictors of suicide, consultation frequency prior to suicide and treatment of depression. We obtained data concerning age, sex, SMI diagnosis, General Practice (GP) consultation rate, antidepressant medication, smoking, and social deprivation using the Carstairs index for each general practice, categorised into quintiles. The GPRD did not include accurate data on deprivation or ethnicity at the level of the individual. The Carstairs index of the GP was the only available measure and is a standard index of deprivation that is calculated for each geographical area on four variables. These are: male unemployment; households with no car; overcrowding and head of households in unskilled manual occupation (Morris and Carstairs, 1991). To assess change in consultation behaviour we compared number of GP visits in the final 3 months on the GPRD database with the number of visits made in the same 3 months of the previous year. This comprised the last year of life in the case of suicide, whereas it was the last year on the data base for the remainder. We performed survival analysis using Cox regression techniques in STATA (version 8.2) (StataCorp, 2003). We estimated hazard ratios for suicide firstly comparing people with and without SMI and then restricting the
135
analysis to people with SMI. In a multivariate analysis we adjusted for major available explanatory variables including age, calendar period and geographical social deprivation. We also used smoking as a “proxy covariate” for the multivariate analysis since it is associated with individual deprivation and with poor physical and mental health. A test of proportional hazards was employed to determine whether hazard ratios were stable across ages. The cohort was stratified into age bands to estimate risk of suicide in different periods of life in the three main SMI diagnostic groups. Population attributable risk (PAR) was calculated using the hazard ratio for suicide and choosing the prevalence of SMI in the GPRD at the midpoint of the cohort (1994). 3. Results The cohorts comprised 46,136 people with SMI and 300,426 without SMI. The median follow-up time was 4.7 years in the SMI group, but 4.3 years in the comparison group. The majority of people attracted only one SMI diagnosis during their time in the GPRD (35,097/46,136 = 76.1%). The most common SMI diagnoses were schizophrenia (40.2%), bipolar affective disorder (23.3%), delusional disorder (19.2%), depressive psychosis (5.9%) and schizoaffective disorder (5.3%) (Osborn et al., 2007). This afforded statistical power to examine suicide rates separately in schizophrenia and bipolar affective disorder and the remaining SMI diagnosis was grouped together for further analyses. Where a person had attracted more than one diagnosis we applied two rules. Diagnoses of schizophrenia, schizoaffective disorder or bipolar disorder were given priority over delusional disorder, brief psychotic episodes, unipolar depressive psychoses or non-specified psychoses. Where a diagnosis had changed between schizophrenia, bipolar disorder or schizoaffective disorder over time we utilised the most recent diagnosis. In SMI, the crude rate for GP recorded suicide was far greater than in the comparison group (0.58 vs. 0.047 per 1000 person years), yielding an overall unadjusted hazard ratio of 12.97 (Table 1). This increased risk was observed irrespective of SMI diagnosis, sex or age in the under 70s. In the oldest group, the three-fold excess risk was still significant. After adjustment in the multivariate analysis, hazard ratios for suicide in SMI were smaller in magnitude. The individual variable responsible for this difference in magnitude was GP prescription of antidepressants.
136
D. Osborn et al. / Schizophrenia Research 99 (2008) 134–138
Table 1 Suicide mortality in people with and without severe mental illness Total
Suicides % suicide Crude hazard ratio
Adjusted hazard ratio a
SMI vs. comparison group
Diagnosis
Sex
Age bands
Social deprivation
Within SMI group only Sex
Comparison SMI Comparison Schizophrenia Bipolar disorder Other SMI Female comparison Female SMI Male comparison Male SMI Comparison 18–30 SMI 18–30 Comparison 30–50 SMI 30–50 Comparison 50–70 SMI 50–70 Comparison 70+ SMI 70+ Comparison 20% most deprived SMI 20% most deprived Comparison 20% least deprived SMI 20% least deprived
Female Male Age bands b 18–30 30–50 50–70 70+ Social deprivation quintiles 1 least deprived 20% 2 3 4 5 most deprived 20% Consultation change in final 3 months b2 more visits 2–4 more visits 4 or more visits Antidepressant in last 6 months No Yes
300,246 72 46,136 143 300,246 72 18,555 48 10,742 41 16,839 54 156,781 18 24,608 54 143,465 54 21,528 89 102,653 12 9600 24 134,455 33 20,223 65 88,937 15 17,265 46 50,152 12 11,121 8 63,312 16
0.02 0.31 0.02 0.26 0.38 0.32 0.01 0.22 0.04 0.41 0.01 0.25 0.02 0.32 0.02 0.27 0.02 0.07 0.03
1.00 12.97 1.00 10.99 15.20 13.67 1.00 18.05 1.00 10.96 1.00 19.56 1.00 13.14 1.00 16.39 1.00 3.25 1.00
(9.75, 17.25) (7.61, 15.85) (10.32, 22.37) (9.60, 19.48) (10.49, 31.06) (7.81, 15.37) (9.76, 39.17) (8.64, 19.99) (9.15, 29.37) (1.33, 7.94)
1.00 7.72 1.00 7.00 8.74 7.85 9.82 1.00 6.75 1.00 13.26 1.00 7.80 1.00 9.19 1.00 2.08 1.00
(5.66, 10.51) (4.78, 10.25) (5.80, 13.17) (5.38, 11.46) (5.55, 17.38) (4.65, 9.81) (6.24, 28.15) (4.92, 12.36) (4.93, 17.11) (0.81, 5.31)
12,469 31,825
32 9
0.26 0.03
10.26 (5.60, 18.79) 1.00
5.30 (2.77, 10.15) 1.00
3536
16
0.45
16.25 (7.13, 37.04)
10.52 (4.25, 26.05)
24,608 21,528 9600 20,223 17,265 11,121 3536 5526 8507 9542 12,469 35,162 6361 4613 33,757 12,379
54 89 24 65 46 8 16 24 28 29 32 70 38 35 62 81
0.22 0.41 0.25 0.32 0.27 0.07 0.45 0.43 0.33 0.30 0.26 0.20 0.60 0.76 0.18 0.65
1.00 1.70 1.00 1.05 0.85 0.25 1.00 0.87 0.72 0.63 0.53 1.00 3.53 5.45 1.00 3.27
(1.20, 2.40) (0.66, 1.68) (0.52, 1.39) (0.11, 0.55) (0.46, 1.65) (0.39, 1.33) (0.34, 1.16) (0.29, 0.96) (2.38, 5.24) (3.63, 8.20) (2.35, 4.55)
1.00 1.94 1.00 1.13 0.98 0.30 1.00 0.88 0.73 0.63 0.54 1.00 3.22 4.51 1.00 3.56
(1.37, 2.76) (0.71, 1.81) (0.59, 1.61) (0.14, 0.69) (0.47, 1.66) (0.40, 1.36) (0.34, 1.15) (0.30, 0.99) (2.16, 4.80) (2.98, 6.83) (2.55, 4.98)
a
Adusted for sex, age, calendar period, geographical social deprivation, antidepressant prescription in last 6 months and smoking. b Age bands were compared using a Poisson model as a Cox model could not be estimated due to co-linearity between the age bands. Bold text: 95% CI's do not overlap zero. SMI: Severe mental illness.
Hazard ratios for suicide in different age bands varied with SMI diagnosis (Table 2). In bipolar disorder and other SMI, highest rates occurred in the older age groups, but differences across age groups were not significant. Conversely, the hazard ratios for suicide in schizophrenia decreased significantly with age (test for proportional hazards p = 0.001). Stratifying by age band with the comparison group as the baseline, the suicide HR for
schizophrenia was 22.6 (95% CI 10.1–50.4) in 18–30 year olds; 11.2 (6.6–18.8) in 30–50 year olds and 8.77 (3.9– 19.5) in 50–70 year olds. Of the 48 suicides in people with schizophrenia, 13 (27.1%) occurred in the 18–30 year olds, 25 (52.1%) in the 30–50 year olds, 10 (20.8%) in the 50–70 year olds, and none in the over 70 year olds. In 1994 the overall GPRD prevalence of SMI was 0.87% yielding an unadjusted population attributable
D. Osborn et al. / Schizophrenia Research 99 (2008) 134–138 Table 2 Hazard ratios for suicide in SMI by diagnosis and age group, compared to controls 18–30 years
30–49 years
50–70 years
Schizophrenia 22.6 (10.1–50.4) 11.2 (6.6–8.8) 8.8 (3.9–19.5) Bipolar 14.2 (4.0–50.5) 14.8 (8.2–26.6) 24.8 (12.5–49.4) disorder Other SMI 13.2 (5.4–32.4) 13.5 (7.9–23.0) 21.4 (10.8–42.6) (psychosis) (95% confidence intervals).
risk percent of 9.4%. The PARs decreased over time, from 10.3% in 1988 to 7.7% in 2001. Within the SMI group, increasing age was associated with high relative rates of suicide until 70 after which there was a decreased rate (Table 1). Those completing suicide were more likely to be male and were over three times more likely to have received antidepressants in the last 6 months. In the 3 months prior to suicide, GP consultations increased by a mean of 1.90 (95% CI 1.34–2.47) visits, compared to the same 3 months of the previous year. Higher rates of suicide were associated with lower social deprivation scores. 4. Discussion We report a major excess of suicide deaths in schizophrenia, bipolar disorder and other SMI. Our data indicate that approximately one in ten of the suicides recorded in UK general practice may be attributable to SMI. Although the risk of suicide in schizophrenia relative to the comparison group was highest in people aged under 30, it was still raised eight-fold in 50 to 70 year olds. Seventy-three percent of suicides in schizophrenia occurred in people aged 30 and over, demonstrating that the risk must be assumed throughout the illness. This counters current wisdom that the majority of suicides occur early in schizophrenia and that suicide-prevention strategies are best delivered early in the course of the illness. The relative risk of suicide in other SMI was elevated throughout life, but rose sharply, after 50. While suicide rates are raised across all groups with SMI, those in the least socially deprived areas and those treated for depression are the most at risk. The importance and novelty of these findings lie in the large community nature of the sample, rather than relying on opportunistic follow-up of relatively small numbers of people from less generalisable settings such as inpatient psychiatric units. The GPRD also provided a
137
valid community comparison group and allowed us to control for other explanatory factors. Limitations include the fact that the GPRD lacks validated accurate data regarding social, demographic, economic or psychiatric status at the level of the individual. Data extraction is a major undertaking in the GPRD and we deliberately chose variables which are recorded reliably during clinical practice, to ensure the validity of our results. Furthermore, although recording of causes of death and diagnosis of SMI are validated in the GPRD, it is a working clinical database and some degree of mis-coding is unavoidable. Patients who are not in contact with general practitioners such as the homeless may be underrepresented in this sample. Our finding that SMI patients consult more often and are more likely to be depressed and prescribed antidepressants before their suicide may be a useful clinical warning sign. In the UK, family practitioners each have approximately 20 patients with SMI on their lists and are being urged to monitor them more vigilantly. Patients with SMI consult their doctors more often than average (Nazareth et al., 1993b) and are usually well known to them. Although suicide remains an unusual event in SMI, the presence of depression and increased consultations in primary care could alert GPs to monitor the extent of suicidal intent. The false negative rate would be high, but over half the suicides in SMI were associated with 2 or more GP consultations over 3 months, compared to the same period in the preceding year. Role of funding source The study was funded by a Trial Platform grant from the UK Medical Research Council. Reference: G0301032. The funder had no further involvement in the execution or writing of the study, or decision to submit for publication. Contributors DO, IN & MK designed the study, wrote the protocol, supervised the analysis, interpreted the results and contributed to the final manuscript. DO wrote the initial draft. GL designed the study, analysed the data, interpreted the findings and contributed to the final manuscript. GL had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Conflict of interest None for all contributing authors. Acknowledgements We are grateful to Dr Irene Petersen and Amir Islam for advice and assistance with data management.
138
D. Osborn et al. / Schizophrenia Research 99 (2008) 134–138
References Department of Health, 1999. A national Service Framework for Mental Health. Department of Health, London. Hawton, K., Sutton, L., Haw, C., et al., 2005. Schizophrenia and suicide: a systematic review of risk factors. British Journal of Psychiatry 187, 9–20. Jick, H., Kaye, J.A., Jick, S.S., 2004. Antidepressants and the risk of suicidal behaviors. JAMA 292, 338–343. Limosin, F., Loze, J., Philippe, A., Casadebaig, F., Rouillon, F., 2007. Ten-year prospective follow-up study of the mortality by suicide in schizophrenic patients. Schizophrenia Research 94 (1–3), 23–28. Morris, R., Carstairs, V., 1991. Which deprivation? A comparison of selected deprivation indices. Journal of Public Health Medicine 13, 318–326.
Nazareth, I., King, M., Haines, A., et al., 1993a. Accuracy of diagnosis of psychosis on general practice computer system. BMJ 307, 32–34. Nazareth, I., King, M., Haines, A., et al., 1993b. Care of schizophrenia in general practice. British Medical Journal 307, 910. Osborn, D.P.J., Levy, G., Nazareth, I., Petersen, I., Islam, A., King, M., 2007. Relative risk of cardiovascular and cancer mortality in people with severe mental illness from the United Kingdom's general practice research database. Archives of General Psychiatry 64, 242–249. Palmer, B.A., Pankratz, V.S., Bostwick, J.M., 2005. The lifetime risk of suicide in schizophrenia: a reexamination. Archives of General Psychiatry 62, 247–253. StataCorp, 2003. Stata Statistical Software: Release 8.2. Stata Corporation, College Station, TX.