Acta Psychiatr Scand 2015: 131: 174–184 All rights reserved DOI: 10.1111/acps.12383
© 2014 John Wiley & Sons A/S. Published by John Wiley & Sons Ltd ACTA PSYCHIATRICA SCANDINAVICA
Meta-analysis
Meta-analysis of suicide rates among psychiatric in-patients Walsh G, Sara G, Ryan CJ, Large M. Meta-analysis of suicide rates among psychiatric in-patients.
G. Walsh1, G. Sara2,3,4,
C. J. Ryan2,5,6, M. Large1,7 1
Objective: To examine factors associated with the number of psychiatric admissions per in-patient suicide and the suicide rate per 100 000 in-patient years in psychiatric hospitals. Method: Random-effects meta-analysis was used to calculate pooled estimates, and meta-regression was used to examine between-sample heterogeneity. Results: Forty-four studies published between 1945 and 2013 reported a total of 7552 in-patient suicides. The pooled estimate of the number of admissions per suicide calculated using 39 studies reporting 150 independent samples was 676 (95% CI: 604–755). Recent studies tended to report higher numbers of admissions per suicide than earlier studies. The pooled estimate of suicide rates per 100 000 in-patient years calculated using 27 studies reporting 95 independent samples was 147 (95% CI: 138–156). Rates of suicide per 100 000 in-patient years tended to be higher in more recent samples, in samples from regions with a higher whole of population suicide rate, in samples from settings with a shorter average length of hospital stay and in studies using coronial records to define suicide. Conclusion: Rates of in-patient suicide in psychiatric hospitals vary remarkably and are disturbingly high. Further research might clarify the extent to which patient factors and the characteristics of in-patient facilities contribute to the unacceptable mortality in psychiatric hospitals.
Mental Health Services, The Prince of Wales Hospital, Sydney, NSW, 2Discipline of Psychiatry, University of Sydney, Sydney, NSW, 3NSW Ministry of Health, Sydney, NSW, 4School of Population Health, University of Queensland, Brisbane, QLD, 5Centre for Values, Ethics and the Law in Medicine, University of Sydney, Sydney, NSW, 6Department of Psychiatry, Westmead Hospital, Westmead, NSW, and 7School of Psychiatry, University of New South Wales, Sydney, NSW, Australia
Key words: suicide; in-patients; psychiatric hospitals Matthew Large, The Kiloh Centre, The Prince of Wales Hospital, Barker Street, Randwick, NSW 2031, Australia. E-mail [email protected]
Accepted for publication November 27, 2014
Summations
• Suicide rates in psychiatric hospitals are a potentially important measure of the performance of mental health services.
• In-patient suicide rates, whether measured by admissions per suicide or by the number of suicides per 100 000 in-patient years, are disturbingly high.
• The reported rates of in-patient suicide vary remarkably. Considerations
• The number of suicides per 100 000 in-patient years appears to have increased in the last three dec• •
174
ades, probably reflecting the increased acuity of a smaller number of admitted patients in the era of deinstitutionalized community mental health care. In-patient suicides may be underreported in studies that do not utilize coronial records to identify cases. Admissions per suicide might prove to be a more important measure of in-patient suicide than suicide rates per 100 000 patient years because it appears to be less influenced by length of stay or rates of suicide in the general community.
Suicide rates among psychiatric in-patients Introduction
Aims of the study
Psychiatric disorders are strongly associated with suicide (1, 2), and patients living with mental illness are often hospitalized because of perceived suicide risk (3, 4). Although one of the aims of hospitalization is to prevent suicide, rates of suicide among those admitted to psychiatric hospitals are greatly elevated when compared to rates of suicide in the community (5, 6). The rate of in-patient suicide varies considerably from institution to institution, and there may be important factors operating at the ward or hospital level influencing suicide rates. To date, however, there is little information about what might constitute a high or low rate of in-patient suicide. One barrier to comparing the incidence of inpatient suicide is that there is no universally accepted way of expressing the rate of inpatient suicide. In-patient suicide is usually reported in one of two ways. Most frequently, it is reported as the proportion of psychiatric admissions that end with a suicide. This figure can also be reported as the number of admissions that can be expected, on average, between in-patient suicides (7–12). In-patient suicide rates may also be reported as the probability that a patient in a particular in-patient bed will suicide in a given year. This measure is usually expressed as the number of suicides per 100 000 in-patient years to avoid reporting the very small annual figures, and it is similar to the way in which community suicide rates are reported (12–14). In any given ward or hospital, the average length of stay provides a mathematical explanation for the relationship between these two rates. A hospital with an average length of stay of one year would have the same proportion of admissions that end in a suicide as it would suicides per in-patient year. Note that if a particular proportion of admissions ended in a suicide, a hospital with a short average length of stay would have more suicides per in-patient year than a hospital with a long length of stay and vice versa. Both methods of reporting reflect the huge variation in in-patient suicide rates observed between units. Recent studies have reported the number of admissions per suicide to be as high as one in 113 admissions (11) or as low as one in 1300 admissions (12). Similarly, suicide rates per 100 000 in-patient years have been reported to vary between 200 and 920 even among similar hospitals in the same year and in the same city (14).
Our primary aim was to calculate pooled estimates of both the number of admissions per suicide and the suicide rate per 100 000 in-patient years. Secondary aims were to ascertain whether the observed heterogeneity in these two rates could be explained by one or more of four independent variables chosen on an a priori basis: the year(s) in which the suicides occurred; the average length of hospital stay; the whole of population suicide rate; and whether coronial records were used to establish suicide numbers. Material and methods
We conducted a meta-analysis of published reports of in-patient suicide in general mental health settings. The methods used were based on the ‘MOOSE’ guidelines for conducting systematic reviews in epidemiology (15). Identification and selection of studies
A systematic review of the literature was conducted to locate English language papers that reported the number of admissions per suicide, the proportion of admissions ending in a suicide or the suicide rate per in-patient year in general psychiatric care settings. Searches were conducted in databases MEDLINE, PsychINFO or EMBASE for papers published between January 1945 and December 2013 using the terms ‘suicide’ or ‘suicides’ or ‘suicidal’ AND ‘inpatient’ or ‘hospital’ or ‘in-patient’. All papers that were likely to meet the inclusion criteria below were examined in full text, and their references were hand-searched for further studies (Fig. 1). Two authors (GW and ML) conducted searches independently. Studies were included if they met all of the following criteria i) The study reported in-patient suicide rates by either: a) reporting the number of suicides and the number of admissions or other data that allowed the calculation of these two figures (e.g. the number of suicides and the proportion of admissions ending in a suicide), or b) reporting the number of suicides and the number of in-patient years or bed years or other data that allowed the calculation of these figures (e.g. the number of admissions and average length of hospital stay or the rate of suicide per in-patient or bed year). ii) The study was conducted in a general mental health setting, defined as a psychiatric ward, a 175
Walsh et al. Search strategies 1. 834 Titles containing (suicide or suicides or suicidal) and (in-patient or hospital or inpatient) in PreMEDLINE/MEDLINE1948 to December 2013, EMBASE 1974 to December 2013, PsycINFO 1967 to December 2013, after the removal of duplicates 2. 32 Additional papers titles identified by hand searches of included studies and previous reviews of in-patient suicide
724 Titles excluded 142 Abstracts reviewed
86 Papers examined in full text
44 Papers met inclusion criteria or reporting data allowing the calculation of either or both of i) The number of admissions per in-patient suicide or ii) The suicide rate per 100 000 in-patient years
56 Abstracts excluded • 24 Non-English language • 11 Reviews or qualitative papers • 7 General hospitals or nonacute psychiatric hospitals • 5 Examined suicide attempts only • 5 Unable to obtain full-text • 4 Included out-patients 42 Papers excluded • 11 No admission or patient year data • 9 Examined specific diagnostic or demographic groups • 8 No suicide data • 5 Duplicate studies • 3 Included outpatients • 4 Non-acute psychiatric units • 1 Included a mix of medical and psychiatric in-patients • 1 Considered only inhospital suicide
group of psychiatric wards, a psychiatric hospital or a group of psychiatric hospitals in a designated geographic area. iii) The study defined in-patient suicide as a suicide of a registered in-patient regardless of whether the suicide occurred on hospital grounds or while on approved or unapproved leave. Studies were excluded if they: i) included data on suicide attempts that could not be separated from suicide data. ii) defined in-patient suicide only as suicide occurring within psychiatric wards. iii) reported on settings other than general psychiatric settings such as medical and surgical wards, prisons and special forensic hospitals. Data collected
Two authors (GW & ML) independently extracted the following data from each study: i) the number of suicides. 176
Fig. 1. Flow chart of searches for studies reporting rates of suicide by registered psychiatric in-patients.
ii) the number of years over which the study was conducted. iii) the number of admissions for the period of the study. iv) the number of in-patient years or the number of available bed years. v) the year or years in which the suicides occurred. If the suicides were collected over more than one year, this value was recorded as the year at the midpoint of the period of data collection, unless the study reported data for multiple years, in which case each year’s data were entered as a single sample. vi) the average length of in-patient stay recorded in days. vii) The rate of suicide in the whole population living in the area in which the study was performed. This was characterized as the national suicide rate in the relevant country in the year(s) in which the data were collected as reported either in the study, or in studies where
Suicide rates among psychiatric in-patients this was not reported, the suicide rate reported for that country, at the time as reported by the World Health Organization (see http://www. who.int/mental_health/media/). This was chosen as a moderator because of the possibility that in-patient suicide rates might reflect community rates. viii) whether suicides were defined using coronial records or hospital records. A study was characterized as having used coronial records to define suicide cases if a national mortality database was used in any way to identify or define suicide cases. This study characteristic was chosen as a moderator because of the possibility that suicide might be underreported in locally held mortality databases. We intended to examine demographic and diagnostic data as independent variables potentially explaining between-study heterogeneity. However, these analyses could not be conducted though because of a paucity of data concerning the characteristics of the in-patients in primary research. Statistical analysis
Pooled estimates of suicide per admission and the number of suicides per 100 000 in-patient years were made using a random-effects meta-analysis using comprehensive meta-analysis (CMA) V3 (Biostat, Engelwood, NJ, USA). A random-effects model was chosen a priori because we did not presume the studies were of similar patient groups or used similar methods and because we were aware of the high degree variation in the rates of suicide between studies. When a study reported more than one sample, for example, over different periods of time or from different hospitals, each sample was included separately, resulting in a larger number of samples than studies. The number of suicides per admission (a small fraction) was converted to the number of admissions per suicide (rounded to the nearest integer) after the analysis. Between-study heterogeneity was assessed using Q-value and I-square statistics with a significance level of P < 0.05. Possible publication bias was assessed with funnel plots of the scatter of effect sizes vs. the inverse variance and was formally tested with an Egger’s linear regression. The hypothetical effect of missing studies was calculated using Duval and Tweedie’s trim-and-fill method (16). We planned to perform a post hoc sensitivity analysis after excluding outlier studies. A method-of-moments meta-regression was used to examine the extent to which heterogeneity in the number of admissions per suicide and the number
of suicides per 100 000 in-patient years was associated with four independent moderator variables that were chosen on an a priori basis: the year(s) in which the suicides occurred; the average length of hospital stay; the whole of population suicide rate; and whether coronial records were used to establish suicide numbers. Pooled estimates were also tabulated according to the era in which the suicides occurred and geographic region in which the studies were performed. Average length of stay was transformed using a natural logarithm because of a highly skewed distribution of values. Moderator variables that were significantly associated with admissions per suicide or suicides per 100 000 in-patient years at P < 0.05 were entered into a random-effects multiple meta-regression model. All significance tests were conducted in a twotailed form. Results Searches
Forty-four studies that met inclusion criteria were located after one study was excluded because it only reported on suicides that occurred within hospital wards (17). The 44 studies included 39 that reported the number of suicides and the number of admissions and 27 that reported either the number of suicides per 100 000 in-patient years or sufficient data for this rate to be calculated. Twenty-two studies reported both types of data (Table 1). Admissions per suicide
Thirty-nine studies reported 6 832 071 admissions in 150 independent samples [mean admissions per sample 45 561, standard deviation (SD) 144 824] of which 6178 ended in a suicide (see Table (SI) SI1 Raw data, SI 2. Forrest plot). The pooled estimate of the number of admissions per suicide was 676 (95% confidence interval (CI) 604–755, I-square = 94.1, range 115–4230, 1st quartile 377, median 729 and 3rd quartile 1128). No clear evidence of publication bias was identified using an examination funnel plots (SI 3 Funnel plot). Egger’s regression intercept was significant (intercept 2.71, df = 148, t-value = 5.2, P < 0.001). Duval and Tweedie’s trim-and-fill method identified one hypothetically missing study and adjusted the pooled estimate of the number of admissions per suicide from 676 to 671. The number of admissions per suicide fell in a stepwise fashion in the periods before 1960, 1960 to 1979, 1980 to 1999, indicating a rising probability of in-patient suicide but rose again after 2000, 177
178
Medlicott & Medlicott (1969) (50) Modestin & Kopp (1988) (51) Modestin & Wurmle (1989) (52) Modestin & Hoffmann (1989) (53)
Matakas & Rohrbach (2007) (49)
Lim (1988, 1991) (46, 47) Lipschutz L (1942) (48) Madsen et al. (2012, 2013) (6, 12)
Langley & Bayatti (1984) (44) Levy & Southcombe (1953) (45)
Li et al. (2008) (43)
Kapur (2013) (42)
Goldney (1985) (38) Gorenc & Bruner (1985) (39) Havaki-Kontaxaki (1994) (40) Hesso (1977) (41)
Farberow (1971) (34) Fernando & Storm (1984) (35) Gaertner (2002) (36) Gale (1980) (14) Ganesvaran & Shah (1997) (37)
Dong (2005) (9)
Corcoran & Walsh (1999) (32) Deisenhammer (2000) (33)
Chatteron (1999) (29) Coakley (1996) (30) Copas & Robin (1982) (31)
Achte et al. (1966) (24) Ajdacic-Gross et al. (2009) (25) Bassett & Tsourtos (1993) (18) Blain & Donaldson (1995) (26) Brecic (2009) (27) Chapman (1965) (28)
Study
Eight hospitals, Helsinki, Finland Zurich Canton, Switzerland Queen Elizabeth Hospital, Woodville, Australia Seven psychiatric hospitals in Northern England Vrapce Psychiatric Hospital, Zagreb, Croatia Topeka Veterans Administration Hospital, Topeka, Kansas, USA Unidentified Hospital, Queensland, Australia Cavan/Monaghan Mental Health Service, Ireland All psychiatric wards and hospitals in England and Wales All psychiatric wards and hospitals in Ireland Innsbruck University Hospital, Psychiatric State Hospital of Tyrol, Austria All public psychiatric wards and hospitals in Hong Kong, China Multiple Veteran Administration Hospitals, USA Chase Farm Hospital, Middlesex, England Tuebingen University Hospital, Germany Five state hospitals, New York City, USA North Eastern Metropolitan Psychiatric Services, Melbourne, Victoria, Australia Glenside Hospital, South Australia, Australia Ten Bavarian Psychiatric Hospitals, Germany Psychiatric Hospital of Attica, Athens, Greece All Norwegian, Swedish, and Finnish psychiatric hospitals All psychiatric wards and hospitals in England and Wales Guangzhou Psychiatric Hospital, Guangzhou, China Exe Vale Hospital, North Devon, England Eastern State Hospital, Medical Lake, Washington, USA Woodbridge Hospital, Singapore Eloise Hospital, Eloise, Michigan, USA All psychiatric wards and hospitals in Denmark Cologne Psychiatric Hospital, Cologne, Germany Ashburn Hall, Dunedin, New Zealand Two psychiatric hospitals, Switzerland Two psychiatric hospitals, Switzerland Hospitals in the Canton of Berne, Switzerland
Location
Table 1. Studies reporting rates of psychiatric in-patient suicide published in English (1945–2013)
1882–1968 1960–1981 1977–1986 1971–1981
1984–2005
1985–1986 1929–1941 1997–2006
1972–1981 1891–1945
1956–2005
1997–2008
1972–1982 1950–1976 1959–1987 1930–1974
1959–1966 1976–1981 1965–1999 1975–1977 1973–1993
1997–1999
1983–1992 1987–1994
1996–2006 1974–1983 1967–1973
1952–1963 1992–2004 1980–1991 1975–1985 1996–2006 1946–1962
Period
Hospital records Police reports Police reports Police reports
Coronial records
Coronial records Hospital records Coronial records
Hospital records Hospital records
Hospital records
Coronial records
Hospital and coronial records Hospital records Hospital and coronial records Coronial records.
Hospital Records Coronial records Hospital records Hospital records Hospital records
Coronial records
Hospital records Police records
Hospital records Hospital and coronial records Coronial records
Hospital and coronial records Psychiatric case register Hospital records Coronial records Hospital records Hospital records
Suicide ascertainment
12 149 115 49
19
12 18 219
40 58
77
1942
21 306 32 1506
650 8 61 60 103
93
146 44
18 8 696
48 141 9 50 25 18
Suicides N.
390 294 289 221
634
336 653 1314
262
751
1139
986 1687 906 526
1401 260 444 531 272
371
1849 764
171 2063 2004
3032 1278
1015 619 421
Admissions per suicide
267 1045
2159
261 42 1067
40
98
588
382 60 11 109
462
174
9
322 1066 10 490 238
Suicides per 100 000 patient years
Walsh et al.
100 20 Hospital and coronial records 1967–1987
19 Hospital records 1930–1938
496 1970–1981
Hospital records
285
844
3631 27 248 1052 59 30 Coronial records Hospital and coronial records 1967–1992 1949–1951
412 79 Coronial records 1984–1993
1312 97 30 1960–1970 1989–1999
Hospital records Hospital records
413 27 1984–1989
Coronial records
730 Coronial records 1963–1992
112
875 551
828 1389 41 Hospital records
Zain (1991) (63)
Woolley & Eichert (1941) (62)
Wolfersdorf (1988) (61)
Taiminen & Helenius (1994) (59) Temoche (1964) (60)
Steblaj (1999) (58)
Sletten (1972) (57) Spiessl (2002) (11)
Read (1993) (56)
Powell (2000) (55)
Neuner (2008, 2011) (10, 54)
Psychiatric University Hospital Regensburg, Germany Psychiatric hospitals in Oxfordshire, Berkshire, Buckinghamshire, and Northamptonshire, England Wellington Area Health Board, Wellington, New Zealand Five Missouri state hospitals, USA Psychiatric State Hospital, Regensburg, Germany University Psychiatric Hospital, Ljubljana, Slovenia The Psychiatric Clinic of Turku, Finland All psychiatric wards and hospitals in Massachusetts, USA Four State Mental Hospitals of Baden-Wurttemberg, FRG Sheppard and Enoch Pratt Hospital, Baltimore, Maryland, USA University Hospital Kuala Lumpur, Malaysia
1995–2004
Admissions per suicide Study
Table 1. (Continued)
Location
Period
Suicide ascertainment
Suicides N.
Suicides per 100 000 patient years
Suicide rates among psychiatric in-patients indicating a falling probability of in-patient suicide (Table 2). Studies from Australasia reported the lowest number of admissions per suicide followed by Continental Europe, USA, Nordic countries and the UK and Ireland (Table 3). Random-effects meta-regression suggested an association between the year the suicides occurred and the number of admissions per suicide, such that the earlier the suicides occurred, the higher the probability of an admitted patient committing suicide (Table 4). However, it was apparent that the association between year and admissions per suicide was not linear, with numbers of admissions falling and then rising (Table 2). The year of the suicide accounted for
© 2014 John Wiley & Sons A/S. Published by John Wiley & Sons Ltd ACTA PSYCHIATRICA SCANDINAVICA
Meta-analysis
Meta-analysis of suicide rates among psychiatric in-patients Walsh G, Sara G, Ryan CJ, Large M. Meta-analysis of suicide rates among psychiatric in-patients.
G. Walsh1, G. Sara2,3,4,
C. J. Ryan2,5,6, M. Large1,7 1
Objective: To examine factors associated with the number of psychiatric admissions per in-patient suicide and the suicide rate per 100 000 in-patient years in psychiatric hospitals. Method: Random-effects meta-analysis was used to calculate pooled estimates, and meta-regression was used to examine between-sample heterogeneity. Results: Forty-four studies published between 1945 and 2013 reported a total of 7552 in-patient suicides. The pooled estimate of the number of admissions per suicide calculated using 39 studies reporting 150 independent samples was 676 (95% CI: 604–755). Recent studies tended to report higher numbers of admissions per suicide than earlier studies. The pooled estimate of suicide rates per 100 000 in-patient years calculated using 27 studies reporting 95 independent samples was 147 (95% CI: 138–156). Rates of suicide per 100 000 in-patient years tended to be higher in more recent samples, in samples from regions with a higher whole of population suicide rate, in samples from settings with a shorter average length of hospital stay and in studies using coronial records to define suicide. Conclusion: Rates of in-patient suicide in psychiatric hospitals vary remarkably and are disturbingly high. Further research might clarify the extent to which patient factors and the characteristics of in-patient facilities contribute to the unacceptable mortality in psychiatric hospitals.
Mental Health Services, The Prince of Wales Hospital, Sydney, NSW, 2Discipline of Psychiatry, University of Sydney, Sydney, NSW, 3NSW Ministry of Health, Sydney, NSW, 4School of Population Health, University of Queensland, Brisbane, QLD, 5Centre for Values, Ethics and the Law in Medicine, University of Sydney, Sydney, NSW, 6Department of Psychiatry, Westmead Hospital, Westmead, NSW, and 7School of Psychiatry, University of New South Wales, Sydney, NSW, Australia
Key words: suicide; in-patients; psychiatric hospitals Matthew Large, The Kiloh Centre, The Prince of Wales Hospital, Barker Street, Randwick, NSW 2031, Australia. E-mail [email protected]
Accepted for publication November 27, 2014
Summations
• Suicide rates in psychiatric hospitals are a potentially important measure of the performance of mental health services.
• In-patient suicide rates, whether measured by admissions per suicide or by the number of suicides per 100 000 in-patient years, are disturbingly high.
• The reported rates of in-patient suicide vary remarkably. Considerations
• The number of suicides per 100 000 in-patient years appears to have increased in the last three dec• •
174
ades, probably reflecting the increased acuity of a smaller number of admitted patients in the era of deinstitutionalized community mental health care. In-patient suicides may be underreported in studies that do not utilize coronial records to identify cases. Admissions per suicide might prove to be a more important measure of in-patient suicide than suicide rates per 100 000 patient years because it appears to be less influenced by length of stay or rates of suicide in the general community.
Suicide rates among psychiatric in-patients Introduction
Aims of the study
Psychiatric disorders are strongly associated with suicide (1, 2), and patients living with mental illness are often hospitalized because of perceived suicide risk (3, 4). Although one of the aims of hospitalization is to prevent suicide, rates of suicide among those admitted to psychiatric hospitals are greatly elevated when compared to rates of suicide in the community (5, 6). The rate of in-patient suicide varies considerably from institution to institution, and there may be important factors operating at the ward or hospital level influencing suicide rates. To date, however, there is little information about what might constitute a high or low rate of in-patient suicide. One barrier to comparing the incidence of inpatient suicide is that there is no universally accepted way of expressing the rate of inpatient suicide. In-patient suicide is usually reported in one of two ways. Most frequently, it is reported as the proportion of psychiatric admissions that end with a suicide. This figure can also be reported as the number of admissions that can be expected, on average, between in-patient suicides (7–12). In-patient suicide rates may also be reported as the probability that a patient in a particular in-patient bed will suicide in a given year. This measure is usually expressed as the number of suicides per 100 000 in-patient years to avoid reporting the very small annual figures, and it is similar to the way in which community suicide rates are reported (12–14). In any given ward or hospital, the average length of stay provides a mathematical explanation for the relationship between these two rates. A hospital with an average length of stay of one year would have the same proportion of admissions that end in a suicide as it would suicides per in-patient year. Note that if a particular proportion of admissions ended in a suicide, a hospital with a short average length of stay would have more suicides per in-patient year than a hospital with a long length of stay and vice versa. Both methods of reporting reflect the huge variation in in-patient suicide rates observed between units. Recent studies have reported the number of admissions per suicide to be as high as one in 113 admissions (11) or as low as one in 1300 admissions (12). Similarly, suicide rates per 100 000 in-patient years have been reported to vary between 200 and 920 even among similar hospitals in the same year and in the same city (14).
Our primary aim was to calculate pooled estimates of both the number of admissions per suicide and the suicide rate per 100 000 in-patient years. Secondary aims were to ascertain whether the observed heterogeneity in these two rates could be explained by one or more of four independent variables chosen on an a priori basis: the year(s) in which the suicides occurred; the average length of hospital stay; the whole of population suicide rate; and whether coronial records were used to establish suicide numbers. Material and methods
We conducted a meta-analysis of published reports of in-patient suicide in general mental health settings. The methods used were based on the ‘MOOSE’ guidelines for conducting systematic reviews in epidemiology (15). Identification and selection of studies
A systematic review of the literature was conducted to locate English language papers that reported the number of admissions per suicide, the proportion of admissions ending in a suicide or the suicide rate per in-patient year in general psychiatric care settings. Searches were conducted in databases MEDLINE, PsychINFO or EMBASE for papers published between January 1945 and December 2013 using the terms ‘suicide’ or ‘suicides’ or ‘suicidal’ AND ‘inpatient’ or ‘hospital’ or ‘in-patient’. All papers that were likely to meet the inclusion criteria below were examined in full text, and their references were hand-searched for further studies (Fig. 1). Two authors (GW and ML) conducted searches independently. Studies were included if they met all of the following criteria i) The study reported in-patient suicide rates by either: a) reporting the number of suicides and the number of admissions or other data that allowed the calculation of these two figures (e.g. the number of suicides and the proportion of admissions ending in a suicide), or b) reporting the number of suicides and the number of in-patient years or bed years or other data that allowed the calculation of these figures (e.g. the number of admissions and average length of hospital stay or the rate of suicide per in-patient or bed year). ii) The study was conducted in a general mental health setting, defined as a psychiatric ward, a 175
Walsh et al. Search strategies 1. 834 Titles containing (suicide or suicides or suicidal) and (in-patient or hospital or inpatient) in PreMEDLINE/MEDLINE1948 to December 2013, EMBASE 1974 to December 2013, PsycINFO 1967 to December 2013, after the removal of duplicates 2. 32 Additional papers titles identified by hand searches of included studies and previous reviews of in-patient suicide
724 Titles excluded 142 Abstracts reviewed
86 Papers examined in full text
44 Papers met inclusion criteria or reporting data allowing the calculation of either or both of i) The number of admissions per in-patient suicide or ii) The suicide rate per 100 000 in-patient years
56 Abstracts excluded • 24 Non-English language • 11 Reviews or qualitative papers • 7 General hospitals or nonacute psychiatric hospitals • 5 Examined suicide attempts only • 5 Unable to obtain full-text • 4 Included out-patients 42 Papers excluded • 11 No admission or patient year data • 9 Examined specific diagnostic or demographic groups • 8 No suicide data • 5 Duplicate studies • 3 Included outpatients • 4 Non-acute psychiatric units • 1 Included a mix of medical and psychiatric in-patients • 1 Considered only inhospital suicide
group of psychiatric wards, a psychiatric hospital or a group of psychiatric hospitals in a designated geographic area. iii) The study defined in-patient suicide as a suicide of a registered in-patient regardless of whether the suicide occurred on hospital grounds or while on approved or unapproved leave. Studies were excluded if they: i) included data on suicide attempts that could not be separated from suicide data. ii) defined in-patient suicide only as suicide occurring within psychiatric wards. iii) reported on settings other than general psychiatric settings such as medical and surgical wards, prisons and special forensic hospitals. Data collected
Two authors (GW & ML) independently extracted the following data from each study: i) the number of suicides. 176
Fig. 1. Flow chart of searches for studies reporting rates of suicide by registered psychiatric in-patients.
ii) the number of years over which the study was conducted. iii) the number of admissions for the period of the study. iv) the number of in-patient years or the number of available bed years. v) the year or years in which the suicides occurred. If the suicides were collected over more than one year, this value was recorded as the year at the midpoint of the period of data collection, unless the study reported data for multiple years, in which case each year’s data were entered as a single sample. vi) the average length of in-patient stay recorded in days. vii) The rate of suicide in the whole population living in the area in which the study was performed. This was characterized as the national suicide rate in the relevant country in the year(s) in which the data were collected as reported either in the study, or in studies where
Suicide rates among psychiatric in-patients this was not reported, the suicide rate reported for that country, at the time as reported by the World Health Organization (see http://www. who.int/mental_health/media/). This was chosen as a moderator because of the possibility that in-patient suicide rates might reflect community rates. viii) whether suicides were defined using coronial records or hospital records. A study was characterized as having used coronial records to define suicide cases if a national mortality database was used in any way to identify or define suicide cases. This study characteristic was chosen as a moderator because of the possibility that suicide might be underreported in locally held mortality databases. We intended to examine demographic and diagnostic data as independent variables potentially explaining between-study heterogeneity. However, these analyses could not be conducted though because of a paucity of data concerning the characteristics of the in-patients in primary research. Statistical analysis
Pooled estimates of suicide per admission and the number of suicides per 100 000 in-patient years were made using a random-effects meta-analysis using comprehensive meta-analysis (CMA) V3 (Biostat, Engelwood, NJ, USA). A random-effects model was chosen a priori because we did not presume the studies were of similar patient groups or used similar methods and because we were aware of the high degree variation in the rates of suicide between studies. When a study reported more than one sample, for example, over different periods of time or from different hospitals, each sample was included separately, resulting in a larger number of samples than studies. The number of suicides per admission (a small fraction) was converted to the number of admissions per suicide (rounded to the nearest integer) after the analysis. Between-study heterogeneity was assessed using Q-value and I-square statistics with a significance level of P < 0.05. Possible publication bias was assessed with funnel plots of the scatter of effect sizes vs. the inverse variance and was formally tested with an Egger’s linear regression. The hypothetical effect of missing studies was calculated using Duval and Tweedie’s trim-and-fill method (16). We planned to perform a post hoc sensitivity analysis after excluding outlier studies. A method-of-moments meta-regression was used to examine the extent to which heterogeneity in the number of admissions per suicide and the number
of suicides per 100 000 in-patient years was associated with four independent moderator variables that were chosen on an a priori basis: the year(s) in which the suicides occurred; the average length of hospital stay; the whole of population suicide rate; and whether coronial records were used to establish suicide numbers. Pooled estimates were also tabulated according to the era in which the suicides occurred and geographic region in which the studies were performed. Average length of stay was transformed using a natural logarithm because of a highly skewed distribution of values. Moderator variables that were significantly associated with admissions per suicide or suicides per 100 000 in-patient years at P < 0.05 were entered into a random-effects multiple meta-regression model. All significance tests were conducted in a twotailed form. Results Searches
Forty-four studies that met inclusion criteria were located after one study was excluded because it only reported on suicides that occurred within hospital wards (17). The 44 studies included 39 that reported the number of suicides and the number of admissions and 27 that reported either the number of suicides per 100 000 in-patient years or sufficient data for this rate to be calculated. Twenty-two studies reported both types of data (Table 1). Admissions per suicide
Thirty-nine studies reported 6 832 071 admissions in 150 independent samples [mean admissions per sample 45 561, standard deviation (SD) 144 824] of which 6178 ended in a suicide (see Table (SI) SI1 Raw data, SI 2. Forrest plot). The pooled estimate of the number of admissions per suicide was 676 (95% confidence interval (CI) 604–755, I-square = 94.1, range 115–4230, 1st quartile 377, median 729 and 3rd quartile 1128). No clear evidence of publication bias was identified using an examination funnel plots (SI 3 Funnel plot). Egger’s regression intercept was significant (intercept 2.71, df = 148, t-value = 5.2, P < 0.001). Duval and Tweedie’s trim-and-fill method identified one hypothetically missing study and adjusted the pooled estimate of the number of admissions per suicide from 676 to 671. The number of admissions per suicide fell in a stepwise fashion in the periods before 1960, 1960 to 1979, 1980 to 1999, indicating a rising probability of in-patient suicide but rose again after 2000, 177
178
Medlicott & Medlicott (1969) (50) Modestin & Kopp (1988) (51) Modestin & Wurmle (1989) (52) Modestin & Hoffmann (1989) (53)
Matakas & Rohrbach (2007) (49)
Lim (1988, 1991) (46, 47) Lipschutz L (1942) (48) Madsen et al. (2012, 2013) (6, 12)
Langley & Bayatti (1984) (44) Levy & Southcombe (1953) (45)
Li et al. (2008) (43)
Kapur (2013) (42)
Goldney (1985) (38) Gorenc & Bruner (1985) (39) Havaki-Kontaxaki (1994) (40) Hesso (1977) (41)
Farberow (1971) (34) Fernando & Storm (1984) (35) Gaertner (2002) (36) Gale (1980) (14) Ganesvaran & Shah (1997) (37)
Dong (2005) (9)
Corcoran & Walsh (1999) (32) Deisenhammer (2000) (33)
Chatteron (1999) (29) Coakley (1996) (30) Copas & Robin (1982) (31)
Achte et al. (1966) (24) Ajdacic-Gross et al. (2009) (25) Bassett & Tsourtos (1993) (18) Blain & Donaldson (1995) (26) Brecic (2009) (27) Chapman (1965) (28)
Study
Eight hospitals, Helsinki, Finland Zurich Canton, Switzerland Queen Elizabeth Hospital, Woodville, Australia Seven psychiatric hospitals in Northern England Vrapce Psychiatric Hospital, Zagreb, Croatia Topeka Veterans Administration Hospital, Topeka, Kansas, USA Unidentified Hospital, Queensland, Australia Cavan/Monaghan Mental Health Service, Ireland All psychiatric wards and hospitals in England and Wales All psychiatric wards and hospitals in Ireland Innsbruck University Hospital, Psychiatric State Hospital of Tyrol, Austria All public psychiatric wards and hospitals in Hong Kong, China Multiple Veteran Administration Hospitals, USA Chase Farm Hospital, Middlesex, England Tuebingen University Hospital, Germany Five state hospitals, New York City, USA North Eastern Metropolitan Psychiatric Services, Melbourne, Victoria, Australia Glenside Hospital, South Australia, Australia Ten Bavarian Psychiatric Hospitals, Germany Psychiatric Hospital of Attica, Athens, Greece All Norwegian, Swedish, and Finnish psychiatric hospitals All psychiatric wards and hospitals in England and Wales Guangzhou Psychiatric Hospital, Guangzhou, China Exe Vale Hospital, North Devon, England Eastern State Hospital, Medical Lake, Washington, USA Woodbridge Hospital, Singapore Eloise Hospital, Eloise, Michigan, USA All psychiatric wards and hospitals in Denmark Cologne Psychiatric Hospital, Cologne, Germany Ashburn Hall, Dunedin, New Zealand Two psychiatric hospitals, Switzerland Two psychiatric hospitals, Switzerland Hospitals in the Canton of Berne, Switzerland
Location
Table 1. Studies reporting rates of psychiatric in-patient suicide published in English (1945–2013)
1882–1968 1960–1981 1977–1986 1971–1981
1984–2005
1985–1986 1929–1941 1997–2006
1972–1981 1891–1945
1956–2005
1997–2008
1972–1982 1950–1976 1959–1987 1930–1974
1959–1966 1976–1981 1965–1999 1975–1977 1973–1993
1997–1999
1983–1992 1987–1994
1996–2006 1974–1983 1967–1973
1952–1963 1992–2004 1980–1991 1975–1985 1996–2006 1946–1962
Period
Hospital records Police reports Police reports Police reports
Coronial records
Coronial records Hospital records Coronial records
Hospital records Hospital records
Hospital records
Coronial records
Hospital and coronial records Hospital records Hospital and coronial records Coronial records.
Hospital Records Coronial records Hospital records Hospital records Hospital records
Coronial records
Hospital records Police records
Hospital records Hospital and coronial records Coronial records
Hospital and coronial records Psychiatric case register Hospital records Coronial records Hospital records Hospital records
Suicide ascertainment
12 149 115 49
19
12 18 219
40 58
77
1942
21 306 32 1506
650 8 61 60 103
93
146 44
18 8 696
48 141 9 50 25 18
Suicides N.
390 294 289 221
634
336 653 1314
262
751
1139
986 1687 906 526
1401 260 444 531 272
371
1849 764
171 2063 2004
3032 1278
1015 619 421
Admissions per suicide
267 1045
2159
261 42 1067
40
98
588
382 60 11 109
462
174
9
322 1066 10 490 238
Suicides per 100 000 patient years
Walsh et al.
100 20 Hospital and coronial records 1967–1987
19 Hospital records 1930–1938
496 1970–1981
Hospital records
285
844
3631 27 248 1052 59 30 Coronial records Hospital and coronial records 1967–1992 1949–1951
412 79 Coronial records 1984–1993
1312 97 30 1960–1970 1989–1999
Hospital records Hospital records
413 27 1984–1989
Coronial records
730 Coronial records 1963–1992
112
875 551
828 1389 41 Hospital records
Zain (1991) (63)
Woolley & Eichert (1941) (62)
Wolfersdorf (1988) (61)
Taiminen & Helenius (1994) (59) Temoche (1964) (60)
Steblaj (1999) (58)
Sletten (1972) (57) Spiessl (2002) (11)
Read (1993) (56)
Powell (2000) (55)
Neuner (2008, 2011) (10, 54)
Psychiatric University Hospital Regensburg, Germany Psychiatric hospitals in Oxfordshire, Berkshire, Buckinghamshire, and Northamptonshire, England Wellington Area Health Board, Wellington, New Zealand Five Missouri state hospitals, USA Psychiatric State Hospital, Regensburg, Germany University Psychiatric Hospital, Ljubljana, Slovenia The Psychiatric Clinic of Turku, Finland All psychiatric wards and hospitals in Massachusetts, USA Four State Mental Hospitals of Baden-Wurttemberg, FRG Sheppard and Enoch Pratt Hospital, Baltimore, Maryland, USA University Hospital Kuala Lumpur, Malaysia
1995–2004
Admissions per suicide Study
Table 1. (Continued)
Location
Period
Suicide ascertainment
Suicides N.
Suicides per 100 000 patient years
Suicide rates among psychiatric in-patients indicating a falling probability of in-patient suicide (Table 2). Studies from Australasia reported the lowest number of admissions per suicide followed by Continental Europe, USA, Nordic countries and the UK and Ireland (Table 3). Random-effects meta-regression suggested an association between the year the suicides occurred and the number of admissions per suicide, such that the earlier the suicides occurred, the higher the probability of an admitted patient committing suicide (Table 4). However, it was apparent that the association between year and admissions per suicide was not linear, with numbers of admissions falling and then rising (Table 2). The year of the suicide accounted for
