Identification of cases

The 373 patients who died by suicide within a year of their final hospital admission formed the overall patient group; 75 of these died while in hospital. The cases in this study were the 59 in-patients (79%) for whom case notes could be found. In-patient suicides who died in hospitals outside Wessex were excluded owing to the difficulties in obtaining ethical approval to inspect the notes of these individual patients.

Matching with control patients

Each index patient was matched with two controls using the following sequential criteria: gender, age, diagnostic group, ward type and admission date. Every index patient was matched with at least one control.

Data collection

One of us (J.M.A.S.) extracted demographic, clinical and health-care data from medical case notes onto the specifically designed study pro forma, with additional data (e.g. legal status and observation level) noted for index patients. When not recorded in the case notes, some further information relating to the circumstances of death (e.g. time of day, place and method of death) was derived from coroners' records, obtained as part of an ongoing audit of suicides in Wessex.

Statistical procedures

We performed univariate and multiple regression analyses using conditional logistic regression (Reference CollettCollett, 1991) in STATA 6 (StataCorp, 1999). We identified 60 potential variables for the in-patient deaths and 105 variables for the out-patient deaths, as already described. Stepwise backward elimination of variables, with probability for rejection set at 0.1, was used to produce a parsimonious model. Non-conditional logistic regression was used in separate comparisons of in-patient and out-patient suicides and their controls.

Study sample

Of the 75 patients who died before discharge, 51 suicides (68%) and 24 open verdicts (32%) were returned. Six cases were excluded and in another ten the case notes were unavailable. The notes of five of these cases were located but they were stored in a room contaminated with blue asbestos and therefore were inaccessible. Of the remaining 59 cases, 47 (80%) had two controls and 12 had one identifiable control. There were no differences in age and gender distribution between the 75 in-patient suicides and the 59 cases included in the study, but it was not possible to compare diagnoses.

The only pre-admission factor that differentiated index patients from controls was a history of deliberate self-harm (Table 1). There was no significant difference between the cases and controls in length of illness, number of previous admissions or family history of either mental illness or suicide.

Table 1

Clinical characteristics of cases and matched controls

	Control group		Index group		P	Test
	n=106	%	n=59	%
Length of psychiatric illness	9.9 years		10.3 years		0.248	t-test
Number of admissions prior to index admission
None	36	34	14	24 [UNK]
One	12	11	13	22 [UNK]	0.124	χ2 (2 d.f.)
Two or more	58	55	32	54 [UNK]
Readmitted to different hospital	13	12	12	20	0.180	FE
Readmitted under different consultant	27	25	16	27	0.166	χ2
Readmitted with different diagnosis	8	8	3	5	0.399	FE
Concomitant physical illness	32	30	17	29	0.853	χ2
Family history of completed suicide	7	7	7	12	0.269	χ2
Family history of mental illness	41	40	27	46	0.491	χ2
History of deliberate self-harm	52	49	43	73	0.003	χ2
Additional Axis II disorder	15	14	8	14	0.916	χ2
History of problem drinking	28	26	14	24	0.704	χ2
History of illicit/prescribed substance misuse	18	17	7	12	0.380	χ2

Admission variables

Of the 59 cases, 37 (63%) had been admitted to old ‘mental hospitals’, 21 to district general hospital psychiatric units and 1 to a private hospital. Sixteen (27%) in-patient deaths occurred within the first week of admission, 32 (54%) between 1 week and 3 months, 7 (12%) after 3 months and 4 (7%) after more than 1 year.

Gender differences between index patients

The male/female in-patient suicide ratio was 1.2:1. Significantly more males (13, 40.6%) than females (2, 7.4%) suffered from schizophrenia or a schizophrenia-like illness (Table 2). A lifetime history of deliberate self-harm was present in 25 females (92.6%) and 17 males (53.1%).

Table 2

Diagnoses of 59 in-patient suicides and 234 discharged patient suicides

	In-patient suicides (n=59)						Discharged patient suicides (n=234)
	Male		Female		Total		Male		Female		Total
	n=32	%	n=27	%	n=59	%	n=145	%	n=89	%	n=234	%
Male/female ratio	1.2		1				1.6		1
Schizophrenia and schizophrenia-like disorders	13	40.6	2	7.4	15	25.4	31	21.4	10	11.2	41	17.5
Psychotic affective disorders	6	22.2	6	22.2	12	20.3	26	17.9	24	27.0	50	21.4
Non-psychotic affective disorders	12	37.5	15	55.6	27	45.8	60	41.4	44	49.4	104	44.4
Residual diagnoses	1	3.1	3	11.1	4	6.8	28	19	10	11.2	38	16.2
Other			1	3.7	1	1.7			1	1.1	1	0.43

Circumstances of death

Differences between in-patient and out-patient suicides

Relatively more of the in-patient suicides were women (male/female ratio 1.2:1). The diagnostic distribution is shown in Table 2. In both in-patient and out-patient suicides the majority of patients with schizophrenia and schizophrenia-like disorders were male. Psychotic and nonpsychotic affective disorders were present in a similar proportion of in-patient and out-patient suicides.

Comparison between in-patient suicides and matched controls

Table 3 summarises the 18 variables with a P value of <0.20 in the univariate model and the results of the multiple regression used to produce a parsimonious model. The seven independent variables associated with an increased risk of suicide that remained in the model after backward elimination of variables (with probability for rejection set at 0.1) are shown in Table 4.

Table 3

Factors from univariate analysis (P<0.20) and multiple logistic regression

Factor	Controls	Cases	Univariate			Multivariate
	n=106 (%)	n=59 (%)	Odds ratio	95% CI	P	Odds ratio	95% CI	P
Psychiatric history
History of DSH before admission	52 (49)	42 (71)	2.23	1.15-4.34	0.017	3.37	0.93-12.16	0.063
Reason for admission
Involvement with police	11 (10)	13 (22)	2.78	1.08-7.14	0.034	5.07	0.53-48.22	0.157
Aggression with others	12 (11)	11 (19)	1.97	0.73-5.32	0.193	0.73	0.04-14.01	0.835
Suicidal ideation	37 (35)	31 (52)	2.35	1.14-4.85	0.021	2.64	0.45-15.43	0.281
Loss of social support/social factors	50 (47)	20 (34)	0.61	0.31-1.18	0.142	1.13	0.336-3.553	0.830
Planned admission	13 (12)	3 (5)	0.36	0.10-1.30	0.120	0.63	0.66-5.97	0.685
Admitted under MHA	7 (7)	12 (20)	3.27	1.21-8.84	0.019	328.30	8.34-12 920	0.002
Symptoms elicited and noted at Mental State Examination
Depressive symptoms	65 (61)	45 (76)	2.28	1.03-5.01	0.040	52.37	4.84-567.02	0.001
Hopelessness	16 (15.1)	16 (27)	2.26	0.95-5.306	0.064	0.41	0.08-2.06	0.278
Suicidal ideas	34 (32)	28 (47)	2.36	1.08-5.182	0.032	0.42	0.06-3.02	0.388
Hostility	10 (9)	10 (17)	2.25	0.78-6.47	0.134	1.85	0.04-77.06	0.746
Events during admission
MHA applied	20 (19)	22 (37)	2.97	1.32-6.65	0.008	0.02	0.00-1.04	0.052
Violence to property by patient	4 (4)	9 (15)	4.99	1.33-18.71	0.017	6.26	0.51-76.78	0.151
DSH during admission	7 (7)	12 (20)	3.95	1.37-11.39	0.011	1.26	0.23-7.03	0.790
Threat of self-discharge requiring detention	14 (13)	18 (31)	3.40	1.38-8.41	0.008	62.11	1.03-3722.38	0.048
Patient went AWOL	13 (12)	21 (36)	5.55	2.03-15.12	0.001	22.96	2.31-228.31	0.008
Change of consultant during admission	6 (6)	11 (19)	3.97	1.37-11.54	0.011	14.79	0.99-220.46	0.051
Key professional leaving/on leave	3 (3)	7 (12)	4.13	1.06-16.11	0.041	16.42	1.90-141.76	0.011

Table 4

Variables remaining after backward elimination from multiple logistic regression model with probability for rejection set at P < 0.1

Variable	Controls (valid n=106)		Index (valid n=59)		Odds ratio	95% CI	P
	n	%	n	%
History of DSH before index admission	52	49	42	72	2.59	0.86-7.80	0.090
Admission under MHA	7	7	12	20	49.83	4.97-500.01	0.001
Admitted after involvement with police	11	10	13	22	4.59	0.88-24.07	0.071
Depressive symptoms elicited at admission	65	61	45	76	23.45	3.93-140.13	0.001
Events during admission
Significant professional leaving/on leave	3	3	7	12	17.51	2.68-114.50	0.003
Went AWOL	13	12	21	36	13.05	2.25-75.78	0.004
Showed violence to property	4	4	9	15	10.13	1.54-66.63	0.016

Number of risk factors

Significantly more patients (30, 51%) than controls (12, 11.3%) had three or more of the seven risk factors (Table 5). No social factors were found to discriminate between patients and controls. Admission under the Mental Health Act and depressive symptoms elicited on admission were independently associated with increased risk of suicide.

Table 5

Number of suicide risk factors among index and control in-patients

Number of risk factors	Control (n=106)		Index (n=59)
(maximum=7)	n	%	n	%
0	12	11.3	0
1	46	43.4	7	11.9
2	36	34.0	22	37.3
3	11	10.4	24	40.7
4	1	0.9	4	6.8
5	0		2	3.4
Total	106	100	59	100

Comparison between in-patient and out-patient suicides and controls

Index patients

Admission under the Mental Health Act, going absent without leave or showing violence to property during the final admission occurred significantly more frequently in in-patient than out-patient cases (Table 6).

Table 6

Differences between 59 in-patient suicides and 234 discharged patient suicides and between 106 in-patient suicide controls and 431 discharged in-patient controls (univariate logistic regression)

In-patient v. out-patient	In-patient suicides (n=59)		Out-patient suicides (n=234)		Odds ratio	95% CI	P
	n	%	n	%
Index
History of DSH before admission	42	71.2	157	67.1	1.21	0.65-2.27	0.548
Admission under MHA	12	20.3	20	8.6	2.73	1.25-5.97	0.012
Admitted after involvement with police	13	22.0	31	13.2	1.85	0.90-3.81	0.095
Depressive symptoms at mental state examination	45	76.3	167	71.4	1.41	0.72-2.79	0.311
Key professional leaving/on leave	7	11.9	12	5.1	2.49	0.93-6.63	0.068
Went AWOL	21	35.6	29	12.4	3.91	2.02-7.56	<0.001
Showed violence to property	9	15.3	12	5.1	3.33	1.33-8.33	0.010
Control	In-patient controls (n=106)		Out-patient controls (n=431)
History of DSH before admission	52	49.1	179	41.5	1.36	0.89-2.08	0.162
Admitted after involvement with police	11	10.4	45	10.4	0.99	0.49-1.99	0.985
Admissions under MHA	7	6.6	35	8.1	0.80	0.35-1.85	0.603
Depressive symptoms at mental state examination	65	61.3	275	63.8	0.90	0.58-1.39	0.635
Key professional leaving/on leave	3	2.8	4	0.9	3.11	0.68-4.10	0.142
Went AWOL	13	12.3	37	8.6	1.49	0.76-2.91	0.245
Showed violence to property	4	3.8	19	4.4	0.85	0.28-0.55	0.773

There was no significant difference between in-patient and out-patient suicides in the presence of depressive symptoms at admission (76.3% v. 71.4%) or the absence of a significant health-care professional at the time of the event leading to death (2.8% v. 0.9%). The social factors associated with an altered risk of suicide following discharge do not appear to affect the risk of in-patient suicide (Reference King, Baldwin and SinclairKing et al, 2001) (Table 7).

Table 7

Comparison of independent factors associated with increased suicide risk in in-patient suicides and out-patient suicides: factors from multivariate logistic regression analysis remaining after backward elimination of variables with a P value for rejection of 0.10

Factor	In-patients					Out-patients
	Control n=106 (%)	Index n=59 (%)	Odds ratio	95% CI	P	Control n=431 (%)	Index n=234 (%)	Odds ratio	95% CI	P
Social/demographic
Living alone						135 (31)	95 (41)	1.87	1.19-2.93	0.006
Not White						23 (5)	24 (10)	2.22	1.04-4.75	0.039
Index admission
New1 consultant						231 (54)	147 (63)	1.56	1.01-2.41	0.042
Admitted under MHA	7 (7)	12 (20)	49.83	5.0-500	0.001
Psychiatric history
History of DSH	52 (49)	42 (71)	2.59	0.86-7.8	0.09	179 (44.3)	157 (69.8)	3.61	2.34-5.59	<0.001
Reason for admission
Involved with police	11 (10)	13 (22)	4.59	0.9-24.1	0.071
Symptoms elicited and recorded at admission
Suicidal ideation						160 (39.6)	114 (50.7)	2.0	1.3-3.1	0.002
Hopelessness						58 (14.4)	62 (27.6)	1.8	1.07-14.2	0.026
Depressive features	65 (61)	46 (78)	23.45	3.9-140	0.001
Events after admission
Went AWOL	13 (12)	21 (36)	13.05	2.2-75.8	0.004
Showed violence to property	4 (4)	9 (15)	10.13	1.5-66.7	0.016
Key professional leaving/on leave	3 (3)	7 (12)	17.51	2.7-114	0.003
Loss of job						5 (1.2)	12 (6.3)	7.2	2.0-26.2	0.003
New relationship problem						7 (1.7)	11 (4.9)	4.1	1.2-14.2	0.029
Events post-discharge
Key professional leaving/on leave						4 (1.0)	12 (5.3)	13.9	3.0-63.9	0.001

Study limitations

As described in the accompanying paper (Reference King, Baldwin and SinclairKing et al, 2001), this study has three limitations. The first is the reliance on data collected retrospectively from case notes not intended for research purposes. The second is that the data collection was not performed blind to patient outcome. The final limitation is that many variables were analysed, although these were identified as potential factors a priori, with correction for multiple variables.

Study strengths

However, the study design has several advantages. First, data were collected using a standardised instrument, with operationally defined criteria. The patients were identified through the long-running Wessex Suicide Audit with ready access to coroners' records. Because of its size, data could be limited to a period of 8 years, in contrast to a recent study in which data were derived from admissions over a 30-year period (Reference Powell, Geddes and DeeksPowell et al, 2000).

Comparison with previous investigations

The in-patient suicides in this study are similar to those in previous cohorts, irrespective of country, including New Zealand (Reference Read, Thomas and MellsopRead et al, 1993), Australia (Reference Shah and GanesvaranShah & Ganesvaran, 1997), Canada (Reference Proulx, Lesage and GrunbergProulx et al, 1997) and Ireland (Reference Coakley, Carey and OwensCoakley et al, 1996). There are similarities in the proportion of patients with psychotic illnesses, particularly schizophrenia (Reference Modestin and HoffmannModestin & Hoffmann, 1989; Reference Coakley, Carey and OwensCoakley et al, 1996; Reference Proulx, Lesage and GrunbergProulx et al, 1997; Reference Shah and GanesvaranShah & Ganesvaran, 1997), and in those with a history of deliberate self-harm or requiring compulsory admission (Reference Read, Thomas and MellsopRead et al, 1993). Furthermore, there are similarities in deaths from hanging or associated with moving vehicles, and in the prevalence of deaths occurring away from the ward (Reference Proulx, Lesage and GrunbergProulx et al, 1997; Reference Shah and GanesvaranShah & Ganesvaran, 1997). As such, the findings of this study may be widely applicable.

Comparison with out-patient suicides

There is a relative excess of female in-patient suicides, the male/female ratio being 1.2:1, similar to that found in other studies (Reference Proulx, Lesage and GrunbergProulx et al, 1997; Reference Powell, Geddes and DeeksPowell et al, 2000). Certain risk factors for suicide in in-patients, such as admission under the Mental Health Act, involvement of police, violence towards property and going absent without leave, may reflect greater turmoil than is seen in those patients who kill themselves after discharge. Social factors appear relatively more important in out-patient suicides.

Implications for clinical practice

Six of the seven identified risk factors for in-patient suicide are patient-related. Four of these are factors generally recorded on admission (being admitted under the Mental Health Act, or after involvement with police, the presence of depressive symptoms and a history of deliberate self-harm). The other two, namely display of violence to property and going absent without leave, were recorded during admission.

The six patient-related factors are easily recorded, either as part of the standard admission procedure or in routine ward observations. Some factors are common, such as the presence of depressive symptoms, but others occur less often, for example being admitted under the Mental Health Act and/or via the police. Violence to property and being AWOL are relatively infrequent. When health professionals are about to go on leave it may be advisable to make careful arrangements for continuity of care, particularly if other risk factors are present.

Self-poisoning represents a much lower proportion of in-patient than out-patient suicides. Because none of the in-patients who died from a fatal overdose used currently prescribed medication, this suggests that restricting access to potentially fatal means can reduce the number of drug-related deaths. Restricting the means of self-hanging is more difficult, but optimising the ward environment and patient observation has the potential to make some impact on in-patient deaths.

There was no difference in recorded suicide risk in either patients or controls at admission, but in almost half of the cases no statement of risk had been recorded by staff at their last contact with the patient. Continuing the risk assessment process throughout hospital care may allow potential risk and protective factors to be identified. Although the odds ratios for risk factors are high, they occur in only a minority of patients and thus the sensitivity of individual risk factors is low.

The risk of suicide is present in many psychiatric in-patients. Although the ability to predict suicide accurately is limited, the conscientious and continuing assessment of individual risk and thoughtful policies and procedures may prevent some fatal outcomes (Reference SedererSederer, 1994).

CLINICAL IMPLICATIONS

• ▪ Seven independent risk factors distinguish cases from controls.

• ▪ In-patient and out-patient suicides differ in certain characteristics, the in-patients showing greater signs of personal turmoil.

• ▪ Restricting access to potentially fatal methods may prevent some deaths.

LIMITATIONS

• ▪ Data were collected retrospectively, and not blind to patient outcome.

• ▪ The number of in-patient deaths was not great (n=59) and occurred over 8 years.

• ▪ The findings require confirmation in a prospective study.