Risk Factors for Deliberate Self-Harm and Completed Suicide in Young Chinese People with Schizophrenia

Subjects

In four consecutive years from 1 July 2002 to 30 June 2006, all patients who entered the early intervention programme with a chart diagnosis of schizophrenia in the computerized case register were recruited as subjects. The diagnosis of schizophrenia was confirmed using ICD-10 criteria, and further ascertained by consensus with an experienced consultant psychiatrist. Subjects with other principal psychiatric diagnoses were excluded.

Data collection

In the period between January and March 2007, case notes and discharge summaries of the subjects were reviewed. The follow-up period was calculated from the dates of the first and the last outpatient attendance prior to the time of data collection. Subjects were regarded as having completed suicide if the following two criteria were satisfied: (i) they were identified as dead at the end of the study period; and (ii) the cause of death was self-inflicted. The coroner's verdict was not used because it would have underestimated completed suicide cases [6–8]. Due to the fact that suicide intent is very difficult to prove, it was not included in the criteria of completed suicide in the present study. The nomenclature of deliberate self-harm was defined in accordance with guidelines of the World Health Organization/European Study on Parasuicide: ‘an act with non-fatal outcome, in which an individual deliberately initiates a non-habitual behaviour that, without intervention from others, will cause self-harm, or deliberately ingests a substance in excess of the prescribed or generally recognized therapeutic dosage, and which aimed at realizing changes which the subject desired via the actual or expected physical consequences’ [9]. The status of completed suicide and deliberate self-harm was ascertained by case-note review and computerized clinical data, and further clarified by interviewing treating staff. A proforma data retrieval form was used. Data concerning a total of 51 pre-defined independent variables (Appendix I) were collected.

Statistical analysis

Subjects were categorized into event groups and a censored group. Two event groups (completed suicide group and combined group of completed suicide and deliberate self-harm) were identified. The reason for combining completed suicide and deliberate self-harm cases was because a ‘pure deliberate self-harm’ group excluding completed suicide cases could cause selection bias and underestimation of suicidal behaviour.

The time to event, that is, the duration of time from the first attendance to the occurrence of either of the two outcome events, was calculated.

Statistical analysis was performed using SPSS for Windows version 14 (SPSS, Chicago, IL, USA). Univariate Cox proportional hazard model was used to assess the effect of 51 independent variables on the occurrence of dependent variables. Independent variables with p<0.05 were identified as risk factors for suicide. Multivariate Cox proportional hazard model was used to account for interrelationships between risk factors. PASS (version 2002) was used to calculate the post-hoc power for those variables that had been regarded as risk factors in the literature but were found to have insignificant p in the present study. Post-hoc power could estimate the size of Type II error for a particular independent variable.

Sample characteristics

All subjects were Chinese in ethnicity. The male: female ratio was 1:0.8. The mean age was 21.9 years at the time of entering EASY. A total of 90.6% of subjects (n=212) in the cohort were single, 51.3% (n=120) were unemployed, 22.6% (n=53) were students, and 94.0% (n=220) were living with their families. Apart from having a diagnosis of schizophrenia, six subjects had a secondary diagnosis of mild mental retardation, three had borderline intelligence, and two had borderline personality disorder. The prevalence of substance misuse in the cohort was 21.4% (n=50). A total of 52.6% (n=123) of subjects had been admitted to a psychiatric hospital during an average of 2.6 years of follow up; 46.2% (n=108) of subjects were documented to have non-compliance or non-attendance for >1 month.

Rates of suicidal behaviour

Forty-eight subjects (20.5%) had a past history of deliberate self-harm prior to entering EASY. After an average of 2.6 years of follow up, 27.8% (n=65) of subjects had at least one episode of suicidal behaviour. The occurrence rate for suicidal behaviour (n=33) in EASY was 14.1%.

Risk factors for suicidal behaviour

Table 1 shows the comparative analysis of sociodemographic, extrapyramidal side-effect, and ‘level of care’ variables. Subjects with suicidal behaviour tended to have a history of substance abuse (p=0.07; power=0.44, which was inadequate to reject the null hypothesis). Table 2 shows the medications for the two groups of subjects prior to the outcome events or at the end of the follow-up period. The 11 risk factors identified in univariate Cox proportional hazard model are shown in Table 3.

OPEN IN VIEWER

Table 1.

Comparative analysis on sociodemographic, EPSE, and level of care variables

Independent variable	Suicidal behaviour (n=33) (%)	Non-suicide (n=201) (%)	p
Age (years) entering EASY (±SD)	21.1±4.0	21.1±4.1	0.20†
Male: female ratio	57.1	54.2	0.72‡
Marital Status: single	93.9	90.0	0.75§
Unemployed	57.6	50.2	0.71¶
Education level: above Form 5 (>11 years)	40.0	60.9	0.15¶
Accommodation: public housing	63.6	67.2	0.42¶
Positive family history of suicide	18.2	10.4	0.32¶
Positive history of substance abuse	33.3	19.4	0.07‡
Positive drinking history	27.3	18.9	0.38¶
DUP (months)	17.2	21.1	0.43†
Akathisia	6.1	5.5	0.99§
EPSE (akathisia excluded)	15.2	12.4	0.88¶
Input of clinical psychologist	33.3	22.4	0.17‡
Input of occupational therapist	39.4	31.8	0.39‡
Input of psychiatric day hospital	12.2	6.5	0.27§

DUP, duration of untreated psychosis; EASY, Early Assessment Service for Young People with Psychosis; PFU, priority follow up; EPSE, extrapyramidal side-effect.

^†t-test; ^‡χ² test; ^§Fisher's exact test; ^¶χ² test with Yates correction.

OPEN IN VIEWER

Table 2.

Comparative analysis on medication variables

Medications at the end of follow up or event	Suicidal behaviour (n=33) (%)	Non-suicide (n=201) (%)	p
Conventional antipsychotics	42.4	28.9	0.12†
Atypical antipsychotics (without clozapine)	33.3	54.2	0.03†
Long-acting injectable antipsychotics	6.1	10.0	0.75‡
Mood stabilizers	3.0	10.9	0.22‡
Antidepressants	15.2	12.9	0.94§
Clozapine	3.0	1.5	0.46‡

Suicidal behaviour: combined group of completed suicide and deliberate self-harm.

^†χ² test; ^‡Fisher's exact test; ^§χ² test with Yates correction.

OPEN IN VIEWER

Table 3.

Risk factors for suicidal behaviour (n=33; univariate Cox proportional hazard model)

Variables	p	Hazard ratio (95%CI)
Continuing medication	<0.01	0.23 (0.10–0.53)
Atypical antipsychotics (without clozapine)	0.01	0.40 (0.19–0.83)
No. psychiatric admissions†	<0.01	1.52 (1.20–1.94)
Non-compliance	0.04	2.09 (1.04–4.22)
Non-compliance and non-attendance	0.03	2.26 (1.09–4.70)
Past history of attempted suicide‡	<0.01	2.76 (1.37–5.56)
Anxiety§	<0.01	3.10 (1.45–6.58)
Past history of offence	0.01	3.26 (1.33–7.93)
Hallucination§	<0.01	3.51 (1.75–7.04)
Suicidal ideation§	<0.01	6.90 (3.24–14.71)
Depressed mood§	<0.01	7.94 (3.70–16.95)

CI, confidence interval; EASY, Early Assessment Service for Young People with Psychosis.

^†Prior to event or at the end of follow up; ^‡prior to entering EASY; ^§at the time of event or at the end of follow up.

Based on the present sample size, multivariate analysis was able to identify significant models that contained three variables. After performing multivariate analysis on all the possible permutations (each containing three variables) of the 11 risk factors, the variable of ‘non-compliance’ was found to be insignificant. Table 4 details the significant multivariate model that best explains the variance.

OPEN IN VIEWER

Table 4.

Risk factors for suicidal behaviour^† (n=33; stepwise multivariate Cox proportional hazard model^‡)

Variables	p	Hazard ratio (95%CI)
Continuing medication	<0.01	0.17 (0.02–0.42)
No. psychiatric admission	<0.01	1.68 (1.27–2.22)
Presence of depressed mood	<0.01	11.06 (4.93–24.83)

CI, confidence interval.

^†Combined group of completed suicide (n=8) and deliberate self-harm (n=29).

^‡Percentage of variance explained: 13.1.

Risk factors for completed suicide

The results of the univariate and multivariate Cox proportional hazard model are shown in Tables 5 and 6, respectively.

OPEN IN VIEWER

Table 5.

Risk factors for completed suicide (n=8; univariate Cox proportional hazard model)

Variables	p	Hazard ratio (95%CI)
Priority follow up target status	0.01	8.62 (1.74–43.48)
Depressed mood†	<0.01	11.49 (2.70–47.62)
Suicidal ideation†	<0.01	13.89 (3.30–58.82)

CI, confidence interval.

^†At the time of event or at the end of follow up.

OPEN IN VIEWER

Table 6.

Risk factors for completed suicide (n=8; stepwise multivariate Cox proportional hazard model^†)

Variables	p	Hazard ratio (95%CI)
Priority follow up target status	<0.01	16.46 (2.74–98.67)
Suicidal ideation‡	<0.01	22.05 (4.44–109.55)

CI, confidence interval.

†Percentage of variance explained: 13.8.

^‡At the time of event or at the end of follow up.

Risk factors more specific to Hong Kong

The priority follow-up (PFU) system was established in Hong Kong in 1982, whereby psychiatric patients with a history of violence or who were assessed to have a violent disposition were registered. Non-PFU status is given to a patient who is assessed to have no risk of violence. If such a risk is assessed to be present, a patient is categorized as PFU (target) or PFU (subtarget), in increasing order of risk of violence. The present findings suggest that PFU status could indicate suicide risk instead of being merely a proxy to assess violence.

In the literature, hallucination is seldom regarded as a risk factor and may even be associated with a reduced suicide risk [5]. Previous findings, however, are not directly comparable to the present results because the present study includes both suicide completers and attempters as subjects of interest. A recent study found a positive correlation between attempted suicide and hallucination but recruited subjects with first-episode psychosis rather than first-episode schizophrenia [11].

In the present study 9% of subjects had ceased to take medication after an average of 2.6 years of follow up. Evidence-based literature offers no consensus regarding when antipsychotic medication could be discontinued. Local guidelines in Hong Kong also give recommendations based on clinical experience [14]. The present study indicated that continuing medication was associated with a reduced risk of suicidal behaviour. Given this finding, patients should be informed of the possible protective effect of continuing medication on suicide risk before deciding to discontinue antipsychotic medication.

It was a surprise to find that prescribing atypical antipsychotic medication reduced the risk of suicidal behaviour. Literature lends no strong support on ‘anti-suicidal’ effect in atypical antipsychotic medications apart from clozapine [15]. Psychiatrists’ prescribing practice could explain the findings. At EASY, conventional antipsychotic medications remain the first-line treatment for schizophrenia [14], with 73.5% (n=172) of subjects starting treatment with conventional antipsychotic medication. But by the end of follow up, 53% (n=124) were on atypical antipsychotic medication. The chance to receive atypical antipsychotic medication increased as the subjects received EASY follow up for a longer time. Because the time to event was shorter in the event cases (n=33) than in the censored cases, the suicidal subjects (n=33) would be less likely to receive atypical antipsychotic medication. The second possible explanation is the confounding effects of other variables in particular psychiatric symptoms. These variables, however, had already been controlled in multivariate analysis. Third, atypical antipsychotic medication is less likely to cause akathisia, which has been reported to increase suicide risk [16]. Although there was no difference in extrapyramidal side-effects between suicidal and non-suicidal subjects, it was possible that failure to recognize akathisia contributed to suicide. Fourth, there may be a genuine ‘anti-suicidal’ effect due to atypical antipsychotic medication. This hypothesized effect would best be investigated in a randomized controlled study, but ethical consideration would be a serious concern.

The limitations of retrospective cohort design include information bias and confounders. There are also three major limitations of case-note review: non-standardization, underreporting, and rater value judgement. First, the case notes were not originally prepared for research purposes and therefore might have problems of inconsistency and non-standardization. Second, negative entry of a particular variable on past records may be due to its true absence or to missing information. It is possible that certain information enquired about on the data retrieval form had not been regarded by clinicians as relevant to a particular subject and therefore did not appear in the case notes. Third, although a single rater was involved, he was not blinded to the outcome event while collecting the data. The rater's value judgement could result in observer bias.

Theses limitations were balanced against the virtue of the highly specialized service. Approximately two-thirds of the variables were crude and factual information that was gleaned from past records, while the proforma data retrieval form would likely minimize rater value judgement. It is not possible to blind the rater if longitudinal information such as time to event needs to be obtained. The difference between the suicidal and non-suicidal group was unlikely to be due to systematic bias because clinicians could not have known which patients would later deliberately harm themselves or commit suicide. Random bias resulting in false-positive findings is very unlikely because of the highly significant p.

This study had a large Type II error, especially in the analysis involving completed suicide (n=8), while Type II error was less of a problem in the analysis involving the combined group of deliberate self-harm and completed suicide (n=33). There were also limitations related to identification of event cases. For completed suicide it was possible that the treating staff might have overestimated the self-inflicted act as the cause of death, and thereby overestimate the rate of completed suicide. If an attempted suicide was not documented in the case notes, it could result in underestimation of deliberate self-harm. Among the subjects classified as inactive (n=27) who had been discharged from psychiatric service at the time of data collection, the author could account for only eight cases of completed suicide. The remaining 19 subjects could possibly develop suicidal behaviour subsequently. This could underestimate the rate of suicidal behaviour.

Clinicians working in early intervention programmes should be vigilant with regard to patients who pose a high suicide risk. The present findings are useful to the early intervention programme for psychosis in Hong Kong, because the risk factors so identified are locally specific and could be elicited easily in daily clinical practice. In the future, a prospective cohort study to identify risk factors for suicide in young Chinese people suffering from schizophrenia is recommended.