Using the unique national identification number assigned to all Swedish citizens we linked data on all people living in Sweden during 1973-82 (n=9.4 million) to the hospital discharge register (held by the Epidemiological Centre of the National Board of Health and Welfare), cause of death register (Epidemiological Centre), 1970 population and housing census, and education and migration registers (the last three held by Statistics Sweden).
We then identified people aged 10 or older who had been admitted to inpatient care in Sweden during 1973-82 because of suicidal behaviour, defined as a definite suicide attempt (E950-9, international classification of diseases, eighth revision) or an uncertain suicide attempt (E980-9), as registered in the hospital discharge register (n=49
509). In case of more than one admission for a suicide attempt we used the first admission as the index attempt. To avoid confounding effects of being in the asylum seeking process we excluded people who had immigrated within two years before baseline (n=860). We identified cases as those with one of the studied psychiatric diagnoses present at discharge from the index admission for a suicide attempt or at discharge from the first inpatient episode beginning within one week after this index episode, as recorded in the hospital discharge register (n=12
681). Those without a diagnosis of mental disorder within one year after the suicide attempt were used as reference subjects (n=27
004). We did not include people with a different psychiatric diagnosis to those studied, or a diagnosis after one week but within one year after the suicide attempt (n=8964). The study cohort thus consisted of 39
685 people; 18
642 males and 21
043 females, mean age 38.4 (SD=16.5) years and 37.0 (SD 17.0) years, respectively.
The reference group comprised 68% of the study cohort. Reference subjects had attempted suicide but had no coexistent mental disorder within one year after the attempt. In clinical practice, the suicide attempt code in itself has often been considered sufficient to indicate distress when reported to the national hospital discharge register. A proportion of the reference subjects might, however, have had milder forms of psychiatric illness. Thus our reference group consisted of people who had attempted suicide and had none, subclinical, or milder forms of psychiatric morbidity at the index episode and within one year thereafter.
We studied eight psychiatric disorders: schizophrenia (code 295, international classification of diseases, eighth revision), bipolar and unipolar disorder (296.1-296.9), other depressive disorder (296.0, 300.4), anxiety disorder (300, except 300.4), adjustment disorder or post-traumatic stress disorder (307), alcohol abuse or dependence (303), drug abuse or dependence (304), and personality disorder (301). The hospital discharge register has national coverage for psychiatric disorders from 1973. Reporting to this register is mandatory for all healthcare providers, including the few existing private hospitals. Swedish registers are of good quality,17
and the validity of diagnoses for schizophrenia in the hospital discharge register was confirmed during a file based review by psychiatrists.18
For all disorders we used the principal diagnosis. As the cause of death register covers more than 99% of deaths in Swedish residents, including those occurring outside Sweden, the loss of information on death by suicide was minimal.19
We introduced the potential confounders of age, educational level, and immigrant status as covariates in the regression analyses. We measured education on a seven point ordinal scale from not having completed compulsory school (shorter than nine years) to postgraduate education, using the 1970 population and housing census and the education register for 1990, 2000, and 2004. Immigrant status (yes or no) was obtained from the migration register.
Patients were followed from discharge after attempted suicide to a definite or uncertain suicide (E950-9 and E980-9, international classification of diseases, eighth and ninth revisions, and X60-84 and Y10-34, ICD-10), death other than suicide, first emigration, or end of follow-up (31 December 2003). We included uncertain suicides because their exclusion might have led to an underestimation of suicide rates.20
Thus we followed-up patients for 21-31 years.
We used Kaplan-Meier survival curves to plot temporal patterns of suicide after a suicide attempt; we excluded adjustment disorder or post-traumatic stress disorder owing to absence of risk effect and substance misuse owing to low prevalence. For each diagnostic category and sex we determined separately absolute and relative mortality from suicide after a suicide attempt and used Cox regression models to compute hazard ratios (95% confidence intervals) taking time at risk into account. We adjusted the hazard ratios for age, highest level of education, and immigrant status. For each psychiatric disorder we calculated the population attributable fractions of suicide among people who had previously attempted suicide using the formula P×(HR−1/HR), where HR is the hazard ratio derived with Cox regression (adjusted for age, education, and immigrant status) and P the base rate of the disorder among all people who had completed suicide. We used SPSS for Windows (version 15) for the statistical analyses.