This study used a retrospective psychiatry-based case-control design. Data from the period 1 January 2007 to 31 December 2013 were obtained from the Swedish National Cause of Death Registry , psychiatric and medical records, and the statistics of the Swedish total population .
The study catchment area (Örebro County) covers a population of 285,395, with a mix of urban and rural areas. In 2013, the region had suicide rates of 21.7 suicides per 100,000 male inhabitants and nine suicides per 100,000 female inhabitants. These rates were higher than the Swedish national rates for both males (16.2 per 100,000) and females (7.5 per 100,000) during the same period . The specialist psychiatric unit belongs to a general university hospital with 918 beds, of which 136 are part of the psychiatric care unit. In 2013, the psychiatric units offered 40,020 days of inpatient psychiatric care and 131,137 outpatient department visits. Persons outside specialist psychiatry with visits only to community services outside of health care, or those at private clinics, are not included in this study.
Sample and participant selection
Suicide cases were identified using the Swedish National Cause of Death Registry. According to the registry, during the period 1 January 2007 to 31 December 2013, a total of 339 individuals (69.3% men) from Örebro County died secondary to suicide (codes X60–X84) or undetermined intent (codes Y10–Y34) classified in accordance with the International Statistical Classification of Diseases and Related Health Problems, 10th Revision (ICD-10) . The national identification numbers of these individuals were used to search the electronic psychiatric medical records of the University Hospital of Örebro. This revealed that 154 (45.4%) of these 339 individuals had received psychiatric care at this regional centre during the 2 years prior to their death. These 154 individuals (65.6% men) were therefore included in the present study as suicide cases. Cases classified as uncertain suicides were also included, as their exclusion may have led to an underestimation of the suicide rate. The national statistics on causes of death include cases of uncertain cause of death as possible suicides, as analyses indicate that the majority of uncertain cases are probable suicides [3, 36]. In the subsequent text, the term suicide refers to both definite suicide and uncertain suicide.
The 154 control subjects were identified through the electronic psychiatric medical records of the University Hospital of Örebro during the matching procedure (see next section).
Case-control matching procedure
A hospital statistician outside the research team personnel matched the suicide cases with control subjects on the basis of the following: 1) a history of contact with psychiatric services in the year that the suicide case died, 2) age, 3) sex, and 4) primary psychiatric diagnosis. For the primary psychiatric diagnosis, consistency was required in terms of the first two or three digits of the respective ICD-10 code (e.g., F32.2 or F32). When applicable, efforts were made to control for a comorbid diagnosis of psychoactive substance use disorder (F10-F19), as comorbid substance use disorders increase the risk for suicide [37, 38]. For suicide cases below the age of 25 years, a control of similar age (± 2 years) was sought. In older suicide cases, controls with a maximum age difference of 2 years were primarily sought; however, a maximum age difference of 10 years was accepted.
A total of 113 (73.4%) suicide case-matched control pairs met the stringent primary diagnosis matching criteria (e.g., F32.2). Twenty-five (16.2%) pairs were matched on the basis of two digits (e.g., F32), and two (1.3%) pairs were matched on the basis of a diagnostic cluster (e.g., Mood disorders F30–F39). Thirteen suicide cases (8.5%) lacked a primary psychiatric diagnosis. Of these, 12 were matched with controls without psychiatric diagnoses, while a control with a primary diagnosis of depression was selected as the optimal match for the remaining case (0.6%). A total of 15 suicide cases had a comorbid diagnosis of psychoactive substance use (F10–F19). Of these, nine were matched to a control with a similar comorbidity.
For suicide cases younger than 25 years, 16 out of 18 pairs (88.9%) were matched according to the study criteria. For the remaining two pairs, the age-interval was extended to ±3 years and ± 7 years. The majority (87.5%) of the 136 case-control pairs in the age group 25 years or older were matched according to the study criteria (i.e., a maximum age-interval of ±2 years); however, the age-interval was extended to ±3 years in seven pairs, ± 4 years in three pairs, and ± 5 years in five pairs. In the remaining pair, a 96-year-old suicide case was matched to an 85-year-old control.
The mean follow-up time in days was calculated from the date of the first psychiatric care contact included in the study to the date of suicide (for suicide cases), or to the date of last contact included in the study (for controls). The mean follow-up time was 446 days (SD 256.4) for suicide cases and 479 days (SD 254.1) for control subjects, with no significant difference being found between suicide cases and controls according to an independent t-test analysis (t = 1.13, df 306, p = 0.259).
The Swedish National Cause of death registry
The suicide cases in this study were identified from the National Cause of Death registry. This contains information on the deaths of all Swedish citizens, including deaths occurring outside Sweden. The registry does not include information concerning the deaths of people seeking asylum, undocumented migrants, or visitors to Sweden. Information concerning suicide is based on death certificates. These are completed by a physician following a clinical or forensic autopsy, with the cause of death being classified in accordance with the ICD-10 codes . The annual rate of missing data on causes of death for Swedish citizens is less than 2% . For the purposes of the present study, the following data were collected: national identification number, municipality, date of death, underlying cause of death, and medical evaluation of whether death by self-harm was intentional or of uncertain intent.
Data from medical records for suicide cases and controls
Information concerning psychiatric outpatient visits and psychiatric and somatic admissions in the 2 year period prior to each suicide was retrospectively collected from electronic medical records. The data concerned all care received from the 1st January 2005 until the 31st December 2013. For example, for a suicide case with a date of death of 14th March 2007, data were collected from 14th March 2005 to 14th March 2007 inclusive. Care consumption for controls was included from the same year as the death of the case, and the data were collected for 2 years back in time.
For each suicide case and control subject, data-files were created from electronic psychiatric medical records, and if applicable, from the relevant somatic medical records. From these files, data were retrieved concerning the number of psychiatric outpatient visits, the number of psychiatric admissions, the total number of days of hospitalization (from admission to discharge) for each psychiatric admission, the total number of days spent as a psychiatric inpatient, the occurrence of somatic hospitalization, and the ICD-10 diagnoses within the 2 year study period.
Information concerning the interventions implemented during outpatient visits and/or inpatient care in psychiatry (e.g., psychological treatment, prescriptions of psychotropic drugs, and ECT) was gathered from the electronic medical records.
Somatic comorbidity and serious suicide attempts
Data on somatic hospitalization at a specialist medical unit together with diagnoses according to ICD-10 codes (see Appendix for ICD-codes) were used as indicators of somatic comorbidity, and somatic comorbidity was treated as a possible covariate in the analyses, as it is assumed to increase suicide risk. Somatic comorbidity did not include hospitalization secondary to injuries or suicide attempts. A serious suicide attempt was defined as any intentional attempt to end life that led to hospitalization at a somatic specialist medical unit (e.g., due to intoxication). The ICD-10 codes for previous suicide attempts are provided in the Appendix.
Descriptive statistics were used to describe the distribution of psychiatric diagnoses, sex, and age in the suicide cases and control subjects. The chi-square test and Fisher’s exact test were used to test for significant case-control differences in the occurrence (yes/no) of outpatient or inpatient psychiatric care, somatic hospitalization due to comorbidity, somatic hospitalization due to previous serious suicide attempts, and specific psychiatric interventions. The Mann Whitney U-test was used for case-control comparisons involving non-normally distributed variables (e.g., the frequency of health care provision).
Treatment interventions/covariates for which case-control differences (p < 0.1) were found (see Table 2) were evaluated as predictors or covariates using stepwise forward logistic regression models (i.e., a selection method using the likelihood ratio) to examine their association with suicide. This analysis was used to compute odds ratios (ORs) and confidence intervals (CIs) with suicide (yes = 1, no = 0) as a binary dependent outcome variable. Correlations between the included interventions and covariates were examined with Spearman correlation analyses. All statistical analyses were performed using SPSS for windows version 22 (IBM, New York).
A power calculation was performed on the basis of assumptions of 25% exposure to treatment interventions in controls, an OR of 2, and a confidence level of 95%. This showed that to obtain a power of 80% for the detection of significant differences, a sample size of 154 patients per group was required. If the exposure to treatment among controls was only 5%, an OR > 3.2 would be required to detect significant differences with this sample size.