Study design and population

This study consisted of both cohort and nested case–control designs. First, a national cohort study was conducted of all 6 947 191 persons aged ≥18 years who had lived in Sweden for at least 2 years as of 1 January 2002, as identified in the Swedish Total Population Register. This register contains demographic information for nearly 100% of persons living in Sweden since 1968.^{Reference Ludvigsson, Almqvist, Bonamy, Ljung, Michaelsson and Neovius18} Within this cohort, a nested case–control study was conducted of all 2601 persons who died by suicide during 2002–2015 and who had a registration of AUD within the previous 2 years, as identified using national population registries (as described below). Each of these individuals was matched to 10 controls randomly sampled from the general population who had the same birth year, birth month and gender, and who were still living in Sweden on the respective individual's death date (the index date).

We assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008. All procedures involving human participants were approved by the ethics committee of Lund University in Sweden (no. 2013/736). Participant consent was not required as this study used only anonymised registry-based secondary data.

Alcohol use disorder (AUD) ascertainment

Alcohol use disorder was identified using nationwide diagnoses and alcohol-related convictions reported during 2000–2015. First, ICD codes for AUD were identified from all primary or secondary diagnoses in the Swedish Hospital, Out-Patient, and Primary Care Registries (ICD-8: 291, 303, 357F, 425F, 535D, 571A–571D, 980, V79B; ICD-9-CM: same codes as ICD-8, plus 305A; ICD-10-CM: F10, except F10.0, Z50.2, Z71.4, E24.4, G31.2, G62.1, G72.1, I42.6, K29.2, K70.0–K70.9, K85.2, K86.0, O35.4). The Swedish Hospital Registry started in 1964 and initially included all hospital discharge diagnoses from six populous counties in southern Sweden, but was gradually expanded to cover >99% of the national population since 1987.^{Reference Ludvigsson, Andersson, Ekbom, Feychting, Kim and Reuterwall19} The Out-Patient Registry started in 2001 and contains out-patient diagnoses from all specialty clinics nationwide. The Primary Care Registry initially included all primary care diagnoses from two populous counties covering 20% of the national population starting in 1998, but was expanded in 2001 to cover approximately 45% of the national population and since 2008 has covered 75%.^{Reference Sundquist, Ohlsson, Sundquist and Kendler20} In addition, AUD was identified in all individuals having at least two convictions for drink-driving (law 1951:649) or drunk in charge of a maritime vessel (law 1994:1009), using nationwide data from the Suspicion and Crime Registers.

Suicide ascertainment

The study cohort was followed up for suicide deaths from 1 January 2002 through 31 December 2015, using nationwide data from the Swedish Cause of Death Registry. This registry includes deaths and ICD codes for cause of death among all persons registered in Sweden since 1960, with compulsory reporting nationwide. All intentional deaths were identified using ICD-10-CM codes X60–X84, and deaths of undetermined intent were identified using ICD-10-CM codes Y10–Y34. Prior studies have suggested that substantial numbers of suicides may be misclassified as deaths of undetermined intent^{Reference Ohberg and Lonnqvist21,Reference Bjorkenstam, Johansson, Nordstrom, Thiblin, Fugelstad and Hallqvist22} and that such misclassification may be more common in persons with AUD.^{Reference Lindqvist and Gustafsson23,Reference Allebeck, Allgulander, Henningsohn and Jakobsson24} In the present study, intentional deaths and deaths of undetermined intent were analysed together as the primary outcome and separately as secondary outcomes.

Ascertainment of healthcare encounters

In the nested case–control study, all healthcare encounters in in-patient, specialty out-patient and primary care out-patient settings during the year prior to the index date, regardless of diagnosis, were identified using the Swedish Hospital, Out-Patient and Primary Care Registries (as described above). To exclude encounters that were the direct result of a fatal suicide attempt, hospital admissions were included only if the discharge date preceded the recorded date of death. In addition, all healthcare encounters were excluded if they occurred on the death date and contained a diagnosis directly related to mortality (i.e. ICD-10-CM R96–R99: ‘ill-defined and unspecified causes of mortality’).

Other study variables

Sociodemographic characteristics and psychiatric disorders that may be associated with AUD, suicide and healthcare utilisation were examined as adjustment variables. Sociodemographic factors were identified from the Total Population Register and national census data and were 100% complete. These variables included age (continuous and categorical (18–24, 25–34, 35–44, 45–54, 55–64, 65–74, ≥75 years) to allow for a non-linear effect), gender, marital status (married/unmarried) and education level (≤9, 10–12, >12 years) at baseline. Psychiatric disorders included drug use disorders (ICD-8: 304; ICD-9-CM: 292, 304–305, except 305.0 and 305.1; ICD-10-CM: F11–F16, F18–F19; Suspicion Register codes 3070, 5010–5012; Crime Register codes for laws covering narcotics (law 1968:64) and drug-related driving offences (law 1951:649)), affective disorders (ICD-8: 296.0–296.8, 300.4; ICD-9-CM: 296A–296E, 296W, 300E, 311; ICD-10-CM: F30–F39, except 32.3), anxiety/phobia disorders (ICD-8: 300.0, 300.2; ICD-9-CM: 300A, 300C; ICD-10-CM: F40–F41), psychotic disorders (ICD-8: 291, 295, 297–299; ICD-9-CM: 291–292, 295, 297–299; ICD-10-CM: F20–F25, F28–F29, F32.3, X.5 in F10–F19), personality disorders (ICD-8/9-CM: 301; ICD-10-CM: F60), and other psychiatric disorders (ICD-8: 300.1, 300.5–300.9; ICD-9-CM: 300B, 300F–300H, 300W, 300X, 307B, 307F; ICD-10-CM: F43–F45, F48, F50).

Statistical analysis

In the national cohort analysis, Poisson regression with robust s.d. was used to determine incidence rate ratios (IRRs) and 95% confidence intervals (CIs) for suicide associated with AUD. Analyses were conducted using three models: (a) unadjusted, (b) adjusted for sociodemographic factors (age, gender, marital status, education level) and (c) further adjusted for psychiatric disorders (as above).

In the nested case–control analysis, Poisson regression with robust s.d. was used to compute prevalence ratios (PRs) and 95% CIs for the prevalence of a healthcare encounter in specific time intervals before the index date (<2 weeks, 1 month, 3 months, 6 months, 1 year) among the people with AUD compared with the controls. In addition, generalised linear models with a Poisson distribution, identity link function and robust s.d. were used to compute prevalence differences (PDs) and 95% CIs for those same prevalences in the people with AUD versus controls. Each of these models was performed both unadjusted and adjusted for covariates (as above). Poisson model goodness-of-fit was assessed using Pearson and deviance tests and was met in each model.

Interactions between AUD and gender or age were examined in relation to healthcare encounter prevalence on both the additive and multiplicative scale. In a sensitivity analysis, the analyses of healthcare utilisation were repeated after identifying AUD on the basis of any reported lifetime history rather than within the previous 2 years. All statistical tests were two-sided and used an α-level of 0.05. All analyses were conducted using Stata version 15.1 for Windows.