This study used a time series design describing the annual suicide rates between 2010 and 2021 for patients with CUD, AUD, and other SUDs.

Cases were identified through a registry linkage of all individuals who died by suicide or undetermined causes of death (X60-X84; Y10-Y34; Y870; Y872) recorded in the Norwegian Cause of Death Registry (NCDR) between January 1st, 2010 and December 31st, 2021, and who had been in contact with specialized SUD or mental health services, as recorded in the Norwegian Patient Registry (NPR) during the last year before suicide. The unique personal identifier assigned to all Norwegian citizens at birth or immigration was used to link the data. We included all individuals with any incident CUD (ICD-10 code F12), AUD (F10), or other SUDs (F11; F13-F16; F18-F19) [24] recorded in the observation period. The number of patients with SUDs who died by suicide was aggregated by year, gender, and age.

A general treatment population was identified in the NPR as the annual number of individuals with any SUD recorded (F10-F16; F18-F19) in contact with SUD or mental health services from January 1st, 2010, to December 31st, 2021. These data were aggregated by year, gender, and age. Cells with counts below three were removed when data were delivered.

The NCDR includes information on causes of death in the Norwegian population. The registry has a high degree of completeness and good data quality [25]. The NPR contains information about contacts with specialized health services in Norway. The registry contains personally identifiable information for contact with mental health services since 2008 and for SUD services from 2009. The coverage of the registry is very good from 2010 onwards [26].

For the cases, we recorded the first CUD from the NPR between 2008 and 2021. For the comparison groups, we ascertained the first diagnosis of AUD (F10) and other SUDs (F11; F13-F16; F18-F19) in patients not diagnosed with CUD. Moreover, we recorded whether the diagnosis was primary or secondary, recorded in SUD or mental health services, and the follow-up time from recording the diagnosis to suicide in person-years. We also identified any registration of comorbid mental disorders categorized using ICD-10 codes for psychosis or bipolar disorders (F20-F31), depressive or anxiety disorders (F32-F49), personality disorders (F60-F69), and ADHD (F90). In patients with CUD we also recorded if AUD or other SUDs were present. From the NCDR, we used the date of death to extract the year of the suicide, along with gender and age at the time of death. The latter was categorized into ten-year bands (10–19, 20–29, 30–39, 40–49, 50–59, 60-).

For the general treatment population, we used aggregated counts of the number of patients treated with a CUD, AUD, or other SUDs diagnosed in specialized services by year, gender, and age group.

Differences between the groups were examined using Chi-squared tests for categorical variables and Mann-Whitney tests for continuous variables. Suicide rates were estimated from data aggregated by year, gender, and age group. We used the number of patients who died by suicide as the numerator and the general treatment population with the same diagnosis as the denominator.

We estimated overall, gender-specific, and age-specific suicide rates and the annual number of suicides and crude suicide rates from 2010 to 2021. Next, we analyzed the annual trends of suicide rates between 2010 and 2021 by estimating incidence rate ratios (IRRs), using the suicide rate in 2010 as a reference. Confidence intervals for the IRRs were estimated using the Delta method, and two-tailed Wald tests were used to test the significance of the IRRs.

To examine the trend of the suicide rates over the whole period, we used Poisson regression, regressing the suicide rate over year. We built both a crude model and a model adjusted for gender and age. Groups with missing data in the population were excluded from the gender- and age-adjusted model. Furthermore, we adjusted the crude model separately for annual counts of psychosis or bipolar disorders, depressive or anxiety disorders, personality disorders, ADHD, and AUD or other SUDs to assess whether any of these variables could explain the trends in suicide rates. In models finding a crude association between the suicide rate and year, the explanatory potential of the covariates was considered. The explanatory potential was considered present if the coefficient for the year was non-significant and the coefficient for the covariate was significant after adjustment. Data were analyzed with R version 4.2.2 [27], and the figures were generated with the ggplot2 package [28].

The Norwegian Directorate of Health has given an exemption from patient confidentiality (ref: 16/27835-12) to access the individual data for the cases since this is a retrospective study where informed consent is impossible to obtain. The linkage was further approved by the Regional Committee for Medical and Health Research Ethics South-East Norway (reference: 32494). The aggregated data on the population were delivered as anonymous data where no formal approvals were needed.