Self-harm in individuals with substance use disorders: Predictive factors and risk model

Substance use disorders are associated with an elevated risk of self-harm. Currently, clinical and structured assessment of self-harm risk typically relies on evidence from the general population samples. The aim of this study was to develop a risk model for self-harm that incorporates predictors specific to individuals with substance use disorders.

Using national registers, we identified a population-based cohort of 449 720 individuals with substance use disorders in Sweden between 2006 and 2020. We tested independence and strength of a range of socio-demographic and clinical factors, obtained through linkage of population-based registers, with a Cox proportional hazards model, and estimated the risk of self-harm. For the risk model, 361 120 individuals were allocated to the development sample and 88 600 to external validation based on different geographical regions. We assessed self-harm risk over five predetermined follow-up periods-within 7 days, 1 month, 3 months, 6 months and 12 months-following a healthcare contact for substance use disorders.

In the development sample, self-harm rates ranged from 0.6% to 3.5%, and in the validation sample from 0.5% to 3.6%. Ten risk factors were retained in the final risk model. Strongest associations with subsequent self-harm were for clinical factors: previous self-harm [hazard ratio (HR) = 3.17, 95% confidence interval (CI) = 3.08-3.26] and comorbidity of mental disorders (HR = 2.63, 95% CI = 2.50-2.72). Recent psychotropic medication use, including antidepressant (HR = 1.29, 95% CI = 1.23-1.38) and antipsychotic treatments (HR = 1.34, 95% CI = 1.24-1.44), was associated with increased risk, even after adjusting for psychiatric comorbidity, likely reflecting greater clinical severity and complexity. Across follow-up periods, performance was good in terms of discrimination, with area under the curve (AUCs) ranging from 0.73 (95% CI = 0.71-0.76) to 0.79 (95% CI = 0.78-0.80). In relation to calibration, expected-to-observed risk ratios were 1.00 to 1.04 and Brier scores 0.01 to 0.04 across follow-up periods. We used the model to generate a simple web-based risk calculator [Oxford Self-hArM after substance use disorders (OxSAMS)].

Modifiable clinical factors appear to have the strongest associations with increased risk of self-harm in people with substance use disorders. Structured tools, taking account of the different strengths of those factors, could inform clinical decision-making and provide a baseline assessment for training and research.