Prediction of 5-year mortality risk in 784,892 people with mental illness: Development and validation of a novel clinical prognostic model (MortOx)

Individuals with psychiatric disorders have significantly higher mortality rates than the general population. Despite identifying risk factors, few attempts have been made to systematically use this information to stratify mortality risk. To address this gap, we developed and externally validated a risk prediction model using national healthcare and social register data from two countries. Nationwide register data were used to create a Swedish cohort (n=530,201) for model development and a Finnish cohort (n=254,691) for external validation. Participants, aged 15-60 years at assessment, had diagnoses of common mental illnesses (schizophrenia-spectrum disorder, bipolar disorder, depression, or anxiety disorders) made in specialist care. A multivariable logistic regression model assessed predictors of 5-year mortality risk. Model performance was evaluated using discrimination (area under the curve [AUC]) and calibration metrics, including intercept, slope, and visual plots. Internal validation employed bootstrapping. A total of 18,619 (3.5%) Swedish and 11,206 (4.4%) Finnish patients died within 5 years of assessment in secondary care. The model incorporated 25 predictors across four domains: sociodemographic factors, clinical characteristics, somatic comorbidities, and psychotropic medication use. External validation demonstrated excellent discrimination (AUC = 0.803; 95% CI: 0.799-0.807). The higher mortality rate in the Finnish cohort required recalibration of the intercept, and post-adjustment calibration was good (intercept: 0.00; 95% CI: -0.02; 0.02; slope: 0.99; 95% CI: 0.98-1.01). Model findings were translated into a web-based calculator (MortOx) for research, training, and potential clinical use. A transdiagnostic prognostic model based on 25 predictors accurately predicts 5-year mortality risk in mental illness. Linkage to interventions is needed to evaluate clinical impact.