Abstract
BACKGROUND: Mental disorders are highly prevalent, and comorbidities between physical and mental health conditions are common. Physical comorbidities and family health histories may improve the accuracy of mental disorder risk prediction. We developed prediction models for mental disorder risk using comprehensive individual and family mental and physical health histories. METHODS: We conducted a population-based cohort study using administrative Health data in Manitoba, Canada, and included adults between 1977 and 2020 with linkages to at least one parent and one grandparent. Mental disorders (mood and anxiety, substance use and psychotic disorders) for individuals, parents and grandparents were identified in inpatient and outpatient Health records. Predictors included demographics, family history of mental disorders and 130 health conditions in individuals, parents and grandparents. We used the Least Absolute Shrinkage and Selection Operator (LASSO) logistic regression to build prediction models that sequentially included health conditions in individuals, parents and grandparents. Predictive performance was evaluated using the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, positive predictive value, negative predictive value and Brier score. RESULTS: Of 125 070 individuals identified, 109 359 had no preexisting mental disorder. 52.9% were males and 52.8% were urban residents. 39 651 (36.3%) had a recorded diagnosis of mental disorders during follow-up. Predictive models incorporating Health histories of individuals, parents, and grandparents achieved the best predictive performance. Amongst all mental disorders, psychotic and substance use disorders had the highest AUCs of 0.78 (95% confidence interval (CI) 0.75–0.81) and 0.75 (95% CI 0.73–0.76), respectively. Key predictors included comorbid mental disorders, gastrointestinal conditions, female infertility and family history of dementia, gastrointestinal and metabolic conditions. CONCLUSIONS: Individual and family histories of physical and mental conditions improved mental disorder risk prediction, though accuracy was only moderate, highlighting the need for further refinement of risk prediction. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1186/s12888-025-07323-z.