Development of death-risk score based on epidemiology of six mental disorders and application to mortality reduction via modifiable health behaviors.

Abstract

Prediction models for mortality rarely incorporate diverse psychiatric conditions, alongside clinical, behavioral, and demographic risk factors. We aimed to develop and validate a mortality risk score using a Gradient Boosting Survival Model (GBSM) that enables stratified risk prediction and supports behavioral interventions. We conducted a retrospective cohort study using data from the National Health Insurance Service (NHIS), including 8,962,548 adults (20-80 years) with or without a diagnosis of six major categories of mental disorders: mood, anxiety, psychotic, stressor-related, somatic-psychobiological disorders, and addictions (2015-2019). A GBSM was developed to predict all-cause mortality with Brier score, C-index, and area under the receiver operating characteristic curve (auROC). GBSM demonstrated enhanced predictive performance with a C-index of 0.860 and an auROC of 0.864. The risk score (range 0-100) showed a clear dose-response gradient with all-cause mortality. Compared to the lowest quartile, hazard ratios (95% confidence intervals) for mortality were 2.09 (1.95-2.23), 4.75 (4.48-5.05), and 35.8 (33.8-37.8) in the 2nd to 4th quartiles, respectively (P < 0.001). Additionally, sex-stratified analyses were conducted to assess the association between changes in lifestyle behaviors and mortality risk across score ranges. Among males, newly initiated smoking was associated with increased mortality risk across most (20-80) score groups. Increased physical activity was linked to reduced mortality, particularly among females with intermediate scores (40-50). We present the score that quantifies mortality risk, demonstrates strong predictive accuracy, and identifies high-risk groups for sex-specific behavioral interventions.