A Prediction Model with Lifestyle in Addition to Previously Known Risk Factors Improves Its Predictive Ability for Cardiovascular Death

Abstract

This longitudinal cohort study aimed to create a novel prediction model for cardiovascular death with lifestyle factors. Subjects aged 40–74 years in the Japanese nationwide Specific Health Checkup Database in 2008 were included. Subjects were randomly assigned to the derivation and validation cohorts by a 2:1 ratio. Points for the prediction model were determined using regression coefficients that were derived from the Cox proportional hazards model in the derivation cohort. Models 1 and 2 were developed using known risk factors and known factors with lifestyle factors, respectively. The models were validated by comparing Kaplan-Meier curves between the derivation and validation cohorts, and by calibration plots in the validation cohort. Among 295,297 subjects, data for 120,823 were available. There were 310 cardiovascular deaths during a mean follow-up of 3.6 years. Model 1 included known risk factors. In model 2, weight gain, exercise habit, gait speed, and drinking alcohol were additionally included as protective factors. Kaplan-Meier curves matched better between the derivation and validation cohorts in model 2, and model 2 was better calibrated. In conclusion, our prediction model with lifestyle factors improved the predictive ability for cardiovascular death.