Development And Validation Of A Stroke Risk Prediction Model Based On A Large Prospective Cohort Population In The Inner Mongolia

Objective: Based on prospective cohort population data in Inner Mongolia,this study describes the current prevalence of stroke among residents in the region.A prediction model for stroke occurrence risk was established using multiple factor Cox proportional risk regression model,competitive risk model,and random survival forest model,and the prediction model was validated.The aim is to provide a reliable tool for predicting and evaluating stroke risk,To provide information and reference for screening high-risk populations for stroke in Inner Mongolia and developing primary stroke prevention strategies.Methods This study is a prospective follow-up study,with a population from the national "Early Screening and Comprehensive Intervention Project for High Risk Population of Cardiovascular Disease" in 2015.The study subjects were selected using a multi-stage stratified cluster sampling method in Inner Mongolia from 2015 to 2021.The study subjects were permanent residents aged 35-75.Divide the research subjects into a training set and a validation set at a ratio of 7:3.Based on the training set,establish a stroke incidence risk prediction model using a multi factor Cox proportional risk regression model,a competitive risk model,and a random survival forest model.Internally validate the model in the validation set.Evaluate the predictive models established by the three methods using out of pocket error rate and comprehensive Brier score.Results A total of 29391 subjects were included in this study.The median follow-up time of the study population was 3.25 years.During the follow-up period,the number of stroke cases was 1643.The crude incidence rate of stroke was 5.59%,and the age standardized incidence rate was 4.46%.The prediction model of stroke risk based on multivariate Cox proportional risk regression includes age,gender,marital status,education level,farmers,physical exercise,obesity,diabetes,hypertension,and dyslipidemia.The C statistic of training set is 0.716(95%CI: 0.702-0.731),the C statistic of validation set is 0.725(95% CI: 0.703-0.747),and the comprehensive Brier scores of training set and validation set are 0.237 and 0.242,respectively,And the calibration and standard lines in the training and validation sets are consistent,indicating that the predicted risk of stroke based on the model is consistent with the actual occurrence risk.The decision curves in the training and validation sets show good returns for the prediction model within the high-risk threshold range.The results of the multi factor competitive risk model showed that age,gender,marital status,education,smoking,farmers,physical exercise,obesity,diabetes and hypertension were independent risk factors for stroke.The C statistic in the training set was 0.722(95% CI: 0.714-0.739),while the C statistic in the validation group was 0.716(95% CI: 0.708-0.745).The comprehensive Brier scores in the training and validation sets were 0.223 and 0.245,respectively.The calibration and standard lines in the training and validation sets were consistent,indicating that the predicted risk of stroke based on the model was consistent with the actual risk of occurrence.The decision curves in the training and validation sets were within the high-risk threshold range,The prediction model has good returns.The random survival forest model is established based on the method of minimum depth to filter independent variables,with an out of band error rate of30.26% in the training set and 28.93% in the validation set;The comprehensive Brier scores for the training and validation sets are 0.348 and 0.374.The out of pocket error rate and comprehensive Brier score of the competitive risk model are lower than those of the Cox proportional risk regression model and the prediction model based on random survival forest,indicating that the stroke risk prediction model based on the competitive risk model in this study has better discrimination and calibration.Conclusions The risk prediction model for stroke occurrence in Inner Mongolia based on the competitive risk model has better predictive performance compared to Cox proportional risk regression and random survival forest models.The competitive risk model can accurately predict whether individuals will experience stroke and provide information reference for stroke prevention work in healthcare departments.