Population-split-based risk assessment model of venous thromboembolism in Chinese medical inpatients

AbstractsO_ST_ABSObjectiveC_ST_ABSInpatients with high risk of venous thromboembolism (VTE) usually face serious threats to their health and economic conditions. Many studies using machine learning (ML) models to predict VTE risk neglected an important statistical phenomenon, fuzzy feature, and achieved inferior results. Considering the effect of fuzzy feature, our study aims to develop a VTE risk assessment model suitable for Chinese medical inpatients. Materials and MethodsInpatients in the medical department of Peking Union Medical College Hospital (PUMCH) from January 2014 to June 2016 were collected. A new ML VTE risk assessment model was built through population splitting. First patients were classified into different groups based on values of VTE risk factors, then trustless groups were filtered out, and finally ML models were built on training data in unit of groups. Predictive performances of our method, five traditional ML models, and the Padua model were compared. ResultsThe fuzzy feature was verified on the whole dataset. Compared with the Padua model, the proposed model showed higher sensitivities and specificities on training data, and higher specificities and similar sensitivities on test data. Standard deviations of predictive validity of five ML models were larger than the proposed model. DiscussionThe proposed model was the only one which showed advantages on both sensitivity and specificity over Padua model. Its robustness was better than traditional ML models. ConclusionThis study built a population-split-based ML model of VTE for Chinese medical inpatients and it may help clinicians stratify VTE risk and guide prevention more efficiently.