Study On PIM Risk Early Warning Model In Elderly Hypertensive Patients

Objective: This study aimed to develop and interpret a risk warning model of PIM for elderly hypertensive patients based on real-world data.And the secondary goal is to explore the related factors,incidence and types of PIM occurrence during real-world data extraction,systematic review and meta-analysis.Methods: First,systematic review and meta-analysis were used to summarize and analyze the studies on PIM of Chinese elderly published at home and abroad based on Beers 2019,China Criteria 2017 and STOPP/START 2014,in order to obtain the relevant factors affecting the occurrence of PIM.So it can provide the basis for feature extraction table of real world data.Moreover,the incidence and types of PIM among the elderly in China can be learned from the pooled analysis.Secondly,the feature extraction table was used to extract the data of elderly hypertension patients from two hospitals in Chengdu,so as to prepare for the establishment of elderly hypertension PIM prediction model.At the same time,statistical analysis of real world data was carried out to explore the occurrence of hypertension PIM and related risk factors in the elderly patients.Finally,based on the real world data,six machine learning algorithms were used to establish the PIM prediction model of hypertension in the elderly under full variables or after feature screening variables,then the model with good performance was selected for Bayesian optimization,and the model with the best prediction effect was obtained.After the SHAP method was used to explain the model.Results: 1.The literature review found 26 characteristics positively correlated with PIM occurrence in elderly patients,including the number of drugs,number of diseases,length of hospital stay,age,and gender.And the incidence of PIM was 30% to 47% under the three criteria.In addition,benzodiazepines were the most common type of PIM in the elderly,followed by diuretics,ACEI,and clopidogrel.2.A total of 741 elderly patients with hypertension were enrolled,and statistical analysis of real world data demonstrated that the three drug categories with the highest frequency of PIM occurrence in elderly hypertensive patients were benzodiazepines,diuretics and ACEI.Under the three criteria,the incidence of PIM was 41.7% to 59.4%,and the overall incidence was 74.5%.The warning factors of PIM in hypertensive patients included 11 characteristics,such as central nervous system drugs and alpha-receptor blockers.3.The Light GBM and XGBoost models were the most effective in predicting PIM in elderly hypertension patients based on full variables and feature screening.Among the two models,XGBoost model has the better performance.The Bayesian method was used to optimize it,and finally the BSXGB model achieved the best performance with an AUC of 0.863.The SHAP method revealed that the two most important features were central nervous system drugs and the number of total drugs.Conclusion: This study successfully developed a prediction model for PIM in elderly patients with hypertension(BSXGB),which provides a more intelligent method for identifying PIM.And it is expected to promote the rational use of drugs in elderly hypertensive patients.Meanwhile,it provides reference for the development of other intelligent models of machine learning for chronic diseases of the elderly.