Study On The Survival Prediction Model Of AIDS-related Cryptococcal Meningitis Based On XGBoost

In view of the extremely high mortality rate of AIDS-related cryptococcal meningitis,this paper established three XGBoost-based survival prediction models(BecCox,EXSA,Hit Boost)on related datasets.By comparing the prediction performance of the above three models and Cox proportional risk model,the mortality prediction model with high accuracy was found,and the mechanism of the prediction results was explained.The model was used to identify patients at high risk,which provided a theoretical basis for taking measures to reduce the mortality of patients.It further enriches the research topic of machine learning in survival prediction model.First,we took the clinical trial data of patients with AIDS-related cryptococcal meningitis--ACTA dataset as the research object.On the basis of data cleaning,missing value filling,and feature extraction,the predictive performance of XGBoost-based survival prediction models BecCox,EXSA,Hit Boost and traditional Cox proportional hazards model was compared.The results showed that the BecCox model has better performance in terms of prediction accuracy on the ACTA dataset,with an average consistency index of 0.746,which is 5.8% higher than the average C-index of the CPH model.In addition,the model identified the key factors affecting HIV-associated cryptococcal meningitis as cerebrospinal fluid yeast count,hemoglobin content and serum glucose levels.This enriched the research and conclusions of the XGBoost-based survival prediction model on the AIDS-related cryptococcal meningitis dataset.Secondly,we introduced the SHAP model to explain the prediction value’s generation mechanism of the BecCox model from the overall and individual perspectives.On the basis of the BecCox model,we enriched the research and conclusions on the interpretability of the BecCox model,and improved the trust of the innovative survival prediction model in the field of healthcare.Finally,we used Jenks Natural Breaks clustering algorithm to select the number of clusters,and based on the prediction results of the BecCox model,the risk rating of the patients was carried out,and the rationality of the risk rating results was verified on the test set.The average survival rates of different risk groups under different treatment methods were compared,and the optimal treatment method for each risk group was obtained.The obtained results can be used as auxiliary information to assist doctors in clinical treatment.It enriched the research on patients’ risk rating and the application of survival prediction model.