Objective To predict risk factors affecting the prognosis of HIV/AIDS patients with Candida infection,providing clinicians with predictors for early identifying high-risk patients. Methods Clinical data were collected from HIV/AIDS patients with Candida infection admitted to an infectious disease hospital in Guangxi from January 2012 to June 2019. Patients were divided into a death group and a survival group according to their different prognostic outcomes. Cases were randomly selected and matched using propensity score matching (PSM) at a ratio of 1∶3 (death:survival) to construct the model. The data were split into a training set and a testing set at a ratio of 7∶3. Various machine learning models were built,and the optimal model was selected as the final prediction model by comprehensively evaluating the model performance. Finally,the SHAP values were used to interpret the features of the model and analyze the influencing factors of patients' prognostic outcomes. Results A total of 3098 HIV/AIDS patients with Candida infection were collected. From 2012 to June 2019,the in-hospital mortality rate of HIV/AIDS patients with Candida infection showed a linear and stable downward trend (P=0.043). After applying PSM,data from 1620 cases were used to construct six different machine learning models,among which the XGBoost model had the best performance (training/testing set,AUC=0.98/0.85,sensitivity=0.93/0.75,specificity=0.93/0.84). Respiratory failure,urea,and LDH levels were thought to be the three major factors affecting the prognostic outcomes of HIV/AIDS patients with Candida infection. Conclusions The XGBoost model showed good predictive performance in predicting prognostic outcomes of HIV/AIDS patients with Candida infection. The model can provide early warning for the identification of high-risk patients and assist clinicians to take personalized treatment measures promptly,which is of great significance for guiding clinical decision-making.
维普
万方数据
