Constructure of death risk model for AIDS inpatients receiving antiretroviral treatment at Yunnan Infectious Disease Hospital
Objective To estimate the risk of mortality among hospitalized AIDS patients by constructing models and analyzing laboratory parameters and comorbidities recorded during their inpatient episodes.This approach aims to enhance early warning systems and reduce mortality risks for hospitalized HIV patients.The effectiveness of model construction using balanced data was compared to determine the most efficient approach.Methods Samples of AIDS inpatients were chosen from January 1,2018,to December 31,2022,at Yunnan Infectious Diseases Hospital.These patients were then divided into training and validation sets with a ratio of 7:3,respectively.Predictive and validation models were constructed,ROC curves were plotted,and the optimal model was employed to create a column chart.Additionally,a comparison of model effects before and after balancing data was conducted using a confusion matrix.Results A total of 3 198 patients were included,of which 111 died.Patients with cardiovascular disease(CVD),AIDS-related lymphoma(ARL),severe bacterial infection(SBI),old age,elevated activated partial thrombin time(APTT),low albumin(ALB),high creatinine(Cr),low CD4 cell count,high neutrophil ratio(NEUT％),and a high red blood cell(RBC)number count were identified as risk factors for death in AIDS patients.The area under ROC(AUC)for constructing a prediction model on the training and validation sets was 0.903(95％CI:0.872-0.935)and 0.805(95％CI:0.721-0.890),respectively.However,the improvement in the performance of constructing a logistic regression model after balancing the data was not deemed significant.Conclusions The age of AIDS inpatients,APTT,ALB,Cr,CD4 cell count,NEUT％,RBC count,and the combination of CVD,SBI,and ARL demonstrate significant efficacy in predicting death risk.Additionally,the logistic regression model proves stables,even for unbalanced data.
HIV/AIDSin-hospital mortality riskprediction modelnomogramsimbalanced data
NETL
NSTL
万方数据
