The Prognosis Of HIV/AIDS:Based On Subdistribution Proportional Hazard Regression Model

ObjectiveAcquired immunodeficiency syndrome(AIDS)continues to be a major public health problem threatening human life and social stability.With the start of antiretroviral therapy(ART),AIDS-related mortality shows an overall declining trend,while the morbidity and overall death rate shows an increasing trend.HIV/AIDS patients may die of non-AIDS related causes,which precludes the occurrence of AIDS related death.Conventional survival analyses are usually used to explore the prognosis of HIV/AIDS,while these methods take non-AIDS related death as censored data or combine the two causes of death,ignoring the existence of competing events.Only two studies in China explored the prognosis of HIV/AIDS based on the competing risk model,but didn't assess the model performance.This study evaluated the predictors of AIDS-related mortality and the model performance based on a retrospective cohort conducted in a high-risk area of Henan province,so as to provide targeted prevention guidelines,personalized management and improve the prognosis of HIV/AIDS patients.MethodsA retrospective cohort study was conducted involving 13,142 HIV/AIDS patients in a high-risk area of Henan province between 1995 and 2016.Microsoft Excel 2010 was used to establish the database,and all the statistical analyses were performed with R 3.5.1 software.The cumulative incidence of each endpoint event was estimated by Kaplan-Meier and cumulative incidence function(CIF),respectively.The difference between Kaplan-Meier and CIF was compared.Gray's test was used to compare the cumulative incidence curves of different characteristics groups.Statistically significant variables in univariate analysis were used to fit the subdistribution proportional hazard regression model.Proportional hazards assumption was verified based on Schoenfeld residual analysis.The discrimination of the model was assessed by time-dependent receiver operating characteristic curve(time-dependent ROC).The calibration of the model was evaluated by calibration plot.Brier score was calculated to assess the overall model performance.Finally,the stability of the model was tested by 10-fold cross validation.Results1.In all,3,386(25.8%)of 13,142 HIV/AIDS patients died of AIDS-related diseases,719(5.5%)died owing to non-AIDS-related causes,and 9,037(68.8%)were still alive or lost to follow-up.The mean survival time of HIV/AIDS patients who died of AIDS-related diseases was 2.8(0.9,5.7)years,and it was 4.1(1.3,9.0)years among HIV/AIDS patients who died of non-AIDS-related diseases.2.Kaplan-Meier method overestimated the cumulative incidence,and the bias would be more obvious with the extended follow-up time.CIF closed to unbiased estimate of cumulative incidence.3.Results of subdistribution proportional hazard regression model showed that patients who initiated antiretroviral therapy(ART)within 15 days of HIV/AIDS diagnosis(sHR=0.37,95%CI=0.33-0.41)and patients with higher baseline CD4~+T lymphocyte counts(200-350:sHR=0.37,95%CI=0.33-0.40;>350:sHR=0.24,95%CI=0.22-0.27)had a lower risk of AIDS-related mortality.However,patients with AIDS stage at diagnosis(sHR=5.70,95%CI=4.91-6.60),those infected through plasma(sHR=2.10,95%CI=1.81-2.45),and patients aged above 60 years(sHR=1.79,95%CI=1.44-2.22)had a higher risk of AIDS-related mortality.4.In the first 15 years,AUC(t)values were above 0.70,the probability of model prediction closed to the actual probability and the Brier scores were below 0.20,which showed the overall model performance was good.The results of 10-fold cross validation were similar to those before the validation.Conclusions1.In the presence of competing risk,CIF is a better method to estimate the cumulative incidence than Kaplan-Meier.2.Early diagnosis and antiretroviral therapy should be emphasized for improving survival of HIV/AIDS patients,especially elders,males,and patients infected through blood transmission.3.Subdistribution proportional hazard regression model has a good performance in the short and medium term,which provides method reference for the construction of HIV/AIDS risk prediction model in the future.