Application Of Logistic Regression And Time-dependent Cox Regression In Prognosis Analysis Of HIV/AIDS

ObjectiveAIDS(Acquired Immune Syndrome)is a public health problem which is harmful to human health.This study retrospectively analyzed the data of HIV/AIDS in Henan province,we aims to establish a short term(5 years)and long-term(10 years)fatality prediction model of HIV/AIDS,the difference of classical Cox model and time-depedent Cox model in multivariate analysis were compared,the influence factors of fatality and survival time of HIV/AIDS were discussed,so as to prolong the survival time of HIV/AIDS,and provide a theoretical reference to improve the prevention and control effect of AIDS.MethodsA retrospective study was conducted to collect the data of HIV/AIDS in Henan Province during the period of 1995-2015.According to the predetermined inclusion and exclusion criteria to select the research object.An Excel database was created and used IBM SPSS Statistics 21.0 and SAS9.1 for statistical analysis.Logistic regression model was used to explore the influencing factors of 5 year and 10 year fatality of HIV/AIDS.We established the prediction models of 5 year and 10 year fatality of HIV/AIDS and used ROC curve to evaluate its diagnostic value.The life table method(life table)was used to estimate the survival probability of HIV/AIDS and draw the survival curve.We used log-rank test as the univariate analysis to compare the survival curve of different characteristics group.Statistically significant variable variables in univariate analysis were used to fit the Cox regression model and time-dependent Cox regression model for multivariate analysis.Results1 Analysis of influencing factors and prediction model evaluation of HIV/AIDS fatality: Influence factors of HIV/AIDS fatality in 5 years including gender(OR=0.747),age at diagnosis(OR=1.296),marital status(OR1=0.349,OR2=1.620),route of infection(OR2=2.306),CD4+ count level at baseline(OR=0.325),presenting clinical symptoms or not at baseline(OR=7.334),whether receiving HARRT at baseline(OR= 0.090),experience as a peer educator or not(OR=0.351);Influence factors of HIV/AIDS fatality in 10 years including gender(OR=0.684),age at diagnosis(OR=1.565),educational level(OR=0.833),marital status(OR1=0.486,OR2 =2.282),route of infection(OR1=0.468,OR2=2.501),CD4+ count level at baseline(OR=0.384),presenting clinical symptoms or not at baseline(OR=3.136),whether receiving HARRT at baseline(OR= 0.020);The area under the ROC curve of the 5 and 10 year fatality prediction models established by Logistic regression were 0.883(95%CI:0.874~0.891)and 0.938(95%CI : 0.933~0.944),respectively,The sensitivity and specificity were 79.1%,81.4% and 85.8% and 88.7% at the cut-off point.2 Influence factors of the survival time of HIV/AIDS: The Log-rank test showed the survival distribution of all variables except the nation ware statistically significant(P<0.05).The time-depedent Cox model showd a higher CD4+ count level at baseline(HR=0.301,95%CI:0.270~0.335)was a protective factor,it had the same effect for receiving HARRT at baseline(HR=0.301,95%CI:0.270~0.335),both the time-dependent effect of these two variables were significant(P<0.05).Conclusions1 we could use the Logistic regression analysis to establish the prediction model of short-term or long-term fatality for HIV/AIDS.In addition to gender,age at diagnosis,and other characteristics of their own as the influence factor of prognosis for HIV/AIDS,Raising the CD4+ count level in an earlier period,Receiving the HARRT earlier could reduce the risk of dying from AIDS related diseases.2 As an extension of Cox regression model,the time-dependent Cox model could estimate the time-dependent effect of the variables,provide more information of the model when a covariate does not satisfy the assumption of PH,which has a certain application value in prognosis analysis of HIV/AIDS.