A Western Blot-based Logistic Regression Model To Identify Recent HIV-1 Infections

Objective To establish a new technique to identify recent HIV-1 infections based on WB band patterns and relative grey ratio.Methods Collect study participants in HIV voluntary counseling testing clinics of Centers for Disease Prevention and Control(CDCs)of Liuzhou,Qinzhou,Guigang,Chongzuo cities.Participants were assigned into two separate sets(training set and validation set)according to their entering time points.Both recent and previous HIV-1 infections were included in each set of the study population.Their baseline characteristics and peripheral blood samples were collected by qualified investigators.Western blot confirmatory test for HIV-1 and BED-CEIA were performed using plasma separated from the peripheral blood sample.Alpha View Software was applied to analyzed relative grey ratio of WB bands.Spss16.0 was used to conduct Mann Whitney test and Pearson chi-square test between recent infection and previous infection on the relative grey ratio and frequency of WB bands presented.Med Calc software(version 11.4.2.0)was utilized to analyze for ROC curve of the model and BED-CEIA.Results1.A total of 153 eligible subjects were included in the study,with 81 in training set and 72 in validation set.In training set,43 were recent HIV-1infections and 38 were previous HIV-1 infections.In validation set,the number was 26 and 46,respectively.Of the total subjects,139(90.85%)of the participants were male and only 14(9.15%)were female.Age ranges were 19 to60 years old with an average of 42.0±12.0.Differences of gender,age and treatment status between recent HIV-1 infections and previous HIV-1 infections were not significant in both sets(p>0.05).The location distribution was,however,significant different between recent HIV-1 infections and previous HIV-1 infections(p<0.05).2.Estimated duration of infection(EDI)of recent and previous HIV-1infections in training set was 183.5±66.8 days and 1699.2±1182.2 days,respectively.The numbers in validation set was 212.0±83.0 days and 1349.6±723.9 days,respectively.There existed no significant difference between recent HIV-1 infections in training set and validation set,neither in previous infections(p>0.05).CD4+T cell counts were also not significantly different between recent and previous infections in both training and validation sets(p>0.05).3.Most of the recent HIV-1 infections were infected through homosexual transmission(69.57%),and the majority of previous infections were infected through heterosexual behaviour(58.33%).Totally,heterosex was the leading route of transmission(45.35%),followed by homosex(32.68%),and intravenous drug transmission was the least(21.57%).4.There was a similar trend in the WB bands present frequency of recent and previous HIV-1 infections between the training set and validation set.The bands of gp41,gp120 and gp160 were observed in all of the participants.For the rest bands,the frequency in recent infections was lower than that of previous infections,but significant difference was only observed in p55 band.5.Relative grey ratio of gp41 and gp160 showed significant difference between rencent and previous HIV-1 infections in both training and validation sets(p<0.05).6.There was a significant difference in univariate analysis of relative grey ratio on bands of p24,p39,p31,p6,gp41,gp160 and the present status of p55band in training set.Four of the variables were stepwise included into the multivariate regression model,which were the present status of p55(X1),relative grey ratio of p31(X2),gp41(X2)and gp160(X4),achieving following equation:Logit(z)=-0.731-1.098*X1-1.477*X2-2.117*X3-2.822*X4.The overall fit of the model was significant(X~2=65.760,p=0.000).Sensitivity,specificity and agreement rate were 93.02,92.11 and 92.59%in training set,respectively.In validation set,the rates were 88.46,84.78 and 86.11%,respectively.Effect of gender,age,treatment,CD4+T cell counts,infection routes and residency on the model was not observed.7.Sensitivity,specificity and agreement rate of BED-CEIA were 76.74,86.84 and 81.48%in training set,respectively.In validation set,the rates were50.00,84.78 and 72.22%,respectively.8.The accuracy of WB band-based multivariate logistic regression model in identifying recent HIV-1 infection was higher than that of BED-CEIA method(p<0.05).The overall misclassification rate of WB band-based model was10.46%,lower than that of BED-CEIA method(22.88%).Conclusions Multivariates Logistic regression model according to WB band-based relative grey ratio and frequency has better performance in distinguishing recent HIV-1 infections from previous infections.In addition,the model was not affected by gender,age,treatment CD4+T cell counts and residency.Therefore,it is worthy of further study and introduction to practical application to solve the problem of identification of recent HIV-1 infections.