Application of logistic regression model to analyze outcome of HIV Western Blot-indeterminate patients

ObjectiveTo use logistic regression model to analyze the relationship between the initial screening results, WB bands and positive conversion in follow up in persons with indeterminate HIV antibodydetection results and provide evidence for the improvement of the prediction of patient outcome.MethodsThe results of ELISA (Bio-Rad), a rapid kit (Abbot Determine HIV-1/2), WB band in 2012–2016 were collected. The threshold of ELISA was determined by ROC curve. The risk factors of western blot bands and the predictive model were established by logistic regression.ResultsAmong 79 indeterminate samples, the positive coincidence rate of the first two initial screening results was as high as 96.29%. The specific threshold value of the Bio-Rad was 8 by ROC curve. S/CO＞8 indicated a high probability of positive follow-up. The most common band patterns of HIV antibody-indeterminate samples were p24, gp160＋p24, p160＋gp120＋p24 respectively. The results of the follow-up confirmatory tests showed that 77.27% of p24 band pattern was HIV antibody-negative and the positive rate of gp160＋p24 with or without other band was high.Logistic regression model was used to screen the three significant factors of gp160, p24 and p17.ConclusionThe positive initial results, S/CO＞8, gp160＋p24 had or had no other brands, the follow-up results were positive.Combined initial screening and the band results could give an objective and accurate prediction.