Comparative study of predictive ability of AIDS incidence in HIV positive people using Markov model according to two criteria, WHO and CDC in CD4 cell categorization

Background: The Multi state Markov models have extensively application with categorization of laboratory marker of CD4 cells for evaluation of HIV disease progression. These models with different states result in different effects of covariates and prediction of HIV disease trend. The main purpose of this study was comparison of four and five states models with the three- state in order to select the model with better prediction ability of occurrence of HIV and finally death in HIV positive people. Methods: A total of 305 HIV positive people were included in this cohort study in the Iran AIDS center in Imam Khomeini Hospital in Tehran that entered during March 1995 to January 2005 and then fallowed up to October 2014. The three continuous- time Markov models of three-, four- and five- state models were fitted to data to describe the evolution of a HIV disease Trend over different states. For comparison of models, two criteria of modification of Akaike’s criterion (DRAIC) and likelihood cross-validation criterion (DRLCV) along with their 95% tracking interval was used. For fitting of these models and estimation of transition matrix and the hazard ratio of gender and treatment independent variables, the msm package of R project for statistical computing, version R 3.2.4 (www.r-project.org) was used. Results: The results showed that the four- state model has more prediction ability than five-state model for evaluation of HIV disease Trend. In the four-state model, the progression hazard ratio to death for people who received highly active antiretroviral therapy (HAART) was 0.64 lower than who didn’t get this therapy. Moreover, the progression hazard ratio for men was 2.33 fold in comparison to women. The disease progression hazard ratio to death was 4.9 fold for men in comparison to women. Conclusion: The (DRAIC) and (DRLCV) criterions showed that the four-state model has more predictive ability of the progression trend of HIV disease in comparison of five-state model.