Background:  We previously characterized HIV resistance development in a large population-based cohort of individuals (HOMER) starting their first HAART regimen by systematic genotyping. Here we describe a statistical model to predict the emergence of resistance, adjusting for the simultaneous effects of adherence, baseline plasma viral load, and initial treatment regimen.

Methods:  Logistic regression was used to create a statistical model of HIV drug resistance for drug-naive patients initiating HAART between August 1996 and November 2005 (n = 1975, incorporating 5137 resistance tests). The model included 818 individuals initiating non-boosted protease inhibitor (PI) regimens, 353 initiating boosted PI regimens, and 804 initiating non-nucleoside reverse transcriptase inhibitor (NNRTI) -based regimens. Several baseline predictors were incorporated in the model, including baseline plasma viral load, CD4 cell count, age, gender, history of injection drug use, and adherence estimated by prescription refill.

Results:  Of the total, 539 individuals (27%) developed resistance during a median of 4.6 years of follow-up (interquartile range 2.5 to 7.2). The model predicted no difference in the odds of developing key resistance mutations between non-boosted PI regimens (reference group) and NNRTI-based HAART regimens (odds ratio [OR] = 0.86, 95% confidence interval (CI) 0.68 to 1.08), but greatly reduced odds for PI-boosted regimens (OR = 0.23, 95%CI 0.16 to 0.34). A skewed, non-linear relationship with adherence was confirmed, as was a strong association of resistance with increasing plasma viral load and a weaker association with decreasing CD4 cell count and a history of injection drug use. Of particular interest, the dependence of resistance selection on baseline plasma viral load and on patient adherence was markedly decreased for boosted PI regimens; for example, the estimated probability of resistance at the worst adherence strata for boosted PI was equal to or lower than that observed at any adherence strata for unboosted PI or NNRTI regimens.

Conclusions:  The statistical model closely reflected the previously presented systematic characterization. In a “real world” population of individuals initiating first HAART, individuals starting on a boosted PI regimen had a 5-fold lower odds ratio of developing HIV drug resistance. Furthermore, the model shows that the relationship between resistance and adherence (as well as resistance and plasma viral load) is less strong for individuals initiating boosted PI regimens than it is for those starting non-boosted PI or NNRTI-based regimens.