Since most sites reported that patients were CD4 tested at least annually, CD4 monitoring was classified

into two categories: at least three times and fewer than three times per year. The two exceptions monitored patients at least annually when resources were available to do so. Clinical disease progression was determined as a new diagnosis of an AIDS-defining illness (CDC category C) or death from any cause. Patient follow-up commenced at HAART initiation and ended at date of death, AIDS-defining illness or most recent contact, whichever was the earliest. Surrogate endpoints were HIV RNA viral suppression (<400 copies/mL) and change Epacadostat in CD4 cell count from baseline at 12 months post-HAART. Surrogate

marker values closest to the target date were selected from windows of 9–15 months. Patients contributing data to each analysis are shown in Fig. 1. For eligible patients, baseline comparisons by country income (χ2, Fisher’s exact or Cochrane – Armitage test for trend) were performed as appropriate. Determinants of 12-month HIV RNA suppression and change in CD4 cell count were assessed via logistic regression and linear regression, respectively. Proportional hazards models were used to evaluate predictors of time to progression to new AIDS-defining illness or death. Analyses were based on an intention to continue treatment approach in that we did Proteasome inhibitor not take into account regimen changes, interruptions or failure post-HAART. Forward stepwise techniques were used to determine the best fitting models. To identify significant variables Thymidine kinase and important confounders, binary covariate P-values and multi-categorical parameter P-values (from tests for trend/heterogeneity) of <0.2, in univariate analyses, were considered for inclusion in multivariate models. Final multivariate models consisted of covariates remaining significant at the 0.05 level. For each endpoint, a base predictive patient model was determined from significant patient covariates. Then, because of our a priori interest in the role of site resourcing

on outcomes, individual estimates of country income and reported frequencies of VL and CD4 testing were assessed for statistical significance after adjustment for the base patient model. Analyses were performed using sas software version 9.1.3 (SAS Institute Inc., Cary, NC, USA) and stata software version 8.2 (STATA Corp., College Station, TX, USA). Of 3346 patients recruited to TAHOD, 2333 (69.7%) fulfilled the inclusion criteria. Of these, 79% had at least 6 months of retrospective data available and 13% were mono- or dual-ARV experienced. Patient demographics, clinical parameters and prescribed HAART regimen are summarized in Table 1. One hundred and seventy-six of the mono- and dual-experienced patients recycled one or two previously used ARVs in the HAART regimen.