Time-Dependence of Survival Predictions Based on Markers of HIV Disease

Abstract

To determine whether the ability of baseline clinical stage, viremia, and CD4 cell counts to predict mortality in HIV-1-infected patients changes with duration of follow-up. Three hundred ninety-four patients were followed for an average of 29 months by the Swiss HIV Cohort Study, a practice-based registry of HIV-1-infected patients in Switzerland. Predictor variables were the baseline clinical stage, CD4 cell count, circulating HIV-1 RNA level, and the RNA/CD4 ratio; the outcome was death. The changes in relative risks of death over time were examined using survival models that extend the Cox model to allow for nonproportionality of hazards. During 949 person-years of follow-up, 169 patients died (mortality rate 17.8 per 100 person-years). Compared with clinical stage A, patients in stages B and C at baseline had much higher mortality rates in the subsequent year. The prognostic ability of stage C decayed over time (P=0.03). By contrast, the relative risks associated with a 2-fold difference in CD4 counts remained remarkably stable, at ~0.6 (P=0.81 for the time-dependence test). Relative hazards associated with a 10-fold difference in HIV RNA per milliliter and in HIV RNA per CD4 cell both tended to increase over time, but this trend failed to reach statistical significance (P=0.21 and P=0.08, respectively). Time-dependence patterns of prognostic ability varied widely among predictors, displaying gradual decay (clinical stage), stability (CD4 cells), and a trend to progressive increase (viremia). These results may affect clinical monitoring of HIV-infected patients and the interpretation of cohort studies of HIV-infected patients.