Purpose: To analyse data from the OPTIMA trial to estimate input parameters in a decision model assessing cost-effectiveness in patients with advanced HIV for whom standard therapies had failed. The input parameters required to estimate such models often differ from the summary statistics on treatment efficacy available from trials. However, the resulting information is valuable to a range of potential analyses.
Methods: We conducted survival analysis to model lifetime disease progression. Events of interest were time to death and time to first AIDS defining event (ADE). By considering events both individually and jointly, we were able to calculate appropriate conditional probabilities. Assumptions were required to (i) link continuous time events to laboratory measures taken at discrete intervals; (ii) address missing data; and (iii) determine appropriate parametric distributions for survival time. We determined CD4 and viral load levels at time of ADE or death by last value carried forward. Examination of Kaplan-Meier curves, information criterion and the requirements of the model informed the selection of distribution.
Results: Table 1 reports hazard ratios for time to first ADE and time to death. Baseline covariates describe prognosis for different sub-groups. Time-varying covariates illustrate how risk of events change as a patient's disease progresses. Risk of ADE or death decreases as CD4 increases, and increases with viral load.
| Table 1 | Baseline covariates | Time-varying covariates | ||
| Covariate | ADE | death | ADE | death |
| Absolute CD4 (per 100 cell change) | 0.53** | 0.49** | 0.43** | 0.51** |
| Log viral load | 1.16* | 1.17* | 1.21** | 1.08 |
| AIDS at baseline | 1.32 | 1.47* | - | - |
| 1 SAE | - | - | 1.03 | 1.22 |
| 2+ SAEs | - | - | 1.03 | 3.6* |
Conclusions: Utilising exponential distributions where appropriate produced output more suited for use in Markov models. The significance of cumulative number of SAEs signified the need for additional health states to more accurately model survival. Inclusion of baseline covariates allowed disease progression to be modelled for different sub-groups. These survival analyses were complemented by further analyses of the OPTIMA data regarding costs, quality of life, and duration of treatment response. The resulting decision model can evaluate new anti-retroviral therapies in an advanced HIV population less often recruited to clinical trials.