Objective: HIV mutation accumulation has great implications for pharmacoeconomics and clinical care, yet scarcity of data has hindered its representation in decision analytic models. Our objective is to determine the accuracy with which mutation accumulation and other unmeasured parameters could be estimated during model calibration.
Methods: We used a second-order Monte Carlo simulation of HIV natural history that had been calibrated by varying two unmeasured parameters (mutation accrual rate and probability of adherence) to minimize differences between estimated and observed clinical outcomes (time to treatment failure and survival). We compared these estimated values first with only those results that had been already published at the time of model calibration, and second including results that were published after model calibration.
Results: The value for mutation accrual rate assigned during calibration was 0.014 mutations per month for antiretroviral-naïve patients, at the lower bound of the results for nine heterogeneous studies published at the time of calibration (pooled 95% confidence interval [CI] 0.014-0.039 mutations per month). In contrast, this estimate accurately anticipated results from 11 larger and more homogeneous studies published after calibration (pooled 95% CI for antiretroviral-naïve patients, 0.012-0.015 mutations per month). The value for probability of adherence assigned during calibration (75%) was also within the range of published results (pooled 95% CI 62-76%).
Conclusion: Estimates for unobserved parameters derived during model calibration were not only within the range of clinical observations, but anticipated with accuracy clinical results that were not yet available. It may be feasible to use models to estimate unobserved parameters.