Systematic validation of disease models for pharmacoeconomic evaluations. Swiss HIV Cohort Study

Pharmacoeconomic evaluations are often based on computer models which simulate the course of disease with and without medical interventions. The purpose of this study is to propose and illustrate a rigorous approach for validating such disease models. For illustrative purposes, we applied this approach to a computer-based model we developed to mimic the history of HIV-infected subjects at the greatest risk for Mycobacterium avium complex (MAC) infection in Switzerland. The drugs included as a prophylactic intervention against MAC infection were azithromycin and clarithromycin. We used a homogenous Markov chain to describe the progression of an HIV-infected patient through six MAC-free states, one MAC state, and death. Probability estimates were extracted from the Swiss HIV Cohort Study database (1993-95) and randomized controlled trials. The model was validated testing for (1) technical validity (2) predictive validity (3) face validity and (4) modelling process validity. Sensitivity analysis and independent model implementation in DATA (PPS) and self-written Fortran 90 code (BAC) assured technical validity. Agreement between modelled and observed MAC incidence confirmed predictive validity. Modelled MAC prophylaxis at different starting conditions affirmed face validity. Published articles by other authors supported modelling process validity. The proposed validation procedure is a useful approach to improve the validity of the model.