OBJECTIVES: Different modeling approaches have been used to estimate the costeffectiveness of antipsychotics used to reduce psychotic symptoms of schizophrenia. This study systematically reviewed schizophrenia modelling studies, to examine the relationship between the modeling approach used in schizophrenia studies and their reported outcomes.
METHODS: A systematic literature review of MEDLINE, EconLit, Embase, and the Cochrane Library and an Internet search identified published results of schizophrenia modeling studies from 2000 to 2011. Two independent reviewers performed searches according to a prespecified protocol limited to English-language articles from any country.
RESULTS: Eighty-three publications reported 80 individual modelling studies that met the inclusion criteria. Fifty-seven studies reported results of 71 pairs of antipsychotic drug comparisons (drug A vs. drug B) as incremental cost-effectiveness ratios (ICERs), such as cost per quality-adjusted life-year (QALY), which allowed a comparison of results. The majority of the economic evaluations used a Markov (23 studies) or decision-tree model 23 studies); 9 studies used a discrete-event simulation (DES) model, and 2 studies used a microsimulation model. Among the 11 comparisons with contradictory results, we focused on the following drug comparisons of atypical antipsychotics with the most studies: risperidone long-acting injection versus oral olanzapine, oral risperidone versus oral olanzapine, oral risperidone or oral olanzapine versus ziprasidone, and oral olanzapine versus oral aripiprazole. Overall, model structure, time horizon, and patient population did not affect study results. Differences among studies with contradictory results generally reflected definition of response and relapse rates in the health states, validity of clinical data sources for transition probabilities, and assignment of utilities to estimate QALYs.
CONCLUSIONS: The cost-effectiveness results of the majority of models were in agreement regardless of the model structure. In models with contradictory results, most differences could be explained by definition of response, relapse, discontinuation, or adverse-event rates and/or by the selection of sources.
