Decision analytic modeling tpdm 1 assessing the predictive ability of a deterministic model and stochastic model

Formulary decisions are often based upon the safety, efficacy, and projected costs of medications. Models used to predict costs are rarely assessed to determine their predictive ability. OBJECTIVE: The purpose of this study was to assess the ability of a decision analytic deterministic model and a regression analytic stochastic model to predict the diabetesspecific costs incurred during the 12 months after metformin was added to an HMO formulary. The ability of the stochastic model to predict total healthcare costs was also assessed. METHODS: The deterministic model, a decision tree, was constructed within an equilibrium framework using literature-based probabilities and internal costs to predict diabetes-specific costs. A regression model was constructed using medical and pharmacy claims data to predict diabetes-specific cost and total healthcare costs. RESULTS: The total diabetes-specific cost estimate predicted by the decision analytic model came within 5% of actual costs. The model underestimated the diabetes-specific medical costs and overestimated the diabetes-specific pharmacy costs. The diabetes-specific regression model produced an estimate that was within 18% of the actual costs. The total healthcare cost model estimate was within 7% of the actual costs. The total and diabetes-related medical cost estimates were within 6% and 46% of the actual costs respectively. The total and diabetes-specific pharmacy costs were within 20% and 49% of the actual costs, respectively. CONCLUSION: Further research is needed to refine model construction techniques. A decision tree constructed with internal data should be used to predict disease-specific costs when only medical and pharmacy claims data from the previous year are available, and a regression model should be used to predict total healthcare costs.