OBJECTIVES. We discuss tradeoffs and errors associated with alternative approaches to modeling health economic decisions. Through an application in pharmacogenomics (PGx), we assessed model accuracy, optimal decisions and computation time for an identical decision scenario modeled four ways.

METHODS. We developed four models to assess decisions to use utilize genetic testing to guide drug selection for individuals with a genetic variant that affects metabolism of a chronic disease medication: (1) coupled-time differential equations [DEQ]; (2) a cohort-based discrete-time state transition model [MARKOV]; (3) an individual discrete-time state transition model [MICROSIM]; and (4) discrete event simulation [DES]. For stochastic models (DES, MICROSIM) we considered simulation sizes up to 10 billion patients. For discrete-time models (MARKOV, MICOSIM) we considered three cycle lengths (1 day, 1 month, 1 year) and alternative approaches to embedding literature-based rates as probabilities.

RESULTS. Based on the gold-standard DEQ model, PGx testing yielded an ICER of $103,212/QALY vs. a reference no-genetic-testing strategy. At a willingness-to-pay threshold of $100,000/QALY, MARKOV with transition probabilities converted using standard formulas (P(t)= 1 – e^-rt) resulted in different optimal decisions depending on the cycle-length used, while MARKOV results were nearly identical to DEQ for all cycle lengths when transition probabilities were embedded using a transition intensity matrix. Among stochastic models, model convergence with DEQ was achieved with substantially fewer simulated patients for DES (1 million) vs. MICROSIM (1 billion). Model run-time was fastest for deterministic solutions (DEQ, MARKOV), and was considerably faster for DES (median=48.2sec for 10 million patients) vs. MICROSIM (median=1,198sec for 10mil patients and an annual cycle).

CONCLUSIONS. Properly embedded MARKOV models provided the most favorable mix of accuracy and run-time. However, improper embedding of transition probabilities in using widely-used rate-to-probability conversion formulas for state-transition models was shown to yield different optimal decisions depending on the cycle length used.