OBJECTIVES: Non-alcoholic Steatohepatitis (NASH) is a disease with its early progression characterized by five stages of hepatic fibrosis from F0 (no fibrosis) to F4 (compensated cirrhosis). The fibrosis progression rate (FPR) between fibrosis stages has been estimated by Singh, 2015 for each baseline stage as the total number of stages progressed divided by the total number of person years of follow up. The FPR has been used in Markov cost-effectiveness models as a proxy for transition probabilities between fibrosis stages (Tapper, 2015; ICER, 2016). This research argues that model specific transition probabilities calibrated to match detailed study data give a better model of disease progression than FPR.

METHODS: The observed data include the number of patients for all 25 combination of initial and final fibrosis stages, and the total number of patient years for each initial stage. Our Markov-model assumed that transition is possible to the next stage in either direction that leaves 8 transition probability parameters to be calibrated. The calibration objective was to match the observed and modelled final patient number distribution at the modelled time horizon. For the calibration, the time horizon was set to the number of patient years for each initial stage.

RESULTS: The model generated distribution of patients closely match the observed distribution from the data source. The resulting transition probabilities are also consistent with the estimated FPR.

CONCLUSIONS: The calibration method for transition probabilities in Markov-models of NASH is recommended over the use of FPR. The method incorporate all available information from the data source instead of the FPR that is only a summary statistic of the observed progressions. Furthermore, the calibrated transition probabilities are fully consistent with the model by allowing for transitions in both directions from each state and multiple transitions from each state over the model time horizon.