OBJECTIVES: To characterize the progression of Duchenne muscular dystrophy (DMD) for healthcare decision makers, a natural history model (NHM) of disease progression across eight health states has been developed by Project HERCULES. Estimation of transition probabilities for this model, using traditional approaches, has been challenging because of limited longitudinal data. We applied an advanced estimation approach to refine, derive, and calibrate transition probabilities based on real-world DMD data.

METHODS: Published age distributions per health state were used to estimate the durations of time spent in each state using the additivity of cumulants and the Markov property of the NHM. Probability curves for health state membership over time were estimated with the method of moments using four distributions (exponential, gamma, lognormal, Weibull). Resulting survival curves across states were compared with time-to-event curves extracted from published DMD sources. The fitted model closest in aggregate to the published sources was selected as having the best fit.

RESULTS: Patients spend on average 11.0 years in ambulatory states, 1.5 years in the transfer state, and 15.7 in non-ambulatory states, with a mean survival of 28.3 years. According to the Weibull-based probability curves, which gave the best fit, median ages at loss of ambulation, full ventilation, and death are 11.0 years (95% confidence interval (CI): 6.1, 16.1), 19.7 years (95% CI: 12.6, 27.8) and 27.9 (95% CI: 17.8, 41.2), respectively, which is consistent with recent published literature. These results are similar across the gamma and lognormal distributions, and the corresponding probability curves aligned better with real-world data relative to the exponential distribution.

CONCLUSIONS: Using the cumulant method to bridge data gaps yielded estimated health state transitions that accorded with multiple published sources. This congruency will be helpful to inform and enhance the use of the HERCULES model for decision-making involving cost-effectiveness analyses and health technology appraisals.