OBJECTIVES:

Evaluations of medical technologies typically require extrapolation of results to new populations and time horizons. Furthermore, new frameworks such as generalized risk-adjusted cost-effectiveness make estimation of outcome distributions increasingly important. Rare disease settings are particularly challenging because sample sizes are small and there is often a high degree of heterogeneity in outcomes. We present a new approach for predicting real-world outcome distributions for first-in-class treatments of rare disease and consider endpoints based on assessment scores collected longitudinally.

METHODS:

Prediction of real-world outcomes proceeds in five steps. First, a Bayesian Markov ordered logit model represents the ordinal nature of the response variable and the persistence of scores over time. Second, the model is fit to real-world data comprised of control observations. Third, relative treatment effects are estimated from the pivotal randomized clinical trial. Since sample sizes are small, dynamic borrowing techniques are used to augment the control group with real-world data and the treatment group with phase 1 trial data. Fourth, the parameters of the Markov ordered logit model are updated using the estimated distribution of relative treatment effects. Fifth, a posterior predictive distribution for the target population at timepoints of interest using the new treatment is simulated using the Markov ordered logit model.

RESULTS:

The method is illustrated with predictions of real-world health outcomes in Duchenne muscular dystrophy.

CONCLUSIONS:

We propose an innovative Bayesian framework that is broadly applicable to any first-in-class treatment in which both clinical trial and external control data are available. In our application, we employ a Markov ordered logit model that is (to our knowledge) a novel method for parameterization of health economic models. Our approach ensures that simulated distributions of health outcomes are consistent with the underlying data and provides a new use case for dynamic borrowing techniques more commonly used in regulatory contexts.