Objective:

• Predict health outcomes for patients with Duchenne muscular dystrophy (DMD) treated with a novel gene therapy in a real-world setting.

Challenges:

• The gene therapy is being evaluated in a phase 3 randomized clinical trial (RCT) so there is no-real world data for patients treated with the gene therapy. Additionally, the characteristics of patients and conditions in clinical trials are different than in the real-world.
• DMD is a rare disease so sample sizes are small (99 patients across both arms in the phase 3 RCT).
• Follow-up [and representation of the real-world treatment population] is limited in available datasets, so predictions at longer time horizons [and addressing the entire real-world population of interest] require statistical extrapolation.
• The true statistical model relating patient and clinical characteristics to outcomes is unknown.

Methods:

• Predict outcomes by using real-world data for the standard of care (SoC) to estimate baseline risk and the RCT to estimate relative treatment effects.
• Build Bayesian model pipelines (BMPs) to optimize and evaluate the prediction model for SoC, automate the process of making predictions with new data, and facilitate quantification of parameter, structural, and sampling uncertainty.
• Leverage shrinkage priors to mitigate overfitting and improve out-of-sample prediction performance.
• Model baseline risk as a flexible function of time to facilitate extrapolation.
• Incorporate Bayesian model averaging into the BMPs to weight different plausible statistical models and quantify structural uncertainty.