Predicting Treatment Effects of a New Drug in a Target Population Based on Randomized Controlled Trial Results

Background: Treatment effects derived from randomized controlled trials may not be generalizable to real-world populations that have different distributions of effect modifiers. Predicting effects for a target population in clinical practice can provide useful insights into the expected benefits and risks, especially shortly after a drug's launch when observational data are scarce.

Objectives: We developed outcome prediction models leveraging individual-level trial data and used these models to predict treatment effects in a real-world US Medicare population.

Methods: We used data from the Randomized Evaluation of Long-Term Anticoagulation Therapy trial, which investigated the effect of dabigatran vs warfarin on the occurrence of stroke or systemic embolism (stroke/se) among patients with atrial fibrillation (AF). Outcome prediction models were developed within each treatment group by fitting Weibull parametric survival models in the trial data and were validated using bootstrapping. The target population was defined as Medicare fee-for-service enrollees with AF who initiated dabigatran or warfarin between 2011—2017. We predicted rates, rate differences (RD), and 95% confidence intervals (CIs) of stroke/se, cardiovascular death, major bleeding, gastrointestinal (GI) major bleeding, and minor bleeding based on the distribution of measured characteristics of the target population.

Results: Model calibration was good, and discrimination was adequate, e.g., optimism corrected Harrell's C-index of 0.63 (95% CI, 0.61—0.71) for dabigatran and 0.64 (0.62—0.70) for warfarin for stroke/se. The trial and target populations had similar mean CHADS2 score (2.2 [SD: 0.9] vs 2.2 [1.1]) but different mean age (71 vs 79 years). The predicted RDs were highly overlapping between RCT and target populations, indicating similar benefits of dabigatran vs warfarin for stroke/se (trial: -6.1 [95% CI: -9.2, -3.0]; target: -5.5 [-6.0, -5.0]) and cardiovascular death (-4.1 [-8.1, -0.1]; -2.1 [-2.6, -1.6]), although shifting RDs toward smaller benefits. The predicted RDs for major bleeding (-2.2 [-7.1, 2.8]; 1.6 [0.8, 2.4]), GI major bleeding (5.3 [2.3, 8.3]; 7.5 [7.0, 8.0]), and minor bleeding (-21.6 [-34.5, -8.8]; 6.1 [4.3, 7.9]) suggested more bleeding harms in the target population.

Conclusions: The predicted effect estimates suggested that, compared with warfarin, the expected benefits of dabigatran would be similar in terms of reducing stroke/se or cardiovascular death, but risks of bleeding would be greater in the target Medicare population compared with the effects in the trial population.