Augmenting treatment arms with external data through propensity-score weighted power-priors with an application in expanded access 205 9♥ Table 2: Characteristics of the patients participating in the vemurafenib expanded access program (EAP) and trial. Clinical Program Characteristic EAP, N = 2411 TRIAL, N = 1321 Age at enrolment 53 (13) 50 (15) Gender assigned at birth Female 51 (39%) Female 95 (39%) 51 (39%) Male 146 (61%) 81 (61%) Melanoma stage M1a 22 (9.1%) 33 (25%) M1b 26 (11%) 18 (14%) M1c 182 (76%) 80 (61%) Unresectable Stage III 11 (4.6%) 0 (0%) ECOG performance status Grade 0 112 (46%) 61 (46%) Grade 1 98 (41%) 71 (54%) Grade 2 30 (12%) 0 (0%) Grade 3 1 (0.4%) 0 (0%) Objective Response Rate 129 (54%) 75 (57%) 1Mean (SD); n (%) Analysis and outcome In the first stage of our analysis, we estimate the probability of patients being in the trial conditional on their baseline characteristics. Frail patients (with a ECOG score ≥ 2) were not allowed to participate in the trial and we expect these patients not to be integrated in our analysis. The propensities are depicted in Figure 6. The 42 patients with weight 0 on the lefthand side are indeed all 31 patients with ECOG 2 and 3, as well as 11 additional patients with a baseline melanoma stage of ’Unresectable Stage III’ the latter category was also not present in the trial. The primary outcome of the trial and the expanded access program was the Objective Response Rate (ORR), defined uniformly as the fraction of patients with a complete response (CR) or partial response (PR). The estimate of ORR in the trial was 53% (95% confidence interval (CI): 44%-62%). In a Bayesian reanalysis, given a U(0,1) prior for the ORR, the posterior is Beta(75+1, 132-75+1) distributed with posterior mean 56.7 % and 95% posterior credible interval (48.3%, 65.0%). The estimated ORR in the expanded access program was 54% (95% CI: 47%-60%), in a Bayesian reanalysis leading to a posterior that is Beta(129+1, 241129+1) with a mean of 53.5% and 95% posterior credible interval (47.2%, 59.7%). The analysis with the ProPP leads to a posterior mean of 56.4%, with a 95% posterior credible interval of (49.4%, 63.3%). The different methods are depicted in Figure 7.
RkJQdWJsaXNoZXIy MTk4NDMw