Importance Geographic atrophy (GA) enlargement rates are generally stable over time for any given eye, a property which may help improve GA clinical trial efficiency by preserving statistical power with smaller sample sizes. One strategy is to incorporate a run-in phase into randomized clinical trials (RCT) to collect natural history data before randomization, while another strategy is a single-arm trial, in which all participants receive the study drug and enlargement rates are compared between run-in and treatment phases. Objective To evaluate whether run-in phase RCTs or single-arm trial designs might improve trial efficiency using power analyses and trial simulations. Design, Setting, and Participants This diagnostic/prognostic study uses in silico clinical trial power analyses. Parameters were derived from participants in the GA Minocycline Trial with baseline GA area of 2.5 to 17.5 mm 2 . Data were analyzed from November 2025 to March 2026. Main Outcomes and Measures The primary outcome was the required sample size for a given power, using square root GA area as the trial outcome. Sample size requirements for 80% and 90% power were estimated using linear mixed-model analyses and validated empirically by trial simulations (10 000 replications), assuming 25% treatment effect and α = .05. First, RCTs (1:1 randomization) with vs without a run-in phase were compared, keeping total follow-up constant. Second, RCTs were compared with a single-arm trial of equal duration. Results The Geographic Atrophy Minocycline Trial comprised 133 visits, 28 eyes, and 28 participants. The sample size estimate was smaller for an RCT with vs without a run-in phase. Sample size was incrementally smaller with longer run-in duration. For 2-year trials with 90% power, the sample size was 156 for an RCT vs 82, 40, and 26 for RCTs with run-in phases of 3, 6, and 9 months, respectively. Sample size estimates were smaller for single-arm trials than RCTs. The sample size for a 2-year single-arm trial (9-month run-in phase and 15-month treatment) was 14 vs 26 for an RCT (same phase durations). For all comparisons, similar patterns were observed for trials with 1-year duration, 80% power, nonsubfoveal GA only, or untransformed enlargement rates. Conclusions and Relevance For GA clinical trials, this diagnostic/prognostic study suggests incorporating run-in phases into RCTs or using single-arm designs can improve trial efficiency by exploiting the inherent stability of GA enlargement rates. These sample size estimates may apply to a wide range of GA trial designs.
Hou et al. (Thu,) studied this question.