Abstract
Background and aims
Spontaneous intracerebral hemorrhage is a dynamic, multi-stage condition with marked heterogeneity in hematoma volume and location and clinical time windows, motivating adaptive trial strategies. Using CHAIN as an illustrative case, we assessed whether a Bayesian adaptive enrichment design could lead to conclusions different from a conventional fixed design by identifying and prioritizing potential responder subgroups.
Methods
We fitted separate models for subgroup and overall treatment effects, using a Gaussian likelihood for the standardized outcome. Under limited effective sample sizes and potentially time-varying effects, we compared scenarios with 2 vs 3 interim looks, stopping thresholds of 20% vs 30%, and three alternative resampling strategies. Six pre-specified variables were tested one at a time as enrichment targets, and one extra exploratory subgroup was derived from literature review and expert input.
Results
A total of 72 sets of results were generated. Enrichment was triggered in all scenarios. In the 12 scenarios of Location, lobar was enriched and ultimately obtained positive results, including early effective cessation. All other scenarios obtained negative results. In 80.6% (58/72) of scenarios, the adaptive enrichment design produced outcomes different from the original result. As one scenario, under the 25/50/75% three-look design (20% stopping threshold; full-period resampling), the exploratory volume-strata analysis, enrichment favored the 15–50 mL subgroup, at the third interim analysis estimated 0.06 (95% CrI 0.02–0.10), indicating a positive signal.
Conclusions
Bayesian adaptive enrichment can support timely identification of clinically plausible responder subgroups in ICH and may lead to different trial decisions, but depends critically on prespecified, evidence-grounded enrichment targets.
Conflict of interest
All authors declare that they have no conflicts of interest.