01_capture_binary

Example 01 — CAPTURE (1997)

Trial

The CAPTURE Investigators (1997). Randomised placebo-controlled trial of abciximab before and during coronary intervention in refractory unstable angina: the CAPTURE Study. Lancet 349:1429-1435.

Phase 3 trial of abciximab vs placebo in refractory unstable angina patients planned for PCI. 30-day composite endpoint (death + MI + urgent reintervention).

Published design

• α = 0.05 (two-sided), 80% power, 1:1 randomization, fixed-sample.
• Assumed 30-day event rate 15% (placebo) vs 9% (active).
• Planned N = 1,400 (sponsor inflated ~30% over textbook minimum for protocol-deviation buffer; the agent gets the textbook number).

Reproduction

R -e 'source("examples/01_capture_binary/run.R")'

Expected: ~960–1,100 patients (textbook minimum at the published assumptions). The 1,400 published number reflects sponsor padding documented in the SAP.

What this example demonstrates

• design_binary with comparison = "superiority" and design_class = "fixed".
• Reasoning chain populated with two entries — one fda_guidance for alpha, one llm_precedent for the control event rate. Demonstrates the citation-trail pattern.