A pharmaceutical company sponsors a double-blind RCT of a novel antihypertensive agent versus placebo in 300 Indian adults with stage 2 hypertension. The primary outcome is change in systolic blood pressure (SBP) at 8 weeks. At week 4, an interim analysis reveals that the treatment arm has achieved a mean SBP reduction of 18 mmHg versus 6 mmHg in the placebo arm (p = 0.008). The Data Safety Monitoring Board (DSMB) is asked whether the trial should continue to completion. The DSMB notes that the pre-specified stopping rule was p < 0.001 for interim efficacy. What is the most appropriate recommendation by the DSMB?

Explanation

## Interim Analysis and Multiple Testing in RCTs **Key Point:** When a pre-specified stopping rule is NOT met at interim analysis, the trial must continue to completion. O'Brien-Fleming boundaries — not Bonferroni correction — are the gold-standard method for adjusting the final p-value threshold in trials with planned interim looks. ### Understanding the Interim Analysis Problem Conducting multiple looks at accumulating data inflates the Type I error rate (false positive risk). To control this, stopping rules and statistical adjustment methods must be **pre-specified** in the Statistical Analysis Plan (SAP) before the trial begins. | Finding | Interpretation | |---------|----------------| | Interim p = 0.008 | Suggestive but does **NOT** meet the pre-specified stopping threshold of p < 0.001 | | Pre-specified stopping rule: p < 0.001 | Conservative threshold protecting against Type I error inflation from interim looks | | DSMB recommendation | Continue to planned completion; apply O'Brien-Fleming adjustment to the final analysis | ### Why Option D (O'Brien-Fleming) Is Correct O'Brien-Fleming (O-F) boundaries are the **regulatory standard** (FDA, EMA) for group sequential trials: - They use a **very stringent threshold at early interim looks** (e.g., p < 0.001 at the midpoint), preserving most of the α budget for the final analysis - The **final analysis threshold is only slightly reduced** (e.g., p < 0.048 instead of p < 0.05), maintaining statistical power - They are pre-specified and control the overall Type I error at α = 0.05 across all looks - The current interim result (p = 0.008) is consistent with the O-F framework — it is below the final threshold but above the interim stopping boundary, so continuation is appropriate ### Why Option A Is Wrong The interim p-value (0.008) does **not** meet the pre-specified stopping rule of p < 0.001. Stopping early without meeting the pre-specified boundary violates protocol integrity, inflates Type I error, and may overestimate treatment effect (the "winner's curse"). The DSMB must adhere to the pre-specified plan. ### Why Option B (Bonferroni) Is Inferior Bonferroni correction divides α equally across all looks (e.g., α/2 = 0.025 per look for 2 analyses). This is **overly conservative** — it reduces power unnecessarily at the final analysis and does not reflect the asymmetric spending logic of group sequential designs. O'Brien-Fleming is preferred because it spends very little α early and preserves power at the final look. Bonferroni is also not the pre-specified method in this scenario. ### Why Option C Is Wrong Reducing sample size mid-trial without a formal adaptive design pre-specification is a **major protocol violation**. It compromises statistical power and introduces bias. No credentialed DSMB would recommend this. ### DSMB Decision Framework ``` Interim result meets pre-specified stopping boundary? ├── YES (p < 0.001) → Stop for efficacy; offer drug to placebo arm └── NO (p = 0.008 ≥ 0.001) → Continue to completion └── Apply O'Brien-Fleming adjustment to final analysis ``` **High-Yield (Park's Textbook of Preventive & Social Medicine / Friedman's Fundamentals of Clinical Trials):** O'Brien-Fleming group sequential boundaries are the most widely used method in regulatory RCTs. They maintain overall α = 0.05 while allowing interim monitoring without inflating false-positive rates. Bonferroni is a simpler but less efficient alternative not suited for group sequential designs. **Mnemonic: DSMB Responsibilities — SAFE** - **S**afety monitoring (adverse events) - **A**dherence to protocol - **F**utility assessment (stopping for lack of efficacy) - **E**fficacy assessment (stopping for overwhelming benefit)