During the interim analysis of a double-blind RCT comparing a novel tuberculosis regimen with standard therapy in 800 Indian TB patients, the Data Safety Monitoring Board (DSMB) observes a statistically significant difference in cure rates (new regimen: 92%, standard: 85%, p = 0.008) at 50% enrollment. The DSMB chair asks: what is the most appropriate next step?

Explanation

## Interim Analysis and Alpha Spending in RCTs **Key Point:** Interim analyses must be planned *a priori* in the protocol with pre-specified stopping rules and alpha-spending functions to maintain the overall Type I error rate at 0.05. A single interim p-value of 0.008 at 50% enrollment does **not** automatically justify stopping the trial. ### Why Continuing with Adjusted Alpha Spending is Correct (Option B) 1. **Alpha spending framework**: Pre-planned stopping boundaries (e.g., O'Brien–Fleming, Pocock) allocate the total Type I error budget across interim and final analyses. At 50% enrollment, the O'Brien–Fleming interim boundary is approximately **p < 0.001** — far stricter than the nominal p < 0.05. The observed p = 0.008 **fails** this interim boundary and therefore does NOT justify stopping. 2. **Prevents false positives**: Without adjustment, repeated testing inflates the probability of a Type I error well above 5%. Stopping at the first favorable signal risks declaring a false positive. 3. **Trial integrity**: The DSMB must apply the pre-specified stopping rule, not react to a nominally significant p-value. Continuing enrollment with the pre-planned adjusted alpha spending is the methodologically and ethically correct action. 4. **Regulatory alignment**: ICH-E9 guidelines and Indian GCP (Schedule Y) both require that interim analyses and stopping rules be pre-specified in the protocol. Ad hoc stopping violates these standards. ### Interim Analysis Framework | Rule | Interim Boundary (50% enrollment) | Final Boundary | Key Feature | |------|-----------------------------------|----------------|-------------| | **O'Brien–Fleming** | p < 0.001 | p < 0.048 | Conservative early; preserves final power | | **Pocock** | p < 0.016 | p < 0.016 | Uniform boundaries; less power loss | | **Fixed alpha (unadjusted)** | p < 0.05 | p < 0.05 | Inflates Type I error to ~10% | **High-Yield:** At 50% enrollment, the observed p = 0.008 does NOT cross the O'Brien–Fleming interim boundary (~p < 0.001). The correct action is to continue enrollment and apply the pre-planned alpha-spending function at the next analysis point. ### Why Other Options Are Wrong - **Option A (Stop immediately)**: Stopping at p = 0.008 without checking the pre-specified interim boundary violates alpha spending and risks a false positive. Premature stopping also overestimates treatment effects (Pocock & Simon effect). - **Option C (Unblind to investigators)**: Unblinding compromises the double-blind design, introduces performance and ascertainment bias, and is not indicated unless the stopping boundary is crossed. - **Option D (Exclude interim data)**: Discarding interim data wastes information, violates the pre-specified protocol, and is statistically and ethically unjustifiable. **Clinical Pearl:** The DSMB's role is to protect both efficacy *and* safety using pre-specified, adjusted stopping rules — not to stop at the first nominally favorable signal. In TB trials, premature stopping can expose future patients to a regimen whose superiority may be a false positive. **Mnemonic:** **DSMB** = **D**ata **S**afety **M**onitoring **B**oard uses **pre-specified, adjusted** stopping rules (ICH-E9 / Indian GCP compliant). *Reference: Friedman LM, Furberg CD, DeMets DL. Fundamentals of Clinical Trials, 5th ed.; ICH-E9 Statistical Principles for Clinical Trials; Indian GCP Guidelines (Schedule Y).*