Interim Analysis

Clinical trials are long and expensive. Sometimes, researchers peek at the data halfway through. This is an Interim Analysis.

If the drug looks amazing, they might stop the trial early. "It is unethical to withhold this cure from the control group!"

The Problem

Randomness fluctuates. At the start of a race, the underdog might be leading. If you stop the race then, the underdog wins.

If you stop a trial at the exact moment the drug looks best, you lock in a "high score" that is partly luck.

Studies stopped early for benefit consistently overestimate the treatment effect. When repeated, the effect shrinks.

The Rules

You can do interim analyses, but you must pay a penalty. You must make your p-value threshold much stricter (e.g., p < 0.001 instead of 0.05). This is the O'Brien-Fleming rule.

The Diagnosis

If a "breakthrough" drug comes from a trial that was stopped early, be careful. The real-world effect will likely be smaller.