Interim analyses and early stopping
- Concept
Looking at a trial's data before it ends lets it stop for clear benefit, harm or futility, but every look has a cost.
What it is
An interim analysis reviews a trial's results while it is still running. It is done by an independent data monitoring committee, with rules fixed in the protocol, to protect participants: the trial can stop for benefit, harm or futility. Because every look is one more test, strict thresholds are used (group sequential designs); and trials stopped early for benefit tend to overestimate the effect.
Key facts
- Overall chance of a false positive when testing repeatedly at p < 0.05: 2 looks ≈ 8 %, 3 ≈ 11 %, 5 ≈ 14 %.
- O'Brien-Fleming-type rules require very small p values at early looks and close to 0.05 at the final one.
Worked example
- Trial planned for 2 000 patients with two interim analyses; rule: stop for benefit only if p < 0.001.
- First analysis at 600: deaths 15/300 (5 %) versus 30/300 (10 %). RR = 0.05 / 0.10 = 0.50.
- Chi-square: 22.5 deaths expected per arm; (7.5² / 22.5) × 2 + (7.5² / 277.5) × 2 = 5.0 + 0.41 ≈ 5.41; p ≈ 0.02.
- p ≈ 0.02 does not cross the 0.001 threshold: the committee recommends continuing.
- At the end: 70/1 000 (7 %) versus 95/1 000 (9.5 %); RR = 0.07 / 0.095 ≈ 0.74.
- Reading: stopping at the first look would have reported a 50 % reduction instead of the final 26 %.
Reasons to stop a trial
| Reason | What the data show | What is required | Risk |
|---|---|---|---|
| Benefit | Large difference favouring the intervention | Crossing a very strict, pre-set threshold | Overestimating the effect |
| Harm | More adverse events or deaths in the treated arm | A looser threshold: safety weighs more | Stopping for a chance excess |
| Futility | The effect is so far off it will not be shown even if the trial continues | Low conditional probability of success | Missing a small but real effect |
Markers
- Monitoring committee
- Futility
- O'Brien-Fleming
- Sequential design
In practice
- Trials with serious outcomes (death, myocardial infarction) almost always have a monitoring committee.
- A trial stopped early for benefit with few events deserves caution when read.
- Adaptive designs allow, under pre-set rules, changes to sample size or dropping arms.
Common pitfalls
- Looking at the data with no plan and stopping when p drops below 0.05: high risk of a false positive.
- Investigators knowing the interim results: they may change how they recruit or measure.
- Trusting the effect size of a truncated trial as if it were final.
Test yourself
Think of the answer, then tap the question to see it.
What are the three reasons for stopping a trial early?
Clear benefit, harm to participants, and futility (the effect can no longer be shown even if the trial continues).
Why are thresholds stricter than p < 0.05 used at interim analyses?
Because each analysis is one more test; without correction the overall chance of a false positive exceeds 5 %.
Why do trials stopped early for benefit tend to exaggerate the effect?
They usually stop at a moment when chance favoured the intervention, with few events; with more follow-up the effect drifts back towards its true value.
Connections
Educational content on epidemiological and statistical methods; the examples use made-up numbers to teach the calculation. Not a clinical or legal guide: current Colombian rules are in the official sources.
Sources: Establishment and Operation of Clinical Trial Data Monitoring Committees (U.S. Food and Drug Administration, 2006); Adaptive Design Clinical Trials for Drugs and Biologics Guidance for Industry (U.S. Food and Drug Administration, 2019); Chapter 8: Assessing risk of bias in a randomized trial (Cochrane, 2024).
Spotted a mistake?We are always improving for the community. Tell us what is wrong and our team will review it.Report a mistake
Representative model, for educational purposes only.