Related Experiment Videos
Subgroup analyses in randomized trials: risks of subgroup-specific analyses; power and sample size for the
Sara T Brookes1, Elise Whitely, Matthias Egger
1Department of Social Medicine, University of Bristol, Whiteladies Road, Bristol, BS8 2PR, UK. sara.t.brookes@bristol.ac.uk
Journal of Clinical Epidemiology
|April 7, 2004
Summary
Subgroup analyses in clinical trials often yield unreliable results, even when formal interaction tests are used. The power to detect treatment-subgroup interactions is frequently insufficient, potentially leading to misinterpretation of findings.
Area of Science:
- Clinical Trials
- Biostatistics
- Epidemiology
Background:
- Guidelines recommend formal interaction tests for subgroup analyses in clinical trials.
- However, inappropriate subgroup-specific analyses persist, and trials often lack power for interaction detection.
Purpose of the Study:
- To quantify error rates in subgroup analyses.
- To assess the reliability of subgroup-specific tests versus formal interaction tests.
- To evaluate the power of interaction tests in trials designed for overall treatment effects.
Main Methods:
- Monte Carlo simulations were employed.
- Quantified risks of misinterpreting subgroup analyses.
- Assessed the power of interaction tests in relation to overall treatment effect detection.
Main Results:
- Subgroup-specific tests showed considerable unreliability, with significant effects in one subgroup observed in 7%–64% of simulations.
- A trial with 80% power for the overall effect had only 29% power for an interaction effect of the same magnitude.
- Detecting interactions with equivalent power to overall effects requires quadrupling sample size, with greater increases for smaller interactions.
Conclusions:
- The potential for spurious results from subgroup analyses may be underestimated.
- Formal interaction tests perform as expected, but subgroup-specific analyses are less reliable.
- Clinical trial design needs to account for the low power and potential for false positives in subgroup analyses.