Related Experiment Videos
The relation between treatment benefit and underlying risk in meta-analysis
S J Sharp1, S G Thompson, D G Altman
1Medical Statistics Unit, London School of Hygiene and Tropical Medicine.
BMJ (Clinical Research Ed.)
|September 21, 1996
Summary
Analyzing treatment benefits in clinical trials requires careful risk assessment. Common methods are flawed due to regression bias, leading to misleading conclusions about patient subgroup effectiveness.
Area of Science:
- Biostatistics
- Clinical Epidemiology
- Medical Research Methodology
Background:
- Meta-analyses commonly assess treatment benefit variation across patient risk groups.
- Identifying patient subgroups who benefit most or least from interventions is a key goal.
Purpose of the Study:
- To identify flaws in common meta-analysis methods for assessing treatment effect variation by patient risk.
- To propose more statistically appropriate methods for this type of analysis.
Main Methods:
- Critique of standard meta-analysis techniques using control group event proportions to estimate patient risk.
- Evaluation of alternative methods like using average event proportions or L'Abbé plots.
- Discussion of bias due to regression to the mean in these approaches.
Main Results:
- The conventional method of using control group event proportions as a risk measure is fundamentally flawed.
- Regression to the mean bias significantly distorts results, especially with small trials or low risk variability.
- Alternative methods like average proportions or L'Abbé plots also suffer from bias.
Conclusions:
- Common meta-analysis approaches for assessing treatment benefit variation by patient risk are unreliable.
- These flawed methods can lead to seriously misleading conclusions about intervention effectiveness in different patient groups.
- Statistically sound methods, preferably those analyzing baseline patient characteristics, should replace current practices.