May 21, 2002Statistics in Medicine
Statistical methods for assessing the influence of study characteristics on treatment effects in ‘meta‐epidemiological’ research
Authors
Discussion
Loading...
Member takes
Implication
Methodological evaluation demonstrates that accounting for heterogeneity and confounding alters bias estimates in meta-epidemiology, highlighting the need for meta-meta-analytic models.
Key Points
- To evaluate statistical approaches for measuring how trial characteristics and biases influence treatment effect estimates in meta-epidemiological studies.
- Analyzed data from evaluations of publication bias (39 meta-analyses, 394 published and 88 unpublished trials) and language bias (29 meta-analyses, 297 English and 52 non-English trials).
- Compared standard logistic regression models (with and without robust standard errors) against a meta-meta-analytic framework adjusting for between- and within-meta-analysis heterogeneity and trial characteristic confounding.
- Both within-meta-analysis and between-meta-analysis heterogeneity substantially influence treatment effect estimates and variance in meta-epidemiological analyses.
- Clear confounding exists between publication status and trial quality characteristics, demonstrating the necessity of adjusting for multiple study-level factors simultaneously.
Cite This Study
Sterne et al. (2002) studied this question.