Meta-regression detected associations between heterogeneous treatment effects and study-level, but not patient-level, factors

J Clin Epidemiol. 2004 Jul;57(7):683-97. doi: 10.1016/j.jclinepi.2003.12.001.

Abstract

Objective: Two investigations evaluate Bayesian meta-regression for detecting treatment interactions.

Study design and setting: The first compares analyses of aggregate and individual patient data on 1,860 subjects from 11 trials testing angiotensin converting enzyme (ACE) inhibitors for nondiabetic kidney disease. The second explores meta-regression for detecting treatment interaction on 671 covariates, including the baseline risk, from 232 meta-analyses of binary outcomes compiled from the Cochrane Collaboration and the medical literature.

Results: In the ACE inhibitor study, treatment effects were homogeneous so meta-regression identified no interactions. Analysis of individual patient data using a multilevel model, however, discovered that treatment reduced glomerular filtration rate (GFR) more among patients with higher baseline proteinuria. The second investigation found meta-regression most effective for detecting treatment interactions with study-level factors in meta-analyses with >10 studies, heterogeneous treatment effects, or significant overall treatment effects. Under all three conditions, 46% of meta-regressions produced strong interactions (posterior probability >0.995) compared with 6% otherwise. Baseline risk was associated with the odds ratio in 6% of meta-analyses, half the rate found using maximum likelihood.

Conclusion: Meta-regression can detect interactions of treatment with study-level factors when treatment effects are heterogeneous. Individual patient data are needed for patient-level factors and homogeneous effects.

Publication types

  • Research Support, U.S. Gov't, P.H.S.

MeSH terms

  • Adult
  • Algorithms
  • Angiotensin-Converting Enzyme Inhibitors / therapeutic use
  • Bayes Theorem
  • Female
  • Humans
  • Kidney Diseases / drug therapy
  • Male
  • Meta-Analysis as Topic*
  • Middle Aged
  • Models, Statistical*
  • Randomized Controlled Trials as Topic / methods
  • Regression Analysis
  • Treatment Outcome*

Substances

  • Angiotensin-Converting Enzyme Inhibitors