Meta-Regression for Dose-Response Research

Meta-analysis is frequently used as a quantitative method for synthesizing the results from a set of related studies and providing an overall summary of the effectiveness of a treatment or intervention. Yet, it provides little or no information on why there is difference across studies. Meta-regression can help estimate interaction between covariates and treatment, explain the heterogeneity among studies, and, with this knowledge, then determine when, where, for whom and what treatment is beneficial. Covariates can be characteristics of studies (e.g., intensity, frequency and duration of exercise intervention), or study-level summaries of participants' characteristics (gender, ethnicity, age, etc.). The choice of covariates should be based on the knowledge of the subject matter and the number of covariates should be controlled to avoid a false-positive result. The statistical models for meta-regression analysis could be grouped as “fixed-” or “random-effects” models. A fixed-effects model does not take into account variability in between-study differences. To examine “residual heterogeneity,” a random-effects model should be employed. The causes of heterogeneity should be examined via the inclusion of covariates at both person and study levels. Software for meta-regression includes SAS, S-Plus, STATA, MLwiN, HLM etc. After key concepts and procedure of meta-regression are introduced, how to use meta-regression for dose-response research will be illustrated step-by-step using the example of strength training intervention for hypertension reduction.