Theoretical and Empirical Limits in Synthesis: Findings from Integrating Individual Data Across Similar and Diverse Randomized Trials

Title: Theoretical and Empirical Limits in Synthesis: Findings from Integrating Individual Data Across Similar and Diverse Randomized Trials

In standard meta-analysis, summary statistics from similar randomized trials are combined by standardizing outcomes, most often on effect size or odds ratio scales, and then adjusting for trial level covariates in meta-regression. Typically meta-analyses are limited to main effect analysis. With syntheses that are built around individual level data across multiple trials, there is more opportunity to conduct more refined analyses, including rare subgroup analyses, moderation analyses, and mediation analyses. The challenge, however, with combining such data is that the trials may differ along three major dimensions, including population characteristics, trial level characteristics (including which measures and which times are used for assessments), and intervention characteristics. Using 1) an ongoing synthesis study of up to 40 randomized trials focusing on adolescent depression, 2) an overview of 42 meta-analyses on prevention, and 3) a synthesis of 42 antidepressant trials, we discuss the limits of findings that can come from such synthesis studies. We also present an example that uses only three trials of the Familias Unidas preventive intervention; this intervention focuses on preventing drug abuse, HIV risk behaviors, and depression in Hispanic adolescents. We discuss the synthesis of a broad set of related preventive interventions that differ on measures assessed, intervention, and population characteristics, particularly for mediation, moderation based on individual as well as study level characteristics, and comparative effectiveness.