Individual participant data meta-analytic methods 1993 - 2012: A critical methodological quality survey

Over the past 40 years, research syntheses exploring the links between medical and behavioral factors almost always have focused on the combination of studies' published summary statistics. Aggregated data meta-analysis (AD-MA) introduces potential ecological or aggregation bias. Therefore, individual-participant data meta-analysis (IPD-MA) has been recognized as the gold standard for combining evidence from existing studies. The methods involved in IPD-MA have improved since the first pooled data was analyzed in 1993 but it is not known to what extent IPD-MA exhibit high quality among all of the steps and variety of statistical complexity that can still be applied. IPD-MAs were retrieved from electronic databases with pre-established inclusion criteria: applied meta-analytic procedures; used of individual raw data; and had medical and/or behavioral content. Two trained coders extracted the methodological content and evaluated the quality of the statistic procedures. The number of IPD-MAs has increased significantly the last two years. Some analysts do not take into account the study cluster and have used conventional pooled averages or regression models, still including aggregation bias. Complex models such as multivariate regressions, mixed- and hierarchical models have been used only very recently. IPD-MA provide more thoughtful insights into behavioral and medical relationships but to date none have applied a sufficient advanced statistical model to capture the richness of IPD-MA and to achieved contemporary quality standards. Our results offer some guidance to conduct IPD-MA and to which new statistical models and techniques applied to IPD-MA are most needed to address medical and behavioral factors simultaneously.