A Comparison of Meta-analytic Results Using Literature vs Individual Patient Data

OBJECTIVE: To compare the meta-analytic results from published literature vs those obtained from pooling original patient data. DATA SOURCES: Individual patient data from 15 completed or ongoing trials on paternal white blood cell immunization as treatment for recurrent miscarriage were obtained through the American Society for Reproductive Immunology. STUDY SELECTION: Only randomized controlled trials were selected. Within these eight selected trials, 202 patients were from four published studies, 43 were added by the same investigators after publication, and 140 were from four unpublished trials. DATA EXTRACTION: Individual patient data were collected using a standardized form and double data entry. DATA SYNTHESIS: Using the fixed treatment effect model, we found that the effect of immunization, denoted as the relative live-birth ratio (RR), was greater by pooling summary data from published articles (RR, 1.29; 95% confidence interval CI, 1.03 to 1.60) than by pooling all individual patient data from the same investigators (RR, 1.17; 95% CI, 0.97 to 1.37) or when individual patient data were pooled from four unpublished trials (RR, 1.01; 95% CI, 0.74 to 1.28). A similar diminishing treatment effect for the same comparisons was observed using the random treatment effect model (RR, 1.38; 95% CI, 0.89 to 1.87 using published summary data; RR, 1.18; 95% CI, 0.98 to 1.42 using individual patient data; RR, 1.01; 95% CI, 0.74 to 1.28 using unpublished trials). CONCLUSIONS: Meta-analytic results may differ depending on the use of various data sources: whether the source was from summary statistics in the literature, from individual patient data provided by trialists, or from unpublished trials.