Patient-level compared with study-level meta-analyses demonstrate consistency of D-dimer as predictor of venous...

We compared the performance of aggregate data (AD)--based and individual patient data (IPD)-based meta-analyses to synthesize evidence on the ability of D-dimer to distinguish recurrence risk in patients with unprovoked venous thromboembolism (VTE) who stopped anticoagulation.

We compared the results of the published AD-based rate ratio of VTE recurrence for positive vs. negative D-dimer, estimated by a mixed-effect Poisson model, with those of the IPD-based hazard ratio obtained by a Cox regression stratified by trial. We performed three additional analyses to investigate the methodological reasons for differences between the two approaches, comparing the IPD Cox regression with AD generated from IPD Poisson regression (to control for differences in population on study), AD time-to-event meta-analysis, and AD generated from IPD meta-regression.

Published analyses agreed in direction and statistical significance when estimating the prognostic value of D-dimer even if IPD estimates suggested a stronger effect. The additional analyses suggested that differences in study populations might explain this slight difference. Poor reporting in published studies precluded a true comparison of AD- and IPD-based assessments of heterogeneity sources.

AD and IPD meta-analyses yielded similar estimates of D-dimer effect to distinguish risk for recurrent VTE. The IPD approach was justified by the need to investigate sources of heterogeneity.