Adjustment of publication bias using a cumulative meta-analytic framework.

Abstract

Publication bias has the potential to adversely impact clinical decision making and patient health if alternative decisions would have been made had there been complete publication of evidence. The objective of our analysis was to determine if earlier publication of the complete evidence on rosiglitazone's risk of myocardial infarction (MI) would have changed clinical decision making at an earlier point in time. We tested several methods for adjustment of publication bias to assess the impact of potential time delays to identifying the MI effect. We then performed a cumulative meta-analysis (CMA) for both published studies (published-only data set) and all studies performed (comprehensive data set). We then created an adjusted data set using existing methods of adjustment for publication bias (Harbord regression, Peter's regression, and the nonparametric trim and fill method) applied to the limited data set. Finally, we compared the time to the decision threshold for each data set using CMA. Although published-only and comprehensive data sets did not provide notably different final summary estimates [OR=1.4 (95 percent confidence interval [CI]: .95-2.05) and 1.42 (95 percent CI: 1.03-1.97)], the comprehensive data set reached the decision threshold 36months earlier than the published-only data set. All three adjustment methods tested did not show a differential time to decision threshold versus the published-only data set. Complete access to studies capturing MI risk for rosiglitazone would have led to the evidence reaching a clinically meaningful decision threshold 3years earlier.