Assessing risk of bias associated with missing dichotomous outcome data in meta-analyses : application in five cochrane systematic reviews -

Abstract

Missing participant data relates to trial participants for whom outcome data are not available for systematic review authors. There is no consensus on how systematic review authors should assess risk of bias associated with missing data for a given meta-analysis. One proposed approach is to evaluate the impact of different assumptions about missing data on the pooled effect estimate.To assess how different assumptions about the outcome of participants with missing data alters statistically significant pooled effect estimates of patient-important dichotomous outcomes in five Cochrane systematic reviews.We conducted this study using a series of five recently updated Cochrane systematic reviews addressing different clinical questions about anticoagulation in patients with cancer. We considered patients with missing data those described as having withdrawn consent, being lost to follow-up or having outcome not assessable. We focused on outcomes for which the primary meta-analysis, a complete case analysis, revealed statistically significant pooled effect estimates. We applied nine assumptions about the outcome of participants with missing data. Four of these assumptions are commonly used (e.g., best case scenario and worst case scenario). The remaining five assumptions are considered more plausible as they are based on incidences observed among participants followed-up in the trials. We assessed the number of assumptions under which each pooled effect estimate loses significance and changes direction.