The pros and cons of including abstracts in systematic reviews: findings from the Multiple Data Sources Study (MUDS)

Article type
Year
Authors
Fusco N1, Dickersin K1, Scherer RW1, Bertizzolo L2, Saldanha I1, Vedula SS3, Li T1, Mayo-Wilson E1
1Center for Clinical Trials and Evidence Synthesis, Johns Hopkins Bloomberg School of Public Health, USA
2Department of Medicine, University of Milan, Italy
3Laboratory for Computational Sensing and Robotics, Johns Hopkins University, USA
Abstract
Background: Only about 60% of RCTs reported in conference abstracts (abstracts) are published in full (e.g. journal articles) and publication is associated with positive trial results. Cochrane, the Institute of Medicine and others recommend searching for abstracts to include in systematic reviews to minimize reporting biases (i.e. by identifying otherwise unpublished trials and outcomes).

Objective: Our objective was to examine abstract reporting in two case examples: gabapentin for neuropathic pain and quetiapine for bipolar depression.

Methods: We conducted electronic searches of bibliographic databases and trial registers; handsearched conference proceedings and reference lists; used materials from litigation; and accepted ad hoc notification of reports. Two independent reviewers performed each of these tasks with disagreements handled by discussion: screening of citations, reading full text for eligibility, and extraction of data. Results were stored in Systematic Review Data Repository. We compared abstract data with aggregate data from public (e.g. journal articles) and 'hidden' sources (e.g. clinical study reports).

Results: We identified 21 and seven trials about gabapentin and quetiapine, respectively. Sometimes we found it difficult to assign a report to one trial. Not all trials had been registered and no relevant data were present when they had. We identified one trial for each example only through ClinicalTrials.gov. We found one trial reported only in an abstract, and it lacked essential information about the intervention and comparator groups, risk of bias, and results. Most abstracts did not contain meta-analyzable data (16/20 gabapentin; 15/20 quetiapine), frequently containing little information about number of groups, participants, interventions and comparators, and study duration. Abstracts and journal articles sometimes reported different information about the same trial, leading to ambiguity about unique trial identification.

Conclusions: In the two examples we examined, abstracts alone may present insufficient information to contribute to a systematic review or meta-analysis, and may contribute to double-counting of trials.