A snowballing technique to ensure comprehensiveness of search for systematic reviews: A case study

Article type
Vedula S1, Mahendraratnam N1, Rutkow L1, Kaufmann C1, Rosman L2, Twose C2, Dickersin K1
1Johns Hopkins Bloomberg School of Public Health, USA
2Welch Medical Library, Johns Hopkins University, USA
Background: Increasingly, trial data are becoming available through internal company documents made accessible through litigation. We performed a systematic review of how such documents had been used in the past and how they were identified. Systematic reviews, such as ours, that address complex questions pose a challenge in terms of searching for potentially eligible studies.

Objective: The objective of this study was to evaluate a 'snowballing’ technique to iteratively revise a search strategy used for a systematic review of studies using data from internal industry documents.

Methods: We identified a set of 19 'index’ articles that were eligible for inclusion in our systematic review. From this set of 'index’ articles, we developed an initial search strategy for PubMed and Embase using a combination of indexing terms and keywords. After de-duplicating abstracts retrieved from the two databases, two reviewers independently screened the titles and abstracts, and subsequently the potentially eligible full text articles, against pre-specified eligibility criteria. We will examine the indexing terms and text words for allstudies categorized as eligible and revise our initial search strategy using additional index terms and text words. We will compare the results of our 'snowball’ approach to a 'reference standard’ to estimate the recall and precision of our approach.

Results: The initial strategy retrieved 7061 results in PubMed and 2244 results in Embase. Independent screening of titles and abstracts and full-text articles is in progress. Findings related to our analyses of recall andprecision will be described in our presentation at the Cochrane Colloquium.

Conclusions: Searches for systematic reviews addressing complex questions using a 'snowballing’ approach could be tested for systematic reviews addressing other types of questions.