Abstract:
Introduction: This thesis examines methods for updating searches for systematic
reviews of healthcare interventions. Systematic reviews endeavour to find and synthesize
all relevant research as a basis for the practice of evidence-based medicine. They are more
useful if they are complete and up-to-date.
Materials and Methods: The sample was 93 meta-analyses in allopathic medicine.
Newer randomized controlled trials (RCTs) were sought through MEDLINE searches,
and were assessed for relevance by physicians. Two Boolean searches, two similarity
searches and one non-database search approach were tested. The Boolean searches were
based on a simple subject search paired with a filter selecting only RCTs from Abridged
Index Medicus journals or with the balanced Clinical Query. The two similarity searches
were Support Vector Machine (SVM) and a Related Article search of PubMed based on
the three newest and three largest studies from the original review.
Main Results: Clinical Query provided good recall but with large retrievals.
Abridged Index Medicus RCT had smaller retrieval sizes and lower recall, but did detect
many large studies. The Related Article search showed the highest recall. Recall with
SVM was lower, but retrievals were smaller. RCTs that cited the systematic review being
updated were also tested but identified only a small proportion of new evidence.
Relative performance of the test searches was consistent regardless of whether the
intervention was a drug, device or procedure. All searches showed variability across
clinical areas, but Related Articles RCT showed the most consistency. The pairing of
Related Article RCT and Clinical Query gave excellent recall of new relevant material.
Conclusions: Meta-analysts can identify new evidence through a simple structured
Boolean search paired with a related articles protocol. By building on the evidence base
formed in the original review, related article searching may replace time-consuming nondatabase
methods necessary in conducting original reviews.
