A shared latent space matrix factorisation method for recommending new trial evidence for systematic review updates

摘要

Clinical trial registries can be used to monitor the production of trialevidence and signal when systematic reviews become out of date. However, thisuse has been limited to date due to the extensive manual review required tosearch for and screen relevant trial registrations. Our aim was to evaluate anew method that could partially automate the identification of trialregistrations that may be relevant for systematic review updates. We identified179 systematic reviews of drug interventions for type 2 diabetes, whichincluded 537 clinical trials that had registrations in ClinicalTrials.gov. Wetested a matrix factorisation approach that uses a shared latent space to learnhow to rank relevant trial registrations for each systematic review, comparingthe performance to document similarity to rank relevant trial registrations.The two approaches were tested on a holdout set of the newest trials from theset of type 2 diabetes systematic reviews and an unseen set of 141 clinicaltrial registrations from 17 updated systematic reviews published in theCochrane Database of Systematic Reviews. The matrix factorisation approachoutperformed the document similarity approach with a median rank of 59 andrecall@100 of 60.9%, compared to a median rank of 138 and recall@100 of 42.8%in the document similarity baseline. In the second set of systematic reviewsand their updates, the highest performing approach used document similarity andgave a median rank of 67 (recall@100 of 62.9%). The proposed method was usefulfor ranking trial registrations to reduce the manual workload associated withfinding relevant trials for systematic review updates. The results suggest thatthe approach could be used as part of a semi-automated pipeline for monitoringpotentially new evidence for inclusion in a review update.