Article type
Year
Abstract
Background: Screening search results to identify eligible studies for inclusion in systematic reviews is time consuming. Machine learning aims to reduce the workload of screening, but data evaluating the performance are limited.
Project outline: We are therefore conducting a retrospective case study by comparing the performance of machine learning technology to the ‘gold standard’ of duplicate manual screening.
Methods: We included data from published Cochrane Heart Reviews for which search results are available to Cochrane Heart.
Results: Preliminary results for six (out of 40) reviews were presented at the Cochrane UK and Ireland Symposium in Birmingham, UK, in March 2016. These showed that at least 60% of the screening workload could have been saved with no loss in recall. Final results for 40 reviews will be presented at the Colloquium.
Conclusions: Machine learning represents a potential strategy to reduce the workload of screening for systematic reviews. Further research evaluating the performance of machine learning systems and in other fields are needed before this method can be widely adopted
Project outline: We are therefore conducting a retrospective case study by comparing the performance of machine learning technology to the ‘gold standard’ of duplicate manual screening.
Methods: We included data from published Cochrane Heart Reviews for which search results are available to Cochrane Heart.
Results: Preliminary results for six (out of 40) reviews were presented at the Cochrane UK and Ireland Symposium in Birmingham, UK, in March 2016. These showed that at least 60% of the screening workload could have been saved with no loss in recall. Final results for 40 reviews will be presented at the Colloquium.
Conclusions: Machine learning represents a potential strategy to reduce the workload of screening for systematic reviews. Further research evaluating the performance of machine learning systems and in other fields are needed before this method can be widely adopted