Evaluation of database contributions to a review on predictive analytics for disease progression: A SWAR using the R package CiteSource

Background: Significant advancements in the use of artificial intelligence and  machine learning (AI/ML) for the prediction of disease progression has  led to a rapidly growing literature in this space. Systematic and  scoping reviews in this area could benefit from methods research to  improve efficiency of the review process. In the context of a scoping  review on AI/ML applications in predicting the progression of chronic  kidney disease, we are conducting a study within a review (SWAR) to  evaluate database contributions to this topic, which lies at the  intersection of biomedical and computer science research.   
Objectives: We aim to evaluate the usefulness of five databases for reviews in  predictive analytics and disease progression: Scopus, Medline, CINAHL,  ACM Digital Library, and IEEE Xplore. We are interested in overlap  across databases, unique contributions of each database, and what each  database contributes to a set of benchmark studies and the different  screening stages of the review.  
Methods: We will use a new R package called CiteSource to assess database overlap  and contribution to benchmark studies and review stages. The unique  records from each database will be further evaluated on characteristics  such as publication year, journal titles, topic, and keywords.   
Results: We conducted a preliminary analysis of database contribution prior to  screening and found that Scopus finds twice as many studies compared  with Medline. There are early indications that many of these results are  from Medline-indexed journals and are found owing to the liberal  application of Emtree terms, searched in Scopus as part of the Keywords  field. Moreover, the computer science database ACM Digital Library  contributes relatively few, but mostly unique, records to the search. We  will investigate these findings in more depth as well as other aspects  of the multiple databases searched.   
Conclusions: Understanding the contributions of multidisciplinary and  discipline-specific databases to the searches for our scoping review can  inform decisions about source selection for future reviews on  predictive analytics for disease progression. This SWAR provides a case  study using the R package CiteSource, presenting new opportunities for  methods research related to source selection and search strategy  validation.  

Presented at the 2023 Cochrane Colloquium, London, UK