An updated and more efficient search strategy to identify primary care-relevant clinical prediction rules

Objectives: The aim of the study was to develop an improved search strategy for clinical prediction rules. Study Design and Setting: We first refined a list of 30 primary care -relevant journals and improved the efficiency of the Haynes Narrow Filter/Teljour/Murphy Inclusion Filter with 26 items by removing one term (Modified Haynes 26 filter). We then developed the "Royal College of Surgeons in Ireland (RCSI) filter"and compared it with the modified HNF/TMIF26 for its ability to detect prediction rules in the primary care literature. All abstracts and, if necessary, full text were reviewed independently in parallel by primary care physicians. The key outcomes were the percentage of prediction rules identified out of the total identified by both search strategies (sensitivity) and the number of articles that had to be reviewed to identify them (efficiency). Results: The Modified Haynes 26 filter returned 1,701 abstracts vs. 1,062 for the RCSI filter. The RCSI filter identified 105 of 111 of all prediction rules identified by either filter, compared with 107 of 111 by the Modified Haynes 26 filter (94.6% vs. 96.4%; P = 0.52). In addition, 9.9% of abstracts found using the RCSI filter were prediction rules, compared with only 6.3% using the Modified Haynes 25 filter (P = 0.001). Conclusion: We have developed a novel "RCSI filter"that more efficiently identifies prediction rules in the medical literature. (C) 2020 Elsevier Inc. All rights reserved.