Objective: To compare the performance of Retraction Watch Database (RWD), PubMed, and Web of Science (WoS) in identifying retracted publications (RP) in medicine.

Methods: This cross-sectional study analyzed RP in 131 high-impact journals spanning nine disciplines: anesthesiology, dermatology, general internal medicine, gynecology/obstetrics, neurology, oncology, pediatrics, psychiatry, and radiology. Using RWD, PubMed, and WoS, we retrieved all publications that were retracted in these journals. The total number of RP was defined as the combined count across the three databases. We calculated the proportion of RP retrieved by each database overall, by journal, and by discipline.

Results: A total of 878 RP were identified. Anesthesiology accounted for the most RP (n  = 382), followed by general internal medicine (n  = 125) and gynecology/obstetrics (n  = 116). RWD retrieved the highest number (815; 92.8%), followed by PubMed (758; 86.3%) and WoS (734; 83.6%). Performance varied across disciplines: RWD captured 75-99%, PubMed 52-97%, and WoS 58-96%. RWD outperformed the others in eight of nine disciplines; the exception was gynecology/obstetrics, where PubMed performed better.

Conclusion: RWD demonstrated superior coverage compared to PubMed and WoS, though performance varied by discipline. Combining databases offers a more comprehensive approach to retraction identification.