Background: Stroke care data come from multiple sources, each capturing different aspects of care. The Florida Stroke Registry (FSR), a statewide quality improvement initiative using Get With The Guidelines–Stroke data, partners with hospitals many of which utilize RAPID AI, an advanced platform providing automated analyses of both ischemic and hemorrhagic stroke imaging. Linking FSR clinical data with RAPID AI imaging metrics presents a valuable opportunity to enhance real-world stroke research. This study outlines the linkage methodology, key challenges, and potential to improve stroke outcomes research. Methods: A structured pipeline linked de-identified patient-level data from the Florida Stroke Registry (FSR) with RAPID AI imaging metrics (Case-Scan Linkage). Data were collected from January 2023 to December 2024 across 31 participating hospitals with DUAs with RapidAI as well. Patients >90 years were excluded to minimize re-identification risk. Linkage used two methods: (1) deterministic matching by arrival date, age, gender, and facility ID; and (2) probabilistic matching allowing a one-day difference between RAPID scan date and FSR admission date. Ambiguous matches were excluded. Imaging within 120 minutes of the initial RAPID AI scan were compiled as “initial imaging,” while duplicate or later studies of the same modality were classified as follow-up. Results: The FSR dataset included 29,804 cases, and the RAPID AI dataset contained 239,630 scans from 79,522 unique patients. Using FSR cases as the reference, a 50% linkage rate was achieved (45% deterministic, 5% additional probabilistic), while using RAPID AI unique patients as the reference, the linkage rate was 19%. Linkage yielded 14,827 patients matched with a total of 49,093 scans. The final diagnosis of linked cases included 11,275 ischemic strokes (76.0%), 1,167 TIAs (7.9%), 1,717 ICH (11.6%), and 413 SAH (2.8%). Conclusion: This study demonstrates the successful linkage of comprehensive clinical data from the FSR with advanced RAPID AI imaging metrics across multiple hospitals. The integration of these datasets enables robust real-world stroke research by combining detailed clinical and imaging information, ultimately supporting improved understanding and outcomes in stroke care.
Fakoori et al. (Thu,) studied this question.