(155) How much data are lost in longitudinal studies of medical devices using hospital databases? Hospital data linked with claims data for Atrial Fibrillation ablation treatment.

Background: Discontinuity in follow-up due to patients receiving care outside the index hospitals may result in incomplete capture of study outcomes within the hospital database. Linkage between hospital database and claims data could allow us to improve longitudinal follow-up of patients.

Objectives: To assess the sensitivity of atrial fibrillation (AF) recurrence post ablation procedures by linking hospital records database with claims data.

Methods: We identified patients who underwent catheter ablation for AF between 2018 and 2020 from the Premier Hospitals Database and linked them with Komodo claims data using tokenization. AF recurrence, defined by repeat ablation, AF-related hospitalization, or direct cardioversion (DCCV), was examined within one-year post-ablation in both the hospital database and linked data. Sensitivity of AF recurrence was calculated by dividing the number of cases identified in the hospital database to number of cases in the linked database for each component. Subgroup analyses were conducted based on patients' age, sex, insurance type, and characteristics of the index hospital (teaching status and bed size).

Results: Among 4689 patients, there were 6.2% vs 7.5% of patients with repeat ablation, 5.2% vs 6.6% with AF hospitalization, and 9.6% vs 13.4% with DCCV in hospital and linked data respectively. Overall sensitivities were 0.83 for repeat ablation, 0.79 for AF hospitalization, and 0.72 for DCCV. Sensitivity of all three components of AF recurrence increased with age. Male patients exhibited lower sensitivity (0.70) than females (0.77) for DCCV. Sensitivity across different insurance types varied: repeat ablation showed similar sensitivity, while AF hospitalization was lower in patients with commercial insurance, and DCCV was lower in those with Medicaid. Sensitivity of repeat ablation increased with larger hospital bed size, while no clear trend was observed for AF hospitalization and DCCV. Sensitivity of repeat ablation was higher in teaching hospitals (0.88) compared to non-teaching hospitals (0.73).

Conclusions: Using linked hospital and claims data as the “golden standard” for identifying AF recurrence events, hospital data captured 83% of repeat ablations, 79% of AF hospitalizations, and 72% of DCCVs. Recurrent events treated at the same hospital demonstrated higher sensitivity in the hospital data compared to events that could occur in other hospitals or outpatient settings. Sensitivity varied based on patients’ demographics and hospital characteristics.