Presentation (CTI)

OBJECTIVES: To perform a significance-based comparison of clinical and diagnostic features in atrial fibrillation (AF) patients with and without subsequent stroke, and to identify real-world data (RWD) derived features that enhance or complement the Framingham Heart Study (FHS) AF-stroke risk model.
METHODS: Using an administrative claims database, we conducted a retrospective cohort analysis of AF patients with a subsequent stroke event (N=1,365) and AF patients without stroke (N=4,399). Structured diagnosis and procedure or utilization features were summarized as binary indicators. Feature prevalence was calculated for each cohort and statistical significance assessed using p-values (p<0.05). To reduce outcome leakage, direct post-stroke sequelae and rehabilitation/testing artifacts were excluded from interpretation.
RESULTS: Among a broad set of candidate features, the most significant and clinically plausible predictors of stroke in AF patients were TIA (p<0.001) and prior stroke history (p<0.001), both core predictors in the FHS study. Additional significant associations included cerebrovascular disease and related vascular diagnoses (p<0.05). Neurologic vulnerability features, including altered brain structure or lesion history, brain injury, encephalopathy, dementia or impaired cognition, obtundation, and seizures, were modestly enriched (all p<0.05). Diabetes and hypertension diagnosis codes were statistically significant but more prevalent in the non-stroke cohort. Several preventative care or utilization features (e.g., vaccination codes, infections, inflammatory diagnoses, and outpatient visit frequency) reached statistical significance; they were more prevalent in the non-stroke cohort, likely reflecting differential healthcare utilization or preventive care rather than increased stroke risk.
CONCLUSIONS: This RWD analysis supports the continued relevance of key FHS AF-stroke predictors, particularly cerebrovascular history, in contemporary populations. While numerous additional features achieved statistical significance, many reflected utilization patterns rather than clinically meaningful stroke risk. These findings highlight the importance of pairing statistical testing with clinical plausibility when extending traditional risk stratification frameworks using RWD.