Natural language processing (NLP)-assisted assessment of neurologic severity and functional outcomes in factor Xa inhibitor–related intracranial hemorrhages

Introduction Factor Xa inhibitors (FXai) are associated with a risk of serious bleeding, including intracranial hemorrhage (ICH), which can lead to long-term disability. Functional status, often measured using the modified Rankin Scale (mRS), is a key outcome after ICH but is rarely recorded in structured electronic medical record (EMR) fields. This study used natural language processing (NLP) in the retrospective derivation of mRS scores from unstructured clinical notes and evaluate clinical correlates of functional outcomes in a national cohort of US veterans with FXai-related ICH. Methods This retrospective cohort study used data from the Veterans Affairs (VA) Informatics and Computing Infrastructure (VINCI) from March 2014 to May 2022. Adults (≥18 years) hospitalized for direct or indirect FXai-related ICH and treated with reversal or replacement agents were included. Functional outcomes were assessed using mRS scores derived via NLP of unstructured clinical notes. Glasgow Coma Scale (GCS) scores were extracted using an algorithm to support model adjustment and NLP reliability. Results The cohort included 269 patients (mean age 75.7 years; 96.7% male). An NLP algorithm was used to retrospectively derive mRS scores, and GCS scores were extracted using an algorithm to support NLP accuracy and model adjustment. Favorable functional status (mRS 0–3) was observed in 36.1% at baseline, increasing to 58.3% at 90 days. Higher baseline mRS and Charlson Comorbidity Index (CCI) were independently associated with worse functional outcomes (mRS ≥ 4) at discharge and 90 days ( P ≤ 0.039). Discussion Retrospective derivation of mRS scores from clinical documentation is feasible using NLP. This approach demonstrates functional outcome assessment in large-scale EMR-based studies when direct measurement is unavailable. Because long-term recovery after ICH is often defined by functional status rather than survival alone, incorporating mRS into retrospective analyses may improve the evaluation of interventions for FXai-related ICH.