BACKGROUND While stroke is well recognized as a critical disease, current treatment options are limited or absent depending on stroke subtype. Inpatient stroke encounters carry critical information that can be used to investigate the mechanisms of stroke and patient outcomes, however, this data is stored in forms designed to support administrative functions instead of research. To support improvements in stroke patient care and treatment, a substantive research data platform is needed. OBJECTIVE To our role as a Learning Healthcare System and improve research across ischemic and hemorrhagic stroke subtypes, we seek to a comprehensive research repository of stroke data using the Houston Methodist Electronic Health Record (EHR) system. METHODS Dedicated processes have been developed to import data from patients with primary acute ischemic stroke, intracerebral hemorrhage, transient ischemic attack, and subarachnoid hemorrhage from the EHR under an IRB-approved protocol. Developed Extract, Transform, Load (ETL) processes convert EHR data to enable cross-sectional and longitudinal research efforts. Imaging data from patient encounters are withdrawn to assess cortical characteristics and small vessel disease. Patient information needed to interface with other local and national databases are retained. Patient outreach has been developed to assess functional outcomes at 30-, 90-, 180, and 365-days after discharge. Dashboards have been developed to provide investigators with data summaries to support access. RESULTS As of June 2022, the database contains 18,061 total patients, including 1,810 intracerebral hemorrhage patients, 13,444 acute ischemic stroke patients, 1,215 subarachnoid hemorrhage, patients, and 3,168 transient ischemic attack patients. Imaging data from CT is available on 96.1-98.6% of patients, with MRI available on 52.7-89.9%. Outcomes assessment has been able to contact 56.0% of post-stroke patients, with 71.3% of responders providing consent for assessment. CONCLUSIONS A highly curated, clinically-focused research platform for stroke data will establish a foundation for future research that can fundamentally improve patient care and outcomes after stroke.

BACKGROUND Neuroimaging is the gold standard diagnostic modality for all suspected stroke patients. However, the unstructured nature of imaging reports remains a major challenge to extracting useful information from electronic health records (EHR) systems. Despite the increasing adoption of natural language processing (NLP) for radiology reports, information extraction for many stroke imaging features has not been systematically evaluated. OBJECTIVE In this study, we propose an NLP pipeline, which adopts the state-of-the-art ClinicalBERT model with domain-specific pre-training to extract 13 stroke imaging features from head computed tomography (CT) imaging notes. METHODS We utilized the model to generate structured datasets with information on the presence or absence of common stroke features for 24,924 stroke patients. We compared the survival characteristics of patients with and without features of severe stroke (midline shift, perihematomal edema, or mass effect) using the Kaplan-Meier curve and log-rank test. RESULTS Pre-trained on 82,073 head CT notes with 61 million words and fine-tuned on 200 annotated notes, our HeadCT_BERT model achieved an average Area Under Receiver Operating Characteristic curve (AUROC) of 0.9831, F1 score of 0.8683, and accuracy of 97%. Among patients with acute ischemic stroke, admissions with any severe stroke feature in initial imaging notes were associated with lower probability of survival (P-value < .001). CONCLUSIONS Our proposed NLP pipeline achieved high performance and has the potential to improve medical research and patient safety.

Background Neuroimaging is the gold-standard diagnostic modality for all patients suspected of stroke. However, the unstructured nature of imaging reports remains a major challenge to extracting useful information from electronic health records systems. Despite the increasing adoption of natural language processing (NLP) for radiology reports, information extraction for many stroke imaging features has not been systematically evaluated. Objective In this study, we propose an NLP pipeline, which adopts the state-of-the-art ClinicalBERT model with domain-specific pretraining and task-oriented fine-tuning to extract 13 stroke features from head computed tomography imaging notes. Methods We used the model to generate structured data sets with information on the presence or absence of common stroke features for 24,924 patients with strokes. We compared the survival characteristics of patients with and without features of severe stroke (eg, midline shift, perihematomal edema, or mass effect) using the Kaplan-Meier curve and log-rank tests. Results Pretrained on 82,073 head computed tomography notes with 13.7 million words and fine-tuned on 200 annotated notes, our HeadCT_BERT model achieved an average area under receiver operating characteristic curve of 0.9831, F1-score of 0.8683, and accuracy of 97%. Among patients with acute ischemic stroke, admissions with any severe stroke feature in initial imaging notes were associated with a lower probability of survival (P<.001). Conclusions Our proposed NLP pipeline achieved high performance and has the potential to improve medical research and patient safety.

Background Although stroke is well recognized as a critical disease, treatment options are often limited. Inpatient stroke encounters carry critical information regarding the mechanisms of stroke and patient outcomes; however, these data are typically formatted to support administrative functions instead of research. To support improvements in the care of patients with stroke, a substantive research data platform is needed. Objective To advance a stroke-oriented learning health care system, we sought to establish a comprehensive research repository of stroke data using the Houston Methodist electronic health record (EHR) system. Methods Dedicated processes were developed to import EHR data of patients with primary acute ischemic stroke, intracerebral hemorrhage (ICH), transient ischemic attack, and subarachnoid hemorrhage under a review board–approved protocol. Relevant patients were identified from discharge diagnosis codes and assigned registry patient identification numbers. For identified patients, extract, transform, and load processes imported EHR data of primary cerebrovascular disease admissions and available data from any previous or subsequent admissions. Data were loaded into patient-focused SQL objects to enable cross-sectional and longitudinal analyses. Primary data domains (admission details, comorbidities, laboratory data, medications, imaging data, and discharge characteristics) were loaded into separate relational tables unified by patient and encounter identification numbers. Computed tomography, magnetic resonance, and angiography images were retrieved. Imaging data from patients with ICH were assessed for hemorrhage characteristics and cerebral small vessel disease markers. Patient information needed to interface with other local and national databases was retained. Prospective patient outreach was established, with patients contacted via telephone to assess functional outcomes 30, 90, 180, and 365 days after discharge. Dashboards were constructed to provide investigators with data summaries to support access. Results The Registry of Neurological Endpoint Assessments among Patients with Ischemic and Hemorrhagic Stroke (REINAH) database was constructed as a series of relational category-specific SQL objects. Encounter summaries and dashboards were constructed to draw from these objects, providing visual data summaries for investigators seeking to build studies based on REINAH data. As of June 2022, the database contains 18,061 total patients, including 1809 (10.02%) with ICH, 13,444 (74.43%) with acute ischemic stroke, 1221 (6.76%) with subarachnoid hemorrhage, and 3165 (17.52%) with transient ischemic attack. Depending on the cohort, imaging data from computed tomography are available for 85.83% (1048/1221) to 98.4% (1780/1809) of patients, with magnetic resonance imaging available for 27.85% (340/1221) to 85.54% (11,500/13,444) of patients. Outcome assessment has successfully contacted 56.1% (240/428) of patients after ICH, with 71.3% (171/240) of responders providing consent for assessment. Responders reported a median modified Rankin Scale score of 3 at 90 days after discharge. Conclusions A highly curated and clinically focused research platform for stroke data will establish a foundation for future research that may fundamentally improve poststroke patient care and outcomes.