Machine learning and natural language processing methods to identify ischemic stroke, acuity and location from radiology reports.

Accurate, automated extraction of clinical stroke information from unstructured text has several important applications. ICD-9/10 codes can misclassify ischemic stroke events and do not distinguish acuity or location. Expeditious, accurate data extraction could provide considerable improvement in identifying stroke in large datasets, triaging critical clinical reports, and quality improvement efforts. In this study, we developed and report a comprehensive framework studying the performance of simple and complex stroke-specific Natural Language Processing (NLP) and Machine Learning (ML) methods to determine presence, location, and acuity of ischemic stroke from radiographic text. We collected 60,564 Computed Tomography and Magnetic Resonance Imaging Radiology reports from 17,864 patients from two large academic medical centers. We used standard techniques to featurize unstructured text and developed neurovascular specific word GloVe embeddings. We trained various binary classification algorithms to identify stroke presence, location, and acuity using 75% of 1,359 expert-labeled reports. We validated our methods internally on the remaining 25% of reports and externally on 500 radiology reports from an entirely separate academic institution. In our internal population, GloVe word embeddings paired with deep learning (Recurrent Neural Networks) had the best discrimination of all methods for our three tasks (AUCs of 0.96, 0.98, 0.93 respectively). Simpler NLP approaches (Bag of Words) performed best with interpretable algorithms (Logistic Regression) for identifying ischemic stroke (AUC of 0.95), MCA location (AUC 0.96), and acuity (AUC of 0.90). Similarly, GloVe and Recurrent Neural Networks (AUC 0.92, 0.89, 0.93) generalized better in our external test set than BOW and Logistic Regression for stroke presence, location and acuity, respectively (AUC 0.89, 0.86, 0.80). Our study demonstrates a comprehensive assessment of NLP techniques for unstructured radiographic text. Our findings are suggestive that NLP/ML methods can be used to discriminate stroke features from large data cohorts for both clinical and research-related investigations.

Enhanced Detection of Edema in Malignant Anterior Circulation Stroke (EDEMA) Score: A Risk Prediction Tool.

BACKGROUND AND PURPOSE: Rapid recognition of those at high risk for malignant edema after stroke would facilitate triage for monitoring and potential surgery. Admission data may be insufficient for accurate triage decisions. We developed a risk prediction score using clinical and radiographic variables within 24 hours of ictus to better predict potentially lethal malignant edema. METHODS: Patients admitted with diagnosis codes of cerebral edema and ischemic stroke, NIHSS score (National Institute of Health Stroke Score) of ≥8 and head computed tomographies within 24 hours of stroke onset were included. Primary outcome of potentially lethal malignant edema was defined as death with midline shift ≥5 mm or decompressive hemicraniectomy. We performed multivariate analyses on data available within 24 hours of ictus. Bootstrapping was used to internally validate the model, and a risk score was constructed from the results. RESULTS: Thirty-three percent of 222 patients developed potentially lethal malignant edema. The final model C statistic was 0.76 (confidence interval, 0.68-0.82) in the derivation cohort and 0.75 (confidence interval, 0.72-0.77) in the bootstrapping validation sample. The EDEMA score (Enhanced Detection of Edema in Malignant Anterior Circulation Stroke) was developed using the following independent predictors: basal cistern effacement (=3); glucose ≥150 (=2); no tPA (tissue-type plasminogen activator) or thrombectomy (=1), midline shift >0 to 3 (=1), 3 to 6 (=2), and 6 to 9 (=4); >9 (=7); and no previous stroke (=1). A score over 7 was associated with 93% positive predictive value. CONCLUSIONS: The EDEMA score identifies patients at high risk for potentially lethal malignant edema. Although it requires external validation, this scale could help expedite triage decisions in this patient population.