Abstract TMP12: Artificial Intelligence Prediction Of Delayed Cerebral Ischemia After Cerebral Aneurysm Rupture

Abstract

Background: Aneurysmal subarachnoid hemorrhage (aSAH) results in significant mortality and disability, which is worsened by the development of Delayed Cerebral Ischemia (DCI). Tests to identify patients with DCI prospectively are needed. Objective: We created a machine learning (ML) system based on clinical variables to predict DCI in aSAH patients and to determine which variables have the most impact on DCI prediction. Methods: We performed a retrospective cohort study of aSAH patients from January 2006 to September 2014. The ML algorithm was trained on age, sex, HTN, diabetes, hyperlipidemia, CHF, CAD, smoking history, family history of aneurysm, Fisher Grade, Hunt and Hess score, and external ventricular drain (EVD) placement. Prediction outcome of the ML algorithm was DCI+, which was defined as new neurologic deterioration that could not be attributed to aneurysm re-bleeding, hydrocephalus, infection, seizure, hyponatremia or other metabolic abnormality. SHAP was used to explain and visualize the role of each feature’s contribution to the model prediction. Results: 500 aSAH patients were identified and 369 met inclusion criteria: 70 patients developed DCI (DCI+) and 299 did not (DCI-). Random Forest was selected for this project after a 5-fold cross validation. 276 cases (222 DCI- and 54 DCI+) were used for training and 93 cases (77 DCI- and 16 DCI+) were used for testing the algorithm. The Random Forest ML algorithm predicted DCI: Accuracy: 81.7%, Sensitivity: 12.5%, Specificity: 96.1%, PPV: 40%, and NPV: 84.1%. SHAP value demonstrated Age, EVD placement, Fisher Grade, and Hunt and Hess score, and HTN had the highest predictive values for DCI. Lower age, absence of hypertension, higher Hunt and Hess score, higher Fisher Grade, and EVD placement increased risk of DCI. Conclusion: ML models based upon clinical variable predict DCI with high specificity and modest accuracy. The addition of imaging or other biomarkers may improve the sensitivity of the ML algorithm.