Zanaty M, Park BJ, Seaman SC, Cliffton WE, Woodiwiss T, Piscopo A, Howard MA, Abode-Iyamah K. Predicting Chronic Subdural Hematoma Recurrence and Stroke Outcomes While Withholding Antiplatelet and Anticoagulant Agents.
Front Neurol 2020;
10:1401. [PMID:
32010052 PMCID:
PMC6974672 DOI:
10.3389/fneur.2019.01401]
[Citation(s) in RCA: 11] [Impact Index Per Article: 2.8] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Download PDF] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 11/11/2019] [Accepted: 12/20/2019] [Indexed: 12/27/2022] Open
Abstract
Introduction: The aging of the western population and the increased use of oral anticoagulation (OAC) and antiplatelet drugs (APD) will result in a clinical dilemma on how to balance the recurrence risk of chronic subdural hematoma (cSDH) with the risk of withholding blood thinners.
Objective: To identify features that predicts recurrence, thromboembolism (TEE), hospital stay and mortality. To identify the optimal window for resuming APD or OAC.
Methods: We performed a retrospective multivariate analysis of a prospectively collected database. We then build machine learning models for outcomes prediction.
Results: We identified 596 patients. The rate of recurrence was 22.17%, that of thromboembolism was 0.9% and that of mortality was 14.78%. Smoking, platelet dysfunction, CKD, and alcohol use were independent predictors of higher recurrence, while resolution of the SDH was protective. OAC use had higher odds of developing TEEs. CKD, developing a new neurological deficit or a TEEs were independent predictors of higher mortality. We find the optimal time of resuming OAC to be after 2 days but before 21 days as these patients had the lowest recurrence of bleeding associated with a low risk of stroke. The ML model achieved an accuracy of 93, precision of 0.84 and recall of 0.80 for recurrence prediction. ML models for hospital stay performed poorly (R2 = 0.33). ML model for stroke was overfitted given the low number of events.
Conclusion: ML modeling is feasible. However, large well-designed prospective multicenter studies are needed for accurate ML so that clinicians can balance the risks of recurrence with the risk of TEEs, especially for high-risk anticoagulated patients.
Collapse