Predicting Anticoagulation Need for Otogenic Intracranial Sinus Thrombosis: A Machine Learning Approach.

Objective  The role of anticoagulation (AC) in the management of otogenic cerebral venous sinus thrombosis (OCVST) remains controversial. Our study aims to better define when AC is used in OCVST. Methods  MEDLINE, EMBASE, and The Cochrane Library were searched from inception to February 14, 2019 for English and English-translated articles. References cited in publications meeting search criteria were searched. Titles and abstracts were screened and identified in the literature search, assessing baseline risk of bias on extracted data with the methodological index for nonrandomized studies (MINORS) scale. Random effects meta-regression followed by random forest machine learning analysis across 16 moderator variables between AC and nonanticoagulated (NAC) cohorts was conducted. Results  A total of 92% of treated patients were free of neurologic symptoms at the last follow-up (mean 29.64 months). Four percent of AC and 14% of NAC patients remained symptomatic (mean 18.72 and 47.10 months). 3.5% of AC patients experienced postoperative wound hematomas. AC and NAC recanalization rates were 81% (34/42) and 63% (five-eights), respectively. OCVST was correlated with cholesteatoma and intracranial abscess. Among the analyzed covariates, intracranial abscess was most predictive of AC and cholesteatoma was most predictive of NAC. Comorbid intracranial abscess and cholesteatoma were predictive of AC. Conclusion  The present study is the first to utilize machine learning algorithms in approaching OCVST. Our findings support the therapeutic use of AC in the management of OCVST when complicated by thrombophilia, intracranial abscess, and cholesteatoma. Patients with intracranial abscess and cholesteatoma may benefit from AC and surgery. Patients with cholesteatoma can be managed with NAC and surgery.