Impact of middle meningeal artery embolization on neurological complications and discharge outcomes in chronic subdural hematoma: A machine learning-based comparative study.

INTRODUCTION: Chronic subdural hematoma (cSDH) predominantly affects older adults, often those with prior head trauma, anticoagulation therapy, or chronic comorbidities. Traditional management involves surgical evacuation; however, middle meningeal artery (MMA) embolization has emerged as a less invasive alternative with the potential for fewer complications. This study compares outcomes of surgical evacuation, MMA embolization, and combination therapy using machine learning to identify predictors of neurological complications and discharge disposition. METHODS: Patients with cSDH from the 2017-2020 National Inpatient Sample (NIS) were categorized into surgical evacuation, MMA embolization, or combined treatment groups. Patient demographics, clinical variables, and outcomes were analyzed. Tree-based classifiers (Random Forest, Decision Tree, LGBM, CatBoost) were employed to predict two primary outcomes: post-interventional neurological complications and home discharge. Model performance was evaluated using classification accuracy and area under the receiver operating characteristic curve (AUC). Feature importance was assessed via minimal depth analysis, and partial dependence plots were used to visualize key predictors. RESULTS: Among 5,754 patients with cSDH, 4,872 underwent surgical evacuation, 726 received MMA embolization, and 156 received both. Surgical evacuation was more common in males, Medicare beneficiaries, and patients with chronic kidney disease or anticoagulant use (all p < 0.05). MMA embolization was more frequent in privately insured patients and urban hospitals. The combined approach was associated with higher rates of neurological and pulmonary complications (p = 0.001), longer hospital stays, and higher total costs (both p < 0.001). MMA embolization showed the highest rate of discharge to home (40.2%, p = 0.004). Among machine learning models, Random Forest achieved the highest accuracy (98.7%), identifying ischemic stroke history, insurance status, age, and treatment type as key predictors of outcomes. Surgical evacuation had the highest predicted complication probability (0.45), while MMA embolization had the lowest (0.20) and the highest predicted home discharge rate (0.58). CONCLUSION: MMA embolization was associated with discharge disposition and neurological complication outcomes after adjustment for measured baseline characteristics; however, these findings should be interpreted as associative rather than causal because treatment allocation was nonrandom and residual confounding from unmeasured radiographic and clinical variables remains possible. Combined surgical evacuation plus MMAE was associated with longer length of stay, higher hospitalization cost, and higher observed neurological and pulmonary complication rates. Model-derived partial dependence findings were exploratory and should not be interpreted as causal estimates or direct observed event rates. Tree-based machine learning models, particularly the Random Forest Classifier, identified ischemic stroke history, insurance status, age, and treatment type as key predictors of outcomes. Prospective multicenter studies incorporating granular clinical, radiographic, procedural, and longitudinal follow-up data are needed to define the comparative effectiveness of MMA embolization and surgical evacuation strategies in cSDH.