The NIS, developed by the Healthcare Cost and Utilization Project (HCUP), is the largest all-payer healthcare database in the United States [1]. HCUP provides survey weights to calculate national estimates and temporal trends for inpatient admissions. Due to public availability of the database, institutional review board approval was not necessary. Patient information in the NIS is de-identified so that patient consent is not required.

NIS admissions between 2012-2019 were identified by International Classification of Diseases, Ninth and Tenth Revision, Clinical Modification (ICD-9-CM and ICD-10-CM) and Procedure Coding System (PCS). Inclusion criteria included age ≥18 years and a diagnosis of SDH (ICD-9 432.1, 852.2, 852.3; ICD-10 I62.00-I62.03, S065*). Treatment with MMAe (ICD-9 39.72, 39.75, 39.76; ICD-10 03LG, 03LM, 03LN) or surgery (using previously published ICD-9 and 10 codes) was identified [12]. Patients were excluded if they had diagnoses or received treatment for aneurysms, neoplasms, precerebral arteriopathies, or dissections [4]. MMAe admissions were individually reviewed to ensure the suitability of their inclusion.

Patient characteristics included age, sex, race, All Patient Refined Diagnosis Related Groups (APR-DRG) illness severity and mortality risk scores, long-term use of anticoagulant therapy, coagulopathy, chronic kidney disease, and Elixhauser comorbidity index [6].

Outcomes studied included cost of admission, hospital length of stay (LOS), nonroutine discharge disposition (destination other than home), and complications. Neurological complications included intraoperative nervous system complications, ischemic or hemorrhagic sequelae, facial weakness, and vision loss/blindness. Seizures were analyzed but not classified as a complication due to the difficulty in identifying preexisting seizure conditions versus new postoperative seizures based on ICD codes. Complications were defined by presence of at least one of the following: neurological, respiratory, thromboembolic, cardiac, infectious, or renal.

All analyses were completed within complex samples functions, accounting for NIS stratification and weighting in accordance with HCUP guidelines [11]. Descriptive statistics were performed for baseline analyses of demographic and clinical data. Univariable analysis of continuous variables was completed using independent samples t-tests or ANOVA and Wilcoxon rank-sum tests for parametric and nonparametric distributions, respectively. Chi-squared tests were used for categorical variables. Propensity score matching (PSM) was performed to reduce potential sampling bias between cSDH patients who received no intervention versus those who underwent MMAe only; patients who underwent both MMAe and surgery were excluded. Propensity scores were calculated based on the probability of undergoing MMAe using multivariable logistic regression in which patient and hospital characteristics were used as covariates. Separate PSM analyses were performed for MMAe versus conservative management, and MMAe versus surgery. Patients were matched 1:1 using the nearest neighbor method without replacement and a caliper of 0.2 times the standard deviation of the propensity scores. Analyses were performed using Stata version 17.0 (StataCorp LP, USA).

A total of 227,116 SDH admissions met inclusion criteria (Table 1). Of these, 381 (0.17%) patients underwent MMAe (64.3% MMAe only, 35.7% MMAe plus surgery). There was a trend toward lower age for patients undergoing MMAe only compared to no-intervention (67.1 vs. 69.4 years, p=0.057). MMAe patients had higher Elixhauser comorbidity indices (p<0.001), were more likely to have elective admissions (13.9 vs. 7.5%, p<0.001), treatment at large hospitals (75.9 vs. 64.6%, p=0.001), teaching hospitals (96.3 vs. 78.2, p<0.001), and hospitals located in the Northeast (36.7 vs. 18.3%, p<0.001). Race, history of coagulopathy or long-term anticoagulation, and chronic kidney disease did not differ between groups (p>0.05). MMAe patients were less likely to have a history of trauma (58 vs. 76.7%, <0.001).

PSM was performed to account for potential biases in patient treatment groups. MMAe patients had no difference in complications or LOS compared to no-intervention patients (Table 2), but higher cost ($31,170 vs. $10,768, p<0.001) and lower rate of nonroutine discharge (53.8% vs. 64.3%, p=0.024).

PSM was also completed for MMAe versus surgery (Table 3). MMAe patients had shorter median LOS (5 vs. 7 days, p<0.001), and lower rates of neurological complications (2.7% vs. 7.1%, p=0.029) and nonroutine discharge (53.8% vs. 71.7%, p<0.001). There was a trend towards increased cost of MMAe relative to surgery ($31,170 vs. $27,639, p=0.086) that did not meet statistical significance.

Limitations of this study include biases in patient selection, as utilization of MMAe may vary between interventionalists, institutions, and geographic regions. While we utilized PSM to reduce the effects of these biases, a prospective multicenter trial would provide a higher level of evidence. There are numerous limitations to the NIS database, including its cross-sectional design without post-discharge outcome data, lack of granularity regarding SDH size, laterality, or chronicity, and the use of billing codes for both diagnoses and procedures. Furthermore, large databases such as the NIS, are subject to coding biases, which may be of greater concern in relatively new procedures [21]. Thus, there may be underrepresentation of complications that were not captured with billing codes. Previous procedures cannot be identified, so it is possible some MMAe patients had surgical evacuation completed on prior encounters. Moreover, the NIS does not temporally order diagnosis codes with procedures, which can make it challenging to differentiate baseline deficits from new post-procedure deficits.