Manual chart review on the ICH patients in the department of Neurology at the only academic comprehensive stroke center (Alabama center or academic center) located within Alabama was performed. The comprehensive chart review combined the medical record number with hospital patient record to extract additional variables such as demographic and clinical characteristics. Four hundred twenty-five patients were used from the registry from 2016 to 2019. Patients aged 18 or above who were diagnosed with ICH or transferred to hospital because of ICH were included in the database. The Institutional Review Board (IRB) of the academic center approved the data collection for and analysis of data using our institutional stroke registry [5, 14].

The NIS dataset is the largest publicly available all-payer inpatients dataset in the US., sampling 20% of inpatient discharges from all US community hospitals. It currently contains data from more than seven million hospital stays each year in 48 states and the District of Columbia [15]. We used the NIS data between 2016 and 2018 and included 68,525 patients with ICH. The NIS dataset includes comprehensive information such as demographic characteristics, hospital characteristics, and outcome. All baseline characteristics of patients, hospital characteristics, diagnosis codes, and procedure codes were recorded in the dataset. The analysis of NIS data was considered exempt from review by the IRB.

The timeframe difference between the datasets (2016–2019 for the Alabama comprehensive center dataset vs. 2016–2018 for the NIS dataset) is primarily due to the data availability for each cohort. The NIS dataset includes 68,525 cases, while the Alabama dataset includes 425 cases. While the datasets cover slightly different timeframes, we believe this discrepancy does not significantly impact the comparison, as the overall trends in stroke care and outcomes are likely consistent across this short period.

Patients aged 18 years or older, diagnosed with ICH, who were hospitalized between 2016 and 2019 in the academic stroke center data and between 2016 and 2018 in the NIS data, were included in the study. International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) code I61.0-I61.9 were used to select ICH patients from NIS data. Patients whose discharge disposition was transferred out in the NIS dataset were excluded.

Our primary independent variable was the hospital area subgroup variable in the combined NIS and academic comprehensive stroke center datasets. The subgroups in the NIS dataset were stratified by variables originally collected in the NIS dataset: hospital bed size (small/medium/large), location/teaching status of hospital (rural/urban non-teaching/urban teaching), region of hospital (Northeast/Midwest/South/West), and census division of hospital (New England/Middle Atlantic/East North Central/West North Central/South Atlantic/East South Central/West South Central/Mountain/Pacific). We restricted the NIS data to ICH patients in large, urban teaching hospitals in the South, and then stratified by census divisions: South Atlantic, East South Central, and West South Central. After stratification, we created a variable with four levels of census divisions (South Atlantic Large Urban Teaching/East South Central Large Urban Teaching/West South Central Large Urban Teaching/academic comprehensive stroke center). A second variable was created by combining the academic comprehensive stroke center dataset and East South Central Large Urban Teaching Hospital into a single category, resulting in a 3-level census division variable with South Atlantic Large Urban Teaching Hospitals and West South Central Large Urban Teaching Hospitals as the other two levels.

Our primary outcomes were in-hospital mortality and length of stay (LOS).

Sociodemographic variables such as Age (18–44/45–64/65–84/85 +), gender (male/female), race and ethnicity (White/Black or African American/Asian/Other) were included in the study. We further included insurance and procedures (clot evacuation/decompression included craniotomy, craniectomy, cranioplasty, burr hole, clot evacuation, clot aspiration, Hydrocephalus-related included extra ventricular device (EVD) and ventriculoperitoneal shunting (VPS) or repair of vascular malformations included clipping, coiling, venous thrombectomy, arteriovenous malformation (AVM) embolization and AVM resection). Procedures in the NIS dataset were selected and coded using the International Classification of Diseases, Tenth Revision, Procedure Coding System (ICD-10-PCS). Lastly, death was included as a covariate in the analysis of the association between census divisions and LOS in the adjusted linear regression models.

Descriptive statistics were generated for baseline demographic and clinical characteristics across the four groups: academic comprehensive stroke center, South Atlantic large urban teaching hospitals, East South Central large urban teaching hospitals, and West South Central large urban teaching hospitals. Continuous variables were analyzed using one-way analysis of variance (ANOVA), while categorical variables were analyzed using Pearson’s chi-square test or Fisher’s exact test, where appropriate.

To assess the association between hospital area and in-hospital mortality, we performed multivariable logistic regression, adjusting for age, gender, race, insurance, and procedures. Separate models were created for the academic comprehensive stroke center data and each of the NIS subgroups (South Atlantic, East South Central, and West South Central large urban teaching hospitals). A final combined model included data from academic comprehensive stroke center and the NIS South Large Urban Teaching Hospitals, using the four-level hospital area variable (South Atlantic large urban teaching, East South Central large urban teaching, West South Central large urban teaching, and academic comprehensive stroke center). The odds ratios (ORs) and corresponding 95% confidence intervals (CIs) were calculated for the association between each hospital area and in-hospital mortality, using South Atlantic large urban teaching hospitals as the reference group.

To examine the association between hospital area and LOS, we used multivariable linear regression, adjusting for age, gender, race, insurance, procedures, and in-hospital death. Linear regression models were built separately for each hospital area and a combined model was created using the four-level hospital area variable. Patients with a hospital stay longer than 30 days were excluded from the analysis to reduce skewness in LOS distribution. Beta coefficients (β) and corresponding 95% CIs were calculated for each hospital area, with South Atlantic large urban teaching hospitals as the reference group. The linear regression analysis assessed whether the LOS differed significantly between academic comprehensive stroke center, East South Central, and West South Central large urban teaching hospitals.

All statistical analyses were performed using SAS version 9.4. A two-sided p-value of < 0.05 was considered statistically significant.