The Nationwide Inpatient Sample (NIS) is part of the Healthcare Cost and Utilization Project (HCUP) which is sponsored by the Agency for Healthcare Research and Quality (AHRQ) of the US Department of Health & Human Services and represents the largest all-payer inpatient care database in the USA [17
]. Datasets from year 2000 to 2007 were analyzed that included 7.4–8.0 million hospital admissions per year from 28 to 40 states and 994–1,044 nonfederal hospitals, respectively. This constitutes a 20% representative and stratified sample of US hospitals. The NIS contains only deidentified data.
Variables included demographic parameters, admission type, length of stay, discharge disposition, death, hospital location, and charges. In addition, relevant diagnoses and procedures in the database were identified using the ICD-9-CM system (The International Classification of Diseases, 9th Revision, Clinical Modification): adult patients (≥18 years) with IVM were selected using code 747.81 (). Intracranial hemorrhages were identified using codes 430, 431, and 432.x, seizure using code 345.xx, and headache with 346.xx and 784.0, unruptured cerebral aneurysms with 437.3, and acute ischemic stroke with 433.xx and 434.xx, respectively. Procedures were selected using codes 88.41 (cerebral angiogram), 39.72 and 39.79 (embolization), 01.59 (surgery), 92.3x (radiosurgery), 39.5x (aneurysm clipping), and 96.7x (mechanical ventilation). The NIS collects up to 15 diagnoses and procedures for each admission. Thus, relevant modes of presentation, including intracranial hemorrhage, seizure, and headache were identified by selecting primary diagnoses only in addition to diagnoses found at “ANY” rank.
Analyses were weighted with provided discharge weight data that are used to create national estimates. For the trend analysis data were stratified into four two-year blocks. Univariate analyses included frequency (absolute number and %) and average values (SD; standard deviation) with respective comparative analyses (χ2-test, t-test for independent samples). In multivariate logistic regression models, the risk (odds ratio and 95% confidence interval; OR 95%-CI) of death and unfavorable discharge outcome (death or discharge, other than routine or to short-term hospital) was determined with input of significant and relevant variables from the univariate analysis. In addition, outcome was assessed for patients presenting with hemorrhage versus without. Outcome was further stratified into one group undergoing surgery and the other without surgery. Furthermore, factors were determined that were associated with hospital charges in multivariate linear regression models. A statistical test was considered significant when P < 0.05. All statistics were calculated with SPSS 15.0 (Chicago, IL).