We used the 2000 Nationwide Inpatient Sample (NIS), a database of hospital inpatient stays, developed by the Healthcare Cost and Utilization Project. As the largest all-payer inpatient care database that is publicly available in the United States, the NIS data set represents 20% of non-federal US hospitals. These data include a stratified random sample of 994 hospitals in 28 states, encompassing approximately 7.5 million inpatient stays.
Outcome and Predictor Variables
The primary outcome for this study was the in-hospital mortality rate. Secondary outcomes included total charges and length of stay. The independent variables studied included demographic characteristics (ie, age, gender, race, median income of zip code of residence, and health insurance status), comorbidity, HIV status, hospital admission source, and hospital characteristics (ie, region, location, and teaching status). All independent variables were examined as categoric variables, except for age, which was used as a continuous variable.
We assessed comorbidity in this sample by using the Deyo-adapted Charlson comorbidity index (DCI), a method that is used to estimate the risk of death from comorbid disease using ICD-9, clinical modification, administrative databases. Patients with three or more comorbid illnesses were categorized as one group due to the small number of patients with comorbidities in this range. Patients were considered to have HIV if any of the 15 hospital discharge diagnoses had an ICD-9 code of 042.
We used weighted descriptive statistics to characterize the patient sample, using proportions or means with SDs where appropriate. The means and SDs of continuous variables were compared using the Student two-tailed t test. Differences for categoric variables were determined by x2 test or the Cochran-Mantel-Haenszel test for trend. A p value of < 0.01 was considered to be statistically significant.