Inpatient charges among HIV/AIDS patients in Rhode Island from 2000–2004

The Rhode Island Department of Health's Center for Health Data and Analysis collects Hospital Discharge Data (HDD) each year on inpatients from all 13 licensed acute care hospitals and the rehabilitation facility in Rhode Island from 2000–2004. The data reported to the Department of Health includes information on demographic data, clinical items, and hospital charges[16]. The data were restricted to the years 2000 to 2004, to discharges with specific ICD 9 codes, and to a patient population aged 18+. Because the file did not include a unique medical identifier, it was not possible to distinguish between patients with a single admission during the study period versus multiple admissions. The final study population included 1927 HIV/AIDS discharges. International Classification of Disease (ICD-9) diagnosis code 042–044 was used to identify HIV/AIDS admissions. We decided to include diagnoses that were not primary (> 50%) based on the need to decrease the large amount of false negatives that would result by excluding secondary and tertiary diagnoses of HIV/AIDS. The Rhode Island HDD is a public use file. We obtained consent to use this data from the Rhode Island Department of Health.

Inpatient charges (in dollars) were measured as the total charges of an admission to the hospital. Inpatient charges do not include professional fees, and non-covered charges. Non-covered charges are charges that patients or insurance providers are not required to pay. Emergency department charges incurred prior to admission are included in total charges. Inpatient charges were adjusted for inflation. Charges for subsequent years were deflated to 2000 dollars. This was done by taking into account the average annual percent growth in National Health Expenditures. This measure estimates the amount spent on all health services, supplies, health-related research, and construction activities in the United States during the calendar year.

The independent variables included, LOS, and point of entry for hospitalization. LOS was measured in days. Point of entry for hospitalization had two categories: Emergency Room, and patients who were referred to the hospital through a physician, hospital clinic referral, or a facility referral. Specific data on income, education, and other components of SES were not available in the HDD data. Therefore, we utilized the point of entry for hospitalization variable as a proxy for socioeconomic status (SES). Previous studies have shown that patients referred through the emergency room are of lower SES when compared to patients with other types of hospital referrals (physician, clinic, or facility) [17, 18]. The demographic variables included in this study were age, race, and gender. Age was a continuous variable that was measured in years. Race category consisted of six categories: white, black, Hispanic, Asian, unknown and other.

Rhode Island HDD data from 2000–2004 were analyzed using the Statistical Analysis System (SAS, version 9.1; SAS Institute, Cary, NC). We performed a descriptive analysis of demographic and clinical characteristics by year, using one-way Analysis of Variance (ANOVA) and two-sample t-tests for continuous outcomes and Pearson's chi-square test of independence or Fisher's exact test for categorical outcomes. Fisher's exact test replaced Pearson's chi-square test when the expected cell count was less than 5 for a particular descriptive analysis. We used histograms and the Kolmogorov-Smirnov test to determine if inpatient charges and LOS followed a normal distribution. Due to the skewness of the data a natural log transformation was taken for LOS and inpatient charges. ANOVAs and t-tests were performed on the log transformed data which was subsequently back-transformed to generate the geometric means and standard errors for LOS and inpatient charges. Geometric means and standard deviations reported are likely to be smaller than normal calculations involving untransformed data. The reason for this is that the natural log transformation reduces the influence of extreme observations and outliers resulting in exponentiated geometric values that inherit the reduced influence of extreme observations. We used 2 multivariable linear regression models to compare the coefficients for different years across the models to understand how they change when LOS is controlled. The 2 models examined factors associated with inflation adjusted inpatient charges, with LOS excluded from the first model and included in the second model. Comparing the 2 models enabled us to examine how LOS mediates the effects of sociodemographic and clinical variables. The inflation adjusted inpatient charges were used as the outcome variable. The other independent variables in these models included year, age, race, gender, and point of entry for hospitalization. To meet the assumption of normality of the linear regression model, we performed a natural log transformation for inpatient charges and LOS. Regression coefficients were then exponentiated to obtain the estimated mean percentage increase in inpatient charges for each independent variable. We used a test of linear trend to examine if certain factors are related to each other linearly. All tests were two-sided with a type I error (α) of 0.05. The Bonferroni adjustment of α/k (k number of comparisons) for post-hoc ANOVA analysis was used to address the multiple comparisons problem.

Inflation adjusted inpatient charges differed by point of entry for hospitalization in 2001, and gender from 2002–2004 (Table 1). In 2001, patients that were admitted through an emergency room ($9,760 ± 3.9) had significantly higher inpatient charges than patients who were admitted by physician, clinic, or facility referral ($7,663 ± 2.6). Males had significantly higher inpatient charges than women in 2002 ($12,544 ± 2.8 versus $9,404 ± 2.4), 2003 ($16,652 ± 2.7 versus $14,561 ± 2.4), and 2004 ($18,966 ± 3.0 versus $14,815 ± 2.7).

The test of trend showed a significantly increasing trend for inpatient charges among HIV/AIDS admissions from 2000–2004 (Table 2). Average charges per HIV/AIDS admission was least in 2000 ($8,771 ± 4.6) and greatest in 2004 ($17,367 ± 2.9). The greatest one-year increase in charges (35.6%) and LOS (12.8%) occurred from 2003–2004. LOS was found to be significantly higher in 2004 compared to 2001 (p = 0.001), after adjusting for multiple comparisons by using a Bonferroni adjustment. Age, race, LOS and inpatient charges differed significantly by year. However, the test of trend showed that LOS did not have a significantly increasing linear trend from 2000–2004.

In both multivariable models inpatient charges significantly increased over time based on the test of trend after adjusting for LOS, calendar year gender age, race and point of entry for hospitalization (Table 3, 4). In the first model that excluded LOS, inflation adjusted log inpatient charges were significantly higher in 2002 (31%), 2003 (49%), and 2004 (101%) compared to 2000. Males had a 17% (95% CI, 6–29) significant mean increase in inpatient charges than females. This model (R² = 0.0646) accounted for only 6.46% of the variance in log inpatient charges. This small R² is primarily due to not having LOS in the model.

In the model that included LOS, compared to 2000, inpatient charges increased significantly in 2001 (11%), 2002 (31%), 2003 (45%) and 2004 (73%) respectively. LOS and other race were also found to be significantly associated with log inpatient charges in this model. For each 1 day increase in LOS, inpatient charges increased on average by 166% (95% CI, 159–173) after adjusting for other factors. Admissions that belonged to the other race category had a 19% (95% CI, 3.3–37.5) mean increase in inpatient charges compared to whites. Inpatient charges were significantly lower among blacks (-11%, 95% CI, -16.8,-6.4) and Asians (-11%, 95% CI, -17.8,-4.5) compared to whites. Males had a 10% (95% CI, 4.6–15.9) increase in inpatient charges compared to females in this model. The results of the multivariable model generated a R² of 0.7473, showing that our model accounted for 74.73% of the variance in log inpatient charges.