Secondary Logo

Journal Logo

Epidemiology & Social Science

Recent Trends in HIV-Related Inpatient Admissions 1996–2000

A 7-State Study

Fleishman, John A., PhD; Hellinger, Fred H., PhD

Author Information
JAIDS Journal of Acquired Immune Deficiency Syndromes: September 1st, 2003 - Volume 34 - Issue 1 - p 102-110
  • Free


The advent of highly active antiretroviral therapy (HAART) in 1996 has had a major impact on the treatment of HIV infection and has dramatically altered patterns of health care utilization. Early studies reported significant drops in inpatient censuses of patients with HIV infection after the diffusion of HAART. 1 Analyses of hospital discharge data for 7 states between 1993 and 1997 showed that the number of HIV-related inpatient admissions rose between 1993 and the first half of 1995 and then declined precipitously in 1996 and 1997, coincident with the diffusion of HAART. 2

Most research to date has examined data from the first few years of the HAART era. It is not clear if the trends seen in 1996 and 1997 have continued more recently. Some evidence suggests that the initial trend of steady reductions in inpatient utilization may be reversing. Analyses of inpatient costs in a large national sample of HIV-infected patients showed a 43% reduction between 1996 and mid-1997, followed by an upturn in costs per month in 1998; this was interpreted by the authors as reflecting increased inpatient service use. 3 Data from a large urban clinic revealed a decline in hospital admission rates between 1995 and 1997, followed by a plateau in 1997 and an increase in 1998. 4

This study examines temporal trends in HIV-related hospital utilization from 1996 through 2000 using comprehensive data from all admissions to community hospitals in 7 states. Employing the same methodology as a prior analysis of discharge data, 2 we examine whether a rise in admissions after 1997 is evident in a large database that includes recent data from multiple hospitals and states.

Prior analyses 2 found that declines in HIV-related inpatient admission rates between 1993 and 1997 were not uniform. Although declines occurred generally, the magnitude was greatest for white men and least for black men. In addition, declines in privately insured hospital admissions were greater than for admissions covered by Medicaid. We use more recent data to examine whether sociodemographic disparities in HIV care are narrowing over time.

In addition to number of admissions, length of stay (LOS) is important in assessing overall change in inpatient resource utilization. Prior analyses showed a consistent yearly decline in LOS between 1993 and 1997. 2 In the current treatment environment, if inpatient admissions are predominantly for patients experiencing treatment failure or severe HIV-related complications, increases in LOS could be observed. To examine this, we also analyze trends in average inpatient LOS between 1996 and 2000.


Hospital discharge data were obtained from the Healthcare Cost and Utilization Project (HCUP), State Inpatient Database (SID). The SID contains hospital discharge abstract data covering inpatient stays from all short-term non-Federal hospitals in participating states; data from long-term or specialized (eg, psychiatric or substance abuse treatment) institutions were not included in the analyses. SID data include primary and all secondary diagnoses for each inpatient stay, the gender and race/ethnicity of the patient, and information on the primary payer for inpatient services.

This study uses data from 7 states: California, Colorado, Kansas, Maryland, New York, New Jersey, and South Carolina. These are the same states used in an earlier analysis 2 and were chosen to provide continuity with the prior study. States were originally selected, in part, because they provided complete time trend data in the HCUP and because data on patients' race/ethnicity were available. 2 Most of these states have relatively high HIV prevalence; Kansas and South Carolina were included as comparison states with relatively low HIV prevalence.

We identified HIV-related hospitalizations by examining all primary and secondary diagnoses listed in the discharge abstract. All hospitalizations with primary or secondary ICD-9-CM diagnosis codes that included 042.0 through 044.9 were selected as HIV-related hospitalizations. After October 1994, the ICD-9-CM system included only 1 code (042) for HIV disease and AIDS. 5 We searched for codes in the range of 042 to 044 in all years in case use of these codes persisted. In a prior study, the 042 code had a positive predictive value of 89% for AIDS and 99% for HIV; based on review of medical records, 97% of persons with an HIV diagnosis on their hospital discharge abstract were infected with HIV. 6 To be consistent with prior analyses, we excluded admissions with only a V08 code from analyses.

We analyzed SID data from 1996 through 2000. This period covers the beginning of widespread diffusion of HAART and includes the most recent data available. Data from 1996 and 1997 have been published previously but are included here to provide a basis for comparison. Overall, 362,230 HIV-related inpatient admissions were identified for the 7 states in 1996 through 2000.

Payer categories were Medicare, Medicaid, private (including private health maintenance organization [HMO]), self-pay, no charge, or other. The relatively small number of admissions (716 [0.2%]) with a payer code of no charge were combined with self-pay. Race/ethnicity categories were white, black, Hispanic, Asian or Pacific Islander, Native American, other, or missing. Because the “other” category may reflect measurement error, these admissions were combined with those missing race/ethnicity data. Admissions for Native Americans (755 [0.2%]) were combined with those for Asian/Pacific Islanders into an “other minority” category. Ten demographic groups were formed by combining the patient's gender with race/ethnicity.

In descriptive analyses, 186 admissions were deleted due to missing data for gender or payer. Because missing data for race were more frequent (2%–4% per year), a separate “missing” category was created for this variable. For each combination of state, year, demographic group, and primary coverage, we calculated the total number of inpatient admissions (regardless of cause) and the number of HIV-related admissions.


We first examined the number of HIV-related admissions by year as a function of state, payer, and demographic group (ie, gender-race/ethnicity combination). The units of analysis consisted of each combination of state (7 categories), year (5 categories), payer (5 categories), and gender-race/ethnicity (10 categories). Initial analyses examined bivariate relationships between year and each of the other variables.

Regression analyses were conducted to examine the effects of each variable controlling for the others. Because the data were counts, standard linear regression was not appropriate. 7 Poisson regression suggested the presence of overdispersion in the data. Consequently, we used negative binomial regression, which incorporates an additional parameter to deal with overdispersion and thereby provides better estimates of standard errors than Poisson regression. For simplicity, and because certain combinations of variables had relatively small numbers of HIV-related admissions, the regression analyses excluded the racial/ethnic categories of “other minority” and “missing” and the insurance categories of “self-insured” and “other.” With these exclusions, regression analyses were based on 291,535 admissions. (Analyses that included omitted categories of demographic group and insurance produced results similar to those reported below.)

Demographic groups, states, and payers varied in their population size, which would affect the number of admissions. To control for these variations, regression analyses included the log of the total number of admissions for all causes as an offset (ie, its coefficient was constrained to equal 1). The regression can then be interpreted as an analysis of HIV admission proportions (ie, annual number of HIV-related admissions divided by annual number of admissions for all causes). Controlling for total inpatient admissions has the benefit of controlling for secular trends in overall inpatient utilization. (Analyses that did not include this offset produced similar results.)

A second set of analyses assessed inpatient LOS as a function of the same set of independent variables. For descriptive analyses, we removed 6027 admissions (1.7%) reported as lasting for 0 days or for more than 365 days as well as those missing information on LOS, resulting in a sample of 356,017 admissions. In regression analyses, to parallel analyses of admissions, we excluded observations in which demographic group was other or missing and in which insurance was self-pay or other; this yielded 286,631 admissions. The unit of analysis was the individual admission. We examined mean LOS for each time period as a function of state, payer, and demographic group. We conducted linear regression analyses using log-transformed LOS to reduce skewness in the distribution.

Because analytic interest focused on differential change over time, regression models included 3 interactions between year and state, demographic group, and payer, respectively. In addition, the remaining 2-way interactions (state with demographic group, state with payer, demographic group with payer) were included in the model to capture differences in demographic and insurance profiles across states.


HIV-Related Inpatient Admissions

Overall, Medicaid was the primary payer for the majority of admissions (53%), with private coverage accounting for 16% and Medicare for nearly 20%. Including cases with missing race/ethnicity data, 22% of inpatient admissions in this 5-year period were for white men, 29% for black men, 12% for Hispanic men, 18% for black women, nearly 5% for white women, 5% for Hispanic women, and 1% for other minorities of both genders.

The number of HIV-related hospital admissions declined over time, but the rate of decline slowed in more recent years (Tables 1–3). Overall, the number of HIV-related admissions in the 7 states was 93,337 in 1996, 73,119 in 1997, 68,051 in 1998, 65,186 in 1999, and 62,351 in 2000. HIV-related admissions dropped 33% from 1996 to 2000. Admissions declined by 22% between 1996 and 1997, but the decline between 1999 and 2000 was only 4%.

Trends in Numbers of HIV-Related Hospital Admissions and Mean Length of Stay by Demographic Group
Trends in Numbers of HIV-Related Hospital Admissions and Mean Length of Stay by Primary Payer
Trends in Numbers of HIV-Related Hospital Admissions and Mean Length of Stay by State

Among demographic groups (see Table 1), the general picture of a declining but decelerating curve was most evident among white, black, and Hispanic persons of both genders. The other demographic categories revealed departures from a strictly monotonic decline, with several showing a slight increase in 1999 followed by a drop in 2000. Reasons for these minor deviations from the general pattern are not clear. For white men, the aggregate annual number of admissions dropped by 45% between 1996 and 2000. For black and Hispanic men, the proportional decline was not as great (28% and 38%, respectively). Among women, blacks showed the smallest decline (17%).

Admissions in which Medicaid was the primary payer declined by 35% between 1996 and 2000, with relatively minor changes between 1998 and 2000 (see Table 2). Admissions covered by private insurance showed the same pattern but with a greater overall decrease (46%). In contrast, the number of Medicare admissions dropped between 1996 and 1997 but then increased in 1998 and 1999, with 2000 showing virtually no change from 1999.

New York and New Jersey showed monotonic decreases in numbers of HIV-related admissions, with the largest change occurring between 1996 and 1997 and subsequent changes being relatively smaller (see Table 3). The trend in California was similar, with the exception of a slight increase in 1999. The overall drop in admissions was evident in the other states, but the pattern was more irregular. The magnitude of the decrease between 1996 and 2000 varied across states from 48% in Kansas and 39% in New York to 16% in Maryland and 13% in South Carolina.

Regression Models

In the negative binomial regression model for HIV-related admissions between 1996 and 2000, all 2-way interactions were statistically significant (P < 0.001). We compared numbers of admissions predicted by this model as well as by a model with only main effects with observed numbers (data not shown); predictions were much closer to observed numbers for the model with interactions than for the main effects model, suggesting that the interactions be retained. (Coefficients from this regression are available from the authors on request.)

For each demographic group, the coefficients of the interaction with year for 1999 and 2000 were of similar magnitude (eg, the demographic-year interaction coefficients for Hispanic men in 1999 and 2000 were 1.06 and 1.05, respectively) and not statistically different. Similarly, for each state and for each category of insurance, the coefficients for 1999 and 1998 did not differ statistically (eg, the insurance-year interaction terms for Medicaid were 1.34 and 1.32 for 1999 and 2000, respectively). The main effects of year were also not significantly different in 1999 and 2000 (0.42 and 0.38, respectively). A revised model was estimated, constraining effects for 1999 and 2000 to be equal. A likelihood-ratio test comparing the constrained and unconstrained models produced χ2 = 14.95, with 14 degrees of freedom (P = 0.38). Altogether, these results show that the pattern of admissions is unchanged in 1999 and 2000.

For each combination of year, state, payer, and demographic group, the number of admissions predicted from the regression is the product of the total number of admissions (the offset) and the predicted proportion of admissions. To summarize the implications of the regression model, Table 4 presents the predicted proportions for each combination of demographic group, payer, and year. (For ease of interpretation, predicted proportions were averaged across states, weighted proportional to the state's relative number of HIV-related admissions within each combination of demographic group, payer, and year.)

Predicted Admission Proportions by Demographic Group, Payer, and Year

The predicted proportions present a “smoothed” picture consistent with the results in Tables 1 and 2. For every combination of payer and year, black men had the highest estimated proportion of HIV-related admissions. Within each gender, blacks had the highest proportions and whites the lowest. Proportions for white, black, and Hispanic women were uniformly lower than corresponding ones for men. Black women had higher proportions than white men. Admission proportions for Medicaid were uniformly higher than for Medicare, and Medicare proportions were generally higher than proportions for private coverage.

For each combination of demographic group and payer in Table 4, the estimated proportions show a pattern of decelerating decline over time. There is a large drop between 1996 and 1997, followed by a smaller one in 1997 through 1998 and essentially no change between 1999 and 2000. (The estimated proportions are not exactly identical in 1999 and 2000 due to weighting by relative admissions across states.) The pattern of decelerating decline observed in Tables 1 through 3 thus appears in multivariate analyses.

For white men with private insurance, the estimated proportion was 0.63 lower in 1997 than in 1996 (0.0056421/0.0089912) and 0.47 lower in 2000 than in 1996. In contrast, for black men with Medicaid, the 1997 proportion was 0.85 of the 1996 value and the 2000 proportion was only 0.76 of the 1996 value. Black women on Medicaid showed an even smaller decline (0.93 between 1997 and 1996 and 0.85 between 2000 and 1996).

The interaction of state and year was not significant for Colorado and New Jersey and was of borderline significance for Kansas (P = 0.0504) and South Carolina (P = 0.07). Interactions with year were significant for California and Maryland. For 1999/2000, the coefficients for California, Maryland, and South Carolina had values above 1.0 (range: 1.12–1.31). Values above 1.0 indicate that compared with New York in 1996, estimated numbers of admissions were higher, reflecting a less steep rate of decline. In Kansas, the interaction coefficient of state and years 1999/2000 was below 1.0 (0.82), indicating a steeper trend in this time period relative to New York in 1996.

Inpatient Length of Stay

Across all states, payers, and years, mean LOS was lowest for white men (8.28 days, median = 5) and next lowest for white women (8.72 days, median = 6). LOS for Hispanic patients was 9.63 days for men and 9.48 days for women (median = 6 for both), whereas LOS for black patients was 9.92 days for men and 9.50 days for women (median = 6 for both). Mean LOS was higher for Medicaid admissions (9.73 days) than for private (9.09 days) or Medicare (8.88 days) admissions; the median LOS was 6 days for all payers. Mean LOS was longest in New York (10.61 days) and New Jersey (10.23 days) and shortest in Colorado (6.23 days).

Mean LOS for HIV-related admissions showed a pattern of steady decline from 1996 though 2000 (see Tables 1–3). Overall, mean LOS was 10.39 days in 1996, and 9.51, 9.12, 8.94, and 8.85 days in subsequent years. Median LOS was 7 days in 1996, 6 days in 1997 through 1999, and 5 days in 2000. On average, mean LOS was 8% lower in 1997 than in 1996, 4% lower in 1998 than in 1997, 2% lower in 1999 than in 1998, and 1% lower in 2000 compared with 1999. Thus, the rate of decrease in LOS was itself slowing over time.

In the regression analysis of logged LOS, the R2 for a model with all 2-way interactions was 0.0356 and the R2 for a model with only main effects was 0.0324, suggesting that the interactions did not add substantially to explained variation. Comparison of predicted with observed LOS for each model (data not shown) suggested that both models had similar predictive accuracy. Therefore, interpretation focused on a model with only main effects. Coefficients appear in Table 5.

Regression Analysis of Logarithm of Length of Stay for HIV-Related Admissions (1996–2000)

Because the dependent variable has been log-transformed, the coefficients can be interpreted as percentage changes. Thus, according to the model, there was a 9% drop in LOS between 1997 and 1996, a 13% decrease between 1998 and 1996, and greater than 14% decreases between 1996 and each of the last 2 years. The difference in the 1998 and 1999 coefficients was statistically significant, but there was no significant difference between 1999 and 2000.

The other regression results are also consistent with the bivariate results in Tables 1 through 3. In general, the effects were small in magnitude, as reflected in the low R2 value.


The sharp decline in HIV-related hospital admissions observed between 1996 and 1997 has not persisted in subsequent years. Although admissions have generally continued to drop over time, the magnitude of the year-to-year decrease has recently diminished. The trend in 1999 could not be statistically distinguished from that in 2000. On the other hand, there was no general trend for admission rates to increase between 1998 and 2000; prior reports suggesting general increases in HIV-related inpatient admissions may have been premature.

Presumably, the diffusion of HAART in 1996 and 1997 led to improvements in clinical status and reduced need for inpatient HIV care, which was reflected in marked declines in inpatient admissions. Unfortunately, our data do not contain specific information on use of antiretroviral medications. The attribution of HAART as the factor accounting for declines in inpatient use is thus based on circumstantial evidence; however, analyses of data from multiple HIV practices in 1999 show that receipt of HAART was related to lower hospital utilization. 8

The smaller magnitude of declines in admission rates in subsequent years may be due to higher rates of complications of HAART and treatment failure. In addition, the prevalence of persons living with AIDS has been increasing steadily between 1996 and 2000 9; the larger pool of potential inpatients may also contribute to a slowing of the decline in inpatient admissions. There is no simple relationship between AIDS prevalence and inpatient use, however; prevalence was rising in 1996 through 1997 while inpatient utilization was dropping dramatically.

Although the general pattern of decelerating decline in inpatient hospitalizations was observed for different demographic groups, payers, and states, the extent and rate of decline varied. Admissions for white men showed the greatest decreases in 1997 through 1998 and the least amount of subsequent deceleration. In contrast, admissions for black men and women showed relatively small declines. These findings are consistent with previous reports of racial differences in HIV-related inpatient admissions, which have consistently shown that white HIV patients are less likely to have an inpatient admission than black HIV patients. 8,10–12 Admissions for black and Hispanic women were also consistently higher than those for white women. Differential access to potent therapy could result in different patterns of admission rates over time. Black patients were less likely than white patients to receive combination antiretroviral therapy by the end of January 1998. 12 These results thus point to persisting disparities in inpatient resource utilization, which remain in the HAART era.

In the present study, HIV-related admission proportions for women were lower than those for men. Other recent data, in contrast, show that women were more likely than men to have an inpatient episode. 8 The different gender effects arise because this study uses the hospitalization as the unit of analysis, whereas the person is the unit of analysis in much previous research. The majority of persons with HIV infection are men; women's lower admission numbers reflect the relative numbers of men and women infected with HIV.

Admissions for patients with private insurance showed the greatest decreases in 1996 through 1997 and the least amount of deceleration in 1998 through 2000 compared with Medicaid and Medicare admissions. For most combinations of demographic group, state, and year, the estimated proportion of HIV-related admissions was lowest for private insurance, higher for Medicaid, and highest for Medicare. In contrast, in a prior study of HIV admissions between 1993 and 1997, the proportion of Medicare admissions was significantly lower compared with privately insured admissions. 2 The number of Medicare admissions increased between 1998 and 1999. Consistent with reports 13 that increasing numbers of persons with HIV infection are becoming eligible for Medicare, Medicare admissions were 23% of the total in 2000 compared with 17% in 1996.

Trends in admissions also varied across states, with California and Maryland differing from New York. In California, admission proportions for blacks with Medicaid actually rose between 1997 and 1998 and between 1998 and 1999 compared with declines in New York. In Maryland, Medicaid and Medicare admission proportions rose between 1997 and 1998 for all demographic groups compared with declines in other states. One prior study that observed increases in admissions in 1998 was based on data from Baltimore 4 and thus may reflect a local trend.

Kansas and South Carolina were included in the analyses as relatively low-prevalence states, which might differ in utilization patterns from higher prevalence states. Trend coefficients for Kansas were lower than for New York (ie, below 1.0), whereas comparable coefficients for South Carolina were higher than for New York (ie, above 1.0 [data not shown]). Thus, despite their both having low HIV prevalence, these states differed in their utilization trends, with Kansas showing steeper declines in admissions.

Inpatient LOS also showed a pattern of decline over the entire study period, with the rate of decline slowing in recent years. This pattern is in contrast to prior results 2 that showed a steady yearly decline in LOS between 1993 and 1997. As in prior results, admissions for white patients and those with private insurance tended to have the shortest LOS, whereas those for black men and those covered by Medicaid were generally longer. It may be the case that individuals in the latter groups enter the hospital with more serious conditions or in a more debilitated state and that those in the former groups are in better condition, perhaps as a result of better access to outpatient care or better compliance with HAART.

States differed in average LOS, with New York and New Jersey consistently having the highest mean LOS. Other studies have also found regional differences in LOS. For example, the 1999 National Hospital Discharge Survey found that the average LOS for all causes in short-term hospitals ranged from 5.7 days in the Northeast to 5.0 days in the South and 4.6 days in the West. 14 Geographic variation in LOS is not a phenomenon restricted to HIV-related hospitalizations.

It is possible that the same individual patient could have been admitted multiple times and would therefore be counted more than once in the analysis. From the standpoint of examining aggregate trends in inpatient resource utilization, however, temporal and demographic differences in utilization do provide important information even if they include multiple admissions for the same person. Treating multiple admissions for the same individual as independent observations will lead to underestimating standard errors in statistical analyses; it will have less impact on means or regression coefficients. Because most coefficients were significant well beyond the 0.001 significance level, underestimation of standard errors had minimal effect on the substantive implications of the analyses. Ascertaining whether the rate of readmissions varies across different groups of patients with HIV infection does have clinical and policy implications and may provide evidence for disparities in care.

HIV-related admissions were identified by the presence of specific ICD-9 codes as primary or secondary diagnoses. A potential limitation concerns accuracy of coding. It is possible that some HIV-related admissions were not assigned ICD-9 codes reflecting HIV infection. Other admissions may have received an ICD-9 code in the range of 042 to 044 even though the purpose of the admission was to treat a condition unrelated to HIV infection (eg, an injury). Such false-negative and false-positive results will influence estimates of overall admissions; however, it is less clear that errors or omissions in ICD-9 diagnosis coding would produce the temporal patterns of associations documented in this study.

Regression analyses of admissions controlled for the total number of admissions for each year, demographic group, state, and insurance category. The estimates thus pertain to the proportion of total admissions that were related to HIV. If estimates of HIV or AIDS prevalence had been used instead, the picture might have changed. For the 7 states in this analysis, midyear estimates of the number of persons living with AIDS were 110,355 in 1998, 120,439 in 1999, and 128,699 in 2000. 15–17 Using these estimates in the denominator, HIV-related admission rates were 0.617, 0.541, and 0.484 in these 3 years. Changes in the denominator have a significant impact on changes in these rates over time. Unfortunately, yearly state-by-state data on AIDS prevalence broken down simultaneously by gender, race/ethnicity, and insurance do not appear to be publicly available.

Variation in total inpatient admissions over time could potentially affect estimates of changes in HIV admission proportions. For the 7 states in this study, however, variation in total (all cause) inpatient admissions was minimal. The total number of inpatient admissions in the 7 states was 8,772,952 in 1996 and 8,787,882 in 1997, rising to 8,903,585 in 1998 and 9,070,494 in 2000. In addition, regression analyses that did not include total number of admissions (data not shown) produced results similar to those reported in Tables 4 and 5. These findings suggest that the observed temporal patterns of HIV-related admissions do not merely reflect a general trend that applies to all admissions, regardless of cause.

This study used data from comprehensive discharge databases in 7 states. The findings have greater generalizability than results based on data from particular hospitals, states, demographic groups, or insurance databases. Ancillary analyses of data from Florida, Illinois, and Pennsylvania show a similar pattern over time. Between 1998 and 2000, HIV-related admissions declined by 7% in Illinois (7280 to 6778) and by 2% in Pennsylvania (5828 to 5683) and were virtually unchanged in Florida (17,517 to 17,549). These results reinforce the conclusion that the main findings are not confined to a few atypical states.

The trends toward fewer HIV-related inpatient hospitalizations and shorter LOS in recent years are welcome, but the possible bottoming out of the trend raises serious questions. In particular, specific factors that have contributed to these trends remain to be identified. The increased use of HAART has probably been a major factor underlying observed trends, but the possible effects of increased rates of treatment failure and complications of HAART require further scrutiny. In addition, more work needs to be done to explain persisting demographic differences in inpatient utilization. Trends in HIV-related inpatient resource utilization require continued monitoring.


1. Torres RA, Barr M. Impact of combination therapy for HIV infection on inpatient census. N Engl J Med. 1997; 336:1531–1532.
2. Fleishman JA, Hellinger FJ. Trends in HIV-related inpatient admissions from 1993 to 1997: a seven-state study. J Acquir Immune Defic Syndr. 2001; 28:73–80.
3. Bozzette SA, Joyce J, McCaffrey DF, et al. Expenditures for the care of HIV-infected patients in the era of highly active antiretroviral therapy. N Engl J Med. 2001; 344:817–823.
4. Gebo KA, Diener-West M, Moore RD. Hospitalization rates in an urban cohort after the introduction of highly active antiretroviral therapy. J Acquir Immune Defic Syndr. 2001; 27:143–152.
5. Fasciano NJ, Cherlow AL, Turner BJ, et al. Profile of Medicare beneficiaries with AIDS: application of AIDS casefinding algorithm. Health Care Financ Rev. 1998; 19:19–38.
6. Rosenblum L, Buehler JW, Morgan MW, et al. HIV infection in hospitalized patients and Medicaid enrollees: the accuracy of medical record coding. Am J Public Health. 1993; 83:1457–1459.
7. Long JS. Regression Models for Categorical and Limited Dependent Variables. Thousand Oaks, CA: Sage Publications; 1997.
8. HIV Research Network. Hospital and outpatient health services utilization among HIV-infected patients in care in 1999. J Acquir Immune Defic Syndr. 2002; 30:21–26.
9. Klevens RM, Neal JJ. Update: AIDS—United States, 2000. MMWR. 2002; 51:592–593.
10. Mor V, Fleishman JA, Dresser M, et al. Variation in health service use among HIV-infected patients. Med Care. 1992; 30:17–29.
11. Fleishman JA, Hsia DC, Hellinger FJ. Correlates of medical service utilization among people with HIV infection. Health Serv Res. 1994; 29:527–548.
12. Shapiro MF, Morton SC, McCaffrey DF, et al. Variations in the care of HIV-infected adults in the United States: results from the HIV Costs and Services Utilization Study. JAMA. 1999; 281:2305–2315.
13. Fasciano NJ, Thornton CV, Turner BJ, et al. Profile of Medicare Beneficiaries with AIDS: Application of an AIDS Case-Finding Algorithm. Executive Summary: Final Report for Contract 500-97-0004. Princeton, NJ: Mathematica Policy Research; 1999.
14. Popovic JR, Hall MJ. 1999 National Hospital Discharge Survey. Advance Data from Vital and Health Statistics, No. 319. Hyattsville, MD: National Center for Health Statistics; 2001.
15. Centers for Disease Control and Prevention. HIV/AIDS Surveill Rep. 1998; 10:5.
16. Centers for Disease Control and Prevention. HIV/AIDS Surveill Rep. 1999; 11:5.
17. Centers for Disease Control and Prevention. HIV/AIDS Surveill Rep. 2000; 12:5.

AIDS; inpatient hospital use; inpatient length of stay; resource use

Copyright © 2003 Wolters Kluwer Health, Inc. All rights reserved.