D Methodological Details of HCSUS Analyses
This appendix provides details of the social area analysis presented in Chapter 5 as a potential method for identifying predictors of resource needs.
SAMPLE
The HIV Cost and Service Utilization Study (HCSUS) (Bozzette et al., 1998; Shapiro et al., 1999) was an interview study of a national probability sample of noninstitutionalized persons with HIV infection. As part of the HCSUS interviews, patients reported where they received care. Using that information, patients were assigned the zip code of the clinic in which they received most of their care. This information was used to identify the county where each patient received care.
Two sources of information were then used to characterize each county. The first is U.S. census data, which are available for geographic units below the Eligible Metropolitan Area (EMA) level. The second is the Area Resource File (ARF), which is a database containing over 6,000 variables for each county in the United States. ARF contains data on health facilities, health professions, measures of health care utilization and expenditures, health status, economic activity, health training programs, and demographic and environmental characteristics. The ARF data are compiled from more than 50 sources, such as vital statistics data and national surveys.1 Both of
1 |
For more information please refer to the following website: http://www.arfsys.com/overviewAccess.htm. |
these data sources represent information that is collected and reported in a standardized manner, is publicly available, provided at the county level, and is routinely updated. The U.S. Census is only conducted every 10 years. The Current Population Survey is conducted monthly and updated estimates are provided annually, although those data are not available at the county level. The ARF data are updated annually.
HCSUS interviewed a total of 2,864 patients. Using the location information gathered in interviews, it was possible to attach a zip code to 2,360 of these patients and link information from their interviews to information from the February 1996 ARF (Landon et al., 2002). This ARF was chosen because it corresponds to the time when the patients were interviewed for the HCSUS. These 2,360 patients resided in 82 counties with a mean count of 29 patients per county (ranging from 1 to 260). A total of 504 patients were not linked because of a missing or unmatched zip code either in the HCSUS sample or in the ARF file.
DEPENDENT VARIABLES
As examples of indicators of need, the Committee created two dependent variables using data from the HCSUS study. The first variable measured the number of needs reported by the patients. In the HCSUS interview, each patient was asked the following questions:
-
Did you need income assistance such as SSI, SSDI, AFDC, or health care benefits from Medicaid or the Veterans Administration in the last 6 months?
-
Did you need to find a place to live in the last 6 months?
-
Did you need home health care in the last 6 months?
-
Did you need mental health or emotional care or counseling in the last 6 months?
-
Did you need drug or alcohol treatment in the last 6 months?
The Committee created a variable called “number of needs,” which is simply the number of these questions that the respondent answered affirmatively.
The HCSUS interview also asked whether patients had received highly active antiretrovial therapy (HAART). Although this is not a direct indicator of resource needs, differences in use of appropriate therapies might be an indicator of variations in resources and/or difficulties treating patients appropriately. The working definition of HAART use, which is based on the Department of Health and Human Services/Henry J. Kaiser Foundation definition used in the Guidelines for the Use of Antiretroviral Agents in HIV Infected Adults and Adolescents, was: taking
certain combinations of nucleoside reverse transcriptase inhibitors (e.g., zidovudine and lamuvidine) plus certain protease inhibitors (PIs, e.g., Nelfinavir), combinations of PIs (e.g., ritonavir and saquinavir), or the combination of a PI plus non-nucleoside reverse transcriptase inhibitors (e.g., Nevirapine). Each patient was classified as taking HAART, if he/ she reported taking any one of the HAART combinations. According to guidelines published close to the time of the 1996 baseline interview, 99 percent of the sample was eligible for HAART because they had either CD4 < 500, or HIV RNA > 10,000 copies per ml, or they had symptomatic HIV or AIDS (Carpenter et al., 1996).
PREDICTORS
The ARF and census data provide examples of how to characterize the “social area” of the persons interviewed. The ARF file contains thousands of variables, but the Committee selected several that it considered representative of the kinds of variables that are likely to be related to resource needs for HIV care. Specifically, the following variables were selected for the regression models:
-
Total general practitioners in 1996 divided by the total population in 1990,
-
Total number of medical specialists in 1996 divided by the total population in 1990,
-
Percent of the population that was black in 1990,
-
Percent of the population that is foreign born in 1990,
-
Percent of population that lives in urban areas in 1990,
-
Percent of population who live in poverty in 1990,
-
Percent of population who are college graduates in 1990.
In addition to the variables specified above, dummy variables representing region (Northeast, Midwest, South, West) were included in the models to control for general regional differences.
ANALYSES AND RESULTS
To find potential predictors for reported needs and HAART use, the Committee estimated two regression models. The model that predicted reported needs was a linear regression model, estimated using ordinary least squares. The model that predicted HAART use was estimated using a logistic regression. Given that many of the counties had multiple patients, a hierarchical model (Bryk and Raudenbush, 1992) was used, in which the county was specified as the second-level cluster (Landon et al.,
TABLE D-1 Results of Regression Analysis in HCSUS sample
|
Regression Coefficient (Standard Errors) |
|
Predictor Variables |
Reported Needs |
HAART Treatment |
Number of general practice MDs per 100 persons |
−10.58 (4.04)** |
−36.51 (10.20)*** |
Number of medical specialists per 100 persons |
2.45 (0.65)*** |
−2.29 (1.90) |
Percent African American |
0.00 (0.00) |
−0.03 (0.01)** |
Percent foreign born |
−0.65 (0.53) |
−1.41 (0.87) |
Percent urban |
−0.46 (0.36) |
0.58 (0.91) |
Percent of families below poverty level |
0.01 (0.01) |
0.08 (0.03)** |
Percent of persons over 25 years of age with 4 or more years of college |
−3.27 (1.05)** |
3.74 (2.56) |
Midwest |
0.07 (0.14) |
1.29 (0.32)*** |
South |
0.14 (0.09) |
0.59 (0.30) |
West |
0.22 (0.15) |
1.10 (0.37)** |
R-Square |
0.02 |
0.05 |
(*p < 0.05, **p < 0.01, ***p < 0.001) |
2002). Because measures of association, rather than estimates of central tendency, were of interest, the data were not weighted prior to analysis. Results of the analyses are reported in Table D-1. Standard errors are shown in parentheses.
Results
The models indicate that the education in an area, the number of general practice physicians, and the number of specialists in an area are statistically significant predictors of the number or needs reported by individuals. Specifically, persons reported more needs if they lived in areas with fewer college educated persons, fewer general practitioners, and more medical specialists. The relationships with education and general practitioners seem reasonable. The relationship for medical specialists is counterintuitive, but it might be a reflection of an emphasis on more expensive care at the expense of more basic services. The strongest predictors of not receiving HAART therapy were living in a county with a high percent of African Americans, a high percentage of families below the poverty level, and an area with more general practitioners.
Discussion
The HCSUS study was not designed to collect information about patient needs thought to be most related to resource needs and the sample was not designed to support analyses of the relationships between county characteristics and needs. Thus, these results do not necessarily describe relationships that could be used to guide allocation decisions. Rather, these analyses were conducted to illustrate the kinds of modeling that could be done to assess whether publicly available measures of area characteristics could be used to predict intercounty variability in resource needs. Assessing the usefulness for such an approach for RWCA allocations would require measures of need directly relevant to RWCA allocation decisions and data that could be aggregated to the level used for allocations, such as EMAs.
REFERENCES
Bozzette SA, Berry SH, Duan N, Frankel MR, Leibowitz AA, Lefkowitz D, Emmons CA, Senterfitt JW, Berk ML, Morton SC, Shapiro MF. 1998. The care of HIV-infected adults in the United States. HIV Cost and Services Utilization Study Consortium. New England Journal of Medicine 339(26):1897–904.
Bryk AS, Raudenbush SW. 1992. Hierarchical Linear Models. Applications and Data Analysis Methods. New York: Sage Publications.
Carpenter CC, Fischl MA, Hammer SM, Hirsch MS, Jacobsen DM, Katzenstein DA, Montaner JS, Richman DD, Saag MS, Schooley RT, Thompson MA, Vella S, Yeni PG, Volberding PA. 1996. Antiretroviral therapy for HIV infection in 1996. Recommendations of an international panel. International AIDS Society-USA. Journal of the American Medical Association 276(2):146–54.
Landon BE, Wilson IB, Wenger NS, Cohn SE, Fichtenbaum CJ, Bozzette SA, Shapiro MF, Cleary PD. 2002. Specialty training and specialization among physicians who treat HIV/AIDS in the United States. Journal of General Internal Medicine 17(1):12–22.
Shapiro MF, Morton SC, McCaffrey DF, Senterfitt JW, Fleishman JA, Perlman JF, Athey LA, Keesey JW, Goldman DP, Berry SH, Bozzette SA. 1999. Variations in the care of HIV-infected adults in the United States: Results from the HIV Cost and Services Utilization Study. Journal of the American Medical Association 281(24):2305–15.