AIDS home- and community-based waivers: effects on use of services, expenditures, and survival.

1. Introduction

The prognosis, survival, and quality of life for persons living with HIV/AIDS (PLWHAs) improved dramatically in recent years due to the development of highly active antiretroviral drug therapies (Hogg et al. 1998; Palella et al. 1998). Concomitantly, advances in medical treatment enabled the majority of PLWHAs to obtain care on an outpatient basis or in their home. Currently, state Medicaid programs are the primary payers of medical care services for PLWHAs. To illustrate, Medicaid spending on care for PLWHAs amounted to about $3.3 billion in 1997, compared to $1.3 billion by Medicare and around $1.2 billion under the Ryan White Comprehensive AIDS Relief Emergency (CARE) Act (Sambamoorthi et al. 1999). In an effort to control increasing Medicaid expenditures yet simultaneously provide high-quality care, a number of state Medicaid programs implemented home- and community-based waiver initiatives for persons with AIDS (PWAs) during the early 1990s. Currently, 13 states provide waiver services for PWAs; these states are identified in Table 1. Enrollment in the program by PWAs in these state Medicaid programs ranges from a low of 17 enrollees in Iowa to a high of 6700 in Florida. Initially authorized under section 2176 of Omnibus Reconciliation Act (OBRA) 1985, the Medicaid 1915c waiver enables states to expand the array of home- and community-based services that are available to Medicaid beneficiaries with AIDS. Waiver services are regarded as an add-on, that is, an additional bundle of up to 24 services that include case management, nursing care, meals-on-wheels, personal care services, and homemaker services that are not available to beneficiaries under the traditional Medicaid program. The services provided under the waiver are summarized in Table 2 for the 13 states currently participating in the program.

The presumption behind this initiative is that if home- and community-based waiver services are used in lieu of inpatient services, then waiver participants should incur lower expenditures than non-waiver participants, and state Medicaid programs should realize cost savings. This presumption is questionable for at least two reasons. First, empirical evidence from evaluations of home-care waiver programs for elderly Medicare beneficiaries suggests that home- and community-based care tends to complement rather than substitute for inpatient services and, thus, results in higher costs per beneficiary (Lindsey, Jacobson, and Pascal 1990). Second, although AIDS-specific waivers are currently functioning in 16 states, only a few published studies evaluated the impact of waiver enrollment on use of services, monthly patient expenditures, and outcomes (Merzel et al. 1992; Anderson and Mitchell 1997; Crystal, Sambamoorthi, and LoSasso 1998; Mitchell and Anderson 2000). These studies found that monthly expenditures and use of inpatient services are lower for waiver participants than for either a control group of non-waiver participants or the entire population of PWAs. Nonetheless, with the exception of the recent study by Mitchell and Anderson (2000), these studies are based on data that predate the development of highly effective antiretroviral therapies for the treatment of HIV/AIDS. Thus, it is unclear whether home- and community-based waiver initiatives for PWAs are able to generate cost savings in this era of highly effective but expensive drug treatment regimens.

In this study, we analyze Florida Medicaid eligibility and claims data for PWAs spanning the years 1996 through 1997 to evaluate how participation in the Medicaid waiver program for PWAs affects the types of services received by patients, monthly patient expenditures, and health. Importantly, antiretroviral combination therapies were available to Medicaid recipients with AIDS throughout the time period. First, we examine whether persons who choose to participate in the waiver program in Florida are more or less likely to receive antiretroviral combination drug therapies and inpatient care. Second, we also examine whether waiver enrollment has any impact on monthly patient expenditures and survival. Considering that the waiver offers a bundle of home- and community-based services that are not readily available to other Medicaid beneficiaries with AIDS, we expect to find significant differences in the types of services used by waiver participants in comparison to those not enrolled in the waiver. These services may be viable alternatives to costly inpatient care. We do not know, however, whether waiver patients are more or less likely to receive recent combination drug therapies and whether the overall treatment packages differentially affect survival and/or monthly patient expenditures.

The remainder of this article is organized as follows. The second section provides a brief synopsis of the relevant literature. Section three describes the data, sample inclusion/exclusion criteria, and variable construction. Section four outlines the estimation strategy. In section five we report the results. A final section contains concluding remarks.

2. Literature Review

Empirical evaluations of Medicare home- and community-based waiver initiatives for elderly persons suggest that home- and community-based services tend to complement rather than substitute for institutional care and, thus, result in higher costs per beneficiary (Lindsey, Jacobson, and Pascal 1990). However, this conclusion may not be applicable to home- and community-based waivers designed for PWAs for at least three reasons. First, previous research failed to account for the possible nonrandom selection of home- and community-based services in evaluating the impact of the waiver on costs. Second, the waiver initiatives for the elderly were designed to substitute home care for nursing home care, whereas the goal of the AIDS waivers was to substitute home care for more expensive inpatient hospital services. Finally, the elderly and AIDS populations differ with respect to demographic composition, services received, and disease duration. This suggests that findings based on elderly populations are probably not applicable to PWAs (Weissart, Cready, and Pawalek 1988).

The limited research that evaluated the effects of enrollment in AIDS-specific home- and community-based waiver initiatives suggests that these programs yield cost savings. For example, Merzel et al. (1992) examined the New Jersey waiver experience and estimated average monthly treatment costs per Medicaid enrollee of $2400 in 1988; they concluded that the average costs for PWAs enrolled in the waiver were substantially lower than the $5000 monthly estimate suggested by previous studies. In a more recent unpublished study, Crystal, Sambamoorthi, and Lo Sasso (1998) also examined the impact of the New Jersey Medicaid home- and community-based waiver program on the costs of care, use of inpatient and outpatient services, and access to care. Their multivariate analyses based on Medicaid data for PWAs diagnosed between 1988 and 1992 show that New Jersey waiver participants used substantially fewer inpatient services and more outpatient services in comparison to those receiving traditional care, yet there was no difference in overall monthly costs of care between the two groups. Further, waiver participation appears to reduce socioeconomic differences in access to outpatient services. Using more recent data from the New Jersey AIDS waiver, Sambamoorthi et al. (1999) compared the use and costs of home care between waiver and non-waiver enrollees. Their findings suggest that waiver participation appears to reduce racial and risk group differences in the probability of using home care, although injection drug users were less likely to participate in the waiver. Irrespective of waiver enrollment, injection drug users incurred significantly lower monthly home-care expenditures.

The only other evaluation of a Medicaid AIDS-specific waiver program compares participants and nonparticipants in Florida. Anderson and Mitchell (1997) found that, after controlling for program selection, PWAs enrolled in the home- and community-based waiver in Florida during its first two years of operation (1990-1991) incurred monthly Medicaid expenditures that were 22-27% lower than nonparticipants. Using more recent data for the years 1993 through 1997, Mitchell and Anderson (2000) evaluated the effects of waiver participation and recently developed antiretroviral drugs on monthly expenditures for Florida Medicaid recipients with AIDS. They found that, after controlling for gender and race/ethnicity, monthly Medicaid expenditures for waiver participants were significantly lower than for non-waiver enrollees. The major reason for the cost difference is that non-waiver enrollees incurred higher inpatient costs than PWAs enrolled in the waiver. Although waiver participants incurred higher drug expenditures, these drug costs represent only a fraction of the higher inpatient costs incurred by non-waiver enrollees. While their analyses are the first to evaluate the impact of the recently developed AIDS drags on patient costs, they provide only indirect estimates because they examined a time period that includes three years prior to and two years after the availability of combination therapies. To directly measure the impact of the new antiretroviral therapies on patient costs, one should focus solely on the time period during which these drug treatments were available. Furthermore, their more recent analysis is descriptive and, thus, does not control for nonrandom waiver selection and other confounding factors.

Our study attempts to address the limitations of existing research by examining the effects of waiver participation on the use of services, monthly patient expenditures, and survival during a time period when the antiretroviral combination therapies were available to Medicaid recipients.

3. Data and Sample Construction

The data for this research came from records of Medicaid claims for medical services in Florida that were delivered between January 1, 1993, and December 31, 1997. All claims were screened for diagnosis codes or prescribed drugs used to treat HIV/AIDS-related illness. The population of Florida Medicaid recipients with HIV/AIDS was identified using a protocol developed by staff of the Florida Medicaid program in conjunction with clinical advice from physicians who specialize in the treatment of AIDS. We recognize that identifying persons with HIV/AIDS-related illness from claims data is a difficult task. Although Medicaid program staff refined this algorithm extensively during the course of its development, there may be a few cases in our sample that do not actually have HIV/AIDS. The specific algorithm developed by Florida Medicaid program staff involves extensive lists of diagnosis and procedure codes and for this reason is not reported in detail here. This algorithm is available upon request from the authors.

Sample Inclusion/Exclusion Criteria

Records are available for every person in the state who filed at least one Medicaid claim during the period 1993-1997 and is HIV positive or has full-blown AIDS. In this study, we focus on the treatment of adults and thus exclude all claims for persons on Medicaid who are less than age 18. If a Medicaid recipient turned 18 between 1993 and 1997, we included this person in our analysis but excluded all claims filed at ages less than 18. (1) We also exclude all persons who, between January 1993 and December 1997, turned age 65. Our rationale for this exclusion is that we do not want to confound our analysis of Medicaid with the switch to Medicare at older ages. Finally, we focus the analysis on Medicaid recipients with AIDS who are white, black, and Hispanic. Thus, we exclude Medicaid recipients who are classified as other race.

The first recorded claim for a combination drug therapy among the patients who meet these criteria occurred in December 1995. Because we are interested in differential access to these therapies, we further limit the sample for our multivariate analysis to only PWAs who entered the Medicaid program prior to December 1995 and were still alive on January 1, 1996. We also exclude women who were pregnant during the period that combination therapies were available (December 1995-December 1997). Hence, our analyses are based only on claims filed for services rendered to individuals when combination therapies were available throughout the period (December 1995-December 1997). The sample is comprised of two groups: waiver participants and a control group of non-waiver participants who had at least one claim for inpatient services. After making these exclusions, the final sample includes 10,836 PWAs enrolled in the Florida Medicaid program.

Variable Construction

For each person in our database, we have records for each month in which a claim for Medicaid services is filed. Each record contains information about the claim as well as basic demographic information about the patient. Claims information includes the number and types of services provided, pharmaceuticals prescribed, amounts paid by Medicaid for services and drugs, diagnosis codes, and date of death. Demographic information includes age, race, gender, and county of residence. We have no information on diagnosis codes prior to January 1993; we therefore cannot determine the month in which they were first diagnosed with HIV or AIDS if they were on Medicaid prior to January 1993. We also have no information on income or employment of recipients and their families at any point in time. We construct proxy measures from information on county per capita income and county urbanization.

We collapse the claims records for each person so that we have one record per person. This file contains information on waiver enrollment and our three categories of dependent variables--treatments received, expenditures, survival--and our independent variables. Use is measured with two variables. Inpatient care is a dummy variable equal to one if the patient experienced one or more hospitalizations from December 1995 to December 1997. Combination drug therapy is also measured with a dummy variable equal to one if the therapies were ever prescribed and received during the same two-year period. Total expenditures are measured from December 1995 to December 1997; average monthly expenditures are equal to total expenditures over this period divided by the number of months that the patient was alive during this same period. In the expenditure models, we use the log of average monthly expenditures to adjust for skewness that typically exists with Medicaid expenditure data. Only persons with positive claims during the December 1995-December 1997 period are included in the expenditure analyses. Survival is measured over the period of time in which combination drug therapies are available. All PWAs in our sample survived through December 1995 when combination therapies became available to Medicaid patients in Florida. Our survival measure is, therefore, the number of months alive from December 1995 until date of death or December 1997, the last point of observation. (2)

We relied on our physician consultant, Paul Arons, M.D., of the Bureau of HIV/AIDS of the Florida Department of Health, to guide us in constructing the drug treatment regimens for HIV/AIDS. During the study period, there were three types of drugs used to treat HIV infection: nucleoside analogs, nonnucleoside analogs, and protease inhibitors. The five nucleoside analogs available during the study period were: zidovudine (azt), didanosine (ddi), zalcitabine (ddc), stavudine (d4t), and lamivudine (3tc). The two nonnucleoside analogs available were nevirapine and deavirdine. Furthermore, there were four protease inhibitors available to Medicaid recipients during the study period: saquinavir, ritonavir, indinavir, and nelfinavir. Drugs from these three groups are used in combination with each other to reduce HIV viral load and to increase the amount of CD4 lymphocyte cells. Recommended treatment for PLWHAs with antiretrovirals was published in two consensus statements issued by a panel of HIV/AIDS experts (Carpenter et at. 1996, 1997). In Florida, the single drug therapies (nucleoside analogs and nonnucleoside analogs) were available to Medicaid patients in 1993, while the combination therapies were not available until December 1995.

Based on the advice of Dr. Arons, we classified combination drug therapy used in the Florida Medicaid program into two groups: (i) two nucleoside analog drugs, including zidovudine (azt), didanosine (ddi), zalcitabine (ddc), stavudine (d4t), and lamivudine (3tc); (ii) one protease inhibitor and a pair of nucleoside analogs. (Physicians may substitute a nonnucleoside analog for a protease inhibitor if a patient develops a resistance to a protease inhibitor.) Using the pharmaceutical claims on each patient between December 1995 and December 1997, we identified individuals who received either of the two drug therapies. The drug therapy variable is equal to one if any claim during the period of study indicated either (i) or (ii).

Waiver participation is a dummy variable equal to one if the patient was enrolled in the waiver during the patient's last month in the data. The three criteria for enrollment in the waiver as established by the Florida Medicaid program are (i) a physician must diagnose and certify the individual has AIDS as defined by Center for Disease Control criteria; (ii) the individual must be disabled as defined by the Social Security Administration, although not all persons who meet the definition are actually receiving Supplemental Security Income (SSI) payments; and (iii) based on the physician's evaluation, the individual must be determined to be at risk of either hospitalization or institutionalization in a nursing home. According to the Florida Medicaid program, 95% of PWAs who enroll in the waiver are determined to be at risk of either hospitalization or institutionalization in a nursing home. Given these criteria, the only appropriate way to construct a comparable control group from the cohort of non-waiver participants is to select those PWAs who were hospitalized. The control group includes those persons who were not enrolled in the waiver and had at least one claim for inpatient care at some point between January 1993 and December 1997. We exclude from our analysis the group of non-waiver participants who received no inpatient care because these individuals are not at risk of hospitalization and, thus, would not be eligible to enroll in the AIDS waiver. (3)

The exogenous variables in our models include demographic characteristics, community characteristics, and stage of disease. The demographic characteristics are race, gender, age at first claim, and county characteristics for county of last claim. Race/ethnicity is measured with two dummy variables for black or Hispanic; white is the omitted category. Gender is a dummy variable equal to one if the patient is female. Age is measured in years. County characteristics include median per capita income and percentage of the county population residing in urban areas. In the waiver selection and input choice models, we also control for other characteristics of the county that influence the choice of home- and community-based care: pharmacies per Medicaid patient, home health care agencies per population, and hospices per population. These three categories of health care services represent substitutes for services provided under the waiver. We expect to find that demand for home-based waiver care is lower in communities with more pharmacies, home health care agencies, and hospices.

We do not have laboratory data with which to measure the severity of AIDS in each patient. However, the claims record contains detailed diagnosis information that we used to construct three indicators of patient health. We measure the state of health at the beginning of our analysis period-December 1995. This state of health at the start of the period is predetermined and assumed to be uncorrelated with tastes and other unobservable factors that influence the choice of the waiver and the types of services received in the future.

The first health variable measures the severity of AIDS. Using the information reported in the diagnosis code field on each claim, we extract the 32 AIDS-defining diagnoses identified by HIV/ AIDS physicians; these diagnoses were employed in the screening algorithm developed by Medicaid program staff. We construct 32 dummy variables indicating whether the PWA ever had a diagnosis of each condition prior to December 1995. We then estimate a common factor, latent AIDS severity measure, using factor analysis with rotation. The indicators ([D.sub.a]) are assumed linearly related to the common factor (A) and specific factors (v) for person (i) as follows in Equation 3.1:

(3.1) [D.sub.ai] = [A.sub.i][B'.sub.a] + [v.sub.i],

where [B.sub.a] is a set of factor loadings and vi is the vector of specific factors. A, [B.sub.a], and v are all estimated. The correlation matrix for the specific factors ([SIGMA]) is used to rotate the common factor to produce estimates with mean zero, variance unity (Harman 1976). (4) The estimated common factor proxies for severity of disease, with higher values representing greater severity. The common factor Ai is orthogonal to the specific factors [v.sub.i]. The estimate of the first common factor for person i is found by normalizing [B'.sub.a][[summation of].sup.-1]-[B.sub.a] = 1 so that [A.sub.i] [B'.sub.a][[summation of].sup.-1][D.sub.ai]. (5) The factor loadings are reported in Appendix Table A1. All of the factor loadings should be positive, and all but three are positive; the negative ones are small in absolute value and presumably a result of sampling error.

We use factor analysis to define a second measure of health reflecting the presence of other comorbidities. To control for other comorbidities, we construct a series of 15 dummy variables to identify the following broad categories of health conditions: infectious disease, malignant neoplasm, immunity problem, blood disease, nervous system disorder, circulatory problem, respiratory problem, pneumonia (not AIDS related), digestive system problem, genitourinary problem, skin problem, musculoskeletal problem, drug dependency, other symptoms as yet undiagnosed, and injuries. We have many reported diagnoses from the first claim through December 1995. We collapse all of these reported diagnoses over the 1993-1995 period into dummy variables. Do is the set of indicators of other health problems. We estimate the common factor as in Equation 3.1 with mean zero and variance unity and use this as our measure of other health. The rotated factor loadings are all positive and are reported in Appendix Table A2.

Third, we construct a separate dummy variable to identify women who were pregnant at some point during the three-year time period before December 1995. This variable is equal to one if a female PWA was ever pregnant during 1993-1995 and is equal to zero if she was not pregnant or if the PWA was male.

4. Empirical Models and Results

Descriptive Evidence

Table 3 reports descriptive statistics for all PWAs and then stratifies the sample by waiver participation. About half of the 10,836 persons in the sample are enrolled in the waiver.

As anticipated, use of services varies significantly between waiver and non-waiver enrollees. Waiver participants are much more likely to use combination drag therapy (59%) compared to only 25% of non-waiver enrollees. The reverse holds for inpatient care; close to 35% of waiver participants use inpatient care compared to almost 66% of non-waiver participants. These differences in the use of services have implications for expenditures. On average, expenditures during the two-year period are $1290 a month among all patients. After controlling for waiver enrollment, it appears that average monthly expenditures of waiver participants are $1622 compared to $961 for non-waiver participants. This significant difference in monthly expenditures is linked to differences in the services provided to patients in the two groups. Monthly expenditures on drugs are about $600 higher among waiver participants, while their expenditures on inpatient care are only $150 lower. These differences explain most of the observed differential in average expenditures between the two groups of patients. (6)

Between 1996 and 1997, about 17% of these patients died, and the average survival time was 22 months. We find that waiver participants were more likely to die before December 1997 than non-waiver participants, and non-waiver participants lived, on average, two months longer than waiver participants; 22% of waiver participants died during the interval, and 12.5% of non-waiver participants died. However, based on our indices of health, we find that waiver participants were in worse health and were much less likely to have been pregnant. This suggests that sicker patients are selecting the waiver program, and this selection may be driving the mean differences in survival that we observe.

Regarding race/ethnicity, about 42% were black, 8% were Hispanic, and 50% were white. Women accounted for 47% of PWAs. The average age of patients was 38, and over 80% reported living in urban areas of their counties. Waiver and non-waiver participants differed in these demographic characteristics. Waiver participants were more likely to be white and male than non-waiver participants. Only 26% of waiver participants were female compared to 67% of non-waiver participants. The majority of waiver participants were white (58%) compared to 43% of non-waiver enrollees. Waiver participants were about three years older on average than non-waiver participants and were more likely to live in urban areas.

The descriptive evidence indicates that the waiver is more likely to be chosen by white men and sicker patients than is traditional care. This suggests that some groups of PWAs--in particular, women and blacks--may have less information about the waiver option and its relative benefits. We also find evidence that receipt of combination drug therapies, use of inpatient care, expenditures, and survival vary by waiver status. Whether participation in the waiver actually does increase expenditures and lower survival cannot be determined from these descriptive results, however, because we have not, as yet, controlled for the differential effect of health and individual patient characteristics on these outcomes. The results of our multivariate analyses that control for the nonrandom selection of the waiver option on use of services, expenditures, and survival are reported below.

Probit Estimates of Waiver Selection, Combination Drug Therapy, and Inpatient Care

In this section we describe the empirical models of waiver participation, treatment, expenditures, and survival, and we present the results from estimating these models. The first model examines the selection of the waiver and the use of inpatient care and combination drug therapy. We assume that participation in the waiver by PWAs is not random, so that the residuals in a waiver choice model might be correlated with the residuals in the treatments received, expenditure, and survival models. The first step in our analysis is, therefore, to estimate a model of waiver participation. The propensity to participate in the waiver program ([P.sup.*]) and, therefore, to receive waiver services and treatments is affected by the characteristics of the patient and the availability of comparable services in the community (X). The characteristics of the patient include race, gender, age, and health. County income and county urbanization proxy for household resources. The variables that identify selection of the waiver separately from survival or expenditures include the availability of pharmacies and home health care agencies in the county. The selection equation is estimated with probit and is given in Equation 4.1 below,

(4.1) [P.sup.*.sub.i] = [[beta].sub.p][X.sub.i] + [u.sub.pi],

where P = 1 if [P.sup.*] [greater than or equal to] 0 and P = 0 if [P.sup.*] < 0, [[beta].sub.p], is a vector of coefficients, i = 1, ..., n persons in the sample.

In the service model, we classify health care services (H) into two broad groups: inpatient care ([H.sub.1]) and the receipt of combination drug therapies ([H.sub.2]); [H.sub.1] and [H.sub.2] are dummy variables equal to one if the service is received and equal to zero otherwise. Services are affected by patient and community characteristics (X) and by participation in the waiver (P). We estimate selection of each type of service jointly with selection of the waiver program (Eqn. 4.1) using bivariate probit (Maddala 1983). The propensity to receive service [H.sub.j] ([H.sup.*.sub.]i) is modeled in Equation 4.2 as

(4.2) [H.sup.*.sub.ji] = [[beta].sub.j][X.sub.i] + [[alpha].sub.j][P.sub.i] + [u.sub.ji],

where [H.sub.ji] = 1 if [H.sup.*.sub.ji] (j = 1,2) [greater than or equal to] 0 for patient i and [H.sub.ji]=0 if [H.sub.ji] < 0, [[beta].sub.j] and [[alpha].sub.j] are vectors of coefficients, and [u.sub.pi] and [u.sub.ji] follow a bivariate normal distribution. (7) We test whether [u.sub.pi] and [u.sub.ji] are correlated.

Table 4 presents estimates of the single equation probit models of waiver participation, the receipt of inpatient care, and the receipt of combination therapies. The table also contains the bivariate probit estimates predicting the propensities to receive inpatient care or combination therapies. With the exception of age, local urban population, and Hispanic ethnicity, all the variables are significant determinants of waiver participation. As suggested by the descriptive statistics, whites and Hispanics are more likely to select the waiver than blacks, and men are more likely to select the waiver than women. White men with AIDS are more likely to be bisexual or homosexual and, therefore, better connected to AIDS information networks in the local community. In addition, this group of men, in comparison to women and blacks, is better educated and had well-paying jobs prior to contracting AIDS. Most of these men spent down their assets in order to qualify for Medicaid. Women on Medicaid are more likely than men to be black and to have children; the majority of women qualify for Medicaid under the Aid to Families with Dependent Children (AFDC) criteria. On average, they are poorer and less educated than men on Medicaid.

We also find that the higher the severity of the AIDS condition, as measured by our AIDS health index, the more likely the PWA is to join the waiver. Persons with a higher score on the other comorbidities index and women who have been pregnant are less likely to enroll. The effect of other comorbidities on waiver selection is much weaker than the effect of AIDS severity.

Community characteristics are also important to the selection of the waiver. The waiver is more likely to be chosen by PWAs residing in wealthier counties in Florida. We also find that PWAs who reside in counties with greater availability and access to home health agencies and pharmacies are less likely to join the waiver; concentrations of home health agencies and pharmacies in the county are indicative of the availability of other health services in the county, and these services tend to substitute for waiver care.

From the waiver selection model, we obtain predicted values of the probability of choosing the waiver treatment. These predicted values instrument for waiver selection in the structural models below. As in Bound, Jaeger, and Baker (1995), we regress waiver selection on the instruments and examine the strength of the correlation between actual waiver selection and predicted waiver selection. The R square from this regression is 0.25 and the F statistic is 302, which indicates a strong and significant correlation between waiver selection and its instruments (Bound, Jaeger, and Baker 1995).

Using these estimates of waiver participation, we next evaluate the effect of the waiver on the type of care received by patients. In particular, we estimate the impact of waiver participation on the receipt of combination drug therapies and the use of inpatient care. From Table 4, we find that there are significant effects of waiver participation on the type of care received. Waiver participants are significantly more likely to receive combination drug therapy but are significantly less likely to receive inpatient care relative to non-waiver participants. This suggests that waiver services substitute for inpatient care, yet they appear to be complementary to the receipt of antiretroviral drug therapy. Waiver services are designed to be used in lieu of inpatient services, and our results indicate that, in practice in Florida, waiver services do substitute for inpatient services. Waiver services promote the use of antiretroviral therapies because the case manager informs the patient that these drug protocols are the most effective treatments available. Our health service results are significant and consistent with the descriptive statistics in Table 3.

We find that both the receipt of drug therapy and inpatient care vary with demographic characteristics and patient health status. In the reduced form and structural models, women are less likely to receive combination drug therapies but are more likely to be hospitalized than men, after controlling for health. Blacks are more likely to receive combination drug therapies than whites, although racial differences in the receipt of inpatient care received by PWAs, once we control for waiver participation, are negligible. Health is also an important determinant of the types of services that each patient receives. Patients with more severe cases of AIDS are more likely to obtain the combination drug therapies, although severity of AIDS has no impact on the use of inpatient care. Patients who score higher on the other comorbidities index and women who have ever been pregnant are less likely to receive either drug therapy or inpatient care. County characteristics also have some impact on the care received. Patients in counties with more hospice care are less likely to receive combination therapy or inpatient care, while patients in counties with more home health agencies are more likely to receive both drug therapy and inpatient care. Pharmacy availability is associated with a lower probability of inpatient care, yet it has no impact on the odds of receiving combination therapy.

Waiver Enrollment and Average Monthly Expenditures

To determine whether the waiver is a less expensive treatment option than standard care, we next examine the costs of this treatment per month using Medicaid expenditures as our measure of cost. The log of average monthly expenditures (E) throughout the two-year period is a function of exogenous variables (Z) and participation in the waiver (P). The relationship between E and Z and P is given in the treatment effects model below; the model for E is given in Equation 4.3 and is estimated jointly with Equation 4.1,

(4.3) [E.sub.i] = [[beta].sub.ei][Z.sub.i] + [[alpha].sub.e][P.sub.i] + [u.sub.ei],

where [[beta].sub.e] and [[alpha].sub.e] are coefficients and [u.sub.pi] and [u.sub.ei] are assumed to follow a bivariate normal distribution. We test whether [u.sub.pi] and [u.sub.ei] are correlated. (8) We identify selection separately from E by including three indicators of the demand for and supply of local health services: hospice care in the county, home health agencies per population, and pharmacies per Medicaid recipient in the county. We test for the validity of these overidentifying restrictions using the Sargan statistic (Hayashi 2000). We estimate a chi squared of 3.052, which is not statistically significant. We cannot reject the overidentifying restrictions that the instruments are uncorrelated with the error term; from this test result, the instruments are valid in the expenditure model.

The expenditure results are presented in Table 5. We find that waiver enrollment has a positive effect on monthly costs if we assume that participation in the waiver program is exogenous. However, when we control for nonrandom participation in the waiver program by PWAs, we find that the waiver enrollment results in lower average monthly Medicaid expenditures. After adjusting the coefficients to account for the log dependent variable, the ordinary least squares (OLS) results suggest that waiver enrollees incur monthly expenditures that are 68.5% higher than non-waiver participants. In contrast, the instrumental variable model (IV) results imply that monthly patient expenditures are 53% lower for waiver participants relative to non-waiver participants. Average monthly expenditure is predicted to be $552.45 lower on average if a nonparticipant were enrolled in the waiver. If all eligible non-waiver enrollees were to enroll in the waiver program, the aggregate savings per month to the Florida Medicaid program are predicted to be $3 million. (9) We also find, as expected, that expenditures are lower for women on average. Racial differences in monthly Medicaid expenditures are negligible. (10) Sicker patients, using either index of health, are more expensive patients, as expected, but women who were pregnant are much less costly than either men or other women in our sample. Expenditures are also higher in more urban areas of the state and in counties with higher income. These community results are consistent with expectations.

Waiver Enrollment and Survival

To determine the effect of the waiver on survival, we estimate three versions of a treatment effects model of survival. First, we estimate a standard regression in which waiver selection is assumed exogenous. Second, we estimate an instrumental variables model as in Equation 4.3 but replace log expenditures with log survival. We test for the validity of the overidentifying restrictions imposed on the model. The Sargan statistic is equal to 0.027, which is statistically insignificant. We cannot reject the overidentifying restrictions; the instruments are valid in the survival model. Third, we estimate the survival model using a Weibull hazards model to adjust for the probability that the patient's survival spell is still open at the end of the data. The baseline hazard (ho) is assumed to be equal to [pt.sup.p-1]; if p is greater than one, then the hazard increases with duration or the probability of death increases over time. The waiver variable in the hazard model is the predicted probability of waiver participation derived from the probit model Equation 4.1.

Table 6 presents the results from the estimation of regression and hazard models of survival. The coefficients in the regression models are marginal impacts on log duration (survival from December 1995 to December 1997). For the hazard model, we report the odds ratios derived from the hazard model coefficients. An odds ratio greater than one indicates that the independent variable increases the probability of death, whereas an odds ratio less than one means the independent variable is associated with a lower probability of death.

We find that waiver participation, irrespective of whether it is exogenous, endogenous, or a predicted probability, has no effect on survival in either the regression or the hazard model. Thus, although waiver participants use a package of services different from their nonparticipant counterparts, these differences in treatments received have no impact on survival. The descriptive statistics suggest that PWAs enrolled in the waiver survive about one month less than nonparticipants. However, these mean differences do not control for differences in the other characteristics of patients that may explain survival differences. Once we control for differences in other patient characteristics, we find no effect of waiver participation on survival. The survival results in combination with the finding that waiver enrollees incur lower monthly expenditures than nonenrollees suggest that services available under the AIDS-specific waiver yield a significantly more cost-effective form of treatment than services available to nonenrollees under the traditional Medicaid program.

Contrary to waiver enrollment, some patient characteristics are linked to differences in survival. Blacks have lower survival and higher odds of dying than other patients. Conversely, women are less likely to die than men. We also find that sicker patients, as measured by both the AIDS index and the comorbidity index, are significantly more likely to die than less sick patients. We find, however, a protective effect of pregnancy among women. Women who were ever pregnant prior to December 1995 were less likely to die during the 1996-1997 period. Finally, the odds of dying appear to be significantly higher in urban areas. (11)

5. Conclusions and Policy Implications

State Medicaid agencies in 13 states implemented AIDS-specific home- and community-based waiver programs during the late 1980s and early 1990s to provide home- and community-based services to PWAs as an alternative to more expensive hospital-based care. The waiver program predated the availability of antiretroviral therapies, and antiretroviral therapies are available to all PWAs on Medicaid irrespective of whether they enrolled in the waiver. The waiver program offers a different form of treatment to PWAs; waiver enrollees receive home-based care, less inpatient care, and case management. The waiver, through the provision of case management services, promotes the use of antiretroviral therapies and reduces the asymmetrical information problems that partly explain why some PWAs do not receive specific services and drug therapies. If home- and community-based care is used in lieu of more expensive inpatient care, then Medicaid expenditures per AIDS patient should fall, and this, in turn, should generate program savings.

Although some published research demonstrated that participants in AIDS waivers incur lower monthly expenditures than non-waiver participants (Merzel et al. 1992; Anderson and Mitchell 1997), these studies are based on data from a time period that predates the availability of recently developed antiretroviral combination therapies. In this study, we analyzed Florida Medicaid claims data for PWAs during the period December 1995 through December 1997 to evaluate the effects of waiver participation on use of inpatient care, receipt of combination drug therapies, monthly expenditures, and survival.

Our findings indicate patients on the waiver program in Florida receive a different form of treatment in comparison to other PWAs on Medicaid; specifically, waiver treatment uses significantly less institutional care and relies more heavily on antiretroviral drug therapy. However, participation in the waiver program is a choice and is not randomly selected by eligible patients. White men and persons who have greater AIDS-related severity are more likely to enroll in the waiver program than other patients. Recognizing that waiver participation is correlated with residuals in the service use model has significant implications for the analyses evaluating the impact of waiver participation on monthly expenditures. After controlling for this residual correlation through instrumental variables estimation, waiver patients are found to incur significantly lower expenditures per month than non-waiver participant patients. The IV estimates suggest that average monthly expenditures are 50% lower for waiver enrollees relative to non-waiver participants. The differences can be attributed to lower hospitalization expenses and are detected only when we control for the nonrandom selection of the program and other characteristics of patients. On the other hand, waiver participation does not seem to affect the survival of PWAs. Once we control for other patient characteristics, survival is equivalent in both groups of patients.

These results are based on PWAs who qualify for Medicaid and thus meet stringent income and wealth limits. While these findings can be generalized to Medicaid recipients with AIDS in other states, they may or may not be applicable to PWAs who are covered by private insurance. It is also possible that family members bear some of the financial and non financial (caregiver) burdens required to provide care of the family member with AIDS. For example, the opportunity cost for a family member to care for a PWA at home under the waiver initiative may be higher than under traditional care. A second limitation of our findings is that we do not have the data to quantify these family costs. However, if we observe a Medicaid recipient opting for the waiver program, we assume that this choice is optimal for the family and, therefore, the family is better off.

Given these qualifications, the waiver program seems to be an efficient treatment choice. Patients who opt for this form of care are less expensive than other patients. Further, it is likely that they are receiving a more comfortable, preferable type of care based in their homes and not in institutions. The case management component of the AIDS-waiver program provides each PWA with a plan of care that ensures regularity of care and promotes compliance with recommended medications. These results suggest that waiver initiatives for the treatment of AIDS are cost effective and should be considered by Medicaid programs in other states. Moreover, home- and community-based care initiatives may also be viable policy options for providing cost-effective services to persons with other chronic health problems.

Appendix

Table A1. Estimated Factor Loadings for AIDS Severity Index

                                                            Factor
AIDS Indicators (a)                                        Loadings

Salmonella                                                  0.05929
Tuberculosis-other                                          0.28314
Tuberculosis-primary                                        0.11131
Herpes                                                      0.20045
Cytomegalic inclusion disease                               0.20778
Candidiasis                                                 0.34403
Coccidiosis                                                 0.09888
Coccidioidomycosis                                          0.02951
Histoplasmosis                                              0.09559
Toxoplasmosis                                               0.16384
Infectious/parasitic disease                                0.78648
Other malignant neoplasm, skin                              0.10191
Kaposi sarcoma                                              0.16584
Malignant neoplasm, cervix                                 -0.02687
Lymphoma                                                    0.00742
Meningitis                                                  0.05684
Pneumonia                                                   0.09301
Encephalitis                                                0.17386
Malnutrition                                                0.12997
Other mycobacterial disease                                 0.23874
Progressive multifocal leukoencephalopathy                  0.05684
Reticulosarcoma                                            -0.01551
Burkitt tumor                                              -0.01222
Deficient cell immunity                                     0.1438
Immune mechanistic disease                                  0.17951
Autoimmune disease                                          0.01421
Immunity deficiency                                         0.31518
Pneumocytosis                                               0.78004
HIV disease                                                 0.52396
HTLV-III/LAV (b)                                            0.3044
HIV disease                                                 0.40821
Positive serum or viral culture findings for HTLVLA (b)     0.21164

(a) Presence of condition = 1.

(b) HTLV is human T-cell lymphocytic virus; III is type III; LAV is
lymphadenopathy-associated virus.

Table A2. Factor Loadings for Comorbidity Index

Indicators of Other Health Conditions (a)    Factor Loadings

Infectious disease                               0.51559
Neoplasm                                         0.39736
Metimmune disease                                0.48544
Immune disorder                                  0.28804
Blood disease                                    0.4367
Nervous disorder                                 0.49236
Circulatory problem                              0.53823
Respiratory problem                              0.55605
Pneumonia                                        0.37818
Digestive problem                                0.53476
Genitourinary problem                            0.41461
Skin disease                                     0.41601
Musculoskeletal problem                          0.47664
Other diagnosis                                  0.59167
Injuries                                         0.46771
Drug dependency                                  0.12262

(a) Presence of condition = 1.

Table 1. AIDS Population Estimates and Waiver Enrollees by State

State                 No. of PWAs (a)    AIDS Rate (per 100,000)

California (b)            45,428                  12.5
Colorado (b)                3025                   7.5
Delaware (b)                1367                  31.1
Florida (c)               38,742                  31.1
Iowa (b)                     623                   3.1
Illinois (b)              10,717                  10.6
Missouri (b)                4418                   8.0
North Carolina (b)          4886                   9.4
New Jersey (c)            15,495                  22.1
New Mexico (b)               997                   6.3
Pennsylvania (b)          11,839                  13.6
South Carolina (b)          4941                  18.2
Virginia (b)                6147                  14.4

State                 No. of Medicaid Waiver Enrollees

California (b)                      1424
Colorado (b)                         144
Delaware (b)                         410
Florida (c)                         6700
Iowa (b)                              17
Illinois (b)                        1294
Missouri (b)                          72
North Carolina (b)                    21
New Jersey (c)                       787
New Mexico (b)                        60
Pennsylvania (b)                      90
South Carolina (b)                  1073
Virginia (b)                         564

(a) Counts of PLWAs as of December 2001.

(b) Counts of participants in AIDS waiver as of June 2001.

(c) Counts of participants in AIDS waiver as of June 2002.

Table 2. Services Available under Medicaid AIDS-Specific Home- and
Community-Based Waivers

                                                   State

Type of Service (a)                      CA    CO    DE    FL    IA

Case management                          T           T     T
Skilled nursing care                     T     T           T     T
Homemaker services                       T     T           T
Environmental modifications              T                 T     T
Transportation                           T     T           T     T
Medical equipment and supplies           T     T     T     T     T
Attendant care                           T           T           T
Psychosocial counseling                  T           T
Home delivered meals                     T           T     T
Nutritional counseling                   T           T           T
Nutritional supplements                              T
Personal care                                  T     T     T
Consumable medical services                                T
Foster parent personal care                                T
Education and support services chores                      T     T
Day health care                                T
Home modification                                          T
Community respite care                                     T
Companion services                                         T
Adaptive equipment                                         T
Massage or physical therapy                                T
Chores/housekeeping                                        T
Pest control                                               T
Electronic monitoring                          T

                                                   State

Type of Service (a)                      IL    MO    NC    NJ    NM

Case management                          T                 T     T
Skilled nursing care                     T     T     T     T     T
Homemaker services                       T                       T
Environmental modifications              T
Transportation                                 T
Medical equipment and supplies                 T
Attendant care                           T           T
Psychosocial counseling
Home delivered meals                     T     T     T
Nutritional counseling
Nutritional supplements
Personal care                            T     T           T     T
Consumable medical services
Foster parent personal care                                T
Education and support services chores
Day health care                                      T     T
Home modification
Community respite care                               T
Companion services                                   T
Adaptive equipment                       T
Massage or physical therapy
Chores/housekeeping                            T
Pest control
Electronic monitoring

                                               State

Type of Service (a)                      PA    SC (b)    VA

Case management                                  T       T
Skilled nursing care                     T       T       T
Homemaker services                       T
Environmental modifications                      T
Transportation
Medical equipment and supplies           T
Attendant care                                   T
Psychosocial counseling
Home delivered meals
Nutritional counseling                   T       T       T
Nutritional supplements                  T       T       T
Personal care
Consumable medical services
Foster parent personal care
Education and support services chores
Day health care                          T
Home modification
Community respite care                                   T
Companion services                               T
Adaptive equipment
Massage or physical therapy
Chores/housekeeping
Pest control
Electronic monitoring

(a) T = service available in the state program.

(b) South Carolina also offers extended prescription drugs under the
waiver.

Table 3. Descriptive Statistics for Variables Used in the Empirical
Models

                                                  All PWAs
                                                (N = 10,836)

Dependent variables (12/95-12/97)
  Waiver selection (=1)                            0.49
  Received combination drug therapy (=1)           0.415
  Received inpatient care (=1)                     0.50
  Average monthly expenditures ($)              1290 (1887)
  Average monthly inpatient expenditures ($)     227 (758)
  Average monthly drug expenditures ($)          631 (1196)
  Died after 12/95 (=1)                            0.172
  Length of survival, months                      22.48
Independent Variables
  White (=1)                                       0.50
  Black (=1)                                       0.42
  Hispanic (=1)                                    0.08
  Female (=1)                                      0.47
  Per capita income in the zip code ($)           12,764
  Percentage of zip code urban                    81.3
  AIDS index as of 12/95                           0
  Comorbidity index as of 12/95                    0
  Pregnant before 12/95                            0.18
  Home health agencies per county population       0.074
  Pharmacies per Medicaid patient                  0.396

                                                   Waiver
                                                Participants
                                                 (N = 5396)

Dependent variables (12/95-12/97)
  Waiver selection (=1)                              --
  Received combination drug therapy (=1)           0.592
  Received inpatient care (=1)                     0.346
  Average monthly expenditures ($)              1622 (2030)
  Average monthly inpatient expenditures ($)     155 (598)
  Average monthly drug expenditures ($)          953 (1415)
  Died after 12/95 (=1)                            0.219
  Length of survival, months                      21.71
Independent Variables
  White (=1)                                       0.58
  Black (=1)                                       0.34
  Hispanic (=1)                                    0.08
  Female (=1)                                      0.26
  Per capita income in the zip code ($)           13,463
  Percentage of zip code urban                    84.4
  AIDS index as of 12/95                           0.198
  Comorbidity index as of 12/95                   -0.046
  Pregnant before 12/95                            0.05
  Home health agencies per county population       0.069
  Pharmacies per Medicaid patient                  0.393

                                                 Non-waiver
                                                participants
                                                 (N = 5440)

Dependent variables (12/95-12/97)
  Waiver selection (=1)                              --
  Received combination drug therapy (=1)           0.254
  Received inpatient care (=1)                     0.658
  Average monthly expenditures ($)               961 (1670)
  Average monthly inpatient expenditures ($)     299 (882)
  Average monthly drug expenditures ($)          312 (812)
  Died after 12/95 (=1)                            0.125
  Length of survival, months                      23.25
Independent Variables
  White (=1)                                       0.43
  Black (=1)                                       0.50
  Hispanic (=1)                                    0.07
  Female (=1)                                      0.67
  Per capita income in the zip code ($)           12,070
  Percentage of zip code urban                    78.3
  AIDS index as of 12/95                          -0.197
  Comorbidity index as of 12/95                    0.045
  Pregnant before 12/95                            0.42
  Home health agencies per county population       0.08
  Pharmacies per Medicaid patient                  0.400

Table 4. Probit Models of Waiver Selection, Receipt of Combination
Drug Therapy, and Use of Inpatient Care (a)

Variable                       Waiver Selection (b)

Intercept                            1.135 ***
                                    (0.125)
Black (=1)                          -0.157 ***
                                    (0.030)
Hispanic (=1)                       -0.055
                                    (0.052)
Female (=1)                         -0.674 ***
                                    (0.031)
Age (divided by 100)                 0.003
                                    (0.144)
Per capita income ($1000s)           0.015 ***
                                    (0.003)
Percentage of zip code area         -0.035
  that is urban                     (0.040)
AIDS index                           0.231 ***
                                    (0.016)
Comorbidity index                   -0.101 ***
                                    (0.016)
Pregnant before 12/95               -0.909 ***
                                    (0.047)
Home health agencies                -8.069 ***
  per population                    (0.512)
Pharmacies per                      -0.480 ***
  Medicaid patient                  (0.228)
Waiver participant                      --

                                Combination Drug Therapy (c)

Variable                         Probit      Bivariate Probit

Intercept                      -0.273          -1.065 ***
                               (0.119)         (0.156)
Black (=1)                      0.157 ***       0.226 ***
                               (0.029)         (0.038)
Hispanic (=1)                  -0.005           0.025
                               (0.050)         (0.065)
Female (=1)                    -0.345 ***      -0.162 ***
                               (0.030)         (0.040)
Age (divided by 100)            0.099           0.001
                               (0.140)         (0.002)
Per capita income ($1000s)     -0.000          -0.004
                               (0.003)         (0.004)
Percentage of zip code area     0.375 ***       0.424 ***
  that is urban                (0.039)         (0.049)
AIDS index                      0.133 ***       0.078 ***
                               (0.015)         (0.020)
Comorbidity index              -0.068 ***      -0.043 ***
                               (0.015)         (0.020)
Pregnant before 12/95          -0.550 ***      -0.350 ***
                               (0.043)         (0.054)
Home health agencies            0.432           1.831 ***
  per population               (0.471)         (0.591)
Pharmacies per                 -0.226          -0.111
  Medicaid patient             (0.213)         (0.268)
Waiver participant                 --           0.825 ***
                                               (0.029)

                                    Inpatient Care (d)

Variable                         Probit      Bivariate Probit

Intercept                       0.184           0.964 ***
                               (0.111)         (0.150)
Black (=1)                      0.091 ***       0.046
                               (0.028)         (0.038)
Hispanic (=1)                   0.041           0.025
                               (0.048)         (0.063)
Female (=1)                     0.371 ***       0.174 ***
                               (0.030)         (0.040)
Age (divided by 100)           -0.610 ***       0.700 ***
                               (0.140)         (0.179)
Per capita income ($1000s)     -0.015          -0.012 ***
                               (0.003)         (0.004)
Percentage of zip code area     0.093           0.080
  that is urban                (0.036)         (0.048)
AIDS index                     -0.076 ***      -0.013
                               (0.015)         (0.022)
Comorbidity index              -0.014          -0.049 ***
                               (0.015)         (0.020)
Pregnant before 12/95          -0.345 ***      -0.597 ***
                               (0.039)         (0.054)
Home health agencies            3.687 ***       1.853 ***
  per population               (0.436)         (0.580)
Pharmacies per                 -0.634 ***      -0.755 ***
  Medicaid patient             (0.189)         (0.252)
Waiver participant                 --          -0.858 ***
                                               (0.029)

(a) Probit coefficients and standard errors are in parentheses.

(b) Waiver selection is equal to one if the PWA chooses the waiver and
is equal to zero if the PWA chooses traditional care.

(c) Combination drug therapy is equal to one if the PWA is given one of
the two combination drug treatments described in section B and is equal
to zero if the PWA does not receive any combination treatment over the
two-year period.

(d) Inpatient care is equal to one if the PWA is hospitalized or placed
in a nursing home and is equal to zero if the PWA is not placed in an
institution during the two-year period.

*** Significant at the 1% level.

Table 5. Average Monthly Expenditures Models: OLS and IV (a)

                                       Log Monthly Expenditures:
Variables                                       OLS (b)

Intercept                                  5.611 *** (0.083)
Waiver participant (=1)                    0.522 *** (0.031)
Black (=1)                                 0.062 *** (0.03)
Hispanic (=1)                               0.095 ** (0.053)
Gender (female = 1)                       -0.238 *** (0.001)
Age                                        0.013 *** (0.001)
AIDS index                                 0.149 *** (0.017)
Comorbidity index                          0.252 *** (0.015)
Ever pregnant (=1)                        -0.832 *** (0.043)
Per capita income (divided by 1000)            0.003 (0.003)
Percentage of area that is urban           0.409 *** (0.039)
F statistic                                    324.73
[R.sup.2]                                      0.255
Sample size                                     9510

                                       Log Monthly Expenditures:
Variables                                        IV (b)

Intercept                                  6.243 *** (0.132)
Waiver participant (=1)                   -0.545 *** (0.170)
Black (=1)                                   -0.0002 (0.034)
Hispanic (=1)                                  0.060 (0.057)
Gender (female = 1)                       -0.513 *** (0.056)
Age                                        0.013 *** (0.002)
AIDS index                                 0.240 *** (0.023)
Comorbidity index                          0.240 *** (0.018)
Ever pregnant (=1)                        -1.091 *** (0.061)
Per capita income (divided by 1000)        0.009 *** (0.003)
Percentage of area that is urban           0.432 *** (0.041)
F statistic                                   263.11
[R.sup.2]                                     0.159
Sample size                                    9510

(a) Regression coefficients and standard errors are in parentheses.

(b) Monthly expenditures are equal to the average of all Medicaid
claims over the two-year period.

** Significant at the 5% level.

*** Significant at the 1% level.

Table 6. OLS and IV Regressions Predicting Survival, Weibull Hazard
Model Predicting the Probability of Death (a)

Variables                               Survival: OLS (b)

Intercept                                    3.045 (0.030)
Waiver participant (=1)                      -0.06 (0.011)
Black (=1)                              -0.062 *** (0.011)
Hispanic (=1)                                0.007 (0.019)
Gender (female = 1)                       0.059 ** (0.012)
Age                                         0.0006 (0.0005)
AIDS index                               0.062 *** (0.006)
Other health index                      -0.055 *** (0.006)
Ever pregnant (=1)                       0.100 *** (0.016)
Per capita income (divided by 1000)         -0.001 (0.001)
Percentage of area that is urban            -0.016 (0.014)
Parameter p (d)
F statistic                                     66
[R.sup.2]                                      0.057
Likelihood ratio chi squared (10)
Sample size                                   10,836

Variables                               Survival: IV (b)

Intercept                                   3.028 (0.053)
Waiver participant (=1)                    -0.032 (0.071)
Black (=1)                             -0.060 *** (0.012)
Hispanic (=1)                               0.008 (0.019)
Gender (female = 1)                     0.066 *** (0.022)
Age                                    0.0006 *** (0.0005)
AIDS index                             -0.064 *** (0.008)
Other health index                     -0.053 *** (0.006)
Ever pregnant (=1)                      0.107 *** (0.024)
Per capita income (divided by 1000)        -0.001 (0.001)
Percentage of area that is urban           -0.017 (0.014)
Parameter p (d)
F statistic                                    63.02
[R.sup.2]                                      0.057
Likelihood ratio chi squared (10)
Sample size                                   10,836

                                           Weibull
Variables                              Hazard Model (c)

Intercept                                     --
Waiver participant (=1)                   1.135
Black (=1)                                1.493 ***
Hispanic (=1)                             0.878
Gender (female = 1)                       0.721 ***
Age                                       1.004 ***
AIDS index                                1.214 ***
Other health index                        1.290 ***
Ever pregnant (=1)                        0.344 ***
Per capita income (divided by 1000)       1.000
Percentage of area that is urban          1.231 ***
Parameter p (d)                           0.969 ***
F statistic
[R.sup.2]
Likelihood ratio chi squared (10)           694.36
Sample size                                 10,836

(a) Cells contain coefficients with standard errors in parentheses.

(b) Survival is the number of months alive during the two-year period.

(c) Cells contain the odd ratios derived from the hazard coefficients.
An odds ratio greater than one indicates that the independent variable
increases the probability of death.

(d) p greater than one in the Weibull Hazard Model indicates that the
probability of dying increases over time.

*** Significant at the 1% level.

This research was supported by funding from the Agency for Health Care Quality and Research under grant HS 09560. We thank Paul Arons. M.D., Medical Director of the Bureau of HIV/AIDS. Florida Department of Health, for his clinical expertise and helpful discussions.

(1) PWAs under 18 were excluded from the analysis because AIDS manifests itself differently in infants and children than in adults. As a consequence, treatment protocols recommended for infants and children differ from those recommended for adults.

(2) We estimate survival in months as opposed to days because the exact day of death was missing for a substantial number of persons although the month and year of death were always reported.

(3) We consulted with Florida Medicaid program staff to develop the appropriate control group given that we had no access to laboratory data (CDC counts) on these patients. The staff, which included the medical director of the bureau of HIV/AIDS, agreed on the appropriateness of our selection criteria.

(4) In the absence of a direct medical measure of severity, but assuming that there is an unobserved latent severity measure, factor analysis generates a valid estimate of severity. If all of the factor loadings were equal, the result would be the average of the indicators, so that measure is a special case here.

(5) This normalization could be done with any positive definite matrix, but in using the variance-covariance matrix for v, the variance in A is equal to one.

(6) Mitchell and Anderson (2000) present descriptive evidence on expenditures for waiver enrollees and non-waiver enrollees for the entire 1993-1997 period; expenditures during the years before and after combination drug therapy were available and included in the calculation of mean expenditures per month. With the inclusion of these early claims, the mean difference in expenditures was higher for non-waiver participants than for waiver enrollees. On average, total monthly expenditures among non-waiver participants were more than double the monthly expenditures of waiver participants. This was largely due to the very large mean difference in inpatient expenditures. On average, inpatient expenditures were 510% higher among non-waiver participants than among waiver enrollees.

(7) The choice of the waiver is associated with a different package of services compared to those available under traditional care. We therefore include the same set of explanatory variables (X) in the selection ([P.sup.*]) and the service ([H.sup.*.sub.j]) models.

(8) An alternative estimation strategy is to estimate a seemingly unrelated selection-adjusted system of equations. This approach is a more complex econometric model because of the selection adjustment and the dichotomous dependent variables. The major shortcoming of our approach is that we assume the input demand equations are independent.

(9) If the 5440 nonparticipants were on the waiver, estimated expenditure savings are ($552.45 per non-waiver participant PWA multiplied by 5440 non-waiver participant PWAs) = $3,005,309.

(10) We estimated our expenditure models separately by race but do not report the results in this article. We find that expenditures are lower under the waiver for blacks and Hispanics, but the waiver has no effect on the expenditures of whites.

(11) We estimate the survival models separately by race and find no effect of the waiver on the survival of whites, blacks, or Hispanics.

References

Anderson, Kathryn H., and Jean M. Mitchell. 1997. Expenditures on services for AIDS patients under a Medicaid home and community-based waiver program: Are selection effects important? Medical Care 35:425-39.

Bound, John, Dean A. Jaeger, and Regina M. Baker. 1995. Problems with instrumental variables when the correlation between the instruments and the endogenous explanatory variable is weak. Journal of the American Statistical Association 90: 443-50.

Carpenter, Charles C. J., Margaret A. Fischl, Scott M. Hammer, et al. 1996. Antiretroviral therapy for HIV infection in 1996: Recommendations of an international panel. Journal of the American Medical Association 276:146-54.

Carpenter, Charles C. J., Margaret A. Fischl, Scott M. Hammer, et al. 1997. Antiretroviral therapy for HIV infection in 1997: Updated recommendations of the international AIDS society-USA panel. Journal of the American Medical Association 277:1962-9.

Crystal, Stephen, Usha Sambamoorthi, and A. Lo Sasso. 1998. Impact of case-managed HIV care program on health care access and expenditures. Unpublished paper, Rutgers University.

Harman, Harry H. 1976. Modern factor analysis. 3rd edition. Chicago: University of Chicago Press.

Hayashi, Fumio. 2000. Econometrics. Princeton, NJ: Princeton University Press.

Hogg, Robert S., Katherine V. Heath, Benita Yip, Kevin J. P. Craib, Michael V. O'Shaughnessy, Martin T. Schechter, and Julia S. G. Montaner. 1998. Improved survival among HIV infected individuals following initiation of antiretroviral therapy. Journal of the American Medical Association 279:450-4.

Lindsey, Phoebe A., Peter D. Jacobson, and Anthony H. Pascal. 1990. Medicaid home and community-based waivers for acquired immunodeficiency syndrome patients. Health Care Financing Review Annual Supplement:109-18.

Maddala, G. S. 1983. Limited-dependent and qualitative variables in econometrics. Cambridge, UK: Cambridge University Press.

Merzel, Cheryl S., Stephen Crystal, Usha Sambamoorthi, Daniel Kraus, and Carol Kurland. 1992. New Jersey's Medicaid waiver for acquired immunodeficiency syndrome. Health Care Financing Review 13:27-44.

Mitchell, Jean M., and Kathryn H. Anderson. 2000. Effects of case management and new drugs on Medicaid AIDS spending. Health Affairs 19:233-43.

Palella, Frank J., Kathleen M. Delaney, Anne C. Moorman, Mark O. Loveless, Jack Fuhrer, Glen A. Satten, Diane J. Aschman, and Scott D. Holmberg. 1998. Declining morbidity and mortality among patients with advanced HIV infection. New England Journal of Medicine 338:853-60.

Sambamoorthi, Usha, Sara R. Collins, Stephen Crystal, and James Walkup. 1999. Home-care use and expenditures among Medicaid beneficiaries with AIDS. Health Care Financing Review 20:161-77.

Weissart, W. G., C. M. Cready, and J. E. Pawalek. 1988. The past and future of home- and community-based long-term care. Milbank Quarterly 66:309-88.

Kathryn Anderson * and Jean Mitchell ([dagger])

* Department of Economics, Vanderbilt University, Box 351819 Station B, Nashville, TN 37235-1819, USA; E-mail [e-mail]; corresponding author.

([dagger]) Georgetown Public Policy Institute, Georgetown University, 3600 N Street NW, Suite 200, Washington, DC 20(107, USA; E-mail [e-mail].

Received August 2002; accepted November 2003.