Customize your JAMA Network experience by selecting one or more topics from the list below.
Anderson KH, Mitchell JM. Differential Access in the Receipt of Antiretroviral Drugs for the Treatment of AIDS and Its Implications for Survival. Arch Intern Med. 2000;160(20):3114–3120. doi:10.1001/archinte.160.20.3114
Recently published research based on selected samples of patients treated at human immunodeficiency virus clinics documents that use of more intensive antiretroviral drug therapies is responsible for significant declines in morbidity and mortality in persons living with human immunodeficiency virus or acquired immunodeficiency syndrome (PLWHAs). In this study, we evaluate whether receipt of more recently developed antiretroviral therapies varies by sex and race/ethnicity in a large population-based sample of PLWHAs and whether receipt of such drugs has any impact on survival.
Analysis of Florida Medicaid eligibility, enrollment, and claims data for PLWHAs for 1993 through 1997. Receipt of 2 nucleoside analogs (TWONUKES) and receipt of 1 protease inhibitor and a nucleoside combination (PI+NUKES) was constructed from claims data. The probability of dying was constructed from eligibility and enrollment data.
The probabilities of receiving TWONUKES and PI+NUKES are 0.16 and 0.09, respectively, lower for women relative to men (P<.01 for both). Blacks are more likely to receive TWONUKES than whites, whereas the reverse is true for Hispanics; this probability is almost 0.04 higher for blacks and 0.03 lower for Hispanics relative to whites (P<.01). In contrast, blacks are significantly less likely to receive PI+NUKES (P<.01). Both drug variables have large statistically significant negative effects on the probability of death. The PLWHAs who received PI+NUKES are 60% as likely to die each month (P<.01). Receipt of TWONUKES lowers the relative hazard of death by close to 66% each month (P<.01). Survival varies significantly by sex and race/ethnicity. Controlling for receipt of drug therapy and diagnosed health throughout the period, women are 56% as likely to die as men (P<.01). Hispanics are almost 14% less likely to die each month relative to whites (relative hazard, 0.87), and blacks are 20% more likely to die than whites (relative hazard, 1.21).
States need to investigate why women are less likely to receive antiretroviral drug therapies than men and to consider policies that might foster better access to antiretroviral therapies for women with acquired immunodeficiency syndrome because these efforts might yield even further reductions in mortality in women. Given the large reductions in mortality that accompany receipt of antiretroviral therapies, states need to foster policies that promote widespread use of new drug treatment protocols.
TREATMENT of acquired immunodeficiency syndrome (AIDS) has become more promising yet potentially more costly in recent years with development of protease inhibitors and nonnucleoside analog reverse transcriptase inhibitors. Such drug treatment protocols have been found to suppress viral replication, increase production of CD4+ cells, reduce morbidity, and prolong survival.1-4 Recently published research documenting such findings is based on selected samples of patients treated at human immunodeficiency virus (HIV) clinics. There are significant advantages to studying clinical samples because detailed information is available on patient socioeconomic characteristics, attendance at clinical follow-up appointments, and laboratory data such as CD4 cell counts and viral load. Such information is important in understanding the relation between access, compliance, degree of immunosuppression, and clinical response. Nonetheless, findings based on clinical samples are subject to sample selection bias if such persons living with HIV or AIDS (PLWHAs) are more prone to take the recommended medications as prescribed by their physicians than PLWHAs who are not treated at HIV specialty clinics. Another source of selection bias is that such patients tend to be relatively homogeneous with respect to demographic and socioeconomic characteristics. If this is the case, one cannot generalize the results derived from selected clinical samples to the entire population of PLWHAs.
Given that the new drug therapies are costly ($12,000-$20,000 per person-year), it seems likely that their expense can deplete the financial resources of PLWHAs so that they qualify for Medicaid, state medically needy programs, or the AIDS Drug Assistance Program. The PLWHAs enrolled in state Medicaid programs represent a lower-income, population-based sample that is not subject to the potential biases that limit the generalizability of findings based on selected clinical samples. Although within a state Medicaid program resource constraints do not explain differential access to care, the receipt of the new drug therapies and their impact on survival are expected to vary by sex and race/ethnicity for at least 4 reasons.5,6 First, there might be differences by sex and ethnicity in the risk and effectiveness of different treatments. Second, there might be differences in the socioeconomic and psychological backgrounds of patients that require differences in treatments. Differences by sex and race in income and residence might reflect differential access to care and the ability to pay for the best treatment. Third, demographic differences in treatment might reflect patients' preferences. Women, eg, might be concerned about effects of some treatments on a fetus or might be more time constrained and unable to maintain regular care than men (CDC Fact Sheet, accessed at CDC Web site, November 1998). Finally, there might be sex or racial bias in the provision of care. If physicians or case managers base treatment on "inflexible attitudes about men and women (their ailments, time constraints, treatment preferences, psyches, etc.) . . . ," then differences in treatment may result, and these differences are not consistent with "sound medical practice."5 It is difficult to empirically isolate all 4 of these causes of differential treatment, but with adequate controls for stage of disease and socioeconomic circumstances, explanations for remaining differences likely center on differences in preferences and provider bias. Separating these causes is not possible with our data.
Previous research7-9 based on selected samples of homosexual and bisexual men found that use of HIV-related drug therapies was most common among educated white men who had health insurance and displayed symptoms of HIV. Results of other studies10,11 suggest that income is an important determinant of access to and use of HIV-related drug therapy. Regarding sex, some research12-16 has found that women with HIV disease have lower rates of medication use than men. The limited evidence on race is conflicting. For example, Moore et al12 reported that blacks were significantly less likely than whites to receive antiretroviral drug therapy or Pneumocystis carinii pneumonia prophylaxis. On the other hand, Smith and Kirking16 found that the odds of using either antiretroviral or P carinii pneumonia medications were higher in blacks compared with whites.
Although results of published research suggest that use of antiretroviral drug therapies for the treatment of HIV and AIDS varies by race and sex, these studies are based on data from the late 1980s and early 1990s and consequently do not capture the use of more recently developed drug therapies. Moreover, most previous research is based on data from selected samples. To our knowledge, no previous research has examined whether use of more recently developed drug therapies varies by sex and race/ethnicity in a large population-based sample and whether receipt of such drugs has implications for survival.
In this study, we analyzed Florida Medicaid eligibility and claims data for PLWHAs for January 1, 1993, through December 31, 1997, to evaluate whether receipt of antiretroviral drug therapies varies by sex and race/ethnicity and whether such differences in access have any impact on survival. During fiscal years 1997 and 1998, the Florida Medicaid program served 18,673 beneficiaries with HIV or AIDS. The Florida Department of Health estimates that there were approximately 82,500 PLWHAs during this period. Assuming that the Florida Medicaid program had successfully identified all of their recipients with HIV or AIDS, the proportion of the HIV-infected population that is actually receiving Medicaid services would be about 23%. These estimates are probably conservative because there are some PLWHAs who did not use any services during fiscal years 1997 and 1998.
Examination of PLWHAs enrolled in the Florida Medicaid program has potentially significant policy implications for at least 3 reasons. First, Florida ranks third among the states with about 10% of reported AIDS cases in the United States, although Florida accounts for only 5% of the US population.17 Second, the Florida AIDS population is demographically diverse in that it comprises a higher proportion of women, blacks, Hispanics, and children compared with the US AIDS population.17 Third, results based on an analysis of the use of antiretroviral drug therapies and their effects on survival in PLWHAs enrolled in state Medicaid programs can be generalized to lower-income populations because such findings are not subject to the potential sample selection bias that might hamper clinical samples.
Data for this research came from records of Medicaid claims for medical services in Florida provided between January 1, 1993, and December 31, 1997. All claims were screened for diagnosis codes or prescribed drugs used to treat HIV- or AIDS-related illness. Florida Medicaid recipients were 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- or 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 which it identifies individuals for inclusion in our sample who do not actually have HIV or AIDS.
Records are available for every person in the state who filed at least 1 Medicaid claim during this period and is HIV positive or has full-blown AIDS according to the screening criteria developed by Medicaid program staff. In this study, we focus on the treatment of adults in the working age population (aged 18-64 years). Thus, we excluded all claims for persons receiving Medicaid who are younger than 18 years. Yet, if a Medicaid recipient turned 18 years old between January 1, 1993, and December 31, 1997, we included this person in our analysis but excluded all claims filed at ages younger than 18 years. We also excluded all persons who, at any point between January 1, 1993, and December 31, 1997, turned age 65 years. Because our analysis focuses on Medicaid recipients with AIDS who are white, black, and Hispanic, we excluded Medicaid recipients classified as "other race." Finally, to focus on the population who are potentially eligible to receive the more recently available AIDS drug therapies, we excluded persons who died before January 1, 1996, and those who did not use any services between January 1, 1996, and December 31, 1997. After making these exclusions, the sample consisted of 24,646 PLWHAs enrolled in the Florida Medicaid program between January 1, 1993, and December 31, 1997.
For each person in our database, we have records for each month in which a claim for Medicaid services was filed. Each record contains information about the claim and 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, sex, and county of residence. We have no information on diagnosis codes before January 1993; we cannot determine, therefore, the month in which patients were first diagnosed with HIV or AIDS if they were receiving Medicaid before January 1993. We also have no information on education, income, or employment of recipients and their families at any point in time. We constructed proxy measures from information on county per capita income, percentage of the population in the county with a college education, and urbanization to control for the effects of education and income.
We relied on our physician consultant, Paul Arons, MD, Bureau of HIV/AIDS, Florida Department of Health, to define the drug therapies for the treatment of HIV and AIDS. During the study 3 types of drugs were used to treat HIV: nucleoside analogs (zidovudine, didanosine, zalcitabine, stavudine, and lamivudine), nonnucleoside analogs (nevirapine and deavirdine), and protease inhibitors (saquinavir mesylate, ritonavir, indinavir, and nelfinavir). Drugs from these 3 groups are used in combination with each other to reduce HIV viral load and to increase the level of CD4 lymphocyte cells. Recommended treatment for PLWHAs with antiretroviral drugs was published in 2 consensus statements issued by a panel of HIV and AIDS experts.18,19 In Florida, the single drug therapies were available to Medicaid patients in 1993 and the combination therapies were not available to Medicaid patients until the spring of 1996.
Using the pharmaceutical claims on each patient between 1993 and 1997, we constructed 2 variables to measure use of the more recently developed drug therapies. The first identifies individuals who received 2 nucleoside analog drugs (TWONUKES). If more than one drug therapy is used in conjunction with 1 or more other drugs, this is referred to as a combination drug regimen. The second variable identifies combination drug therapies consisting of 1 protease inhibitor and a pair of nucleoside analogs (PI+NUKES). (Physicians may substitute a nonnucleoside analog for a PI if a patient develops a resistance to a PI.)
We collapsed the Medicaid claims records to construct a single record for each person. This file contains information on race; sex; date of first claim between January 1, 1993, and December 31, 1997; age at first claim; county characteristics for county of last claim; date of death; length of life by December 31, 1997; diagnoses; and drug therapies. Race/ethnicity is measured with 2 dummy variables for black or Hispanic; white is the omitted category. Sex is a dummy variable equal to 1 if the patient is a woman. Age is measured in years. Date of first claim is an index equal to 1 for January 1993 and increasing to 60 for December 1997. Lifetime is measured as months alive after first claim between January 1, 1993, and December 31, 1997. County characteristics include median per capita income, percentage of the county population in urban areas, and percentage of the county population with a college education. We do not have county information for every person in our sample, and we created a dummy variable equal to 1 if county data were missing; 1794 persons report no county code.
Using information reported in the diagnosis code field on each claim, we identified all AIDS-defining diagnoses and all AIDS-supporting diagnoses (Table 1). We used factor analysis to construct an index of AIDS-related conditions. This index proxies for severity of disease, with higher values representing greater severity. To control for other comorbidities, we constructed a series of dummy variables to identify whether a person ever had a diagnosis code identifying 1 of the following broad categories of physical health problems: infectious disease, malignant neoplasm, blood disease, nervous system disorder, circulatory system problem, respiratory problem, pneumonia, digestive system problem, genitourinary problem, skin problem, musculoskeletal problem, drug dependency, and other symptoms as yet undiagnosed. To eliminate potential collinearity problems between the disease indicators, we again used factor analysis to construct an index of other comorbidities. We also constructed a separate dummy variable to identify women who were pregnant at some point during the 5-year study. Finally, we identified PLWHAs with mental health problems by constructing a dummy variable to identify those with at least 1 diagnosis code indicating a mental disorder.
We constructed 2 dependent variables to measure the use of antiretroviral drug therapies: receipt of TWONUKES and receipt of PI+NUKES. Because the dependent variables are dichotomous, we estimated each model using probit analysis. The independent variables in this model include sex; race; age at first claim; month of first claim; county measures of income, urban residence, and college education; and controls for health status (the AIDS index, comorbidities index, and dummy variables indicating pregnancy and the presence of mental health disorders). Next, we estimated a Cox proportional hazards model measuring survival until December 31, 1997 (the last point of observation). The independent variables include the 2 drug therapy variables (TWONUKES and PI+NUKES), sex, race, age, month of first claim, county characteristics, and controls for disease severity and comorbidities.
Table 2 reports descriptive statistics for all PLWHAs and for the subsamples of men, women, whites, blacks, and Hispanics. Among 24,646 PLWHAs who were still alive as of January 1, 1996, 41% received TWONUKES and 20% received PI+NUKES. Receipt of the AIDS drug therapies, however, varies significantly by sex: a higher proportion of men received each of the drug therapies. About 57% of men received TWONUKES and 30% received PI+NUKES. In contrast, only 27% of women received TWONUKES and almost 12% received PI+NUKES. On the other hand, without controlling for other factors, differences in the receipt of the 2 drug therapies by race/ethnicity are small and for the most part are not statistically significant. Between 1996, and 1997, about 13% of PLWHAs died, and the average survival time or time until the last point of observation was almost 38 months. Men are significantly more likely to die than women (19.8% vs 7.7%), and mean survival time for men is shorter, about 31 months compared with almost 44 months for women. Although it seems that Hispanics are significantly less likely to die than whites or blacks, this may be an artifact of our data because the period for which we observe them is shorter than for whites or blacks. Specifically, Hispanics were enrolled in Medicaid for a shorter period than whites or blacks. Regarding other characteristics, 49% of PLWHAs were black, 42% were white, and 9% were Hispanic. Women account for 53% of the sample. The mean age of PLWHAs is 37 years, and more than 77% reside in an urban area. Mean per capita income in the county is just less than $12,000, and less than 10% of the population in each county has a college degree.
Results of the estimation of the probit models are presented in Table 3. The pseudo R2 values indicate that the independent variables account for about 15% and 11% of the variation in the probability of receiving TWONUKES and PI+NUKES, respectively. Moreover, the models have good predictive power because the predicted probabilities of receiving each drug therapy are close to the actual observed probability in the sample. We report the marginal effects of each independent variable on the probability of receipt of the drug therapy. The marginal effects represent the change in the probability associated with the presence of each characteristic relative to the reference group.
The results show that significant sex differences exist for receipt of both drug therapies. The probability of receiving TWONUKES is 0.16 lower for women relative to men (P<.01). Similarly, the likelihood of receiving PI+NUKES is almost 0.09 lower for women compared with men (P<.01). With respect to race/ethnicity, blacks are more likely to receive TWONUKES than whites, whereas the reverse holds for Hispanics; this probability is almost 0.04 higher for blacks and 0.002 lower for Hispanics relative to whites (P<.01). In contrast, blacks are significantly less likely to receive PI+NUKES, but this difference is small. Moreover, differences between whites and Hispanics are negligible.
Furthermore, it seems that the likelihood of receiving each of the drug therapies increases with age. The probability of receiving TWONUKES is greater in counties with higher per capita income but lower in urban areas and counties with a greater proportion of college graduates. Residence in an urban area increases the receipt of PI+NUKES, yet the proxies for income and education have trivial effects. Date of first claim matters; the later the first claim, the more likely the patient is to receive TWONUKES or PI+NUKES. The effect of date of first claim may be related to the availability of combination therapies. We found that the index of AIDS-related conditions is positive and highly significant for both drug therapies. This implies that greater severity of AIDS increases the likelihood that a PLWHA will receive each of the 2 drug therapies. This is also the case for the index measuring the presence of comorbidities. As expected, women who were pregnant at some point during the 5 years are less likely to receive each of the drug therapies. This can be attributed to providers' reluctance to recommend combination therapies for pregnant women given the possible adverse effects these drugs might have on the fetus.5 Finally, mental disorders have a small positive impact on the probability of receiving TWONUKES but negligible effects on receipt of PI+NUKES.
These hazard model results are reported in Table 4. The coefficients reported are the marginal effects of the independent variables on the relative hazard rate. Both drug variables have large statistically significant negative effects on the probability of death. The PLWHAs who received PI+NUKES are 60% as likely to die each month (P<.01). Similarly, PLWHAs who received TWONUKES are 40% as likely to die each month (P<.01). We also found that survival varies significantly by sex and race/ethnicity. Controlling for the receipt of drug therapy and diagnosed health throughout the period, women are 56% as likely to die as men (P<.01). Regarding race/ethnicity, Hispanics are almost 14% less likely to die each month relative to whites (relative hazard, 0.86) and blacks are 20% more likely to die than whites (relative hazard, 1.21).
Other results are for the most part as expected. The probability of death increases with age and is also higher in urban and higher-income counties. As expected, persons with more advanced stages of AIDS, as measured by the AIDS index variable, are significantly more likely to die. Likewise, the presence of other comorbidities increases the probability of death. Women who were pregnant at some point during the 5 years are 28% as likely to die each month compared with other PLWHAs (P<.01). This suggests that pregnant women are healthier or that pregnancy has protective effects. Those who had a diagnosis of a mental health disorder are 72% as likely to die relative to those without such mental problems (P<.01). It seems likely that persons with mental disorders have a less serious form of the disease rather than suggesting that mental disorders reduce the probability of dying of AIDS.
The effects of race and sex in the survival model discussed in the previous paragraph are the direct effects of these demographic characteristics on survival; they measure the effect of race/ethnicity and sex on survival after controlling for the indirect effects of the receipt of drug therapy. For example, women are significantly less likely than men to receive either drug therapy. Because use of antiretroviral drug therapy increases survival in women, the indirect effect of being female on survival is negative. Overall, if we estimate a survival model excluding the drug therapy variables, the effect of sex should still be positive but smaller because we no longer control for this indirect effect.
To demonstrate the importance of these indirect effects, we present in Table 4 the marginal impacts from the hazard estimation that excludes the drug variables. We find, as expected, that the effect of sex is less positive than in our structural model. Women would have higher survival if they were given antiretroviral drug therapy at the same rate as men; the relative hazard of death for women decreases from 0.66 in the model without drugs to 0.56 in the model with drugs. The differences in the effects of race on survival in the structural and reduced from models are small.
Finally, the estimated effects of AIDS and other health conditions on the probability of death are smaller when we exclude the drug therapy measures. These findings, in conjunction with the significance and magnitude of the marginal impacts of the drug variables, suggest that the exclusion of the drug variables results in omitted variables bias.
In this study, we analyzed Florida Medicaid eligibility and claims records for PLWHAs for 1993 through 1997 to examine whether use of antiretroviral drug therapies for the treatment of AIDS varies by sex and race/ethnicity. We also examined whether receipt of antiretroviral drug therapies, sex, and race/ethnicity have any impact on survival. To our knowledge, this study is the first to examine these issues using a large, nonclinical, population-based sample. The results of our analyses, therefore, are not subject to the potential biases associated with analyses of the behavior of PLWHAs treated at HIV and AIDS specialty clinics.
Our most important finding is the persistent and significant sex difference in the prescription of antiretroviral therapies for the treatment of AIDS. Women are much less likely than men to receive the antiretroviral therapies we examined (TWONUKES or PI+NUKES). These significant sex differences in the receipt of antiretroviral therapies persisted even after controlling for the severity of AIDS and other comorbidities. Nonetheless, we also found that despite their lower use of antiretroviral therapies, women are more likely to survive than men under any scenario. Clearly, the persistent lower use of antiretroviral drugs by women with AIDS suggests that their survival with AIDS could be significantly improved if their receipt of antiretroviral medications were at parity with that of men.
Our findings corroborate the low use of antiretroviral drug therapies by women in the HIV Epidemiology Research Study, as reported by Solomon et al,20 and in published studies by Stein et al13 and Turner et al.15 One possible explanation is that in our sample women were healthier than men. This seems unlikely given that we controlled for the severity of AIDS and the presence of comorbidities. Alternatively, physicians might have been reluctant to prescribe antiretroviral therapies for pregnant women because these drugs might have adverse effects on the fetus (CDC Fact Sheet, accessed at CDC Web site, November 1998). Yet, the sex differences in receipt of antiretroviral therapies that exist cannot be attributed to pregnancy because the model included a control to identify women who were ever pregnant during the sample period. As expected, pregnant women were significantly less likely to receive antiretroviral therapies. Furthermore, the lower use of antiretroviral therapies by women cannot be attributed to the fact that they are eligible to receive medications paid for by sources other than the Florida Medicaid program. Although Medicaid provides drug coverage to all eligible enrollees with HIV or AIDS, Medicaid recipients are not eligible for the AIDS Drug Assistance Program, which was established under the Ryan White Comprehensive AIDS Resource Emergency Act to provide access to medications for HIV-positive persons who lack other health care coverage.21
However, whether further prescription of drug therapy is warranted for individual women in our sample cannot be assessed. Women may on average be less likely to comply with this therapy than men, or women may prefer some other form of treatment. In addition, retrenchment may be important; if women are at lower risk for progression of the disease in general, then therapy should be and likely is postponed. Our results are consistent with any of these 3 explanations.
Second, our findings show that receipt of recently developed antiretroviral therapies, in particular, either TWONUKES or PI+NUKES, has a large significant negative effect on the probability of death. Our findings corroborate the results of recently published research1-4 based on selected clinical samples that documents that receipt of antiretroviral drug therapies is associated with significant reductions in mortality. Given that previous research on use of recently developed antiretroviral therapies is based on selected clinical samples, this is the first study to document that recently developed antiretroviral drug therapies yield significant reductions in mortality in large population-based samples.
Our findings have important implications for state Medicaid programs in the development of their policies for PLWHAs. First, it seems that states need to explore the reasons behind the sex differences in access to drug therapy and to consider policies that foster better access to antiretroviral therapies for women with AIDS; these efforts might yield even further reductions in mortality in women. Second, given the large reductions in mortality that accompany receipt of antiretroviral therapies, states need to foster policies that promote widespread use of the drug treatment protocols. Although these promotion policies will result in higher pharmaceutical expenditures, inpatient costs will decline and are likely to dominate the larger expenditures on drugs.22
Despite the significance of our findings, our analyses have some inherent limitations. First, because the time frame we analyzed represents a transition period between monotherapy, dual combination therapy, and triple combination therapy, one could argue that it is misleading to draw conclusions based on this period of great change in therapy. We contend that it is important to examine "transition" periods because access problems are apt to be more acute in the periods immediately after the availability of new drug therapies. Second, because of data availability we are only able to evaluate sex and racial or ethnic differences in the receipt of potent combination therapies for a relatively short period (Spring 1996 through December 1997). Although this might be a limitation of our study, we contend that it is valuable to examine differences in access to combination therapies during the period when they first became available because if access differences exist initially they may persist unless specific efforts are made to eliminate them. Third, our outcome indicator—survival—is limited, but it is the only measure that is available from Medicaid enrollment and claims records. The lack of detailed information on health indicators such as viral load and CD4 counts suggests that state Medicaid programs should make efforts to routinely collect this information.
Accepted for publication May 18, 2000.
This research was supported by grant HS 09560 from the Agency for Health Care Research and Quality, Rockville, Md.
We thank Paul Arons, MD, Bureau of HIV/AIDS, Florida Department of Health, Tallahassee, Fla, for his clinical expertise and helpful discussions.
Corresponding author: Jean M. Mitchell, PhD, Georgetown Public Policy Institute, Georgetown University, 3600 N St NW, Suite 200, Washington, DC 20007 (e-mail: email@example.com).
Create a personal account or sign in to: