Variations in the Care of HIV-Infected Adults in the United States: Results From the HIV Cost and Services Utilization Study

Context Studies of selected populations suggest that not all persons infected with human immunodeficiency virus (HIV) receive adequate care.

Objective To examine variations in the care received by a national sample representative of the adult US population infected with HIV.

Design Cohort study that consisted of 3 interviews from January 1996 to January 1998 conducted by the HIV Cost and Services Utilization Consortium.

Patients and Setting Multistage probability sample of 2864 respondents (68% of those targeted for sampling), who represent the 231,400 persons at least 18 years old, with known HIV infection receiving medical care in the 48 contiguous United States in early 1996 in facilities other than emergency departments, the military, or prisons. The first follow-up consisted of 2466 respondents and the second had 2267 (65% of all surviving sampled subjects).

Main Outcome Measures Service utilization (<2 ambulatory visits, at least 1 emergency department visit that did not lead to hospitalization, at least 1 hospitalization) and medication utilization (receipt of antiretroviral therapy and prophylaxis against Pneumocystis carinii pneumonia).

Results Inadequate HIV care was commonly reported at the time of interviews conducted from early 1996 to early 1997 but declined to varying degrees by late 1997. Twenty-three percent of patients initially and 15% of patients subsequently had emergency department visits that did not lead to hospitalization, 30% initially and 26% subsequently of those who had CD4 cell counts below 0.20,×10⁹/L did not receive P carinii pneumonia prophylaxis, and 41% initially and 15% subsequently of those who had CD4 cell counts below 0.50×10⁹/L did not receive antiretroviral therapy (protease inhibitor or nonnucleoside reverse transcriptase inhibitor). Inferior patterns of care were seen for many of these measures in blacks and Latinos compared with whites, the uninsured and Medicaid-insured compared with the privately insured, women compared with men, and other risk and/or exposure groups compared with men who had sex with men even after CD4 cell count adjustment. With multivariate adjustment, many differences remained statistically significant. Even by early 1998, fewer blacks, women, and uninsured and Medicaid-insured persons had started taking antiretroviral medication (CD4 cell count adjusted P values <.001 to <.005).

Conclusions Access to care improved from 1996 to 1998 but remained suboptimal. Blacks, Latinos, women, the uninsured, and Medicaid-insured all had less desirable patterns of care. Strategies to ensure optimal care for patients with HIV requires identifying the causes of deficiency and addressing these important shortcomings in care.

The US health care system is going through many changes that raise concerns about adequacy of care, including new constraints on public programs¹ and changes in the insurance options available to the privately insured.^2-4 Ascertaining the impact of the emerging organization of services on those most likely to benefit from care is important but difficult. Neither national surveys of households or medical providers nor most studies of specific diseases provide sufficient information. The former studies do not include enough patients with specific conditions of interest to analyze issues of particular concern in the care of those conditions and do not collect enough disease-specific information.⁵ The latter studies usually cannot be broadly generalized, either because they are local or use some form of purposive sampling to assemble a nonrepresentative cohort. Consequently, evidence of disparities in care at 1 or a few institutions may challenge assumptions about adequacy of care to certain sectors of the population, but nationally representative data are needed to determine if broader attention is merited.

We conducted a comprehensive study of health care for a national probability sample of US adults with 1 important chronic disease, human immunodeficiency virus (HIV) infection.⁶ Adequate treatment for this disease is important because of the clear benefit of therapy and because HIV-infected persons are a paradigmatic example of patients who need complex treatment.⁷ Studies with limited or uncertain generalizability have raised concerns about the possibility of inequities in realized access or receipt of needed HIV care.^8-12 In this article, we examine variations in the care received by a national sample representative of the adult HIV-infected population receiving regular medical care in the 48 contiguous United States from early 1996 to early 1998.¹³

The HIV Cost and Services Utilization Study (HCSUS) used multistage national probability sampling to select the study cohort.^6,13 The reference population was persons at least 18 years old with known HIV infection who made at least 1 visit for regular care in the contiguous United States to other than a military, prison, or emergency department (ED) facility between January 5 and February 29, 1996, except for 1 city where sampling began and ended 2 months later.

In the first stage of sampling, we randomly selected 28 metropolitan statistical areas and 24 clusters of rural counties that together contained nearly 70% of all acquired immunodeficiency syndrome (AIDS) cases in the United States.^14,15 In the second stage, we randomly selected 58 institutional or individual physicians known to care for patients with HIV infection (known providers) in urban areas and 28 in rural areas, who had been identified by local physicians or public health officials. Using data from the American Medical Association Master File, we randomly sampled approximately 4000 physicians in relevant specialties among whom 87 physicians (other providers) in urban areas and 23 in rural areas had confirmed in a screening survey that they cared for eligible patients.

In the first stage, we set sampling rates proportionate to caseload based on data obtained from the Centers for Disease Control and Prevention and local health departments. In the second stage, sampling rates were proportionate to caseload reported by the providers. In the third stage, we set sampling rates to equalize probabilities within subgroups while increasing the overall sampling rate for women and members of private staff-model health maintenance organizations (HMOs).⁶ In the third stage of sampling, subjects were randomly selected with the appropriate predetermined probability from anonymous lists of those who received outpatient or inpatient care from participating providers during January and February 1996. We removed duplicate patient codes from the patient lists and adjusted for the fact that a patient might see more than 1 provider using multiplicity weights as described below.

After the replacement of a single urban provider with an equivalent one in the same community, we obtained agreement to participate from 100% of known urban providers and 79% in rural areas, 70% of other urban providers and 83% in rural areas, and 84% of selected patients. The baseline coverage rate, or the ratio of the population directly represented to the population that would have been represented if we had complete cooperation at all stages, was 68% for long-form interviews and 87% for long, short, and proxy interviews and forms completed by providers when no interview was possible (nonresponse forms). The short-form and proxy interviews were administered when subjects were too ill or otherwise unable to complete the full interview themselves. Nonresponse data included vital status, date of death if applicable, sex, age, race or ethnicity, insurance status, risk-exposure group, lowest and most recent CD4 cell counts, date of CD4 cell count, any AIDS-defining illness, and any inpatient stays in the last 6 months.

We performed 3 rounds of interviews: baseline, first follow-up, and second follow-up. Respondents who did not complete a long-form interview in the previous survey wave were not approached for the subsequent wave. Centrally trained personnel from the National Opinion Research Center, Chicago, Ill, used computer-assisted personal interviewing to conduct in-person interviews.¹⁶ We approached anonymously selected subjects for interview only after providers or their agents obtained permission. The institutional review board of RAND, Santa Monica, Calif, and, if available, a local review board reviewed all forms and materials.

In the analyses described below, our baseline sample consisted of 2864 long-form respondents interviewed between January 1996 and April 1997 (71% of all sampled subjects). The first follow-up sample consisted of 2466 respondents interviewed between December 1996 and July 1997 (69% of all surviving sampled subjects); the second follow-up sample consisted of 2267 respondents interviewed between August 1997 and January 1998 (65% of all surviving sampled subjects).

Using all available data, we constructed a baseline analytic weight for each respondent to adjust the sample to represent the entire reference population, not just the proportion directly represented by the sample. Each weight, which can be interpreted as the number of persons represented by that respondent, is the product of a sampling weight, which adjusts for differential sampling probabilities; a multiplicity weight, which adjusts for patients who could have entered the sample through visits to multiple providers; and a nonresponse weight, which adjusts for differences in rates of cooperation.^6,17,18 The sum of the baseline weights is an estimate of the size of the target population represented by HCSUS. The first follow-up weight is the product of the baseline analytic weight and an attrition weight, which accounts for the attrition of baseline respondents who were not successfully interviewed at first follow-up; the second follow-up weight takes into account attrition of first follow-up respondents not successfully interviewed at second follow-up. Respondents who died prior to a wave are not considered eligible for that wave.

The analyses presented herein are based primarily on 3 cycles of 90-minute long-form interviews. At each wave, we asked all subjects about use of medical services and HIV medications during the preceding interval (6 months at baseline or since the last interview). We also inquired about their clinical status including their latest and lowest CD4 cell counts. We asked respondents who did not recall the exact value of their CD4 cell counts to provide a range.

We imputed missing values for essential covariates using a standard "hot-deck" strategy¹⁹ with further details given by Duan et al.¹⁷ Briefly, for each variable being imputed, we classified all respondents into imputation classes based on observed values for other variables. Then, for each respondent missing a value for the variable being imputed, we randomly selected a donor value from those respondents not missing a value in the same imputation class. At baseline, we had to impute 4.9% of CD4 cell count values, less than 4% for income, less than 2% for insurance status, 1.4% of values for ambulatory visits, and less than 0.5% of missing values in other essential variables. The follow-up data had comparable amounts of missing values, which were imputed using the same approach as at baseline.

Information on the use of protease inhibitors (PIs) and nonnucleoside reverse transcriptase inhibitors (NNRTIs) was obtained from baseline, first follow-up, and second follow-up interviews. To obtain comprehensive information on their use through January 1998, an additional, brief interview was completed with most subjects in early 1998. We asked respondents for the date on which they began use. Thus, for any particular month, we could determine the proportion who had used the medications by that month. We imputed the date of probable first use for the 10% of respondents who had not used these medications at baseline and did not complete a follow-up interview.¹⁷

Our model of utilization or access to needed HIV care assumes that a need for care exists and that a less-than-optimal pattern of care is reflected in infrequent office visits among persons known to have HIV disease, in the use of EDs when there is no need for hospitalization, in increased hospitalizations, and in failure to take appropriate anti-HIV medications.

We constructed 3 measures of service and 3 measures of pharmaceutical utilization to gain insight into the adequacy of the access to needed care. The measures of service utilization included having, in the 6 months prior to the interview: (1) fewer than 2 ambulatory visits, (2) at least 1 ED visit that did not lead to a hospitalization, and (3) at least 1 hospitalization. Two visits were selected as an initial threshold for adequate care because 1 visit every 3 months is about the upper limit of an acceptable interval for monitoring disease events and response to treatment in most patients. While some hospitalizations are probably inevitable as HIV disease progresses, higher hospitalization rates clearly result from failure to receive indicated outpatient therapy. Ambulatory care that is sufficiently frequent, high in continuity, and adequate in quality should prevent some complications and decrease the rate of hospitalization, even in the advanced stages of the disease. Persons with good continuity of ambulatory care should, likewise, be better able to avoid using the ED in nonemergent situations.

The measures of medication utilization included having taken: (1) prophylaxis against Pneumocystis carinii pneumonia (PCP) in the 6 months preceding the interview for those subjects whose lowest CD4 cell count was less than 0.20 × 10⁹/L, the accepted threshold for this treatment; (2) at least 1 antiretroviral medication at any time prior to the baseline interview or second follow-up interview, among subjects with a CD4 cell count of less than 0.50 × 10⁹/L (Table 1, Table 2, and Table 3); and (3) either PI or NNRTI therapy by a specified date (December 1996 for the initial analyses presented in Table 1 and Table 3, and January 1998 for the analyses presented in Table 2), also among subjects with CD4 cell counts of less than 0.50 × 10⁹/L. We asked about lifetime use of medications in these last 2 categories because of the possibility that subjects might have discontinued therapy prior to the time covered by the interview. We conducted additional analyses of these 2 variables that included all subjects irrespective of CD4 cell count. We constructed the variable indicating use of newer antiretroviral therapies from baseline, first follow-up, and second follow-up interview data as previously noted.¹⁷ To gain insight into the stability of differences in the use of these medications across population subgroups, we also examined patterns of dissemination of these therapies from January 1996 to January 1998, thereby encompassing a period during which there was growing awareness and wide publicity of the effectiveness of these treatments.²⁰

For the baseline sample and using the baseline long-form data, we estimated the weighted proportions and odds ratios for the utilization measures across categories defined by lowest CD4 cell count (Table 1). For each measure of utilization, we then adjusted for possible confounding via weighted logistic regression. We adjusted all SEs and statistical tests to compensate for the complex sampling design and differential weighting using linearization methods available in the SUDAAN (Research Triangle Institute, Research Triangle Park, NC) and Stata software packages (Stata Corp, College Station, Tex).²¹

We estimated 2 main effects models with different covariates to determine which covariates were most important in explaining differences among subgroups in the access measures. Model A includes CD4 cell count; model B adds age, sex, race or ethnicity, exposure group, insurance coverage, education, and region to model A. (Results are shown as statistical significance in Table 1 and Table 2 both with and without adjustment for CD4 cell count [model A]. Table 3 presents the results of the model B analyses.)

To assess the sensitivity of findings to model specifications, we also conducted a number of exploratory analyses. In addition to the main analysis of 2 or more ambulatory visits in the preceding 6 months, we examined variations in receipt of 3 or more and 4 or more visits. The pattern of variations in disparities were very similar with these other measures. We stratified the highly associated sex and risk variables in a single, 7-category variable for alternative analyses (male injection drug users, female injection drug users, men having sex with men, male heterosexuals, female heterosexuals, male other, or female other, with the hierarchy among women being injection drug users, heterosexuals, and others), again with little or no impact on the analyses. We examined variation in the rate of hospitalization within CD4 cell count strata because of the strong effect of disease progression. Finally, we examined the patterns of PI and NNRTI use by additional specific dates during 1996. (Results of the regressions discussed in the text that are not presented in the tables are available from the authors on request.)

To assess change over time, we estimated the proportions for the utilization measures across subgroups at the second follow-up. To separate the effects of time vs attrition, we focused on the second follow-up sample only. As a result, our reference population for this analysis is the baseline reference population who survived until the second follow-up. We estimated this population's proportions at second follow-up, and conducted tests of these proportions adjusted for CD4 cell count. We also analyzed the percentage change since baseline in utilization proportions for this population, and we discuss some of these findings in the text.

The HCSUS data are weighted to represent 231,400 adults who received care in the 48 contiguous United States outside of EDs, prisons, or the active military during a 2-month period in early 1996.¹³ While a majority of these persons represented by the HCSUS met the standards represented by our 6 utilization measures during 1996 and 29% were in compliance with all of them, many did not. Overall, 15% of the represented population had fewer than 2 ambulatory visits in the previous 6 months, 23% had at least 1 ED visit that did not lead directly to hospitalization, and 19% were hospitalized at least once during the same period (Table 1). Forty-one percent of those with CD4 cell counts of less than 0.50 × 10⁹/L had not received PI or NNRTI therapy by the end of 1996, and 30% of persons with CD4 cell counts of less than 0.20 × 10⁹/L had not taken medications to prevent PCP in the 6 months prior to their interview, but only 8% had never taken any antiretroviral therapy by the time of their initial interview.

Use of needed care differed by CD4 cell count, age, sex, race or ethnicity, HIV risk–exposure group, insurance, socioeconomic status, and region (Table 1). The CD4 cell count was associated with care for all 6 measures tested, with a lower CD4 cell count being associated with more use of medications, hospitalization, and ambulatory care. The relationship between age and use of needed care was mixed; HIV-infected persons who were at least 50 years old were less likely to use EDs but were more likely not to take indicated PCP prophylaxis. Women were more likely than men to use the ED and to be hospitalized and were more likely not to have taken indicated PCP prophylaxis or to have started PI or NNRTI therapy by the end of 1996. Race or ethnicity was associated with the adequacy of care for 5 of 6 measures. Considering individual comparisons, care received by blacks and Latinos was significantly less optimal than that received by whites on 6 and 4 of 6 measures, respectively.

Exposure route was associated with care received on 4 of 6 measures. Those having either injection drug use or heterosexual contact as their risk factor had less favorable patterns of use than men who had sex with men (Table 1). The care received by both uninsured patients and those with Medicaid insurance was unfavorable compared with that received by the privately insured on 5 of 6 access measures. Medicare beneficiaries, most of whom had also had Medicaid, had high rates of ED and hospital use, similar to subjects who had Medicaid only. However, persons with Medicare coverage were more likely to have started PI or NNRTI therapy and were less likely to be deficient in PCP prophylaxis compared with those with Medicaid coverage alone. Those enrolled in HMOs had a pattern of care very similar to others with private insurance, except that they were more likely to have received appropriate PCP prophylaxis. Those with the most education had a more desirable pattern of care than others for 4 of 6 measures, including PI and NNRTI therapy and PCP prophylaxis. The pattern of care varied by income in a similar way, with the lowest income group having the least favorable rates for most measures (data not shown). Finally, care varied by region for 4 of 6 measures. Patients in the Northeast were more likely to use an ED, be hospitalized, or miss indicated PCP prophylaxis, while those in the South were more likely to have not used a PI or NNRTI.

We reassessed compliance with the standards during the latter half of 1997 and early 1998, after a median of 15.1 months of follow-up (Table 2). At the second follow-up interview, 16% of patients represented had fewer than 2 visits per 6 months, 16% had an ED visit without hospitalization within the prior 6 months, and 14% were hospitalized at least once during the same period. In addition, 15% had not received indicated PI or NNRTI therapy by January 1998, 3% had never taken any antiretroviral therapy, and 26% had not taken indicated PCP prophylaxis. Overall, the percentage who were in compliance with all 6 indicators had increased from 29% to 47% and the percentage who were noncompliant with 2 or more measures had fallen from 34% to 17%.

Relative to baseline values, there was virtually no change in the proportion meeting the standard of 2 outpatient visits in 6 months (Table 2). Interestingly, there was an increase of 26% in the proportion not having at least 3 ambulatory visits, perhaps reflecting improving health in the population (data not shown). For the other 5 indicators, there were declines in the proportions of subjects not meeting the standards in the indicators. Of note, the improvement in proportion of persons with CD4 cell counts not receiving PCP prophylaxis (from 30%-26%) was much more modest than the decline in the proportion not having taken PI and NNRTI therapy (41%-15%). At each point in time, almost one quarter of persons who had initiated PI and NNRTI therapy did not receive indicated PCP prophylaxis. Significant predictors of improvement were CD4 cell count and race for 3 indicators each, and exposure and region for 1 indicator each (data not shown). Despite these improvements, there continued to be significant differences in compliance by CD4 cell count for 5 of 6 measures, by exposure group, insurance type, and region for 4 each, and by sex and race or ethnicity for 3 each (Table 2).

Although the unadjusted values given accurately describe variations in the care actually received by HIV-infected US adults, we constructed multiple logistic models to test the independent fixed effects of the factors listed above. We evaluated the 6 measures with respect to 7 demographic characteristics, for a total of 42 assessments. The CD4 cell count adjustment had little impact: in the initial period, of 30 unadjusted significant overall effects within the 7 demographic variable categories (sex, age, race or ethnicity, insurance status, exposure group, education, and region) in our 6 measures (out of a possible 42), 28 effects remained significant with CD4 cell count adjustment (Table 1). Twenty-two of 42 unadjusted effects were significant in the second period examined, and 20 of these remained significant with CD4 cell count adjustment (Table 2). Although some overall effects within demographic variable categories lost significance, in models that included all 7 categories of variables listed above plus CD4 cell count, 15 of 42 possible effects (other than by CD4 cell count) remained significant in the initial period (Table 3). Eleven of 18 possible effects were significant in the multivariate models, for insurance, race or ethnicity, and age (including 5 of the 6 access indicators for insurance, 3 of the 6 indicators for race or ethnicity, and 3 of the 6 indicators for age), compared with 14 significant effects for these 3 variables with CD4 cell count adjustment alone. On the other hand, only 4 of 24 possible effects were significant in the multivariate models for sex, education, exposure group, and region compared with 15 effects with CD4 cell count adjustment alone.

Even with the differential improvement rates, significant differences in the presence of deficiencies persisted for 20 of 48 possible predictors at follow-up in the full multiple regression model. These included race or ethnicity for 1 of 6 indicators; age for 2; sex, insurance, and exposure group for 3 each; region for 4; and CD4 cell count for 5 (data not shown). While there was less evidence of an independent overall effect of race or ethnicity at follow-up, important CD4 cell count–adjusted differences remained between whites and blacks in PI and NNRTI use and between whites and Latinos in the pattern of ambulatory visits (Table 2).

To explore further the differences in access to a key component of care over time, we examined the proportion of the HIV-infected adult population that had ever used PI or NNRTI over a 2-year period. Overall, use of these drugs grew very rapidly throughout this period among persons with a CD4 cell count lower than 0.50 × 10⁹/L, increasing from 17% in January 1996 to 59% in January 1997, and to 85% in January 1998 (Figure 1). Generally, use varied inversely by CD4 cell count (Figure 1). In addition, there were clear and substantial differences in cumulative use of these medications by sex, race, and insurance with women, minorities, and the underinsured being less likely to have received these drugs (Figure 1 and Figure 2). The 24% difference in initial use of these medications in late 1996 (P<.001, Table 1) between blacks and whites had declined to 8% in early 1998, but remained significant, even with CD4 cell count adjustment (P=.016, Table 2). Similarly, the difference in treatment rates between those uninsured and those privately insured declined from 26% (P <.001) to 12% (P=.016). The rate for Medicaid beneficiaries was within 2% of the rate for the uninsured at the later date and significantly different from that of the privately insured (P<.001); both remained significantly different from those who were privately insured, even with multivariate adjustment (P=.026, Medicaid; P=.048, no insurance). Finally, the difference between men and women (12% in late 1996) declined very little and was 9% in early 1998 (P=.003, P<.001, respectively, with CD4 cell count adjustment).

Even though the differences between certain groups abated over time, the lag in dissemination to traditionally disadvantaged groups meant that these groups waited much longer before having an opportunity to try these agents. For example, on average men waited 11.2 months from the beginning of 1996 while women waited 13.5 months (P≤.001); blacks waited 13.5 months while whites waited 10.6 (P<.001); uninsured persons waited 13.9 months while Medicaid beneficiaries waited 12.4; and those with private fee-for-service insurance waited 9.4 (P≤.001; P<.002, respectively). There were other contrasts of interest: southerners waited 13.4 months compared with 10.6 to 10.8 months elsewhere (P=.03). Overall, in a multiple regression model, CD4 cell count, sex, race, education, insurance, and risk or exposure were the independent significant predictors of delay in time to first use of a PI or NNRTI (P<.05).

This first detailed examination of medical care for a specific chronic condition in a nationally representative sample of Americans has yielded a mixed picture. Realized access, or use of needed care, is good for many HIV-infected patients. At the time of the first study interview, about 80% met each of our individual standards for continuity of outpatient services. About 90% of those with indications had received antiretroviral therapy, and nearly 60% had received PIs within a year of these agents' approval. However, 30% of those with indications denied taking medications to treat or prevent PCP. Moreover, less than 30% of patients were in compliance on all of the measures for which they were eligible. Finally, the deficiencies in HIV care were unevenly distributed across the population.

Disparities in HIV care cut across important population categories, with disadvantaged populations having the least favorable patterns of use. Compared with whites, use was less optimal for blacks and Latinos for 5 and 3 measures, respectively, of the 6 measures studied, in models that were statistically adjusted for CD4 cell count, age, sex, and exposure group. Addition of insurance to the models attenuated these effects for many of the measures, suggesting that insurance contributes importantly to racial and ethnic differences in receipt of care. Women and persons whose exposure group was not men having sex with men fared worse on most measures; however, most of these effects were mitigated or eliminated by covariate adjustment, suggesting that at least some of the less desirable pattern of care experienced by women at that time was related to insurance coverage and other demographic characteristics, notably race or ethnicity.

Those who lacked health insurance fared worse in terms of most measures. Surprisingly, those with Medicaid also were deficient in many of the categorical measures relative to private insurance. We found no evidence of an unfavorable pattern of care among private HMO enrollees. Along with Medicare beneficiaries (most of whom must be followed by the health care system long enough to establish 2 years of disability), they were most likely to receive indicated prophylaxis against PCP.

The results of the interview in late 1997 and early 1998 indicate important changes in the pattern of care. The proportion of persons taking PI or NNRTI therapy for the first time continued to rise until near the end of that time, by which time 85% of all persons with CD4 cell counts of less than 0.50 × 10⁹/L had tried these agents. There continued to be differences in use across demographic groups, but many were of lesser magnitude than those groups that had been evident a year earlier. For example, the difference between blacks and whites had decreased from 24% to 8% and the disparities by insurance category declined by about half. On the other hand, there was no further narrowing of the gap between men and women.

At the same time, some other access measures had improved. Hospitalization rates and ED use both declined overall, and the gaps declined among subgroups. By early 1998, almost all subjects had taken an antiretroviral medicine at some time if their CD4 cell counts were less than 0.50 × 10⁹/L. On the other hand, the proportion of persons taking PCP prophylaxis had improved only very modestly, and 26% of those with an indication reported not receiving any in the preceding 6 months.

Thus, there was a substantial improvement in access to care in 1997 and 1998 (with the proportion not deficient in any access measures for which they were eligible increasing from 29% to 46%), but some important deficiencies remained, and disparities by race, insurance status, and sex persisted, although their magnitude had lessened. Overall, the HCSUS data show that disenfranchised groups in the United States are less likely to receive medications for HIV that are potentially lifesaving, yet being in an advantaged group does not guarantee access to ongoing effective care.

As noted by Bozzette et al,¹³ the study represents only persons receiving ongoing care and used a sampling strategy that is likely to underrepresent those with very poor access, those who are less compliant, and those who are relatively healthy; however, it does account for the 32% of nonrespondents to the full interview with proxy and nonresponse data. The study uses self-report data. Patient reports are subject to recall bias. To minimize this, subjects were shown photographs of antiretroviral medications when questioned about their use. Patient reports of CD4 cell count have been shown to be reliable and valid in relation to critical thresholds at which the therapies queried in this article are recommended (<0.20, or <0.50 × 10⁹/L).²² Data on use of newer medications represent the practices through early 1998 for persons who received ongoing care for HIV disease in early 1996 but not for persons who may have entered care after the eligibility period for this study. It is also possible that intergroup differences in use of services and medications reflect, in part, patient choice and/or nonadherence rather than social or system factors. The findings relating to regional differences in hospitalization rates may reflect, in part, regional differences in practice style; they are consistent with national data on overall hospitalization rates.²³ Finally, the 6 measures of access are not measuring independent phenomena, and there is some correlation among them.

Nonetheless, several conclusions can be drawn from the patterns of deficiencies in care uncovered by this study. First, being uninsured and having HIV is a serious problem and places the individual at risk of not obtaining adequate care. Second, having Medicaid does not result in as appropriate of a pattern of use for needed treatments as does having private insurance. Third, blacks and Latinos under care for HIV also have less favorable patterns of use of needed services than do whites, patterns that cannot be explained by other characteristics of these patients. Fourth, use varies by certain other characteristics, including sex, exposure group, income, and education. While these disparities diminished over time, they did not disappear. In the initial period, patient insurance status, race or ethnicity, and age were the most important explanatory variables; in the second period studied, sex, exposure group, insurance, and region were each an independent predictor of access for at least 3 of the 6 access variables. Strategies to ensure optimal care for persons infected with HIV in the US population need to take into account these patterns of deficiency and the fact that no one characteristic of disenfranchised populations is responsible for them.

What are the implications of these findings for the future of care for HIV disease? Up-to-date treatment for those infected with HIV offers the prospect of long-term survival, but such therapy is likely to continue to evolve rapidly. Patients need to be in continuous relationships with providers who can monitor their clinical status and their compliance with therapy and who can assess their need for modifications in their therapy if they become resistant to their regimens. As newer therapies become available, members of the medical community must be concerned that such treatments may not disseminate therapy at similar rates throughout the populations who need them, as the HCSUS data show occurred with the PI and NNRTI medications. A lag in obtaining newer therapies may well place patients at risk of death and other serious complications.

In the last 2 years, HIV has become a highly treatable disease.^7,20 For much longer, it has been clear that good medical care can prolong life by preventing complications and attacking the infecting organism. While access to HIV care is good for many adults and improving for others, it is still not nearly optimal. These data show a pattern of less access to high-quality HIV care for disadvantaged groups across the United States, and these data suggest the need for comprehensive efforts to improve patient care. Such efforts should include steps to ensure that lags in access to newer HIV treatments are not recapitulated with each improvement in treatment.

There have been major advances in recent years in the care of many other chronic diseases, including heart disease, diabetes mellitus, and some cancers. Representative national data regarding care for these conditions would help us to determine the extent to which variations in care are contributing to inequities in the health care of the US population. Such populations need to be monitored longitudinally because of the likelihood that the patterns of any disparities in care may well change over time as they have in the case of HIV disease.²⁴