PMCCPMCCPMCC

Search tips
Search criteria 

Advanced

 
Logo of jgimedspringer.comThis journalToc AlertsSubmit OnlineOpen Choice
 
J Gen Intern Med. 2003 April; 18(4): 233–241.
PMCID: PMC1494839

Physician Specialization and Antiretroviral Therapy for HIV

Adoption and Use in a National Probability Sample of Persons Infected With HIV
Bruce E Landon, MD, MBA,1,2 Ira B Wilson, MD, MSc,3 Susan E Cohn, MD, MPH,8 Carl J Fichtenbaum, MD,9 Mitchell D Wong, MD, PhD,7 Neil S Wenger, MD,7 Samuel A Bozzette, MD, PhD,4,5,6 Martin F Shapiro, MD, PhD,4,7 and Paul D Cleary, PhD2

Abstract

BACKGROUND

Since the introduction of the first protease inhibitor in January 1996, there has been a dramatic change in the treatment of persons infected with HIV. The changing nature of HIV care has important implications for the types of physicians that can best care for patients with HIV infection.

OBJECTIVE

To assess the association of specialty training and experience in the care of HIV disease with the adoption and use of highly active antiretroviral (ARV) therapy (HAART).

DESIGN

Observational cohort study of patients under care for HIV infection and their physicians.

PATIENTS AND SETTING

This analysis used data collected from a national probability sample of noninstitutionalized persons with HIV infection participating in the HIV Costs and Service Utilization Study and their primary physicians. We analyzed 1,820 patients being cared for by 374 physicians.

MEASUREMENTS

Rates of HAART use at 12 months and 18 months after the approval of the first protease inhibitor.

RESULTS

Forty percent of the physicians were formally trained in infectious diseases (ID), 38% were general medicine physicians with self-reported expertise in the care of HIV, and 22% were general medicine physicians without self-reported expertise in the care of HIV. The majority of physicians (69%) reported a current HIV caseload of 50 patients or more. In multivariable models controlling for patient characteristics, there were no differences between generalist experts and ID physicians in rates of HAART use in December 1996. When compared to ID physicians, however, patients being treated by non-expert general medicine physicians were less likely to be on HAART (odds ratio [OR], 0.32; 95% confidence interval [95% CI], 0.17 to 0.61). Patients being treated by low-volume physicians were also much less likely to be on HAART therapy than those treated by high-volume physicians (OR, 0.26; 95% CI, 0.14 to 0.48). These findings were attenuated by June 1997, suggesting that over time, the broader physician community successfully adopted HAART therapy. This finding is consistent with prior research on the diffusion of innovations.

CONCLUSIONS

Similar proportions of patients treated by expert generalists and ID specialists were on appropriate HAART therapy by December 1996 and July 1997. Patients treated by non-expert generalists, most of whom were the lowest-volume physicians, were much less likely to be on appropriate ARV therapy in the earlier time period. Our findings demonstrate that expert generalists who develop specialized expertise are able to provide care of quality comparable to that of specialists.

Keywords: HIV, AIDS, specialism, physicians' practice patterns, HAART therapy

Since the introduction of protease inhibitors in January 1996, there has been a dramatic change in the treatment of persons infected with HIV.13 Numerous clinical trials have demonstrated the effectiveness of multidrug regimens for treatment of HIV, and national guidelines for the treatment of HIV infection released since 1996 support their use.25 Since that time, AIDS morbidity and mortality in the United States have fallen dramatically.6,7 Potent combination chemotherapy for HIV requires detailed knowledge of multiple side effects and drug interactions and requires frequent monitoring and adjustments to achieve maximal benefit.3 Drug therapy for HIV has become the single largest component of medical costs for patients with HIV.8

Changes in HIV care have important implications for the types of physicians that can best care for patients with HIV infection. Even before the use of protease inhibitors, data from a sample comprised mostly of family practitioners suggested that at least some experience in treating HIV was associated with improved survival.9 A more recent study using case vignettes found that both training in infectious diseases (ID) and HIV experience were associated with the use of appropriate antiretroviral (ARV) regimens and that, controlling for experience, ID physicians were more likely to choose appropriate ARV regimens. These findings, however, were attenuated after excluding physicians who do not manage HIV infection by themselves.10 Another recent study demonstrated that providers with higher HIV caseloads were earlier adopters of complex antiretroviral regimens.11 Current guidelines recommend that these complex antiretroviral medications be prescribed only by physicians with “extensive experience.”3 No study in the current era of highly active antiretroviral therapy (HAART), however, has examined actual prescribing behavior and its relationship to provider characteristics such as experience and formal training.

The HIV Costs and Services Utilization Study (HCSUS), a study of a nationally representative sample of persons receiving care for HIV, provides a unique opportunity to study the introduction and adoption of a new therapy for the treatment of a chronic medical condition by physicians, because HCSUS began collecting data just as protease inhibitors were first licensed.1,12 In this study, we examine the relationship between physician specialty and subspecialty training, experience with HIV as defined by their volume of HIV patients, self-rated HIV expertise, and objective HIV knowledge and adoption and use of HAART therapy in this large nationally representative sample of persons with HIV infection.

METHODS

Participants and Data Collection

Patient Sample

The HIV Cost and Services Utilization Study used a multistage probability sample of persons in care for HIV in the contiguous United States.1214 In the first stage of sampling, we randomly selected metropolitan statistical areas and clusters of rural counties that, in the aggregate, contained nearly 70 percent of all U.S. cases of AIDS.15,16 In the second stage, we randomly selected providers (typically institutions) who had been identified by local physicians or public health officials as caring for patients with HIV infection, as well as other providers who had reported caring for patients with HIV infection during a screening survey of approximately 4,000 randomly selected physicians. In the third stage, patients seen by sampled providers were randomly sampled during January and February 1996. The study enrolled 2,864 patients and these patients completed a baseline interview and 2 subsequent interviews about aspects of their care, including costs, access to care, and antiretroviral use.1,2,17

Physician Sample

As part of the second patient interview (conducted in the first half of 1997) subjects were asked to identify all of their providers, including the physician most important for their HIV care. Eighty-five percent of patients were able to identify such a physician. In cases in which the patient was unable to identify a principal HIV physician, we created an algorithm to match the patient with the physician most likely to be responsible for his/her care. First, we selected the physician the patient said he or she saw most recently for HIV care and if there was no such physician, then we selected a non-HIV primary care physician that they identified. We repeated this process using the baseline survey if no physician had been identified in the follow-up survey. We were able to identify a physician for 2,642 patients (92% of the sample).18

Physician Data Collection

We confirmed physician mailing addresses and names for 551/692 identified physicians (80%) by telephoning their practice site or local contacts, checking the AMA website, and conducting Internet searches to obtain addresses.18,19 We sent these 551 physicians a survey with a $25 check as a token of appreciation for study participation. Nonrespondents received 3 follow-up calls and were offered the chance to complete the survey by phone. The surveys were administered between September 1998 and July 1999.

Of the 551 clinicians contacted, 412 (75%) completed the survey (60% of the 692 identified clinicians). Thirty-three of these were nurses, nurse practitioners, or physician's assistants (including 1 with incomplete specialty data) whom 76 HCSUS participants had incorrectly identified as physicians; we eliminated these. In addition, the patients that identified 5 of the physicians were subsequently deemed to be ineligible because they were cared for at multiple sampling sites, thus leaving a total of 374 useable physician responses.

Analytic Variables

Outcome Variables

Our working definition of HAART use, which is based on the Department of Health and Human Services/Henry J. Kaiser Foundation Guidelines for the Use of Antiretroviral Agents in HIV-infected Adults and Adolescents was “taking certain combinations of nucleoside reverse transcriptase inhibitors (e.g., zidovudine and lamuvidine) plus certain protease inhibitors ([PIs], e.g., nelfinavir), combinations of PIs (e.g., ritonavir and saquinavir), or the combination of a PI plus a non-nucleoside reverse transcriptase inhibitor (e.g., nevirapine).”20 Each patient was classified as taking HAART if he/she reported taking any 1 of the HAART combinations. According to guidelines published close to the time of the baseline interview,2,21 99% of the sample were eligible for HAART because they had either CD4 <500 cells/mm3 or HIV RNA >10,000 copies per mL, or they had symptomatic HIV or AIDS. While these guidelines have subsequently been modified, that should not affect our findings related to the primary questions under study.

We focused on the use of HAART by December 1996 and June 1997 to capture the diffusion of HAART therapy after its introduction in December 1995. In addition, this time period corresponds to the time when the patients identified their physicians. The HAART by December 1996 variable was constructed from both the baseline and follow-up surveys, while the HAART by June 1997 also included data from the second follow-up survey. In each round of interviews, patients were asked to report whether they had taken specific antiretroviral medications and at what point they started taking them. These data were used to create a dichotomous variable indicating HAART use as described in Anderson et al.22

Predictor Variables

Data on physician specialization

For these analyses, we used selected physician characteristics: physician training and specialization, experience treating HIV, self-rated expertise, and knowledge of HIV practices.18 Physicians were classified as being trained in either infectious diseases or general medicine/other for the purposes of this analysis. The survey also asked about the physicians' current and past HIV practice experience, including the number of patients currently cared for (categorized based on our previous analyses as low [0 to 19], medium [20 to 49], and high [>49]), the cumulative number of patients cared for in the previous 5 years, and the proportions with various risk behaviors for acquisition of HIV in their patient population. We created variables indicating whether the physician said that 50% or more of his or her patients had intravenous drug use, male-to-male sex, or male-to-female sex as a risk behavior for HIV acquisition. The physicians were also asked whether they considered themselves “specialists” (hereafter referred to as “experts”) in the treatment of HIV.

Only 6 of 152 (4%) of ID physicians in the sample did not consider themselves HIV experts. To examine the joint effects of training and expertise in multivariable models, we therefore created a composite training/expertise variable with 3 levels: ID physicians, generalist experts, and generalist non-experts. Because current and cumulative caseloads were highly correlated (r = .58), we used only current caseload in these analyses.

Physician knowledge

The physician survey also included a series of questions that asked about the physicians' knowledge of current HIV treatment practices, including questions related to prophylaxis against opportunistic infections and screening (5 questions), and viral load testing and antiretroviral therapy (6 questions). The number of correct responses out of 11 were summed to form a knowledge score. “Don't know” responses were classified as incorrect. The Cronbach's α for the resulting scale was 0.64.18 Because of the skewed distribution of the score, we also created a dichotomous variable (“high” HIV knowledge) indicating whether the physician was correct on 80% or more of these questions.

Patient control variables

Additional information relating to sociodemographic and clinical characteristics, including age, gender, education, income, insurance status, mode of acquisition of HIV, and lowest ever CD4 count were included as control variables in all analyses.1,23 We imputed missing values for essential patient control variables using standard “hot deck” imputation for less than 5% of CD4+ lymphocyte counts, less than 3% of insurance and income values, and less than 0.5% of other missing values.13,17 Finally, because patient attitudes are related to the use of antiretroviral medications, patients who do not want to take ARVs might be more likely to choose non-expert generalist physicians. We therefore included a question from the baseline survey that measured patients' attitudes toward antiretroviral therapy. Patients were asked if, after considering the benefits and risks of ARVs, they thought that these medications were definitely or probably worth taking, probably not worth taking, or definitely not worth taking. For multivariable models, these responses were dichotomized to definitely or probably worth taking versus not worth taking. For 39 respondents missing this information from the baseline survey, we used their response on the follow-up survey.

Analyses

Missing Data and Weighting

We compared the demographic characteristics of patients who could not identify a physician and of patients whose physician did not respond to the survey with those of patients whose physician did respond to the survey. All analyses are weighted to reflect the national noninstitutionalized population under care for HIV infection.13 In addition, we also applied a weight to account for physician nonresponse. All analyses were performed using SUDAAN software version 8 (Research Triangle Institute, Research Triangle Park, NC) to account for the effects of the complex sample design and clustering of patients within physicians.

Bivariate and Multivariate Relationships

We examined bivariate relationships between ID training, HIV caseload, self-assessed expertise, and the composite training/expertise variable and the use of HAART therapy with χ2 tests or t tests, as appropriate. We estimated a series of multivariable logistic regression models for the main outcomes of interest. We first constructed a baseline model that controlled for patient-level effects.22 We then estimated separate models that examined the individual effects of specialty training, the joint composite training/expertise variable, and current HIV caseload in models that only controlled for patient-level effects. The final model examined the joint effects of these variables together as well as HIV knowledge.

Sensitivity Analysis

A 2-part propensity score analysis was also conducted to test for the possibility that some of the differences among physicians were due to the fact that certain types of patients were more likely to be treated by specialized or experienced physcians.24 For the first model, the probability of seeing a general physician (as opposed to an ID physician) was modeled, and for the second model, the probability of seeing a general medicine expert (as opposed to a non-expert) was modeled. Both models used the patient-level predictors from the HCSUS surveys, primarily related to sociodemographic factors, CD4 count, and risk factors. Weights were developed to create balanced samples, and the regressions were repeated using a combination of these weights and the sampling weights mentioned above. Because these analyses were not substantively different from those that did not incorporate these adjustments, we present the models that did not incorporate these weights.

Role of the Funding Sources

The funding sources had no role in the collection, analysis, or interpretation of the data or in the decision to submit the manuscript for publication.

RESULTS

A total of 374 physician responses were linked to 1,820 (64%) of the 2,864 persons enrolled in HCSUS. Of the 1,044 patients that were not linked to a physician, 222 did not identify any specific physician as responsible for their care in either the baseline or follow-up surveys, and 76 identified a nurse, nurse practitioner or physician's assistant. Therefore, the number of patients who were either linked to a physician or did not identify any physician is 2,042 (1,820 plus 222), accounting for 71% of the HCSUS cohort.

Physician Characteristics

Forty percent of the physicians were formally trained in infectious diseases (representing physicians caring for 46.5% of HIV cases under care), 38% were general medicine physicians with expertise in the care of HIV (representing 44.5% of patients under care), and 22.2% were general medicine physicians without expertise in the care of HIV (representing 9.1% of patients) (Table 1). The majority of physicians (69%) reported a current HIV caseload of 50 patients or more. No information was available on nonresponding physicians.

Table 1
Provider Characteristics

Patient Characteristics

Table 2 presents a description of the patient sample by the main types of physicians. Seventy-eight percent of the patients were men and approximately 32% were African American. Twenty-three percent had intravenous drug use as their primary risk factor for HIV acquisition, while 49% said that they were males who had sex with males. A comparison of the demographic characteristics of patients who were included and excluded from the study sample revealed no significant difference (data not shown).

Table 2
Patient Characteristics by Type of Physician*

Patients being seen by general medicine experts were more likely to have no insurance (21.6% for general medicine experts as compared to 10.6% for non-experts and 7.7% for ID physicians). In addition, patients of general medicine experts had lower income (22.7% with annual income <$5,000 versus 11.5% and 17.7% for general medicine non-experts and ID physicians, respectively) and less-advanced HIV disease (22.5% with a lowest CD4 count <50 cells/mm3 as compared to 27.6% and 28.5% of patients being cared for by general medicine non-experts and ID physicians, respectively). Finally, a higher proportion of patients being treated by general medicine non-experts felt that ARVs were probably or definitely not worth it (31.5% versus 18% for general medicine experts and 15.9% for ID physicians; P < .01 for both comparisons).

Use of and Adoption of HAART Therapy

Bivariate Relationships

Overall, 39.5% of the population were taking HAART therapy by December 1996. This proportion increased to 65.8% by June 1997 (Table 3). Patients being cared for by ID physicians were more likely to be on HAART therapy than those being cared for by non-expert general medicine physicians at both time periods (44.1% vs 32.9% by 12/96 and 70.5% vs 59.7% by 7/97 P < .05). The rate for patients being cared for by general medicine experts fell between the rates observed for ID physicians and general medicine non-experts, although the rates were closer to those of general medicine non-experts.

Table 3
Bivariate Relationship Between Predictors and Outcome Variables

Multivariable Relationships

In multivariable models that examined HAART use by December 1996 after controlling for patient characteristics, patients being treated by general medicine physicians (Fig. 1, model 1) were far less likely than patients of ID physicians to have been on HAART (odds ratio [OR], 0.69; 95% confidence interval [95% CI], 0.52 to 0.95). After classifying generalist physicians according to their self-rated expertise, however, only patients being treated by non-expert general medicine physicians were less likely to be on HAART (OR, 0.32; 95% CI, 0.17 to 0.61) when compared to ID physicians (Fig. 1, model 2). There was no significant difference between general medicine experts and ID physicians (OR, 0.80; 95% CI, 0.59 to 1.10). In models that examine physician HIV caseload, patients being treated by low-volume physicians were also much less likely to be on HAART therapy (OR, 0.26; 95% CI, 0.14 to 0.48) than those treated by high-volume physicians (Fig. 1, model 3). In models including both the training/expertise variables and patient caseload, because of the close relationship between these 2 variables, only patients with low-volume physicians with between 0 and19 current HIV-infected patients had lower odds of being on HAART by December 1996 (data not shown). Although HIV knowledge is associated with both specialty/expertise and caseload, HIV knowledge was not an important predictor after controlling for the other physician characteristics.18

FIGURE 1
Multivariable predictors of HAART use. Results are presented as odds ratios with their associated 95% confidence intervals. Base model controls for the following patient-level variables: age, gender, education, income, insurance status, mode of acquisition ...

Predictors of HAART use by June 1997 were consistent with those seen in the December 1996 models, although the effects of specialty training and caseload were attenuated and no longer statistically significant. These findings suggest that over time, this innovation diffused successfully throughout the physician community.

DISCUSSION

One year after the introduction of the first protease inhibitor, 30% to 40% of eligible patients with HIV infection were taking HAART therapy. This number had risen to 60% to 70% just 6 months later. In this study, specialty training and expertise of the physicians caring for these patients were important predictors of early adoption and use of HAART. At both 1 year and 11/2 years after the approval of the first protease inhibitor, we found that patients of non-expert generalist physicians had lower rates of HAART therapy than either patients of physicians with specialty training in ID or patients of generalists physicians with expertise in the care of HIV patients. These findings, however, were attenuated by June 1997 and no longer statistically significant, suggesting that over time, HAART therapy was successfully adopted by the broader physician community such that generalist physicians who lacked expertise in HIV care were also using this important therapy. These findings are consistent with the literature on the diffusion of innovation.25

This is the first study that we are aware of to simultaneously examine the influence of formal training and expertise (including caseload) on the treatment patterns of a nationally representative population of patients with a chronic medical condition. These findings suggest that generalist physicians with appropriate experience have the potential to provide high-quality care to patients with complex chronic illnesses, at least as measured by the specific measure of technical quality that we examined in this study, and that their patients get similar access to important new therapeutic advances. Conversely, physicians with little experience should not be relied upon as the primary caregivers for complex conditions unless they are able to co-manage the patients with an expert provider. Consequently, guidelines and treatment algorithms should be designed to incorporate the possibility of differential levels of expertise among generalists (and specialists) and should acknowledge that generalist physicians are often able to provide high-quality care as well. While guidelines for the treatment of HIV infection have evolved over the last several years, we examined measures of HAART use that were thought to be state of the art during the time period of our study. Thus, these changes in recommendations would not affect our main conclusions with respect to physician specialization and the adoption and use of state-of-the-art treatments for a complex chronic medical condition.

In research conducted before the introduction of protease inhibitors, Kitahata et al. found that patients of physicians who had treated at least 5 HIV-infected patients had lower mortality.9 Findings from Stone et al., who also used a cutoff of 5 patients to define low-volume physicians, also indicated that experience in HIV care was beneficial, but other cutoffs were not reported.10 In our study, few physicians were caring for 5 or fewer patients. Our data suggest that as the AIDS epidemic has matured, physicians who care for HIV-infected patients tend to have a higher caseload. Thus, as the complexity of care has increased, HIV caseloads have increased, suggesting that physicians require a higher threshold of patients to maintain competence. Despite this, low-volume physicians continue to care for almost 10% of the population with HIV infection.

Some have hypothesized that patients who want to pursue the most advanced and aggressive therapy will preferentially seek out physicians trained in infectious diseases and/or the most-experienced clinicians. While patients of ID physicians did have slightly more advanced HIV disease, the vast majority of patients seen by each class of physicians rated taking antiretroviral therapy as “definitely or probably worth it.” Nonetheless, fewer patients of generalist non-experts agreed with this statement, so patient attitudes were adjusted for in our multivariable analyses. Thus, while patient attitudes toward taking ARVs were important predictors of being on HAART therapy, differing attitudes did not account for differences in use of HAART therapy across types of physicians.

Studies of other conditions have shown that specialists are usually more knowledgeable than generalists about diagnostic techniques26,27 and efficacious therapies.28,29 In addition, specialists tended to provide care deemed appropriate at higher rates than generalists when processes of care are examined (using chart reviews or patients' reports) for acute myocardial infarction ,30,31 unstable angina,32 asthma,33 acute arthritis34,35, multiple sclerosis,36 and depression.37 In addition, studies have demonstrated that high-volume physicians have better outcomes for treatment processes such as angioplasty and many surgical procedures.3841 Our findings extend these studies by examining formal training and caseload simultaneously and suggest that caseload, as a measure of expertise, might be as important or more important than specialty training. In addition, we also show that while specialized physicians adopt new treatment strategies more rapidly, these innovations successfully diffuse to the broader physician community over time. To the extent that a particular disease condition is experiencing rapid introduction of new treatment modalities, however, expert clinicians (whether by virtue of specialty training or other modes of developing expertise) will generally provide more-current therapy to patients.

Our study has several limitations. First, although we were able to identify appropriate links to physicians for over 70% of the sample, there might have been unmeasured differences between linked and unlinked patients. Including patient characteristics and weights adjusting for nonresponse in our models, however, largely controlled for these potential biases. Furthermore, propensity score analysis results were consistent with our original findings. Second, while we carefully identified the physicians caring for the HCSUS patients, some patients may have switched physicians at some point in time. Consequently, we focused our analysis on the time period closest to the identification of the physicians. In addition, we would expect that any misclassifications would make it more difficult to detect true differences. Third, while this is among the largest studies to date of this type, small differences can sometimes have a larger cumulative impact when they apply to large numbers of patients. Thus, we might have lacked power to detect important differences between ID physicians and general medicine experts. Finally, we only examined 1 important aspect of care for patients with HIV infection. We might have observed different results if we had examined other processes of care, such as the appropriate treatment of opportunistic infections.

Our findings demonstrate that expert generalists, whether by virtue of experience or interest, are able to provide care of quality equal to that of specialists, at least as measured by the use of HAART therapy. Guidelines and strategies to improve care for HIV patients should therefore promote the use of expert generalists as well as ID physicians. In addition, physicians without such expertise care for a small but substantial proportion of patients. Developing strategies to obtain appropriate expert input for these patients might also lead to improved outcomes.

Acknowledgments

We are indebted to Lin Ding, PhD for assistance with expert statistical programming; to Deborah Collins for assistance with manuscript preparation; to Shirley Nederend and Dana Perry for assistance with survey administration.

Supported by grants from the Robert Wood Johnson Foundation, the Agency for Healthcare Research and Quality (grant no. R-01HS10227), and the Aetna Quality Forum, and by a cooperative agreement (U-01HS08578) between RAND and the Agency for Health Care Policy and Research. Dr. Bozzette is a Health Services Research and Development Senior Research Associate of the Department of Veterans Affairs.

REFERENCES

1. Shapiro MF, Morton SC, McCaffrey DF, et al. Variations in the care of HIV-infected adults in the United States. Results from the HIV Cost and Services Utilization Study. JAMA. 1999;281:2305–15. [PubMed]
2. Carpenter CCJ, Fischl MA, Hammer SM, et al. Antiretroviral therapy for HIV infection in 1996. Recommendations of an international panel. International AIDS Society-USA. JAMA. 1996;276(2):146–54. [PubMed]
3. Carpenter CC, Cooper DA, Fischl MA, et al. Antiretroviral therapy in adults: updated recommendations of the International AIDS Society-USA Panel. JAMA. 2000;283:381–90. [PubMed]
4. Carpenter CC, Fischl MA, Hammer SM, et al. Antiretroviral therapy for HIV infection in 1997: updated recommendations of the International AIDS Society-USA panel. JAMA. 1997;277:1962–9. [PubMed]
5. Carpenter CC, Fischl MA, Hammer SM, et al. Antiretroviral therapy for HIV infection in 1998: updated recommendations of the International AIDS Society-USA Panel. JAMA. 1998;280:78–86. [PubMed]
6. Centers for Disease Control and Prevention. HIV AIDS Surveillance Report 11 (no.1) Atlanta, Ga: U.S. Department of Health and Human Services, Centers for Disease Control and Prevention; 1999.
7. Palella FJ, Delandy KM, Moorman AC. Declining morbidity and mortality among patients with advanced human immunodeficiency virus infection. N Engl J Med. 1998;338:853–60. [PubMed]
8. Bozzette SA, Joyce G, McCaffrey DF. Expenditures for the care of HIV-infected patients in the era of highly active antiretroviral therapy. N Engl J Med. 2001;344:817–23. [PubMed]
9. Kitahata MM, Koepsell TD, Deyo RA, Maxwell CL, Dodge WT, Wagner EH. Physicians' experience with the acquired immunodeficiency syndrome as a factor in patients' survival. N Engl J Med. 1996;334:701–6. [PubMed]
10. Stone VE, Mansourati FF, Poses RM, Mayer KH. Relation of physician specialty and HIV/AIDS experience to choice of guideline-recommended antiretroviral therapy. J Gen Intern Med. 2001;16:360–8. [PMC free article] [PubMed]
11. Kitahata MM, Van Rompaey SE, Shields AW. Physician experience in the care of HIV-infected persons is associated with earlier adoption of new antiretroviral therapy. J Acquir Immune Defic Syndr. 2000;24:106–14. [PubMed]
12. Bozzette SA, Berry SH, Duan N. The care of HIV-infected adults in the United States. HIV Cost and Services Utilization Study Consortium. N Engl J Med. 1998;339:1897–904. [PubMed]
13. Frankel MR, Shapiro JF, Duan N, et al. National probability samples in studies of low-prevalence diseases. Part II: designing and implementing the HIV cost and services utilization study sample. Health Serv Res. 1999;34(5 Pt 1):969–92. [PMC free article] [PubMed]
14. Shapiro MF, Berk ML, Berry SH, et al. National probability samples in studies of low-prevalence diseases. Part I: perspectives and lessons from the HIV cost and services utilization study. Health Serv Res. 1999;34:951–68. [PMC free article] [PubMed]
15. Kish L. Survey Sampling. New York: John Wiley; 1965.
16. Lam NSN, Liu KB. Use of space-filling curves in generating a national rural sampling frame for HIV/AIDS research. Professional Geographer. 1996;48:321–32.
17. Duan N, McCaffrey D, Frankel M, et al. (RAND). HCSUS baseline methods technical report. In press.
18. Landon BE, Wilson IB, Wenger NS, et al. Specialty training and specialization among HIV physicians who treat HIV/AIDS in the United States. J Gen Intern Med. 2002;17:12–22. [PMC free article] [PubMed]
19. American Medical Association. Available at: http://www.ama-assn.org/aps/amahg.htm. Accessed February 10, 2003.
20. Centers for Disease Control and Prevention. Guidelines for the Use of Antiretroviral Agents in HIV-Infected Adults and Adolescents. Department of Health and Human Services and Henry J. Kaiser Family Foundation. Morb Mortal Wkly Rep. 1998;47:43–82. [PubMed]
21. Report of the NIH panel to define principles of therapy of HIV infection. Ann Intern Med. 1998;128(12 Pt 2):1057–78. [PubMed]
22. Andersen R, Bozzette S, Shapiro M, et al. Access of vulnerable groups to antiretroviral therapy among persons in care for HIV disease in the United States. HCSUS Consortium. HIV Cost and Services Utilization Study. Health Serv Res. 2000;35:389–416. [PMC free article] [PubMed]
23. Cunningham WE, Markson LE, Andersen RM, et al. Prevalence and predictors of highly active antiretroviral therapy use in patients with HIV infection in the United States. HCSUS Consortium. HIV Cost and Services Utilization Study. J Acquir Immune Defic Syndr. 2000;25:115–23. [PubMed]
24. Rubin DB. Estimating causal effects from large data sets using propensity scores. Ann Intern Med. 1997;127(8S):757–63. [PubMed]
25. Rogers EM. Diffusion of Innovations. 4th ed. New York: The Free Press; 1995.
26. Hnatiuk O, Moores L, Loughney T, Torrington K. Evaluation of internists' spirometric interpretations. J Gen Intern Med. 1996;11:204–8. [PubMed]
27. Dolan NC, Martin GJ, Robinson JK, Rademaker AW. Skin cancer control practices among physicians in a university general medicine practice. J Gen Intern Med. 1995;10:515–9. [PubMed]
28. Ayanian JZ, Hauptman PJ, Guadagnoli E, Antman EM, Pashos CL, McNeil BJ. Knowledge and practices of generalist and specialist physicians regarding drug therapy for acute myocardial infarction. N Engl J Med. 1994;331:1136–42. [PubMed]
29. Chin MH, Friedmann PD, Cassel CK, Lang RM. Differences in generalist and specialist physicians' knowledge and use of angiotensin-converting enzyme inhibitors for congestive heart failure. J Gen Intern Med. 1997;12:523–30. [PMC free article] [PubMed]
30. Ayanian JZ, Guadagnoli E, McNeil BJ, Cleary PD. Treatment and outcomes of acute myocardial infarction among patients of cardiologists and generalist physicians. Arch Intern Med. 1997;157:2570–6. [PubMed]
31. Jollis JG, DeLong ER, Peterson ED, et al. Outcome of acute myocardial infarction according to the specialty of the admitting physician. N Engl J Med. 1996;335:1880–7. [PubMed]
32. Schreiber TL, Elkhatib A, Grines CL, O'Neill WW. Cardiologist versus internist management of patients with unstable angina: treatment patterns and outcomes. J Am Coll Cardiol. 1995;26:577–82. [PubMed]
33. Legorreta AP, Christian-Herman J, O'Connor RD, Hasan MM, Evans R, Leung KM. Compliance with national asthma management guidelines and specialty care: a health maintenance organization experience. Arch Intern Med. 1998;158:457–64. [PubMed]
34. Panush RS, Carias K, Kramer N, Rosenstein ED. Acute arthritis in the hospital: comparison of rheumatologic with nonrheumatologic care. J Clin Rheumatol. 1995;1:74–80. [PubMed]
35. Evans TI, Wheeler MT, Small RE, Breitbach SA, Sanders KM, Roberts WN. A comprehensive investigation of inpatient intravenous colchicine use shows more education is needed. J Rheumatol. 1996;23:143–8. [PubMed]
36. Vickrey BG, Edmonds ZV, Shatin D, et al. General neurologist and subspecialist care for multiple sclerosis: patients' perceptions. Neurology. 1999;53:1190–7. [PubMed]
37. Norquist G, Wells KB, Rogers WH, Davis LM, Kahn K, Brook R. Quality of care for depressed elderly patients hospitalized in the specialty psychiatric units or general medical wards. Arch Gen Psychiatry. 1995;52:695–701. [PubMed]
38. Katz JN, Losina E, Barrett J, et al. Association between hospital and surgeon procedure volume and outcomes of total hip replacement in the United States Medicare population. J Bone Joint Surg Am. 2001;83:1622–9. [PubMed]
39. Jollis JG, Peterson ED, DeLong ER, et al. The relation between the volume of coronary angioplasty procedures at hospitals treating Medicare beneficiaries and short-term mortality. N Engl J Med. 1994;331:1625–9. [PubMed]
40. Hannan EL, Racz M, Ryan TJ, et al. Coronary angioplasty volume-outcome relationships for hospitals and cardiologists. JAMA. 1997;279:892–8. [PubMed]
41. Canto JG, Every NR, Magid DJ. The volume of primary angioplasty procedures and survival after acute myocardial infarction. N Engl J Med. 2000;342:1573–80. [PubMed]

Articles from Journal of General Internal Medicine are provided here courtesy of Society of General Internal Medicine