In the late 1980s, Medicaid-insured human immunodeficiency virus (HIV)–infected patients with Pneumocystis carinii pneumonia (PCP) were 40% less likely to undergo diagnostic bronchoscopy and 75% more likely to die than were privately insured patients, whereas rates of use of other, less resource-intensive aspects of PCP care were similar. We reviewed 1395 medical records at 59 hospitals in 6 cities for the period 1995–1997 to examine the impact of insurance status on PCP-related care. Medicaid patients were only one-half as likely to undergo diagnostic bronchoscopy as were privately insured patients, yet we found no evidence that mortality was greater among patients who received empirical treatment. The bronchoscopy rates were primarily related to patients' personal insurance status. A weaker hospital-level effect was seen that was related to hospitals' Medicaid/private insurance case mix ratios. The situation has evolved from one in which Medicaid coverage was associated with underuse of bronchoscopy and poorer survival among empirically treated persons with HIV-related PCP to one in which empirical therapy is effective in treating this disease and expensive diagnostic procedures may be overused for privately insured patients.
The relationship between health insurance and the quality and intensity of medical care is a major concern in the United States. Persons covered by Medicaid insurance are less likely to receive diagnostic tests and therapeutic procedures than are those with private insurance [1–5], which suggests that there is a gap in access to quality medical care. These considerations are especially relevant for HIV-infected individuals, for whom Medicaid has become the primary source of health insurance [5–9].
Pneumocystis carinii pneumonia (PCP) is one of the most common serious complications of AIDS and is the subject of the largest number of AIDS-related quality-of-care studies reported to date. Earlier in the AIDS epidemic, critically ill HIV-infected patients with suspected PCP who were covered by Medicaid were 40% less likely to undergo diagnostic bronchoscopy and 75% more likely to die during hospitalization than were privately insured patients. However, the rates of use of less resource-intensive aspects of PCP care were similar across insurance groups . Whether Medicaid patients with HIV-related PCP were less likely to undergo bronchoscopy because they received care at hospitals that provide lower-intensity medical care (hospital-level effect) or whether physicians chose to provide a lower level of care to Medicaid patients (patient-level effect) is not known. Previous studies have had conflicting results. At one large-volume hospital, diagnostic bronchoscopy rates were one-half as high for Medicaid patients with HIV-related PCP as for similar but privately insured patients, which supports a patient-level effect theory . In contrast, in another study that included 2174 patients with HIV-related PCP from 82 hospitals, patients at hospitals with large Medicaid case loads had lower bronchoscopy rates, and those at hospitals with lower Medicaid case loads had higher bronchoscopy rates . The latter findings support the theory that insurance case mix contributes to hospital-level practice patterns for resource-intensive procedures.
In the present study, we evaluated whether insurance-related variations in care for patients with HIV-related PCP persisted into the second decade of the AIDS epidemic and, if so, whether these variations primarily were related to patient-level or hospital-level factors. We obtained clinically detailed information for 1395 individuals with HIV-related PCP who received care at 59 hospitals in 6 US cities in the period 1995–1997.
The study population consisted of all HIV-infected individuals with confirmed or probable PCP who received a portion of their medical care in a study institution in the years 1995–1997. The sampling methodology has been described in detail elsewhere [11, 12]. In brief, the study institutions represented a random sample of hospitals in 6 major US cities: Chicago, Miami, Los Angeles, New York, Seattle, and Phoenix. Four cities had high HIV incidence rates (Chicago, Miami, Los Angeles, and New York), and 2 cities had moderate HIV incidence rates (Seattle and Phoenix). All medical records that had the International Classification of Diseases, 9th Revision clinical modification codes for PCP (136.3) and HIV (042–044) were screened for inclusion in the study. Subjects had to be ⩾18 years old and meet 1 of 2 other eligibility criteria: (1) probable PCP diagnosis (written in the progress note or in the discharge summary note by the attending physician or consultant) or (2) documentation in the medical record of PCP diagnosis confirmed by cytologic examination of a specimen obtained through bronchial washing/brushing, a sputum sample, a transbronchial/open lung biopsy specimen, or an autopsy specimen. Study exclusion criteria included (1) receipt of inpatient medical care for the current episode of PCP at another hospital, (2) diagnosis of cancer (except Kaposi sarcoma) reported in the admitting note or the discharge sheet, and (3) failure to find supporting evidence of HIV or PCP in the medical record. Institutional review board approval was obtained at Northwestern University and at each of the study hospitals before data collection.
A total of 1395 medical records were abstracted by trained medical record reviewers, which represents a 90% success rate in record retrieval. Quality was ensured by review of all abstracted data by 2 physician overreaders. Interrater reliability was assessed through reabstraction of 10% of the charts by a different abstractor. Agreement of >95% was observed in sociodemographic, severity of illness (SOI), process of care, and outcome data.
Hospitals were characterized according to ownership structure as for-profit, private not-for-profit (including church-affiliated and volunteer), or public (county/state-operated). Teaching affiliation was categorized according to the American Hospital Association guide . As described in our previous studies, hospital HIV-related PCP experience was characterized by the total number of patients with HIV-related PCP who had received care at that institution in the period 1995–1997 (hospitals with <35 cases were considered to have a low level of experience, and hospitals with ⩾35 cases were considered to have a high level of experience) [12, 14, 15].
Patients were categorized by insurance type: Medicaid (including HMO Medicaid) or private (including Medicare). Uninsured or self-pay patients were not included in this study (this group was heterogeneous and included affluent self-pay and destitute patients). Sociodemographic data included age, sex, race, and employment status. Clinical information obtained about disease characteristics included previous HIV diagnosis, history of AIDS, medical comorbidities, use of PCP and Mycobacterium avium complex prophylaxis, and use of antiretroviral medications before admission. To measure SOI at admission, a recently published staging system based on hierarchically optimal classification tree analysis was used . This 5-category staging system for predicting mortality is based on 3 predictors: serum albumin level, arterial blood gas oxygenation levels, and history of wasting before admission [12, 16]. The mortality rate increased with stage: 3.7% for stage 1, 8.5% for stage 2, 16.1% for stage 3, 23.2% for stage 4, and 49.1% for stage 5 .
Diagnosis, management, and treatment were evaluated as “processes of care.” For diagnosis, we focused on the use of blood cultures, bronchoscopy, cytologic examination of sputum, and confirmation of PCP. Measures of patient treatment included initiation of PCP medications within 2 days after admission and use of adjunctive corticosteroids within 3 days after initiation of anti-PCP medications in persons with an alveolar-arterial oxygen gradient of >35 mm Hg or partial pressure of oxygen of <70 mm Hg. These processes of care are supported by National Institutes of Health consensus statements for PCP care . Our primary outcome was in-hospital mortality (excluding individuals who left the hospital against medical advice).
We compared patient characteristics and outcomes by insurance type, using χ2 tests for categorical variables and Wilcoxon rank sum tests for continuous variables. Linear regression was used to estimate the effect of the proportion of patients with PCP who were covered by Medicaid on PCP-related bronchoscopy rates at a particular hospital, weighted by the number of patients at each hospital. Only hospitals that treated ⩾5 patients with PCP were included in the analysis of bronchoscopy rates. This analysis was first performed for all patients combined and then for Medicaid and privately insured patients separately. The relationship between insurance type and diagnostic bronchoscopy rates was further assessed using a generalized linear random effects model (PROC GLIMMIX, SAS statistical software, version 8.0; SAS), assuming a binomial distribution and a logit link function to estimate adjusted ORs (aORs). The patient-level insurance effect (Medicaid versus private insurance) and the hospital-level insurance effect (the percentage of patients with Medicaid at the hospital) were assessed in models in which hospital was accounted for as a random effect. SAS statistical software, version 8.0 (SAS), was used for all analyses. Statistical significance was defined as P < .05, using 2-sided tests.
Of the 59 study hospitals, 13 were in Chicago, 12 were in Los Angeles, 20 were in New York, 4 were in Miami, 4 were in Phoenix, and 6 were in Seattle. The hospitals varied in ownership, teaching affiliation, and level of HIV/AIDS experience. The majority (68%) were volunteer or church-affiliated hospitals; 14% were for-profit hospitals, and 17% were public hospitals. Medicaid patients and privately insured patients tended to receive care at different types of hospitals (table 1). Public hospitals provided care for 30.1% of the Medicaid patients, compared with 11.5% of privately insured patients (P < .001). Hospital HIV/AIDS experience was not statistically significantly associated with bronchoscopy rates (aOR, 1.07; 95% CI, 0.74–1.55) or inpatient mortality (aOR, 0.72; 95% CI, 0.43–1.21). Sensitivity analyses indicated that our findings were qualitatively similar over a wide range of cutoff points for high- versus low-level HIV/AIDS experience.
Patients in the 2 health insurance categories differed significantly in demographic characteristics, HIV risk groups, and disease characteristics (table 1). Medicaid patients were more likely to be nonwhite (77.3% of Medicaid patients were nonwhite, vs. 48.4% of privately insured patients; P < .001), female (26.7% vs. 11.2%; P < .001), and unemployed (52.1% vs. 15.2%; P < .001) and to have a history of injection drug use (33.3% vs. 9.4%; P < .001). Insurance status was not significantly related to previous HIV care. Medicaid patients were just as likely as privately insured patients to have received antiretroviral therapy (33.2% vs. 37.9%; P = .07) or PCP prophylaxis (51.8% vs. 51.8%; P = .99) or to have received a diagnosis of HIV infection (82.7% vs. 82.6%; P = .94) before admission. SOI at admission did not vary with insurance (10.8% of Medicaid patients were stage 4 or 5, vs. 12.7% of privately insured patients; P = .32). Medicaid-insured individuals were less likely to have CD4 cell counts <50 cells/mm3 than were privately insured individuals (62.6% vs. 54.5%; P < .01).
Hospitals varied greatly in the rates of bronchoscopy performed among patients with HIV-related PCP (range, 0%–93%; figure 1). Bronchoscopy rates were higher in hospitals with smaller proportions of Medicaid patients. For each increase of 10% in the proportion of Medicaid patients, the estimated bronchoscopy rate decreased by 2.4% (P = .01; figure 1). The negative association between bronchoscopy rate and proportion of Medicaid patients within a hospital was observed both among the Medicaid patients and among the privately insured patients within individual hospitals (P = .77, by test for interaction in an adjusted mixed effects model; figure 2). This trend toward lower bronchoscopy rates in hospitals with higher numbers of Medicaid patients was notable. Of the 23 hospitals with appreciable numbers of both privately insured patients and Medicaid patients, 57% had higher hospital-specific bronchoscopy rates for privately insured patients than for Medicaid patients, and 26% had equivalent rates; only 17% had higher bronchoscopy rates for Medicaid patients than for privately insured patients. In addition to the hospital-level effect, a patient-level effect was observed. Overall, bronchoscopy was performed less often for Medicaid patients than for privately insured patients (27.4% vs. 42.4%; P < .0001; table 1). To see whether hospital- and patient-level effects contributed independently to the trend, multivariate random effects regression models were run that were adjusted for both effects, in addition to SOI, injection drug use, hospital discharge against medical advice, preadmission receipt of PCP prophylaxis, and a history of PCP. The aOR for the patient-level effect was statistically significant (aOR, 0.72; P = .02). However, the hospital-level effect was not statistically significant (aOR for an increase of 10% in the proportion of Medicaid-insured patients, 0.93; P = .13). Of note, bronchoscopy use was not associated with SOI (data not shown).
Initiation of timely anti-PCP medications or use of adjunctive corticosteroids was unrelated to insurance status (table 1). Hospital-specific rates of timely initiation of anti-PCP medication and appropriate adjunctive corticosteroid use were similar for Medicaid patients and privately insured patients (OR, 0.87 [95% CI, 0.55–1.83], and OR, 0.96 [95% CI, 0.67–1.37], respectively) and did not vary according to hospital case mix of Medicaid versus privately insured patients with PCP. Mortality for Medicaid patients was similar to that for privately insured patients (11.6% vs. 11.9%; P = .88). Moreover, even after adjustment for differences in patient-level clinical and sociodemographic factors and hospital-level characteristics, Medicaid patients and privately insured patients with PCP had similar odds of dying during the hospitalization (aOR, 1.01; 95% CI, 0.69–1.48; P = .96).
During 1995–1997, critically ill patients with AIDS who had suspected PCP and were covered by Medicaid were only one-half as likely to undergo diagnostic bronchoscopy as were patients with private insurance. Our findings are similar to those reported elsewhere for persons with HIV-related PCP who received care during 1987–1990 , as well for persons with a wide variety of other medical illnesses [3, 4]. In interpreting our results, we attempted to evaluate both patient- and hospital-level factors associated with variations in diagnosis, treatment, and outcomes for these 2 groups of patients.
Several studies have reported that hospital-level characteristics are important determinants of the type and intensity of HIV-related care that is provided [5, 18]. Similarly, in this study, public hospitals and private not-for-profit hospitals at which most patients with HIV-related PCP were covered by Medicaid had low bronchoscopy rates, whereas private for-profit and not-for-profit hospitals at which most patients with HIV-related PCP were privately insured had high bronchoscopy rates. However, we found that having private health insurance was a more important determinant of bronchoscopy use than was type of hospital. Of the 23 hospitals with appreciable numbers of both privately insured and Medicaid patients with HIV-related PCP, in only 1 in 6 were bronchoscopy rates higher among Medicaid patients than among privately insured patients. Similarly, Loue et al.  reported that the odds of undergoing a diagnostic bronchoscopy were only 50% as great for Medicaid patients as for privately insured patients with PCP. These findings raise the concern that resource rationing as a function of health insurance may be occurring. In contrast, reimbursement considerations would be expected to have a smaller impact on the use of less expensive treatments, such as trimethoprim-sulfamethoxazole and adjunctive corticosteroids. Consistent with this, we found small insurance-related differences in hospital-specific rates for the use of PCP medications and large insurance-related differences in hospital-specific rates of bronchoscopy use. In particular, our findings suggest that rates of use of essential resources, such as HAART, for HIV-infected individuals covered by Medicaid are similar to those for privately insured HIV-infected individuals, whereas rates of use of resources for which there is clinical debate over the necessity of use are lower for individuals who are covered by Medicaid.
During the 2 decades of the AIDS epidemic, there has been debate over whether there is a need for diagnostic bronchoscopy in HIV-infected individuals with pneumonia. Recent consensus panels of AIDS experts and quality—of—PCP care studies have focused on the use of PCP medications. In the present study, we found no evidence that mortality rates were higher when empirical PCP treatment was administered. This differs from what we reported in the 1980s, when we found that empirically treated patients were 50% more likely to die than were HIV-infected individuals with confirmed PCP diagnoses . Early in the AIDS epidemic, there was a higher level of concern over possible concomitant untreated pulmonary Kaposi sarcoma or non-PCP pulmonary infections that might have complicated the care of persons presenting with a pneumonia syndrome. In the second decade of the epidemic, support for empirical treatment appears to dominate. Thus, the debate may be turning from the issues of the last decade, when underuse of bronchoscopy, unconfirmed PCP diagnosis, and poorer survival were all associated with Medicaid insurance, to questions about the overuse of this procedure for privately insured patients.
Our study has limitations that must be acknowledged. First, although our data encompass a large number of patients, hospitals, and geographic areas, the generalizabilty of our results remains unproven. Second, our findings are for the period 1995–1997, which overlaps with the introduction of HAART into general use. The incidence of AIDS-defining opportunistic infections, including PCP, among HIV-infected patients has decreased during the HAART era . Although PCP has become increasingly less common, the illness continues to represent a major problem for patients who do not receive or who are noncompliant with HAART regimens, and its incidence may again increase as more patients develop multidrug-resistant HIV infection. Third, our data were obtained from medical record reviews. Incompleteness in recording procedures or miscoding of procedures may have occurred. However, we used reliable methods for ensuring accurate abstraction of data and also reviewed the procedure codes of each individual record.
In conclusion, during the mid-1990s, type of health insurance was strongly associated with the intensity of diagnostic evaluation for HIV-related PCP. However, contrary to the situation earlier in the HIV epidemic, when empirically treated individuals with PCP experienced higher mortality rates than did persons with confirmed diagnoses, mortality rates were similar for Medicaid patients versus privately insured individuals and for empirically treated patients versus patients with cytologically confirmed cases.
Hines Veterans Affairs Hospital, Hines (J.P.P.), Department of Medicine, Loyola University Stritch School of Medicine, Maywood (J.P.P.), Department of Preventive Medicine (M.D.-K., J.S.C.) and Department of Medicine (L.P., C.L.B.), Northwestern University Feinberg School of Medicine, Veterans Affairs Chicago Healthcare System, Lakeside Division (C.L.B.) and Westside Division (A.M.A.), Department of Medicine, University of Illinois College of Medicine (A.M.A.), Department of Medicine, Cook County Hospital (R.A.W.), Rush Medical College (R.A.W.), Department of Medicine, Sinai Health System (D.B.), Chicago, Illinois; Department of Medicine, Veterans Affairs Greater Los Angeles Healthcare System (M.B.G.), and University of California Los Angeles School of Medicine (M.B.G.), Los Angeles; Division of Infectious Diseases, University of Miami School of Medicine, Miami, Florida (R.C.); and AIDS Center, Mount Sinai Medical Center, New York (J.J.), and Department of Medicine, State University of New York Health Science Center at Brooklyn, Brooklyn (J.D.).
We acknowledge the invaluable assistance of the nurse abstractors in each of the cities for data collection and the staffs of each of the hospitals that assisted with identification of medical records for review.