Skip Navigation LinksHome > February 1, 2008 - Volume 47 - Issue 2 > Practice Makes Perfect: A Volume-Outcome Study of Hospital P...
JAIDS Journal of Acquired Immune Deficiency Syndromes:
doi: 10.1097/QAI.0b013e31815e402a
Epidemiology and Social Science

Practice Makes Perfect: A Volume-Outcome Study of Hospital Patients With HIV Disease

Hellinger, Fred PhD

Free Access
Article Outline
Collapse Box

Author Information

From the Agency for Healthcare Research and Quality, Rockville, MD.

Received for publication May 31, 2007; accepted October 12, 2007.

This is a US government work. There are no restrictions on its use with the exception of any previously printed figures and tables.

Correspondence to: Fred Hellinger, PhD, Agency for Healthcare Research and Quality, Rockville, MD 20850 (e-mail: fhelling@ahrq.gov).

Collapse Box

Abstract

Objective: There is considerable evidence that patients with HIV fare better in hospitals that treat more HIV-positive patients. Yet, it is possible that much of this benefit is attributable to the care provided by physicians who treat high volumes of HIV-positive patients. This study examines the relation between 2 measures of volume (the number of HIV-positive patients treated in a hospital and the number of HIV-positive patients treated by the attending physician) and the probability of dying in the hospital.

Data: This study uses discharge data from 43,325 patients hospitalized with HIV disease in 5 states (Colorado, Maryland, New Jersey, New York, and Washington State) in 2002. These data were obtained from the Agency for Healthcare Research and Quality's Healthcare Cost and Utilization Project State Inpatient Databases.

Study Design: Volume-outcome studies have demonstrated an inverse relation between the number of HIV-positive patients treated at a hospital and the mortality rate for these patients. Yet, the most current of these studies is based on data more than a decade old, and none of these account for the volume of HIV-positive patients treated by the physician. This study uses multivariate logistic regression analyses to estimate the impact of hospital and physician volume on patient mortality.

Results: This study found that when measures of physician and hospital volume are included in a regression equation explaining patient mortality, only the variable measuring physician volume remains statistically significant. Moreover, when a variable is defined for each patient based on the quartile rankings of the patient's hospital volume and the patient's physician volume, the quartile ranking of physician volume is a better predictor of survival than the quartile ranking of hospital volume.

Conclusion: These findings suggest that the volume of patients treated by the attending physician is the key measure of volume associated with the survival of hospitalized HIV-positive patients.

As a general rule, empiric studies indicate that the probability of a surgical patient dying in a hospital is inversely related to the number of the same procedures performed at the hospital and that this relation is strongest for highly technical surgical procedures (eg, pancreatectomy, gastrectomy, prostatectomy) as opposed to routine procedures. In a recent review article, Begg and Scardino1 declare that

Although not all studies have demonstrated a statistically significant volume-outcome trend, the overall pattern in the results is overwhelmingly consistent. High-volume hospitals demonstrate lower mortality rates, with the magnitude of the trend varying considerably by procedure.

In a 2002 review article, Halm and colleagues2 examined empiric studies of the relation between volume and outcome published between January 1980 and December 2000. Each of the studies reviewed included volume as an independent variable, health outcome as a dependent variable, and data from more than 1 institution. Of the 135 studies examined by these authors, 118 studies (70%) revealed a statistically significant relation between volume and improved health outcomes, and none of these studies revealed a statistically significant relation between volume and poorer health outcomes.

Halm and colleagues2 concluded (p. 511) that “The strongest associations were found for AIDS treatment and for surgery on pancreatic cancer, esophageal cancer, abdominal aortic aneurysms, and pediatric cardiac problems.” Halm and colleagues2 based their conclusion regarding the relation between volume and outcome for HIV-positive patients on 6 studies of hospitalized HIV-positive patients and 1 study of HIV-positive patients treated at a staff-model health maintenance organization between 1984 and 1994. Indeed, the only strong evidence of an inverse relation between the volume of services provided to a specific type of nonsurgical patient and the mortality rate for that type of patient comes from studies of patients with HIV disease, and most of these studies focus on patients with HIV-related Pneumocystis carinii pneumonia (PCP).

Most of the early volume-outcome studies of hospitalized HIV-positive patients focused on PCP because PCP was the most common cause of hospitalization. Specifically, Turner and colleagues3 found that 33% of hospitalized patients with AIDS in New York State in 1985 had PCP, and Turner and Ball4 found that 35% of 10,508 hospitalized patients with AIDS in the 258 hospitals in the Hospital Cost and Utilization Project (HCUP) sample in 1986 and 1987 had PCP. Mortality rates for hospitalized patients with AIDS were quite high in the early years of the epidemic. For example, in the aforementioned study by Turner and colleagues,3 the mortality rate was 22%, and in the aforementioned study by Turner and Ball,4 it was 17%.

In one of the most widely cited studies of the relation between volume and mortality for patients with AIDS, Bennett and colleagues5 demonstrated an inverse relation between AIDS familiarity and mortality for 257 patients with AIDS who had PCP and were treated at 15 California hospitals between October 1986 and October 1987 (Table 1). The authors used administrative data from a modified version of the Uniform Hospital Data Set (UHDS), and they measured AIDS familiarity as the number of patients with AIDS treated at a hospital per 10,000 discharges. In a subsequent study, Bennett and colleagues6 confirmed their finding that patients at hospitals that treat more patients with AIDS who have PCP have lower mortality rates. In this study, they utilized data for 3126 persons with AIDS who had PCP and were treated at 1 of 73 New York City hospitals in 1987 to demonstrate that the chances of death decreased when care was provided at a hospital with a higher caseload of patients with PCP.

Table 1
Table 1
Image Tools

Bennett and colleagues7 extended their research to examine mortality at 30 days after a hospital admission and the number of patients with PCP treated at a hospital using data from 140 Veterans Administration medical centers on 3981 patients treated between 1987 and 1991. The authors found that the 30-day mortality rate of patients with PCP treated at a hospital that treated 50 or more such patients was 27% less than the rate experienced by patients treated a hospital that treated 4 or fewer such patients.

In a study of 10,538 adult discharges of patients with AIDS treated at 258 hospitals in fiscal year 1987 from throughout the United States, Turner and Ball4 found that the higher the number of patients with AIDS treated at a hospital, the lower was its mortality rate for patients with AIDS. This study examined all patients with AIDS (not just those with PCP) using a multivariate logistic regression that included each opportunistic infection as a separate independent dichotomous variable in addition to other independent variables representing patient age, race, and gender. The geographic region the hospital was located in also was included as an independent variable, as was the number of patients with AIDS treated at the hospital. They found that an increase of 100 cases treated at a hospital resulted in a 4% decrease in the probability of dying in the hospital. The authors adjusted for the severity of illness using dichotomous variables for the existence of each opportunistic infection and patient age, gender, race, and geographic region of the country.

In another study of the relation between the number of patients with AIDS treated at a hospital and mortality, Stone and colleagues8 used multivariate logistic regression analysis to demonstrate the inverse relation between the number of patients with AIDS treated at a hospital and mortality using data on 806 patients with AIDS hospitalized in 40 Massachusetts hospitals in 1987. This study measured the severity of illness using the Severity Classification for AIDS Hospitalizations (SCAH) system. Prior studies used a variety of methods to adjust for case severity in multivariate models. In many studies, dichotomous variables representing AIDS-defining conditions and other comorbidities were included as well as a simple count of the number of diagnoses reported on the hospital discharge abstract.

In a more recent study, Hogg and colleagues9 demonstrated that patients in hospitals with more than 100 patients with AIDS per year had a 36% lower chance of dying in the hospital than patients in hospitals with an average of 1 patient or less with AIDS per year. This study adjusted for patient age, gender, year of admission, and severity of illness using the 4 World Health Organization (WHO) clinical stages of AIDS. The study utilized data on 38,075 HIV-positive patients treated at Canadian hospitals between March 1987 and April 1994.

In a study of 7901 adult hospitalized patients with HIV disease using the 1994 California Hospital Discharge Data Set, Cunningham and colleagues10 found that the mortality rate varied significantly based on the number of patients with AIDS treated at a hospital. In particular, they found that hospitals in the lowest quartile of AIDS volume had a 12.3% mortality rate for patients with AIDS and that hospitals in the top quartile of AIDS volume had a 7.6% mortality rate after adjusting for age, gender, race, insurance, teaching affiliation, and severity of illness.

In a 1990 study, Bennett and colleagues11 extended their research to examine the relation between the number of patients treated by a hospital and their utilization of resources. They utilized data from 15 California hospitals between October 1986 and October 1987 that they had examined in a previous study of the relation between hospital volume and mortality. Although they found that the average resource utilization between hospitals with marked variation in the number of patients with PCP did not diverge, they found that hospitals that treated more patients with PCP used more resources (longer intensive care unit [ICU] stays, more frequent bronchoscopies, and higher total charges) treating survivors than hospitals that treated fewer patients with PCP and that the opposite was true for nonsurvivors.

Horner and colleagues12 examined resource utilization by patients with PCP using medical record data on 890 patients in 56 hospitals in 3 cities (Chicago, Los Angeles, and Miami), and they found that Medicaid patients were approximately 40% less likely to undergo bronchoscopy than privately insured patients. They also found that Medicaid patients were 75% more likely to die in the hospital. In a more recent study of the resource utilization by patients with PCP, Parada and colleagues12a examined the medical records of 1395 patients with PCP treated at 59 hospitals in 6 cities (Chicago, Los Angeles, New York, Miami, Phoenix, and Seattle) and found that Medicaid patients were half as likely to undergo diagnostic bronchoscopy as privately insured patients and that patient characteristics were better predictors of resource utilization than hospital effects. Mortality rates did not vary by insurance status in this study, however.

Although there are no studies of the influence of physician volume on the mortality rates or resource utilization for hospitalized patients with HIV, there is evidence that patients treated by physicians who treat large numbers of HIV-positive patients in outpatient settings are more likely to receive appropriate therapy8,13,15 and to experience more favorable outcomes.14,15 In addition, it is certainly possible that much of the benefit experienced by patients in hospitals with high volumes of HIV-positive patients is attributable to the experience of the attending physician.

The current study is the first to examine the relation between the volume of hospitalized HIV-positive patients seen by a physician and survival. It is also the first study to examine the relation between volume and outcome for hospitalized HIV-positive patients after the diffusion of highly active antiretroviral therapy. The findings in this study address ongoing concerns about the appropriateness of care provided by physicians who treat few HIV-positive patients. Indeed, this study indicates that HIV-positive patients in hospitals fair better when they are cared for by physicians who treat many HIV-positive patients, and this finding buttresses findings from other studies indicating that HIV-positive patients cared for in outpatient settings by physicians who treat many HIV-positive patients fare better.

Back to Top | Article Outline

HIGHLY ACTIVE ANTIRETROVIRAL THERAPY

Powerful new drugs (protease inhibitors and nonnucleoside reverse transcriptase inhibitors) diffused rapidly throughout 1996 and 1997, and therapy involving these drugs is referred to as highly active antiretroviral therapy (HAART).16-18 At that time, a number of clinical studies provided information about the positive response of patients to HAART, and this information resulted in rapid changes in clinical practice.19,20,20a Moreover, tests to measure the level of HIV-1 RNA in plasma became widely available in 1996, and these tests were used to ascertain whether HAART was successful in suppressing HIV-1 replication.

Drug interactions and adverse drug effects became a serious concern after the diffusion of powerful new antiretroviral drugs (ie, protease inhibitors, nonnucleoside reverse transcriptase inhibitors), and knowledge about the benefits and risks of various drug combinations became essential because of the escalating possibility of adverse reactions from the expanding number of potent drugs used by patients with compromised immune systems. Today, there are 30 HIV medications in 6 classes (nucleoside reverse transcriptase inhibitors [13], non-nucleoside reverse transcriptase inhibitors [3], protease inhibitors [11], fusion inhibitors [1], entry inhibitors [1], and integrase inhibitors [1]). The first fusion inhibitor (enfuvirtide) was approved by the FDA on March 13, 2003. The first entry inhibitor (maraviroc) was approved on August 6, 2007, and the first integrase inhibitor (raltegravir) was approved on October 12, 2007.

The greater prospect for acute drug reactions and interactions associated with the diffusion of forceful anti-HIV medications has been more than compensated for by the influence of these drugs on the lifespan of persons with HIV disease. This influence is evidenced by the reduction in the number of deaths attributable to HIV disease in the United States immediately after the diffusion of HAART. The number of persons who died from HIV disease in the United States in 1995 was 51,117, and this number fell to 22,245 in 1997.21

The dramatic changes in the death rate attributable to HIV disease, hospital admission rates for persons with HIV disease, and the diagnostic profile of hospitalized patients with HIV disease that have occurred since the diffusion of HAART suggest the need for timely studies of the volume-outcome relation for HIV-positive patients in hospitals.

Back to Top | Article Outline

DATA

Information from 43,325 hospital discharges of persons with HIV disease in 5 states (Colorado, Maryland, New Jersey, New York, and Washington State) for the year 2002 is examined in this study. The 5 states in this study account for approximately 28% of the persons living with AIDS in the United States in January 2002.22

Hospital discharge data were obtained from the HCUP, State Inpatient Databases (SID), which are maintained by the Agency for Healthcare Research and Quality (AHRQ).23,24 In general, the SID contains the universe of a state's hospital inpatient discharge records and is composed of annual state-specific files that share a common structure and common data elements. Most data elements are coded in a uniform format across all states, and the uniform format facilitates cross-state comparisons. In addition, the SIDs are well suited for research that requires complete enumeration of hospitals and discharges within states. The discharge data for the states in this study also include a code number for the attending physician and a code number for each hospital.

This study uses a conservative approach for distinguishing hospitalized patients with HIV disease. Just those patients with a primary or secondary code diagnosis (most states include up to 6 secondary diagnoses) that contains 042 are acknowledged as patients with HIV disease. Thus, patients who are infected with HIV but who are not identified as such are not included in this analysis.

After October 1994, the International Classification of Diseases, Ninth Revision, Clinical Modifications (ICD-9-CM) system included only 1 code (042) for HIV disease and AIDS.25 In the most comprehensive study of the accuracy of diagnostic coding for persons with HIV disease, it was determined that 97% of persons with an HIV diagnosis on their hospital discharge abstract were infected with HIV.26 In that study, more than 7000 hospital records of persons in 6 states with diagnostic codes indicative of HIV disease were examined, and it was determined whether or not an individual was infected using AIDS surveillance data from state health departments and a review of the medical charts. The predictive accuracy of using codes for AIDS-related illnesses was poor. For example, it was determined that only 38% of patients with the diagnostic code for PCP (136.3) were infected with HIV.

Back to Top | Article Outline

METHODS

This study examines the relation between volume and mortality with the individual patient as the unit of observation. The relation between volume and outcome is examined through the use of multivariate logistic regression equations, where mortality is a function of volume and other independent variables.

A number of volume variables are defined for each patient observation, including the number of patients with HIV disease seen at the patient's hospital (Hospvol) and the number of HIV-positive patients treated by the patient's attending physician (MDvol). In addition, dichotomous variables representing the quartile of the volume of the patient's hospital and physician were constructed. For example, the variable QHospHighest/MDmedLow is set equal to 1 for a patient treated at a hospital whose volume of HIV-positive patients is more than the hospital volume of 75% of HIV-positive patients and by an attending physician whose volume is more than the physician volume of 25% of HIV-positive patients and less than the physician volume of 50% of HIV-positive patients. The variable QHosphighest is set equal to 1 for all patients treated at a hospital whose volume of HIV-positive patients is more than the hospital volume of 75% of HIV-positive patients, and the variable QMDmedLow is set equal to 1 for all patients whose attending physician's volume is more than the physician volume of 25% of HIV-positive patients and less than the physician volume of 50% of HIV-positive patients.

Other independent variables, including gender, race/ethnicity, teaching status, age, and dummy variables for each hospital and for each state, were also included to estimate hospital fixed effects and state fixed effects. The insurance status of a patient was categorized as Medicare, Medicaid, private, self-pay, or other, and hospital ownership was categorized as government, for-profit, or nongovernmental not-for-profit. Race was categorized as black, white, Hispanic, or other.

Because observations of patients treated by the same physician may be correlated (thus violating the assumption that explanatory variables are independently distributed), we estimated our model using generalized estimating equations (GEEs).27 A GEE is a technique to estimate data that are correlated. If data from patients treated by the same physician are correlated over time and this correlation is not taken into account, the standard errors for the coefficients derived using least squares estimates are underestimated.

The teaching status of each hospital was determined using a dichotomous variable indicating whether or not the hospital was a member of the Council of Teaching Hospitals (COTH). As a general rule, COTH membership implies that the hospital is a major teaching facility. The COTH was created by the Association of American Medical Colleges (AAMC) in 1965, and the approximately 400 COTH member institutions instruct approximately three quarters of the physician residents in the United States. Furthermore, these hospitals provide an enormous range of complex products and services to patients in a clinical setting intended to promote research.

In this study, we used dichotomous variables representing 31 chronic conditions to control for the severity of illness. Of these, 27 were obtained from a study by Elixhauser and colleagues,28 and each of these is identified by a set of diagnoses. Elixhauser and colleagues28 selected these conditions because of their power to explain mortality for heterogeneous sets of discharge records with all reasons for hospitalizations and for homogeneous subsets of these discharge records for patients with the same illness. In addition, we included dichotomous variables for 4 more conditions (cytomegalovirus, PCP, toxoplasmosis, and Mycobacterium avium intracellulare), and we included the number of diagnoses reported on the discharge abstract.

A logistic regression was used to estimate the impact of volume and the other independent variables on the probability of a patient dying in the hospital. Logistic regression was used because if linear regression is used, the predicted values may be greater than 1 and less than 0 and because one of the assumptions of a linear regression is that the errors of prediction are normally distributed.29 Because the dependent variable is either 1 or 0, however, the errors of prediction cannot be normally distributed.

Back to Top | Article Outline

RESULTS

Descriptive Statistics

Table 2 presents some descriptive statistics about the 43,325 hospital admissions of persons with HIV disease in 392 hospitals in the 5 states in our sample in 2002. The average age of an HIV-positive patient was 43 years old, whereas 37% were female, 54% were black, and 16% were Hispanic. The median of the variable indicating the number of HIV-positive patients treated in a hospital was 597 (ie, one half of all patients were cared for in a hospital that treated fewer than 597 HIV-positive patients during the year), and the median of the variable indicating the number of HIV-positive patients treated by the attending physician was 17 (ie, one half of all patients were treated by a physician who treated fewer than 17 HIV-positive patients during the year). The probability of dying in the hospital was 6% (2697 HIV-positive patients died in the hospital), and the average length of stay was 8.9 days. Medicaid was the primary payer for 56% of the patients, and Medicare was the primary payer for another 21%. Consequently, more than three fourths of the HIV-positive patients in hospitals in our sample had their care covered by public payers. The most common AIDS-defining condition was PCP, which accounted for 8% of cases, followed by cytomegalovirus, which accounted for 2% of cases.

Table 2
Table 2
Image Tools
Back to Top | Article Outline
Mortality

In the first equation in Table 3, we find that patients treated in hospitals that treat more HIV-positive patients have a lower probability of dying in the hospital (the probability of dying in a hospital decreased by 3% for each increase of 100 HIV-positive patients treated by a hospital). The second equation in Table 3 reveals that patients treated by physicians who treat more HIV-positive patients have a lower probability of dying in the hospital (the probability of dying in a hospital decreased by 2.4% for each increase of 10 HIV-positive patients treated by a physician). The third equation in Table 3 indicates that the coefficient of the hospital volume variable becomes statistically insignificant when the physician volume variable is included in the equation explaining probability of death.

Table 3
Table 3
Image Tools

Table 4 presents the odds ratios for variables representing the combined quartiles (where first quartile = lowest, second quartile = medlow, third quartile = medhigh, and fourth quartile = highest) of hospital volume (Hosp) and physician volume (MD). The coefficients for these variables were derived from equations that included the same set of non-volume-related independent variables as found in Table 3. The coefficients reflect the probability of dying in a hospital for patients in a given quartile for hospital and physician volume in relation to the probability of dying in the hospital for all patients not in this quartile.

Table 4
Table 4
Image Tools

For example, the coefficient of the variable QHosphighest/MDlowest is 1.21, and it is statistically significant. This implies that the odds of death for an HIV-positive patient treated at a hospital whose volume of HIV-positive patients is more than that of 75% of HIV-positive patients and who has an attending physician whose volume is less than the physician volume of 75% of HIV-positive patients is 21% greater than that of a patient not in this category.

The odds ratio for the variable QMDhighest is 0.74, which indicates that a patient treated by a physician who treated more patients than the physicians of 75% of the patients (75% of patients have MDvol ≤49) had a 26% lower chance of dying than a patient whose physician treated fewer than 49 HIV-positive patients. The odds ratio for the variable QHosphighest is 0.89, which indicates that a patient treated at a hospital whose volume of HIV-positive patients is more than the hospital volume of 75% of HIV-positive patients (75% of patients have Hospvol ≤922) had a 11% lower chance of dying than similar patients treated in hospitals with fewer than 922 HIV-positive patients.

Each of the odds ratios for the aggregated variables for physician volume is statistically significant, and the odds ratios for the third and fourth quartiles are <1, whereas those for quartiles 1 and 2 are >1 (see Table 4). Conversely, the odds ratios for 2 aggregated quartile variables for hospital volume were not statistically significant and the odds ratio for the second quartile is >1. These results and the finding that the odds ratio of hospital volume becomes statistically insignificant when the physician volume variable is added to the survival function (see Table 3) suggest that the key measure of volume affecting the probability of dying in the hospital is the volume of patients seen by the attending physician. In addition, the correlation coefficient between physician volume and hospital volume was 0.185, and a sizable proportion (28%) of physicians who treated a relatively large number of HIV-positive patients were treating patients in a hospital that treated a low volume of HIV-positive patients.

Back to Top | Article Outline

DISCUSSION

Scientists hypothesize that awareness and rigorous practice shape maps in our brain that coordinate the actions of neurons.30 Moreover, scientists hypothesize that when we think about what we are doing, and we do it over and over again, our brain maps become more defined and our performance becomes more skilled. The findings of volume-outcome studies support this way of thinking.

Volume-outcome studies of hospitalized HIV-positive patients have focused on how experiential learning relates to the probability of a patient dying in a hospital. Yet, these studies apply to HIV care before the diffusion of new and powerful antiretroviral drugs, and these studies do not account for the effect of the volume of patients seen by a physician. This study is the first volume-outcome study of hospitalized patients with HIV disease that uses data collected after the diffusion of protease and nonnucleoside reverse transcriptase inhibitors, and it is the first study to examine the impact of physician and hospital volume on the probability of a patient dying in the hospital.

Yet, this study has limitations. First, this study is based on a review of administrative data. This is a notable limitation, because administrative records do not possess detailed information about the immune status of patients. Indeed, a study of the relation between volume and outcome in the treatment of HIV-positive hospitalized patients that included data from medical records could provide valuable information about the volume-outcome relation for hospitalized HIV-positive patients. Such a study, however, would likely involve a relatively small number of hospitalized patients because of the high cost of reviewing medical records, and there are likely to be a relatively large number of hospitalized patients without a CD4 cell count or viral load measure included in their medical record.

Second, this study does not follow patients after their hospital stay, and it would be valuable to have information on the mortality rate of discharged patients 1, 2, and even 3 months after their discharge. If certain hospitals are more likely to transfer extremely sick patients to other acute care hospitals or to home health care possibly involving a hospice, the findings of this study would not provide a completely accurate assessment of the relation between volume and mortality. A small percentage of HIV-positive patients are transferred to other acute care hospitals (1.6% in New York State),31 however, and it is unclear how many severely sick HIV-positive patients are discharged because of their desire to die at home. There is evidence, however, that state of residence, gender, and race are related to the probability that a person with HIV dies in a hospital,32 and our analyses include gender, race, and state of residence. Thus, our results are biased only to the extent that the probability of a patient being discharged to die at home, in a hospice, or in a nursing home is systematically related to the volume of HIV care provided in a hospital or by a physician after adjusting for the impact of gender, race, state of residence, the existence of chronic conditions, age, insurance status, and number of diagnoses.

Third, this study includes a relatively small number of hospitals that treat large numbers of HIV-positive patients. Even though this study examines the records of 43,325 HIV-positive patients, only 39 hospitals treat more than the 469 HIV-positive patients, and one half all patients are treated in hospitals that treat <469 patients. Consequently, the 50% of the patients who are treated in hospitals that treat more than 469 patients are treated in 39 hospitals, whereas the total number of hospitals that treat HIV-positive patients in our sample is 322. Nevertheless, a relatively large number of physicians (379) treated more than 17 HIV-positive patients in our study, and 50% of the patients in our study were treated by a physician who treated fewer than 17 HIV-positive patients.

Fourth, the relation between volume and mortality observed in this study may reflect selective referrals patterns related to the quality of care.33,34 In particular, if physicians are more likely to refer their patients to hospitals that provide high-quality care (ie, low-mortality hospitals), the relation between volume and mortality in a given hospital is, to some extent, a consequence of selective referral.

Finally, it is important to note that hospitalization rates for HIV-positive patients have been steadily decreasing over time; thus, current studies may be measuring the rate of “unlearning” and not the effect of increased volume. This suggests that physicians with more patients have a lower rate of “unlearning” than physicians with few patients.

Today, HIV disease is generally perceived as a moderately expensive chronic illness rather than as an unusually expensive fatal illness. Indeed, concerns about survival and costly hospital stays have to a great extent been replaced by concerns about the ramifications of alternative drug regimens and the value of resistance testing. Yet, empiric studies suggest that the probability of an HIV-positive patient dying in a hospital is inversely related to the number of HIV-positive patients treated at the hospital, and this study suggests that the volume of patients treated by a physician is the important factor. Nonetheless, this conclusion is not definitive, because this study does not include the data necessary to construct good measures of the severity of illness and it does not include data on 30-day mortality rates.

Nevertheless, the conclusion that physician volume is more important than hospital volume in explaining mortality rates for HIV-positive patients is not entirely unexpected. The most appropriate entity for measuring the effect of experience for surgical patients may be the volume of patients treated by operating room staff in performing similar types of procedures. Yet, outcomes of hospitalized HIV-positive patients do not depend on operating room staff and are less dependent on complex medical equipment. Their survival may depend more on the skill and expertise of their physician, which suggests that the most appropriate entity for measuring the effect of volume is the number of HIV-positive patients seen by the patient's attending physician.

Indeed, a previous volume-outcome study of patients with heart disease found that the number of patients with acute myocardial infarction treated by a physician was more closely related to their survival than the number of such patients treated at the hospital.35 The same study found that hospital volume was a more important factor than physician volume in the treatment of patients with heart disease who underwent coronary catheterization and coronary artery bypass graft surgery.

Although we found that patients treated at a hospital that treats relatively large numbers of HIV-positive patients fare better than patients treated at hospitals that treat relatively few HIV-positive patients, this effect was not statistically significant when the number of HIV-positive patients seen by the patient's physician was included in the equation. This is important because the relatively high-volume threshold for physicians (median of 17 cases per year) is reached by only approximately 5% of the physicians treating HIV-positive patients. Indeed, standards for treating patients with HIV change rapidly, and it is more difficult for physicians who treat a small number of HIV-positive patients to keep abreast of and to assimilate these changes into their practice.

Physicians who treat more HIV-positive patients are likely to be exposed to a greater number of AIDS-related conditions, and repeated care of patients with these conditions may increase their ability to manage care effectively. Nev-ertheless, these findings do not constitute justification for the regionalization of HIV care in hospitals with extensive experience treating HIV-positive patients. They do, however, suggest the need for research into the techniques and mechanisms by which physicians who treat a large number of HIV-positive patients deliver high-quality care.

Back to Top | Article Outline

REFERENCES

1. Begg CB, Scardino PT. Taking stock of volume-outcome studies. J Clin Oncol. 2003;21:393-394.

2. Halm EA, Lee C, Chassin MR. Is volume related to outcome in health care? A systematic review and methodologic critique of the literature. Ann Intern Med. 2002;137:511-520.

3. Turner BJ, Kelly JV, Ball JK. A severity classification system for AIDS hospitalizations. Med Care. 1989;27:423-437.

4. Turner BJ, Ball JK. Variations in inpatient mortality for AIDS in a national sample of hospitals. J Acquir Immune Defic Syndr Hum Retrovirol. 1992;5:978-987.

5. Bennett CL, Garfinkle JB, Greenfield S, et al. The relation between hospital experience and in-hospital mortality for patients with AIDS-related PCP. JAMA. 1989;261:2975-2979.

6. Bennett CL, Adams J, Gertler P, et al. Relation between hospital experience and in-hospital mortality for patients with AIDS-related Pneumocystis carinii pneumonia: experience from 3126 cases in New York City in 1987. J Acquir Immune Defic Syndr Hum Retrovirol. 1992;5:856-864.

7. Bennett CL, Adams J, Bennett RL, et al. The learning curve for AIDS-related Pneumocystis carinii pneumonia: experience from 3981 cases in Veterans Affairs Hospitals 1987-1991. J Acquir Immune Defic Syndr Hum Retrovirol. 1995;8:373-378.

8. Stone VE, Seage GR 3rd, Hertz T, et al. The relation between hospital experience and mortality for patients with AIDS. JAMA. 1992;268:2655-2661.

9. Hogg RS, Raboud J, Bigham M, et al. Relation between hospital HIV/AIDS caseload and mortality among persons with HIV/AIDS in Canada. Clin Invest Med. 1998;21:27-32.

10. Cunningham WE, Tisnado DM, Lui HH, et al. The effect of hospital experience on mortality among patients hospitalized with acquired immune deficiency syndrome in California. Am J Med. 1999;107:137-143.

11. Bennett CL, Gertler P, Guze PA, et al. The relation between resource use and in-hospital mortality for patients with acquired immunodeficiency syndrome-related Pneumocystis carinii. Arch Intern Med. 1990;150:1447-1452.

12. Horner RD, Bennett CL, Rodriguez D. Relationship between procedures and health insurance for critically ill patient with Pneumocystis carinii pneumonia. Am J Respir Crit Care Med. 1995;152:1435-1442.

12a. Parada JP, Deloria-Knoll M, Chmiel JS, et al. Relationship between health insurance and medical care for patients hospitalized with human immunodeficiency virus-related Pneumocystis carinii pneumonia, 1995-1997: Medicaid, Bronchoscopy, and Survival. Clin Infect Dis. 2003;37:1549-1555.

13. Willard CL, Liljestrand P, Goldschmidt RH, et al. Is experience with human immunodeficiency virus disease related to clinical practice? Arch Fam Med. 1999;8:502-509.

14. Kitahata MM, Koepsell TD, Deyo RA, et al. Physicians' experience with acquired immunodeficiency syndrome as a factor in patients' survival. N Engl J Med. 1996;334:701-706.

15. Landon BE, Wilson IB, McInnes K, et al. Physician specialization and the quality of care for human immunodeficiency virus infection. Arch Intern Med. 2005;165:1133-1139.

16. Hellinger FJ. HIV patients in the HCUP database: a study of hospital utilization and costs. Inquiry. 2004;41:95-105.

17. Lane HC, Neaton JD. When to start therapy for HIV infection: a swinging pendulum in search of data. Ann Intern Med. 2003;138:680-681.

18. Moore RD. Cost effectiveness of combination HIV therapy: 3 years later. Pharmacoeconomics. 2000;17:325-331.

19. Bozzette SA, Berry SH, Duan N, et al. The care of HIV-infected adults in the United States. N Engl J Med. 1998;339:1897-1928.

20. Gebo KA, Fleishman JA, Moore RD. Hospitalizations for metabolic conditions, opportunistic infections, and injection drug use among HIV patients. J Acquir Immune Defic Syndr. 2005;40:609-616.

20a. Palella FJ, Delaney KM, Moorman AC, et al. Declining morbidity and mortality among patients with advanced human immunodeficiency virus infection. HIV Outpatient Study Investigators. N Engl J Med. 1998;338:853-860.

21. US Centers for Disease Control and Prevention. HIV/AIDS Surveillance Report: US HIV and AIDS Cases Reported Through June 2001, vol. 13, no. 1. Atlanta, GA. 2001.

22. US Centers for Disease Control and Prevention. HIV AIDS Surveillance Report, vol. 14. 2003.

23. Merrill CT, Elixhauser A. Hospitalization in the United States, 2002: HCUP Fact Book No. 6. AHRQ Publication No. 05-0056, June 2005. Rockville, MD: Agency for Healthcare Research and Quality. Available at: http://www.ahrq.gov/data/hcup/factbk6/. Accessed November 13, 2007.

24. Steiner C, Elixhauser A, Schnaier J. The Healthcare Cost and Utilization Project: an overview. Eff Clin Pract. 2002;5:143-151.

25. Fasciano NJ, Cherlow AL, Turner BJ, et al. Profile of Medicare beneficiaries with AIDS: application of AIDS casefinding algorithm. Health Care Financ Rev. 1998;19:19-38.

26. Rosenblum L, Buehler JW, Morgan MW, et al. HIV infection in hospitalized patients and medicaid enrollees: the accuracy of medical record coding. Am J Public Health. 1993;83:1457-1459.

27. Liang K-Y, Zeger SL. Longitudinal data analysis using generalized linear models. Biometrika. 1986;73:13-22.

28. Elixhauser A, Steiner C, Harris DR, et al. Comorbidity measures for use with administrative data. Med Care. 1998;36:8-27.

29. Greene WH. Econometric Analysis. 2nd ed. New York: Macmillan Publishing Company; 1993.

30. Kolb B. Brain Plasticity and Behavior. Hillsdale, NJ: Lawrence Erlbaum; 1995.

31. Department of Health, State of New York. AIDS in New York State. 2001-2002 ed. Albany, NY: State of New York, Department of Health; 2006.

32. Kelly JJ, Chu SY, Buehler JW, and the AIDS Mortality Project Group. AIDS deaths shift from hospital to home. Am J Public Health. 1993;83:1433-1437.

33. Bennett CL, Cassileth BR, DeHovitz JA, et al. Score cards for AIDS care. J Acquir Immune Defic Syndr Hum Retrovirol. 1995;9:286-290.

34. Bennett CL, Deneffe D. Does experience improve hospital performance in treating patients with AIDS? Health Policy (New York). 1993;24:35-43.

35. Kelly JV, Hellinger FJ. Heart disease and hospital deaths: an empirical study. Health Serv Res. 1987;22:369-395.

Cited By:

This article has been cited 1 time(s).

Health Policy
Determinants of length of stay for psychiatric inpatients: Analysis of a national database covering the entire Korean elderly population
Chung, W; Oh, SM; Suh, T; Lee, YM; Oh, BH; Yoon, CW
Health Policy, 94(2): 120-128.
10.1016/j.healthpol.2009.09.004
CrossRef
Back to Top | Article Outline
Keywords:

hospital care; hospital experience; hospital mortality; physician experience

© 2008 Lippincott Williams & Wilkins, Inc.

Login

Search for Similar Articles
You may search for similar articles that contain these same keywords or you may modify the keyword list to augment your search.