Objective: To determine whether changes in health maintenance organization (HMO) penetration or payer mix affected inhospital mortality and treatment patterns of patients with acute myocardial infarction (AMI).

Study Design: Observational study using patient-level logistic regression analysis and hospital and year fixed effects of data from the Agency for Healthcare Research and Quality's Healthcare Cost and Utilization Project Nationwide Inpatient Sample, a geographically diverse sample of 20% of the hospitalized patients in the United States.

Patients and Methods: Discharges of patients (n = 340 064) with a primary diagnosis of acute myocardial infarction who were treated in general medical or surgical hospitals that contributed at least 2 years of data to the HealthCare Cost and Utilization Project Nationwide Inpatient Sample from 1989 to 1996. In-hospital mortality and rates of cardiac catheterization, angioplasty, or coronary artery bypass grafting for Medicare patients or non-Medicare patients were the main outcome measures.

Results: Among Medicare patients, increases in HMO penetration were associated with reduced odds of receiving cardiac catheterization, angioplasty, or coronary artery bypass grafting of 3% to 16%, but were not associated with any change in mortality risk. Increases in the number of HMOs within a metropolitan statistical area, our measure of HMO competition, were associated with small but significant increases in the odds of cardiac catheterization and angioplasty of about 2%. There was no pattern of changes in cardiac procedure rates or in-hospitality mortality among non-Medicare patients.

Conclusion: Increases in HMO penetration reduced cardiac procedure rates by statistically significant but small amounts among Medicare patients with AMI, without affecting mortality rates.

(Am J Manag Care. 2004;10:505-512)

The growth of managed care as a mechanism to lower healthcare costs has led to concerns that this may adversely affect quality of care. Relative reductions in payment may affect the quality of care because hospitals have fewer resources with which to provide services. The overall effect on patient care of declines in reimbursement by a particular insurer comprises 2 phenomena: a direct effect on the care of patients with that particular insurance, who may receive fewer beneficial services, and an indirect or spillover effect on other patients within the same hospital, whose care is altered because of hospital-wide resource constraints.

Direct comparisons of quality of care between health maintenance organization (HMO) and fee-for-service (FFS) patients have produced mixed results.^1-3 However, commercial HMO penetration has been shown to have spillover effects on the cost of care among Medicare FFS patients, as lower costs among Medicare patients have been observed in areas with high commercial HMO penetration.^4,5 In addition, higher HMO and managed care (HMO or preferred provider organization) penetration rates have been shown to be associated with reduced access to care,³ less use of costly diagnostic testing,⁶ and reductions in hospital cost growth.^7-9 The effect of HMOs on quality of care for hospitalized patients could be mediated by several mechanisms besides reduced availability of resources. Growth in managed care could bring about changes in physician practice patterns, system-wide availability of resources and new technology, and changes in the mix of services available to favor more cost-effective technologies and services.¹⁰

The quality effects for non-HMO patients of the observed spillover effects of HMOs on hospital costs are unknown. While reductions in hospital cost growth could lead to fewer resources being available for patient care and to worse outcomes, HMOs have greater ability to shop for better quality providers than individuals. Two cross-sectional analyses have examined the association between HMO penetration and quality of care received by Medicare FFS patients. Heidenreich et al¹¹ used data from 1994 to 1995 on Medicare FFS patients with acute myocardial infarction (AMI) in a cross-sectional analysis of geographic areas with different levels of HMO penetration and found that higher HMO penetration was associated with higher use of beneficial medications such as β-blockers and aspirin and lower rates of coronary angiography, but was not associated with AMI mortality rates. Mukamel et al¹² examined 1990 data for Medicare patients and found a statistically significant but small reduction in 30-day mortality risk of roughly 0.15% among patients in the highest (>24.5%) HMO market penetration group relative to those in the lowest (<10.6%) HMO penetration group.

The cross-sectional design of the Heidenreich and Mukamel studies makes inferences about causality problematic. Explanations such as a tendency by HMOs to locate in higher cost markets because of greater potential for cost savings may explain the observed associations, as higher cost markets may also have better outcomes.¹³ In addition, both of the studies focus only on the Medicare population, and the effect of managed care on patient quality may have been more significant in hospitals with greater proportions of patients younger than 65 years, particularly if a significant portion of patients is part of a vulnerable population group, such as the uninsured. The role of HMO competition in this context is also important to explore. Previous work has demonstrated that, while HMO profits have been shown to be lower in markets with more competition, HMO market penetration was not independently associated with profit rates.¹⁴ This suggests that competition between HMOs may limit the ability of HMOs to reduce prices paid to hospitals and might also reduce the degree of cost reductions and the potential effect on quality of care.

In this analysis, we use data from 1989 through 1996 from the Agency for Healthcare Research and Quality's Healthcare Cost and Utilization Project Nationwide Inpatient Sample (HCUP NIS) to examine the association between changes in HMO penetration and changes in outcomes and treatment patterns for all AMI patients. Using discharges from repeated cross-sectional samples of hospitals makes the selection bias inherent in single cross-sectional studies less problematic and allows us to control for baseline quality differences among patients in different hospital markets to examine how outcomes and treatment patterns changed over time in association with changes in HMO penetration. In contrast to previous work, we studied not only Medicare patients but also non-Medicare patients younger than 65 years, because the effects of price competition and reduced reimbursement may be most direct in this group.

While it is important to study Medicare and non-Medicare patients, there is some evidence that within a given marketplace outcomes and treatment patterns may be similar for all patients, although the price being paid for similar services may vary widely.¹⁵ For this reason, we focus on intertemporal changes in mortality and cardiac procedure use for all patients within each of our 2 main groupings of patients (the Medicare and non-Medicare groups), as opposed to looking at changes in these measures for patients with specific insurance types. In addition, we examine the role of HMO market concentration to determine whether the effects observed are a function of the competitiveness of HMO markets. We adjust for changes in the proportion of uninsured, Medicaid, and Medicare patients, as reimbursement for these payers has been low relative to FFS indemnity; therefore, increases in the proportion of these low-paying payers may also have affected the process and outcomes of care delivered within hospitals.

METHODS

Data Sources

Discharges with AMI were chosen for this study because AMIs are common, the treatment objective generally is to ensure survival, and the quality of care affects the likelihood of survival.^16,17 In addition, all patients with recognized AMI are admitted on an emergency basis, usually to the nearest hospital. Therefore, there is little opportunity for selection bias to affect the decision about who gets admitted or risk-adjusted mortality over time.

The database for this study was constructed from 3 primary sources: the HCUP NIS, the Area Resource File, and the HMO Geographical Enrollment Data file (D. R. Wholey, PhD, Division of Health Services Research and Policy, University of Minnesota, unpublished data, 2001). We used data from 1989 to 1996 and included individuals who had a principal discharge diagnosis of AMI (International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9-CM] code 410.XX), excluding subsequent care (code 410.X2) (807 743 observations). These codes were chosen to focus on new AMI admissions, using the same approach taken by the California Hospital Outcomes Project.¹⁸ We excluded patients who were not treated in hospitals located within metropolitan statistical areas (MSAs) (n = 100 701), as our HMO data were limited to patients in MSAs; patients with a coded primary diagnosis of AMI who were discharged alive after a length of stay of less than 2 days (n = 32 110), because it is likely that these patients did not experience an AMI; individuals younger than 18 years and older than 85 years (n = 47 483); patients with a length of stay greater than 30 days (n = 9105); patients for whom neither discharge alive or dead was coded (n = 636); patients treated in governmental hospitals (n = 58 210); and patients with age missing (n = 4071), as age is an important risk adjuster. Because we were not able to determine their ultimate outcome, we excluded individuals who were transferred to another hospital (52 591 observations). In addition, we excluded patients who went to hospitals at which uninsurance rates, Medicare rates, or Medicaid rates were coded as 0% for any year, on the basis that the insurance variable was likely miscoded (n = 122 359). To use hospital fixed effects to control for unobserved heterogeneity among hospitals, we excluded discharged patients who were treated in hospitals for which we had only a single year of data (n = 40 413). The final sample of this unbalanced panel consisted of 340 064 discharge observations from 457 hospitals from 1989 to 1996.

The Area Resource File was used to link the individual patient-level data records from the NIS with MSA codes. The HMO Geographical Enrollment Data file provided the commercial HMO penetration and the number of HMOs within each MSA for each year being studied.

Measure of HMO Penetration and Payer Mix

We used an MSA-level measure of HMO penetration and the number of HMOs within each MSA as a measure of HMO competition. We used an MSA-level measure rather than hospital-level measures because it is less likely to be endogenous, as HMOs might choose to contract with better quality hospitals but would be unlikely to make selective contracting decisions on the market level based on differing levels of quality. Using an MSA-level measure of HMO penetration also protects against potential hospital-level selection bias, which could result from significant changes in admission patterns of patients, with different-than-average AMI mortality in conjunction with increases in HMO penetration.

Each discharge was grouped into 1 of 4 quartiles based on the market-level HMO penetration for that hospital in that year: HMO penetration between 0% and less than 9.3% (quartile 1), HMO penetration between 9.3% and 14.5% (quartile 2), HMO penetration between 14.6% and 21.4% (quartile 3), and HMO penetration greater than 21.4% (quartile 4).

We calculated the percentage of uninsured, Medicare, and Medicaid patients in each hospital per year in the sample using the NIS data.

Measurement of In-hospital Mortality and Treatment Patterns

The principal outcome measure was death that occurred during the initial hospitalization, provided the length of stay was less than or equal to 30 days. To examine changes in treatment patterns, we examined cardiac catheterizations (ICD-9-CM codes 37.21-37.23 and 88.55-88.57), coronary artery bypass graft (CABG) procedures (codes 36.11-36.19), and coronary angioplasties or stents (codes 36.01, 36.02, 36.05, and 36.06) during the first 30 days of the initial hospitalization.

Risk Adjustment

The method for risk adjustment developed by Elixhauser and colleagues¹⁹ at the Agency for Healthcare Research and Quality was used, as this method has tested well in terms of superior discrimination vs other risk adjustment approaches for use with administrative data.²⁰ Variables in the risk adjustment algorithm include age, sex, peripheral vascular disease, paralysis, other neurological disorders, chronic pulmonary disease, diabetes mellitus, diabetes mellitus with chronic complications, hypothyroidism, renal failure, liver disease, peptic ulcer disease, AIDS, lymphoma, metastatic cancer, solid tumor without metastasis, rheumatoid arthritis, coagulopathy, obesity, weight loss, fluid and electrolyte disorders, chronic blood loss anemia, deficiency anemias, alcohol abuse, drug abuse, pyschoses, depression, and hypertension.

Statistical Analysis

We used patient-level logistic regressions with hospital dummy variables to control for unobserved heterogeneity among hospitals, and year dummy variables to control for intertemporal changes common to all hospitals in the sample. Hospital-level dummy variables are included because, even with the inclusion of many control variables, there may be unobservable characteristics of areas that are not included but that are correlated with HMO market share and treatment patterns or in-hospital mortality.

To test for spillover effects, primarily whether changes in the proportion of patients with HMO insurance affected the outcomes and treatment patterns of patients with Medicare insurance, the sample was divided into 2 subsamples: Medicare and non-Medicare patients. We examined treatment patterns and outcomes at the discharge level in both subsamples. The dependent variables were mortality and 3 cardiac procedures: catheterization, angioplasty, and coronary artery bypass graft (CABG). In this way, by examining the Medicare subsample, we were able to test whether there was a spillover effect from changes in HMO penetration, Medicaid, or the percentage of uninsured on the outcomes and treatment patterns of the Medicare population. We examined the non-Medicare patient subsample separately, as nearly all studies of HMO spillovers have examined Medicare patients only, and this allowed us to examine whether there were changes in outcomes and treatment patterns based on changes in payer mix within the non-Medicare population.

Patient-level logistic regression analysis was used to examine the relationship between changes in HMO penetration, HMO market concentration, and the outcome and treatments of interest. To control for baseline quality differences among the hospitals, hospital fixed-effects dummy variables were used. Through the use of hospital fixed effects, the outcome (death or survival) is interpreted as a function of within-hospital changes in payer mix, as variation in time-invariant hospital characteristics is accounted for by the hospital fixed-effects dummy variables. Because we used quartiles of HMO penetration to measure HMO penetration and because HMO penetration increased in every market during the study, the estimated regression coefficients examine the effect of interquartile increases in HMO market share on treatment patterns and in-hospital mortality. Patient characteristics such as age, sex, and clinical risk factors were used as controls. Coronary artery bypass graft surgery, angioplasty, and cardiac catheterization were modeled as dependent variables in a fashion similar to that used for mortality. We also controlled for intertemporal changes common to all hospitals in the sample by using dummy variables for each year.

To compensate for the fact that our analyses of HMO penetration were correlated at the MSA level, we adjusted our standard errors for clustering at the MSA level. Analyses were done using SAS software (SAS Institute, Cary, NC), with adjustment of standard errors for clustering done using STATA (Stata Corporation, College Station, Tex).

RESULTS

Descriptive Data

The hospitals included in the NIS changed from year to year based on the NIS sampling criteria. In 1989, our sample included 180 hospitals, and in 1996, our sample included 238 hospitals. Our total sample included 457 hospitals that contributed data for at least 2 years, which enabled the use of hospital fixed effects. Mean HMO penetration increased from 18.8% in 1989 to 27.5% in 1996. There were no large changes in the mean number of HMOs within each MSA. The percentage of uninsured decreased from 6.0% to 4.1%. Changes in payer mix for each of the other classes of payers were small (Table 1). Overall, in-hospital mortality for AMI patients declined from 14.6% to 9.3% during this period, while cardiac catheterization, angioplasty, and CABG rates rose substantially. Hospitals in the sample in 1989 and hospitals in the sample in 1996 were not necessarily the same hospitals.

Measurement of the HMO effect on changes in outcomes and treatments over time has the potential to be biased if there are underlying unmeasured differences in areas that attracted HMOs to a greater degree than others. For instance, HMO penetration might be more likely to increase in areas in which the population is younger or healthier. To investigate whether there were systematic differences between markets in which HMO penetration increased or did not, we compared observed patient characteristics in these 2 types of markets (Table 2). There were 96 124 patients in hospitals that did not change their quartile of HMO penetration during the study. There were 243 940 patients in hospitals that experienced interquartile change in HMO penetration. Although there are several statistically significant differences in the proportion of patients with these comorbidities between these 2 groups, the magnitude of the differences in all cases is small and does not favor one group or the other.

Regression Analysis Results

We present results separately for the Medicare patients and the non-Medicare patients.

Medicare Patients. Among Medicare patients, we found extensive evidence for spillover effects from increases in HMO penetration, particularly in terms of changes in treatment patterns. Odds ratios (ORs) that were significantly lower than 1 were found with each interquartile increase of HMO penetration for cardiac catheterization and angioplasty and for patients in hospital markets in which HMO penetration increased into the highest quartile for CABG (OR, 0.84; 95% confidence interval [CI], 0.74-0.95) (Table 3). For each of these procedures, the OR stayed the same or decreased with increasing levels of HMO penetration, such that the ORs were lowest in hospital markets in which HMO penetration increased into the highest quartile. The magnitude of these coefficients means that, for every interquartile increase in HMO penetration (roughly 5%-7%, depending on the quartile), the odds of a Medicare AMI patient receiving cardiac catheterization decreased by 12% to 17%, the odds of receiving an angioplasty decreased by 7% to 12%, and the odds of receiving a CABG decreased by 3% to 16%.

There was no evidence that increases in HMO penetration had a significant effect on inpatient mortality, as patients in hospitals that experienced increases from the first quartile to the second quartile of HMO penetration had a slightly increased odds of mortality (OR, 1.07; 95% CI, 1.01-1.13), but patients in hospitals in which HMO penetration increased into the third or fourth quartile did not experience an OR of mortality significantly different than 1. Increases in the number of HMOs within an MSA, our measure of HMO competition, were associated with small but significant increases in cardiac catheterization (OR, 1.03; 95% CI, 1.01-1.04) and angioplasty (OR, 1.02; 95% CI, 1.00-1.04) rates. There were no significant effects of the change in the number of HMOs on mortality or CABG rates.

Increases in the hospital's proportion of uninsured or Medicaid patients had no statistically significant effect on the mortality risk for Medicare patients or the odds of receiving cardiac catheterization, angioplasty, or CABG. Increases in the percentage of Medicare patients were associated with reductions in the odds of mortality (OR, 0.71; 95% CI, 0.53-0.96) and in the odds of bypass surgery (OR, 0.50; 95% CI, 0.25-0.99), but had no significant effect on the odds of cardiac catheterization or angioplasty.

Using fixed effects to control for unobserved heterogeneity in a logistic regression analysis may result in biased coefficient estimates in cases in which there are few events per hospital. As a test for robustness and to mitigate the potential for bias, the regression for each dependent variable was also run using a data subsample limited to patients in hospitals with 10 or more occurrences of each dependent variable. We found no evidence of biased coefficients in any of the specified regression models.

Non-Medicare Patients. We found less evidence of changes in treatment patterns and in-hospital mortality due to changes in HMO penetration in the non-Medicare sample. Interquartile changes in HMO penetration had no significant effect on mortality, cardiac catheterization, or CABG rates. Increases into the third quartile of HMO penetration were associated with a reduced odds of angioplasty (OR, 0.90; 95% CI, 0.82-0.99), but increases into the second or fourth quartile of HMO penetration were not associated with ORs of angioplasty that were significantly different than 1. Increases in the number of HMOs within an MSA, our measure of HMO competition, were associated with a small but statistically significant increase in angioplasty rates (OR, 1.02; 95% CI, 1.01-1.04), but no significant change in the odds of mortality, cardiac catheterization, or bypass surgery (Table 4).

Increases in the percentage of uninsured had no significant effect on the odds of mortality, cardiac catheterization, or angioplasty, but were associated with a significant reduction in the odds of all non-Medicare patients receiving bypass surgery (OR, 0.17; 95% CI, 0.03-0.84). Increases in the percentage of Medicaid patients were also associated with significant reductions in the odds of bypass surgery (OR, 0.15; 95% CI, 0.05-0.49), but had no significant effect on the odds of mortality, cardiac catheterization, or angioplasty. The reduction in the odds of receiving bypass surgery from a 1% increase in the percentage of uninsured or Medicaid patients is large, but given that the baseline rates of each of these is only about 4%, a 1% increase represents a potentially substantial additional financial burden. Changes in the percentage of Medicare patients had no significant effects on any of these measures within the non-Medicare population.

There was no evidence of bias in any of the study variable coefficients in subsamples of the data that were limited to patients in hospitals with 10 or more occurrences of the dependent variable.

DISCUSSION

Our results indicate that increases in HMO penetration from 1989 to 1996 significantly affected treatment patterns for Medicare AMI patients within the HCUP NIS, and these effects were largest in the markets in which HMO penetration increased into the highest quartile. The relative risk of receiving cardiac procedures was reduced by about 10% to 15% at hospitals with increases in HMO penetration into 1 of the 3 upper quartiles. Given the rates at which patients received these procedures, the magnitude of this relative decrease is about 1% to 2% for CABG, 2% to 3% for angioplasty, and 5% to 7% for cardiac catheterization. We found no evidence of any increased mortality risk from increases in HMO penetration. Increases in the number of HMOs, suggesting greater HMO competition, were associated with modest increases of approximately 2% in the rates of cardiac catheterization (eg, 1.0%, given a baseline rate of about 50%) and angioplasty (0.5%) after AMI. We believe that these estimates are unbiased by cross-sectional differences in case-mix or hospital characteristics, because we used a measurement framework that provides estimates of within-hospital changes in response to different degrees of change in MSA-level HMO penetration.

In the non-Medicare population, we found little evidence that treatment patterns or in-hospital mortality was affected by increases in HMO penetration. However, increases in the proportion of uninsured patients and Medicaid patients were associated with significant reductions in the odds of receiving bypass surgery for all non-Medicare patients. Because care for such patients (to the degree that Medicaid reimbursement does not cover marginal costs) is funded out of hospital margins, these results are not surprising.

Our results on treatment patterns and mortality in the Medicare population are similar to those found in the study by Heidenreich et al,¹¹ who showed that higher rates of HMO penetration were associated with lower rates of coronary angiography (relative risk, 0.93; 95% CI, 0.86-1.01) but no significant differences in AMI mortality. We did not observe reductions in mortality, as Mukamel and colleagues¹² did (a small reduction of 0.15%). Reductions in cardiac procedure use are consistent with the literature that has demonstrated reductions in expenditures in areas with bigger increases in HMO penetration.^4,5,7-10

Given that there were no changes in mortality risk as a result of changes in HMO penetration, the ramifications of the observed changes in treatment patterns for quality of care are difficult to assess. From the administrative data contained in the NIS, we cannot determine the appropriateness of the treatments that individual AMI patients received. Therefore, we do not know whether the changes in treatment patterns are of a beneficial or detrimental nature, as it depends on whether the procedures received were appropriate and necessary, or not.

An important limitation of this study is that we studied only patients with AMI; therefore, we cannot make inferences about how outcomes and treatment patterns for other conditions were affected by changes in HMO penetration. We were limited to in-hospital measures of mortality, but the mean length of stay for all AMI patients decreased to a similar degree in hospitals in the areas with the biggest and smallest increases in HMO penetration, so these measures would not have been biased by differential changes in length of stay. In addition, some investigations focused on quality of care have found results robust to using inpatient deaths or deaths within 30 days of admission as measures of quality.²¹ The HCUP NIS did not code whether patients were HMO patients; therefore, we could not directly analyze spillover effects among non-HMO patients younger than 65 years or among Medicare patients who were not HMO enrollees. However, we think this was unlikely to have any significant effect on our results, as the percentage of Medicare HMO patients nationwide was small during 1989 to 1996, and Baker¹⁰ demonstrated that overall HMO market share, as opposed to Medicare HMO market share, predicts spending among Medicare patients in several different investigations. It would also be useful to estimate the managed care effects by including preferred provider organization data, but such data were not available on a national level for all MSAs at the time of the study. Furthermore, HMOs have been the managed care entities that have typically most aggressively negotiated with providers, exerted influence over treatment decisions, and attempted to affect the practice patterns of physicians.¹⁰

The observations we have made with these data are reassuring, as they suggest that increases in HMO penetration within hospitals in the HCUP NIS did not affect the availability of resources to a sufficient degree to have a significant adverse effect on AMI mortality. Further work should investigate the effects of changes in payer mix and HMO competition on quality of care for other diseases, as well as in portions of the country that have experienced large increases in the percentage of HMO patients or other changes in payer mix that could cause sufficient financial stress within hospitals to affect the quality of care.

Acknowledgments

We acknowledge helpful suggestions from seminar participants at the 2002 Association for Public Policy Analysis and Management Fall Research Conference, November 7-9, 2002, Dallas, Tex; the 2002 Annual Research Meeting of Academy Health, June 23-25, 2002, Washington, DC; the HealthCare Cost and Utilization Project Partners Meeting, March 2002, Rockville, Md, the Wharton School, and the University of Pennsylvania School of Medicine. We also acknowledge helpful suggestions from Sean Nicholson, PhD, and Irma Arispe, PhD.