|
Are HMO Enrollees Healthier Than Others? Results From The Community Tracking Study
This analysis addresses the question of biased selection into health maintenance organizations (HMOs) by using recent, nationally representative data from the Community Tracking Study (CTS) to compare the health status of nonelderly privately insured persons enrolled in HMO and non-HMO plans. Contrary to the conventional view that HMOs receive favorable selection, we find among the privately insured that HMO enrollees are not healthier and may be slightly less healthy. To help understand that result, we present evidence suggesting that other factors, including cost considerations, may be more important than health when people are deciding whether to enroll in an HMO.
Many point to the lower premiums often associated with health maintenance organizations (HMOs) to suggest that HMOs offer a way to help control rising U.S. health care costs. However, HMOs’ ability to control costs depends on how they achieve lower premiums. To the extent that their cost advantage is the result of favorable selection in enrollment rather than efficiencies in health care delivery and lower payments to providers, their potential to serve as a remedy for high health care spending is limited.1 The conventional view, arising perhaps from a subset of prior studies or from assumptions about the preferences of people in poor health, seems to be that HMOs do enroll a generally lower-risk population. Although considerable evidence suggests that this is true for the Medicare and Medicaid populations, there is only mixed evidence for the privately insured—the majority of the insured population.2 This analysis investigates the question of biased selection in HMOs by using recent, nationally representative data from the Community Tracking Study (CTS) to compare the health status of nonelderly privately insured persons enrolled in HMO and non-HMO plans.3 Contrary to the conventional view about biased selection in HMOs overall, we do not find among the privately insured that HMO enrollees tend to be healthier than non-HMO enrollees. Background and framework. We use the term biased selection to indicate that one type of health insurance plan enrolls persons who on average are better or worse risks than enrollees in other types of plans. In empirical analyses of biased selection, enrollee risk is typically approximated using a measure highly correlated with risk, such as health status or health care use. Despite the conventional view that HMOs receive favorable selection, there is no clear theoretical prediction about whether this actually occurs. For example, in the hypothetical situation where everyone has a choice between HMO and non-HMO options, people who are in worse health, and who therefore expect to have many visits to health care providers, may prefer HMO plans because of their typically lower per visit cost sharing.4 On the other hand, people in worse health may place greater weight on maintaining established ties to particular providers, and so they may prefer non-HMO plans, which impose fewer or no restrictions on provider visits. In addition, people in poor health may tend to avoid HMOs because of commonly held beliefs that managed care plans provide worse quality of care.5 Thus, either favorable or adverse selection may arise depending on how people in poor health weigh lower cost sharing in HMOs against their restrictiveness and perhaps perceived lower quality. Biased selection could also arise from correlations between health status and other characteristics that are related to preferences between HMO and non-HMO plans. For example, older people may be less familiar with managed care or more likely to have long-standing relationships with providers. Since older people also tend to be in worse health, non-HMOs could receive adverse selection simply because their method of care delivery is especially appealing to this group. Similarly, people with lower incomes may prefer HMOs because of the lower premiums and cost sharing, and therefore HMOs could receive adverse selection because of the fact that lower income is often associated with worse health.6 Although it is commonly assumed that biased selection results from consumers’ choices among plans, not everyone has the same opportunity to exercise such a choice.7 To address this, we investigate health status differences for three distinct groups of privately insured persons with different plan-choice opportunities: those in the individual market, those with employer-sponsored insurance and a choice of HMO and non-HMO plans, and those with employer-sponsored insurance but no choice. Biased selection is least likely to be observed in this last group, although it may still occur if workers select jobs on the basis of the type of health plan offered or if employers choose to offer HMO or non-HMO plans based in part on workers’ health status. Although this is not a formal analysis of consumer or employer decision making, separate analysis of these three groups allows us a fuller understanding of the factors behind the overall health status differences between HMO and non-HMO enrollees. Previous research. There is a large literature that examines biased selection in HMOs, and others have already reviewed this literature in detail.8 Conventional wisdom that HMOs experience favorable selection may be based on results for just the Medicare and Medicaid populations, which consistently show that HMO enrollees are lower risks.9 That view could also have developed from the results of a subset of the studies of the privately insured, as discussed below. Many of the studies that focus on nonelderly persons with private insurance, which is the population of interest for this analysis, use samples of employees, typically at a single large employer, who have a choice of different types of plans. Most of these studies analyze measures of health care use (from either before or after the enrollment choice was made, and sometimes represented by expenditures), and they find that people with lower use tend to enroll in HMOs.10 However, those results cannot necessarily be used to conclude that HMOs receive favorable selection, for several reasons. First, the relationship between low utilization and choice of an HMO plan is not universal.11 In addition, many of these studies rely on measures of health care use in the period prior to choosing a health plan, and there is evidence suggesting that such measures may tend to overstate the differences in use between the plan types in subsequent periods because of regression to the mean.12 Lastly, because these studies use samples that come from only a few employers, the ability to generalize from the results is limited. Another common approach to this topic is to use nationally representative survey data and measures of health status.13 The studies using this approach find few statistically significant health status differences between enrollees in HMOs and non-HMOs, and those differences go in both directions, with some health status measures indicating that HMOs receive favorable selection and others indicating adverse selection. Our analysis is most similar to this set of studies.
Data source. The data come from the Community Tracking Study (CTS) 1998–99 Household Survey, a large nationally representative survey of the civilian noninstitutionalized population that tracks changes in the health care system through periodic surveys and site visits.14 The survey collected data on approximately 59,000 persons between August 1998 and October 1999, using telephone interviews that included questions on health insurance coverage, sociodemographic characteristics, and health status.15 Population groups. This analysis uses a subsample of about 34,900 nonelderly persons with private health insurance coverage, of whom 54 percent were enrolled in HMOs and 46 percent in other types of plans. HMO enrollment is defined by the response to a question about whether the plan is an HMO.16 Thus, the sample of HMO enrollees is likely to include enrollees in point-of-service (POS) products, since these are typically marketed as HMOs with a POS option. These enrollment percentages are in line with results from other surveys, which show approximately 50 percent combined enrollment in POS and traditional HMO products.17 We analyze results separately for the following three groups: people with employer-sponsored insurance who had a choice between an HMO and a non-HMO (45 percent of the privately insured), people with employer-sponsored insurance but no choice between plan types (47 percent), and people with nongroup insurance (7 percent). To define whether an HMO/non-HMO choice was available for each person, information is used on the health insurance plans offered by the employers of all adults in the person’s family. Health status measures. We used four health status measures to ensure that our results are robust. Each is an indicator of whether an enrollee’s health status is below a certain threshold.18 Two indicators, overall poor physical health and poor mental health, are based on the physical and mental health summary scores from the SF-12, which are available only for adults.19 Poor physical health is identified by a physical health summary score below the twentieth percentile for the general U.S. population, and poor mental health is defined analogously. The third measure indicates whether the respondent’s self-assessed health status is fair or poor (as opposed to excellent, very good, or good). The final measure is the presence of any chronic conditions that are considered expensive to treat.20 We first present results for the prevalence of poor health status among HMO and non-HMO enrollees. Then, because health status is correlated with many basic sociodemographic variables, we use logit models to estimate the independent effect of health on the probability of a person’s being enrolled in an HMO.21 We show the marginal effects from those logit models for the sample with employer-sponsored insurance and a choice between HMO and non-HMO plan types. Lastly, we present HMO/non-HMO differences in attitudes about trading off lower out-of-pocket costs for health care against limitations on choice of providers. Data limitations. Two data limitations should be noted. The first is that we are evaluating biased selection using measures of health status, which is only an indirect measure of risk and propensity to use services. The second is that these results may be affected by misreporting of plan type by household survey respondents.22 However, such misreporting would affect our results only if its direction varied by health status, and our analysis of additional data in which plan type is identified by health plans indicates that misreporting has not affected our results.23
Health status and plan type. Among the privately insured, HMOs do not receive favorable selection (Exhibit 1  ). Although the differences are not large, HMOs have a higher percentage of enrollees in poor health for three of the four health status measures. The recent growth in HMOs with a POS option providing at least some out-of-network coverage raises the possibility that favorable selection in traditional HMOs is being obscured by adverse selection in the less-restrictive POS products. Although the data used in this analysis cannot address that question, a supplement to an earlier year of these data can, and it shows no evidence of health status differences between POS-model and traditional HMOs.24
Exhibit 1  also examines whether the small health status differences overall are masking any strong HMO/non-HMO health status patterns within any of our three subsamples. We might expect biased selection primarily among the sub-samples of individually insured and people with employer coverage and a choice of plan types, because for them plan type can most directly reflect enrollees’ preferences. In fact, none of the health status differences for those two subsamples is statistically significant. The remaining subsample of those with employer-sponsored insurance but no choice of plan types does have significant differences, but there is no compelling interpretation for that, because only two differences are significant and the magnitudes are not large. Therefore, at least according to these simple descriptive results, it appears that enrollees’ health status is not strongly related to choice of plans. Health status and plan type, controlling for other factors. To explore these descriptive results, we use multivariate analysis to estimate the independent effect of health on the probability of HMO enrollment for the full sample of adults, controlling for various sociodemographic characteristics (results not shown).25 Consistent with the descriptive results, we find that poor health has a small positive effect on the probability of HMO enrollment, although the effect is significant at p<.05 for only two of the four health measures.
Because the full sample combines persons who face very different types of health insurance choices, we reestimated the models on just the subsample of adults with employer-sponsored coverage and a choice between HMO and non-HMO plans, to better assess the factors most important in the decision about whether to enroll in an HMO (Exhibit 2
Health status did not have a significant effect on the likelihood of HMO enrollment for those with a choice of plan types. However, other factors did. For example, age had a significant negative effect, which could reflect the fact that older people are less familiar with HMOs. Income also had a significant negative effect, which supports the notion that those with lower incomes are attracted to HMOs’ lower cost-sharing requirements. Overall, for this sample, the lack of large significant effects of any of the measures of health status, combined with the significant effects of income, suggests that nonelderly people with private insurance put more weight on cost considerations than on health status when choosing coverage.
We also tried including interactions between the health and income variables in the basic model shown in Exhibit 2
Choice of providers versus lower costs.
Exhibit 3
Our analysis has found little biased selection in HMOs among the privately insured, with HMOs tending to have slightly more enrollees in poor health. This result is contrary to the conventional wisdom that HMOs receive favorable selection, although it is generally consistent with other similar studies. To the extent that health status represents risk, these studies in aggregate, based on data spanning 1977 through 1998, suggest that HMOs have not been receiving favorable selection among their nonelderly enrollees for quite some time, if in fact they ever did. These results make a noteworthy contribution to the literature on biased selection in HMOs. First, our measure of choice captures all employer coverage available through the family instead of just what is available from a person’s own employer, and we use this information to identify that segment of the population among which biased selection is most likely to be observed. In addition, we have direct information on attitudes about trading off more limited provider choice for lower out-of-pocket spending, which confirms our multivariate results indicating that cost plays an important role in the HMO enrollment decision. Lastly, the use of timely data in this analysis is important, since many recent developments in the health insurance market could have affected biased selection among the privately insured, such as the growth in HMO enrollment, changes in employers’ health insurance offerings, and changes in managed care products.26 Our results imply that, at least for the privately insured, lower HMO costs cannot be attributed simply to enrollment of a healthier population. Rather, at least some of HMOs’ cost advantages are likely attributable to lower utilization associated with care management tools or to lower fees paid to providers. Our results also suggest that factors other than health status play a role in people’s decisions about whether to enroll in an HMO. In particular, our finding of income’s strong negative effect on HMO enrollment decisions suggests that people are influenced by HMOs’ lower costs. This is reinforced by our finding that HMO enrollees express a much greater willingness to benefit from lower out-of-pocket costs in exchange for having restrictions on their choice of providers. Age also appears to be important in HMO enrollment, most likely reflecting older people’s lack of familiarity with HMOs. Together these results suggest that HMOs’ ability to attract better risks because of older people’s unfamiliarity with them is offset by their appeal to those with lower incomes, who tend to be worse risks. The effectiveness of HMOs in reducing costs for consumers and the importance of cost to consumers in their health insurance decisions have particular salience today as we return to a period of large increases in health care costs among the privately insured.27 In the past few years a strong economy meant that labor markets were tight, insulating employees from premium increases through employers’ willingness to absorb those increases to keep and attract workers. During that period, consumers’ and policymakers’ complaints focused on the restrictions imposed by managed care plans, although as our data show, some people nevertheless still continued to choose HMOs because of their lower costs. However, we are now seeing a weaker economy and looser labor markets, along with larger increases in premiums. In this environment employers are far more likely to pass on cost increases to their employees. As the privately insured once again face large premium increases, our results suggest that cost may return to the forefront among the complaints of privately insured consumers. This possibility should not be forgotten in discussions about managed care regulations that may greatly increase costs or restrict consumers’ choice of health plan features. If the ultimate effect of such regulations is to reduce or eliminate some lower-cost health insurance options, they may not be entirely welcome to all consumers.
Elizabeth Schaefer is a health research analyst and James Reschovsky is a senior health researcher at the Center for Studying Health System Change, in Washington, D.C. The Community Tracking Study is funded by the Robert Wood Johnson Foundation. The authors thank Leif Karell of Social and Scientific Systems for excellent programming assistance and Paul Ginsburg, Ann Greiner, Joy Grossman, Jack Hadley, Lee Hargraves, and Peter Kemper for helpful comments on an earlier version of this paper.
This article has been cited by other articles:
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Top
Data And Methods
Study Results
