Policy Implications of Risk Selection in Medicare HMOs

The Issue of Risk Selection

For more than a decade, Medicare beneficiaries have had the option to enroll in risk-contract health maintenance organizations (HMOs), where the federal payment is pegged at 95 percent of the estimated fee-for-service cost. Two persistent questions have occupied health policy researchers throughout this period: Are Medicare HMO enrollees healthier than elderly individuals who receive fee-for-service care? And if so, does the government payment rate for Medicare HMOs accurately account for the lower expected costs of a healthier population? This Issue Brief examines three recent studies that investigate the extent of biased selection in risk-contract HMOs and its cost to the Medicare program.

HMOs with Medicare risk contracts receive a fixed monthly payment for each enrollee, irrespective of actual service utilization. The adjusted average per capita cost (AAPCC) is calculated based on county-level and enrollee-specific characteristics such as gender, age, Medicaid status, and institutional status. It is widely understood that the health status of risk-contract HMO enrollees -- a factor not captured by the AAPCC formula -- ultimately determines whether the Medicare program achieves the 5 percent savings embedded in the payment mechanism, or realizes any savings at all.

Biased selection occurs when the medical care needs of risk HMO enrollees differ systematically from those of Medicare beneficiaries in the fee-for-service sector. If risk enrollees are consistently healthier and less likely to use medical care than their fee-for-service counterparts, the HMO experiences favorable selection. If, on the other hand, risk enrollees are consistently less healthy, the HMO experiences adverse selection. Whether Medicare saves or loses money on the risk-contract program hinges on the direction and magnitude of this selection bias.

Biased risk selection has created some difficult research challenges, one of the most fundamental being how to disentangle the effects of HMO care management from population differences. Utilization patterns observed in HMOs could reflect both the HMO's practice style and the characteristics of its enrollees. Consequently, researchers cannot use utilization during the period of HMO enrollment as an indicator of population differences. Instead, they must rely on alternative measures such as health status comparisons between HMO and fee-for-service enrollees, and utilization data from the period before enrollment or after disenrollment.

Are Earlier Findings Still Valid?

Research on the earlier years of the risk-contract program indicated that the characteristics built into the AAPCC were not reliable predictors of future medical costs and that Medicare HMO enrollees were healthier than fee-for-service beneficiaries. A 1992 study by Mathematica Policy Research, Inc. (MPR), for example, found that rather than saving 5 percent for each HMO enrollee, Medicare actually paid 5.7 percent more.

Some researchers questioned whether these earlier findings of favorable selection still held true for a newer cohort of Medicare HMO enrollees. HMO enrollment had more than tripled since the MPR study, growing from 1.1 million beneficiaries in January 1990 to 3.5 million in April 1996. Simultaneously, the number of plans offering risk contracts rose from 96 to 202. However, if the health status advantage of HMO enrollees relative to fee-for-service beneficiaries persisted, the financial losses to Medicare would only grow as HMO enrollment expanded.

Three new studies were presented and discussed at a policy research seminar sponsored by the Center for Studying Health System Change. Using new data, one study concluded that HMO enrollees did not differ meaningfully from their fee-for-service counterparts, while the other two found that the pattern of healthier individuals in Medicare HMOs -- as documented in the MPR study -- remained largely intact.

Price Waterhouse: No Bias Found

In a study commissioned by the American Association of Health Plans, Jack Rodgers of Price Waterhouse analyzed the 1992 Medicare Current Beneficiary Survey (MCBS) and found that the characteristics predicting medical care use were roughly equivalent for HMO enrollees and fee-for-service beneficiaries. Rodgers constructed his sample by matching each Medicare risk-contract enrollee with all fee-for-service beneficiaries in the same county who shared the demographic cost factors incorporated into the AAPCC payment formula (referred to as the risk subset). By comparing characteristics of HMO enrollees with this fee-for-service risk subset, Rodgers sought to measure differences between the two groups that were not controlled for by the AAPCC. To estimate the impact of risk-contract HMO enrollment on Medicare spending, he used Medicare costs and characteristics of all fee-for-service beneficiaries to calculate projected costs for each HMO enrollee.

Among the key findings from the Rodgers study: Rates of chronic conditions did not favor either group. HMO enrollees had a greater incidence of diabetes and stroke, while fee-for-service beneficiaries experienced higher rates of cancer and heart disease. The average incidence of activity limitations was comparable between the two populations. After adjusting for AAPCC risk factors, HMO enrollees and the fee-for-service risk subset had similar per capita costs -- predicted total risk HMO costs were just 2 percent below actual fee-for-service risk subset total costs.

Based on these 1992 data, Rodgers concluded that the average monthly cost to Medicare was approximately the same for HMO enrollees and their fee-for-service counterparts, and that Medicare saved approximately 5 percent for each enrollee in a risk HMO.

HCFA Team: HMO Payments Too High

Using 1994 MCBS data, Gerald Riley and a team of researchers at the Health Care Financing Administration (HCFA) examined the selection issue in risk-contract HMOs and reached a different conclusion: the Medicare payment formula for HMOs did not adequately adjust for the better health status -- and correspondingly lower expected costs -- of HMO enrollees. In other words, HMO payments were set too high.

Rather than using a matched sample as Rodgers had done, the HCFA study included all fee-for-service respondents living in counties with HMO respondents and controlled for AAPCC demographic factors statistically. Models incorporating various health status measures to predict expenses were used to estimate costs for both HMO and fee-for-service enrollees.

The HCFA team found that HMO enrollees had better functional status, fewer activity limitations, and were less likely to report their health as fair or poor compared to fee-for-service respondents. More fee-for-service enrollees had chronic conditions, though only the results for heart disease reached statistical significance. Using a model that incorporated demographics, self-reported conditions, functional status, and general health status, Riley found that predicted costs of HMO enrollees were considerably less than those of fee-for-service beneficiaries. The ratio of HMO costs to fee-for-service costs was 0.85. Given this finding, Riley suggested that the AAPCC was higher than warranted and would need to be lowered by about 12 percent, on average, if adjusted to reflect the better health status of HMO enrollees.

PPRC Study: Favorable Selection Confirmed

Findings from a Physician Payment Review Commission (PPRC) study also pointed to favorable selection in Medicare HMOs. According to the research conducted by Christopher Hogan and Donald F. Cox, spending by new HMO enrollees was 37 percent less than among fee-for-service beneficiaries in the six months prior to HMO enrollment. Spending for those who disenrolled from HMOs was 60 percent higher in the six months after they dropped HMO coverage compared to fee-for-service beneficiaries.

The PPRC study examined HMO enrollees' experiences prior to and after disenrolling from a risk-contract HMO using expenditure data from Medicare claims files. HMO beneficiaries who enrolled between July 1989 and June 1994 were compared against a fee-for-service control group.

Because prior-use data do not capture information during the period of HMO enrollment, the researchers examined beneficiaries' experiences during enrollment to assess the relationship between length of enrollment and costs or service use. They found that hospital use and mortality rates among new enrollees were lower than the average for all HMO enrollees during the initial years of enrollment. However, both hospital use and mortality rates rose with longer enrollment duration. This pattern suggests that HMO enrollees' costs tended to increase the longer they remained in the plan -- a regression toward the mean in beneficiaries' costs after enrolling.

The PPRC also found that the best available risk-adjustment models captured only about half of the risk selection observed among new HMO enrollees. These findings -- that preenrollment costs were much lower than risk-adjustment models would predict, and that costs regressed only slowly toward the mean -- suggested that the MPR and Riley estimates of biased selection may actually have been conservative. The real extent of biased selection could well be higher.

Which Results Are More Reliable?

Earlier findings about risk selection appeared to apply to more recent experience as well. The preponderance of evidence indicated that two key conclusions had not changed: favorable selection still existed in risk-contract HMOs, and it was large enough to produce financial losses for the Medicare program. In discussions at the HSC seminar and in a memo on risk selection released by the Congressional Budget Office (CBO) on the same day, the primary methodological criticisms focused on the Price Waterhouse study. Several issues were raised:

The database used in the Price Waterhouse study was considered by many researchers to be inadequate for analyzing the selection issue. The HMO sample in the 1992 MCBS was quite small, including only 371 respondents, whereas the HCFA study using 1994 MCBS data included 863 respondents. This made the Price Waterhouse results highly sensitive to changes in the health status of a small number of individuals.

The sample was not designed to be representative of Medicare's HMO enrollment, which compounded the small sample problem. Additionally, by using the MCBS, the study excluded all individuals who died during the first nine months of the year, meaning it failed to capture the cost impact of differences in mortality rates between HMO and fee-for-service populations -- mortality being lower in the HMO population.

Finally, the Price Waterhouse study did not account for enrollment shifts between HMOs and the fee-for-service sector during the study period, even though such shifts could have affected the average costs of both populations. Because other researchers reported very low costs among new HMO enrollees and high costs among disenrollees, ignoring enrollment changes that occurred throughout 1992 could have biased the results. In contrast, the HCFA study measured HMO membership and health status at the same point in time, avoiding this potential source of bias.

Given these research limitations, the CBO concluded that corrections for the biases in the Price Waterhouse study could more than quadruple its estimate of favorable selection.

Policy Implications

With increasing numbers of Medicare beneficiaries enrolling in HMOs and growing congressional interest in encouraging HMO enrollment, the issue of favorable selection into risk-contract HMOs remained a pressing concern for the health policy community. To the extent that Medicare was overpaying for risk-contract enrollees, financial losses to the program would only worsen as HMO enrollment grew. And if medical savings accounts or other plan types were made available to Medicare beneficiaries, those financial losses could be further compounded.

One approach to improving the accuracy of Medicare HMO payments was to revise the payment formula by developing more effective and sophisticated risk adjusters. A critical component of improving the AAPCC involved obtaining much more data from HMOs. However, HMOs needed incentives to provide such data. One suggestion was to develop an incentive system in which risk HMOs would receive higher payments in exchange for providing HCFA with utilization data on their Medicare enrollees -- or lower payments if they refused.

Because a better-calibrated AAPCC would translate to lower payments to HMOs, such a change could have adverse effects on both Medicare beneficiaries and HMOs. Many risk HMOs at the time charged zero premiums and offered benefits exceeding standard Medicare coverage. Reducing payments to risk HMOs could lead to higher premiums for enrollees and fewer benefits, making HMOs a less attractive option for the elderly. This presented a potential political obstacle to reforming the AAPCC.

Some observers questioned the fairness of reimbursing risk HMOs at only 95 percent of the AAPCC. One proposed solution was to raise the percentage from 95 to 100 at the same time the risk adjustment formula was strengthened. Establishing a level playing field was considered important if policy makers expected HMOs to compete effectively with the fee-for-service program in attracting Medicare beneficiaries.

References

Randall Brown, Jeanette Bergeron, Delores Clement, et al., The Medicare Risk Program for HMOs: Final Summary Report on Findings from the Evaluation Report to the Health Care Financing Administration. Mathematica Policy Research, Inc., Princeton, N.J., February 1993.

Jack Rodgers and Karen Smith, "Is There Biased Selection in Medicare HMOs?" Health Policy Economics Group, Price Waterhouse LLP, Washington, D.C., March 14, 1996.

Gerald Riley, Cynthia Tudor, Yen-pin Chiang, and Melvin Ingber, "Health Status of Medicare Enrollees in HMOs and Fee-for-Service in 1994." Health Care Financing Review, Vol. 17, No. 4, Summer 1996.

"Risk Selection and Risk Adjustment in Medicare." Chapter 15 in the Annual Report to Congress, Physician Payment Review Commission, Washington, D.C., 1996.

Memorandum on biased selection in Medicare HMOs, Congressional Budget Office, Washington, D.C., July 17, 1996.

Sources and Further Reading

CMS -- Medicare Program Information -- Official Medicare program information from the Centers for Medicare and Medicaid Services.

Kaiser Family Foundation -- Medicare -- Research and data on Medicare policy, spending, and enrollment.

Health Affairs -- Peer-reviewed health policy research and analysis.

Commonwealth Fund -- Research on health care system performance and Medicare policy.