In chapter 6, we considered a variety of measures that insurers could use to place people into risk classes that reflected their likely claims experi-ence. In this chapter, we look at some empirical evidence about the pre-

dictive power of alternative measures. In particular, we examine the extent to which demographic, health status, and prior utilization measures predict individual use of health services.

Three key points emerge from this discussion. First, even the most complete set of measures explains only a small proportion of the variance in an individual’s use of health services. If utilization was wholly predictable based on readily available measures, there would be no role for insurance. Instead, we would borrow and lend to even out the peaks and troughs of our spending patterns. It should not be surprising that models have limited predictive power. If our health status has a large random component to it, then by definition it is not predictable.

The second key point is that some sets of measures are better predic-tors of health services use than are others. Demographic characteristics per-form surprisingly poorly. Prior utilization is the best predictor, and various measures of health status fall somewhere in between. However, dismissing the predictive abilities of these other measures would be a mistake. The abil-ity to predict even a couple of percentage points better than others can yield a substantial competitive advantage, provided it can be done at relatively low cost.

Third, statistical modeling has its limits. The presumption in risk adjustment is that statistical methods will eliminate the least costly efforts to attract low utilizers and avoid high utilizers. This may be so. But it may be that other approaches implicitly contain more or better information on the future use of health services than those contained in the statistical models. It may be, for example, that our urban legend from chapter 5, in which a Medicare managed care plan was willing to enroll anyone who could walk up three flights of stairs, is a better predictor than the best statistical model.

Risk adjustment methodology has at least two other uses besides directly determining premiums based on expected utilization. First, the healthcare exchanges required by the Affordable Care Act (ACA) must risk-adjust the enrollment in exchange plans. Although individual consumers are not to face premiums based on their health status, the exchanges must

Copyright 2020. AUPHA/HAP Book.

All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted under U.S. or applicable copyright law.

EBSCO Publishing : eBook Collection (EBSCOhost) – printed on 1/12/2023 11:01 AM via UNIVERSITY OF MARYLAND GLOBAL CAMPUSAN: 2459482 ; Michael A. Morrisey.; Health Insurance, Third EditionAccount: s4264928.main.eds

Health Insurance122

determine whether some plans obtain a more or less healthy draw of enrollees and reduce payment to plans with disproportionately healthy enrollees while raising payments to those with less healthy subscribers. Second, proposals to convert Medicare into a “premium support” or voucher program often require that the voucher be based in part on the health status of the Medicare beneficiary (see chapter 23).

In this chapter, we consider risk adjustment measures in the context of the payment system that Medicare uses to pay Medicare Advantage plans, the managed care plans that provide care to approximately one-third of Medicare beneficiaries. Although this risk adjustment system is used by a payer rather than an insurer, it has the great advantage of being publicly available. It also highlights the key issues.

Medicare Adjusted Average per Capita Costs

Because HMOs do not have a claims database, they were at a disadvantage in participating in Medicare when it was introduced in 1965. After a number of largely unsuccessful efforts, in 1985 Medicare implemented the Adjusted Average Per Capita Costs (AAPCC) payment methodology under authoriza-tion from Congress in the Tax Equity and Fiscal Responsibility Act of 1982 (TEFRA). See Zarabozo (2000) for a history of Medicare’s approaches to paying managed care plans in its first 35 years.

Under TEFRA, Medicare essentially paid participating HMOs a fixed dollar amount for each beneficiary that chose to join the plan. Because HMOs were thought to be more efficient than traditional care providers (recall the “HMO effect” from chapter 5), the legislation prescribed that the capitated rate should be 95 percent of the average Medicare Part A (i.e., hospital) plus Part B (i.e., ambulatory) expenditures per beneficiary. As we speculated in chapter 6, claims experience likely varies by location. Congress appreciated this as well and ordered that the average expenditures be com-puted and applied for each county. These rates were then adjusted by the mix of beneficiaries the plan enrolled, taking into account their age, gender, Medicaid status, whether they were in a nursing home, and whether the ben-eficiary was an active worker with coverage through an employer. Thus, the AAPCC paid 95 percent of the county average Medicare Part A and Part B expenditures adjusted for age, gender, Medicaid, institutional, and active worker status. This method is analogous to a simple manual rating system.

As we saw in chapter 5, the Medicare payment system appears to provide Medicare Advantage plans with substantial incentives for enrolling people with lower-than-average expected claims and avoiding people with above-average claims. One government study found that early Medicare HMO enrollees had expenditures that were only 63 percent of the average

EBSCOhost – printed on 1/12/2023 11:01 AM via UNIVERSITY OF MARYLAND GLOBAL CAMPUS. All use subject to https://www.ebsco.com/terms-of-use

Chapter 7: Risk Adjustment 123

of all beneficiaries in the six months prior to joining the HMO (Physician Payment Review Commission [PPRC] 1994). If this was so, Medicare was consistently overpaying for the services provided by Medicare Advantage plans, and higher-cost beneficiaries may have been effectively denied access to a form of healthcare delivery that they may have preferred.

Improving the Adjusted Average Per Capita Costs Payment Methodology

Almost immediately, Medicare funded research to try to improve its payment system. Such research requires data on the demographic, health status, and healthcare utilization characteristics of a relatively large number of heteroge-neous people over time. Moreover, these people should face the same finan-cial incentives for the use of health services; otherwise their use of services will be distorted.

The RAND Health Insurance Experiment (RAND-HIE) provided a data set that mostly satisfied these conditions. We discuss this experiment in some detail in chapter 8, but for current purposes, it is enough to know that the study randomly assigned people from six sites across the country into different health plans and monitored their use of health services over the four to five years of the experiment during the 1970s. It also recorded demographic and health status characteristics of the participants at baseline. In fact, much of the current knowledge about the measurement of health status had its genesis with this study. Thus, the study is well suited to exam-ine alternative predictive models of utilization based on demographic char-acteristics, subjective and physiological measures of health status, and prior utilization (Newhouse et al. 1989). See Measures of Potential Risk Factors Used in the RAND-HIE Study for a summary of the measures available for consideration.

Measures of Potential Risk Factors Used in the RAND-HIE Study

Demographic Measures (AAPCC Variables)• Age

• Gender

• Location (indicator for each of the six sites in the study)

• Eligible for welfare at baseline


EBSCOhost – printed on 1/12/2023 11:01 AM via UNIVERSITY OF MARYLAND GLOBAL CAMPUS. All use subject to https://www.ebsco.com/terms-of-use

Health Insurance124

Although the analysis was more complicated than suggested by the side-bar, the study team essentially ran a series of regressions in which total inpatient and outpatient expenditures of each individual in year t were the dependent variables and were explained by alternative sets of potential risk characteristics. When prior-year utilization measures were included, these were expenditures in year t − 1. Because the overall RAND study was concerned with the effects of insurance copayment arrangements on expenditures, the regressions also controlled for the health plan in which the person was enrolled. Exhibit 7.1 reports the R2, or percentage of explained variation, for many of the regressions the study team ran. Because their interest was in improving the AAPCC model Medicare used, all of the models include the demographic or AAPCC factors. The AAPCC variables by themselves explain 1.6 percent of total expenditures, 0.7 percent of inpatient expenditures, and 7.2 percent of outpatient varia-tion. Notice that, in general, outpatient expenditures were more predictable than inpatient spending. This probably reflects the greater extent to which

Subjective Health Status Measures• Physical health (based on self-reported measures of role and

personal limitations)

• Mental health (based on self-reported measures of psychological distress, behavioral and emotional control, and positive affect)

• General health (based on self-reported measures of general well-being)

• Disease count (based on the presence of any of 32 chronic conditions)

Physiological Health Status Measures• Dichotomous measures

• Continuous measures (based on 27 measures, including such items as elevated cholesterol, hypertension, diabetes, electrocardiogram abnormalities, active ulcer, anemia, dyspepsia, abnormal thyroid function, and so on)

Prior Utilization• Outpatient expense in prior year

• Inpatient expense in prior year

Note: AAPC = adjusted average per capita costs.

Source: Data from Newhouse and the Insurance Experiment Group (1993) and Newhouse

et al. (1989).

EBSCOhost – printed on 1/12/2023 11:01 AM via UNIVERSITY OF MARYLAND GLOBAL CAMPUS. All use subject to https://www.ebsco.com/terms-of-use

Chapter 7: Risk Adjustment 125

behavioral and chronic factors influence ambulatory use. Notice, too, that the percentage of explained variation is quite small. Age, gender, location, and welfare status explained less than 2 percent of total expenditures.

In principle, Medicare, or any insurer, could ask its subscribers to report their health status and use the responses to assign the subscribers to appropriate risk classes. This study had rather extensive measures of subjec-tive health. When these were added to the AAPCC measures, the model explained 2.8 percent of total expenditures—a 75 percent improvement! Operationally, however, self-reported health status is likely to be problematic for Medicare. Beneficiaries (and the health plans that they wish to join) may have an incentive to report poorer health status in the hope that a higher cap-itation payment will be forthcoming. Confirming that the information that beneficiaries provide is truthful could become a serious and costly challenge.

Alternatively, Medicare could obtain relatively simple dichotomous physiological measures of health status, such as measures from a clinical record that indicate whether the beneficiary has hypertension, diabetes, and so on. These measures were added to the AAPCC measures and are reported in the third row of exhibit 7.1. Together with the demographic factors, they explained 3.8 percent of total expenditures. This finding is a substantial improvement over simply using the demographic measures, but obtaining even simple clinical data is expensive both for Medicare and for the beneficiary.

We could go further and use even more detailed clinical information. For example, we could collect and use data on actual blood pressure, instead of a simple measure of whether the beneficiary has hypertension. We could use a measure of elevated glucose, rather than a simple measure of whether the beneficiary has diabetes. Such continuous measures of physiological health are reported in the fourth row of exhibit 7.1. Together with the AAPCC measures, they explained 4.2 percent of variation in total expenditures. Thus,

EXHIBIT 7.1Percentage of Explained Variation in Healthcare Expenditures Yielded by Alternative Specifications

Note: AAPC = adjusted average per capita costs.

Source: Data from Newhouse et al. (1989).

Total Inpatient Outpatient

AAPCC 1.6 0.7 7.2

AAPCC + Subjective health status 2.8 1.2 11.1

AAPCC + Dichotomous physiological health status

3.8 2.0 13.5

AAPCC + Continuous physiological health status

4.2 2.6 13.0

AAPCC + Prior utilization 6.4 2.8 21.2

All 9.0 5.0 25.1

EBSCOhost – printed on 1/12/2023 11:01 AM via UNIVERSITY OF MARYLAND GLOBAL CAMPUS. All use subject to https://www.ebsco.com/terms-of-use

Health Insurance126

more detailed clinical measures do provide more predictive power, but the costs of collecting such detailed data are probably prohibitive.

Alternatively, Medicare (or another insurer) could use prior utilization data. These measures were added to the AAPCC demographic factors and are reported in the fifth row of exhibit 7.1. This approach explained 6.4 percent of total expenditures, 2.8 percent of inpatient claims, and 21.2 percent of outpatient expenditures. Relative to the other approaches, prior utilization has substantially greater explanatory power. This result probably explains why health insurers tend to focus on prior claims experience when setting insurance premiums. The data exhibit relatively strong predictive power. Moreover, once insurers have a set of subscribers and their claims experience, using those data to predict future use is relatively inexpensive.

We could go further and combine various sets of health status mea-sures. The final row of exhibit 7.1 presents the percentages of explained varia-tion when all the measures of subjective and physiological health and prior utilization were included with the AAPCC measures. The model explained 9.0 percent of variation in total expenditures. Thus, using all these data does improve ability to predict expenditures, but routinely collecting such infor-mation is very expensive.

Implications of Better Risk Adjustment

The final exercise the RAND study team undertook was to estimate the potential profit that an HMO could achieve if it could somehow better predict future Medicare expenditures of potential enrollees than the existing AAPCC formula. The admittedly unrealistic assumption is that the HMO could do this costlessly and would use the information to enroll only profit-able beneficiaries. The results are presented in exhibit 7.2, inflated to 2018 dollars using the Consumer Price Index (all categories of the index included).

EXHIBIT 7.2Profits

from Better Prediction of

HMO Medical Expenditures

Source: Adapted from data in Newhouse et al. (1989).

Additional Variance Explained by HMO

Profit per Enrollee, 1988 Dollars

Profit per Enrollee, 2018 Dollars

0 percentage points $0 $0

1 percentage point $630 $1,313

5.5 percentage points $1,170 $2,438

7.5 percentage points $1,320 $2,751

13 percentage points $1,530 $3,189

18.5 percentage points $1,650 $3,439

EBSCOhost – printed on 1/12/2023 11:01 AM via UNIVERSITY OF MARYLAND GLOBAL CAMPUS. All use subject to https://www.ebsco.com/terms-of-use

Chapter 7: Risk Adjustment 127

Obviously, if HMOs cannot predict expenditures any better than can Medicare’s AAPCC formula, there is no extra profit. However, if they could predict 1 percentage point better and use this information to enroll only healthier people, they would gain profits of $1,313 per enrollee because their costs would be lower than the Medicare payment rate. If HMOs could do 5.5 percentage points better, the profit per enrollee would be $2,438. Notice in exhibit 7.2 that, as we move to greater and greater additional explana-tory power, higher profits are garnered. But notice, too, that the extra profit gets smaller with each increment. One additional point yields $1,313 in profit, but the next 4.5 percentage points only result in an additional $1,125 ($2,438−$1,313). Two additional percentage points beyond that yield only an extra $313. The modeling reported in exhibit 7.1 makes it clear how dif-ficult it would be to get an additional 4.5 percentage points of explanatory power. Using all the information available to the study, the RAND team could only get 7.4 percentage points greater predictive power than the AAPCC.

This finding had important implications for Medicare. If Medicare could improve its AAPCC model enough to predict just a few percentage points better than it currently did, it could remove the easy opportunities for favorable selection that the managed care plans seemed to enjoy. To do bet-ter than, say, the AAPCC plus prior utilization would likely require managed care plans to incur considerable costs of improved predicting for rather mod-est increases in profits. Even plans bent on taking full advantage of favorable selection would find that their efforts were likely to be unremunerative.

Generalizing the RAND Findings

Your first reaction to the RAND findings might be to say: “Surely, one can do better than predicting only 6.4 or 9.0 percent of total expenditures!” Van de Ven and Ellis (2000) provide a detailed summary of the research on risk adjustment in their chapter in Handbook of Health Economics. Exhibit 7.3 reproduces a table from their work that summarizes six major studies of risk adjustment, beginning with the RAND study. With the exception of the study of US HMO enrollees (column 3), all of the results are remarkably similar, with age and sex variables explaining 0.7 to 3.8 percent of variation, and all variables explaining 7 to 9 percent. More recent work by Behrend and colleagues (2007) examined risk adjustment strategies using German sickness fund data from 1997 and 1998. They found that age, gender, and “invalid status” explained 5.1 percent of concurrent healthcare spending. However, adding inpatient hierarchical coexisting conditions explained a total of 37 percent of concurrent expenditures and 12 percent of prospec-tive expenditures. These condition codes, which we discuss next, were able to substantially increase the ability to predict future healthcare expenditures.

EBSCOhost – printed on 1/12/2023 11:01 AM via UNIVERSITY OF MARYLAND GLOBAL CAMPUS. All use subject to https://www.ebsco.com/terms-of-use

Health Insurance128



e et






et a



de V





s et






an P










pe e

t al

. (1




s (1







S pr

























ss fu
























N =



= 2



= 5


N =



N =














All s








al s





























rm 3

6 lik







r ye





















is b









All v
















R2 fr

om V




k Ad



t M


s fr

om S





e: *

All m


s in


e ag

e an

d se

x as


l as



s sh


† Dep


nt v






d at



, whi

ch in


s R

2. ‡ ACG


G m


. §D


CC m


. "Th







ce: T



e w

as p



in H



of H


h Ec




yer a

nd N




. Van





, “R




ent i

n Co



ve H


h Pl

an M






, tab

le 3

, Cop


ht E


ier 2




ed w





EBSCOhost – printed on 1/12/2023 11:01 AM via UNIVERSITY OF MARYLAND GLOBAL CAMPUS. All use subject to https://www.ebsco.com/terms-of-use

Chapter 7: Risk Adjustment 129

Medicare’s Current Approach to Risk Adjustment

The Balanced Budget Act of 1997 (BBA) required Medicare to phase in a new risk adjustment methodology beginning in 2000 (Ingber 2000). The new meth-odology was to better incorporate health status into the capitation rates. Medicare implemented a transitional risk adjustment system based only on inpatient data in 2000 and a full model based on both inpatient and ambulatory data in 2004. In addition, because a risk-adjusted payment system is based on patient health status measures, the BBA requires Medicare Advantage plans and other providers to provide encounter data to the Centers for Medicare & Medicaid Services (CMS). For a detailed discussion of what is now called the CMS Hierarchical Condition Categories (CMS-HCC) model, see Pope and colleagues (2004).

CMS and its contractors developed the payment system by running a series of regression models not unlike those used in the earlier RAND study. In essence they ran a model something like the following:

Expendituresit = a1 * Age (65–69) + a2 * Age (70–74) +a3 * Age (75–79) + . . . + a6 * Male + a7 * Medicaid eligible +a8 * Condition1 + a9 * Condition2 +a10 * Condition3 + . . . +a60 * Condition52 + εit.

The estimated coefficients—the a’s in the equation—tell CMS how much the associated variable contributed to Medicare expenditures, on aver-age. CMS experimented with how age, sex, and other demographic factors were specified and with how alternative measures of the clinical conditions explained contemporaneous expenditures and subsequent expenditures. This experimentation continued until CMS was satisfied that its final model reflected an acceptable compromise across the ten principles summarized in Guiding Principles in Medicare’s Risk Adjustment Approach. As a reading of the principles makes clear, considerable experimentation and judgment is required to develop such a risk-adjusted payment system.

Guiding Principles in Medicare’s Risk Adjustment Approach

The new risk adjustment system was designed to meet ten guiding principles (Pope et al. 2004). These principles relate to insurance underwriting issues, understanding and acceptance by users, and minimization of opportunities to game the system. Briefly, the ten principles are the following:


EBSCOhost – printed on 1/12/2023 11:01 AM via UNIVERSITY OF MARYLAND GLOBAL CAMPUS. All use subject to https://www.ebsco.com/terms-of-use

Health Insurance130

1. The health status–related measures should be clinically meaningful. This principle means that the measures should make sense to a knowledgeable observer and be sufficiently clinically specific to make it difficult for plans to assign a beneficiary with a vaguely defined condition into a higher payment group.

2. The measures should predict both current and future medical expenditures. Thus, a transitory condition, such as an ankle sprain, would not be a useful measure.

3. The measures should be based on large enough sample sizes that they yield accurate and stable predictions. Thus, as we saw in chapter 6, Medicare, as with any insurer, may have to sacrifice some risk categories to gain reduction in variance.

4. Related clinical conditions should be treated hierarchically, while unrelated conditions should increase the level of payment. Thus, someone identified as having had a recent acute myocardial infarction (i.e., a heart attack) and having unstable angina would only be counted as having the more severe condition rather than both. However, someone with unstable angina and lung cancer would be counted as having both.

5. Vague measures should be grouped with low-paying diagnoses to encourage specific coding of health conditions.

6. The measures should not encourage multiple reporting of the same or closely related diagnoses. Thus, the hierarchy of related conditions should be used and only the most severe condition coded.

7. Providers should not be penalized for reporting many conditions. Thus, no condition should have a negative payment associated with it, and a more severe condition must pay at least as much as a less severe manifestation.

8. Transitivity must hold. If condition A results in a greater payment than condition B, and if B is paid more than C, then A should be paid more than C.

9. All of the diagnoses that clinicians use have to map onto the payment system.

10. Discretionary diagnostic codes should be excluded to prevent intentional or unintentional gaming of the system.

The chosen model would be used to determine the annual payment that a Medicare Advantage plan would be paid on behalf of a Medicare ben-eficiary living in a particular county. For example, the base payment for a woman, aged 75–79, living in the community might be $2,475 per month.

EBSCOhost – printed on 1/12/2023 11:01 AM via UNIVERSITY OF MARYLAND GLOBAL CAMPUS. All use subject to https://www.ebsco.com/terms-of-use

Chapter 7: Risk Adjustment 131

If in the last year she had diabetes without complications, that might add $1,024. If she also had unstable angina, that might add $1,785. In this case, the Medicare HMO would be paid $5,284 on her behalf.

The model that Medicare ultimately adopted contains 12 age cat-egories × 2 sex categories × 2 site (community vs. institution) categories for a total of 48, plus 6 Medicaid categories, plus 70 hierarchical condition categories (i.e., the condition codes), plus another 6 condition code inter-actions. In some instances, the costs associated with having two conditions are greater than simply the sum of the costs of each; diabetes together with cerebrovascular disease is an example. The interactions allow Medicare to pay a Medicare Advantage plan more for the care of such patients. There are also categories that relate to the Medicare disabled.

Age and sex explain approximately 1.0 percent of the variation in Medicare expenses. The CMS-HCC model explains 11.2 percent. Exhibit 7.4 compares the predictive ratio of the age and sex model and the CMS-HCC model for the quintiles of Medicare expenditures. The predictive ratio is just the predicted costs of a group divided by the actual cost. If the value is greater than 1, it means that Medicare would be overpaying for the care of people in that group. If the value is less than 1, Medicare would be under-paying. The first quintile of expenditures (i.e., the least expensive one-fifth of Medicare beneficiaries) is shown in the first row. On average, just using age and sex as adjusters (the first column) leads to an overpayment of 266 percent of actual costs. In contrast, an HMO caring for Medicare beneficiaries in the most expensive one-fifth of the distribution would be paid only 44 percent of what their care would cost, on average.

Clearly, the CMS-HCC model is an improvement. While it still over-pays for the less costly quintiles, the overpayments are drastically reduced. Similarly, while it underpays for the most expensive fifth quintile, the pay-ment is much closer to actual costs. These findings have led some researchers

EXHIBIT 7.4Predictive Ratios for Alternative Risk Adjustments

Source: Data from Pope et al. (2004).

Quintiles of Expenditures Age/Sex CMS-HCC Model

First 2.66 1.23

Second 1.93 1.23

Third 1.37 1.14

Fourth 0.95 1.02

Fifth 0.44 0.86

Top 5% 0.28 0.77

Top 1% 0.10 0.69

EBSCOhost – printed on 1/12/2023 11:01 AM via UNIVERSITY OF MARYLAND GLOBAL CAMPUS. All use subject to https://www.ebsco.com/terms-of-use

Health Insurance132

to suggest that future risk adjustment models should continue to employ a CMS-HCC–like model but incorporate a mechanism directly to pay some share of costs for particularly high-cost beneficiaries (Ellis and McGuire 1993; Newhouse 1996).

CMS began phasing in risk-adjusted payments to Medicare Advantage plans beginning in 2000. In 2007, payment rates were based entirely on the CMS-HCC methodology. As with the original AAPCC formula, CMS-HCC continues to establish the basic payment level based on Medicare’s expenditures in geographic regions, usually counties. However, instead of being simple averages as in the AAPCC, the payments are now based on the risk-adjusted mix of beneficiaries in the county. The sidebar Sample Annual Medicare Advantage Payment Under the CMS-HCC Model, Lake County, Illinois, 2018 presents the payment that a Medicare HMO in a county just north of Chicago would receive in 2018 for a 72-year-old woman who is on Medicaid and who has diabetes and congestive heart failure. The base rate for her county of residence is $10,078 per year. The factors for her sex, Medicaid status, and health conditions are added and then multiplied by the base rate to determine the payment to be made to the HMO each month on her behalf.

Sample Annual Medicare Advantage Payment Under the CMS-HCC Model, Lake County, Illinois, 2018

Basic Lake County, Illinois, rate: $10,078

Female, age 70–74: .406

HCC15 diabetes without complications: .118

HCC87 congestive heart failure: .368

Total payment is: $10,078 (1 + .406 + .118 + .368) = $19,067.58

While this format is less intuitive than the dollar-based formats discussed earlier in the chapter, it has the administrative advantage that CMS need not recompute each value every year. New data on average risk-adjusted expenditures and any congressionally mandated across-the-board increases or decreases can simply be applied to the base rates. The relative values of the person-specific components are unaffected. The CMS-HCC is used to risk adjust the payments to Medicare Advantage plans, but the Congress has made the determination of the basic county rate more nuanced (see Current Medicare Advantage Payment Plans Include a Bid-ding Mechanism).

EBSCOhost – printed on 1/12/2023 11:01 AM via UNIVERSITY OF MARYLAND GLOBAL CAMPUS. All use subject to https://www.ebsco.com/terms-of-use

Chapter 7: Risk Adjustment 133

How Well Did the Centers for Medicare & Medicaid Services Hierarchical Condition Categories Model Work in Reducing Favorable Selection?

The more sophisticated risk adjustment mechanism was intended to reduce favorable selection into Medicare Advantage plans by determining payment levels that were closer to actual expenditures. Plans would have reduced incentives to spend resources seeking healthier subscribers. Moreover, if the payment levels are closer to actual patient-specific costs, plans would also have reduced incentives to try to disenroll expensive subscribers (see Current Medicare Advantage Payment Plans Include a Bidding Mechanism).

Current Medicare Advantage Payment Plans Include a Bidding Mechanism

As discussed in more detail in chapter 22, the Medicare Modernization Act of 2004, which provided for prescription drug coverage for Medicare benefi-ciaries, also modified the way Medicare Advantage plans are paid. The man-aged care plans proffer a bid per enrollee per month to Medicare to provide a basic set of benefits consistent with traditional Medicare. If this bid is below the CMS-established benchmark for the county (or region, if applicable), the managed care plan keeps 75 percent of the difference to apply to reduced cost sharing or expanded benefits for enrolled beneficiaries. If it is above the benchmark, the plan charges enrollees an additional premium. However, the CMS-HCC model is used in all cases to adjust the payments for beneficia-ries enrolled by the plan to reflect their demographics and health status. The ACA also made some changes to the payment for Medicare Advantage plans. However, these do not affect the risk adjustment mechanism and, therefore, we defer that discussion to chapter 22.

In recent work (Morrisey et al. 2013) we found mixed effects. We used 10 years of Medicare claims data on more than 3 million beneficiaries from 1999 through 2008 and followed the modeling used by the PPRC (1994) that was discussed in chapter 5. We identified people who switched from traditional Medicare to a Medicare Advantage plan. Then we computed their claims costs in the six months prior to the switch and compared these to the costs of people in the same county, in the same six-month interval, who did not switch from traditional Medicare. We made an analogous calculation for the six months after disenrollment of those switching back to traditional Medicare. In a regression model that accounted for payment levels, a time

EBSCOhost – printed on 1/12/2023 11:01 AM via UNIVERSITY OF MARYLAND GLOBAL CAMPUS. All use subject to https://www.ebsco.com/terms-of-use

Health Insurance134

trend, and county-level fixed effects, we found no effect of the shift to CMS-HCC on the extent of favorable selection into Medicare Advantage plans. On the other hand, we did find that the CMS-HCC reduced the number of disenrollees; these were concentrated among the most costly tail of the distribution. Thus, if one returns to exhibit 7.4, it appears that the reduced disenrollments were concentrated in the fourth quintile. These are the sub-scribers that Pope and colleagues (2004) estimated to shift from net “losers” for a managed care plan to net “winners.”

Brown and colleagues (2011) found a more nuanced story. They examined data on 55,000 Medicare beneficiaries over the period 1994–2006. They found that after risk adjustment using the CMS-HCC, the risk scores of those switching to a Medicare Advantage plan rose relative to those remaining in traditional Medicare. However, given the risk score, Medicare’s expendi-tures for those switching to Medicare Advantage plans actually fell. As Brown and his colleagues characterize it, the plans devoted less selection effort along the beneficiary characteristics that were included in the HCC and more along dimensions that were unmeasured. In their estimates, “After risk-adjustment, those switching into Medicare Advantage plans were over $1,200 ‘cheaper’ than the risk-adjustment formula predicts them to be” (Brown et al. 2011, 2).

This is not to say that risk adjustment practices or underwriting more generally are bad ideas. Rather, it reinforces a point we made in chapters 5 and 6. People know more about their likely use of health services than an insurer does, and they should be expected to use their greater knowledge to their advantage. If an insurer gets the underwriting wrong or if a government program gets the risk adjustment wrong, their costs will be higher than if they got it right.

Risk Adjustment in the Affordable Care Act

The ACA established three forms of adjusting for risk in the exchanges: a permanent risk-adjustment program, a three-year temporary reinsurance pro-gram, and a risk corridor program. The risk adjustment program is modeled on the Medicare Advantage approach that was just described. However, it uses Truvan Health Analytics data on privately insured individuals typically under 65 rather than Medicare data to establish HCCs that apply to the ACA popula-tion. Data provided by insurers on their ACA enrollees allow CMS to compute the average risk level of those enrolled by each insurer, and these are used to adjust payments. If one insurer has a “sicker” enrollment mix it receives a risk-adjustment payment. However, if the insurer has a less “sick” enrollment pool, it makes a payment into the program. This is a zero-sum process. The amounts paid into the program by those with less sick enrollees is equal to the amounts paid out to other insurers. This zero-sum model generated problems for some insurers (see An ACA’s Risk Adjustment Implication).

EBSCOhost – printed on 1/12/2023 11:01 AM via UNIVERSITY OF MARYLAND GLOBAL CAMPUS. All use subject to https://www.ebsco.com/terms-of-use

Chapter 7: Risk Adjustment 135

The reinsurance program was designed to mitigate the costs of extremely expensive enrollees. Fees were collected from essentially all insur-ers, including self-insured large employers. These fees were $63, $44, and $27 per enrollee, respectively, in 2014, 2015, and 2016. Insurers selling plans through the ACA that had high-cost enrollees received monies to com-pensate them partially for these unusually high costs.

The definition of high cost is determined by the “attachment point” established by CMS. In 2014 and 2015, insurers would receive compensation for those with costs above $45,000. In 2016, the attachment point was set at $90,000. There were also limits on the amount that an insurer could draw from the reinsurance fund. See Cox and colleagues (2016) for a more detailed description. Several analysts have argued that a meaningful way to reduce the effects of adverse selection in the ACA would be for the federal government or the individual states to implement a reinsurance pool. As of the summer of 2018, Alaska, Minnesota, Oregon, Wisconsin, Maine, and Maryland have obtained waivers to establish ACA reinsurance programs in their states.

The risk corridor program was designed to mitigate some of the pre-mium concerns faced by insurers who were uncertain about their risks under the ACA. Insurers that had profits that exceeded 3 percent of a threshold were assessed a fee, and those who had profits 3 percent below the threshold received payments.


• In general, risk adjustment models have been able to predict about 12 percent of total claims. Ambulatory use is easier to predict than inpatient use, perhaps because it has a larger behavioral component.

• Demographic characteristics, such as age and gender, are only modestly predictive of future claims experience. While subjective and physiological measures of health status are more predictive, prior utilization provides the most predictive power.

An ACA Risk Adjustment Implication

The ACA risk adjustment process had important implications for insurers, as one Texas insurer told me. In 2014, the organization set its premiums high, attracted high morbidity enrollees, and lost money. The next year, it set lower premiums, attracted lower morbidity enrollees, and made money—then came the risk adjustment fee, and the insurer lost money overall. It dropped out of the ACA.

EBSCOhost – printed on 1/12/2023 11:01 AM via UNIVERSITY OF MARYLAND GLOBAL CAMPUS. All use subject to https://www.ebsco.com/terms-of-use

Health Insurance136

• The Medicare AAPCC is a manual rating program by which Medicare paid Medicare HMOs based on the average costs in the county, adjusted for the age, gender, and Medicaid, institutional, and active-worker status of the beneficiary.

• Medicare currently pays Medicare Advantage plans on the basis of the CMS-HCC model. This manual rating program uses approximately 70 clinical conditions, in addition to demographic and location factors, to determine the amount Medicare will pay HMOs for the care of its beneficiaries.

• The ACA uses an HCC approach to adjust payments to insurers that have enrollment that is above or below average risk.

Discussion Questions

1. How would you describe the CMS-HCC risk adjustment system? Does it use prior utilization, physiological, and demographic information to determine payment rates? How?

2. Suppose a Medicare Advantage plan had been aggressively using some method to attract low utilizers into its plan. In what ways would you expect it to change its behavior, if at all, as a result of the implementation of the new CMS-HCC model?

3. How would a CMS-HCC type model apply to people newly eligible for Medicare?

4. If Medicare Advantage plans must provide Medicare with encounter data on the healthcare utilization of their subscribers, what would you predict about the nature of the underwriting that managed care plans will use when negotiating future contracts with private employers?

5. How might an insurer offering coverage in the state individual insurance exchanges be able to influence the mix of healthy people enrolled in its plan even in the face of a prohibition on using health status?

For the Interested Reader

Cox, C., A. Semanskee, G. Claxton, and L. Levitt. 2016. “Explaining Health Care Reform: Risk Adjustment, Reinsurance, and Risk Corridors.” Kaiser Family Foundation. Published August 17. www.kff.org/health-reform/issue-brief/explaining-health-care-reform-risk-adjustment-reinsurance-and-risk- corridors/.

EBSCOhost – printed on 1/12/2023 11:01 AM via UNIVERSITY OF MARYLAND GLOBAL CAMPUS. All use subject to https://www.ebsco.com/terms-of-use

Chapter 7: Risk Adjustment 137

Morrisey, M. A., M. L. Kilgore, D. J. Becker, W. Smith, and E. Delzell. 2013. “Favorable Selection, Risk Adjustment, and the Medicare Advantage Pro-gram.” Health Services Research 48 (3): 1039–56.

Weiner, J. P., E. Trish, C. Abrams, and K. Lemke. 2012. “Adjusting for Risk Selec-tion in State Health Insurance Exchanges Will Be Critically Important and Feasible, but Not Easy.” Health Affairs 31(2): 306–15.


Behrend, C., F. Buchner, M. Happich, R. Holle, P. Reitmeir, and J. Wasem. 2007. “Risk-Adjusted Capitation Payments: How Well Do Principal Inpatient Diagnosis-Based Models Work in the German Situation?” European Journal of Health Economics 8 (1): 31–39.

Brown, J., M. Duggan, I. Kuziemko, and W. Woolston. 2011. “How Does Risk Selection Respond to Risk Adjustment? Evidence from the Medicare Advan-tage Program.” National Bureau of Economic Research. Published April. www.nber.org/papers/w16977.

Cox, C., A. Semanskee, G. Claxton, and L. Levitt. 2016. “Explaining Health Care Reform: Risk Adjustment, Reinsurance, and Risk Corridors.” Kaiser Family Foundation. Published August 17. www.kff.org/health-reform/issue-brief/explaining-health-care-reform-risk-adjustment-reinsurance-and-risk- corridors/.

Ellis, R. J., and T. McGuire. 1993. “Supply Side and Demand Side Cost Sharing in Health Care.” Journal of Economic Perspectives 7 (4): 135–51.

Fowles, J. B., J. P. Weiner, D. Knutson, A. M. Tucker, and M. Ireland. 1996. “Taking Health Status into Account When Setting Capitation Rates.” Journal of the American Medical Association 276 (16): 1316–21.

Ingber, M. J. 2000. “Implementation of Risk Adjustment for Medicare.” Health Care Financing Review 21 (3): 119–26.

Lamers, L. M. 1999. “The Simultaneous Predictive Accuracy for Future Health Care Expenditures of DCGs, PCGs, and Prior Costs.” Paper presented at the iHEA Conference, Rotterdam, Netherlands, June 6–9.

Morrisey, M. A., M. L. Kilgore, D. J. Becker, W. Smith, and E. Delzell. 2013. “Favorable Selection, Risk Adjustment and the Medicare Advantage Pro-gram.” Health Services Research 48 (3): 1039–56.

Newhouse, J. P. 1996. “Reimbursing Health Plans and Health Providers: Efficiency in Production Versus Selection.” Journal of Economic Literature 34: 1236–63.

Newhouse, J. P., and the Insurance Experiment Group. 1993. Free for All? Lessons from the RAND Health Insurance Experiment. Cambridge, MA: Harvard University Press.

EBSCOhost – printed on 1/12/2023 11:01 AM via UNIVERSITY OF MARYLAND GLOBAL CAMPUS. All use subject to https://www.ebsco.com/terms-of-use

Health Insurance138

Newhouse, J. P., W. G. Manning, E. B. Keeler, and E. M. Sloss. 1989. “Adjusting Capitation Rates Using Objective Health Measures and Prior Utilization.” Health Care Financing Review 10 (3): 41–54.

Physician Payment Review Commission (PPRC). 1994. “A Comparison of Alterna-tive Approaches to Risk Measurement.” Selected External Research Series, no. 1. Washington, DC: Physician Payment Review Commission.

Pope, G. C., J. Kautter, R. P. Ellis, A. S. Ash, J. Z. Avanian, L. I. Iezzoni, M. J. Ing-ber, J. M. Levy, and J. Robst. 2004. “Risk Adjustment of Medicare Capitation Payments Using the CMS-HCC Model.” Health Care Financing Review 25 (4): 119–41.

Pope, G. C., K. W. Adamache, R. K. Khandker, and E. G. Walsh. 1998. “Evaluating Alternative Risk Adjusters for Medicare.” Health Care Financing Review 20 (2): 109–29.

Van Vliet, R. C., and W. P. M. M. Van de Ven. 1992. “Towards a Capitation Formula for Competing Health Insurers: An Empirical Analysis.” Social Science and Medicine 34 (9): 1035–48.

Van de Ven, W. P. M. M., and R. P. Ellis. 2000. “Risk Adjustment in Competitive Health Plan Markets.” In Handbook of Health Economics, edited by A. J. Culyer and J. P. Newhouse, 755–845. Amsterdam, Netherlands: Elsevier.

Zarabozo, C. 2000. “Milestones in Medicare Managed Care.” Health Care Financ-ing Review 22 (1): 61–67.

EBSCOhost – printed on 1/12/2023 11:01 AM via UNIVERSITY OF MARYLAND GLOBAL CAMPUS. All use subject to https://www.ebsco.com/terms-of-use

Calculate your order
Pages (275 words)
Standard price: $0.00
Client Reviews
Our Guarantees
100% Confidentiality
Information about customers is confidential and never disclosed to third parties.
Original Writing
We complete all papers from scratch. You can get a plagiarism report.
Timely Delivery
No missed deadlines – 97% of assignments are completed in time.
Money Back
If you're confident that a writer didn't follow your order details, ask for a refund.

Calculate the price of your order

You will get a personal manager and a discount.
We'll send you the first draft for approval by at
Total price:
Power up Your Academic Success with the
Team of Professionals. We’ve Got Your Back.
Power up Your Study Success with Experts We’ve Got Your Back.
Open chat
Hello. Can we help you?