History of Risk Adjustment in the US

• Abstract
• Introduction
• Why Is Risk Adjustment Necessary?
• The Road to Risk Adjustment
• Where Are We Now?
• What Have We Learned?
• References

Abstract

It has long been recognised that risk segmentation in health insurance markets is problematic, often resulting in the sickest individuals being denied health insurance coverage. Over the last several decades a variety of attempts have been made to address the problems of risk segmentation in the health insurance market in the United States, as researchers, policy-makers, and health plan representatives have attempted to increase health coverage and improve the efficiency of the market. A variety of tools have been developed for assessing risk in different population groups and risk pools. These tools, in turn, have been used as part of risk adjustment mechanisms in the private market, for state employee groups, for Medicaid, and for Medicare. Great strides have been made toward addressing risk segmentation in the US. However, despite significant investment in the development of risk assessment tools and strategies for implementation, health-based payments have not been widely adopted in the private insurance market, nor universally adopted among public payers.

Introduction

Addressing the issue of risk segmentation in the health insurance market is a two-step process involving both risk assessment and risk adjustment. Risk assessment can be defined as a means of predicting the deviations of each individual’s expected cost from the average enrollee’s cost, and risk adjustment would be the method used to compensate health plans ^{[ b ]} according to the amount of risk they assume. For example, plans with high risk would receive payments through a risk adjustment mechanism, while plans with low risk would make payments into a risk adjustment system. The purpose of this article is to describe risk adjustment efforts in the US and to summarise lessons learned to date ^{[ 1 ]}.

Why Is Risk Adjustment Necessary?

In a competitive and voluntary health insurance market, like that in the US, health plans have a strong financial incentive to attract the healthiest enrollees, while excluding sicker, higher risk enrollees. This incentive encourages health plans to compete on the basis of risk selection rather than on the basis of cost efficiency and quality of health care.

Therefore, plans often design benefit packages and promotional strategies that maximise their appeal to healthy consumers. In addition, consumers always maintain an information advantage over the health plans, since consumers know more about their own health status and likely health care utilisation over the short term, resulting in enrollees who are healthy exercising different insurance preferences than those who are sick. The combination of plans’ attempts to risk select among enrollees and individuals’ selection of plans and providers likely to conform best to their own needs results in a market with risk segmentation among plans and providers ^{[ 2 ]}.

The result of these incentives and strategies is that some health plans attract a relatively healthy group of enrollees, while others acquire a sicker pool of enrollees and therefore the risk is higher. Plans with healthier enrollees stand to be rewarded financially, while plans with sicker enrollees from the higher risk pool will suffer financial losses. This, in turn, can result in plans denying coverage to sicker individuals and groups. In addition, plans and providers not compensated for the actual risk they are covering may have an incentive to stint on care, thereby threatening the quality of health care provided to the sickest people. Plans with higher risk populations for which they are not adequately compensated might also choose to leave the market entirely. Finally, without risk adjustment, each of these scenarios could result in inequities in financial access to health care, with sicker individuals being forced to pay much higher premiums or being denied coverage at any cost ^{[ 2 ]}.

In the private insurance market, risk adjustment is a corrective tool designed to reorient the incentives for health plans and enrollees, reducing the negative consequences of enrolling high-risk users by compensating plans according to the health risk of plans’ enrollees. Public payers use health-based payments to set initial managed care rates, as well as to move money among plans according to the risk pool they attract ^{[ 2 ]}. An adequate risk adjustment mechanism provides incentives for health plans to compete on quality of care rather than on price alone. Risk adjustment comprises a number of approaches including:

Many tools for assessing and adjusting risk have been developed over the last decade in the US. The tools vary in the data they require, with some using inpatient data, some using outpatient data and some using both. In addition, tools have been developed using pharmacy data. Risk adjustment tools are tailored to different populations and can be used for adjusting risk prospectively, concurrently or retroactively. These tools have been implemented in a variety of settings in both the private and public health insurance markets, where there continues to be much discussion about the actual impact of various risk adjustment mechanisms, the most appropriate settings in which to implement them, and the best processes for implementation.

The Road to Risk Adjustment

The first step in implementing a satisfactory risk adjustment system is developing an adequate risk assessment tool. Since the late 1960s, many efforts have been undertaken to develop and test a variety of risk assessment mechanisms, especially for individual major medical insurers. Early risk assessment efforts were used primarily as a tool for health insurance underwriting and to price individual premiums ^{[ 3 ]}. While the risk assessment tools that have been developed and tested vary significantly, they all use some combination of demographic and/or health status data. Demographic models generally comprise some or all of the following variables: age, sex, family status, location and welfare status. Measures of health status can include self-reported health status (through surveys), clinical diagnoses and data reflecting prior utilisation. Health status models also tend to include demographic predictors of cost ^{[ 4 ]}.

In the mid-1990s, discussions about risk adjustment centred on reaching a common understanding about the meaning of risk selection in terms of access to care for vulnerable populations and how risk adjustment might, in theory, affect the insurance market and contribute to lessening the incentives for risk selection. At that time, there were several private sector risk adjustment demonstrations under development. There was little understanding about when and how to trade improvements in accuracy of risk measurement for simpler models ^{[ 2 ]}. In 1994, participants at a conference addressing risk adjustment concluded that improving the accuracy of risk adjustment models was important, but that it was more important to begin experimenting with risk adjustment mechanisms and resolving implementation issues for purchasers and plans ^{[ 5-6 ]}.

By the late 1990s, the health insurance market had moved significantly toward addressing the problem of risk segmentation in the health care market ^{[ 2 ]}. Many of the risk adjustment tools had been implemented in a variety of demonstration settings for many different populations, including those with private employer coverage, government employee coverage, Medicaid^{[ c ]} and Medicare^{[ d ]}. In fact, by 1997, the US Balanced Budget Act of 1997 required the Health Care Financing Administration (HCFA) to implement a risk-adjusted payment system for Medicare+Choice^{[ e ]} plans by 1 January 2000 ^{[ 7 ]}.

Case studies of the Washington State Health Care Authority ^{[ 8 ]}, the Health Insurance Plan of California ^{[ 9 ]}, the Minnesota Buyers Health Care Action Group ^{[ 10 ]} and the Colorado Medicaid programme ^{[ 11 ]}, prepared for a 1998 conference ^{[ 12 ]}, illustrated how risk adjustment models blending demographic and health status data were used by private and public payers to address risk segmentation in a variety of ways. In addition, Medicare had had experience in developing and implementing risk adjustment for a variety of demonstrations ^{[ 13 ]}. In each case those implementing risk adjustment faced technical difficulties in data collection and model development, including how to account for geographic variations for purchasers with enrollees in multiple markets. In addition, the support of health plans, employers, and providers was hard-won, since it was clear that some would incur financial losses as a result of risk adjustment. However, taken together, these demonstrations showed that it is possible to overcome both the technical and the political hurdles faced by those confronting the problem of risk segmentation in the health insurance market ^{[ 14 ]}. The variation among the demonstrations arose in part from the different constraints and markets faced by private and public payers and in part from the different goals that each set for health-based payments. For example, state employers (like Washington’s Health Care Authority) faced constraints as public entities, but insured a population that more closely resembled private sector employees than beneficiaries of public programmes. Private employers relied on risk adjustment to maintain choice in the marketplace. On the other hand, Medicaid programmes had the goal of using health-based payments to respond to provider or advocacy pressure or to create an environment where providers taking care of the sickest people could survive economically ^{[ 2 ]}.

By 1998, there was general consensus that risk segmentation in the health care market was problematic. There were several examples in both the public and private markets demonstrating that implementation of health-based risk adjustment was feasible, and an increasing number of states were developing risk adjustment mechanisms for their Medicaid programmes. In addition, HCFA was struggling to meet its mandate to move beyond demographic-only risk adjustment to health-based payments. Therefore, the remaining question was "why have health-based payments not been adopted more widely by purchasers in the private sector?" ^{[ 2 ]} Analysts concluded that it was difficult to develop incentives for the adoption of risk-adjusted payments in the private market. Some employers did not believe that they could adequately assess risk, while others argued that the administrative cost and effort were not worth the relatively small monetary transfers. However, some researchers pointed out that employers really did not know, in the absence of risk assessment, how significant their monetary transfers might be. In addition, while the case studies showed that health-based payments were feasible, they also identified technical obstacles, political issues and research gaps. In particular, purchasers and health plans needed to develop better databases and quality assurance mechanisms, including reliable encounter data. Many researchers posited that once Medicare and Medicaid implemented health-based payments more broadly, other purchasers would begin to adopt them. Once providers and consumers were exposed to health-based payments, it was thought that they might begin to insist that private purchasers implement them as well. And, health plans would probably want to simplify their administrative processes by having similar payment mechanisms across payers. In addition, over time, proponents of risk-adjusted payments hoped to educate purchasers so that they would understand the financial advantages of adequate compensation of health plans for the services they were providing, as well as how health-based payments promoted access and equity. Finally, researchers hoped to be able to provide better information about the relationship between incentives for purchasers and incentives for plans, providers, and consumers and the interaction between health-based payments and insurance market rating restrictions such as medical underwriting and community rating. They also planned to continue the development of risk assessment tools to permit risk adjusted payments for more complex chronic conditions involving the need for social support services and for long term care ^{[ 15 ]}.

Where Are We Now?

Well, we are now four years further down the road. It is time to review the US situation once again and to ask whether risk adjustment has been adopted more widely by the private sector, by Medicaid programmes, and by Medicare. The simple answer is that progress toward addressing the issues identified in 1998 has been mixed.

By late 2000, Richard Kronick PhD, and colleagues at the University of California at San Diego reported that many states and Medicaid health plans were comfortable with the concept of risk adjustment. Kronick noted that there was still some concern that risk adjustment was complicated and difficult to do accurately. However, he added that many of the actuaries working for state Medicaid agencies were conversant with risk adjustment technology and recommended its use to their clients ^{[ 16 ]}.

At the time of the 1998 conference discussed above ^{[ 2 ]}, only the US states Colorado and Maryland had implemented health-based payments. However, as Medicaid programmes increasingly rely on managed care organisations to provide care to their enrollees, the importance of health-based payments grows. As with other payers, if managed care organisations receive capitated payments that are not adjusted for the health status of enrollees, there is a financial incentive to enroll and retain only low-risk, low-cost patients.

By the end of 2000, nearly three years later, eight states (Maryland, Colorado, Oregon, Michigan, Minnesota, Delaware, Tennessee and Utah) employed some type of risk adjustment in their Medicaid programmes. Utah was in the process of moving from a marker diagnosis tool, relying on just a few diagnoses to assess risk, to the chronic illness and disability payment system (CPDS; refer below) and two other states, New Jersey and Washington, planned implementation in 2000 and 2001, respectively ^{[ 16 ]}. At the time of writing this paper, each of those states has institutionalised diagnosis-based risk adjustment for its Medicaid programme. Pennsylvania and Oklahoma are scheduled to implement health-based payments for Medicaid in 2003, and a number of other states are exploring risk adjustment ^{[ 17 ]}.

There are several tools that payers, including Medicaid, can use to make health-based payments. Kronick and his colleagues have developed two such tools, the disability payment system (DPS) and the CDPS. The CDPS is designed to more accurately reflect health expenditures of Temporary Assistance for Needy Families (TANF) recipients in addition to disabled Medicaid beneficiaries ^{[ 16 ]}.

Although the early adopters of diagnostic risk adjustment mostly used state-specific weights (Colorado, Maryland, New Jersey, Michigan and Minnesota), most recent implementations use national weights (Oregon, Pennsylvania, Oklahoma, Tennessee and Utah). Colorado and New Jersey are considering switching from state-specific weights to national weights. National weights are estimated from utilisation in a variety of states. If diagnosis-based risk adjustment in Medicaid is be more widely adopted and become routine, a reliable set of national weights is needed. For most states, it is too expensive and time consuming to estimate state-specific weights, and small and medium-sized states would need to aggregate many years of data in order to do so reliably ^{[ 17 ]}. Currently, Medicaid programme staff and researchers are discussing the need for recalibrating the CPDS weights based on newer data, and on data reflecting managed care utilisation patterns rather than fee-for-service data that have been used to develop the weights currently in use. Some contend that the weights are not likely to change payments significantly, while others point out that new drugs or procedures may lead to large changes in some diagnoses over time and that it might not be acceptable to continue to use weights that are based on fee-for-service utilisation data ^{[ 17 ]}. In any case, it is clear that the implementation of health-based payments for Medicaid programmes is feasible, has been accepted in several states and is being considered in others. There is reason to believe that, as implementation barriers are addressed and national weights are more widely accepted, additional Medicaid programmes will implement health-based payments.

Health-based payments have not been adopted widely among employer groups. To date, public employers and purchasing alliances have taken the most interest in exploring risk assessment and risk adjustment ^{[ 18 ]}. By the summer of 2001, health-based payments had been implemented by the Washington Health Care Authority, the Health Insurance Plan of California (PacAdvantage), The Buyer’s Health Care Action Group and the Missouri Consolidated Health Care Plan. In addition, several other employers were investigating risk assessment and risk adjustments and had run simulations. Employers exploring health-based payments include: the Massachusetts Group Insurance Commission, General Motors, the Pacific Business Group on Health, the Minnesota Department of Employee Relations, General Electric and California PERS (Public Employees’ Retirement System). Various approaches to risk adjustment have been considered by employers, with the choice of model often depending on the availability of the data. Most employers and purchasing coalitions can easily obtain eligibility information and inpatient data. However, they have much more difficulty outpatient data. Therefore, most of the risk assessment models explored by employers have been based on inpatient data alone. Recently, employers have begun investigating the use of prescription drug data for risk assessment and adjustment since that is easily obtained through pharmacy benefit managers ^{[ 18 ]}.

As noted earlier, the Balanced Budget Act of 1997 expanded managed care options for Medicare beneficiaries under the Medicare+Choice (M+C) programme and required that a risk adjustment mechanism be developed and implemented by January 2000 ^{[ 7 ]}. The Health Care Financing Administration (HCFA), now the Centers for Medicare and Medicaid Services (CMS), which administers the Medicare program opted to phase in its risk adjustment mechanism in order to be less threatening to health plans concerned that their enrollees would be identified as "healthier," resulting in payment transfers out of the plan. In addition, the Medicare programme has a long history of implementing payment changes in phases. In January 2000, the Principal Inpatient Diagnostic Cost Group (PIPDCG) risk adjuster was incorporated into the Medicare payment structure for Medicare capitated health plans. Hospitals participating in M+C plans were required to submit encounter data sufficient to determine eligibility and permit computation of the amount that would have been paid under Medicare fee-for-service. In addition, the Medicare programme proceeded with plans to implement a comprehensive prospective risk adjuster requiring collection of outpatient and physician encounter data by 2004, with outpatient data collection to begin in 2001. Most agreed that a risk adjustment approach, relying entirely on inpatient data, could lead to incentives for health plans to provide inpatient care, even when less expensive outpatient treatment might be more appropriate. While CMS considers it important to fully implement a comprehensive risk adjustment model, the agency remains concerned about the burden such a model imposes on health plans ^{[ 19 ]}.

In response to health plan concerns, CMS announced in June 2001 that it had suspended, until 1 July 2002, the required filing of physician and hospital outpatient department encounter data and would continue to explore and implement a risk adjustment process balancing accuracy and administrative burden. In January 2002, CMS convened a public meeting where the agency agreed to accept a "thinner" data set from the health plans. In March 2002, CMS announced that it had "chosen a ’selected significant condition’ model using diagnoses that are ’significant’ because they are relevant for risk adjusted payment and have been included because of their statistical and clinical significance for the Medicare population . . . The model is a refinement of the Hierarchical Condition Category (HCC) model." CMS simultaneously announced that "to implement the new risk adjusted payments beginning January 1, 2004, M+C organizations will need to begin collecting information on the selected diagnoses from July 1, 2002.^{[ 20 ]}"

What Have We Learned?

Researchers, health plans, employers and policymakers have taken great strides toward addressing risk segmentation in the health insurance market. However, despite significant investment in the development of risk assessment tools and strategies for implementation, as well as the funding of both public and private sector demonstrations, health-based payments have not been universally adopted. Employer groups that have implemented risk adjustment mechanisms continue to face issues of data availability. In addition, they must attempt to mesh their risk adjustment goals with their overall health insurance purchasing strategies, including quality monitoring.

In nearly all cases, employers and Medicaid programmes have found that it was important to develop and maintain a collaborative process with the health plans and providers, including sharing of data and documentation. Many states and employer groups also conducted simulations to identify the impact of proposed risk adjustment mechanisms on plans prior to implementation. These simulations also provided an opportunity for the plans to improve their information systems before real money was at stake. States and employer groups, like the Medicare programme, faced challenges in obtaining and using encounter data. And, early in the simulations, the ability to provide encounter data varied among plans. However, once providers understood the importance of collecting complete and accurate encounter data, they made adjustments to their systems to improve their data collection and its quality ^{[ 19 ]}. As we look forward to the eventual implementation of risk adjustment by the Medicare programme, some posit that more employer groups and states will seriously consider health-based payments to become consistent with the major payers. Only time will tell.

References

Footnotes