Minnesota began a prepaid managed care demonstration, under an 1115 waiver (Weiner et al., 1998), for its public health care programs in 1985. A key component of this demonstration is that the State pays participating managed care organizations a fixed, prepaid premium or monthly capita tion payment for each health plan enrollee. Capitation is defined as a contract arrangement whereby a purchaser agrees to pay health plans a fixed payment per capita/enrollee per month in return for which health plans assume responsibility for the provision of all covered services for their enrolled populations (Hurley, Freund, and Paul, 1993). Health plans are then effectively at risk for additional health care costs that exceed capitation revenues.
There are formally three distinct prepaid public program populations administered by the State of Minnesota: (1) the Prepaid Medical Assistance Program (PMAP), (2) Prepaid General Assistance Medical Care (PGAMC), and (3) Prepaid Minnesota Care (PMNC). PMAP is Minnesota's Medicaid managed care program that serves low-income residents, including Aid to Families with Dependent Children (AFDC)-eligible families, pregnant females, children, and the elderly. PMAP operates in 83 of Minnesota's 87 counties and continues to expand.
PGAMC is a managed care program that serves low-income Minnesota residents who do not qualify for PMAP or other State or Federal health insurance programs. PGAMC primarily serves single or marfled Minnesota residents between the ages of 21 and 64 who have no children.
PMAP and PGAMC were implemented in 1985. There were approximately 249,000 PMAP enrollees and 28,000 PGAMC enrollees in calendar year (CY) 2000.
PMNC is a subsidized insurance program for Minnesotans who have somewhat greater assets than people eligible for PMAP or PGAMC, but no other access to public or private health insurance. PMNC began in October 1992 and converted to a prepaid managed care program in 1996. PMNC is jointly funded by the Federal Government, a 1.5-percent tax on health care provider revenues, and enrollee premiums, which are assessed on the basis of a sliding scale. Approximately 165,000 persons were enrolled in PMNC in 2000. This article focuses exclusively on the PMAP and PMNC programs.
From the inception of these programs and until January 1, 2000, capitation payment rates have been based on combinations of the demographic characteristics, age, sex, prepaid program, region, institutional status, Medicare coverage status, pregnancy status, parental status, and family income. Rates for some programs were further adjusted by region. Over time, experts and policymakers have become increasingly concerned about biased selection, the profit realized by enrolling healthier, low-cost enrollees, and the financial penalties that can result for health plans from enrolling sicker, high-cost enrollees, and generally the inadequacy of demographic-based capitation ratesetting systems (Newhouse et al., 1989; Fowles et al., 1996).
As a result, it has become apparent that the demographic basis of capitation rates does not sufficiently reflect the relative health based risk of prepaid populations. To address this need, a number of methods of measuring population health status have been developed in recent years, supported in part by the Federal Government for possible application to the Medicare Program (Weiner et al., 1996; Ellis et al., 1996; Medicare Payment Advisory Commission, 1998).
Risk-Adjustment Mandate in Minnesota
Proposals to set capitation rates for these programs on the basis of health stares-based risk adjustment first surfaced in the Minnesota health policy arena in 1993, as a part of State and national level health care reform proposals (Minnesota Departments of Health and Human Services, 1996). In 1994, legislation was passed that required the submission of a report by the Minnesota Departments of Health and Commerce regarding the implementation of risk adjustment. The report, submitted to the Minnesota legislature in early 1995, recommended the development of a risk-adjustment mechanism for the PMAP, PGAMC, and PMNC programs. Legislation passed during that session required the Minnesota Departments of Health and Human Services to jointly develop the risk-adjustment system, in consultation with a stakeholder advisory committee--the Public Programs Risk Adjustment Work Group (PPRAWG).
Risk-Adjustment Method Evaluation Criteria
The main criterion by which we chose to evaluate risk-adjustment options reflected the fundamental objective of this initiative: to improve the accuracy with which capitation payments are targeted to the illness burden of prepaid populations. Given this objective, a secondary priority was to minimize the administrative impacts of adopting risk adjustment. Given these priorities, the two major criteria by which we chose to evaluate alternative risk-assessment methods were: non-random or skewed group predictive performance and administrative feasibility (Minnesota Department of Health and Minnesota Department of Commerce, 1995). Throughout our evaluation, however, we also monitored individual level predictive performance (i.e., [R.sup.2] statistics) for any inconsistencies with non-random group-level results.
Non-random/skewed group predictive performance refers to the relative degree to which alternative risk-adjustment methods can produce unbiased predictions for skewed or non-random subpopulations (Dunn et al., 1996). Methods that under-predict for certain non-random subpopulations and overpredict for others create clear incentives for risk selection, to the extent that the subpopulations are readily identifiable by participating health plans.
By administrative feasibility, we mean the degree to which alternative methods can be implemented efficiently. To maximize administrative feasibility, the State chose to build on existing work in this area, and focus on models and methods that would utilize existing claims, encounter, and eligibility data systems, or minimize changes in those systems wherever possible. In fact, key features of the risk-adjustment model implemented for these programs resulted directly from the adaptation of risk-adjustment methods to the existing capitation ratesetting systems.
PAYMENT MODEL DEVELOPMENT AND KEY CHARACTERISTICS
Selection of Population-Based Method
The State considered two basic types of capitation risk-adjustment methods: (1) targeted conditions methods, and (2) population-based methods. Targeted conditions methods essentially focus on a small subset of specific clinical conditions that are usually treated in inpatient settings, and account for a small fraction of a given population and a significant, but relatively small proportion of expenditures (Managed Risk Medical Insurance Board, 1995). The principal advantage of a targeted conditions method is that it would require the submission of the number of recipients with each of the targeted conditions only each year. This is significantly easier than if conditions typically treated in ambulatory settings were included, because inpatient diagnosis data are usually more accessible. One important disadvantage of such a method, however, is that it creates an incentive to find recipients with the targeted conditions, and reduces incentives for ambulatory care.
Population-based diagnosis classification methods, on the other hand, focus on entire populations. That is, when assessing the risk of any given entity (e.g., enrollees of a benefit plan or health plan), all enrollees would be included in the assessments, including those for whom there was (1) no health care utilization, (2) only ambulatory utilization, or (3) both inpatient and ambulatory utilization during the year. An important advantage of these methods is that they minimize incentives to avoid enrolling people with particular types of conditions, because all enrollees are assessed and a comprehensive array of both ambulatory- and inpatient-based conditions are included. However, one disadvantage is that they require participating health plans to maintain both inpatient and ambulatory data management systems.
Nevertheless, in 1997 the Minnesota Departments of Health and Human Services recommended that the risk-adjustment mechanism should be based on a population-based diagnosis classification method. Once the State narrowed the scope of required encounter data to elements that were uniformly collected across health plans, the participating health plans indicated their support for this recommendation through the PPRAWG.
Selection of ACG Case-Mix System
Three diagnosis classification systems were initially proposed as candidates for the risk-adjustment model--ACGs (then known as ambulatory care groups) (Weiner et al., 1996), diagnosis cost groups (Ellis et al., 1996), and the disability payment system (Kronick et al., 1996). After some discussion, key stakeholders supported this proposal. Based on considerable preliminary testing we conducted, and studies conducted by others that have addressed the relative predictive accuracy of these methods (Dunn et al., 1996), staff recommended the Johns Hopkins ACG case-mix system as the basis of the risk-adjustment model. Based on those recommendations, and that the ACGs were the most widely used population-based case-mix system, the Minnesota Commissioners of the Departments of Health and Human Services recommended to the legislature that the ACG case-mix system be the basis of the capitation risk-adjustment payment model in the Minnesota prepaid public programs (Minnesota Department of Health, 1998).
At the time, within the ACG case-mix system there were two options for diagnosis classification: ambulatory diagnostic groups (ADGs) and ACGs, now known as adjusted clinical groups (Johns Hopkins University School of Hygiene and Public Health, Health Services Research and Development Center, 2000). The ADG classification system is a multivariate system, whereas the ACG system is a mutually exclusive categorization system. The ACG system was selected primarily because it could be implemented with only minor changes to the existing capitation payment system managed in Minnesota. A model based on ADGs, however, because it would have been a multivariate, additive model, would have required significant changes to those systems. ACGs could simply replace the existing demographic rate cells as new rate cells in the existing payment system.
As of January 1, 2000, Minnesota …