clinical groups (ACGs, formerly referred to as ambulatory care groups) and diagnostic cost groups (DCGs) (Newhouse et al., 1997). Both methods use diagnostic information to improve predictability as compared with the demographic adjustments used by Medicare to pay health plans.
ACGs were developed at The Johns Hopkins University to classify risks by using diagnoses reported in ambulatory visits (Starfield et al., 1991). The system assigns diagnoses to a risk group based on five clinical dimensions, such as a condition’s duration, severity, and diagnostic certainty. Both inpatient and ambulatory versions are available.
Initial development on DCGs was conducted through a consortium of researchers at Boston University, Health Economics Research, and the Harvard University School of Medicine, and is based on inpatient diagnoses for prior hospitalizations. The DCG model has been expanded to include both inpatient and ambulatory information and to account for multiple medical conditions patients may experience (Ellis et al., 1996).
Another method is clinical risk groups (CRGs), developed by 3M Health Information Systems (Averill et al., 1999). This clinical classification system assigns each patient to a risk group that relates past clinical information to the amount and type of health services the individual will consume in the future. Additionally, a survey-based approach has been developed at Kaiser-Permanente Health Plan for the working-age population. This method uses a chronic disease checklist and self-reported health status and functional status (using the RAND SF-36) to assess health risk (Hornbrook and Goodman, 1996).
The challenges involved in developing a fair and adequate risk-adjustment system cannot be underestimated. All current methods are limited in their ability to predict variation in expenditures. The Health Care Financing Administration implemented a transitional risk-adjustment system in January 2000 using a form of the DCG model, and is moving forward with an expanded model that will include inpatient, hospital outpatient, and physician encounter data. A number of models for this more comprehensive approach are being considered. It should be noted that improving risk-adjustment methods will likely necessitate more clinical information (rather than just claims information), which in turn will require significantly improved information systems (Dudley et al., 2000).
The goals for risk adjustment need to be balanced with the goals of quality improvement. In risk adjustment, the objective is to identify the subpopulation at risk of high utilization and high cost, whereas in quality improvement, the objective is to identify all patients with a particular condition who could benefit from treatment (Dudley et al., 2000). On the one hand, if there is a potential for higher payment, health care organizations are likely to identify as many at-risk patients as possible and collect whatever information is necessary. On the other hand, doing so could bias quality measures through possible upcoding and might not provide incentives to design efficient and effective systems of care. The potential for payment methods to be based on similar patients with common conditions,