ACG Products

Tools

• Adjusted Clinical Groups (ACGs): Because many risk adjustment applications are more appropriately performed with mutually-exclusive categories, as might be found in actuarial-cell tables, we constructed mutually exclusive "Adjusted Clinical Groups" which classify the morbidity pattern experiences over a time period for each member of a population can be assigned. ACGs therefore are a taxonomy of population morbidity. To arrive at the ACG cells, the 32 ADGs are first collapsed into 12 categories called Collapsed ADGs (CADGs). The 23 most frequently occurring combinations of CADGs (commonly referred to as MACs) form the main branches of the ACG decision tree. A last branch, called MAC 24, is reserved for those persons with uncommon morbidity patterns. MACs may form terminal nodes or they may be further subdivided using age, gender, or the presence of conditions (as indicated by the presence of specific ADGs). Terminal nodes of the tree, the ACGs, are formed by subdividing the MACs based on clinical and statistical criteria using a recursive partitioning methodology.

• Expanded Diagnosis Clusters (EDCs): CD diagnostic codes are assigned to 230 Expanded Diagnosis Clusters or disease categories. The 190 EDCs are organized into 27 categories called Major Expanded Diagnosis Clusters (MEDCs). As a stand-alone tool, EDCs can be used to select patients with a specific condition or combination of conditions, or to compare the distribution of conditions in one population with another.

• Risk Scores: The ACG-PM is a statistical model that includes ACGs, age, gender, EDCs, and diagnosis-based markers for high likelihood of future hospitalization and for significant levels of activity restriction-"frailty."

  The acgPM permits the rapid identification of high-risk patients who may benefit from care management services. The acgPM remains grounded in the disease burden perspective unique to the ACG System. This system's focus on commonly occurring patterns of morbidity and assessment of all types of medical need has repeatedly proved to have many advantages over comparable case-mix adjustment approaches that are centered only on disease or episode categories. Also, our predictive model's straight forward approach to integrating clinically-relevant risk factors offers advantages over "black box" strategies based very complex or un-meaningful "data-mining" or "artificial" intelligence statistical strategies. None of these complex strategies have been able to establish superior performance to the acgPM.

  The acgPM has two types of outputs for prediction. The first score is expressed as a probability that represents the likelihood that a member will be a high resource user in the coming year. We have defined high resource users as those falling within the top two percent of resource use for the predicted time period. acgPM users can exercise some discretion in setting their own definition of high resource use. There are model performance tradeoffs to be made when selecting this resource use threshold. These are discussed later in this chapter. The acgPM risk score will be especially useful to case managers in identifying patients for targeted intervention.

  The second output of the acgPM model is the estimated year two costs expressed in terms of relative weights. These weights can be used to calculate expected resource dollars. Health plans and others will find the model a useful tool for rate setting and financial related decisions.