Our analysis of commercially available risk prediction algorithms found minimal differences among the various products’ ability to predict high cost patients in the following year. A majority of the algorithms are exclusively claims based, though some include the ability to augment risk predictions with clinical data. Clinicians have played a critical role in improving our ability to identify high-risk patients. We have found that when physicians review a pre-selected set of their own patients, they have some ability to discriminate among patients who are likely to benefit from care management and those who are not. Clinicians are prone to overemphasizing recent events, but review of a list of patients who are predicted of becoming high cost mitigates this problem. Commercially available algorithms can help create an initial list, but physicians can add perspective on such important factors as social support and executive functioning. This additional information improves the specificity of the initial algorithm outputs, allowing clinicians to play an important role in refining the lists of patients eligible for high-risk care management.
The high cost of labor and space make high-risk care management programs among the most costly programs for an ACO. Care management requires a skilled nursing workforce (among others), which should be embedded into the primary care office for optimal effect [6,7].Given the high costs, there is understandable pressure to increase the ratio of patients per care manager. We have found that the optimal ratio is approximately 200 patients per care manager, with a third of the patients having active complex care management issues, a third being passively surveyed, and a third requiring modest care coordination. We continue with our attempts to refine how we manage this critical aspect of care management programs.
How can managers demonstrate that the investment in care coordination is impacting the ACO’s TME trend? Demonstrating a return on investment is difficult because a population of high-cost patients will inevitably show reduced costs in the following year (a phenomenon called regression to the mean). Isolating a well-matched control group to demonstrate program effectiveness would have the unintended consequence of reducing the potential effectiveness of the program. This situation is complicated by the different risk profiles of high-risk patients in different payer categories. For example, potentially avoidable Medicare costs are dominated by hospitalizations and end-of-life issues, Medicaid costs by mental illness and substance abuse, and commercial costs by specialty issues. In lieu of better management tools to assess the performance of our program, we have depended to date on process measures (eg., enrollment targets), patient surveys, and we are experimenting with some limited outcomes metrics (eg, admissions/1000).
Mental Health Integration
Another important lever for medical trend reduction within an ACO is the integration of mental health services into primary care. While our efforts in this complex area are only about a year old, some of our early lessons may prove valuable to others. First, we have worked hard to make the case for investment in mental health services, requiring assembling the evidence both for the magnitude of the problem as well as the effectiveness of available solutions.