As the United States health care system struggles to recognize and improve the health of the population through management and policy reforms, we look to primary care for guidance. Decades of experimentation in primary care delivery and payment illustrate the correlation of good care and good outcomes, the importance of compensating providers of all types for efficient use of high-value resources, and the value of rewarding behaviors that improve individual and population health outcomes.1
Alignment of payment for services with the goals of care is complicated by the increasing responsibility assigned to primary care providers. Primary care is increasingly held accountable for delivering care while coordinating people, services, and agencies to provide and manage public health, social service, health education, juvenile justice, chronic disease management, and end-of-life care. Services are diverse and may include evaluation, screening, counseling, and pain management, as well as minor surgeries and major invasive procedures. The recipients of these services may be relatively healthy, benefitting mainly from preventive care, or they may suffer from serious and chronic illness. Those served experience an intensity and severity of illness rivaling that experienced in many subspecialty settings. “The severity of depressive symptoms in patients who receive treatment in primary care is equivalent to that of patients treated in psychiatric settings. For example, approximately 43% of such primary care patients report some degree of suicide ideation within the previous week.”2 Highest risk patients, as identified by the Duke Health Profile and the provider-completed Duke Severity of Illness Checklist, accounted for more than a quarter of primary care clinic patients.3 This growing responsibility and accountability should be shared with others able to provide comprehensive primary care, such as other types of clinicians, teachers, peers, families, and community members.
For a comprehensive primary care system to deliver services resulting in a healthy population, several components are essential, including detailed and comprehensive information on costs and risk, systems to manage that data, and payment models compensating providers for actual costs and creating transparent consequences for providers, patients, and communities.4,5 In this issue of Medical Care, Ash and Ellis use available data to predict costs for a defined set of primary care services delivered by clinicians, while risk adjusting the payments and performance rewards for patient characteristics and outcomes. This research sheds light on the importance of understanding the individual patient, adding to our awareness of the challenges we face in addressing community needs, particularly among those who do not use services.
Interesting questions arise when considering the application of this work. If a primary care practice, as structured today, were to rely on the Ash and Ellis model to predict its costs per bundle, provider or practice, what is the likelihood that the clinician or practice manager could differentiate individual need and resource use, adequately manage the care, and manage the financial risk? What is the extent of predictive error across categories of risk, for example, among patients in the highest quartile of risk? A manager may be better able to recruit appropriate personnel or engage communities, to assure responsive care and good outcomes, if she knows where or at what levels the prediction underestimates costs or overestimates resource intensity.
Ash and Ellis use claims data and expert judgment to create a proxy for total costs for a bundle of services they define as primary care. In itself, this exercise is a contribution to the field. Generating, accessing, and accumulating actual costs across communities and providers remain elusive, but such information is necessary to capture the total resource use, by specialty, involved in our current and expanding primary care system.6 These data are particularly crucial if we are to approach an ideal model reflecting the true nature and functions of primary care, including the underlying integration and communication critical to individual and population health.
As the mix of primary care consumers reflects changes in sociodemographic status, engagement, and morbidity, and as new roles are developed for providers of primary care, managers need to consider how these changes impact the actual use and type of services, and their trajectories. In addition, how would new measures replace or enhance the risk adjustment and its results? Recent research using data on Taiwanese National Health Insurance beneficiaries found that “a simple morbidity trajectory classification over a 3-year period is almost as powerful predictor of prospective medical utilization as more comprehensive baseline risk adjustors.”7
Payers have similar incentives to get the payment right. Pricing the bundles of services based on the proposed models remains challenging, not only because of the changing mixes of services and providers but also because of the need for managers to collect the data and apply the model to their practices. Payers and providers are similarly concerned with intended and unintended consequences, such as incentives and disincentives resulting from the payment formula, implications of the timing and apportionment of payment, and potential shock effects of change implemented too quickly.8
Reformulating bundled payment with risk adjustment, to further reduce the risks of inadequate prediction, can protect patients and providers from risk variation. Compelling work by Newhouse and colleagues favors a blended model of payment that recognizes risk-adjusted predicted costs but with retroactive payment for unexpectedly high costs.9,10 These blended rates encourage attracting both low and high risks.
The authors propose using their model to estimate performance on patient outcomes and incentivize high-value services delivered efficiently. The use of bonuses is widespread across industries and should be considered for promoting the cost, value, and performance objectives of primary care and the behaviors that maximize health and prevent illness. Goroll et al11 proposed performance-based bonuses added to comprehensive bundled payments and detailed the administrative and statistical methodology. Ash and Ellis propose using similar models based on the same data to estimate both expected costs and expected outcomes. If we use the expected outcomes to determine bonuses at the same time that we use the expected costs to determine payments, we should closely consider the independence of errors in prediction from these 2 models. Without more information, one might expect that we might tend to underestimate costs for the same patients for whom we underestimate negative outcomes. If so, for those patients we will have a tendency to both underpay the provider and unfairly distribute bonuses. Similarly, if for some other set of patients we overestimate both costs and negative outcomes, the insurer will have a tendency to overpay the provider and award undeserved bonuses. In this sense, the errors in bonus determination would align with errors in cost estimation.
Such dependence of the errors could cause bonuses to amplify problems caused by inaccurate cost predictions, potentially harming attempts to adjust for risk and incentivize proper treatment. In this light, it is important not only to examine prediction errors on subsegments of the population for both models but also to examine these errors jointly to look for systematic errors across models. Otherwise, bonus programs could do more harm than good in our attempts to properly align incentives in primary care. Use of the proposed model in concert with other incentives schemes, such as bonuses, needs to be carefully evaluated and monitored.
The purpose of bundled payment with risk adjustment is to minimize rejection of patients and to promote delivery of high-value services, such as preventive services, screening, and care coordination for chronic illness. Experiments using these methods provide important information for expanding models to communities populated by those who use care, but should not be penalized for being more complicated cases or more chronically ill, and by those who do not use services but perhaps should. Tying the methods to performance incentives should also be tested in the same communities to better understand how incentives can be aligned to benefit public and individual health.
Transformation of the primary care system in the United States is accelerating. As noted by Crabtree et al,12 the mechanistic and traditional view of primary care stifles progress in quality improvement and integration of services. “A theoretical perspective that views primary care practices as dynamic complex adaptive systems…” provides a better framework for conceptualizing comprehensive service delivery and payment models and designing creative, yet feasible, systems for improving care and health.12 Organizations within and across communities, cities, and regions are working examples of how this perspective results in improved processes and outcomes in housing, social service delivery, juvenile justice, education, employment, and health care. In these examples, financing and payment rely in most cases on bundled services, with sensitive risk adjustors, and retrospective adjustment for unexpectedly high cost care (K.L.G., unpublished data, 2012). Comprehensive integrative systems providing the mix of services and people necessary to promote health to individuals and communities are under development.13 These types of integrated service models, engaging communities, are critical to the future of health care and improved population health. Their success depends on a payment system designed to be sensitive, reflective, and innovative. The Ash and Ellis contribution takes us another step toward recognizing the benefits and limitations of currently available administrative data and predictive modeling, and furthers necessary transformations to make primary care the foundation for health.
1. Reschovsky JD, Ghosh A, Stewart K, et al. Paying more for primary care: can it help bend the Medicare cost curve? Commonwealth Fund Issue Brief Publication, 1585, Vol. 5, March 2012
2. Parkerson GR Jr, Hammond WE, Michener JL, et al. Risk classification of adult primary care patients by self-reported quality of life. Med Care. 2005;43:189–193
3. O’Connor EA, Whitlock EP, Beil TL, et al. Screening for depression in adult patients in primary care settings: a systematic evidence review. Ann Intern Med. 2009;151:793–803
4. Merrell K, Berenson RA. Structuring payment for medical homes. Health Aff. 2010;29:852–858
5. Shen Y, Ellis RP. Cost-minimizing risk adjustment. J Health Econ. 2002;21:515–530
6. Bodenheimer T, Berenson RA, Rudolf P. The primary care–specialty income gap: why it matters. Ann Intern Med. 2007;146:301–306
7. Chang HY, Clark JM, Weiner JP. Morbidity trajectories as predictors of utilization multi-year disease patterns in Taiwan’s National Health Insurance Program. Med Care. 2011;49:918–923
8. Bodenheimer T, Pham HH. Primary care: current problems and proposed solutions. Heath Aff. 2010;29:799–805
9. Newhouse JP, Buntin MB, Chapman JD. Risk adjustment and Medicare: taking a closer look. Health Aff. 1997;16:26–43
10. Newhouse JP. Risk adjustment: where are we now? Inquiry. 1998;35:122–131
11. Goroll AH, Berenson RA, Schoenbaum SC. Fundamental reform of payment for adult primary care: comprehensive payment for comprehensive care. J Intern Med. 2007;22:410–415
12. Crabtree BF, Nutting PA, Miller WL, et al. Primary care practice transformation is hard work: insights from a 15-year Developmental Program of Research. Med Care. 2011;49:S28–S35
13. Hofer AN, Abraham JM, Moscovice I. Expansion of coverage under the Patient Protection and Affordable Care Act and Primary Care Utilization. Milbank Q. 2011;89:69–89