The goal of primary care payment reform is to achieve “better value—defined as better outcomes at less cost— … by rewarding physicians for prevention and coordination rather than volume of services.”1
While many have argued the importance of risk adjustment for calculating bundled payments and bonuses for good performance,2-4
little guidance exists regarding how to do so. This paper addresses that gap.
In a 2010 survey, 25 of 26 patient-centered medical home (PCMH) pilots principally relied on fee-for-service (FFS) payments, typically augmented with a management fee under ten dollars per member per month.5
The fee is often slightly higher for the “very sick” than for others, as in the AMA’s 2008 RUC calculations for Medicare’s 2008 Medical Home Demonstration Project.6
Reforms envisioning larger bundled payments typically acknowledge the need for stronger risk adjustment. For example, the Center for Medicare and Medicaid Innovations 2011 Comprehensive Primary Care Initiative (CPCI) retains FFS payments supplemented with a management fee averaging up to $20 per-patient-per-month, and ranging between $60 and $480 per-patient-per-year, depending on the patients’ CMS-HCC score.7
The CPCI also proposes significant bonuses, to be calculated based on shared savings and performance measures. A more radical reform proposed by Goroll et al2
would replace all primary care FFS income with comprehensive monthly bundled payments plus substantial performance-based bonuses. Our current work directly supports the Goroll framework. Its monthly payments are neither intended to cover all services (full capitation), nor to be just an add-on to existing fee-for-service revenues. Rather, we sought to develop a principled approach to computing the “primary care activity level” (PCAL) needed, that is, the cost of all services that primary care practitioners (PCPs) should
provide. These payments rightly vary hugely between fundamentally healthy and highly complex patients.
Although we focus here on primary care payments, our approach is relevant to many other settings. An accountable care organization (ACO) could use our PCAL calculation to set budgets and incentives for its PCPs.8
Or, a model of the outcome “PCAL minus FFS reimbursement” could be used to calculate risk-adjusted case-management supplements to FFS, as proposed for the CPCI. Our paradigm also aligns well with the goals of value-based purchasing, and can be used to produce a risk-adjusted expected value for any population-based outcome that can be modeled in existing large databases.9,10
Our bundled payment model was implemented in 2009 by the Capital District Physician’s Health Plan (CDPHP), a not-for-profit, network-model, physician-guided health plan with 350,000 members concentrated in upstate New York. CDPHP implemented an early version to pay 3 PCMH pilot practices for their CDPHP patients (private HMO, private non-HMO, Medicare Advantage, and Medicaid HMO enrollees) in January 2009.11
This pilot was organized as a “virtual all-payer” system, in that CDPHP financed practices to implement the PCMH as if CDPHP had insured all their patients.3
Another leg of envisioned reform is outcome-based bonuses. Goroll and others have called for large risk-adjusted bonuses (up to 25% of total income) for achieving desired outcomes in cost, quality, and patient experience.2
Although using non-adjusted performance measures may create undesirable incentives for practices to avoid the sickest patients, even crude adjustments are rare.12-15
Here, we explore the importance of risk adjustment for assessing provider performance and examine our models’ performance for patient panels assigned to primary care practices. Our approach is population-based and empirical; it seeks to encourage improved management and outcomes for whole persons. Risk adjustment rewards practices when their patients’ outcomes are better than expected
. Here, “what-is-expected” reflects patient-specific normative relationships calculated via a model. When the model is refit to new data, the norm shifts to reflect the “new normal;” thus, as a delivery system improves, “the bar” rises with it. These models currently rely only on age, sex, and claims-based diagnoses to define both predictors and outcomes. Soon, electronic health records and patient surveys must also be used, both to include non-medical factors as predictors, and quality- and patient-experience data as outcomes.
While bundled base payments allow a PCP to allocate resources efficiently, bonus payments can directly discourage low-value services and encourage activities that promote clinical quality, patient well-being, and satisfaction. Risk-adjusted bonuses are intended to ensure that each practice can earn rewards for doing a good job with its patients, and to mitigate incentives for cherry-picking easy patients and dumping difficult ones.
For each performance measure, we first build a patient-level model to predict its associated outcome from patient characteristics (age, sex and diagnoses). A practice is judged by comparing its patients’ aggregated observed outcome (O) to its model-based expected (E), or predicted, level. We acknowledge, but do not address here, the many issues associated with separating “signal” from “noise” when judging single practices on individual outcomes, or when creating a useful composite score (leading to a practice-level bonus payment) based on multiple measures.13,16,17
Our aim is to demonstrate the feasibility of risk-adjusted performance assessment, and its importance, given that fixed targets punish good providers whose complex patients, even if doing “better than expected,” don’t hit targets that are easier to achieve with healthier patients.