Online Program

Bundled payments to primary care physicians using a risk-adjusted Primary Care Activity Level (PCAL) model
Arlene Ash, Umass Medical School 
*Randall P Ellis, Boston University 


We examine alternative specifications of paying primary care professionals (PCPs) a bundled payment intended to reflect the full cost of providing high quality primary care to their panel of patients. We empirically examine eight different measures of resource need for primary care which differ in the breadth of services included. The original concept we develop is called the Primary Care Activity Level (PCAL). It is the sum of all primary care core services provided by the patient’s own primary provider, plus a share of the primary care services provided by other PCPs, plus smaller proportions of the costs of specialist, emergency department, inpatient, and pharmacy services. The rationale for this broad measure is that patients that require more of these services will also require more attention from their primary care provider.

We predict eight measures of primary care resource need (PCAL) using diagnostic information summarized using the DCG/HCC classification system, and all models are calibrated using a 2006 sample of 1.49 million commercially insured individuals in the Medstat MarketScan dataset. We link patients to a PCP using an algorithm that assigns patients based on the primary care specialty and proportion of primary care services provided, and patients not seeing a primary care doctor are also allocated to PCP using an assignment algorithm using the patient’s county, age, gender. Individual PCAL measures are summed up to the practice level and evaluated for their predictive power and correlations. Actual practices are also summed up to Pseudo practices

Our results show that patients receive nearly a third of core primary care services from someone other than their most commonly used PCP. The cost of primary care services is relatively highly predictable (R2=.37 at the individual level, and .93 at the practice level). Possible PCAL payment models are very highly correlated (?=.98) with each other, suggesting that actual payments will be relatively robust to the choice of what to include or exclude in the PCAL measure. Scatter plots and summary statistics suggest that the proposed PCAL models will result in adequately low risk to individual practices as to be acceptable.