May 18, 2010
BACKGROUND: Some health plans profile physicians on the basis of their relative costs and use these profiles to assign physicians to cost categories. Physician organizations have questioned whether the rules used to attribute costs to a physician affect the cost category to which that physician is assigned. OBJECTIVE: To evaluate the effect of 12 different attribution rules on physician cost profiles. DESIGN: Under each of the 12 attribution rules, a cost profile was created for the physicians in the aggregated claims database and the physicians were assigned to a cost category (high cost, average cost, low cost, or low sample size). The attribution rules differed by unit of analysis, signal for responsibility, number of physicians who can be assigned responsibility, and threshold value for assigning responsibility. SETTING: Four commercial health plans in Massachusetts. PATIENTS: 1.1 million adults continuously enrolled in 4 commercial health plans in 2004 and 2005. MEASUREMENTS: Percentage of all episodes assigned to any physician and percentage of costs billed by a physician that were included in his or her own profile were calculated under each rule. The cost category assignments from a commonly used default rule were compared with those from each of the other 11 attribution rules and the rate of disagreement was calculated. RESULTS: Percentage of episodes that could be assigned to a physician varied substantially across the 12 rules (range, 20% to 69%), as did the mean percentage of costs billed by a physician that were included in that physician's own cost profile (range, 13% to 60%). Depending on the alternate rule used, between 17% and 61% of physicians would be assigned to a different cost category than that assigned by using the default rule. LIMITATION: Results might differ if data from another state or from Medicare were used. CONCLUSION: The choice of attribution rule affects how costs are assigned to a physician and can substantially affect the cost category to which a physician is assigned.