While one year of claims data may allow for detection of acute care episodes, it may omit lengthy or complex episodes, particularly if the profiling algorithm truncates those cases that show no end date within the contract year or capture only episodes with clean periods at both ends.
One year of data is also likely to omit those patients who consistently incur high costs from year to year, whether because of severe and persistent illnesses, or due to high-frequency moderately resource-intensive service needs. Analysis of three years of claims and exposure data from the Society of Actuaries medical claims study (Grazier and G’Sell, 2004) indicates that for claimants with annual claims expense of more than $25,000, over 13 percent have annual claim costs in the subsequent two years of over $25,000; for those with annual claims in one year exceeding $50,000, almost 25 percent have total annual claims exceeding $25,000 in the subsequent year; and for those with annual claims cost exceeding $100,000, over 30 percent have claims exceeding $25,000 in subsequent years. While these data are for patients and not per physician, the effect of such cases on a panel from one year to the next could be misinterpreted if multiple years of data were not captured in the algorithms.
More than one year of data would be needed to establish a fuller picture of use, and to accommodate “clean periods” for episodes that span the limits of inforce coverage contracts or reflect care for chronic conditions. In the White Paper released by Bridges to Excellence (2004), authors recommended “at least two years of data, based on incurred claims” to “develop a statistically reliable determination of provider efficiency.”
Recent research on measuring efficiency and quality has used administrative claims and member data either from commercial carriers or employers, or beneficiary claims data from fee-for-service Medicare. Because of the different payment models reflected in these data sets, care should be taken to ensure internal and external consistency. Within commercial population data, health mainenance organziation (HMO), exclusive provider organization, preferred provider organization (PPO), and traditional indemnity covered care may be captured differently. For instance, HMO encounter data may not incorporate professional fees with the inpatient/hospital records, while traditional-coverage-generated data may have both. Commercial claims data cannot be directly combined with Medicare data, without adjusting for beneficiary, coverage, and charge differences across the payers.
Most claims systems used by commercial carriers or those developed in-house separate pharmacy data systems. If quality is to be incorporated into efficiency measurement, then pharmacy data should be incorporated into the measurement and assessment of the appropriateness of resources (Goldman et al., 2004). If it were combined, then pharmacy data can be edited and aggregated and then linked by unique member identifiers across commercial data sets. If comparable pharmacy data are not available, such as in