ists,” attributing responsibility for services across the providers is usually based on a formula. These formulas differ in their attribution decision rules, and vary the amounts of resources assigned to a responsible provider proportionally or nonproportionally to the primary care, nonprimary care, or total resources consumed across the episode.
The data sources for these efforts have traditionally included encounter and claims data supplied through an employer, insurer, or plan’s administrative data systems. In some cases, the administrative data have been validated against medical records, but these efforts have been inconclusive in determining which source is better than another for these purposes (Hannan et al., 2003). Claims or encounter data at this time are generally more accessible and less expensive to analyze than medical charts or patient surveys, although efforts to identify quality and value metrics continue to explore these sources as well as electronic medical records and online order entry systems (Birkmeyer et al., 1999, 2002, 2003; Fisher et al., 1990a,b, 1992; Malenka et al., 1994; Thomas et al., 2004a).
Different types and amounts of data can be extracted from the same claims data set (Baron et al., 1994; Fisher et al., 1992). Many profiling tools capture and use in their algorithms different numbers of diagnoses, procedures, and different time periods for services. Current episoding algorithms vary in the numbers of episode categories to which diagnoses and procedures are assigned. They also differ in the length of the “clean periods,” those time periods during which no services for the condition are received, thus triggering the end of one episode and the beginning of another.
It is common in profiling methods to aggregate all costs of care that appear with an episode and attribute this total to a provider. But there is also variation in the complexity or severity of the case or in patient characteristics that are not captured in episode categories defined by time of service (Iezzoni et al., 1992a,b, 1994b). Several risk adjustment methods that have been perfected for other purposes as well as for physician efficiency profiling are applied to episodes to explain better the resources identified as inputs in the model.
Risk adjustment is used to adjust claims profiles to account for differences in the health status (and thus expected resource use) of patients served. Without proper adjustment, practice patterns of physicians whose patient panels include greater than average proportions of elderly patients or