A value-based healthcare payment structure in the U.S. ties reimbursement to quality, not just volume. One such model, Episodes of Care, defines common health conditions by time windows and medical codes. Many managing care organizations rely on grouper services to organize their claims into episodes, where they implement their payment incentive structure for each U.S. state. Some grouper services assign responsibility for an episode to a managing physician. This format provides a unique opportunity to study physician variability, with the motivation of understanding what drives episode outcomes such as cost or readmittance.
This work studies physician variability from claims data using diagnosis, procedural and pharmaceutical codes. Claims are grouped by patient and then ordered by date of service, resulting in a single time-ordered sequence for each episode. Pair-wise comparisons using string-matching algorithms follow. Episodes are clustered based on distances from the different algorithms. Post-hoc analyses on the clusters characterize the driving mechanisms of episode outcomes as well as extract optimal treatment paths. This work examines the impact of algorithm selection.