Understanding Practice Variation With A Clinical Pathways System: Differences By Physician And Practice Factors, And Changes In Practice Over Time.

2079 Background: Clinical oncology pathways aim to support clinical decision-making and reduce unwarranted practice variation across an enterprise. The Dana-Farber Cancer Institute (DFCI) implemented web-based oncology pathways with DFCI-customized content in each disease center and at each of its satellites. Our pre-specified aim was an on-pathway rate of 70-85%. Methods: Treatment decisions were electronically captured as on- or off- pathway. Monthly metrics about usage and on-pathway rate were shared with users on a monthly basis. Physicians were categorized into quintiles based on the calculated on-pathway performance during the first 90 days of each individual’s use of the platform. On-pathway rates were then calculated for days 91-360 to study changes in behavior over time. Physician and practice factors were examined to determine any differences by initial on-pathway quintile classification. Results: 122 physicians were eligible for inclusion in this analysis (minimum 5 navigations in each study period). On-pathway rates showed significant variability in the initial 90-day period: quintile 1 median 100%, quintiles 2-4 80.2%, and quintile 5 50% (Table). In the follow-up period, median on-pathway rates shifted into the pre-specified goal range for all groups. Physicians in quintiles 1 or 5 of initial on-pathway rate were more likely to have fewer total navigations than were physicians in quintiles 2-4 (p=0.003). While no other physician or practice characteristic differed significantly by on-pathway rate group, physicians in the first or last quintile were more likely to be in an academic setting, have a PhD, or navigate fewer pathways. Conclusions: Over time, the deployment of a web-based clinical pathways program resulted in greater uniformity in physician practice, based on on-pathway rate. Familiarity with the pathways platform and its navigation, monthly feedback about usage, and evolution of content over time are some factors that might have played a role. [Table: see text]