Impact of an early identification model for oncology case management.

Abstract

e13568 Background: Evidence shows that cancer case management improves patient’s quality-of-life and reduces hospital admissions [1] . When case management (CM) is done by payers, there is a start delay since case managers must wait for the diagnosis information through authorization to start a case. This delay hinders the case manager’s ability to support patients early in their cancer journey as they are making impactful treatment decisions. Researchers at Evernorth Health Services created an Early Identification (EID) machine learning model that allows Evernorth to identify possible breast oncology case management-appropriate patients early. The goal is to shorten the time from diagnosis to engagement in oncology CM. We evaluated how the model impacts CM program metrics. Methods: Implementation of the breast oncology EID program started on April 2022. The list of potential patients identified by the model is sent to case managers on a weekly cadence for outreach and engagement. If the diagnosis is confirmed, they are referred to the oncology care management program. We compared the likelihood of engagement with CM for patients referred to us by the EID model vs. those referred to us by other means (e.g., prior authorization, other case management programs, etc.). We included patients identified by the model between April 2022 and December 2023. We had 3,484 patients in the EID sample and 3,750 in the non-EID initial sample after exclusions. We used Coarsen Exact Matching (CEM) to adjust for demographic, plan type and disease differences between the two groups. In addition, we compared behavioral health integration (measured as having a claim for behavioral health services after engagement in case management) and CM program completion (measured as completing the case goals) for those identified through EID and engaged in CM vs. those engaged in CM and referred to the program through other means. Results: The unadjusted average difference in time from diagnosis to the opening of the oncology CM case is 10 days. The adjusted results show a 21% statistically significant increase in likelihood of engagement for EID identified patients. For engaged patients, the results show a slightly lower rate of CM program completion, but the result is not statistically significant. Finally, EID patients engaged in case management have a 40% statistically significant increase in risk of getting a behavioral health claim after engagement. Conclusions: The Evernorth EID model demonstrates a positive impact on engagement in CM for breast oncology patients as well as a positive impact on behavioral health integration for engaged patients. The model has no significant impact on CM program completion.