Can machine learning help predict the risk of emergency department visits?

Summary of: Coombs LA, Orlando A, Adamson BJ, Griffith SD, Lakhtakia S, Rich A, Shaw P, Wang X, Miksad RA, Mooney K. Prospective validation of a clinical tool developed with machine learning to identify high-risk patients with cancer and reduce emergency department visits. J Clin Oncol 38, 2020 (suppl 29; abstr 254).

Our summary

Huntsman Cancer Institute, an NCI designated cancer center, has a program called Huntsman at Home (H@H) to support targeted identification of oncology patients for supplemental at-home care. In support of H@H, Flatiron Health partnered with Huntsman to complete a prospective evaluation of Flatiron Signal, an oncology-specific machine learning (ML)-based tool that continuously learns from EMR data to predict patients that were high-risk of an ED visit and provide them with more acute clinical care.

Ultimately, patients identified as "high risk" by the tool had 5.4 times greater odds of a 60 day ED visit than those identified as "low risk,” further opening the door for the use of ML to help enhance clinical services at home.

Interested in a partnership with Flatiron?

Contact our team

arrow

Sales CTA Form

Why this matters

Clinicians in oncology are often challenged to identify when patients with cancer are at high risk for adverse outcomes and would benefit from more intensive clinical care, however it is time consuming and inefficient to manually review the charts of each patient on an ongoing basis in an effort to identify such patients.

By applying a machine learning-based tool, preemptive identification of these patients can help clinicians provide acute, hospital-level care and may lead to improved patient outcomes, and help to avoid costly episodes of care.

Read the research

Related articles