(A509) Application of a Machine Learning Algorithm for Managing Real-world Head and Neck Surgical Oncology Referrals

Introduction: The patient referral process often requires input from the medical team, leading to bottlenecks and delays in care. Machine learning is an emerging technology in clinical practice that may improve the efficiency of patient referral management. Our aim was to evaluate the efficacy with which a machine learning model evaluated and managed real patient data in referrals to head and neck surgery.

Methods: Between 12/10/24 - 1/7/25, 77 patients were referred to the UCSF Otolaryngology-Head and Neck Surgical Oncology Center. Clinical data were collected, including the initial referral date, clinical imaging, pathology results, and clinical notes attached to the referral.

The machine learning model utilized natural language processing to identify keywords in a patient’s referral documents. These keywords are used to generate insights including the most likely pathology, level of suspicion, and next steps in patient management. All de-identified patient referrals in this study were inputted into the machine learning model. The machine learning model outputs were compared to what providers recommended in terms of the appropriateness and urgency of the referrals.

Results: The algorithm achieved an 89% accuracy in predicting pathology and urgency. The average time from the referral date to create the first scheduling attempt of a referral, which includes the time to determine the urgency and appropriateness of a referral, was 5.8 days. In contrast, the algorithm is capable of triaging a new referral within 30-90 seconds. Of these referrals, 94.8% (73/77) included a relevant clinical note, 75.3% (58/77) included clinical imaging, and 40.3% (31/77) included pathology results. Further, 41.6% (32/77) of the patient referrals were internal and 58.4% (45/77) of the patient referrals were external.

Conclusions: The high accuracy of the algorithm in assessing the urgency and appropriateness of patient referrals highlights the potential of deep learning to revolutionize the referral process and improve access to care. Future research should explore physician perspectives and evaluate integration into clinical workflows.