A Novel Natural Language Processing Model for Triaging Head and Neck Patient Appointments

Area of Science:

• Medical Informatics
• Oncology
• Otolaryngology

Background:

• Inaccurate patient triage leads to delayed care and increased morbidity/mortality, especially in cancer patients.
• Effective management of clinical capacity is crucial for optimal patient outcomes.
• Head and neck (H&N) cancer patient triage requires precise and timely assessment.

Purpose of the Study:

• To develop and assess a natural language processing (NLP) model for H&N patient triage.
• To evaluate the NLP model's accuracy in categorizing pathology and predicting appointment urgency.
• To determine the model's potential as an adjunctive tool in H&N patient workflows.

Main Methods:

• A retrospective cohort study was conducted at an academic institution.
• An NLP model was developed and applied to referral documents (clinic notes, imaging, pathology reports) of 83 new H&N patients.
• The model predicted pathology type, malignancy risk, and appointment urgency, with final diagnoses serving as the gold standard.

Main Results:

• The NLP model achieved 81.9% accuracy for pathology type and 86.8% for urgency level.
• High sensitivity was observed for various H&N pathologies, including non-endocrine neoplasms (88.9%) and thyroid (88.9%) and parathyroid (100%) pathologies.
• The model demonstrated strong predictive performance for appointment urgency, with a Matthews correlation coefficient of 0.698.

Conclusions:

• The NLP model shows robust performance in predicting H&N diagnoses and urgency from referral documents.
• This tool can assist H&N practice coordinators in screening referrals, potentially optimizing patient care pathways.
• The model's ability to identify urgent cases may significantly improve management of H&N cancer patients.