Development and Validation of a Non-Invasive, Chairside Oral Cavity Cancer Risk Assessment Prototype Using Machine...

Area of Science:

• Oncology
• Medical Informatics
• Machine Learning

Background:

• Oral cavity cancer (OCC) presents significant morbidity and mortality, particularly when diagnosed late.
• Gaps exist in primary care provider (PCP) training for oral cavity examinations.
• Lack of point-of-care (POC) tools hinders identification of high-risk OCC phenotypes, delaying interventions.

Purpose of the Study:

• To develop a prototype OCC risk assessment tool using machine learning (ML).
• To evaluate the performance of ML classifiers for identifying patients at high risk for OCC.
• To explore the integration of this tool into electronic health records (EHRs) as a clinical decision support system.

Main Methods:

• A retrospective dataset from a clinical enterprise data warehouse was utilized.
• Six ML classifiers were compared using a 10-fold cross-validation approach.
• Performance metrics including accuracy, recall, precision, specificity, and AUC were evaluated.

Main Results:

• A voting algorithm achieved 78% accuracy, 64% recall, 88% precision, and 92% specificity.
• The area under the ROC curve was 0.83, and the area under the precision-recall curve was 0.81.
• Multilayer perceptron and AdaBoost classifiers demonstrated performance comparable to the voting algorithm.

Conclusions:

• The developed ML-based OCC risk assessment tool shows promise for clinical application.
• Integration into EHRs as a decision support tool can aid PCPs in identifying and managing at-risk patients.
• This facilitates targeted, personalized interventional strategies for improved OCC outcomes.