Performance of Machine Learning Models for Prognosis Prediction in Oral Cavity Squamous Cell Carcinoma: A Systematic
Sammy Y Gao1, Jonathan M Hughes1, Shaun A Nguyen1
1Department of Otolaryngology-Head and Neck Surgery, Medical University of South Carolina, Charleston, SC 29425, USA.
Cancers
|July 28, 2026
Summary
Machine learning (ML) models show promise for predicting outcomes in oral cavity squamous cell carcinoma (OCSCC). However, inconsistent methodology and limited external validation hinder widespread clinical use of these prognostic tools.
Area of Science:
- Oncology
- Medical Informatics
- Biostatistics
Background:
- Machine learning (ML) models are increasingly used for prognostic prediction in oral cavity squamous cell carcinoma (OCSCC).
- The performance and methodological quality of these ML models are variably reported.
- A systematic review is needed to evaluate contemporary ML-based prognostic models for OCSCC, focusing on independently validated studies.
Purpose of the Study:
- To systematically review and evaluate ML-based prognostic models for clinically relevant outcomes in OCSCC.
- To assess the performance and methodological quality of these models, with an emphasis on independent validation.
- To identify limitations and areas for improvement in ML model development and application for OCSCC.
Main Methods:
- A systematic review was conducted following PRISMA guidelines, searching major databases (PubMed, Scopus, Cochrane Library, CINAHL) up to December 1, 2025.
- Studies evaluating ML or artificial intelligence prognostic models in adult OCSCC patients were included.
- Data extraction and risk-of-bias assessment using PROBAST + AI were performed independently by reviewers.
Main Results:
- Forty studies involving 105,619 patients were included, utilizing various ML architectures (random forests, SVMs, neural networks, deep learning).
- Independently validated models for overall survival generally showed AUCs between 0.80-0.90; recurrence prediction models also demonstrated favorable discrimination.
- Methodological limitations included retrospective designs, small sample sizes, inadequate external validation, and risk of overfitting.
Conclusions:
- ML-based prognostic models for OCSCC exhibit generally favorable predictive performance for survival and recurrence.
- Heterogeneity in methodology and limited external validation impede clinical implementation.
- Future research should focus on prospective multicenter validation, standardized reporting, and reproducible modeling frameworks.
