Can a novel machine learning (ML) model accurately predict the timing of oral squamous cell carcinoma (OSCC) recurrence across four one-year intervals?
BOTTOM LINE
Explore This Issue
July 2023ML methods can interpret complex patterns of patient, clinicopathological, and treatment factors to predict timing of OSCC recurrence.
BACKGROUND: The five-year overall survival rate of oral cavity cancer is 50%. Recurrences are common and carry a significant decrease in survival, and the precise timing of an OSCC recurrence is difficult to predict. ML has shown promising applications in modeling survival and recurrence with greater efficacy.
STUDY DESIGN: Retrospective study.
SETTING: Division of Otolaryngology–Head and Neck Surgery, Department of Surgery, Yale University School of Medicine, New Haven, Conn.
SYNOPSIS: Researchers combined three sets of data inputs with four ML architectures to create 12 models for predicting timing of OSCC recurrence. They retrospectively identified 389 patients with surgically treated OSCC between 2012–2018 from a single health system tumor registry. The primary outcome was the timing of any recurrence (local, regional, or distant), encoded as a five-level gradient across four one-year intervals and one level for no recurrence within four years of surgery. Of the 389 patients studied, 287 had no recurrence and 102 patients had recurrence. Median follow-up time was 25 months for patients with recurrence and 44 months for patients without recurrence. Researchers evaluated their prediction models using mean absolute error (MAE), lower values indicating better prediction of one-year interval recurrence. The best performance was achieved using a 15-variable feature selection data input by two ML models, resulting in an MAE of 0.654 and accuracy of 80.8%. Authors note that the novel ML methodology is a step toward more individualized care for patients.
CITATION: Bourdillon AT, Shah HP, Cohen O, et al. Novel machine learning model to predict interval of oral cancer recurrence for surveillance stratification. Laryngoscope. 2023;133:1652–1659.