Machine learning in personalized laryngeal cancer management: insights into clinical characteristics, therapeutic options, and survival predictions.

Purpose: Over the last 40 years, there has been an unusual trend where, even though there are more varied treatments, survival rates have not improved much. Our study used survival analysis and machine learning (ML) to investigate this odd situation and to improve prediction methods for treating non-metastatic LSCC.

Methods: The surveillance, epidemiology and end results (SEER) database provided the data used for this study's analysis. To identify the prognostic variables for patients with non-metastatic LSCC, we conducted Cox regression analysis and constructed prognostic models using five ML algorithms to predict 5-year survival. A method of validation that incorporated the area under the curve (AUC) of the receiver operating characteristic (ROC) curve was employed to validate the accuracy and reliability of the ML models. We also investigated the role of multiple therapeutic options using Kaplan Meier (K-M) survival analysis.

Results: The study included 63,324 patients, of whom 40,824 were diagnosed with glottic cancer (GC), 21,774 with supraglottic (SuGC) and 726 with subglottic (SC). ML models identified age, stage, and tumor size as the most important factors that affect survival. For SuGC, age, stage, and sex and stage and race for SC. In terms of treatment, best survival therapeutic options for GC and SC were surgery and radiotherapy (RT), whereas SuGC surgery only.

Conclusion: This study underscores the critical role of individualized factors in non-metastatic LSCC management, with surgery often combined with radiotherapy as the optimal treatment for early stage tumors. Despite advancements, stable prognosis highlights the need for continuous refinement of therapeutic strategies to balance tumor control and quality of life.