Laryngeal squamous cell carcinoma (LSCC) prognosis and machine learning insights.

Abstract

e18058 Background: LSCC has seen a rise in cases and deaths over the past 30 years. Considering the evolving therapeutic options available in the treatment landscape, we evaluated treatment approaches and developed a machine learning (ML) model for patients with LSCC without distant metastasis. Methods: Data from 2000 to 2020 were obtained from the National Cancer Institute Surveillance, Epidemiology, and End Results database with localized/regional stages only, including the glottis (GC), supraglottic (SuGC), and subglottic (SGC). Patients who were not diagnosed based on histology, previous history of cancer or other concurrent malignancies, or unknown data were excluded. T-tests and chi-square tests were used to compare variables, while the Kaplan-Meier estimator, log-rank tests, and Cox regression analysis identified prognostic factors for overall survival (OS) and cancer-specific survival (CSS). We constructed prognostic models using ML algorithms to predict the 5-year survival. Patient records were randomly divided into training (70 %) and validation (30 %) sets. A validation method incorporating the area under the curve (AUC) of the receiver operating characteristic curve was used to validate the accuracy and reliability of the ML models. Results: A total of 68,282 patients were included (43,434 with GC, 24,010 with SuGC, and 838 with SGC). Of the patients, 80.8% were males and 63.5% were over 60 years old and white (n=56,171). The tumors were >2 cm in size (85.1%). The median patient age was 64 years, and the median tumor size was 3.2 cm. Of these patients, 36.5% underwent surgery with local tumor excision in 23.9 % and total laryngectomy in 5.7% of patients. White race, regional stage, large tumor size, and chemotherapy were poor prognostic factors for GC. Asian ethnicity, white race, and total laryngectomy were good prognostic factors for SuGC, whereas male sex and older age were poor prognostic factors. In terms of SCG "Surgery + adjuvant radiotherapy" was associated with better prognosis while regional and white race were associated with poor prognosis. Performance metrics for all ML algorithms are summarized in the Table. The factors that contributed the most to GC were the surgery type, stage, and age. Stage, surgery type, and race for SuGC and age, stage, and surgery type for SGC. Conclusions: ML models showed promising accuracy in predicting prognosis, highlighting the valuable contributions of factors such as surgery type, stage, age, and race. These findings offer insights into personalized treatment decisions and future research directions. [Table: see text]