Machine learning for predicting overall survival in early-stage supraglottic cancer: a SEER-based population study

Supraglottic squamous cell carcinoma (SGSCC) represents the second most prevalent form of laryngeal cancer and carries a poor prognosis.

This study aimed to combine clinicopathological and treatment-related factors as integrative inputs to build a machine learning (ML) model to estimate the overall survival (OS) of patients with early-stage SGSCC. Furthermore, we explored the complementary prognostic potential of these input parameters.

A total of 1171 patients with SGSCC were extracted from Surveillance, Epidemiology, and End Results (SEER) public data. We used feature importance analysis to examine the integrative inputs that are associated with OS.

The ML model showed a weighted accuracy of 72.3% in predicting OS. The aggregate feature importance showed that age at diagnosis, marital status, number of malignancies, regional lymph nodes, and radiotherapy are the five most important features for enhancing OS among these patients. We found that as age increases, the chance of OS decreases. Being married, the absence of other primary indicators, surgical treatment, and radiotherapy were associated with improved OS.

Combining clinicopathological and treatment-related factors seems to predict accurately OS in patients with early-stage SGSCC. External independent geographic validation is warranted to evaluate model generalizability.