Advanced-stage tongue squamous cell carcinoma: a machine learning model for risk stratification and treatment planning

Abstract

Background

A significant number of tongue squamous cell carcinoma (TSCC) patients are diagnosed at late stage.

Objectives

We primarily aimed to develop a machine learning (ML) model based on ensemble ML paradigm to stratify advanced-stage TSCC patients into the likelihood of overall survival (OS) for evidence-based treatment. We compared the survival outcome of patients who received either surgical treatment only (Sx) or surgery combined with postoperative radiotherapy (Sx + RT) or postoperative chemoradiotherapy (Sx + CRT).

Material and Methods

A total of 428 patients from Surveillance, Epidemiology, and End Results (SEER) database were reviewed. Kaplan-Meier and Cox proportional hazards models examine OS. In addition, a ML model was developed for OS likelihood stratification.

Results

Age, marital status, N stage, Sx, and Sx + CRT were considered significant. Patients with Sx + RT showed better OS than Sx + CRT or Sx alone. A similar result was obtained for T3N0 subgroup. For T3N1 subgroup, Sx + CRT appeared more favorable for 5-year OS. In T3N2 and T3N3 subgroups, the numbers of patients were small to make insightful conclusions. The OS predictive ML model showed an accuracy of 86.3% for OS likelihood prediction.

Conclusions and Significance

Patients stratified as having high likelihood of OS may be managed with Sx + RT. Further external validation studies are needed to confirm these results.

Chinese Abstract

背景：大量舌鳞状细胞癌 (TSCC) 患者到了后期才被诊断出。

目的：我们的主要目的是开发基于集成 ML范例 的机器学习 (ML) 模型, 将晚期 TSCC 患者根据整体生存 (OS) 可能性进行分级, 以便循证治疗。 我们比较了只接受手术(Sx)的患者或接受手术联合术后放疗 (Sx+ RT) 或术后化放疗 (Sx+ CRT)的患者的生存结果。

材料和方法：对来自监测、流行病学和最终结果(SEER) 数据库的共 428 名患者进行了审查。 用Kaplan-Meier 和 Cox 比例风险模型检查 OS。 另外, 还开发了一个 ML 模型用于 OS 可能性分级。

结果：年龄、婚姻状况、N 分期、Sx 和 Sx+ CRT 被认为是重要的。Sx + RT 的患者显示比 Sx + CRT 或只接受手术有更好的整体生存。 T3N0 亚组获得了类似的结果。对于 T3N1 亚组, Sx+ CRT 似乎更有利于 5 年的整体生存。 在 T3N2 和 T3N3 亚组中, 患者人数很少, 无法得出有意义的结论。 OS预测 ML 模型显示 OS 可能性预测的准确率为 86.3%。

结论和意义：对分级为具有高 OS 可能性的患者可以用Sx + RT进行管理。 需要进一步的外部验证研究来确认这些结果。

Keywords:

• Tongue cancer
• overall survival
• machine learning
• chemoradiotherapy
• radiotherapy
• radiation
• surgery
• SEER

Acknowledgment

Minerva Foundation: Selma and Maja-Lisa Selander’s Fund for Research.

Finska Läkaresällskapet, The Sigrid Jusélius Foundation. The Helsinki University Hospital Research Fund.

Disclosure statement

No potential conflict of interest was reported by the author(s).