Machine learning based on SEER database to predict distant metastasis of thyroid cancer

Area of Science:

• Oncology
• Medical Informatics
• Machine Learning

Background:

• Distant metastasis in thyroid cancer is a critical indicator of poor prognosis.
• Early identification of patients with distant metastasis or high risk is essential for timely intervention.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting distant metastasis in thyroid cancer.
• To provide a computational tool to aid clinicians in diagnosis and treatment planning.

Main Methods:

• Utilized demographic and clinicopathological data from the NIH SEER database (2010-2015).
• Employed logistic regression to identify independent risk factors.
• Compared seven machine learning algorithms including Random Forest, Extreme Gradient Boosting, and Support Vector Machines.
• Evaluated model performance using AUC, sensitivity, accuracy, and F1 score.

Main Results:

• Identified age, gender, race, marital status, histological type, capsular invasion, and lymph node metastasis as independent risk factors.
• The Random Forest model achieved the highest performance, with an AUC of 0.960 and F1 score of 0.908 on the test set.
• The model demonstrated high sensitivity (0.929) and accuracy (0.906) in predicting distant metastasis.

Conclusions:

• Machine learning models, particularly Random Forest, show significant potential for the early diagnosis of distant thyroid cancer metastasis.
• These predictive models can assist physicians in making informed decisions for medical interventions.
• The study provides a valuable reference for clinical practice in managing thyroid cancer patients at risk of distant spread.