Prognostic Models Using Machine Learning Algorithms and Treatment Outcomes of Papillary Thyroid Carcinoma Variants.

BACKGROUND

Hürthle cell (HCC) and columnar cell variants (CCV) are rare subtypes of thyroid cancer.

AIMS: This study used machine learning (ML) to evaluate treatment effectiveness and develop prognostic models.

METHODS

Chi-square tests, Kaplan-Meier curves, log-rank tests, and Cox regression were used. Five ML algorithms constructed prognostic models predicting 5-year survival, validated using the AUC of the ROC curve.

RESULTS

Among 3690 patients, 3180 had CCV and 510 had HCC. ML models showed metastasis, surgery + RT, and age were significant factors for HCC, while the N component of TNM, metastasis, and tumor size were significant for CCV.

CONCLUSION

This study offers a comprehensive approach for treating and assessing prognosis in PTC variants. The ML models developed offer practical tools for personalized clinical decision-making.