Advancing Personalized Treatment in Papillary Thyroid Cancer: A Machine-Learning Model for Predicting Occult Lymph Node Metastasis

Highlights

1. RET fusion positivity and BRAF mutation positivity are independent molecular risk factors for occult central lymph node metastasis (OLNM) in clinically lymph node negative (cN0) papillary thyroid carcinoma (PTC).

2. A random forest model achieved an AUC of 0.906 in the training set and 0.733 in the test set for predicting OLNM risk.

3. The model is available as a web calculator for clinical use.

Background

Papillary thyroid carcinoma (PTC) is the most common type of thyroid cancer, with a generally favorable prognosis. However, the presence of lymph node metastasis, particularly occult metastasis not detectable by preoperative imaging, can significantly impact treatment decisions and patient outcomes. Accurate preoperative prediction of OLNM is crucial for optimizing therapeutic strategies, especially in the era of thermal ablation and active surveillance for low-risk PTC.

Study Design

This retrospective study analyzed data from 961 cN0 PTC patients treated between August 2018 and August 2023. The cohort was randomly divided into training and test sets, with a subset of patients having tumors ≤1 cm extracted for internal validation. Eight machine-learning models were developed incorporating clinical, ultrasonographic, and molecular features. Model interpretability was enhanced using Shapley Additive exPlanations (SHAP).

Key Findings

The study identified RET fusion positivity and BRAF mutation positivity as independent molecular risk factors for OLNM in cN0 PTC, alongside six clinical and ultrasonographic variables. The final predictive model incorporated nine predictors. The random forest model demonstrated optimal performance with an AUC of 0.906 in the training set and 0.733 in the test set, along with low Brier scores indicating good calibration. Internal validation in tumors ≤1 cm showed an AUC of 0.719, confirming model robustness. SHAP analysis revealed tumor size, patient age, and clustered punctate echogenic foci as the top predictors of OLNM.

Expert Commentary

This study represents a significant advancement in personalized risk assessment for PTC patients. The identification of RET fusion positivity as an independent risk factor for OLNM is particularly noteworthy, as it may help refine surgical decision-making. While the model shows promise, further prospective validation is needed to confirm its clinical utility across diverse populations.

Conclusion

The developed random forest model provides a clinically useful tool for predicting OLNM risk in cN0 PTC patients by integrating multimodal data. The web-based calculator facilitates practical implementation in clinical practice, potentially guiding more personalized treatment approaches. Future research should focus on prospective validation and exploration of additional molecular markers to further refine predictive accuracy.

Funding and ClinicalTrials.gov

The study did not report specific funding sources or ClinicalTrials.gov registration information.