Interpretable machine learning distinguishes skip from continuous metastasis in N1b papillary thyroid carcinoma

Area of Science:

• Oncology
• Medical Informatics
• Surgical Pathology

Background:

• Skip metastasis is an underrecognized pattern in papillary thyroid carcinoma (PTC) with lateral lymph node involvement (N1b).
• Accurate identification of skip metastasis is crucial for appropriate patient management.
• Current tools to differentiate skip from continuous metastasis in N1b PTC are limited.

Purpose of the Study:

• To develop and validate prediction models for distinguishing skip from continuous metastasis in N1b PTC.
• To leverage machine learning algorithms for improved diagnostic accuracy.
• To create a clinically deployable tool for individualized risk assessment.

Main Methods:

• Retrospective analysis of 739 N1b PTC patients undergoing bilateral central neck dissection (CND) and lateral neck dissection.
• Training and internal/external validation of ten machine learning algorithms using clinical variables.
• Performance evaluation using discrimination, calibration, clinical utility, and risk reclassification metrics.
• Model interpretability assessed via SHapley Additive exPlanations (SHAP).

Main Results:

• Skip metastasis was present in 14.2% of N1b PTC patients.
• The XGBoost model demonstrated superior and stable discrimination across cohorts.
• The model showed good calibration and favorable clinical utility, outperforming alternative models in risk stratification.
• Key predictors included central lymph node count, metastatic lateral lymph node count, tumor location, age, and size.

Conclusions:

• An interpretable and externally validated machine learning model was developed to differentiate skip from continuous metastasis in N1b PTC.
• This model facilitates individualized risk assessment and may inform tailored surgical strategies.
• Further prospective studies are needed to confirm its impact on clinical outcomes.