Machine learning algorithms for identifying contralateral central lymph node metastasis in unilateral cN0 papillary...

Area of Science:

• Oncology
• Medical Informatics

Background:

• Thyroid cancer incidence is rising globally.
• Surgery is the primary treatment for thyroid cancer.
• Contralateral central lymph node dissection in unilateral cN0 papillary thyroid cancer remains debated.

Purpose of the Study:

• To develop a machine learning model for predicting contralateral central lymph node metastasis.
• Utilize demographic and clinical data for prediction in papillary thyroid cancer patients.

Main Methods:

• Retrospective study of 2225 patients with unilateral cN0 papillary thyroid cancer.
• Comparison of clinical and pathological features between metastatic and non-metastatic groups.
• Construction and comparison of six machine learning models using R software, validated externally.

Main Results:

• Identified independent risk factors for metastasis: male gender, tumor diameter >1cm, multifocality, ipsilateral metastasis, and age <50 years.
• Random forest model demonstrated superior performance.
• External validation confirmed model efficacy; a web calculator was developed.

Conclusions:

• Gender, tumor size, multifocality, ipsilateral metastasis, and age are critical factors for considering contralateral lymph node dissection.
• The random forest-based web calculator can aid clinical decision-making for papillary thyroid cancer management.