Machine learning algorithm predicts skip metastasis in papillary thyroid cancer in retrospective cohort

BackgroundSkip metastasis from papillary thyroid cancer (PTC) is often unpredictable and characterized by lateral lymph node metastasis without central lymph node metastasis. Our objective was to provide a predictive model for skip metastases to cervical lymph nodes based on clinical and demographic data using machine learning.Materials and methodsFrom January 2016 to December 2021, patients who underwent thyroidectomy, central lymph node dissection, and lateral lymph node dissection at the Department of Thyroid Surgery at our hospital, had their clinical and pathological data analyzed retrospectively. Following the identification of five critical characteristics, six machine learning models were developed. Accuracy, sensitivity, specificity, positive predictive value, negative predictive value, F1 score, kappa statistics, and area under the curve were measured in the performance evaluation process, and decision curve analysis was used to determine the clinical advantage. Next, universality was assessed through internal validation. R and Python software was used for all statistical analyses and model construction.ResultsThe incidence of skip lymph node metastases was 13.02% (47/361). Pertinent elements encompassed the number of nodes removed as a result of central lymph node dissection, the existence or non-existence of Hashimoto thyroiditis, the largest tumour size, its bilateral nature, and its multifocal nature. By outperforming alternative models, the random forest model demonstrated excellent performance on the internal validation cohort.ConclusionThis study focused on identifying the risk factors associated with skip metastasis, with the aim of developing an efficient predictive model for this condition using readily available clinical variables. This model can precisely identify skip metastases in PTC using an uncomplicated approach, offering promise for routine clinical use.