Development and Internal Validation of a Predictive Model for Individual Cancer Risk Assessment for Thyroid Nodules

The objective of this study was to develop and validate a predictive model for the assessment of the individual risk of malignancy of thyroid nodules based on clinical, ultrasound, and analytic variables. A retrospective case-control study was carried out with 542 patients whose thyroid nodules were analyzed at our endocrinology department between 2013 and 2018 while undergoing treatment for thyroidectomy. Starting with a multivariate logistic regression analysis, which included clinical, analytic, and ultrasound variables, a predictive model for thyroid cancer (TC) risk was devised. This was then subjected to a cross-validation process, using resampling techniques. In the final model, the independent predictors of the risk of malignancy were: being male, age of the extremes, family history of TC, thyroid-stimulating hormone level >4.7 μU/L, presence of autoimmune thyroiditis, solid consistency, hypoechogenicity, irregular or microlobed borders, nodules that are taller than they are wide, microcalcifications, and suspicious adenopathy. With a cut-off point of 50% probability of thyroid cancer, the predictive model had an area under the receiver operating characteristic curve of 0.925 (95% confidence interval 0.898 to 0.952). Finally, using the 10-fold cross-validation method, the accuracy of the model was found to be 88.46%, with a kappa correlation coefficient of 0.62. A predictive model for the individual risk of malignancy of thyroid nodules was developed and validated using clinical, analytic, and ultrasound variables. An online calculator was developed from this model to be used by clinicians to improve decision-making in patients with thyroid nodules.