Development and Validation of a Clinical Risk Model for Predicting Malignancy in Patients with Thyroid Nodules

Thyroid cancer is the most common malignancy of the endocrine system. Clinically, it is highly important to identify patients at risk of poor prognosis based on patient characteristics and the features of thyroid nodules. The purpose of the current study was to develop and validate a clinical risk model to predict malignancy in patients with thyroid nodules.  

In the analytical cross-sectional study, the data of 650 patients (mean age: 42.36±13.45 years, female: 86.15%) with thyroid nodules who underwent thyroidectomy were analyzed. The samples were patients referred to the specialized endocrinology clinic between 2014 and 2022. A multivariable model was built using demographic, clinical, and Bethesda System data through logistic regression as a generalized linear model (GLM). Interval validity of the model was checked using bootstrap resampling. The discrimination, calibration, and benefits of the model were evaluated using the area under the curve (AUC), Brier score, and decision curve analysis (DCA), respectively. The diagnostic performance of the GLM was compared with five machine learning (ML) algorithms, including linear discriminant analysis (LDA), random forest, neural network, support vector machine, and k-nearest neighbor. 

Out of 650 operated patients, 43% were benign and 57% malignant. The age, gender, history of thyroid diseases in first-degree relatives, type of thyroid disease, thyroid nodule focality, cervical adenopathy, and Bethesda system were significant features in constructing the prediction model based on GLM. The AUC and Brier score of the model were 0.89 and 0.12, respectively. The DCA also showed that the model performed well in clinical practice. Generally, there was no difference among the six ML algorithms in terms of prognostic performance; however, the prognostic parameters of GLM and LDA algorithms were higher than the others.

Developing and validating ML-based prognostic models using demographic, clinical, and Bethesda data may be useful for the treatment management of patients diagnosed with thyroid nodules.