Abstract
OBJECTIVE: Parathyroid carcinoma (PC) and atypical parathyroid tumor (APT) are rare malignant parathyroid disorders with varying degrees of recurrence risk. The aim of this study was to determine an effective model for discriminating PC/APT among solitary parathyroid lesions. METHODS: A total of 439 patients with histologically confirmed primary hyperparathyroidism were retrospectively enrolled. The training cohort comprised 207 patients, the validation cohort comprised 52 patients from Hospital I, and the external validation cohort comprised 180 patients from Hospital II. All patients were diagnosed in the parathyroid adenoma (PA) group and the APT/PC group. The clinical and ultrasonic features of the two patient groups were compared. Multivariate logistic regression analysis was conducted to identify independent risk factors for APT/PC. A nomogram was built based on multivariate logistic regression analysis. Model discrimination was assessed using receiver operating characteristic (ROC) curve analysis. The area under the curve (AUC), sensitivity, specificity, and accuracy were reported. Decision and calibration curve analyses were performed to assess the clinical value and calibration of each model, respectively. RESULTS: In the training set, there were 181 cases of PA and 26 cases of APC/PC. Intact parathyroid hormone (iPTH) [odds ratio (OR): 1.019, 95% confidence interval (CI): 1.008-1.032], shape (OR: 16.625, 95% CI: 5.922-51.883), and relation with the thyroid capsule (OR: 3.422, 95% CI: 1.455-9.152) were independent predictive factors associated with the risk of APT/PC. The AUCs for training and internal and external validation were 0.929, 0.962, and 0.965, respectively. The accuracy, sensitivity, and specificity were 86%, 96%, and 85% in the training cohort; 92%, 100%, and 90% in the validation cohort; and 88%, 100%, and 88% in the external validation cohort, respectively. In addition, calibration plots graphically showed good agreement in the presence of the APT/PC group between risk estimation by the nomogram and histopathologic confirmation of surgical specimens. DCA in the current study showed that the nomogram was more effective than all-patient treatment or no treatment over a wide range of threshold probabilities. CONCLUSIONS: Ultrasonic features in combination with iPTH levels may be an applicable model for predicting potentially malignant parathyroid tumors and has a better potential to facilitate preoperative decision-making.