Development and validation of clinical predictive models for differentiated thyroid carcinoma and cervical lymph node metastasis: A retrospective cohort study

Abstract

Background: For differentiated thyroid carcinoma (DTC), the main negative prognostic factor is cervical lymph node metastasis (CLNM). Two clinical prediction models were established based on the preoperative characteristics of thyroid nodules to provide a more individualized guidance for predicting DTC and CLNM. Methods: Altogether, 679 hospitalized patients with DTC and 294 patients with benign thyroid nodules were retrospectively included, and patients with DTC were further distinguished as having CLNM. Two multiple logistic regression models were established in this retrospective cohort. The considered predictors mainly included ultrasound features of thyroid nodules and lymph nodes, as well as clinical characteristics, including body mass index, age, sex, and thyroid-related laboratory parameters. All cases were used as a training set and internally validated using bootstrap resampling. Results: The receiver operating curve (ROC) of the model for diagnosing benign and malignant thyroid nodules corresponded to an area under the curve (AUC) of 0.873 in the training set and a Brier score of 0.116 for calibration. After bootstrap internal validation, the adjusted AUC was 0.858, and the Brier score was 0.122. The model had better performance than the only ultrasound grading for diagnosing DTC. The ROC of the model for predicting CLNM had an AUC value of 0.745 in the training set, and the Bbrier score was 0.203. After bootstrap internal validation, the adjusted AUC value was 0.720, and the Brier score was 0.214. Conclusion: The two clinical prediction models have good predictive efficacy for diagnosing DTC and assessing its metastasis, providing new diagnostic tools.