Clinical application value of nomogram model based on clinical and ultrasound features in predicting thyroid C-TI-RADS classification optimization

Abstract Backgrounds : A nomogram model based on clinical and ultrasound features was constructed to explore its clinical application value in predicting thyroid C-TI-RADS classification optimization. Methods : Clinical data and ultrasound imaging data of 1,234 patients with thyroid nodules collected from January 2021 to February 2022 of Sichuan Provincial People's Hospital were retrospectively analyzed.All patients underwent preoperative thyroid ultrasound examination and retained standard ultrasound images, evaluated the thyroid nodule C-TI-RADS classification, using the postoperative pathological results as the "gold standard". Independent predictors of C-TI-RADS classification optimization were selected by univariate and multivariate logstic regression analysis, and a nomogram prediction model(*C-TI-RADS) was constructed.The internal validation of the model was performed by Bootstrap resampling. ROC curve was drawn to evaluate the discrimination of the model, and calibration curve and decision curve were drawn to evaluate the consistency and clinical practicability of the prediction model. Results : C-TI-RADS classification, size and number of thyroid nodules, abnormal cervical lymph node ultrasonography, sex and age were independent factors for predicting C-TI-RADS classification optimization (all P < 0.05).The C index of the nomogram prediction model(*C-TI-RADS) constructed based on the above factors was 0.790 (95%CI: 0.765–0.815).Under the optimal cut-off value, the sensitivity was 70.8%, the specificity was 74.4%, and the accuracy was 72.2%.The calibration curve and decision curve showed good consistency and clinical practicability of the model. Conclusions : Nomogram model has good accuracy in the prediction of thyroid C-TI-RADS classification optimization, and can assist ultrasound physician to modify C-TI-RADS classification, which has potential clinical application value.