Preoperative risk score for malignancy in Bethesda III thyroid nodules: a multicentre retrospective study

Bethesda III thyroid nodules represent a heterogeneous cytological category with an estimated malignancy risk of 13-30%, creating uncertainty in clinical decision-making. This study aimed to characterise surgically resected Bethesda III thyroid nodules in a multicentre cohort and to develop a simple preoperative risk score for malignancy prediction using routinely available variables.

We performed a retrospective, observational, multicentre study across four tertiary hospitals in the Region of Murcia, Spain. Adults with Bethesda III thyroid nodules who underwent thyroidectomy between 2020 and 2025 were included. Clinical, biochemical, ultrasonographic, cytological, and histopathological variables were analysed using univariate and multivariable logistic regression. A simplified integer-based risk score was derived from the independent predictors identified in the final model.

Eighty-five nodules were included, with a malignancy rate of 23.5%. Three preoperative variables remained independently associated with malignancy: moderate-to-high ultrasound risk, nuclear atypia on fine-needle aspiration, and lymphocytic thyroiditis on fine-needle aspiration. These predictors were incorporated into a practical score ranging from 0 to 15 points. The model showed acceptable apparent discrimination and calibration, although internal bootstrap validation indicated that cautious interpretation is required because of the limited sample size. A cut-off of 10 points or higher provided the best overall diagnostic performance, and the predicted probability of malignancy ranged from 2.4% to 74.8% across the score spectrum.

In this multicentre cohort of surgically resected Bethesda III thyroid nodules, a simple three-item exploratory score showed promising performance for preoperative malignancy risk stratification. This tool may help support clinical decision-making using variables available in routine practice, although external validation in larger independent cohorts is required before routine clinical implementation.