Prevalence and predictors of malignancy in Bethesda III and IV thyroid nodules undergoing surgery

Indeterminate thyroid cytology remains a major diagnostic challenge. Thyroid nodules classified as Bethesda category (BC) III and IV carry an intermediate risk of malignancy, and their management often involves a choice between active surveillance and diagnostic surgery. This study aimed to assess the prevalence of benign, low-risk, and malignant lesions in surgically treated patients with these nodules, and to identify factors associated with malignancy.

We retrospectively analyzed 618 patients with BC III or IV thyroid nodules, selected from 2,239 patients who underwent thyroid surgery between October 2022 and December 2025 at a Tertiary Referral Endocrine Surgery Centre in Krakow. Histopathology followed the 2022 WHO Classification of Thyroid Tumors.

The distribution of tumor types differed significantly in both BCs III (n=365) and IV (n=253) nodules (both p<0.001), with benign lesions predominating (62.5% in BC III and 55.7% in BC IV), followed by malignant (30.4% and 32.8%) and low-risk lesions (7.1% and 11.5%), respectively. Younger age was associated with a higher risk of malignancy in both BC III (p=0.03) and BC IV (p=0.02). Sex was associated with lesion distribution in BC III (p = 0.016) but not BC IV (p = 0.06), and was not an independent predictor of malignancy. Tumor size ≥1 cm was associated with malignancy in BC IV (p=0.005), but not in BC III (p=0.35). Although age alone showed limited discriminative ability, multivariable analysis identified age <50 years as the only independent predictor of malignancy (adjusted OR 1.74; 95% CI 1.23-2.47; p=0.002) and lymph node metastasis (adjusted OR 4.56; 95% CI 1.19-17.53; p=0.03), while sex, chronic lymphocytic thyroiditis (CLT), and BC were not significant.

In this surgical cohort of BC III and IV thyroid nodules, approximately one-third of lesions were malignant. Younger age was independently associated with malignancy, but its value as an isolated predictor was limited, and the overall discriminative performance of the multivariable model was modest. These findings may support individualized preoperative risk stratification; however, given the retrospective design and the selection of patients for surgery, they do not justify firm conclusions regarding the optimal extent of surgery.