Evaluation of cervical lymph node metastasis after total thyroidectomy for differentiated thyroid carcinoma: construction of a multimodal non-invasive predictive model based on serum thyroglobulin and ultrasonic features.

Background: Accurate assessment of cervical lymph node metastasis (LNM) burden is crucial for formulating individualized secondary surgical strategies in patients with differentiated thyroid carcinoma (DTC) after total thyroidectomy. High-volume metastasis (defined as LNM ≥5) is significantly associated with poor patient prognosis. Preoperative identification of high-volume metastasis prior to secondary surgery is essential for determining the extent of individualized secondary lymph node dissection (LND). However, there is currently a lack of non-invasive predictive tools integrating serum biomarkers and imaging features to quantify this risk. Therefore, this study aimed to develop and validate a multimodal predictive model for preoperative identification of patients with high-volume LNM. Methods: This retrospective study included 97 patients with pathologically confirmed recurrent DTC who underwent secondary LND between May 2020 and December 2024. We analyzed preoperative serum thyroglobulin (Tg) under thyroid stimulating hormone (TSH) suppression (0.1-0.5 mU/L) and ultrasound (US) features, including quantitative metrics (total maximum diameter, total solid component long/short diameter, number of involved neck levels) and qualitative features (microcalcifications, cystic change, hyperechogenicity, abnormal blood flow, aspect ratio ≤2). A data-driven ultrasound score (USscore) was constructed from the qualitative features. Predictors were screened via logistic regression, with internal validation using bootstrap resampling (2,000 repetitions) and clinical utility assessed by decision curve analysis (DCA). Results: Multivariate analysis identified serum Tg ≥2.57 ng/mL [odds ratio (OR) =5.60, 95% confidence interval (CI): 1.31-27.70, P=0.03] and number of involved neck levels (OR =3.38, 95% CI: 1.54-8.49, P=0.003) as independent predictors of LNM ≥5. The multimodal model integrating the USscore (Model 2) achieved an optimism-corrected area under the curve (AUC) of 0.865 (95% CI: 0.783-0.934), with a mean absolute error (MAE) of 0.076, comparable to the baseline model (AUC =0.84, MAE =0.112). DCA showed marginally higher net benefit for Model 2. The USscore demonstrated a dose-dependent association with high-volume metastasis risk (incidence: 68.4% for score ≥5 Conclusions: This study developed and validated a multimodal model integrating serum Tg and sonographic features that effectively predicts the risk of high-volume cervical LNM in postoperative recurrent DTC, providing an objective tool for preoperative assessment and surgical planning.