Abstract
PURPOSE: This study aimed to analyze the ultrasonographic characteristics of differentiated thyroid cancer (DTC) with multigenic co-mutations and to establish a predictive model using contrast-enhanced ultrasonography (CEUS). METHODS: This retrospective study included consecutive patients with pathologically confirmed DTC who underwent preoperative CEUS and next-generation sequencing at the authors' institution between September 2021 and December 2023. Clinical and CEUS features were compared between patients with and without multigenic co-mutations. Bayesian logistic regression (non-informative normal priors) was applied for predictor selection and model development, with Markov-chain Monte Carlo (MCMC) convergence checks and posterior predictive validation. Internal validation was performed using bootstrap resampling (n=1,000 iterations) to evaluate model stability. RESULTS: A total of 116 patients (mean age, 39.84±11.02 years; 33 men) were included, of whom 12 had multigenic co-mutations and 104 did not. Patients with multigenic co-mutations demonstrated a higher incidence of aggressive histological subtypes (25.0% vs. 1.9%, P=0.008) and lymph node metastasis (83.3% vs. 51.9%, P=0.038). Tumor size, enhancement homogeneity, and contrast agent arrival time were identified as significant predictors, with robust posterior distributions (all inclusion probabilities >0.9) and satisfactory MCMC convergence (potential scale reduction factor <1.01). The model achieved an area under the curve (AUC) of 0.873, with posterior predictive checks confirming favorable predicted-observed agreement (coverage ≥0.85). Internal validation with 1,000 bootstrap replicates yielded a consistent AUC of 0.880 (95% confidence interval, 0.745 to 0.978). CONCLUSION: The CEUS-based predictive model demonstrated strong discrimination for detecting multigenic co-mutations in differentiated thyroid cancer; however, external validation is required to confirm its clinical applicability.