Prediction model construction and validation for cervical lymph node metastasis in papillary thyroid microcarci-noma based on shear wave elastography combined with serological markers

CHEN Hua-bin; HUANG Xiao-wen; ZENG Fu-qiang; ZOU Bin; LIN Yong-ping; ZHANG Lu-sheng; XU Xiao-hong

doi:10.13820/j.cnki.gdyx.20250894

Abstract：Objective To establish and validate a predictive model for cervical lymph node metastasis in papillary thyroid microcarcinoma(PTMC)using shear wave elastography(SWE)combined with serological markers.Methods A total of 166 PTMC patients were initially enrolled;6 withdrew from the study due to treatment-related logistical issues.Ultimately,clinical data from 160 patients treated between January 2019 and December 2022 were analyzed.Patients were divided into non-metastasis(n=104)and metastasis(n=56)groups based on pathological results of cervical lymph nodes.Univariate analysis was performed on general information,ultrasound features,and laboratory results.Independent predictors were identified through multivariate binary logistic regression to construct a predictive model.Model calibration and discrimination were evaluated using the Hosmer-Lemeshow test and receiver operating characteristic(ROC)curve.An external validation cohort of 60 patients treated from January 2023 to December 2024 was also analyzed.Results Uni-variate analysis showed that tumor diameter,number of lesions,capsular invasion,maximum elastic modulus(Emax),thyroglobulin(Tg),and thyroid-stimulating hormone(TSH)were associated with cervical lymph node metastasis in PT-MC(all P＜0.05).Multivariate analysis revealed that number of lesions,capsular invasion,Emax,Tg,and TSH were independent predictors(all P＜0.05).The logistic regression model was constructed as:Logit(P)=-14.300+1.370 × number of lesions+1.470 × capsular invasion+0.138 × Emax+0.595 × Tg+6.862 × TSH.The model achieved an AUC of 0.942(95％CI:0.906-0.979),with a maximum Youden index of 0.797,sensitivity of 0.893,and specificity of 0.904.The average absolute error in the calibration curve was 0.023.Decision curve analysis(DCA)indicated good clinical utility.In the validation cohort,the AUC was 0.955(95％CI:0.901-1.000),with sensitivity and specificity both at 0.900,and average absolute calibration error of 0.082.DCA confirmed the model's strong clinical performance.Conclusion The predictive model based on SWE combined with serological markers exhibits excellent performance in predicting cervical lymph node metastasis in PTMC,providing theoretical support for preoperative clinical risk assessment.