Baseline Characteristics
Among 406 patients, LLNM was detected in 128 patients, which accounted for 31.5%. Of note, there were 16 patients (16 of 128) with skip metastases, meaning LLNM without central cervical lymph node metastasis (CLNM) (Table S2). Baseline information and DECT images characteristics of primary foci according to LLNM status were summarized in Table 1. The median age was 48 years (IQR 36 - 57 years, range 22 - 77 years). The majority of patients were female (322 patients, 79.3%; 50 years, IQR 39 - 58 years), and 20.7% (84 patients) were male (41 years, IQR 32 - 55 years). 149 primary foci (36.7%) were with cystic degeneration, 85 primary foci (21.0%) were with calcification, and 104 primary foci (25.6%) were with ETE. The above parameters were statistical significance for differentiation between patients with LLNM and without (P < .05). Check the specific information about other parameters in Table 1.
Result of consistency analysis
The intraclass correlation coefficient (ICC) calculated for the agreement of features extracted by two radiologists ranged from 0.913 to 0.974, reflecting good agreement (P .000). The imaging characteristics of DECT were basically consistent between the two radiologists. The inter-observer and intra-observer consistency analysis for all the parameters was greater than 0.8, which showed good consistency (Table S3, Figure S2).
Comparison of DECT imaging parameters and thyroid functional indicators between patients with and without ipsi-LLNM
Quantitative parameters of patients with and without LLNM were listed in Table 2. Tg, Anti-Tg, Anti-TPO, volume, IC in the arterial phase, and IC in the venous phase were higher in those with LLNM than those without (P < .0001) (Table 2, Figure 2).
Univariate and multivariate logistic regression analysis of risk factors for ipsi-LLNM in patients with PTC
Univariable logistic regression analysis showed that Tg, Anti-Tg, volume, cystic degeneration, calcification, ETE, IC in the arterial phase, and IC in the venous phase were risk factors for predicting the presence of LLNM (P range, .000 - .006). Further multivariable logistic regression analysis showed that among these parameters, Tg (OR, 2.668; 95% CI: 1.590, 4.475; P .000), Anti-Tg (OR, 2.001; 95% CI: 1.202, 3.333; P .008), ETE (OR, 6.335; 95% CI: 3.768, 10.651; P .000), IC in arterial phase (OR, 3.691; 95% CI: 2.170, 6.278; P .000) and IC in venous phase (OR, 2.122; 95% CI: 1.271, 3.541; P .004) were the independent predictors for LLNM. Sex, age, Anti-TPO, HT, nodular goiter, volume, cystic degeneration, and calcification were not related to LLNM in patients with PTC (P > .05) (Table 3).
The cut-off value of each parameter for ipsi-LLNM in patients with PTC
The AUC, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) for differentiating ipsi-LLNM for each parameter were listed in Table 4. ROC curve analysis determined that the optimal cut-off points for Tg, Anti-Tg, IC in arterial phase and IC in venous phase in predicting ipsi-LLNM were 100.01 ng/ml (AUC 0.856, 95%CI 0.818-0.889), 89.43 IU/ml (AUC 0.766, 95%CI 0.721-0.806), 3.4 mg/ml (AUC 0.846, 95%CI 0.807-0.879) and 3.1 mg/ml (AUC 0.777, 95%CI 0.733-0.816), respectively. The specific information of other parameters was listed in Table 4, Figure 3 and S3. There were two examples of predicting ipsi-LLNM, which might help illustrate the predictive value of these independent risk factors (Figure 4 and 5).