Predictive models for contralateral occult thyroid carcinoma and central lymph node metastasis in unilateral multifocal papillary thyroid carcinoma based on ultrasound

Introduction

Thyroid carcinoma (TC) arises from thyroid follicular epithelial or parafollicular cells (1,2). In 2022, there were more than 821,000 cases of TC worldwide, and its overall incidence ranked 7th among malignant tumors. The incidence rate was highest in East Asia and North America, with the 466,100 new cases reported in China accounting for over half of the global disease burden. Data from the National Cancer Registry Center indicate that the incidence of TC has been increasing at an annual rate of 20% (3,4), representing the largest increase among malignant tumors. Contemporary evidence identifies key drivers of TC trends: diagnostic advancements [e.g., high-resolution ultrasonography and fine-needle aspiration biopsy (FNAB)] enable the detection of sub-centimeter micro-nodules, whereas expanded screening programs and heightened public awareness increase early-stage case identification. Environmental factors remain critical, with ionizing radiation exposure, especially in inadequately shielded pediatric imaging, posing established risks. Chronic exposure to endocrine-disrupting compounds (EDCs) from plastics/industrial sources via the food chain disrupts thyroid function and elevates carcinogenic potential, compounded by psychological stress and detrimental lifestyles (5). Gender-specific vulnerabilities further amplify female risk, including miscarriage history, menstrual irregularities, oral contraceptive use, and sex hormone fluctuations, exacerbated by heightened stress susceptibility (6). However, its mortality rate is significantly lower than its incidence rate, positioning it 24^th in terms of lethality. There were an estimated 44,000 deaths from TC in 2022 (3). The overdiagnosis of thyroid cancer is a recognized phenomenon, leading to unnecessary interventions and inefficient resource allocation. The primary treatment modality for TC is surgical intervention. Among the various types of TC, papillary TC (PTC) is the most prevalent (7), accounting for approximately 85–90% of all TC cases. Consequently, PTC constitutes the majority of patients undergoing surgical treatment. Thus, personalized surgical strategy selection is paramount.

For PTC with a primary tumor diameter <1 cm and an absence of high-risk features, lobectomy plus isthmusectomy is recommended. Total thyroidectomy (TT) or near-TT is recommended if any of the following criteria are met: primary tumor diameter >4 cm; tumor diameter >1 cm located in the isthmus; positive resection margins; extrathyroidal extension; vascular invasion; bilateral multifocal disease; clinically apparent lymph node metastasis (≥5 involved nodes or metastatic node diameter ≥3 cm); distant metastasis, and some other factors (8,9). Despite the relative comprehensiveness of the guidelines, challenges persist in certain clinical scenarios. For example, during actual clinical decision-making, pathological indicators, such as the subtypes of PTC or the number of clinical lymph node metastases, are not always accurately obtainable prior to or during surgery. Meanwhile, unilateral multifocal PTC (MPTC) represents a critical clinical decision-making gap in contemporary thyroid surgery. Multifocality is a highly prevalent characteristic in PTC. Most researchers recognize distinct clinicopathological differences between multifocal and unifocal PTC. Relevant meta-analyses indicate that patients with MPTC experience reduced survival rates (10) and poorer prognoses (11). Studies report MPTC incidence rates ranging widely from 18% to 87% (12-14), underscoring the significant clinical value in optimizing treatment strategies for this large subgroup. The selection of the surgical strategy is also influenced by tumor metastasis. Cervical lymph nodes, particularly central lymph nodes, constitute the most common metastatic area of PTC, with an incidence between 30% and 70% (15,16). Lymph node metastasis is more prevalent in MPTC compared to single-focal PTC (17-19). The American Thyroid Association (ATA) guidelines indicate that central lymph node metastasis (CLNM) is as an independent risk factor for recurrence and decreased survival in patients with PTC (9), which holds positive guiding significance for the determination of the surgical extent, recurrence risk assessment, and subsequent treatment. Guidelines for the diagnosis and treatment of thyroid nodules and differentiated TC stipulate that central lymph node dissection (CLND) should be conducted at least on the same side of the lesion under the condition of effective protection of the parathyroid gland and recurrent laryngeal nerve. Simultaneously, the guidelines emphasize that high-resolution ultrasound examination (US) is the preferred imaging modality for the assessment of thyroid nodules. However, the outcomes of US are influenced by the operator’s meticulousness and subjective factors, making it challenging to identify lesions that are deeper or obscured by surrounding structures. Hence, nodules with a diameter of 2 mm or less are at a significant risk of remaining undetected (20). Furthermore, the deep anatomical positioning and intricate internal structure of the central group lymph nodes contribute to chaotic echogenicity in this region. Additionally, the shielding effect by the thyroid gland and the presence of undetectable micrometastases render the sensitivity of US in the evaluation of CLNM insufficient (21,22). At present, the preoperative diagnostic approaches of lymph node metastasis encompass US (23), computed tomography (CT) (24), magnetic resonance imaging (MRI), and lymph node biopsy. Nevertheless, the diagnostic accuracy of CT and US for CLNM is limited (25), whereas MRI and needle biopsy are not commonly utilized. Therefore, it is essential to construct a clinical prediction model for the number of CLNM to assist in surgical decision-making. Currently, owing to the risk of contralateral occult TC (OTC) and clinical lymph node metastasis, a considerable proportion of patients with unilateral multifocal carcinoma still undergo TT, which may result in overtreatment. For non-high-risk patients without contralateral OTC or a large number of CLNM detected through postoperative pathology, thyroid lobectomy (TL) can be performed in accordance with the patient’s consent during the first operation, so as to preserve or even maintain normal thyroid function. Therefore, it is critically important to evaluate the risk of contralateral OTC and CLNM (number ≥5) in patients with MPTC by utilizing the clinicopathological data obtainable before surgery, thereby selecting the most beneficial surgical plan for the patients. When the risk of metastasis is classified as low or moderate, performing TL may be a more prudent option to mitigate the likelihood of overtreatment.

This study aimed to establish clinical models utilizing preoperative clinical data through machine learning (ML) techniques to accurately predict the risks of contralateral OTC and large number of CLNM in patients with unilateral MPTC, thereby providing more explicit guidance for individualized and precise decision-making regarding surgical procedures. We present this article in accordance with the TRIPOD+AI reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-2024-2924/rc).

Methods

Patient selection process

This study collected the clinical and pathological data of PTC patients who underwent initial surgery in Union Hospital, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, China, from May 2020 to May 2024. Figure 1A delineates the surgical decision-making protocol implemented at our institution, which aligns with the Chinese guidelines for TC management.

Figure 1 Study design and patient enrollment. (A) Standardized surgical decision-making flowchart for TC implemented at our research center. (B) Flowchart of patient enrollment for the unilateral MPTC cohort. (C) Research flowchart. CLNM, central lymph node metastasis; FNAB, fine-needle aspiration biopsy; MPTC, multifocal papillary thyroid carcinoma; OTC, occult thyroid carcinoma; PTC, papillary thyroid carcinoma; RAI, radioactive iodine; TC, thyroid carcinoma; US, ultrasound examination.

The inclusion criteria were as follows: (I) complete medical records; (II) preoperative US only reported multifocal nodules (≥2 nodules) in the unilateral gland lobe, with FNAB cytopathology classified as Bethesda category IV or higher; TT with CLND; (III) PTC; and (IV) no previous neck radiotherapy. The exclusion criteria were as follows: (I) partial absence of medical records; (II) preoperative ultrasonography and FNAB confirming any of the following: bilateral malignant nodules, unilateral unifocal nodule, or isthmus-only nodule; (III) TL with CLND; (IV) non-PTC; and (V) previous history of neck radiation.

This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This study was approved by the Ethics Committee of Union Hospital, Huazhong University of Science and Technology (No. 004[2025]). As a retrospective study, it was granted exemption from informed consent requirements under the premise of strict protection of patient privacy.

Data collection and variable selection

Based on numerous reports from the hospital, including both outpatient and inpatient records, we identified patients who presented with unilateral lobe (left or right lobe) characterized by multiple nodules as detected by US. All nodules were subsequently confirmed to be malignant through FANB. These patients then underwent TT. The clinicopathological characteristics of patients with contralateral OTC were investigated. Another outcome measure assessed was the number of CLNM. Due to the insufficient number of lymph nodes harvested by some surgeons during the surgical procedure, data from these patients who did not undergo CLND or had insufficient lymph node yield (e.g., 0 nodes pathologically confirmed) were excluded from the analysis of CLNM. The outcome measures regarding the number of contralateral OTC and CLNM were provided by the pathology department of our center. The study flowchart is presented in Figure 1B,1C.

The preoperative clinicopathological data of patients were collected from the electronic medical record system, including basic information [age, gender, body mass index (BMI)], US characteristics [the number of and the sum of the longest diameter (SLD) of unilateral malignant nodules, calcification, anteroposterior-to-transverse ratio, capsular interruption, isthmus malignant nodule, malignant nodules located near the isthmus], and thyroid function [hyperthyroidism, hypothyroidism, chronic lymphocytic thyroiditis (CLT)]. Figure 2 shows the representative ultrasonographic images.

Figure 2 Representative thyroid ultrasound images of malignant nodules. Ultrasound images demonstrating various presentations of malignant thyroid nodules. Top row (left to right): unilateral unifocal nodule; unilateral multifocal large-diameter nodules (two foci); unilateral multifocal small-diameter nodules (two foci, asterisk indicates a malignant nodule located near the isthmus). Middle and bottom rows: unilateral multifocal malignant nodules (multiple foci).

Data analysis

The software SPSS 7.00 (IBM Corp., Armonk, NY, USA), R software version 4.2.2 (The R Foundation for Statistical Computing, Vienna, Austria; https://www.r-project.org/), and Python (version 3.9.12, Python Software Foundation, Wilmington, DE, USA) were employed for statistical analysis. The chi-square test was utilized for count data. A P value <0.05 was considered statistically significant. Patients enrolled in the study were randomly divided into training and validation datasets at a ratio of 7:3, and there was no significant difference between the two datasets (P>0.05). The variable variance inflation coefficient was calculated to ensure the absence of collinearity between variables. To develop clinical prediction models with desired accuracy and reliability, an ensemble technique based on ML was devised. Under the framework of cross-validation, one algorithm was used for variable selection and another was employed to construct a classification prediction model. The data were tested against 12 ML methods, including least absolute shrinkage and selection operator (LASSO), Ridge, elastic net (Enet), stepwise generalized linear model (Stepglm), support vector machine (SVM), gradient boosting with component-wise linear models (glmBoost), linear discriminant analysis (LDA), partial least squares regression with generalized linear models (plsRglm), random forest (RF), gradient boosting machine (GBM), extreme gradient boosting (XGBoost), and Naive Bayes, with 113 algorithm combinations, and the area under the receiver operating characteristic (ROC) curve (AUC) of each model on the datasets was calculated (AUC index). Finally, the evaluation results of the model were visualized through heat maps. The optimal hyperparameters were obtained through cross-validation grid search, and the optimal hyperparameter combination was used to train the final model. The parameters of the most common models, including decision trees (DT), Enet, logistic regression, multi-layer perceptron (Mlp), RF, regularized SVM (RSVM), XGBoost, k-nearest neighbors (kNN), and LightGBM, were compared. The aforementioned steps aim to select the most optimal model. The ROC curves (26), calibration curves (27), and decision curve analysis (DCA) curves (28) were plotted. Python (shap 0.39.0) was utilized to draw the interpretation of the importance and contribution of SHapley Additive exPlanations (SHAP) to the model, and to interpret the model results by calculating the contribution of each feature to the prediction results (29).

Results

Construction and validation of the contralateral OTC prediction model

Clinical characteristics of the patients

The study collected 12,544 PTC patients who underwent TT. In accordance with the inclusion and exclusion criteria, 112,030 patients were excluded, and 514 patients were enrolled to analyze the clinicopathological factors, among whom 252 (49.03%) had contralateral OTC confirmed by pathology. The baseline characteristics of these patients are presented in Table 1. Hyperthyroidism and hypothyroidism were not considered in the subsequent modeling analyses as they were associated with an underrepresentation of positive outcomes.

Independent variables screening and prediction model development

We integrated 12 ML algorithms and 113 algorithm combinations to compute the AUC index of each model in the training and validation datasets (Figure 3A), among which the RF model had the highest average AUC index of 0.849. We identified the algorithms with an average AUC index greater than 0.8 and employed cross-validation grid search to optimize the hyperparameters to enhance the prediction performance and generalization ability of the model on diverse datasets. The variables selected by different algorithms were sorted in accordance with their significance, the top few variables were selected, and their intersections were taken to construct the final prediction model. When lambda.1se was adopted, five clinicopathological features with nonzero coefficients were screened in the training dataset through LASSO regression analysis (Figure 3B), including malignant isthmus nodules reported by US, the number and SLD of malignant nodules reported by US, capsule interruption reported by US, and CLT. Figure 3C illustrates the relationship between the number of trees and the error in the RF model. The ROC curve indicated that the AUC values of the training and validation datasets were 0.805 and 0.775, respectively. The five-fold cross-validation demonstrated that the average AUC value was 0.747 and the standard deviation was 0.016 (Figure 3D), suggesting that the performance of the model was relatively stable. Figure 3E displays the confusion matrices for both the training and the validation datasets in the RF model. The feature importance measure based on mean accuracy reduction and mean Gini reduction was utilized to screen the important variables as follows: the number and the SLD of malignant nodules reported by US, capsule interruption reported by US, isthmus malignant nodules reported by US, BMI, and gender (Figure 3F). The GBM model was cross-validated to obtain the best parameter combination of mtry =10, trees =1,687, min_n =132. The important variables selected in order of significance were malignant isthmus nodules, the number and the SLD of malignant nodules, capsular interruption, CLT, and calcification (Figure 3G). The ROC curve showed that the AUC values of the training and validation datasets were 0.774 and 0.769, respectively. The five-fold cross-validation revealed that the mean AUC was 0.749, and the standard deviation was 0.024 (Figure 3H). Figure 3I displays the confusion matrices for both the training and the validation datasets in the GBM model. Intersection variables, including malignant isthmus nodules reported by US, the number of malignant nodules reported by US, the SLD of malignant nodules reported by US, and capsular interruption reported by US, were screened and presented in a Venn diagram (Figure 3J).

Figure 3 ML model development and performance for predicting contralateral OTC. (A) Schematic of the 12 ML algorithms and 113 integrated algorithm combinations used in the analysis. (B) Feature selection using LASSO regression analysis to identify significant risk factors. (C) Cross-validation results and the relationship between the error and the number of DT (RF model). (D) ROC curve (RF model). (E) Confusion matrix (left: training dataset; right: validation dataset) (RF model). (F) Variable importance ranking (RF model). (G) Variable importance ranking (GBM model). (H) ROC curve (GBM model). (I) Confusion matrix (left: training dataset; right: validation dataset) (GBM model). (J) Venn diagram illustrating the overlap of important features identified by different models. AUC, area under the receiver operating characteristic curve; BMI, body mass index; CLT, chronic lymphocytic thyroiditis; DT, decision trees; Enet, elastic net; GBM, gradient boosting machine; glmBoost, gradient boosting with component-wise linear models; LASSO/Lasso, least absolute shrinkage and selection operator; LDA, linear discriminant analysis; ML, machine learning; OOB, out-of-bag; OTC, occult thyroid carcinoma; PR, precision-recall; RF, random forest; ROC, receiver operating characteristic; SLD, sum of the longest diameter; Stepglm, stepwise generalized linear model; SVM, support vector machine; XGBoost, extreme gradient boosting.

To visually elucidate the selected variables, we employed SHAP to demonstrate how these variables predict the hazard of contralateral OTC in the clinical model. Figure 4A presents the four variables included in the model, with each sample’s feature influence on the output result depicted using differently colored points. Identified by US, multifocality, large SLD, capsule interruption, and isthmus malignant nodule were identified as factors that elevate the risk of contralateral OTC in patients with PTC. Figure 4B illustrates the ranking of risk factors assessed by mean absolute SHAP values across different datasets, highlighting SLD as the most significant indicator influencing outcome occurrence. Figure 4C further explores both importance and interaction effects among features; specifically, each feature’s total importance value is derived from its interactions with other features. Each point represents an individual sample’s interaction value, with color indicating which feature interacts with the primary feature. Additionally, we present two representative cases to enhance the understanding of model interpretability: patients with contralateral OTC exhibited higher SHAP values (0.68) as shown in Figure 4D, whereas those without OTC demonstrated lower SHAP values (−0.56), as illustrated in Figure 4E. This further substantiates our model’s predictive capability.

Figure 4 SHAP explanation of the predictive model for contralateral OTC. (A) Each row represents a feature, and the abscissa is SHAP value (red points signify a positive effect of feature values on the prediction model, while blue points indicate a negative effect). (B) Ranking of feature importance in terms of SHAP. (C) Summary plot of interactions. Showing the importance and impact of interactions between features. (D,E) Force plots showing SHAP values for individual samples. (D) Contralateral OTC patients. (E) Non-contralateral OTC patients. OTC, occult thyroid carcinoma; SHAP, SHapley Additive exPlanations; SLD, sum of the longest diameter.

Construction and validation of CLNM (number ≥5) prediction model

Clinical characteristics of the patients

The study enrolled 12,060 PTC patients who underwent TT and bilateral CLND. Based on the established inclusion and exclusion criteria, 11,610 patients were excluded from the analysis, resulting in a final cohort of 450 patients whose clinicopathological factors were analyzed. A total of 174 (38.67%) patients exhibited lymph node metastasis with a quantity of five or more nodes. The baseline characteristics of the enrolled patients are shown in Table 2. Subsequent modeling analyses did not include the variables of hyperthyroidism and hypothyroidism.

Independent variables screening and prediction model development

We integrated nine common ML algorithms, including DT, Enet, logistic regression, Mlp, RF, RSVM, kNN, and LightGBM, and used a cross-validation grid search to optimize hyperparameters for model training and evaluation, ultimately comparing the performance of these models (Figure 5). After thoroughly assessing various parameters, such as recall, precision, specificity, ROC curves, calibration curves, and DCA curves, we identified the RF model as the most effective predictive model. Figure 6A presents a schematic diagram of the cross-validation results, and Figure 6B shows the relationship between the error and the number of DT. The ROC curve indicated that the AUC index for both the training and validation datasets was 0.806 and 0.749, respectively (Figure 6C). The five-fold cross-validation yielded an average AUC value of 0.739 with a standard deviation of 0.02 (Figure 6D). Figure 6E displays the confusion matrices for the training and validation datasets. To identify significant variables influencing predictions, based on the ranking of variable importance, we selected the following variables: the SLD of malignant nodules reported by US, capsular interruption reported by US, age, gender, and BMI to construct the final model (Figure 6F).

Figure 5 ML model development and performance for predicting CLNM (number ≥5). (A) Parallel plot illustrating the error indices of each model on the validation dataset. (B) ROC curve of each model on the validation dataset. (C) ROC-AUC index for each model evaluated on the validation dataset. (D) DCA curve representing each model’s performance on the validation dataset. (E) Calibration curve demonstrating the accuracy of predictions from each model on the validation dataset. AUC, area under the receiver operating characteristic curve; CLNM, central lymph node metastasis; DCA, decision curve analysis; DT, decision trees; Enet, elastic net; GBM, gradient boosting machine; KAP, knowledge, attitude, practice; kNN, k-nearest neighbors; MCC, Matthews correlation coefficient; ML, machine learning; Mlp, multi-layer perceptron; NPV, negative predictive value; PPV, positive predictive value; RF, random forest; ROC, receiver operating characteristic; RSVM, regularized support vector machine; Sens, sensitivity; Sepc, specificity; XGBoost, extreme gradient boosting; logistic, logistic regression.

Figure 6 RF model development and performance for predicting CLNM (number ≥5). (A) Diagram of cross-validation results. (B) Diagram illustrating the relationship between the error of the RF model and the number of DT. (C) ROC curve for the model on both training and validation datasets. (D) ROC curve for each fold of the model. (E) Confusion matrix (left: training dataset; right: validation dataset). (F) Importance ranking of variables. AUC, area under the receiver operating characteristic curve; BMI, body mass index; CI, confidence interval; CLNM, central lymph node metastasis; CLT, chronic lymphocytic thyroiditis; DT, decision trees; OOB, out-of-bag; PR, precision-recall; RF, random forest; ROC, receiver operating characteristic; SLD, sum of the longest diameter.

Based on the US results, the models were constructed to clarify the risk of adverse patient outcomes and to guide corresponding surgical decision-making (Figure 7).

Figure 7 Surgical method selection for unilateral MPTC patients. B-US, B-mode ultrasound; BMI, body mass index; CLNM, central lymph node metastasis; MPTC, multifocal papillary thyroid carcinoma; OTC, occult thyroid carcinoma; SLD, sum of the longest diameter.

Discussion

The incidence of TC has increased dramatically in several countries over the past few years, primarily due to the widespread use of imaging techniques and ultrasound-guided FNAB (30). The majority of newly diagnosed TC are subclinical PTC, which typically do not present obvious clinical symptoms or lead to mortality (3). The overall prognosis for this type of TC is favorable, and the mortality rate has remained relatively stable over the past three decades. Consequently, determining how to provide appropriate treatment for low- and medium-risk patients while avoiding insufficient treatment for high-risk patients has emerged as a significant research focus in the field of PTC diagnosis and management. The reoperation for PTC increases the complexity and risk of surgery due to alterations in thyroid anatomy and tissue adhesion from scarring, elevating the likelihood of postoperative complications such as injury to the recurrent laryngeal nerve and parathyroid glands. Additionally, there is an increased risk associated with anesthesia and a greater expenditure of medical resources (31,32). However, performing TT and prophylactic lymph node dissection without careful consideration also heightens the risk of surgery-related complications (33). Furthermore, patients may require high-dose lifelong thyroid hormone replacement therapy. OTC refers to cases where cancer is not identified during preoperative evaluations but is discovered through intraoperatively or postoperatively pathological examination. Reports indicate that the incidence of contralateral OTC in total/subtotal thyroidectomy specimens only ranges from 13% to 56% (34). This might suggest that a significant number of patients undergoing TT may be subjected to overtreatment. Consequently, accurately predicting and evaluating the risks associated with contralateral OTC and CLNM prior to surgery is crucial for selecting optimal surgical method to avoid overtreatment while considering the risk of postoperative recurrence (35,36). We aimed to provide guidance for selecting surgical method based on clinically available information before surgery to guide individualized treatment strategies.

In this study, we developed clinical prediction models for contralateral OTC and CLNM (number ≥5) in patients with unilateral MPTC, utilizing highly correlated variables identified through ML. US features, including capsular interruption, malignant isthmus nodule, and the number and the SLD of malignant nodules, were found to be predictive variables of the risk of contralateral OTC preoperatively. Additionally, factors observed on US, including the SLD of malignant nodules, capsular interruption, and patient’s basic information, including age, gender, and BMI, could predict the risk of CLNM (number ≥5) prior to surgery. The models can be used to guide the decision-making of surgical methods, enabling more effective individualized treatment strategies for PTC patients while compensating for the defect of the difficulty in obtaining pathological factors when making surgical plans. Based on the prediction models, we propose that TL may be feasible for patients classified as low or medium risk; meanwhile, surgical method should be determined according to recurrence risks and patient’s thoughts in high-risk cases. In comparison to previous prediction models used for surgical decision-making in PTC cases, this study emphasizes preoperative indicators that are readily available. It innovatively offers preoperative decision-making information regarding the extent of surgery required contrasting traditional studies that primarily relied on postoperative pathological diagnoses. The comprehensive assessment of the risk of contralateral OTC and CLNM provides more comprehensive evidence support. However, the study’s inclusion/exclusion criteria may introduce selection biases, including: (I) restricted generalizability due to exclusive focus on preoperative US-confirmed unilateral MPTCs (excluding bilateral/unifocal/isthmic variants with potentially distinct biological behaviors); (II) systematic overrepresentation of patients undergoing TT with CLND, potentially underestimating occult metastasis rates in real-world low-risk cohorts; and (III) exclusion of cases with incomplete medical records, favoring “ideal” tertiary-center datasets while obscuring community hospital heterogeneity. Collectively, these factors may limit the extrapolation of findings to broader PTC populations. Therefore, future research will adopt a prospective design aimed at validating and refining our models; additionally, multi-center external validation will be conducted to derive more objective conclusions.

Conclusions

We developed preoperative prediction models utilizing ML methods to assess the risk of contralateral OTC and CLNM (number ≥5) in patients with unilateral MPTC. This study aimed to minimize overtreatment while considering the potential for recurrence, thereby providing guidance for the precise implementation of individualized surgical strategies.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD+AI reporting checklist. Available at https://qims.amegroups.com/article/view/10.21037/qims-2024-2924/rc

Data Sharing Statement: Available at https://qims.amegroups.com/article/view/10.21037/qims-2024-2924/dss

Funding: This study was funded by the National Natural Science Foundation of China (Nos. 82203813, 82203147, 82373055, 82573921, and 82373144), the Hubei Provincial Key Research and Development Program (Nos. 2023BCB011 and 2024BEB020), and the High-Level Talents Program of Hainan Province Natural Science Foundation (No. 823RC594).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://qims.amegroups.com/article/view/10.21037/qims-2024-2924/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This study was approved by the Ethics Committee of Union Hospital, Huazhong University of Science and Technology (No. 004[2025]). As a retrospective study, it was granted exemption from informed consent requirements under the premise of strict protection of patient privacy.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Pacini F, Castagna MG, Brilli L, Pentheroudakis G. Thyroid cancer: ESMO Clinical Practice Guidelines for diagnosis, treatment and follow-up. Ann Oncol 2012;23:vii110-9. [Crossref] [PubMed]
2. Filetti S, Durante C, Hartl D, Leboulleux S, Locati LD, Newbold K, Papotti MG, Berruti AESMO Guidelines Committee. Electronic address: clinicalguidelines@esmo. Thyroid cancer: ESMO Clinical Practice Guidelines for diagnosis, treatment and follow-up†. Ann Oncol 2019;30:1856-83. [Crossref] [PubMed]
3. Sung H, Ferlay J, Siegel RL, Laversanne M, Soerjomataram I, Jemal A, Bray F. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin 2021;71:209-49. [Crossref] [PubMed]
4. Li M, Hu M, Jiang L, Pei J, Zhu C. Trends in Cancer Incidence and Potential Associated Factors in China. JAMA Netw Open 2024;7:e2440381. [Crossref] [PubMed]
5. de Almeida JFM, Ward LS. Thyroid autoimmune diseases and thyroid tumors: Would EBV infection be the link? J Cell Physiol 2019;234:19141-2. [Crossref] [PubMed]
6. Germano A, Schmitt W, Almeida P, Mateus-Marques R, Leite V. Ultrasound requested by general practitioners or for symptoms unrelated to the thyroid gland may explain higher prevalence of thyroid nodules in females. Clin Imaging 2018;50:289-93. [Crossref] [PubMed]
7. Boucai L, Zafereo M, Cabanillas ME. Thyroid Cancer: A Review. JAMA 2024;331:425-35. [Crossref] [PubMed]
8. Moore EC, Zolin S, Krishnamurthy V, Jin J, Shin J, Berber E, Siperstein A. Need for Completion Thyroidectomy in Patients Undergoing Lobectomy for Indeterminate and High-Risk Nodules: Impact of Intra-Operative Findings and Final Pathology. World J Surg 2020;44:408-16. [Crossref] [PubMed]
9. Haugen BR, Alexander EK, Bible KC, Doherty GM, Mandel SJ, Nikiforov YE, Pacini F, Randolph GW, Sawka AM, Schlumberger M, Schuff KG, Sherman SI, Sosa JA, Steward DL, Tuttle RM, Wartofsky L. 2015 American Thyroid Association Management Guidelines for Adult Patients with Thyroid Nodules and Differentiated Thyroid Cancer: The American Thyroid Association Guidelines Task Force on Thyroid Nodules and Differentiated Thyroid Cancer. Thyroid 2016;26:1-133. [Crossref] [PubMed]
10. Cui L, Feng D, Zhu C, Li Q, Li W, Liu B. Clinical outcomes of multifocal papillary thyroid cancer: A systematic review and meta-analysis. Laryngoscope Investig Otolaryngol 2022;7:1224-34. [Crossref] [PubMed]
11. Yan T, Qiu W, Song J, Ying T, Fan Y, Yang Z. Bilateral multifocality, a marker for aggressive disease, is not an independent prognostic factor for papillary thyroid microcarcinoma: A propensity score matching analysis. Clin Endocrinol (Oxf) 2021;95:209-16. [Crossref] [PubMed]
12. Kuhn E, Teller L, Piana S, Rosai J, Merino MJ. Different clonal origin of bilateral papillary thyroid carcinoma, with a review of the literature. Endocr Pathol 2012;23:101-7. [Crossref] [PubMed]
13. Zhao Q, Ming J, Liu C, Shi L, Xu X, Nie X, Huang T. Multifocality and total tumor diameter predict central neck lymph node metastases in papillary thyroid microcarcinoma. Ann Surg Oncol 2013;20:746-52. [Crossref] [PubMed]
14. Katoh R, Sasaki J, Kurihara H, Suzuki K, Iida Y, Kawaoi A. Multiple thyroid involvement (intraglandular metastasis) in papillary thyroid carcinoma. A clinicopathologic study of 105 consecutive patients. Cancer 1992;70:1585-90. [Crossref] [PubMed]
15. Cabanillas ME, McFadden DG, Durante C. Thyroid cancer. Lancet 2016;388:2783-95. [Crossref] [PubMed]
16. Sterpetti AV. Optimization of staging of the neck with prophylactic central and lateral neck dissection for papillary thyroid carcinoma. Ann Surg 2015;261:e30. [Crossref] [PubMed]
17. Kim HJ, Sohn SY, Jang HW, Kim SW, Chung JH. Multifocality, but not bilaterality, is a predictor of disease recurrence/persistence of papillary thyroid carcinoma. World J Surg 2013;37:376-84. [Crossref] [PubMed]
18. Bothra S, Mayilvaganan S. Multifocal Versus Solitary Papillary Thyroid Carcinoma. World J Surg 2017;41:335. [Crossref] [PubMed]
19. Al Afif A, Williams BA, Rigby MH, Bullock MJ, Taylor SM, Trites J, Hart RD. Multifocal Papillary Thyroid Cancer Increases the Risk of Central Lymph Node Metastasis. Thyroid 2015;25:1008-12. [Crossref] [PubMed]
20. Lu WW, Zhang D, Ni XJ. A Review of the Role of Ultrasound Radiomics and Its Application and Limitations in the Investigation of Thyroid Disease. Med Sci Monit 2022;28:e937738. [Crossref] [PubMed]
21. O'Connell K, Yen TW, Quiroz F, Evans DB, Wang TS. The utility of routine preoperative cervical ultrasonography in patients undergoing thyroidectomy for differentiated thyroid cancer. Surgery 2013;154:697-701; discussion 701-3. [Crossref] [PubMed]
22. Zhao H, Li H. Meta-analysis of ultrasound for cervical lymph nodes in papillary thyroid cancer: Diagnosis of central and lateral compartment nodal metastases. Eur J Radiol 2019;112:14-21. [Crossref] [PubMed]
23. Wan F, He W, Zhang W, Zhang H, Zhang Y, Guang Y. Application of decision tree algorithms to predict central lymph node metastasis in well-differentiated papillary thyroid carcinoma based on multimodal ultrasound parameters: a retrospective study. Quant Imaging Med Surg 2023;13:2081-97. [Crossref] [PubMed]
24. Zou Y, Shi Y, Bi H, Tan J, Guo Q, Qin Y, Lu X, Ma X, Yang S, Liu J. A nomogram for risk stratification of central cervical lymph node metastasis in patients with papillary thyroid carcinoma. Quant Imaging Med Surg 2024;14:5084-98. [Crossref] [PubMed]
25. Xing Z, Qiu Y, Yang Q, Yu Y, Liu J, Fei Y, Su A, Zhu J. Thyroid cancer neck lymph nodes metastasis: Meta-analysis of US and CT diagnosis. Eur J Radiol 2020;129:109103. [Crossref] [PubMed]
26. Mandrekar JN. Receiver operating characteristic curve in diagnostic test assessment. J Thorac Oncol 2010;5:1315-6. [Crossref] [PubMed]
27. Giai Gianetto Q, Combes F, Ramus C, Bruley C, Couté Y, Burger T. Calibration plot for proteomics: A graphical tool to visually check the assumptions underlying FDR control in quantitative experiments. Proteomics 2016;16:29-32. [Crossref] [PubMed]
28. Van Calster B, Wynants L, Verbeek JFM, Verbakel JY, Christodoulou E, Vickers AJ, Roobol MJ, Steyerberg EW. Reporting and Interpreting Decision Curve Analysis: A Guide for Investigators. Eur Urol 2018;74:796-804. [Crossref] [PubMed]
29. Lundberg SM, Erion G, Chen H, DeGrave A, Prutkin JM, Nair B, Katz R, Himmelfarb J, Bansal N, Lee SI. From Local Explanations to Global Understanding with Explainable AI for Trees. Nat Mach Intell 2020;2:56-67. [Crossref] [PubMed]
30. Zhang K, Qian L, Chen J, Zhu Q, Chang C. Preoperative Prediction of Central Cervical Lymph Node Metastasis in Fine-Needle Aspiration Reporting Suspicious Papillary Thyroid Cancer or Papillary Thyroid Cancer Without Lateral Neck Metastasis. Front Oncol 2022;12:712723. [Crossref] [PubMed]
31. Kluijfhout WP, Pasternak JD, Lim J, Kwon JS, Vriens MR, Clark OH, Shen WT, Gosnell JE, Suh I, Duh QY. Frequency of High-Risk Characteristics Requiring Total Thyroidectomy for 1-4 cm Well-Differentiated Thyroid Cancer. Thyroid 2016;26:820-4. [Crossref] [PubMed]
32. Hartl DM, Guerlain J, Breuskin I, Hadoux J, Baudin E, Al Ghuzlan A, Terroir-Cassou-Mounat M, Lamartina L, Leboulleux S. Thyroid Lobectomy for Low to Intermediate Risk Differentiated Thyroid Cancer. Cancers (Basel) 2020;12:3282. [Crossref] [PubMed]
33. Christou N, Mathonnet M. Complications after total thyroidectomy. J Visc Surg 2013;150:249-56. [Crossref] [PubMed]
34. Koo BS, Lim HS, Lim YC, Yoon YH, Kim YM, Park YH, Rha KS. Occult contralateral carcinoma in patients with unilateral papillary thyroid microcarcinoma. Ann Surg Oncol 2010;17:1101-5. [Crossref] [PubMed]
35. Mao Y, Xing M. Recent incidences and differential trends of thyroid cancer in the USA. Endocr Relat Cancer 2016;23:313-22. [Crossref] [PubMed]
36. Adam MA, Pura J, Goffredo P, Dinan MA, Reed SD, Scheri RP, Hyslop T, Roman SA, Sosa JA. Presence and Number of Lymph Node Metastases Are Associated With Compromised Survival for Patients Younger Than Age 45 Years With Papillary Thyroid Cancer. J Clin Oncol 2015;33:2370-5. [Crossref] [PubMed]