Development and validation of a multidimensional machine learning-based nomogram for predicting central lymph node metastasis in papillary thyroid microcarcinoma

Abstract

Background: Papillary thyroid microcarcinoma (PTMC), a subset of papillary thyroid carcinoma (PTC), is characterized by tumors ≤10 mm in size. While generally indolent, central lymph node metastasis (CLNM) is associated with higher risks of recurrence and distant metastasis. Existing prediction models for CLNM predominantly depend on isolated clinical or imaging parameters, failing to integrate multidimensional predictors such as clinicopathological, ultrasonographic, and serological features. This limitation significantly undermines their clinical applicability. Therefore, we developed a machine learning-based nomogram that integrates comprehensive predictors to enhance preoperative risk stratification and facilitate personalized surgical decision-making.

Methods: A retrospective study was conducted on 503 PTMC patients who underwent thyroidectomy in Liaoyang Central Hospital between 2020 and 2023. Patients were randomly divided into training (n=352) and validation (n=151) cohorts. Inclusion criteria required preoperative imaging to confirm no cervical lymph node metastasis (LNM), complete clinicopathologic data, and initial surgery with central lymph node dissection, as well as postoperative pathology confirming PTC. Multidimensional predictors (clinical demographics, ultrasonographic features, serological markers, and histopathological characteristics) were analyzed. CLNM was definitively diagnosed via postoperative histopathology. Least absolute shrinkage and selection operator (LASSO) regression was used to identify key predictors, which were incorporated into a logistic regression model. The model's performance was evaluated using receiver operating characteristic (ROC) curves, calibration plots, and decision curve analysis (DCA).

Results: Among 503 enrolled patients (mean age: 48.5 years; male: 24%, female: 76%), CLNM was pathology confirmed in 28.8% (145/503). Age, gender, tumor size, tumor location, and extrathyroidal extension (ETE) were identified as independent predictors of CLNM. The nomogram achieved an area under the curve (AUC) of 0.88 (sensitivity 0.84, specificity 0.76) in the training cohort and 0.78 (sensitivity 0.80, specificity 0.70) in the validation cohort. Calibration plots indicated excellent agreement between predicted and observed probabilities, with mean absolute errors below 0.05. DCA demonstrated clinical utility for threshold probabilities ranging from 15% to 88%. These results suggest that the nomogram has good predictive performance and clinical applicability in assessing the risk of CLNM in PTMC patients.

Conclusions: This Machine learning-based predictive nomogram provides a reliable tool for assessing CLNM risk in PTMC patients, supporting personalized surgical strategies. Further validation in external cohorts is required to confirm its generalizability.

Keywords: Papillary thyroid microcarcinoma (PTMC); lymph node metastasis (LNM); machine learning; nomogram; thyroidectomy.

Figure 1

Study flow chart: shows the process of patient screening and grouping. PTMC, papillary thyroid microcarcinoma; LASSO, least absolute shrinkage and selection operator; ROC, receiver operating characteristic; DCA, decision curve analysis.

Figure 2

LASSO coefficients plot: shows the path of variable coefficients as a function of the strength of regularization and the choice of the optimal λ value. (A) Coefficient path diagram; (B) cross-validation plot. LASSO, least absolute shrinkage and selection operator.

Figure 3

Nomogram: provides a tool for predicting the risk of CLNM. CLNM, central lymph node metastasis; ETE, extrathyroidal extension.

Figure 4

ROC curve: assesses the prediction accuracy of the model in the training and validation groups. (A) Training group ROC curve; (B) validation group ROC curve. AUC, area under the curve; CI, confidence interval; ROC, receiver operating characteristic.

Figure 5

Calibration curve: this figure illustrates the degree of agreement between the model predictions and the actual results. (A) Training group calibration curves; (B) validation group calibration curves.

Figure 6

DCA curve: this figure assesses the clinical decision-making benefit of the model at different thresholds. (A) Training group calibration curves; (B) validation group calibration curves. DCA, decision curve analysis.