Development and validation of prediction models for papillary thyroid cancer structural recurrence using machine...

Area of Science:

• Oncology
• Medical Informatics
• Biostatistics

Background:

• Papillary thyroid cancer (PTC) recurrence affects up to 30% of patients despite good prognosis.
• Predicting PTC recurrence accurately remains challenging due to controversial predictors.
• Investigating machine learning (ML) for enhanced PTC structural recurrence risk prediction.

Purpose of the Study:

• To evaluate the efficacy of machine learning (ML) models in predicting structural recurrence risk in papillary thyroid cancer (PTC) patients.
• To compare the performance of various ML algorithms against the American Thyroid Association (ATA) risk stratification.
• To identify key predictors for structural recurrence in PTC.

Main Methods:

• Utilized data from 2244 PTC patients treated with surgery and radioiodine.
• Analyzed 29 perioperative variables across demographic, tumor, lymph node, and metabolic/inflammatory factors.
• Applied and compared five ML algorithms: logistic regression (LR), support vector machine (SVM), extreme gradient boosting (XGBoost), random forest (RF), and neural network (NN).

Main Results:

• Machine learning models demonstrated superior predictive performance (AUC 0.738–0.767) compared to ATA risk stratification (AUC 0.620).
• The Random Forest (RF) model showed high sensitivity (0.676), specificity (0.784), and negative predictive value (0.964).
• Key predictors identified include non-stimulated thyroglobulin (Tg), lymph node ratio (LNR), and N stage.

Conclusions:

• Machine learning, particularly the RF model, offers a promising approach for accurate risk stratification in PTC patients.
• The developed ML models provide better discrimination, calibration, and interpretability than traditional methods.
• This study highlights the potential of ML to improve clinical decision-making for PTC management.