Interpretable machine learning for thyroid cancer recurrence predicton: Leveraging XGBoost and SHAP analysis.

Area of Science:

• Oncology
• Biostatistics
• Medical Informatics

Background:

• Differentiated thyroid cancer (DTC) recurrence risk is assessed using clinical, laboratory, and pathological features.
• Existing prognostic models require validation and adjustment in large patient cohorts with long-term follow-up.

Purpose of the Study:

• To develop and validate an XGBoost model for accurate DTC recurrence prediction.
• To identify critical risk factors and establish new recurrence risk thresholds.
• To improve patient-centric care through informed decision-making.

Main Methods:

• Retrospective study of 1228 DTC patients (1976-2010).
• Development of an XGBoost model using clinical and biomarker features.
• Application of Shapely Additive exPlanations (SHAP) for model interpretability.

Main Results:

• XGBoost model achieved an AUROC of 0.88 on an independent test set.
• Key predictors identified: tumor size, post-operative thyroglobulin, and thyroglobulin antibody levels.
• SHAP analysis suggested new risk thresholds for tumor size (25 mm) and biomarker levels.

Conclusions:

• The developed XGBoost model provides accurate and interpretable DTC recurrence risk prediction.
• SHAP analysis offers defined risk thresholds to support clinical decision-making.
• The model empowers clinicians for enhanced patient-centric care.