An interpretable nomogram for predicting early acute postoperative hypocalcemia in differentiated thyroid cancer:...

Area of Science:

• Endocrinology
• Oncology
• Medical Informatics

Background:

• Postoperative hypocalcemia (HC) is a common complication after thyroidectomy.
• Clinical management faces a dilemma in balancing calcium supplementation to prevent HC without overtreatment.
• Predicting early acute HC is crucial for differentiated thyroid cancer (DTC) patients.

Purpose of the Study:

• To develop and validate an interpretable machine learning (ML) model for predicting early acute postoperative HC in DTC patients.
• To identify key clinical, biochemical, and surgical predictors of HC.
• To create a tool for individualized perioperative management.

Main Methods:

• Retrospective analysis of 876 DTC patients' data, adhering to TRIPOD-AI guidelines.
• Data split into training (80%) and independent test (20%) sets.
• Seven ML algorithms evaluated; Logistic Regression (LR) selected for transparency and performance.
• Shapley Additive exPlanations (SHAP) and a clinical nomogram used for interpretation.

Main Results:

• Overall incidence of early acute HC was 38.2%.
• The LR model achieved an area under the ROC curve of 0.760 in the test set.
• Key predictors included preoperative serum magnesium, BMI, lymph node dissection, and parathyroid autotransplantation (OR=6.808).
• The LR-derived nomogram showed 0.806 sensitivity at a threshold of 0.372.

Conclusions:

• A transparent and validated LR-based nomogram accurately predicts early acute postoperative HC within 24 hours after thyroidectomy.
• The model integrates surgical and biochemical factors for personalized patient care.
• This tool aids in optimizing perioperative monitoring and calcium supplementation strategies.