Artificial intelligence for predicting transient hypocalcemia after total thyroidectomy

Transient hypocalcemia is a common complication of total thyroidectomy. This study aimed to evaluate whether machine learning (ML)-based models could enhance early risk prediction and support a robust clinical decision support system for hypocalcemia. We proposed a comprehensive ML-based framework using real-world health record-derived data from the preoperative, intraoperative, and postoperative periods of 7,066 patients treated and followed longitudinally at the A.C. Camargo Cancer Center, São Paulo, Brazil, between January 2014 and January 2024. The model was prospectively validated in 149 newly operated patients, of which 31 were confirmed cases of transient hypocalcemia. In this evaluation, the postoperative model achieved an AUC of 0.94 and a precision of 0.88. Known risk factors such as low postoperative parathyroid hormone (PTH) and bilateral neck dissection were reaffirmed, while novel insights such as prolonged surgery duration, earlier start times, and surgeon workload were identified as additional contributors to risk. Our study demonstrates that machine learning models, particularly when integrated with explainable artificial intelligence (XAI) techniques, offer a reliable and interpretable tool for predicting transient hypocalcemia, enhancing personalized postoperative care and enabling timely, risk-based interventions.