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

Postoperative hypocalcemia (HC) is a frequent complication following thyroidectomy. Balancing prophylactic calcium supplementation against avoiding overtreatment remains a clinical dilemma. This study aimed to develop and independently validate an interpretable machine learning (ML) model to predict early acute postoperative HC in patients with differentiated thyroid cancer (DTC).

In accordance with the Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis plus Artificial Intelligence reporting guideline, diverse clinical, biochemical, and surgical parameters from 876 DTC patients (February 2023-January 2026) were retrospectively analyzed. Early acute postoperative HC was defined as biochemical or symptomatic hypocalcemia occurring within the first 24 hours after surgery. The cohort was randomly split into a training set (80%, n = 701) and an independent holdout test set (20%, n = 175). Seven ML algorithms were trained and evaluated. Shapley Additive exPlanations (SHAP) and a clinical nomogram were utilized for model interpretation and clinical application.

The overall incidence of early acute postoperative HC was 38.2% (335/876). While complex ensemble models showed comparable performance, Logistic Regression (LR) was selected as the final model for its optimal balance of favorable discrimination (area under the receiver operating characteristic curve in the test set = 0.760) and clinical transparency. Multivariate and SHAP analyses revealed that preoperative serum magnesium, body mass index, lateral cervical lymph node dissection, and explicit parathyroid autotransplantation dynamics (specifically implantation into the neck muscle, odds ratio = 6.808) were critical drivers of HC. The LR-derived nomogram demonstrated excellent sensitivity (0.806 in the test set) at the optimal threshold of 0.372. Decision curve analysis confirmed superior net clinical benefit across relevant threshold probabilities, and subgroup analyses verified the model's stability across age and sex strata.

We developed and independently validated a transparent, highly discriminative LR-based nomogram for predicting early acute postoperative HC within the first 24 hours after thyroidectomy. Integrating precise surgical factors and biochemical profiles, this tool facilitates individualized perioperative monitoring and tailored calcium supplementation strategies.