构建并比较多种甲状腺癌术后并发低钙血症风险预测模型，筛选核心预测因子，为临床早期预警与精准干预提供科学工具，优化患者围手术期管理。

回顾性纳入2022年4月1日至2025年8月31日惠州市第一人民医院收治的767例分化型甲状腺癌手术患者，以术后是否发生低钙血症为结局事件。通过最小绝对收缩和选择算子（LASSO）回归筛选核心预测因子，基于以上因子构建多变量列线图、决策树及极端梯度提升（XGBoost）3种预测模型。采用分层抽样（8∶2）将样本集划分训练集与测试集，以准确性、精准率、召回率、F1分数及受试者操作特征曲线下面积（AUC）评估模型性能，并通过SHAP方法解析XGBoost模型的变量贡献。

767例患者中，233例（30.4%）发生术后低钙血症。LASSO回归筛选出年龄、体重指数、术后甲状旁腺激素（PTH）、手术范围、复发风险、甲状腺微小乳头状癌、肿瘤数量、糖尿病史及甲状腺炎症9项核心预测因子。3种模型均展现良好的判别能力（AUC≥0.828）：XGBoost模型的AUC最高（0.851），但召回率最低（0.435）；决策树模型召回率（0.587）和F1分数（0.651）最优；列线图模型准确性（0.816）和精准率（0.781）最高，且实现了个体化风险的可视化便捷评估。SHAP分析验证术后PTH和手术范围是贡献度最高的预测因子。

本研究筛选的9项核心预测因子对甲状腺癌术后低钙血症的发生具有重要预测价值。构建的列线图模型兼具准确性、易用性和临床可解释性，可有效辅助医师识别高危患者并制定个体化干预策略，对改善患者预后具有重要意义。

To develop and compare multiple risk prediction models for postoperative hypocalcemia in patients with thyroid cancer, identify core predictive factors, and provide a scientific tool for early clinical warning and precision intervention to optimize perioperative management.

A total of 767 patients who underwent surgery for differentiated thyroid cancer were retrospectively enrolled from April 1, 2022 to August 31, 2025 in Huizhou First People’s Hospital, with the occurrence of postoperative hypocalcemia as the primary outcome. Core predictors were selected using least absolute shrinkage and selection operator (LASSO) regression. Based on these predictors, three prediction models-a multivariate nomogram, a decision tree, and an extreme gradient boosting (XGBoost) model-were constructed. The dataset was split into training and testing sets at an 8∶2 ratio using stratified sampling. Model performance was evaluated using accuracy, precision, recall, F1 score, and the area under the receiver operating characteristic curve (AUC). SHapley additive explanations (SHAP) values were employed to interpret feature contributions in the XGBoost model.

Among the 767 patients, 233 (30.4%) developed postoperative hypocalcemia. LASSO regression identified nine core predictors: age, body mass index (BMI), postoperative parathyroid hormone (PTH) level, extent of surgery, recurrence risk, presence of papillary microcarcinoma, number of tumors, history of diabetes mellitus, and thyroiditis. All three models demonstrated good discriminative ability (AUC≥0.828). The XGBoost model achieved the highest AUC (0.851) but the lowest recall (0.435). The decision tree model showed the best recall (0.587) and F1 score (0.651). The nomogram model yielded the highest accuracy (0.816) and precision (0.781), and enabled convenient, visualized individualized risk assessment. SHAP analysis confirmed that postoperative PTH level and extent of surgery were the most influential predictors.

The nine core predictors identified in this study hold significant value for predicting postoperative hypocalcemia in thyroid cancer patients. Among the models evaluated, the nomogram offers the best balance of accuracy, usability, and clinical interpretability, making it the optimal tool for clinical application. It can effectively assist clinicians in identifying high-risk patients and implementing personalized interventions, thereby improving patient outcomes.