关键词: 甲状腺癌, 甲状旁腺损伤, 影响因素, 预测模型

Abstract: Objective  To analyze the risk factors associated with intraoperative parathyroid injury during thyroid cancer surgery and develop a predictive model to evaluate its predictive efficacy.Methods A retrospective study was conducted involving 76 patients who experienced intraoperative parathyroid injury during thyroid cancer surgery between May 2021 and February 2024.These patients were included in the parathyroid injury group.For the control group,76 patients who underwent thyroid cancer surgery without intraoperative parathyroid injury during the same period were selected.Clinical data,complete blood count parameters,and other relevant variables were collected.Univariate analysis was performed to screen for potential risk factors,and Logistic regression analysis was used to identify independent risk factors.A regression predictive model was established,and the model's goodness-of-fit and predictive power were evaluated.Results Among the 76 patients, 75 had temporary parathyroid function impairment and 1 had permanent parathyroid function impairment. Univariate and multivariate Logistic regression analyses showed that bilateral lymph node dissection, bilateral total lobectomy, capsule invasion, and combined hashimoto's thyroiditis were independent risk factors (P< 0.05).A predictive regression model was developed based on these factors.The Hosmer-Lemeshow test showed a χ2 = 2.064,P=0.356,indicating good model fit.The receiver operating characteristic (ROC) curve revealed that the model had an area under the curve (AUC) of 0.713,with a 95% confidence interval (CI) of 0.639 to 0.787,suggesting good predictive efficacy.Conclusion Hashimoto's thyroiditis,capsular invasion,bilateral lymph node dissection,and bilateral total lobectomy are significant risk factors for intraoperative parathyroid injury during thyroid cancer surgery.The predictive model constructed based on these factors has good predictive efficacy.

Key words: thyroid Cancer, parathyroid Injury, influence factors, predictive model