背景：手术部位感染（Surgical site infection, SSI）是常见的术后并发症，其中器官/间隙（Organ/space, O/S）手术部位感染的危害尤为严重，可显著增加死亡率及医疗负担。目前多数预测模型未区分SSI的具体分型，影响了预测的准确性和临床适用性。因此，建立针对O/S SSI的预测模型具有重要的临床意义。

目的：本研究旨在基于逻辑回归方法，建立并验证腹部肿瘤大手术后O/S SSI的预测模型，以协助临床早期识别高危患者并优化围术期管理。

方法：采用回顾性队列研究设计，纳入2013年1月至2023年12月在北京协和医院接受择期腹部肿瘤大手术的4133例患者数据，分为训练集（3503例）和验证集（630例）。通过结构化提取与人工检索相结合的方式收集术前、术中及术后38个变量。采用逻辑回归方法构建模型，并通过限制性立方样条（Restricted cubic splines, RCS）优化变量线性相关性。模型评估基于区分度（曲线下面积、灵敏度、特异度等）、校准度（Brier评分、校准曲线）及可解释性（列线图、森林图）。

结果：训练集中O/S SSI发生率为12.4%（435例），验证集为10.9%（69例）。最终纳入了18个变量的模型3A总体性能最佳，在训练集中的曲线下面积（Area under the curve, AUC）为0.771 (0.749, 0.793)，灵敏度为0.807 (0.770, 0.844)，特异度为0.603 (0.585, 0.620)，阳性似然比为2.031 (1.906, 2.164)，阴性似然比为0.320 (0.264, 0.389)，阳性预测值为0.224 (0.203, 0.244)，阴性预测值为0.957 (0.947, 0.966)，Brier评分为0.095。在验证集中的AUC为0.733 (0.667, 0.798)，灵敏度为0.652 (0.540, 0.765)，特异度为0.725 (0.689, 0.762)，阳性似然比为2.376 (1.909, 2.956)，阴性似然比为0.479 (0.346, 0.665)，阳性预测值为0.226 (0.168, 0.284)，阴性预测值为0.940 (0.923, 0.966)，Brier评分为0.090。

结论：本研究成功建立了基于逻辑回归的O/S SSI预测模型，具有良好的区分度和校准度，可为临床提供术前风险分层和术后动态监测工具。模型结合了静态与动态变量，具有较高的临床实用性和可解释性，有助于优化围术期管理策略。

Background: Surgical site infection (SSI) is a common postoperative complication, with organ/space SSI (O/S SSI) being particularly severe, as it significantly increases mortality and healthcare burden. Most existing prediction models do not differentiate between specific types of SSI, which limits their accuracy and clinical applicability. Therefore, developing a prediction model specifically for O/S SSI holds significant clinical value.

Objective: This study aimed to develop and validate a predictive model for O/S SSI following major abdominal cancer surgery using logistic regression, to assist in early identification of high-risk patients and optimize perioperative management.

Methods: A retrospective cohort study design was adopted, including data from 4,133 patients who underwent elective major abdominal cancer surgery at Peking Union Medical College Hospital between January 2013 and December 2023. The dataset was divided into a training set (3,503 cases) and a validation set (630 cases). Thirty-eight preoperative, intraoperative, and postoperative variables were collected through structured extraction and manual retrieval. A logistic regression model was constructed, with restricted cubic splines (RCS) used to optimize variable linearity. Model evaluation was based on discrimination (Area Under the Curve [AUC], sensitivity, specificity, etc), calibration (Brier score, calibration curve), and interpretability (nomogram, forest plot).

Results: The incidence of O/S SSI was 12.4% (435 cases) in the training set and 10.9% (69 cases) in the validation set. The final model (Model 3A), incorporating 18 variables, demonstrated the best overall performance. In the training set, the AUC was 0.771 (0.749, 0.793), sensitivity was 0.807 (0.770, 0.844), specificity was 0.603 (0.585, 0.620), positive likelihood ratio was 2.031 (1.906, 2.164), negative likelihood ratio was 0.320 (0.264, 0.389), positive predictive value was 0.224 (0.203, 0.244), negative predictive value was 0.957 (0.947, 0.966), and Brier score was 0.095. In the validation set, the AUC was 0.733 (0.667, 0.798), sensitivity was 0.652 (0.540, 0.765), specificity was 0.725 (0.689, 0.762), positive likelihood ratio was 2.376 (1.909, 2.956), negative likelihood ratio was 0.479 (0.346, 0.665), positive predictive value was 0.226 (0.168, 0.284), negative predictive value was 0.940 (0.923, 0.966), and Brier score was 0.090.

Conclusion: This study successfully developed a logistic regression-based predictive model for O/S SSI with good discrimination and calibration, providing a tool for preoperative risk stratification and postoperative dynamic monitoring. The model combines static and dynamic variables, demonstrating high practicality and interpretability, which can help optimize perioperative management strategies.