探讨原发性肝癌（肝癌）肝切除术后大量腹水的危险因素，并建立其预测的列线图模型。

回顾性分析2018年1月至2022年12月川北医学院附属医院收治的739例肝癌患者临床资料。其中男585例，女154例；年龄21~89岁，中位年龄60岁。根据手术时间及手术科室不同，分为训练集和验证集，其中训练集536例，验证集203例。训练集数据用于模型建立、模型评价与内部验证，验证集数据用于外部验证。通过查询医院科研数据大平台及专病数据库获取纳入研究患者所需的临床资料，在训练集中，应用Lasso回归和二元Logistic回归分析构建术后大量腹水的风险预测模型；使用Bootstrap法对训练集进行1 000次重复抽样进行内部验证；使用验证集数据进行外部验证。采用ROC曲线下面积（AUC）和Calibration校准曲线评估预测模型的预测性能；临床决策曲线分析（DCA）评价预测模型的临床应用价值。

训练集行Lasso回归和Logistic回归分析显示，肝硬化、术后ALB、术中出血量和ALP为术后大量腹水的独立危险因素（OR=3.107，2.321，2.472，2.810；P<0.05）。基于4个独立的危险因素构建预测列线图模型。预测模型总分的最佳截断值为185.5分，总分≥185.5的患者为高危组，总分<185.5的患者为低危组。训练集和验证集中预测模型的AUC分别为0.759（95%CI：0.716~0.802）和0.805（95%CI：0.743~0.867）。Calibration校准曲线显示，肝切除术后大量腹水的预测风险与预测模型估计的实际风险之间具有良好的一致性。DCA表明，预测模型在预测患者发生大量腹水的风险方面具有临床价值，可使患者受益。

基于肝硬化、术后ALB、术中出血量和ALP独立危险因素构建的肝癌患者肝切除术后大量腹水的预测模型具有良好的预测能力和临床适用性。

To investigate the risk factors of massive ascites after liver resection for primary liver cancer (PLC) and construct a nomogram prediction model.

Clinical data of 739 PLC patients admitted to the Affiliated Hospital of North Sichuan Medical College from January 2018 to December 2022 were retrospectively analyzed. Among them, 585 patients were male and 154 female, aged from 21 to 89 years, with a median age of 60 years. According to the operation time and surgical department, all patients were divided into the training set (n=536) and validation set (n=203). The data in the training set were used for model construction, model evaluation and internal validation, and those in the validation set data were utilized for external validation. Clinical data of the enrolled patients were obtained by querying the scientific research data platform and special disease database of the hospital. In the training set, Lasso regression and binary Logistic regression analyses were used to construct the risk prediction model for postoperative massive ascites. Bootstrap method was adopted to conduct 1 000 repeated sampling in the training set for internal validation, and the data in the validation set were used for external validation. The area under the ROC curve (AUC) and calibration curve were employed to evaluate the prediction performance of this prediction model. Decision curve analysis (DCA) was utilized to evaluate clinical application value of the prediction model.

Lasso regression and Logistic regression analyses showed that liver cirrhosis, postoperative ALB, intraoperative blood loss and ALP were the risk factors for postoperative massive ascites in the training set (OR=3.107, 2.321, 2.472, 2.810; all P<0.05). Based on these four independent risk factors, a nomogram prediction model was constructed. The optimal cut-off value of the total score of the prediction model was 185.5. Patients with a total score of ≥185.5 were assigned into the high-risk group, and those with a total score <185.5 were allocated into the low-risk group. The AUC of the prediction model in the training and validation sets was 0.759 (95%CI: 0.716-0.802) and 0.805 (95%CI: 0.743-0.867), respectively. Calibration curve identified high consistency between the predicted risk and the actual risk estimated by the prediction model for of massive ascites after liver resection. DCA demonstrated that the prediction model had clinical value in predicting the risk of massive ascites, and its application could bring clinical benefits to the patients.

Prediction model for massive ascites after liver resection based on the independent risk factors including liver cirrhosis, postoperative ALB, intraoperative blood loss and ALP, in patients with PLC has high predictive ability and clinical applicability.