文献速递肝胆胰外科围术期管理与 ERAS

微创胰十二指肠切除后临床相关胰瘘患者术后出血的预测模型

A predictive model for postoperative hemorrhage in patients with clinically relevant pancreatic fistula following minimally invasive pancreaticoduodenectomy.
PubMed 文献 · Surg Endosc · 2026-03-02

本研究基于五个高容量胰腺中心2010至2024年的回顾性多中心病例对照数据,纳入微创胰十二指肠切除(MIPD)后发生临床相关术后胰瘘(CR-POPF)的患者,构建并外部验证了一个术前四变量预测模型,用于预测迟发性术后胰腺切除术后出血(PPH)。模型在推导队列(n=304)和验证队列(n=187)中均表现出优秀的区分度(AUC分别为0.952和0.925)和良好的校准度。

要点

  • MIPD后CR-POPF患者中迟发性PPH发生率在推导队列为25.6%,验证队列为21.4%。
  • 四个独立预测因子:肿瘤侵犯动脉、术前新辅助化疗、Naples预后评分>2、肿瘤大小>3 cm。
  • 模型AUC推导队列0.952、验证队列0.925,Brier评分0.067,校准良好。

核心结论

  • 四变量术前LASSO-logistic模型可可靠预测MIPD后CR-POPF患者的迟发性PPH。

研究设计与样本

  • 回顾性多中心病例对照研究,纳入五个高容量胰腺中心2010至2024年接受MIPD并发生CR-POPF的连续患者。
  • 推导队列n=304,外部验证队列n=187。
  • 采用LASSO筛选变量并进入多变量logistic回归,通过ROC、校准分析、Brier评分、Hosmer-Lemeshow检验和决策曲线分析评估模型性能。

主要发现

  • 迟发性PPH在推导队列发生率为25.6%(78/304),验证队列为21.4%(40/187)。
  • 肿瘤侵犯动脉OR 10.32(95%CI 2.67-39.93,P<0.01)。
  • 术前新辅助化疗OR 8.28(95%CI 2.41-28.44,P<0.01)。
  • Naples预后评分>2 OR 17.88(95%CI 6.74-47.42,P<0.01)。
  • 肿瘤大小>3 cm OR 9.19(95%CI 3.64-23.15,P<0.01)。

关键数值与推荐

  • 模型区分度AUC推导队列0.952,验证队列0.925,校准良好,Brier评分0.067。
  • Youden指数得出的截断值0.192,在验证队列中区分低风险与高风险组,迟发性PPH发生率分别为5.0%与68.8%。

临床怎么用

  • 利用列线图和个体化风险公式进行床旁应用,早期识别高危患者以制定针对性围术期策略。

以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。

原文

BACKGROUND: Postpancreatectomy hemorrhage (PPH) remains the most lethal complication following pancreaticoduodenectomy. Patients who develop clinically relevant postoperative pancreatic fistula (CR-POPF) are at particularly high risk. A preoperative model tailored to CR-POPF patients is needed to enable early risk stratification and targeted perioperative strategies. METHODS: We conducted a retrospective multicenter case-control study across five high-volume pancreatic centers from 2010 to 2024. Consecutive patients who underwent minimally invasive pancreaticoduodenectomy (MIPD) and subsequently developed CR-POPF were included and split into a derivation (n = 304) and an external validation cohort (n = 187). Variables were screened using least absolute shrinkage and selection operator (LASSO) and entered multivariable logistic regression. Model performance was assessed by receiver operating characteristic analysis, calibration analysis, Brier score, Hosmer-Lemeshow test and decision-curve analysis. A nomogram and individualized risk formula were constructed to facilitate bedside application. RESULTS: Late PPH occurred in 25.6% (78/304) of the derivation cohort and 21.4% (40/187) of the validation cohort. Four independent indicators were identified: Tumor abutment of artery (odds ratio [OR] 10.32; 95% confidence interval [CI] 2.67-39.93; P < 0.01), receipt of preoperative neoadjuvant chemotherapy (OR 8.28; 95% CI 2.41-28.44; P < 0.01), Naples prognostic score > 2 (OR 17.88; 95% CI 6.74-47.42; P < 0.01), and Tumor size > 3 cm (OR 9.19; 95% CI 3.64-23.15; P < 0.01). Discrimination was excellent (AUC 0.952 derivation; 0.925 validation) with good calibration and low prediction error (Brier 0.067). A Youden-derived cutoff (0.192) separated risk groups with clear incidence gradients in validation (low vs high risk: 5.0% vs 68.8%). CONCLUSION: This four-variable, preoperative LASSO-logistic model offers a reliable tool for predicting late PPH among CR-POPF patients after MIPD. Early identification of high-risk individuals may enable targeted perioperative strategies and improved outcomes.

来源:Surg Endosc,查看原文。版权归原作者所有。

引用

  1. Pan Y, Wu D, Lai J, et al. A predictive model for postoperative hemorrhage in patients with clinically relevant pancreatic fistula following minimally invasive pancreaticoduodenectomy. Surg Endosc. 2026. PMID: 41772225. doi:10.1007/s00464-026-12679-8

供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。