文献速递腹部CT

基于CT的临床影像模型术前鉴别胰腺腺鳞癌与导管腺癌的开发与外部验证

Development and external validation of a CT-based clinicoradiological model for preoperative differentiation of pancreatic adenosquamous carcinoma from ductal adenocarcinoma.
PubMed 文献 · Eur J Radiol · 2026-10-23

双中心回顾性研究纳入397例手术证实的胰腺腺鳞癌(PASC)或胰腺导管腺癌(PDAC)患者,开发并外部验证基于增强CT的临床影像模型。整合模型在开发队列AUC为0.919,外部验证AUC为0.846(95%CI 0.781-0.910),但校准差(斜率0.445)。模型外部区分度好但校准不佳,支持用于术前风险分层而非个体概率估计。

要点

  • CT临床影像整合模型外部验证AUC为0.846,区分PASC与PDAC能力良好。
  • 模型校准差(斜率0.445),预测概率不能直接用于个体风险估计。
  • 整合模型优于仅临床模型(ΔAUC 0.179,p<0.001),但未超越仅影像模型。

核心结论

  • 基于CT的临床影像模型在外部验证中保持良好区分度(AUC 0.846),但校准差(斜率0.445),适合风险分层而非个体概率预测。

研究设计与样本

  • 双中心回顾性研究,纳入397例手术证实的PASC或PDAC患者,均有术前增强CT。
  • 开发队列中PASC按年龄、性别和肿瘤位置1:2匹配PDAC患者。
  • 外部验证队列来自第二中心预设诊断特定研究期间连续纳入的PASC和PDAC患者。
  • 采用LASSO回归后多变量逻辑回归开发模型。

诊断标准与关键数值

  • 开发队列整合模型AUC为0.919。
  • 外部验证队列整合模型AUC为0.846(95%CI 0.781-0.910)。
  • 外部验证校准斜率0.445(95%CI 0.309-0.603),校准-in-the-large为0.152(95%CI -0.358至0.656),观察/期望比1.03。
  • 整合模型优于仅临床模型(ΔAUC 0.179,p<0.001),但与仅影像模型无差异(ΔAUC -0.004,p=0.85)。

临床应用与阅片要点

  • 可用于术前区分PASC与PDAC的风险分层,辅助临床决策。
  • 因校准差,模型预测概率不宜直接用于个体概率估计。
  • 需在无选择、患病率代表性的人群中重新校准和前瞻性验证后方可临床使用。

局限性

  • 回顾性研究设计。
  • 外部验证校准差(斜率0.445)。
  • 需进一步重新校准和前瞻性验证。

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)。

英文原文(摘要)· 点击展开

PURPOSE: To develop and externally validate a CT-based clinicoradiological model for preoperative differentiation of pancreatic adenosquamous carcinoma (PASC) from pancreatic ductal adenocarcinoma (PDAC).

MATERIALS AND METHODS: This dual-center retrospective study included 397 patients with surgically confirmed PASC or PDAC and preoperative contrast-enhanced CT. In the development cohort, patients with PASC were matched 1:2 to patients with PDAC by age, sex, and tumor location. At the second center, consecutive eligible patients with PASC and PDAC identified during prespecified diagnosis-specific study periods constituted the external validation cohort. Least absolute shrinkage and selection operator regression followed by multivariable logistic regression was used for model development.

RESULTS: In the development cohort, the integrated model achieved an AUC of 0.919. In the external cohort, discrimination was preserved (AUC, 0.846; 95% CI, 0.781-0.910), but calibration was poor (slope, 0.445; 95% CI, 0.309-0.603). Calibration-in-the-large was 0.152 (95% CI, -0.358 to 0.656), with an observed-to-expected ratio of 1.03. The integrated model outperformed the clinical-only model (ΔAUC, 0.179; p < 0.001), whereas incremental discrimination beyond the imaging-only model was not demonstrated (ΔAUC, -0.004; p = 0.85). Decision curve analysis in the development cohort suggested potential net benefit across selected threshold probabilities.

CONCLUSION: The CT-based clinicoradiological model showed good external discrimination but substantial miscalibration, supporting its potential for preoperative risk stratification rather than individual probability estimation. Further recalibration and prospective validation in unselected, prevalence-representative populations are required before clinical use.

来源:Eur J Radiol,查看原文。版权归原作者所有。

引用

  1. Li Y, Kuang T, Zins M, et al. Development and external validation of a CT-based clinicoradiological model for preoperative differentiation of pancreatic adenosquamous carcinoma from ductal adenocarcinoma. Eur J Radiol. 2026. PMID: 42933252. doi:10.1016/j.ejrad.2026.113329

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。