CEUS模型术前识别局限性透明细胞肾癌不良病理
本研究回顾性纳入370例cT1-T2透明细胞肾细胞癌患者,开发并外部验证基于超声造影、临床及实验室特征的预测模型,用于术前识别WHO/ISUP 3-4级、坏死、侵犯或淋巴结转移等不良病理。模型在训练集、内部测试集和外部验证集的AUC分别为0.810、0.875和0.799。
要点
- 模型对排除不良病理有一定价值,外部验证阴性结果中仅6.7%存在不良病理。
- 阈值0.3051下外部验证阳性预测值61.5%,低于训练集表现,提示个体风险估计存在压缩。
- CEUS征象中有4个被保留为独立预测因子,阅片时需关注这些特征。
核心结论
- 模型在外部验证中显示中等鉴别能力,可用于术前识别较低风险患者。
- 外部预测存在压缩现象,校准斜率为2.380,个体风险估计需前瞻性多中心研究验证。
研究设计与样本
- 回顾性多中心研究,共纳入370例经手术治疗的cT1-T2透明细胞肾细胞癌患者。
- 中心1患者随机分为训练集(210例)和内部测试集(91例),中心2提供外部验证集(69例)。
- 不良病理定义为WHO/ISUP 3-4级、肿瘤坏死、侵犯或淋巴结转移,共138例患者存在不良病理。
- 两位设盲阅片者评估影像,41个术前候选变量进入logistic LASSO回归,采用分层十折交叉验证和1个标准误规则。
- 内部验证使用200次bootstrap重采样完整建模流程。
关键影像表现/征象
- 最终模型保留了7个预测因子,其中4个为CEUS影像特征。
诊断标准与关键数值
- 训练集AUC为0.810(95%CI,0.752-0.868),内部测试集AUC为0.875(95%CI,0.805-0.944)。
- 外部验证集AUC为0.799(95%CI,0.688-0.910),乐观校正后的训练AUC为0.747。
- 校准斜率为2.380,提示外部预测值存在压缩。
- 在训练集推导的阈值0.3051下,外部验证中阳性结果患者不良病理检出率为61.5%(24/39)。
- 同一阈值下外部验证中阴性结果患者不良病理检出率为6.7%(2/30),总体不良病理比例为37.7%。
临床应用与阅片要点
- 该模型可作为术前评估辅助工具,尤其有助于识别不良病理风险较低、可能适合保守管理的患者。
- 外部验证中阴性结果仅6.7%存在不良病理,对临床排除高风险病变有参考价值。
- 外部验证阳性预测值低于训练集,临床应用时需谨慎解读个体风险估计。
局限性
- 外部验证集样本量较小(n=69),可能影响估计的稳定性。
- 外部预测存在明显的校准压缩,提示模型在不同中心间外推时存在系统性偏差。
- 本研究为回顾性设计,存在选择偏倚和混杂因素影响的可能性。
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 model combining contrast-enhanced ultrasound (CEUS), clinical, and laboratory features to identify adverse pathology in clinically localized clear cell renal cell carcinoma (ccRCC).
METHODS: This retrospective study included 370 surgically treated patients with cT1-T2 ccRCC. Center 1 patients were randomly split into training (n = 210) and internal test (n = 91) sets; Center 2 provided external validation (n = 69). Adverse pathology comprised WHO/ISUP grade 3-4, tumor necrosis, invasion, or nodal metastasis. Two blinded readers assessed images. All 41 preoperative candidates entered logistic least absolute shrinkage and selection operator regression with stratified ten-fold cross-validation and the one-standard-error rule. Internal validation used 200 bootstrap resamples of the complete modeling procedure.
RESULTS: Adverse pathology was present in 138 patients. Seven predictors were retained, including four CEUS features. Areas under the receiver operating characteristic curve (AUCs) were 0.810 (95% CI, 0.752-0.868), 0.875 (0.805-0.944), and 0.799 (0.688-0.910) in the training, internal test, and external cohorts, respectively; optimism-corrected training AUC was 0.747. External predictions were compressed (calibration slope, 2.380). At the training-derived threshold of 0.3051, adverse pathology was present in 61.5% (24/39) of external patients with positive results and 6.7% (2/30) with negative results, compared with 37.7% overall.
CONCLUSION: The model showed moderate external discrimination and may support preoperative assessment by identifying patients at lower risk. Prospective multicenter studies are needed to validate individual risk estimates and assess clinical impact.
引用
- Ma, Du, Li, et al. Preoperative identification of adverse pathology in clinically localized clear cell renal cell carcinoma: development and external validation of a contrast-enhanced ultrasound-based model. Abdom Radiol (NY). 2026. PMID: 42747478. doi:10.1007/s00261-026-05795-5
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。