基于游离DNA片段组学模型的胰腺癌早期检测开发与验证
本研究纳入1167名参与者,利用血浆游离DNA浅层全基因组测序和机器学习整合片段组学特征,构建胰腺导管腺癌早期检测模型。训练集AUC为0.992,验证集AUC为0.987;验证集敏感度97.3%、特异度92.8%,外部数据集敏感度90.91%、特异度94.5%。
要点
- 基于cfDNA片段组学与机器学习的模型在训练集区分PDAC与对照的AUC为0.992。
- 验证集敏感度97.3%、特异度92.8%,外部数据集敏感度90.91%、特异度94.5%。
- 模型在非恶性胰腺囊肿组中验证准确性,并纳入两个外部验证队列和一个独立早期数据集评估稳健性。
核心结论
- 本研究强调利用cfDNA片段组学和机器学习进行PDAC早期检测的有效性,作者认为该方法有望通过早期干预显著降低PDAC死亡率,并可能成为肿瘤诊断的突破。
研究设计与样本
- 研究纳入1167名参与者,采集血浆进行浅层全基因组测序,经严格质量评估后,训练队列含166例PDAC患者和167名健康对照,验证队列含112例PDAC患者和111名健康个体。
- 另纳入67例非恶性胰腺囊肿个体验证模型准确性,并纳入两个外部验证队列和一个独立早期数据集评估模型稳健性。
- 分析采用片段组学图谱,整合拷贝数变异、片段大小、突变特征和甲基化模式,并使用机器学习进行分析。
主要发现
- 模型在训练数据集区分PDAC患者与对照的AUC为0.992,在验证数据集为0.987。
- 在cutoff为0.52时,训练集敏感度为93.4%、特异度为95.2%。
- 验证数据集敏感度为97.3%、特异度为92.8%,外部数据集敏感度为90.91%、特异度为94.5%。
关键数值与推荐
- 模型cutoff为0.52时,训练集敏感度93.4%、特异度95.2%,验证集敏感度97.3%、特异度92.8%。
- 外部数据集敏感度90.91%、特异度94.5%。
- 训练集AUC为0.992,验证集AUC为0.987。
临床怎么用
- 原文未据此给出临床推荐。
局限性
- 原文未报告局限性。
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)。
原文
PURPOSE: Pancreatic ductal adenocarcinoma (PDAC), known for its high fatality rate, is often diagnosed in its advanced stages where surgical options are not viable. This highlights the critical need for innovative and effective early detection techniques. This study focuses on the potential of cell-free DNA (cfDNA) fragmentomics integrating advanced machine learning to identify early-stage PDAC with high accuracy. METHODS: Our study included a broad cohort of 1,167 participants, from which plasma was collected and subjected to shallow whole-genome sequencing. After rigorous quality assessments, 166 individuals diagnosed with PDAC and 167 healthy participants were in the training cohort, whereas the validation cohort consisted of 112 patients with PDAC and 111 healthy individuals. A separate group of 67 individuals with nonmalignant pancreatic cysts was also included to validate the model's accuracy. Finally, two additional external validation cohorts and one additional independent early-stage data set were included to evaluate the robustness of model. Our analysis used fragmentomic profiling, integrating copy-number variations, fragment size, mutational signatures, and methylation patterns analyzed using machine learning. RESULTS: The model demonstrated remarkable accuracy in distinguishing patients with PDAC from controls, with an AUC of 0.992 in the training data set and 0.987 in the validation data set. At a cutoff of 0.52, the training set reached a sensitivity of 93.4% and a specificity of 95.2%. In the validation data set, the sensitivity was 97.3% with a specificity of 92.8%, while the external data set demonstrated a sensitivity of 90.91% and a specificity of 94.5%. CONCLUSION: This study underscores the effectiveness of using cfDNA fragmentomics and machine learning for early detection of PDAC. Our approach promises significant potential in reducing PDAC mortalities through early intervention and could serve as a breakthrough in oncologic diagnostics.
引用
- Yin L, Cao C, Lin J, et al. Development and Validation of a Cell-Free DNA Fragmentomics-Based Model for Early Detection of Pancreatic Cancer. J Clin Oncol. 2025. PMID: 40311105. doi:10.1200/JCO.24.00287
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。