Sybil肺癌风险预测模型在以黑人为主的筛查队列中的验证
本研究为回顾性队列研究,纳入3177名参与者的5883次低剂量CT(LDCT),其中64.9%为黑人,评估Sybil深度学习模型预测1-6年肺癌风险的表现。结果显示模型在基线LDCT的AUC从1年0.91(95%CI 87.7-94.6)到6年0.82(95%CI 77.6-86.1),黑人与非黑人表现相当,支持其泛化性。
要点
- Sybil模型在黑人为主的筛查队列中表现出色,1年AUC为0.91,6年AUC为0.82。
- 黑人与非黑人个体的模型性能相当,支持跨种族泛化。
- 模型可能有助于超越当前筛查标准,改善风险导向筛查。
核心结论
- Sybil模型在该以黑人为主的筛查队列中展现出稳健的肺癌风险预测能力,支持其泛化性并突显其在多样化人群中改进风险导向筛查的潜力。
研究设计与样本
- 回顾性队列研究,纳入3177名参与者的5883次LDCT(2014-2025年),使用Sybil模型生成1-6年风险预测,以AUC评估性能。
主要发现
- 队列中黑人占64.9%,共诊断105例肺癌,其中61%在1年内确诊。
- 基线LDCT的AUC从1年0.91(95%CI 87.7-94.6)到6年0.82(95%CI 77.6-86.1)。
- 黑人与非黑人个体的模型性能相当。
临床怎么用
- Sybil模型可用于多样化人群的肺癌风险预测,辅助风险导向筛查决策。
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
INTRODUCTION: Lung cancer remains the leading cause of cancer-related mortality in the United States and worldwide. Black Americans have higher lung cancer incidence and mortality risk despite lower cumulative tobacco exposure. Sybil is an open-source deep learning model that predicts future lung cancer risk from a single LDCT. Sybil was trained using National Lung Screening Trial data in which more than 90% of participants were White; and Sybil's performance in diverse populations remains unclear. METHODS: We conducted a retrospective cohort study of 5,883 LDCTs from 3,177 participants (2014-2025). Sybil generated risk predictions for 1-6 years. Model performance was assessed using AUC. RESULTS: This cohort included 64.9% Black individuals. Among 105 lung cancers, 61% were diagnosed within 1 year. Sybil demonstrated strong discrimination for baseline LDCTs, with AUCs ranging from 0.91 at 1 year (95% CI, 87.7-94.6) to 0.82 at 6 years (95% CI, 77.6-86.1), with comparable performance between Black and non-Black individuals. CONCLUSIONS: Sybil demonstrated robust lung cancer risk prediction in this screening cohort, supporting its generalizability and highlighting its potential to improve risk-based screening beyond current eligibility criteria in diverse populations.
引用
- Pasquinelli MM, Durney CH, Hill T, et al. Brief Report: Validation of the Sybil Lung Cancer Risk Prediction Model in a Predominantly Black Screening Cohort. J Thorac Oncol. 2026. PMID: 42805462. doi:10.1016/j.jtho.2026.106128
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。