预测肺癌筛查不依从性的机会:系统综述与Meta分析
该系统综述与Meta分析检索了2014年4月28日至2025年5月8日间PubMed、Embase和Web of Science中涉及肺癌筛查不依从风险评估的原始研究,纳入2020至2025年发表的9项研究(样本量168至28294)。其中4项研究报告了预测模型性能,合并交叉验证或内部测试的受试者工作特征曲线下面积为0.80(95%CI 0.64-0.90),研究间异质性大(P<.05,Cochran Q检验;I²=98.7%)。作者结论认为可预测个体肺癌筛查不依从风险的机器学习模型尚不成熟,建议放射科医生等利益相关者投资建设大型多中心国家数据库以促进模型开发与验证。
要点
- 4项研究报告预测模型性能,合并交叉验证或内部测试AUC为0.80(95%CI 0.64-0.90),研究间异质性大(I²=98.7%)。
- 可预测个体肺癌筛查不依从风险的机器学习模型尚不成熟,需大型多中心数据库支持开发与验证。
- 作者建议放射科医生等利益相关者投资整理大型多中心国家数据库以识别最需定制干预的患者。
核心结论
- 可预测个体肺癌筛查不依从风险的机器学习模型尚不成熟,放射科医生等利益相关者应投资建设大型多中心国家数据库以促进模型开发与验证。
研究设计与样本
- 系统综述与Meta分析,检索PubMed、Embase和Web of Science中2014年4月28日至2025年5月8日发表的原始研究,方案注册于PROSPERO(CRD420251049715)。
- 纳入9项发表于2020至2025年的研究,样本量168至28294,来自不同机构设置。
- 其中4项研究报告预测模型性能,合并交叉验证或内部测试AUC覆盖3个不同研究人群,样本量168至2430。
诊断标准与关键数值
- 4项研究报告预测模型性能,合并交叉验证或内部测试受试者工作特征曲线下面积为0.80(95%CI 0.64-0.90),覆盖3个不同研究人群,样本量168至2430。
- 研究间异质性大(P<.05,Cochran Q检验;I²=98.7%)。
临床应用与阅片要点
- 作者建议放射科医生和其他利益相关者投资整理大型多中心国家数据库,以促进识别最需定制干预患者的模型开发与验证。
- 作者探索了利用4个国家数据库支持未来模型开发与验证以改善肺癌筛查依从性的可行性。
局限性
- 仅4项研究报告预测模型性能,合并AUC覆盖3个不同研究人群且样本量相对较小(168至2430)。
- 研究间异质性大(P<.05,Cochran Q检验;I²=98.7%)。
- 可预测个体肺癌筛查不依从风险的机器学习模型尚不成熟。
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: The low adherence rate to annual lung cancer screening (LCS) reduces potential mortality benefits. Interventions to improve adherence often do not consider individualized risk for nonadherence, which is vital in informing the design of tailored interventions. The authors systematically reviewed the literature and conducted a meta-analysis on predicting LCS nonadherence risk using machine learning.
METHODS: The authors searched citation databases such as PubMed, Embase, and Web of Science for original studies that mentioned LCS nonadherence risk assessment between April 28, 2014, and May 8, 2025. Study characteristics, nonadherence information, and prediction model performance were extracted. The review protocol was registered with the International Prospective Register of Systematic Reviews (PROSPERO CRD420251049715).
RESULTS: Nine studies published between 2020 and 2025 were included in this systematic review, with sample sizes varying between 168 and 28,294, from various institutional settings. Four of the nine included studies reported prediction model performance, with a pooled cross-validated or internal test area under the receiver operating characteristics curve of 0.80 (95% confidence interval, 0.64-0.90) across three distinct study populations and a relatively small sample size between 168 and 2,430, with large heterogeneity detected across studies (P < .05, Cochran's Q test; I2 = 98.7%). The authors also explored the feasibility of leveraging four national databases to support future model development and validation efforts aimed at improving LCS adherence.
CONCLUSIONS: Machine learning models that can predict the individual risk for LCS nonadherence are underdeveloped. Radiologists and other stakeholders should invest in the curation of large, multicenter, national databases to facilitate the development and validation of models that identify patients who are in the greatest need of tailored interventions to improve adherence and enhance patient outcomes.
引用
- Lin Y, Ding R, Moghanaki D, et al. Opportunities for Predicting Lung Cancer Screening Nonadherence: A Systematic Review and Meta-Analysis. J Am Coll Radiol. 2025. PMID: 41338702. doi:10.1016/j.jacr.2025.08.053
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。