文献速递流行病学与生物统计基础与转化研究

克隆-删失-加权法避免不死时间偏倚的系统方法学综述

Use of clone-censor-weight to avoid immortal-time bias: a systematic methodological review.
PubMed 文献 · J Clin Epidemiol · 2026-08-15

该研究系统检索2010至2025年MEDLINE和Embase,纳入113项使用克隆-删失-加权(CCW)法的观察性研究,评估其实现方式及不死时间偏倚处理。发现CCW实施存在异质性,22项研究(19.5%)仍存在潜在不死时间偏倚。

要点

  • CCW主要用于目标试验模拟框架,但方法学实施和报告不一致。
  • 22项研究(19.5%)因基线后标准或时间零点定义不清仍存在不死时间偏倚。
  • 需统一CCW关键方法学组件并提高报告透明度。

核心结论

  • CCW实施在不同研究间存在差异,尤其在权重统计建模、删失事件考虑及干预效果建模方面。
  • 明确关键方法学组件可支持CCW的一致应用并改善报告。

研究设计与样本

  • 系统方法学综述,检索MEDLINE和Embase从2010年1月1日至2025年3月18日,并手工检索9篇关于TTE或CCW的综述。
  • 由两名独立评价者筛选,对10%研究进行数据提取和偏倚风险评估,因评价者间一致性高,剩余样本由单名评价者评估。
  • 共纳入113项研究。

主要发现

  • 113项研究中103项(91.2%)在目标试验模拟框架内声明实施CCW。
  • 55项研究(48.7%)称该方法为「克隆删失加权」,25项(22.1%)称「CCW」,其余用其他表述。
  • 所有研究均使用逆删失概率加权或逆治疗概率加权来仅处理人为删失(97/113,85.8%),较少联合处理失访(16/113,14.2%)。
  • 权重估计使用合并逻辑回归(n=70,61.9%)或Cox模型(n=26,23%),结局多使用合并逻辑回归(n=58,51.3%)或非参数估计(n=40,35.4%)。
  • 22项研究(19.5%)仍存在潜在不死时间偏倚:21项(18.6%)在实施CCW前基于时间零点后标准排除患者,2项(1.8%)时间零点未明确定义。

临床怎么用

  • 在目标试验模拟中应用CCW时,应明确时间零点定义并避免基于基线后特征排除患者。
  • 需规范报告删失事件、权重模型及结局模型等关键方法学细节。

局限性

  • 仅对10%研究进行双人数据提取和偏倚风险评估,其余由单名评价者完成。
  • 检索限于MEDLINE和Embase,可能遗漏其他数据库的研究。

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

BACKGROUND AND OBJECTIVES: Observational studies evaluating treatment strategies that are indistinguishable at the start of follow-up are prone to immortal-time bias in its misclassification form. The clone-censor-weight (CCW) approach has gained attention for its effectiveness in mitigating immortal-time bias while enhancing the validity of causal inference in observational research within the target trial emulation (TTE) framework. In this methodological systematic review, we assess CCW implementation in observational analyses, evaluate the handling of immortal-time bias, and identify key considerations and potential limitations in CCW application. METHODS: We conducted a comprehensive search in MEDLINE and Embase databases for all studies published from January 1, 2010, to March 18, 2025. We performed an additional manual search by citation searching on nine reviews about the TTE or the CCW method. Screening was conducted by two independent reviewers. Data extraction and evaluation of risk of bias were performed by two reviewers for 10%, and due to high inter-reviewer agreement, a single reviewer evaluated the remaining study sample. Risk of immortal-time bias due to post-baseline eligibility was assessed. RESULTS: Among 113 studies identified, 103 declared implementing the CCW approach within the TTE framework (91.2%). The method was referred as "cloning censoring and weighting" in 55 studies (48.7%), "CCW" in 25 studies (22.1%), or other ways. All studies used either inverse probability of censoring weighting or inverse probability of treatment weighting to account for artificial censoring alone (97/113, 85.8%) and less commonly in combination with loss to follow-up (16/113, 14.2%). Authors used pooled logistic regressions or Cox models for weight estimations (n = 70, 61.9% and n = 26, 23% of studies, respectively), while outcomes were mostly estimated using pooled logistic regressions (n = 58, 51.3%) or nonparametric estimators (n = 40, 35.4%). Potential immortal-time bias was still present in 22 studies (19.5%) where they excluded patients based on post-time zero criteria before implementing CCW (n = 21, 18.6%), or when time zero was not well defined (n = 2, 1.8%). CONCLUSION: CCW implementation varies across studies, particularly in statistical modeling of weights, and considered censoring events, as well as statistical modeling effect of interventions on outcomes. Clarifying key methodological components could support consistent application of CCW and improve reporting. PLAIN LANGUAGE SUMMARY: Researchers increasingly use observational health-care data, such as electronic health records and insurance claims, to evaluate the effects of medical treatments. Unlike randomized clinical trials, these studies can be affected by immortal-time bias if both intervention assignment and eligibility assessment are not set at the start of follow-up. This bias typically occurs when patients must survive long enough to receive a treatment, such as surgery. Patients who die before receiving the surgery are automatically classified as untreated, leading to an unfair comparison. The clone-censor-weight (CCW) approach was developed to reduce this bias in observational studies, with interventions that cannot be distinguished at the start of follow-up. In this paper, we reviewed 113 published studies using this approach to evaluate how researchers implemented the method, whether immortal-time bias was appropriately addressed, and how clearly key methodological details were reported. We found variation in the implementation of the CCW approach, particularly in censoring and weighting processes. Important methodological details were also often incompletely reported, making it difficult to compare studies or reproduce their analyses. Despite the use of CCW approach, 20% of the evaluated studies remained at risk of immortal-time bias, due to eligibility assessment based on characteristics measured after the start of follow-up. Our findings highlight the need for clearer methodological guidance and more transparent reporting to improve the consistency and reliability of studies using the CCW approach.

来源:J Clin Epidemiol,查看原文。版权归原作者所有。

引用

  1. Manzanilla A, Brunetta C, Peyre-Pradat F, et al. Use of clone-censor-weight to avoid immortal-time bias: a systematic methodological review. J Clin Epidemiol. 2026. PMID: 42600711. doi:10.1016/j.jclinepi.2026.112468

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。