文献速递免疫规划卫生政策与健康促进

婴儿期疫苗接种研究中信息性删失引入的选择偏倚

Selection bias introduced by informative censoring in studies examining effects of vaccination in infancy.
PubMed 文献 · Int J Epidemiol · 2019-01-01

该研究通过模拟方法探讨在评估婴儿期疫苗(尤其DTP疫苗)非特异性效应时,右删失(麻疹疫苗接种时)与左删失(排除早期随访)如何引入选择偏倚。结果发现信息性右删失可偏倚向显示DTP有害效应,信息性左删失可偏倚向削弱DTP有益效应,调整混杂或逆概率加权可部分控制偏倚,但需测量预测删失的预后因素。

要点

  • 右删失(麻疹疫苗接种时)若为信息性,可偏倚估计,显示DTP有害效应。
  • 左删失(排除早期随访)若健康状态与DTP均与死亡相关,可削弱DTP有益效应。
  • 调整健康状态混杂或逆概率加权可减少偏倚,但需测量预测删失的预后因素。

核心结论

  • 疫苗非特异性效应估计可能受信息性右删失或左删失偏倚,研究者应评估偏倚可能并采用适当统计方法控制。

研究设计与样本

  • 模拟研究,考察右删失(麻疹疫苗接种时)和左删失(排除早期随访)在有无混杂的多种场景下,估计未调整和调整风险比(HR)的模拟平均值。

主要发现

  • 识别出右删失在麻疹疫苗接种时为信息性并引入偏倚向显示DTP疫苗有害效应的场景。
  • 调整健康状态混杂在部分但非全部场景中消除删失所致偏倚,一个场景需逆概率加权。
  • 左删失偏倚在健康状态和DTP接种均与删失早期随访期间死亡相关时出现,方向为削弱DTP对死亡的有益效应;当DTP效应随时间变化时偏倚更严重。

临床怎么用

  • 在估计疫苗非特异性效应时,应评估信息性删失可能,并采用逆概率加权等适当统计方法控制偏倚。
  • 控制删失偏倚需测量预测删失的预后因素。

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: Many studies have examined 'non-specific' vaccine effects on infant mortality: attention has been particularly drawn to diphtheria-tetanus-pertussis (DTP) vaccine, which has been proposed to be associated with an increased mortality risk. Both right and left censoring are common in such studies. METHODS: We conducted simulation studies examining right censoring (at measles vaccination) and left censoring (by excluding early follow-up) in a variety of scenarios in which confounding was and was not present. We estimated both unadjusted and adjusted hazard ratios (HRs), averaged across simulations. RESULTS: We identified scenarios in which right-censoring at measles vaccination was informative and so introduced bias in the direction of a detrimental effect of DTP vaccine. In some, but not all, situations, adjusting for confounding by health status removed the bias caused by censoring. However, such adjustment will not always remove bias due to informative censoring: inverse probability weighting was required in one scenario. Bias due to left censoring arose when both health status and DTP vaccination were associated with mortality during the censored early follow-up and was in the direction of attenuating a beneficial effect of DTP on mortality. Such bias was more severe when the effect of DTP changed over time. CONCLUSIONS: Estimates of non-specific effects of vaccines may be biased by informative right or left censoring. Authors of studies estimating such effects should consider the potential for such bias and use appropriate statistical approaches to control for it. Such approaches require measurement of prognostic factors that predict censoring.

来源:Int J Epidemiol,查看原文。版权归原作者所有。

引用

  1. López-López JA, Sterne JAC, Higgins JPT, et al. Selection bias introduced by informative censoring in studies examining effects of vaccination in infancy. Int J Epidemiol. 2019. PMID: 31071211. doi:10.1093/ije/dyz092

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。