Bias from conditioning on live-births in pregnancy cohorts: an illustration based on neurodevelopment in children after prenatal exposure to organic pollutants (Liew et al. 2015).
Bias from conditioning on live-births in pregnancy cohorts: an illustration based on neurodevelopment in children after prenatal exposure to organic pollutants (Liew et al. 2015).
复制标题
妊娠队列中活产条件反射产生的偏差：基于产前接触有机污染物后儿童神经发育的说明（Liew et al. 2015）。
DOI：
10.1093/ije/dyv139
复制
发表时间：
2015
影响因子：
7.7
通讯作者：
Parker,SamanthaE
中科院分区：
文献类型：
--
作者：
Werler,MarthaM;Parker,SamanthaE
Liew等人询问，是否观察到产前全氟烷基物质（PFAS）暴露与注意缺陷多动障碍（ADHD）之间的负相关关系可能是由于研究中只包括活产婴儿或所谓的“活产偏见”造成的。在测量PFAS暴露与ADHD之间的关系时，他们提供了一个非常清晰的偏倚结构描述，该偏倚结构可能是由于产前PFAS暴露对胎儿死亡有影响而导致的。他们在三种不同的情况下模拟了PFAS与ADHD风险之间的关系，这些情况基于对暴露、胎儿丢失和ADHD患病率的合理假设，以及结果与已知和未知共变量之间的关联范围。当作者将PFAS与adhd的“真实”关联设定为1.0时，他们确实表明，当PFAS使胎儿死亡风险加倍时，在存在PFAS胎儿死亡关联的强烈混淆时，这种关联是保护性的。当真实关联设置为1.2时，在相同条件下的风险估计接近零，作为向下偏差的进一步证据。换句话说，在某些情况下，作者的研究问题的答案是肯定的，即是否观察到产前PFAS暴露与ADHD之间的负相关可能是由于胎儿死亡的竞争风险而导致的活产偏见。从发病率、治疗费用以及对学习、工作和人际关系的影响来看，多动症是一个重要的结果。因此，确定因果因素是值得的。如果环境污染物导致多动症，我们应该知道并消除暴露的可能性。然而，Liew等人的研究问题似乎放错了地方。在这种情况下，调整掉胎儿死亡的竞争风险来衡量PFAS-ADHD的关联是没有意义的。从理论上讲，当竞争风险将个体从风险人群中移除时，就需要关注竞争风险。ADHD的诊断是基于儿童的行为，因此有风险的人群是儿童，而不是所有的孕妇。从实际的角度来看，在考虑了暴露与胎儿死亡之间的联系之后，暴露与ADHD之间的联系是不可翻译的。在解释PFAS与ADHD风险之间的关联时，PFAS暴露与胎儿死亡之间的关系是无关的。孕妇可能想知道她的胎儿是否患有多动症，但如果她失去了这个胎儿，这个问题就没有意义了。这个问题只对幸存者有意义。可解释的问题是，产前PFAS暴露是否会影响活产婴儿（存活到幼儿期）患多动症的风险。这个可翻译的问题的答案可以为公共卫生干预提供信息。作为奖励，这个答案也将有助于理解神经发育。这两种信息都是相关的、重要的和有效的，即使接触会导致胎儿死亡。观察到的环境毒素与儿童结局之间的保护性关联可能由于活产偏见而存在偏差，这一知识可能被认为是有益的，因为它应该先发制人，防止虚假的健康声明，即接触有毒物质有利于儿童结局。但是，对活产偏见的同样了解意味着，由于它会导致上游后果，暴露在这种情况下是无益的，实际上是危险的。同样，似乎没有什么可以从中获益
Liew et al. 1 ask whether observed inverse associations between prenatal perfluoroalkyl substances (PFAS) exposure and attention deficit hyperactivity disorder (ADHD) could result from including only live-births in a study or a socalled ‘live-birth bias’. They provide a very clear description of a bias structure that could result from prenatal PFAS exposure having an effect on fetal death, when measuring the association between PFAS exposure and ADHD. They conducted simulations of the relationship between PFAS and risk of ADHD in three different scenarios based on reasonable assumptions of the prevalences of exposure, fetal loss and ADHD, and ranges of associations between outcomes and known and unknown covariates. When the authors set the ‘true’PFAS-ADHD association to 1.0, they indeed show the association is protective when PFAS doubles the risk of fetal death in the presence of strong confounding for the PFAS fetal death association. When the true association was set to 1.2, risk estimates were close to the null under the same conditions, as further evidence of the downward bias. In other words, the answer to the authors’ research question—whether observed inverse associations between prenatal PFAS exposures and ADHD could result from live-birth bias due to the competing risk of fetal death—is yes, in some scenarios. ADHD is an important outcome in terms of its growing incidence, cost of treatments and impact on learning, working and relationships. Thus, identifying causal factors is worthwhile. If environmental pollutants cause ADHD, we ought to know and eliminate the potential for exposure. However, Liew et al.’s research question appears to be misplaced. Adjusting away the competing risk on fetal death to measure the PFAS-ADHD association makes little sense in this situation. On theoretical grounds, competing risks are a concern when they remove individuals from the population at risk. 2 Diagnosis of ADHD is based on child behaviorus and therefore the population at risk is children, not all conceptions. From a practical standpoint, an association between an exposure and ADHD after accounting for the association between that exposure and fetal death is not translatable. The relationship between PFAS exposure and fetal death is irrelevant when interpreting an association between PFAS and risk of ADHD. A pregnant woman may wonder if her fetus will have ADHD, but if she loses that fetus, the question is moot. The question only makes sense among survivors. The translatable question is whether prenatal PFAS exposure affects the risk of ADHD among live-births (who survive into early childhood). The answer to this translatable question could inform public health interventions. As a bonus, the answer would also contribute to the understanding of neurodevelopment. Both types of information are relevant, important and valid, even if the exposure causes fetal death. Knowledge that observed protective associations for an environmental toxin and childhood outcomes could be biased due to live-birth bias might be considered helpful in that it should preempt false health claims that the toxic exposures are beneficial for childhood outcomes. But that same knowledge of a live-birth bias means that the exposure is not beneficial and indeed hazardous due to it causing upstream outcomes. Again, there seems little to be gained from