Ioannidis John P A
Department of Hygiene and Epidemiology, University of Ioannina School of Medicine, Ioannina 45110, Greece.
Eur J Epidemiol. 2005;20(9):739-45. doi: 10.1007/s10654-005-2028-1.
Bias is ubiquitous in research. The advent of the molecular era provides a unique opportunity to study the consequences of bias with large-scale empirical evidence accumulated in the massive data produced by the current discovery-oriented scientific effort, rather than just with theoretical speculations and constructs. Here I discuss some empirical evidence about manifestations of bias in molecular epidemiology. Bias may manifest as either heterogeneity or as deviation from the true estimates. The failure to translate molecular knowledge and the failure to replicate information are some typical hallmarks of bias at action. The acquired knowledge about the behaviour and manifestations of bias in molecular fields can be transferred back also to more traditional fields of epidemiology and medical research. Getting rid of false claims of the past is at least as important as producing new scientific discoveries. In many fields, the observed effects sizes that circulate as established knowledge are practically estimating only the net bias that has operated in the field all along. Issues of plausibility (in particular biological plausibility), replication, and credibility that form the theoretical basis of epidemiology and etiological inference can now be approached with large-scale empirical data.
偏差在研究中无处不在。分子时代的到来提供了一个独特的机会，可利用当前以发现为导向的科学研究产生的海量数据中积累的大规模实证证据来研究偏差的后果，而不仅仅是通过理论推测和构建。在此，我将讨论一些关于分子流行病学中偏差表现的实证证据。偏差可能表现为异质性或偏离真实估计值。未能转化分子知识以及未能复制信息是偏差起作用的一些典型特征。在分子领域获得的关于偏差行为和表现的知识也可以反馈到流行病学和医学研究等更传统的领域。摒弃过去的错误论断与产生新的科学发现至少同等重要。在许多领域，作为既定知识流传的观察到的效应大小实际上仅仅估计了该领域一直存在的净偏差。构成流行病学和病因推断理论基础的合理性（尤其是生物学合理性）、可重复性和可信度等问题现在可以通过大规模实证数据来探讨。
