Uher Rudolf
MRC Social, Genetic and Developmental Psychiatry Research Centre, Institute of Psychiatry, King's College London, UK.
Novartis Found Symp. 2008;293:13-26; discussion 26-30, 68-70. doi: 10.1002/9780470696781.ch2.
While interacting biological effects of genes and environmental exposures (G x E) form a natural part of the causal framework underlying disorders of human health, the detection of G x E relies on inference from statistical interactions observed at population level. The validity of such inference has been questioned because the presence or absence of statistical interaction depends on measurement scale and statistical model. Furthermore, the feasibility of G x E research is threatened by the fact that tests of statistical interaction require large samples and their power is substantially reduced by unreliability in the assessments of genes, environmental exposures and pathology. It is demonstrated that concerns about statistical models and scaling can be addressed by integration of observational and experimental data. Judicious selection of genes and environmental factors should limit multiple testing. To overcome the challenge of low statistical power, it is suggested to maximize the reliability of measurement, integrate prior knowledge under Bayesian framework and facilitate pooling of data across studies by use of standardized stratified reporting. Consistencies and discrepancies among studies can be exploited for methodological analysis and model specification.
虽然基因与环境暴露之间的交互生物学效应（基因×环境，G×E）构成了人类健康疾病潜在因果框架的自然组成部分，但G×E的检测依赖于从人群水平观察到的统计交互作用进行推断。这种推断的有效性受到质疑，因为统计交互作用的存在与否取决于测量尺度和统计模型。此外，G×E研究的可行性受到以下事实的威胁：统计交互作用检验需要大样本，并且基因、环境暴露和病理学评估中的不可靠性会大幅降低其检验效能。研究表明，通过整合观察性数据和实验性数据，可以解决对统计模型和尺度的担忧。明智地选择基因和环境因素应能限制多重检验。为克服统计效能低的挑战，建议最大化测量的可靠性，在贝叶斯框架下整合先验知识，并通过使用标准化分层报告促进跨研究的数据合并。研究之间的一致性和差异可用于方法学分析和模型设定。
