Weinberg C R, Umbach D M
Biostatistics Branch, National Institute of Environmental Health Sciences, Research Triangle Park, NC 27709, USA.
Am J Hum Genet. 2005 Oct;77(4):627-36. doi: 10.1086/496900. Epub 2005 Aug 31.
Studies of genetic contributions to risk can be family-based, such as the case-parents design, or population-based, such as the case-control design. Both provide powerful inference regarding associations between genetic variants and risks, but both have limitations. The case-control design requires identifying and recruiting appropriate controls, but it has the advantage that nongenetic risk factors like exposures can be assessed. For a condition with an onset early in life, such as a birth defect, one should also genotype the mothers of cases and the mothers of controls to avoid potential confounding due to maternally mediated genetic effects acting on the fetus during gestation. The case-parents approach is less vulnerable than the case-mother/control-mother approach to biases due to population structure and self-selection. The case-parents approach also allows access to epigenetic phenomena like imprinting, but it cannot evaluate the role of nongenetic cofactors like exposures. We propose a hybrid design based on augmenting a set of affected individuals and their parents with a set of unaffected, unrelated individuals and their parents. The affected individuals and their parents are all genotyped, whereas only the parents of unaffected individuals are genotyped, although exposures are ascertained for both affected and unaffected offspring. The proposed hybrid design, through log-linear, likelihood-based analysis, allows estimation of the relative risk parameters, can provide more power than either the case-parents approach or the case-mother/control-mother approach, permits straightforward likelihood-ratio tests for bias due to mating asymmetry or population stratification, and admits valid alternative analyses when mating is asymmetric or when population stratification is detected.
对风险的遗传贡献的研究可以基于家庭，如病例-父母设计，也可以基于人群，如病例-对照设计。两者都能对基因变异与风险之间的关联提供有力推断，但都有局限性。病例-对照设计需要识别和招募合适的对照，但其优点是可以评估暴露等非遗传风险因素。对于出生缺陷等在生命早期发病的疾病，还应对病例的母亲和对照的母亲进行基因分型，以避免由于孕期母体介导的遗传效应作用于胎儿而导致的潜在混杂。病例-父母方法比病例-母亲/对照-母亲方法更不易受到人群结构和自我选择导致的偏差影响。病例-父母方法还能研究印记等表观遗传现象，但无法评估暴露等非遗传辅助因素的作用。我们提出一种混合设计，即在一组患病个体及其父母中加入一组未患病、无亲缘关系的个体及其父母。对患病个体及其父母都进行基因分型，而仅对未患病个体的父母进行基因分型，不过要确定患病和未患病后代的暴露情况。通过基于对数线性似然分析的拟混合设计，可以估计相对风险参数，比病例-父母方法或病例-母亲/对照-母亲方法具有更强的效力，允许对由于交配不对称或人群分层导致的偏差进行直接的似然比检验，并且在交配不对称或检测到人群分层时允许进行有效的替代分析。
