Pierce Brandon L, Kraft Peter, Zhang Chenan
Department of Public Health Sciences and Department of Human Genetics, University of Chicago, Chicago IL 60615.
Department of Epidemiology and Department of Biostatistics; T.H. Chan School of Public Health, Harvard University, Boston MA 02115.
Curr Epidemiol Rep. 2018 Jun;5(2):184-196. doi: 10.1007/s40471-018-0144-1. Epub 2018 May 18.
In this paper, we summarize prior studies that have used Mendelian Randomization (MR) methods to study the effects of exposures, lifestyle factors, physical traits, and/or biomarkers on cancer risk in humans. Many such risk factors have been associated with cancer risk in observational studies, and the MR approach can be used to provide evidence as to whether these associations represent causal relationships. MR methods require a risk factor of interest to have known genetic determinants that can be used as proxies for the risk factor (i.e., "instrumental variables" or IVs), and these can be used to obtain an effect estimate that, under certain assumptions, is not prone to bias caused by unobserved confounding or reverse causality. This review seeks to describe how MR studies have contributed to our understanding of cancer causation.
We searched the published literature and identified 76 MR studies of cancer risk published prior to October 31, 2017. Risk factors commonly studied included alcohol consumption, Vitamin D, anthropometric traits, telomere length, lipid traits, glycemic traits, and markers of inflammation. Risk factors showing compelling evidence of a causal association with risk for at least one cancer type include alcohol consumption (for head/neck and colorectal), adult body mass index (increases risk for multiple cancers, but decreases risk for breast), height (increases risk for breast, colorectal, and lung; decreases risk for esophageal), telomere length (increases risk for lung adenocarcinoma, melanoma, renal cell carcinoma, glioma, B-cell lymphoma subtypes, chronic lymphocytic leukemia, and neuroblastoma), and hormonal factors (affects risk for sex-steroid sensitive cancers).
This review highlights alcohol consumption, body mass index, height, telomere length, and the hormonal exposures as factors likely to contribute to cancer causation. This review also highlights the need to study specific cancer types, ideally subtypes, as the effects of risk factors can be heterogeneous across cancer types. As consortia-based genome-wide association studies increase in sample size and analytical methods for MR continue to become more sophisticated, MR will become an increasingly powerful tool for understanding cancer causation.
在本文中，我们总结了先前使用孟德尔随机化（MR）方法研究暴露因素、生活方式因素、身体特征和/或生物标志物对人类癌症风险影响的研究。在观察性研究中，许多此类风险因素已与癌症风险相关联，而MR方法可用于提供证据，证明这些关联是否代表因果关系。MR方法要求感兴趣的风险因素具有已知的遗传决定因素，这些因素可作为该风险因素的替代指标（即“工具变量”或IVs），并且在某些假设下，可用于获得一个效应估计值，该估计值不易受到未观察到的混杂因素或反向因果关系导致的偏差影响。本综述旨在描述MR研究如何有助于我们理解癌症病因。
我们检索了已发表的文献，确定了2017年10月31日前发表的76项关于癌症风险的MR研究。常见研究的风险因素包括饮酒、维生素D、人体测量特征、端粒长度、脂质特征、血糖特征和炎症标志物。显示出与至少一种癌症类型的风险存在因果关联的确凿证据的风险因素包括饮酒（对头颈部和结直肠癌）、成人身体质量指数（增加多种癌症风险，但降低乳腺癌风险）、身高（增加乳腺癌、结直肠癌和肺癌风险；降低食管癌风险）、端粒长度（增加肺腺癌、黑色素瘤、肾细胞癌、神经胶质瘤、B细胞淋巴瘤亚型、慢性淋巴细胞白血病和神经母细胞瘤风险）以及激素因素（影响性类固醇敏感癌症的风险）。
本综述强调饮酒、身体质量指数、身高、端粒长度和激素暴露是可能导致癌症病因的因素。本综述还强调了研究特定癌症类型（理想情况下是亚型）的必要性，因为风险因素的影响在不同癌症类型中可能存在异质性。随着基于联盟的全基因组关联研究样本量的增加以及MR分析方法不断变得更加复杂，MR将成为理解癌症病因的越来越强大的工具。
