Mendelian Randomization

Methods for Causal Inference Using Genetic Variants

Stephen Burgess, Simon G. Thompson

• Verlag: Chapman and Hall/CRC
• Jahr: 2021
• ISBN:
  9781000399592
• Auflage: 2
• Thema: Mathematics
• Thema: Probability & Statistics

Mendelian Randomization: Methods For Causal Inference Using Genetic Variants provides thorough coverage of the methods and practical elements of Mendelian randomization analysis. It brings together diverse aspects of Mendelian randomization from the fields of epidemiology, statistics, genetics, and bioinformatics.

Through multiple examples, the first part of the book introduces the reader to the concept of Mendelian randomization, showing how to perform simple Mendelian randomization investigations and interpret the results. The second part of the book addresses specific methodological issues relevant to the practice of Mendelian randomization, including robust methods, weak instruments, multivariable methods, and power calculations. The authors present the theoretical aspects of these issues in an easy-to-understand way by using non-technical language. The last part of the book examines the potential for Mendelian randomization in the future, exploring both methodological and applied developments.

Features

• Offers first-hand, in-depth guidance on Mendelian randomization from leaders in the field
• Makes the diverse aspects of Mendelian randomization understandable to newcomers
• Illustrates technical details using data from applied analyses
• Discusses possible future directions for research involving Mendelian randomization
• Software code is provided in the relevant chapters and is also available at the supplementary website

This book gives epidemiologists, statisticians, geneticists, and bioinformaticians the foundation to understand how to use genetic variants as instrumental variables in observational data.

New in Second Edition: The second edition of the book has been substantially re-written to reduce the amount of technical content, and emphasize practical consequences of theoretical issues. Extensive material on the use of two-sample Mendelian randomization and publicly-available summarized data has been added. The book now includes several real-world examples that show how Mendelian randomization can be used to address questions of disease aetiology, target validation, and drug development

Part I

Understanding and performing Mendelian randomization

1

Introduction and motivation

This book concerns making inferences about causal effects based on observational data using genetic variants as instrumental variables, a concept known as Mendelian randomization. In this chapter, we introduce the basic idea of Mendelian randomization, giving examples of when the approach can be used and why it may be useful. We aim in this chapter only to give a flavour of the approach; details about its conditions and requirements are reserved for later chapters. Although the examples given in this book are mainly in the context of epidemiology, Mendelian randomization can address questions in a variety of fields of study, and the majority of the material in this book is equally relevant to problems in different research areas.

1.1 Shortcomings of classical epidemiology

Epidemiology is the study of patterns of health and disease at the population level. We use the term ‘classical epidemiology’, meaning epidemiology without the use of genetics, to contrast with genetic epidemiology. A fundamental problem in epidemiological research, in common with other areas of social science, is the distinction between correlation and causation. If we want to address important medical questions, such as to determine disease aetiology (what is the cause of a disease?), to assess the impact of a medical or public health intervention (what would be the result of a treatment?), to inform public policy, to prioritize healthcare resources, to advise clinical practice, or to counsel on the impact of lifestyle choices, then we have to answer questions of cause and effect. The optimal way to address these questions is by appropriate study design, such as the use of prospective randomized trials.

1.1.1 Randomized trials and observational studies

The gold standard for the empirical testing of a scientific hypothesis in clinical research is a randomized controlled trial. This design involves the assignment of different treatment regimes at random to experimental units (usually individuals) in a population. In its simplest form, an active treatment (for example, intervention on a risk factor) is compared against a control treatment (no intervention), and the average outcomes in each of the arms of the trial are contrasted. We will often refer to the putative causal risk factor as the ‘exposure’ variable. We seek to assess whether the exposure is a cause of the outcome, and estimate (if appropriate) the magnitude of the causal effect.

While randomized trials are in principle the best way of determining the causal status of a particular exposure, they have some limitations. Randomized trials are expensive and time consuming, especially when the outcome is rare or requires a long follow-up period to be observed. Additionally, in some cases, a targeted treatment which has an effect only on the exposure of interest may not be available. Moreover, many exposures cannot be randomly allocated for practical or ethical reasons. For example, in assessing the impact of drinking red wine on the risk of coronary heart disease, it would not be feasible to recruit participants to be randomly assigned to either drink or abstain from red wine over, say, a 30-year period. Alternative approaches for judging causal relationships are required.

Scientific hypotheses are often assessed using observational data. Rather than by intervening on the exposure, individuals with high and low levels of the exposure are compared. In many cases, differences between the average outcomes in the two groups have been interpreted as evidence for the causal role of the exposure. However, such a conclusion confuses correlation with causation. There are many reasons why individuals with elevated levels of the exposure may have greater average outcome levels, without the exposure being a causal agent.

Interpreting an association between an exposure and a disease outcome in observational data as a causal relationship relies on untestable and usually implausible assumptions, such as the absence of unmeasured confounding (see Chapter 2) and of reverse causation. This has led to several high-profile cases where an exposure has been widely promoted as an important factor in disease prevention based on observational data, only to be later discredited when evidence from randomized trials did not support a causal interpretation [Taubes and Mann, 1995]. For example, observational studies reported a strong inverse association between vitamin C and risk of coronary heart disease, which did not attenuate on adjustment for a variety of alternative risk factors [Khaw et al., 2001]. However, experimental data results obtained from randomized trials showed a non-significant association in the opposite direction [Collins et al., 2002]. The confidence interval for the observational association did not include the randomized trial estimate [Davey Smith and Ebrahim, 2003]. Similar stories apply to the observational and experimental associations between β-carotene and smoking-related cancers [Peto et al., 1981; Hennekens et al., 1996], and between vitamin E and coronary heart disease [Hooper et al., 2001]. More worrying is the history of hormone-replacement therapy, which was previously advocated as being beneficial for the reduction of breast cancer and cardiovascular mortality on the basis of observational data, but was subsequently shown to increase mortality in randomized trials [Rossouw et al., 2002; Beral et al., 2003]. More reliable approaches are therefore needed for assessing causal relationships using observational data. Mendelian randomization is one such approach.

1.2 The rise of genetic epidemiology

Genetic epidemiology is the study of the role of genetic factors in health and disease for populations. We sketch the history and development of genetic epidemiology, indicating why it is an important area of epidemiological and scientific research.

1.2.1 Historical background

Although the inheritance of characteristics from one generation to the next has been observed for millennia, the mechanism for inheritance was long unknown. When Charles Darwin first proposed his theory of evolution in 1859, one of its major problems was the lack of an underlying mechanism for heredity [Darwin, 1871]. Gregor Mendel in 1866 proposed two laws of inheritance: the law of segregation, that when any individual produces gametes (sex cells), the two copies of a gene separate so that each gamete receives only one copy; and the law of independent assortment, that ‘unlinked or distantly linked segregating gene pairs assort independently at meiosis [cell division]’ [Mendel, 1866]. These laws are summarized by the term ‘Mendelian inheritance’, and it is this which gives Mendelian randomization its name...