(For each chapter, clicking on the chapter name expands into a list of contents for that chapter, and clicking on "Summary" provides a brief outline of the contents of the chapter, including links to relevant papers co-written by the authors of the book.)

Preface

Sample available!

PART I: Using genetic variants as instrumental variables to assess causal relationships

Chapter 1: Introduction and motivation

  • 1.1 Shortcomings of classical epidemiology
    • 1.1.1 Randomized trials and observational studies
  • 1.2 The rise of genetic epidemiology
    • 1.2.1 Historical background
    • 1.2.2 Genetics and disease
  • 1.3 Motivating example: The inflammation hypothesis
    • 1.3.1 C-reactive protein and coronary heart disease
    • 1.3.2 Alternative explanations for association
    • 1.3.3 Instrumental variables
    • 1.3.4 Genetic variants as instrumental variables
    • 1.3.5 Violations of instrumental variable assumptions
    • 1.3.6 The CRP CHD Genetics Collaboration
  • 1.4 Other examples of Mendelian randomization
  • 1.5 Overview of book
    • 1.5.1 Structure
  • 1.6 Summary
Summary
Download sample!
 

Chapter 2: What is Mendelian randomization?

  • 2.1 What is Mendelian randomization?
    • 2.1.1 Motivation
    • 2.1.2 Instrumental variables
    • 2.1.3 Confounding and endogeneity
    • 2.1.4 Analogy with a randomized controlled trial
  • 2.2 Why use Mendelian randomization?
    • 2.2.1 Reverse causation and case-control studies
    • 2.2.2 Exposures that are expensive or difficult to measure
  • 2.3 A brief overview of genetics
    • 2.3.1 Reading the genetic code
    • 2.3.2 Using a genetic variant as an instrumental variable
  • 2.4 Summary
Summary

Chapter 3: Assumptions for causal inference

  • 3.1 Observational and causal relationships
    • 3.1.1 Causation as the result of manipulation
    • 3.1.2 Causation as a counterfactual contrast
    • 3.1.3 Causation using graphical models
    • 3.1.4 Causation based on multivariable adjustment
  • 3.2 Finding a valid instrumental variable
    • 3.2.1 Instrumental variable assumptions
    • 3.2.2 Validity of the IV assumptions
    • 3.2.3 Violations of IV assumptions: biological mechanisms
    • 3.2.4 Violations of IV assumptions: non-Mendelian inheritance
    • 3.2.5 Violations of IV assumptions: population effects
    • 3.2.6 Assessing the IV assumptions
    • 3.2.7 Summary of issues relating to IV validity
    • 3.2.8* Definition of an IV as a random variable
    • 3.2.9* Definition of an IV in potential outcomes
  • 3.3 Testing for a causal relationship
    • 3.3.1 Converse of the test
    • 3.3.2 Does Mendelian randomization really assess a causal relationship?
    • 3.3.3 Interpreting a null result
  • 3.4 Estimating a causal effect
    • 3.4.1* Additional IV assumptions for estimating a causal effect
    • 3.4.2* Causal parameters
    • 3.4.3 Parametric IV assumptions
  • 3.5 Summary
Summary

Chapter 4: Methods for instrumental variable analysis

  • 4.1 Ratio of coefficients method
    • 4.1.1 Continuous outcome, dichotomous IV
    • 4.1.2 Continuous outcome, polytomous IV
    • 4.1.3 Binary outcome
    • 4.1.4 Retrospective and case-control data
    • 4.1.5 Confidence intervals
    • 4.1.6 Absence of finite moments
    • 4.1.7 Coverage and efficiency
    • 4.1.8 Reduced power of IV analyses
  • 4.2 Two-stage methods
    • 4.2.1 Continuous outcome - two-stage least squares
    • 4.2.2 Binary outcome
    • 4.2.3* Non-collapsibility
    • 4.2.4* Adjusted two-stage method
  • 4.3 Likelihood-based methods
    • 4.3.1 Full information maximum likelihood
    • 4.3.2 Limited information maximum likelihood
    • 4.3.3 Bayesian methods
    • 4.3.4 Comparison of two-stage and likelihood-based methods
    • 4.3.5* Likelihood-based methods with binary outcomes
  • 4.4* Semi-parametric methods
    • 4.4.1* Generalized method of moments
    • 4.4.2* Structural mean models
    • 4.4.3* Lack of identification for binary outcomes
  • 4.5 Efficiency and validity of instruments
    • 4.5.1 Use of measured covariates
    • 4.5.2 Weak instruments
    • 4.5.3 Overidentification tests
    • 4.5.4 Endogeneity tests
  • 4.6 Computer implementation
    • 4.6.1 IV analysis of continuous outcomes in Stata
    • 4.6.2 IV analyses of binary outcomes in Stata
    • 4.6.3 IV analysis in SAS
    • 4.6.4 IV analysis in R
    • 4.6.5 IV analysis in WinBUGS
  • 4.7 Summary
Summary

Chapter 5: Examples of Mendelian randomization analysis

  • 5.1 Fibrinogen and coronary heart disease
    • 5.1.1 Study design
    • 5.1.2 Genetic instruments
    • 5.1.3 Statistical methodology
    • 5.1.4 Results
    • 5.1.5 Commentary
  • 5.2 Adiposity and blood pressure
    • 5.2.1 Study design
    • 5.2.2 Genetic instruments
    • 5.2.3 Statistical methodology
    • 5.2.4 Results
    • 5.2.5 Commentary
  • 5.3 Lipoprotein(a) and myocardial infarction
    • 5.3.1 Study design
    • 5.3.2 Genetic instruments
    • 5.3.3 Statistical methodology
    • 5.3.4 Results
    • 5.3.5 Commentary
  • 5.4 High- and low-density lipoprotein cholesterol and MI
    • 5.4.1 Study design
    • 5.4.2 Genetic instruments
    • 5.4.3 Statistical methodology
    • 5.4.4 Results
    • 5.4.5 Commentary
  • 5.5 Discussion
Summary

Chapter 6: Generalizability of estimates from Mendelian randomization

  • 6.1 Internal and external validity
    • 6.1.1 Time-scale and developmental compensation
    • 6.1.2 Usual versus pathological levels
    • 6.1.3 Extrapolation of small differences
    • 6.1.4 Different pathways of genetic and intervention effects
    • 6.1.5 Differences in populations
  • 6.2 Comparison of estimates
    • 6.2.1 Cholesterol and coronary heart disease
    • 6.2.2 Blood pressure and coronary heart disease
  • 6.3 Discussion
    • 6.3.1 Using Mendelian randomization in drug assessment
    • 6.3.2 Using Mendelian randomization in drug discovery
    • 6.3.3 Role of estimation in Mendelian randomization
  • 6.4 Summary
Summary

 

PART II: Statistical issues in instrumental variable analysis and Mendelian randomization

Chapter 7: Weak instruments and finite-sample bias

  • 7.1 Introduction
  • 7.2 Demonstrating the bias of IV estimates
    • 7.2.1 Bias of IV estimates in small studies
    • 7.2.2 Distribution of the ratio IV estimate
  • 7.3 Explaining the bias of IV estimates
    • 7.3.1 Correlation of IV associations
    • 7.3.2 Finite-sample violation of IV assumptions
    • 7.3.3 Sampling variation within genetic subgroups
  • 7.4 Properties of IV estimates with weak instruments
    • 7.4.1 Bias of IV estimates
    • 7.4.2 Coverage of IV estimates
    • 7.4.3 Lack of identification
  • 7.5 Bias of IV estimates with different choices of IV
    • 7.5.1 Multiple candidate IVs in simulated data
    • 7.5.2 Multiple candidate IVs in the Framingham Heart Study
  • 7.6 Minimizing the bias of IV estimates
    • 7.6.1 Increasing the F statistic
    • 7.6.2 Adjustment for measured covariates
    • 7.6.3 Borrowing information across studies
  • 7.7 Discussion
    • 7.7.1 Bias–variance trade-off
    • 7.7.2 Combatting weak instrument bias in practice
    • 7.7.3 Bias in study-level meta-analysis
    • 7.7.4 Caution about validity of IVs
  • 7.8 Key points from chapter
Summary
Download sample!
 

Chapter 8: Multiple instruments and power

  • 8.1 Introduction
  • 8.2 Allele scores
    • 8.2.1 Choosing variants to form allele score
    • 8.2.2 Choosing weights in a weighted allele score
    • 8.2.3 Performance of an allele score in IV estimation
  • 8.3 Power of IV estimates
    • 8.3.1 Power with a single IV, continuous outcome
    • 8.3.2 Power with a single IV, binary outcome
    • 8.3.3 Power with multiple IVs
  • 8.4 Multiple variants and missing data
    • 8.4.1 Data from the British Women’s Heart and Health Study
    • 8.4.2 Power and missing data
    • 8.4.3 Methods for incorporating missing data
    • 8.4.4 Results of missing data analyses
  • 8.5 Discussion
    • 8.5.1 Heterogeneity and supplementary analyses
    • 8.5.2 Subsample Mendelian randomization
    • 8.5.3 Relevance to epidemiological practice
  • 8.6 Key points from chapter
Summary

Chapter 9: Multiple studies and evidence synthesis

  • 9.1 Introduction
  • 9.2 Assessing the causal relationship
  • 9.3 Study-level meta-analysis
  • 9.4 Summary-level meta-analysis
    • 9.4.1 Multiple genetic variants in a single study
    • 9.4.2 Single genetic variant in multiple studies
    • 9.4.3 Single common genetic variant in multiple studies
    • 9.4.4 Multiple genetic variants in multiple studies – Genetic associations
    • 9.4.5 Multiple genetic variants in multiple studies – Genetic subgroups
    • 9.4.6 Fixed- and random-effects meta-analysis
    • 9.4.7 Using published summary-level data
    • 9.4.8 Advantages of summary-level meta-analysis
  • 9.5 Individual-level meta-analysis
    • 9.5.1 Modelling in a single study
    • 9.5.2 Model of genetic association
    • 9.5.3 Common genetic variants
    • 9.5.4 Lack of exposure or outcome data
    • 9.5.5 Advantages of individual-level meta-analysis
    • 9.5.6 Combining summary- and individual-level data
  • 9.6 Example: C-reactive protein and fibrinogen
  • 9.7 Binary outcomes
    • 9.7.1 Using summary-level data
    • 9.7.2 Using individual-level data
    • 9.7.3 Combining incident and prevalent cases in a longitudinal study
  • 9.8 Discussion
    • 9.8.1 Precision of the causal estimate
    • 9.8.2 Two-sample Mendelian randomization
    • 9.8.3 Relevance to epidemiological practice
  • 9.9 Key points from chapter
Summary

Chapter 10: Example: The CRP CHD Genetics Collaboration

  • 10.1 Overview of the dataset
    • 10.1.1 Study design
    • 10.1.2 Exposure data: C-reactive protein
    • 10.1.3 Genetic data
    • 10.1.4 Outcome data: coronary heart disease
    • 10.1.5 Covariate data
    • 10.1.6 Validity of the SNPs used as IVs
  • 10.2 Single study: Cardiovascular Health Study
    • 10.2.1 Results
    • 10.2.2 Posterior distributions from Bayesian methods
  • 10.3 Meta-analysis of all studies
    • 10.3.1 Using SNPs one at a time
    • 10.3.2 Using all SNPs
  • 10.4 Discussion
    • 10.4.1 Precision of the causal estimate
    • 10.4.2 Limitations of this analysis
    • 10.4.3 Assessing the IV assumptions
    • 10.4.4 Interpretation of the results
    • 10.4.5 Relevance to epidemiological practice
  • 10.5 Key points from chapter
Summary

 

PART III: Prospects for Mendelian randomization

Chapter 11: Future directions

  • 11.1 Future methodological developments in instrumental variable techniques
    • 11.1.1 Survival data
    • 11.1.2 Non-linear exposure–outcome relationships
    • 11.1.3 Untangling the causal effects of related exposures
    • 11.1.4 Elucidating the direction of causation
    • 11.1.5 Investigating indirect and direct effects
  • 11.2 Future applied developments for Mendelian randomization
    • 11.2.1 High-throughput cell biology: -omics data
    • 11.2.2 Mendelian randomization with GWAS data
    • 11.2.3 Whole-genome sequencing and rare variants
    • 11.2.4 Published data and two-sample Mendelian randomization
  • 11.3 Conclusion
Summary