Mediation analysis in life course epidemiology: methodological innovation and application to studies of obesity and cardiometabolic health

Life course epidemiology is the study of how things that happen during pregnancy, childhood and adulthood influence health and wellbeing. Studying people across their life allows us to examine the dynamic ways in which things change across life and how the pattern of change relates to disease; for example we could ask whether people who become obese as children have the same risk of heart disease as people who are normal weight as children but become obese as adults.

A key aim of life course studies is to understand the chain of events leading from cause to effect. For example, we know that people with low income have greater risk of becoming obese and we may want to understand why, e.g. what role does physical activity play? This type of analysis is called mediation. Mediation studies can tell us what might happen if we intervened on the steps linking cause to effect. For example if we could make blood pressure levels more similar in obese and non-obese people, would the link between obesity and heart attacks be reduced?

At the moment, most mediation studies carry out the analysis in a way that does not reflect the complexity of real life. For example if we are interested in how obesity is related to the structure and function of the heart and whether any link between them was due to blood pressure, most research so far has used obesity measured at one age, blood pressure measured at one age, and structure and function of the heart measured at one age. All of these things change across life, and ignoring these changes means the research could get the wrong answers. In order to do research on mediation that makes better use of data from across life, we need to develop statistical methods that can do this. I will compare several methods for this type of analysis. As well as comparing the methods, I will use them to answer these questions about obesity, heart disease and diabetes:

1. Babies who are born small or who grow very quickly in the first few months of life have a higher risk of diabetes and heart disease as adults. This may be because what happens in early life changes the way the body's metabolic system works. In this case, we would need to change growth before or very soon after birth to reduce the risk of diabetes and heart disease. Alternatively, growth in early life might set people on a path, and only the end of that path that matters; e.g. it could be only obesity in adulthood that causes disease. In this case, we could change body size either in childhood or adulthood and would still be able to reduce the risk of disease. To tell the difference between these two possibilities, I will use information from across the whole life. This research will help us to know what might happen to the risk of diabetes and heart disease if we can prevent or treat obesity at different ages.

2. Obesity can lead to many different health problems. Because of this we might want to try and prevent people from becoming obese, but this is not always possible. Therefore it is also important to find ways to stop obese people from becoming ill. It is possible to reduce the link between obesity and heart disease, for example by using drugs to reduce blood pressure or levels of lipids (fats) in the blood. I will use one of the largest available research studies (UK Biobank) to find out how much the link between obesity and heart disease could be reduced if we intervened on different pathways.

3. Low income, education and social class are all linked with a higher risk of diabetes and heart disease. We know that smoking, physical activity, diet and alcohol explain some, but not all, of this link. However, most research so far has only used information from one time in life. I will use detailed data from across life to understand how we could reduce these unfair differences in the risk of disease.

Mediation analysis is a key focus in life course epidemiology and can offer insight into mechanisms and likely effects of interventions, but measurement error and confounding can cause bias. Current available methods do not easily extend to complex life course questions. Thus methods development is needed. This fellowship combines methods development with applied research on obesity and cardiometabolic health: 1) whether fetal and infant growth affect cardiometabolic health independently of final attained adiposity, 2) the role of blood pressure (BP), lipids, glucose, insulin and DNA methylation in the association between adiposity and cardiometabolic health, 3) the role of behaviours, BP, lipids, glucose, insulin and DNA methylation in socioeconomic inequalities in cardiometabolic health. Through addressing these questions in 7 cohorts and using simulation studies I will develop mediation methods that minimise bias due to measurement and confounding:

1. Longitudinal mediation analysis (exposure, mediator and/or outcome defined by repeated measures) helps overcome measurement error since a trajectory is likely to better represent true levels than a single measure, and can assess more nuanced questions than using single measures. I will compare longitudinal mediation methods (counterfactual methods, structural equation models, multilevel models, structured life course approach)
2. In network Mendelian Randomization (MR) genetic variants are used as instrumental variables (IVs) for both the exposure and mediator to estimate mediation. The genetic IVs alleviate bias due to measurement error and confounding
3. I will extend network MR to enable it to address more complex life course questions by using genetic variants with age-specific gene expression as age-specific IVs

In summary, I will identify potential intervention targets, quantify likely effects of interventions to improve cardiometabolic health, and contribute to advances mediation analysis methods

In addition to scientists working in a closely-related field, I anticipate that the following groups will benefit from my research:

1. Researchers interested in mediation analysis. Analysing the pathways linking cause and effect is crucial in many areas of science, both throughout most areas of health research and beyond into other disciplines. The methodological developments that arise from my research will therefore have broad relevance and may result in improvements to statistical methods used for mediation analysis in the short to medium term, i.e. within the term of this fellowship. To facilitate this, I will ensure rapid publication of my findings, including both papers detailing specific results and review and guideline papers. I will publish in both discipline-specific journals and in broader journals that are more likely to be accessed by the wider scientific community, and I will present at a broad range of conferences.

2. Academics and health care practitioners who design, evaluate or implement interventions designed to prevent or treat obesity, alleviate the health consequences of obesity, or reduce socioeconomic inequalities in obesity and cardiometabolic health. My findings will identify potential intervention targets and quantify the potential effects of interventions targeted at different pathways or different age groups, and will therefore feed in to intervention design in the medium term, including within the duration of this fellowship. To ensure that this happens I will communicate with colleagues in this field of research, e.g. the Centre for the development and evaluation of complex interventions for public health improvement (DECIPHer) and present at relevant conferences.

3. National and international bodies responsible for health and social care guideline development. In the longer term, the results of trials of interventions, to which my results have contributed to the design, will form part of guidelines for the prevention of obesity, the reduction of the health consequences of obesity, and the reduction of socioeconomic inequalities in obesity and cardiometabolic health. Since the methods I develop will contribute to other subject areas, this is likely to also be relevant for different health and social outcomes.

4. The pharmaceutical industry. If my analyses of the role of biological intermediates in disease processes, or analyses by others using the methods I have developed, identify novel potential drug targets (e.g. metabolites) or demonstrate the causality or otherwise of known intermediates in pathways to disease, these results will be beneficial to the pharmaceutical industry. This is likely to be a long-term impact, which I will ensure by prompt publication of results and through discussion with the industry, e.g. via the MRC Integrative Epidemiology Unit advisory board.

5. Organisers, funders and participants of cohort studies. The methodological developments that arise through this fellowship will improve the ability of cohort studies to address complex questions about the life course aetiology of disease, and will enhance the ability of research using cohort studies to reach correct, causal conclusions. This will thus lead to benefits in terms of cost effectiveness for investment in cohort studies, and will mean that the time and effort contributed by participants leads to improved results. This is a long-term impact of my research, which I will ensure by collaborating across a broad range of disciplines in order to promote widespread uptake of the methods I develop.