Statistical Methods for Improving Causal Analyses

We are trying to develop statistical methods to help medical researchers in their search for causes of disease. A lot of medical research involves gathering data from people and using statistical models to tell us which factors cause later health. For example, we might ask a group of people about their diet as children, as teenagers and as adults, and use this to try to tell us whether poor diet causes cancer in later life.
There are several problems with these studies that make it hard to draw conclusions. One is that people agreeing to be in a study are often different from people who don’t agree. Another is that people tend to drop out of a study over time – and again the people who drop out are often different from the people who stay. A third problem is that people change as they go through life – and we might want to know whether the cause of a disease happens at birth, or during childhood, or whether there are chances to prevent the disease even in adults.
Statistical models may not give the right answers if any of these problems occur – and this could mean that the wrong health advice is given, or the wrong treatments developed. We aim to develop methods that can overcome these real-life problems, and help medical researchers to be more confident in their conclusions about causes of disease.

Aim: The aim of this programme is to develop methods for causal inference that are robust to missing data and can investigate change over time, in order to draw unbiased conclusions about realistic problems, using complex observational data.
Importance: Causal inference methods - in particular instrumental variable (IV) and Mendelian randomization (MR) methods - are now straightforward to implement, can be used with summary data, and are widely used by epidemiologists and medical researchers. However, the majority of real-world clinical research settings are more complex than the standard methods allow: data are missing; samples are selected; exposures and outcomes evolve jointly over time; and data from a wide variety of sources need to be integrated. Methods for causal inference, including IV, may be biased by missing data, including individuals missing due to sample selection. Standard IV methods are not able to address complex (and possibly time-varying) relationships between exposure, covariates and outcome. Multiple studies may provide information about the same causal effect, or about different paths in a network of causal effects, and we need to develop better methods to integrate evidence from different study types in order to draw causal inferences.
Objectives:
1. Develop methods to minimise bias due to missing data
2. Develop methods to model complex exposures and outcomes
3. Develop IV methods to examine causal influences of multiple exposures
4. Integrate evidence to improve causal models
Research plans: Part 1 of this programme will develop methods to use study information, and information external to the study, to infer the missing data structure, to inform all types of causal analyses. We will then focus on methods to maximise the robustness of IV methods to different types of missing data. We will pay particular attention to two cases: two-sample IV (using individual or summary data), and the investigation of disease prognosis. Part 2 will extend current methods for modelling trajectories and variability of exposures and outcomes. We will then focus on overcoming some of the current limitations of IV methods, by using structural equation modelling (SEM) and multivariable IV to examine impacts of time-varying exposures. Finally, we will maximise the use of all research data by extending methods to combine and use external information to inform causal models and sensitivity analyses.