Dr Vanessa Didelez is a statistician developing methods to understand better causal mechanisms, the processes linking cause and effect in complex systems in motion that evolve over time, so-called dynamical systems. As many standard methods fail to handle multiple time-varying factors rendering them unusable, she is uniquely combining graphical models with background knowledge and statistical algorithms. Although her anticipated methodology could be applied in many different contexts, she foresees particular benefits for computer scientists, social scientists, geneticists and health professionals. For example, well acquainted with the biomedical community, she helps medical researchers analysing large longitudinal HIV patient data sets to enable the advancement of personalised medicine treatment programmes.

This current research is in its inception, however, based on Dr Didelez's numerous previous breakthroughs, it promises to be fruitful. In 2006, she adapted graphical models developed by computer scientist, Judea Pearl, so they could be generalised to analyse interactions between events that occur in continuous time. This approach has been taken forward to model cellular reaction systems as well as the dynamics underlying HIV infections.

In 2007, she worked on a collaborative project funded by the Medical Research Council (MRC) that included using so-called Mendelian randomisation to study the causal effect of Body Mass Index (BMI) on the risk of asthma in children. The name 'Mendelian randomisation' refers to the 19th century scientist Gregor Mendel who investigated the laws of genetic inheritance in peas. The idea is to exploit nature's randomising when it is impossible for the researcher to randomise. She translated this idea into rigorous mathematics which clarified the underlying assumptions of the statistical methodology and encouraged the health community to adopt it. It is possible that in the future this approach might be used to aid understanding of drug side-effects using large observational data sets from health insurers.

In 2010, she demonstrated, with Phillip Dawid, that if causal questions such as: 'is X causal for Y?' are reformulated into decision problems: 'if we manipulate X will it affect Y?' much conceptual, mathematical and computational clarity can be gained. Taking this approach they discovered that a certain algorithm called G-computation, which some statisticians find quite daunting, turns out to be exactly the same as a simple well-known dynamic programming operation. This is important because the validity of the statistical methodology relies on a thorough understanding of the underlying assumptions, especially when the next question to be investigated is: 'what strategy for manipulating X gives us an optimal result regarding Y?' where X could be a treatment programme and Y the survival of a patient.

"I am very active in ensuring the next generation of applied statisticians learn about and use the appropriate statistical methodology for causal modelling and inference", Dr Didelez says. "We usually know much more about a system than just correlations, we take background information into account. You just have to make that explicit in your models so that you avoid pitfalls others have fallen into in the past. It is a very fundamental technique."