Exploring the use of Causal Inference methods on the MAAS dataset
The Maastricht Aging Study (MAAS) is a longitudinal observational study consisting of nearly 1900 individuals who have been measured over the course of 15 years. The primary aim of this study was to examine age-related cognitive decline and dementia incidence, and to relate this to lifestyle and health factors. However, since it is an observational study, drawing causal conclusions from the data is challenging. Our goal is to make a first attempt to use this dataset to estimate causal effects.
We will do this using methods from the field of causal inference. However, these causal inference methods are inherently subject to very strong assumptions. These assumptions include a lack of confounding (latent variables affecting treatment and outcome), positivity (an assumption about the distribution of variables in your dataset) and consistency (an assumption about the treatment effect that you are estimating). Additionally, since this is a longitudinal dataset, we may encounter the issue of time-varying confounding, where treatment and confounders influence each other over time.
In this project, you will apply causal inference methods that can account for time-varying confounding, such as Marginal Structural Models and the parametric G-formula. You will use these methods to estimate the causal effects of lifestyle factors on risk of dementia and cognitive decline. Beyond applying these methods, you will also explore the validity of the underlying assumptions. Since some of these assumptions are likely to be violated to some extent, we need to evaluate the potential impact of these violations. Key questions include: Can we find out how strong these violations are? And, to what extent do potential violations impact our results?
Contact: Wouter Kant (RU); Marco Loog (RU).