Issues such as missing data, measured and unmeasured confounding, measurement being connected to underlying health status, etc. mean that complex statistical techniques can be necessary to obtain reasonable conclusions about causation from routinely-collected data. Assessing how well these techniques work in practice and identifying optimal analysis approaches is critical to allow researchers to obtain robust and valid conclusions from their analyses. This group undertakes a range of methodological research aiming to provide practical guidance to applied researchers about how best to analyse data from electronic health records for the purpose of causal inference.