Novel statistical methods to unlock the potential of routinely collected health data: COVID-19 & beyond

The recent pandemic has highlighted the importance and great potential of electronic health record data for addressing urgent public health questions in a timely manner, while demonstrating key gaps in existing methodology for this setting. The overall aim of this proposal is to develop statistical methodology to remove barriers that have hampered important public health issues in COVID-19 being fully addressed; and to apply the methods to answer the immediate and arising public health questions in COVID-19 and more broadly. Key goals are to: develop a suite of methodological tools to enable computationally-efficient self-recalibrating risk prediction in EHR databases; develop a framework for assessing the worth of different treatments within EHR databases, accounting for potential high-dimensional confounding and implementing approaches to estimate individual treatment effects within this; create analytic approaches to address causal questions requiring data from two separate sources in the absence of full individual linkage. Data held within the OpenSAFELY platform and the UK Clinical Practice Research Database will be used to motivate the statistical methodological work. Optimal methodological approaches will be applied to address important questions arising in COVID-19 and more broadly, to key questions in chronic disease.