Statistical methods for data integration and hospital electronic health records

In biomedical settings the data that are required to answer important scientific questions relating to diseases and healthcare often neither all have the same form nor all come from a single source.

One aspect of this research programme will be to develop statistical methods that make such evidence syntheses across types and scales of data more practical in typical biomedical settings. Bayesian statistical methods lend themselves naturally to this task since they are amenable to a modular perspective.

However, specifying and fitting suitable Bayesian evidence synthesis models in many settings remains a formidable challenge.

We will develop new methods that make specification of a suitable Bayesian model easier, and new computation methods that make fitting such models faster and more practicable, particularly for sensitive biomedical data. We hope to make systematic and principled approaches to these challenges practical and routine.

Much of the work in the programme will be motivated by the detailed hospital electronic health record (EHR) data.

These data reflect intricate clinical processes and pathways, with varied provenance and reliability, meaning naive analyses have considerable potential to mislead.

This programme of research will develop statistical methodology and software to integrate, synthesise and understand such data in a manner that accurately captures the nature of detailed hospital electronic health record data.

Data integration is required, due to the large number of streams of data of widely varying nature that are recorded in hospitals.

We will exploit and extend ideas of causal inference, since many scientific questions of interest are causal but EHR data are largely purely observational.

We will apply and demonstrate such methods to provide trustworthy answers to priority clinical questions within the NHS and beyond, using data from UK hospital Trusts.

Our applied work will particularly focus on acute medical and critical care settings, where suitable data are most widely available and where changes to clinical processes and treatments could lead to dramatic improvements in patient care.

Our work will be conducted in collaboration with clinical doctors, clinical scientists and others in the NHS as appropriate to ensure the research addresses the most pressing questions that will deliver benefits to the healthcare system at both population and individual patient levels.