The Oxford Martin Programme on

Deep Medicine

The Challenge

Combining expertise in healthcare, epidemiology, computer science and engineering, we are investigating how application of advanced statistics, machine learning and deep learning techniques to large and complex biomedical datasets can lead to better understanding and management of chronic diseases and multimorbidity.

Advances in medicine over the past few decades have led to an unprecedented increase in life expectancy and reduction in major disabilities. More people can expect to live into their 60s and beyond than die prematurely before the age of 60. But this longevity brings with it new problems: a rise in chronic conditions such as diabetes, dementia and heart failure, with many patients suffering from multiple conditions at the same time (multimorbidity), which are poorly understood.

Our limited understanding of complex disease patterns and risks is further compounded by current methods of research, which tend to oversimplify complexity and often underrepresent patients with complicated disease history or concurrent treatments.

The advent of ‘Big Data’ can offer unprecedented opportunities for extending our knowledge and understanding of complex disease patterns and risk, but harnessing its full potential will require new and innovative methods of analysis.

We are developing scalable methods for analysis of biomedical datasets, aim to generate insights into complex disease patterns, risk trajectories and treatment effects, and provide empirical evidence for the value of applying data mining, modelling and machine learning techniques to biomedical data.

Priority areas include the development and testing of models that make use of the depth and breadth of electronic health records to better represent an individual’s medical history. Methods are being developed to assess causal questions and uncertainty of estimations in the context of complex observational data and to contrast them with conventional epidemiological methods, including survival models, Mendelian randomisation and clinical trials.

We seek collaborations with other international partners in both academia and commercial sector for further validation and extension of our work.

featured publication

Predicting the risk of emergency admission with machine learning: Development and validation using linked electronic health records

Emergency admissions are a major source of healthcare spending. This paper published in the Journal PLoS medicine aimed to derive, validate, and compare conventional and machine learning models for prediction of the first emergency admission. It found that that standard machine learning models not only outperformed the best statistical model but they did so with substantially better performance and calibration.

Read now
Plos medicine