AI breakthrough could save lives through disease prediction and prevention

Artificial intelligence (AI) could help predict how an individual patient’s illness will progress and what treatment will be most successful to help them recover, saving lives across the globe, according to new research by the University of Leicester.

A study used AI to map possible outcomes for patients using snapshots of data from many of people at different states of disease, known as Big Data, to create a pathway or trajectory model. 

Patient records can provide vital clues towards determining how their disease will progress and the best course of treatment. 

The research, which focused on diabetic and heart attack patients, was led by Professor Alexander Gorban, at the University of Leicester, and Dr Andrei Zinovyev, at the Institut Curie, in Paris, and their collaborators.

Professor Alexander Gorban, Director of the Centre for Artificial Intelligence, Data Analysis and Modelling (AIDAM) and Professor of Applied Mathematics at the University of Leicester, said: “We as a society are at the tipping point of transforming healthcare through the use of Big Data.

“By analysing large sets of clinical records using artificial intelligence, we can develop patient paths. The key is determining the type of the trajectory the patient is on as early as possible - this will influence the choice of treatment and increase the chances for its success. 

“This is an exciting project which shows that even snapshots of patient records can play a vital role in predicting and preventing disease.”

As part of the study, the team was able to identify clear clinical trajectories in diabetes patients, which set out the probability of readmission to hospital within one month after leaving, having analysed a dataset representing 10 years (1999 - 2008) of clinical care of patients with diabetes at 130 US hospitals and integrated delivery networks.

Reconstructing individual clinical trajectory requires long-term follow-up of a patient, with systematic collection of the information. This data (called longitudinal or diachronic observations) remains difficult to collect.  

Instead, the new technology allows the extraction of typical clinical trajectories from large collections of data on different patients without individual follow-up. Such collections are readily available for the analysis in many clinical centres.

The researchers also state that the suggested methodology is general enough to be applied to many different diseases and types of data.

The paper ‘Trajectories, bifurcations, and pseudo-time in large clinical datasets: applications to myocardial infarction and diabetes data’ by Golovenkin, S. E., Bac, J., Chervov, A., Mirkes, E. M., Orlova, Y. V., Barillot, E.,Gorban A. N. & Zinovyev, A. (2020) was published in the journal GigaScience, 9(11), giaa128. https://doi.org/10.1093/gigascience/giaa128