Southampton Academy of Research

Dr Thomas Daniels is a Consultant Chest Physician at University Hospital Southampton and holder of a Research Leaders Programme (RLP) award.

He has a particular interest in how machine learning can be used to improve healthcare.

Avoiding false alarms

The National Early Warning Score (NEWS2) is a scoring system used across the NHS to identify seriously ill hospital patients. It draws on a combination of physiological measurements, routinely recorded at the patient's bedside, to create a combined score of the state of their health.

The six measurements used are respiration rate, oxygen saturation, systolic blood pressure, pulse rate, level of consciousness or new confusion, and temperature. Each one is given a score between 0 and 3, and these are added together. The higher the total score, the worse the patient’s condition.

While NEWS2 is a useful tool, it has limitations. It is a one-size-fits-all scoring system, Thomas explains, even though what is ‘normal’ varies from one person to the next. A healthy blood pressure, for example, changes as a person gets older. There is also evidence to show that many patients can deteriorate without an abnormal NEWS2 score.

Thom says this can lead to delays in vital changes to treatment for deteriorating patients, and to false alarms, where nurses and doctors spend time examining and running unnecessary tests on improving patients, when they are actually fine.

This is unpleasant for those patients who were unnecessarily disturbed, and wastes doctors’ time, meaning it takes longer for them to see the patients who really are sick.

Using health data

The NEWS2 system only uses six measurements, but Thom says there is now a huge amount of health data that could be used to improve the system and save lives.

“With machine learning, we can use data much more intelligently,” he explains.

“We’re already doing that when Netflix recommends films to us, or when we’re buying stuff off Amazon, or when we’re planning a route to get home using Google Maps. That’s done through machine learning, using large data sets.

“Similarly large data sets are available in hospitals, but we’re not using them for saving lives. Yet we are using them to watch better films. That feels the wrong way around to me.”

He therefore plans to create a new scoring system, which he calls the Computer Assisted Risk Deterioration Score (CARDS). This would use machine learning and large health data sets, including vital signs, blood tests results and demographic information, to generate the new CARDS score.

If all goes to plan, those who use the system would notice very little change except improved accuracy. This is because it would generate a similar combined score, which could be interpreted in just the same way as the current NEWS2 system.

Realising his vision

For his RLP award, Thomas is collaborating with mathematician Prof Mihaela van der Shaar at the University of Cambridge. They intend to adapt an algorithm she has tested on US health data sets to make it suitable for use across the NHS. Their plan is to run this in the background to check its performance, identify any issues and quantify its benefits, before running a randomised controlled trial in the future.

He is also working with Prof Age Chapman, as part of the NIHR Southampton Biomedical Research Centre’s Data, Health and Society theme, to explore the ethical implications of using health data.

“I have this vision of bringing the huge societal advantages of machine learning into the healthcare environment,” he says, “whereas at the moment, the use of data in healthcare is very much stuck in the 20thCentury.”