HypoPredict

Can we really predict hypotension? A lot of people will have their doubts, but mostly because we can’t explain the physiology. That’s the whole point of Bayesian Artificial Neural Networks – finding the proverbial “needle in a haystack”. BANN’s are about using very specific logic to compare large amounts of data to find patterns which can subsequently indicate future events.

We are not, and have no intention, of suggesting we can use this for diagnosis. We are researching benefits of using a prediction technology for alerting clinicians to upcoming events.

Our objective is enabling an earlier intervention as when an episode occurs, mitigating the depth of fall and the length of time it persists. We’ll try to give early notice of an increase in the probability of an episode within the immediate future – say 15 to 30 minutes.

Our results are very “early” – we’ve only just started live monitoring – but, understanding that, and knowing what we know, they are very encouraging.

Patient 85
The image on the right shows two traces showing the actual monitored blood pressure alongside the probability as determined by the BANN- for Patient 85.

We can see BPm readings are unstable during the period 18.30 – 19.30. They even out between 19.30 and 20.00 at which point there’s a decline to below the threshold which continues to trend down for a further 30 minutes.

In the lower chart our Hypopredict programme isn’t noticing very much during the period of instability. Around 19.45 a pattern emerges in advance of the drop at 20.00 and continues through to 20.30. We can see there’s something wrong up to 15 minutes before the drop below the 70 BPm threshold.

Patient 86
In the case of patient 86 we see a different pattern in the traces – see below. BPm readings are stable during the period 15.00 to 16.45 but throughout that period Hypopredict is signalling a probability. By 16.15 it’s giving a solid prediction and continues to do so until the actual drop occurs at 16.47.

Nobody is suggesting these two charts prove anything, but we think they show we’re on to something.

We do have the benefit of knowing the results from our previous phase – when we trained the BANN on legacy data. All of that will be released if, and when, we’re ready to announce results. The BANN is comparing trends in the relationships between 10 different parameters. It compares those trends with a history of relationships and subsequent episodes, and tells us when some similarities appear.

Everybody on our team is excited about this. Maybe there’s a new paradigm emerging. One in which clinicians get a helping hand from engineers and computers, not to do their job for them, but to point them in the right direction, so’s they can do more, for more patients.

Footnote
The Avert-IT project is funded by the EU under FP7. The objectives of the research we’re determined by a group of Europe’s leading clinical scientists – surgeons, intensivists and physicists. This is a revolution in innovation with the clinicians deciding what they need, to do a better job for their patients, software engineers and computer scientists harnessing technology to deliver it, and the European Union helping to pay for it.

If you’d like to know more, or help bring this technology to market contact Steve Reeves – Project Exploitation Manager at steve.reeves@avantrasara.com.