Predictive Monitoring: Heart Attack and Stroke?

Filed under: Innovation,Risk — lenand @ 6:12 am
Tags: , ,

Machine learning via neural networks produces impressive results.  The Blood Glucose prediction hackathon used three data streams: historical blood glucose, insulin dosage and carbohydrate consumption.  This explains approximately 50% of the prediction.

Blood Glucose Predictive Power

Blood Glucose Predictive Power

Adding three more streams would increase this to 85%; other nutrition, activity and other lifestyle factors.  All these could be simply collected from mobile devices and used in other health prediction applications. It shows the value of monitoring data streams for multiple purposes.  This data could also be analysed alongside logs of blood pressure and heart rhythms.

Just think of the value to people who are at risk of heart attack or stroke.   Real-time predictions of heart rate and blood pressure could set alarms that would moderate a person’s behaviour.  An impertinent machine telling you to Stop driving!, sit down! or have a rest! may upset your plans – but it is better than risking your own or another life.

Machine learning is not rule based – it calculates the rules.



Predictive Monitoring: Hackathon success

Filed under: Technology — lenand @ 2:13 pm
Tags: , ,

In the era of big data, analytical methods have to be fast and effective.  Traditional statistical and rule based methods cannot keep pace with volumes and variety of data being collected.  They will miss patterns of data that could improve decision making.

The time has come when there is little option other than starting to consider automatic machine learning on an enterprise scale.  Patterns in data can be identified and used to calculate the probability of events in the future.  The prediction can be done in real time applications, such as intensive care monitoring in hospitals.

There seems to be a view that machine learning, pattern matching and prediction is expensive, slow or inaccurate – or all three!  Here is an example that demonstrates otherwise.  A Hackathon produced a prediction of blood sugar level in a diabetic patient in less than a day.  The machine learning algorithms were not meditated by any additional clinical input.

Blood sugar prediction

Blood sugar prediction

Similar techniques could prove invaluable in care of the elderly, early dementia and psychiatric patient monitoring.

This Predictive Monitoring hackathon was held in Singapore.  The UK should be more proactive in supporting such machine learning innovation, or a lead in vital technology could be lost.

Create a free website or blog at WordPress.com.