Predicting illnesses and preventing potential deaths with computers that learn the symptoms and can detect them before doctors may help to solve a $20bn healthcare problem.
Big data is being used by Amara Health Analytics to predict when patients may be infected and then letting their doctor know via text message.
In particular it tries to predict when sepsis, the presence of harmful bacteria and their toxins through an infection or a wound, is likely to occur or be undiagnosed.
In 2011 sepsis resulted in an aggregate healthcare cost of $20.3 billion in the US.
Steve Nathan, CEO at Amara Health Analytics told Factor: “We are taking a data centric approach to clinical decision support. We’ve started in a hospital setting although our approach applies to other care settings and many different disease states.
“What we’re doing is we’re helping clinicians make those decisions through software that is constantly analysing all of the available data in the hospital so really data driven decision support in real time.”
This could one day lead to systems that tries to diagnose all different types of diseases every minute – although the scale of the data and systems that would be needed for this are not ready yet.
The company collects big data from a variety of sources from the hospitals it works with, it has been testing at four different hospitals in its early phases.
Data streams include admissions data, demographics, lab orders and lab results, data on patients’ vitals and from electronic medical records.
It is then analysed in an attempt to try and pick up on things data doctors may miss – often through no fault of their own.
All the data which they collect is powered by DataStax systems who also work with companies such as Netflix and eBay.
Nathan said: “We do deep natural language processing to find predictive signals in that narrative text. So, for example in sepsis altered mental status [of a patient] is a piece of evidence that is very important and may indicate infection.
“All of the things like natural language processing and all of the analytics that we do, the models that have been trained with machine learning all of that runs on that data server.
“What goes back to the clinician in the hospital is simply a text message that they receive on their smartphone on their tablet and that message calls their attention to an at risk patient.”
He says that in the future the technology could be rolled out to try and detect multiple diseases at once.
“I suppose you could imagine a system that is one single system that is trying to support all possible decision types for all possible disease states minute by minute.
“We’re not trying to do that and I actually think it is a little early for one system to take that on now but maybe in the future.”
With sensitive health data, such as any type of personal information, privacy is a massive concern for those who it belongs to.
To be confident in handing personal details across to a third party, the individual must be confident that they will not be misused.
Nathan said that all the feeds the company receive are transferred through an encrypted private network (VPN) and are not de-identified while the system is running.
He said: “I want to emphasise that patient data privacy is absolutely paramount.
“When it comes to these matters of security in the data centre and our employees being trained and understanding the sensitivity of this data, that’s super important.”