The health sector is a leader in the deployment of big data technologies. Computing spoke to US health analytics firm Amara Health Analytics about the role of big data in hospital alert systems, and also to UK provider FlyingBinary about progress in the UK.
Sepsis is the body's toxic response to infections in the bloodstream. The condition kills millions of people every year but mortality and morbidity rates can be substantially reduced by early treatment with antibiotics. The earlier the condition is detected, the better the chance of success.
Amara's analytics solution continuously ingests data streams from hospital systems that monitor heart-rate, temperature, respiratory function and other physiological factors and combines that with doctors' notes, lab and pharmacy data, operative reports and discharge summaries to predict the onset of sepsis.
"It's predictive modelling applied to clinical analytics. It continuously analyses all the data associated with each patient in a hospital," said CEO Steve Nathan.
"When a certain threshold is met then an alert is sent to the clinician right into their smartphone so they can pay immediate attention to that patient."
There are well-known medical guidelines for the diagnosis of patients who are becoming septic. Typically, Nathan said, these are based on a number of metrics such as heart rate and white blood cell count.
"You can think of a system that focuses on those five or 25 variables as a knowledge-based system. The best human knowledge on that subject has been captured in guidelines and clinical studies."
Amara uses a combination of these guidelines, real-time machine data from monitors, lab results and doctors' notes to improve on the standard alert systems using machine learning.
"What we have found is that significantly better results are obtained if a big data approach is taken. Machine learning operates by having an initial training set of data from which the machine learns to find patterns in a much greater data set that are predictive of the outcome you want to achieve," he explained.
"We've discovered 120 additional variables beyond the initial five or 25 in the standard guidelines. Among those extra 120 variables, and combinations of those, we find a lot more predictive signal that gives us higher level of earliness of alerts and a very low false positive rate."
Because of variations in the way records are kept across different hospitals and laboratories, a system that can cope with inconsistencies on the fly is a must. Amara's software is based on the DataStax Cassandra NoSQL database with Apache Solr for free-text searching of the notes and other unstructured data.
"The NoSQL capability of having flexible data representations with dynamic schema so you don't have to know everything about the data at schema definition time was very important to us," Nathan said.
"Also you get incredible variance in the way that clinicians express things. That variation we handle in our natural language processing pipeline, which sits behind Solr and Cassandra."
Amara plans to apply its predictive systems to other conditions in addition to sepsis, looking initially at all-condition patient deterioration and then possibly at cancer analytics. The firm aims to expand into Europe via a partnership with reseller CapsuleTech.
So will we be seeing Amara or similar systems used by the NHS? Jacqui Taylor, co-founder and CEO of web science company FlyingBinary, says that such systems are here already. With solutions running on its DataStax Enterprise platform, her company offers a number of big data analytics and visualisation services to the NHS and other government bodies through G-Cloud. However, she said, implementation of advanced services is sometimes hidebound by existing contracts.
"Where we're not using the always-on service that Amara are able to use it's actually to do with the legacy problem. [NHS commissioners] have legacy contracts they can't get out of," Taylor said.
In many cases, however, they cannot afford to wait until these contracts end, so they need to be able to make the most of what is already in place.
"They need to start making transformational changes now. So what we've done is to provide legacy platforms that allow them to do that now but that feed them towards the always-on service when that becomes available to them. It's a pragmatic approach."
By transformational change Taylor means moving from the current situation where eight per cent of employees "have access to and are enabled with data" to a figure of 80 per cent. This would enable data-based decision making at every level.
"Within the NHS we are enabling domain experts - clinicians, management boards, commissioners - to unlock the value of the legacy data within the system because they can't change the system itself. You can add other things in, such as streaming data, so that when they are able to unlock the legacy they can move to the always-on service".
FlyingBinary is currently working on having its own predictive analytics system adopted across the NHS.
"With G-Cloud 5 in April we stood up the equivalent of what Amara has done and we're having conversations with the NHS about how we can bring that innovation in," Taylor said.
Sometimes, the power of the mainframe is the most cost effective answer. Computing's Peter Gothard puts Computing's readers' questions on the future of the mainframe to IBM's Z13 expert Steven Dickens.
This Dummies white paper will help you better understand business process management (BPM)