Neurologists at UCLA Medical Center in Los Angeles and IBM IBM -0.65% data scientists are working together to detect dangerous changes in inter-cranial pressure — pressure inside the skull – earlier than current diagnostics detect it, so patients can be helped sooner.
“Medicine has a number of sensor techniques that give you EKG, EEGs, and other wave forms,” said Nagui Halim, an IBM Fellow and Chief Architect of Big Data. “The industry has kind of gotten locked into that model — produce a very sensitive sensor and display the results in a siloed way on some instrument. Then people in the hospital walk in and look at inter cranial pressure and the respiration rate, then come back 20 minutes and make some notes.”
Today, these patients are under constant surveillance by bedside monitors measuring vital signs, but nurses are alerted by the bedside alarm only after brain pressure crosses a critical threshold.
UCLA had received a $1.2 million grant from the National Institute of Neurological Disease and Stroke to implement the predictive system before choosing IBM’s Big Data technology incorporated into bedside monitoring software from Excel Medical Electronics of Jupiter, FL.
IBM’s tool for medical monitoring, called IBM Infosphere Streams, was created to analyze a palette of signals. It can draw from multiple source such as EEK EKG, treatment records and any other patient-related data such as genomics, current state and history.
Then, said Halim, it produces analysis that tells you what is going on at different points in the timeline, history, and current condition leading to predictive analytics such as where the patient state is likely to be at a certain time in the future.
IBM has spent a decade developing the programming model which allows very sophisticated applications to be created.
“It remembers what happened, looks at trends and uses every mathematical technique you ever heard of,” Halim said.
“The idea is that in the health care industry, we partner with people who have deep expertise in specialized disciplines like treating brain injuries. They often don’t know about computing. We are computer people with expertise in mathematics and systems programs.” But, he added, IBM doesn’t have deep expertise in neurology, so the company leaves that to the doctors.
“You bring the two communities together — our experience with applications and analytics – with their particular medical domain. We ask what data is available and what are the key problems to solve. Our job is to bridge between the problems and the raw data.”
At the Ronald Reagan UCLA Medical Center, the medical staff was concerned with inter-cranial pressure spikes. They told the IBM team about the types of data they had and what they had looked at based on experience and intuition.
“We will take the data and apply analytics including machine learning or algorithmic classification. Now if we see something that is not normal, we can alert the medical staff
looking across modalities. We can see where early indicators are that something is not right, so the doctors can get it as early as possible” explained Halim. “Nothing happens suddenly in medicine, it only seems sudden. Conditions increase risk, certain factors are stressing the system, but it has antecedents.”
At a neonatal facility in Toronto, the IBM tools were able to spot serious health problems 24 hours before the symptoms were recognizable by traditional monitoring. Nurses there were highly skeptical, an IBM presenter said at a Big Data conference sponsored by UNCC in Charlotte, NC last year, but the results convinced them that computer monitoring could do something beyond what professionals’ observation could achieve.
“We get test data stats on many points and then we will do offline analysis with exploratory tools to look for trends and correlations,” explained Halim.
“At UCLA, we are going a step at a time,” he added. “We are taking work they had done offline and moving it to real time. Now we are in the real-time domain. The machinery is hooked up to live patients and it is monitoring and doing precise detection with a reduction in false positives.”
False positives had been a real problem in monitoring, he added.
“It is amazing how effective these techniques are. We have been able in different setting to do things people didn’t think was possible.”
Susan Connors, president and CEO of the Brain Injury Association of America, said 1.7 million Americans sustain brain injuries each year. The leading cause is falls, followed by car crashes and then workplace injuries and assault.
“It’s not so much the initial blow that creates so much of the damage, but the swelling that happens afterwards which causes so much disability and impairment. The UCLA-IBM project can identify what is happening way before we can so physicians can take action sooner.” A key data sourse comes from Excel Medical Electronics which makes the BedMasterEx.
Brain injuries cost the country more than $70 million a year in care and lost wages, she added.
At UCLA, IBM has been working with a clinical team that has a sophisticated understand of technology, Halim said.
“For cranial pressure, the physicians have the algorithms. They have developed the techniques and we have helped by pairing their domain experts with our computer guys. My impressions is that it is working as well in real-time as it had been off-line, which is important because offline you have a lot of time to look at the data and analyze it. Now in ICU you now have the potential for getting the information as the patient is starting to suffer.”
Is this a big data problem?
That depends, said Halim. Big Data is not just about size or the three Vs — Volume, Velocity, Variety — he said, but also includes the sophistication of algorithms, and being able to use data in the situational context with the needed immediacy.
“A single patient might not generate that much data, but if you have video analysis and then a ward or an entire hospital and couple that with historical data, you are continuing to increase. Then if you do offline analysis for drug interactions, the problem starts to grow. You might start with modest volumes with tens of patients but over time you could do assessment of more signals from more patients. The human system itself is highly interlinked, so all the major system are interlinked, and its the interlinkages that give you the insight.”
IBM has worked with data across computer processor fabrication labs, seismic exploration, traditional manufacturing and financial services, among its may customers.
Halim said IBM researchers find a pain point in each industry.
“Once you start tackling that, it frees up their minds and they start thinking ahead and we find pent-up demand to get answers to other things. When you start to make progress, there is often a flood of the next round of questions and new ways of thinking about the business.”
In medicine, he said, the doctors who are aggressive in looking for help from technology are the visionaries.
“In all the places where I have been successful in getting technology used, you have to have some kind of visionary in there who gets it and wants to solve some particular hard problem, Halim said. “A lot of people are like that — entrepreneurial, curious — and we find them in organizations all over. We pair up with them and show what we can do, and they show us what their problems and concerns are. Then, in the best cases, we really do advance things. That is true in health care — we find collections of people who want to break the current trends and do better. They are very passionate about their own field, and they are usually the best or at the top of their field.”
He depends on early adopters who are not afraid of failure to push projects forward, he said. Once the leaders have demonstrated results, a community comes together to take the project further, partly motivated by a fear of being left behind.
“You see a change in tone from people saying that looks interesting to asking when can they themselves can get started,” Halim added.