When someone has suffered an irreversibly life-altering event, such as a traumatic brain injury, predictive maintenance of that impaired state is the best we can hope for. Of necessity, people with traumatic brain injuries must be kept under constant monitoring. In most intensive care units (ICUs), caregivers are only alerted when the patient’s brain pressure crosses a critical threshold. At that point, an instant decision must be made to determine if the alarm is false, if the condition is life-threatening, or if immediate action is needed to prevent brain damage or death. Introduction of predictive responses into the ICU has obvious life-saving potential.
I hosted a Twitter chat with UCLA and Excel Medical Electronics that discussed how big data analytics can help predict dangerous brain pressure to prevent further damage. Joining me were:
- Dr. Neil A. Martin, MD, FAANS Professor, ICU informatics researcher & W. Eugene Stern Chair, Dept. of Neurosurgery at UCLA (@UCLAHealth). Dr. Martin specializes in treatment of brain trauma, stroke, brain aneurysms and brain tumors.
- Dr. Xiao Hu, Ph.D., Associate Professor, Dept. of Neurosurgery at UCLA, director of neural systems and dynamics lab online (@UCLAHealth #drhu). Dr. Hu’s interest and expertise are in data driven real-time decision support for patient care in hospital; he develops advanced signal processing and machine learning algorithms for early detection of patient crisis.
- Lance Burton, General Manager at Excel Medical Electronics (@Excel_Medical), which provides BedMasterEx, a solution for acquiring and storing complex physiological data acquired from hospital patient monitoring networks and medical devices.
Drs. Martin and Hu discussed their research into traumatic brain injuries. In particular, they highlighted how they are applying big data analytics in their research, leveraging the data acquired from bedside monitors and using predictive alarms and decision support tools for physicians and other caregivers in ICUs. Here are highlights from the discussion. Note that I’ve placed edited versions of the questions that I and other tweeters asked, and I’ve correlated the doctors’ tweeted responses (without identifying which doctor tweeted what) with the questions so as to call out the conversational flow of the Twitter chat (trying to follow it all in real-time with the naked eye is not advisable):
How will your research help patients with traumatic brain injuries?
- Smarter efficient ICU care. Proactive treatment. Get patients better & home faster.
Tell us more about the kind of data on traumatic brain injury patients that you are getting from bedside monitors at the ICU at UCLA?
- We use streams of data, ECG, EEG, intracranial press., to show early changes in status.
- One such example of changes is brain pressure elevation, others are seizures, loss of blood flow regulation.
- For clinical forecasting of patient status, and early warning of impending crises.
- Data streams contain ECG, blood pressure, brain pressure, brain electrical activities -EEG.
- We are tracking outcomes such as number of nuisance alarms, number of code blue events etc.
- We will be down-playing monitors just as sensors and depended on additional analytics that have access to EMR data for alarms.
- All digital, each sampled is a short integer and sampled at 240 hz.
- Unstructured data to come.
- But modern EHRs are more now using structured data entry elements.
- Medical history exists now as unstructured textual data, so not used in the algorithms yet.
Do you do real-time analytics directly on analog waveforms or do you convert them to a digital representation for processing?
- Waveform data are actually unstructured content as well! You cannot describe them in a relational database. Pure data now.
- Demographics, lab results, monitor waveform parameters.
- So far, we have not used textual data such as notes but we use them to build patient cohort for offline analysis.
- In the future - imaging parameters, genomic and proteomic and metabolomic data.
How much real-time data is involved?
- For a given patient 300mb/day from GE monitors, up to 2.7Gb per day for EEG, and up to 700 alarms per day
Tell us about predictive alarming and how it can supplement historical data in treatment of traumatic brain injury patients in the ICU at UCLA.
- Clinicians check med history at very first encounter.
- VisualAlarm provides a history of alarms for a given patients using event bars and sparklines for trending.
- Med [history] is vital. Can be structured data as ICD-coded comorbidities. New EHR allow this.
- BedMaster recording app from Excel Medical archives all alarms for entire hospital stay.
- BedMaster app lets us archive and easily review all alarms and waveforms. It is key.
- While monitors only announce the current alarms, one by one. So nurses do not have a context of what is going on.
- We have developed a solution for visualizing alarms from patient monitors. This will be tested soon at a cardiac ICU.
- It is a software medical device, prevent secondary insult to the brain such as high brain pressure, lack of blood flow if we can alert clinicians early on.
- We will be down-playing monitors just as sensors and depend on additional analytics that have access to EMR data for alarms.
- First, the predictions are for specific conditions, so they are more actionable compared to conventional monitors alarms.
What sorts of professionals are developing advanced analytics for monitoring and treatment of traumatic brain injury patients in the ICU at UCLA?
- Engineers like us at UCLA work very closely with drs and nurses to formulate problem and find solutions.
- We have computer scientist and electrical engineers in my lab. we also collaborate with biostats folks.
- Subject matter experts’ insight and vision are key.
- In fact, we engineers all have to learn subject matters as well. I am learning reading ECG for example
How is real-time analytics on traumatic brain injury patients presented to physicians and caregivers in the ICU at UCLA?
- Real time data analytics = key. Better communic and collab among all Drs, Nrses equally key.
- Drs & nurses want clinical analysis and prediction.
- Clinical prediction, outcomes meaningful for the patient, quality, value.
- Hospitals want quality, cost parameters.
- The great leap forward comes when clinical data is joined to quality, efficiency, cost data.
- Those are the goals. in this way, we can leverage multi-variate streams as well as additional information from lab tests, notes, etc.
- Clinical+quality+efficiency+cost. HRI. We live it every day.
- HRI = Hlthcare Reform from the Inside. Drs, nurses, researchers have to reinvent Hlthcare.
Is this solution accessible, securely, to wireless/mobile devices carried by caregivers in ICU setting?
- We rely on our hospital wireless /network infrastructure.
- Most cases, similar sensors and hence signals are acquired for out-hospital settings, so directly applicable.
- Signal quality may be one issue to address in out-hospital settings.
- More people are instrumented outside hospital at home etc., the real-time analytics is needed for such applications.
Are these real-time decision support informatics tools only for UCLA, or is the tech available to other institutions?
Lance Burton of Excel Medical Electronics described his company’s role in providing informatics for real-time treatment of brain trauma patients:
- Do trials at UCLA and our consortium members, share analytics to increase value of the platform.
- We have widely published our algorithms.
- Other places can definitely replicate it but there are transnational barriers even within our ICU to bring our algorithm into hands of clincians.
- We develop algorithms for insightful data analysis, to use in immense streams of real time data.
- There are specific algorithms for brain but we have algorithms for other conditions such as cardiac arrest as well.
- Commercialization of this and the labeling of it as a diagnostic will need FDA approval. It is a software medical device.
Lots of clinical data being generated in hospitals most of it relatively unstudied
@UCLAHealth #research tackling most complex
Several collaborations forming with
@UCLAHealth and other @Excel_Medical customers to progress this type of research
- Anticipate progression from academic centers to community hospitals to home health
#BedMasterEx installed in many academic hospitals.
FDA is encouraging predictive v reactive medicine large clinical trials needed such as
- Strong bond between computer science/machine learning and thought leading clinicians
- Two major tech barriers: 1 ability to capture streaming complex data 2 tools to analyze massive amounts of unstructured data
- Lack of standards, complexity, volume and tools needed to capture and analyze.
Continue the discussion & check out these resources
Watch this video that gives another viewpoint of how UCLA tackles brain trauma:
I want to give a shout-out to three tweeters – LeAnna J. Carey (@thehealthmaven), Hurt Blogger (@HurtBlogger), Shelly Lucas (@pisarose) – who virtually assisted me in eliciting these informative responses to our expert panelists. Great job!
If you’re interested in continuing this discussion, leave comments on this blog below.
- Read this press release discussing how UCLA is relying on IBM’s breakthrough big-data technology to help patients with traumatic brain injuries.
- See more information about UCLA Neurosurgery, UCLA Neurosurgery Neural Systems and Dynamics Laboratory (NSDL), and UCLA Neurosurgery’s Clinical Quality Program.
- Learn about Global Care Quest, a provider of visual clinical intelligence to surgeons and clinicians, in which Dr. Martin, is a principal.
- Learn about Excel Medical Electronics.
- Visit our Healthcare page for additional product information, white papers, videos and more
- Download the white paper "Data-driven Healthcare"