IABtalks: February 23rd Industry Presentation & Networking Event | University of Utah School of Medicine | Department of Biomedical Informatics

February 23, 2018
University of  Utah, School of Medicine
Department of Biomedical Informatics

Theme: Natural Language Processing

 Keynote Presentation by Gerasimos Petratos, CEO and Founder of Hiteks Solutions, Inc.

It is great to hear about all the applications of NLP happening amongst the students and in industry, I remember when I was in the DBMI program here in 2001 that NLP was identified as one of the enabling technologies for Health IT; in fact, one my Master’s thesis recommendations was to apply NLP to improve sensitivity and specificity of adverse drug event detection. Why Natural Language in Medicine? Did you know that human language in general can account for an almost infinite amount of ways to describe our outside world? So for clinical language, mainly derived from Greek and Latin, to describe our biological and medical worlds – so to speak – we have borrowed from our broader language system and added clinical terminologies with millions of clinical terms, lab tests, and orderable drugs which amounts to trillions of clinical combinations, clearly more than we can accommodate through picklists and drop-down menus.

I predict that commercial NLP technologies in healthcare will become THE single transformative technology over the next 10 years because of its ability to organize information to support the 4 main Health IT technology domains: Workflow, Analytics, Content, and Mobile: WACM. I’d like to share with you how I came into this field, commercializing an NLP platform that my team built to address the needs of our early clients in the Workflow and Analytics domains

After the DBMI program I entered into Pharmaceutical Medicine, where I quickly learned that human assignment of medical codes could result in a drug being marketed as non-drowsy, even though the reported trial data showed somnolence; and realizing that suppression of safety information once a drug is already on the market can lead to favorable prescription writing. The type of data that was being used included very structured clinical trial data, some EHR data, and insurance claims and I knew that a complete picture of how the patients being treated with our drugs was lacking, along with ethical issues I uncovered, so in 2011 I decided to leave the industry and focus on the exciting changes happening in the provider world. I started Hiteks and assembled a team in whom I was confident could reveal insight in what was actually being documented in the patient charts without having to use nurses and administrative resources to read through them all, sort of an enhanced chart review.
I remember Scott Evans saying, “when you provide complete and timely information to physicians, they can make the right decision at the point of care over 99% of the time”; and I remember Brent James showing us how “Core principles of clinical quality improvement require that you manage what you measure”. So I knew that identifying relevant clinical outcomes and presenting them at the point of care for decision-making should be the goal of my young company.

Decision support through analytics, long a specialization of Utah’s Biomedical Informatics community, is the next frontier for mainstream medicine, and we’ve all heard about the big announcements by Apple and others to help consumers share their data with their providers who will be able to accelerate point of care decision-making. One of the fundamental challenges is how to actually get complete and timely information out of and back into the commercially prevalent EHRs which requires robust and transparent interfaces; the time isn’t quite right because the new FHIR interfaces are still being improved, but it’s close.

So what we did is work with a large health system to feed clinical data into our Insight NLP platform and prototyped a tool that generated this patient summary, and launched it at HIMSS in 2013. The patient summary gave a snapshot of each patient by summarizing the content of all their clinical documentation so that a clinician who glances at the summary knows what the main problems, medications and procedures are. That’s when we got the attention of Epic, who told us that they had unsuccessfully tried working with other vendors on this functionality, and asked us if we would be willing to integrate our application with a new activity within their 2014 version; we knew we were on to something

You see, the current sentiment by health system providers is that the back-office coding and quality reviews need to be moved to the front-end directly to the clinicians to address in a timely manner before they move on to the next patient in their workflow, something that Epic had heard about in the 2 years leading up to that HIMSS conference; It was our shared vision that we could use this type of workflow enhancement to improve problem and medication list management and later CDI – Clinical Documentation Improvement, which enhances documentation, revenues and physician satisfaction, as long as it could be done in real-time. Many new Health IT ventures try to take on too much in terms of featuring new functionality; remember Google Health and its promise to take over the centralization of patient-controlled records by providing not only a hub to collect the records, but also a consent and authorization process to get patients to request the records from their sources?  Google tried to tackle too much and couldn’t sell their idea effectively enough to either consumers nor the source systems that were supposed to automatically transfer data; actually there is no AUTOMATIC in Health IT, it needs to be programmed; Google was way too early and lacked device integration which others are seeing now as ways to on-board consumers.

At Hiteks we’ve constantly tried to push the envelope of innovation, but the uptake only started when we focused on a single differentiator: Speed.  We already had configurability of the NLP engine to enhance accuracy through iterative testing, but our success started when we were able to produce that patient summary in less than 1 second;  And for retrospective analytics were able to process millions of records in a few hours, something that other NLP vendors are not able to do.

Incremental advances are more easily adopted than major breakthrough innovation because it seems that our brains are architected to process information to learn novel and creative ideas in different ways than routine patterns, as described in a recently published book on this topic by Dr. Elkhonon Goldberg called “Creativity: The Human Brain in the Age of Innovation”. Novel information and long-term memory work differently than routine information and short term memory. Knowing this, when you create an innovation and try to commercialize it, remember to position it with what is existing and known, and add in a single descriptor of the difference which ties into a financial benefit

Applying this approach to healthcare IT branding, we can help innovative ideas from being accepted more easily to become part of mainstream medicine, rather than waiting the many years for an innovation to become adopted by relying on moving parts lining up; the late Allan Pryor, one of the original contributors to the HELP system here in Utah lamented to me that his innovation of being the first to automatically interpret EKGs would have been better served as a commercial venture if instead of focusing on the more advanced 12-lead system which only now has reached market penetration, he could have analyzed the 6-lead EKG which was the mainstream

Health IT technology adoption that can impact clinical decision making for higher quality patient care, can be accelerated by how we describe new technology to decision makers, using features that are similar enough to what people already know so that the innovation can be placed into context so that it is tagged as something special by the brain, with the 1 unique characteristic that differentiates it for the activation of the novel-detection.

I remember when I was growing up in New York and my mom would always compare me to my best friend and next-door neighbor Teddy.  Teddy you see was a nice Greek kid- like me- going to the same school but a few years older.  So my mom used him to tell me that I needed to mimic his good traits and behaviors.  “Why can’t you just be more like Teddy?” she said when I wasn’t focusing on my work or caught fooling around, because she heard that Teddy was spending hours at a time on his computer.  What she didn’t realize is that Teddy was playing computer games on his new Commodore64, so when I found that out, I asked for my own computer.  That fueled the start of the Flight Simulator Wars of the mid-80s, and my mom realized her criticality failed when she saw me spending thousands of hours on the computer playing these games. 

Unsurprisingly when I asked to participate in Teddy’s more shady teenage activities I was told to “find a way to be different”, making me realize that I really needed to find another way to differentiate, and it wasn’t by being better at video games than Teddy.  I remembered another neighbor who we called Frankie Long Pants being different because he was the only kid in the neighborhood who wore jeans in the middle of the humid summer; and the mobster movies that always had characters named after their first name and their unique differentiator; unfortunately, before I could differentiate myself, the neighborhood kids had seen the Mikey likes it commercial and because my Greek nickname is Maki from Gerasimaki, they started branding me Maki Likes It…and it took me many years to rebrand myself after that!

If we look at how some recent HealthIT companies have penetrated the market with their differentiation, we need to look no further than the Salt Lake valley:  Health Catalyst has successfully marketed a data warehousing approach that combines traditional elements with some unique algorithms in what they call a “Late Binding” approach; the ability to take an existing set of product features such as data warehousing, and add in 1 key differentiator like “Late Binding”, can result in huge successes in market adoption; the late Pierre Pincetl determined that the best approach to EHR selection was simply adding a “Physician Selection Committee” approach to the decision-making process which ensured that the revenue centers of a health system became stakeholders;  at Hiteks we developed our “Air Binding” approach to NLP pattern recognition which applies an algorithm to our Lexicon and finds hits in clinical documentation in subseconds, allowing us to compete more effectively by being 100 times faster at producing physician CDI suggestions than our competitors.

Novelty in new product functionality requires people to first identify what's normal or mainstream about an idea, which trains the recipient’s brain networks to identify patterns of features they are familiar with, and then introducing that one new and hopefully unique feature that sets it apart, thus training the creative side of the brain with something new and appealing.

Unconsciously we seek to understand what’s novel and what is every day; I notice now that with social media both kids and adults spend hours on their phones.  If you analyze what’s going on here, they are just getting comfortable with what’s normal in their world: Sometimes the posts will show what people are having for dinner, but sometimes they realize something is novel and share it.

The point I’m trying to make is that in your Health IT endeavors to produce innovation, try not to do too much because it will become counter-productive to the goal of getting something out there in use to improve patient care or make some of our complicated processes more efficient; instead focus on the 1 thing that is truly differentiating compared to standard practice

Remember how Utah’s Biomedical Informatics Program was one of 10 or 15 programs around the country similar to it, but it’s the one that had the real-world experience to use clinical data from Intermountain and the U, and implement new ideas in a care setting that the others lacked?  When Homer Warner started the department he was fresh off of his cardiology research where he measured Blood Pressures in the Cath Lab, giving him recognition in Informatics approaches in Cardiology; and then Reed Gardner was able to define the requirements and improve on bedside monitoring through the Medical Information Bus, giving him a recognition in innovative engineering with ICU monitoring.  This focus on choosing 1 feature to improve on is differentiating, and has allowed us to be here today.

I’m proud of my own team’s ability to focus on this single differentiator, because of our Solutions’ Speed, we’re helping millions of patients with safety surveillance in quality areas such as Sepsis and soon Stroke detection, and thousands of physicians with workflow and financial reimbursement through the NoteReader family of activities in Epic.

I took my mom’s constant criticality of asking me to differentiate from my peers to fuel my desire to contribute to the field of Biomedical Informatics by helping advance a single but important innovation into the commercial Health IT world, which is Real-time Insight into a Patient’s Chart so that all downstream interests of decision support, utilization monitoring, clinical research, safety surveillance and much more could be accomplished.  Thanks Mom, Maki actually does like it!


Popular posts from this blog