What the Health! A Blog from ArmadaHealth

Emotional (Rescue) AI

Written by Teresa O'Keefe | Feb 19, 2019 2:17:00 PM

3 minute read

Can Emotional AI rescue us from the imminent invasion of zombie bots!?

Every time I have a hallway conversation with Bharath (we call him “SuperB”), I wish I had a recorder because nearly every exchange could be a blog post.  


The next wave of innovation in AI is set to be fueled by creating a digital heart. ArmadaHealth seeks to inject empathy back into technology to create a more valuable interaction between physicians and patients.”

– Bharath Sudharsan, Director of Data Science, ArmadaHealth 

 

The subject of this post is a conversation we started last year about Emotional AI. The topic was in “skunk works” here at ArmadaHealth and ultimately leveraged Google’s BERT sentiment analysis model which was recently released into the public domain (thanks, Google!). The results have been quite fascinating and SuperB just presented some of the team’s findings at the H20 conference in San Francisco. The subject couldn’t be more timely, as validated by Dr. Eric Topol’s soon-to-be-released new book, Deep Medicine: How Artificial Intelligence Can Revolutionize Health Care — And Make It More Humane.

Emotional AI presentation by our Director of Data Science delivered at H2O World
2019 in San Francisco


To give you some background, we here at ArmadaHealth have developed a multivariate, predictive model using techniques from data science, such as natural language processing (NLP) and machine learning, that help an ArmadaHealth Navigator (or your Care Coordinator) match real-live patients to real-live physicians. The net/net is patient-centric, evidence-based, AI-assisted access to providers. We refer to it as informed access for brevity, as opposed to uninformed access, which is currently the state of the industry. We'll be writing more about this subject in future posts. As you might imagine, this is a complex, multi-dimensional initiative.


Herein Lies the Problem
At the core of what we do, we concentrate on defining the optimal physician profile to treat every known condition or diagnosis. In order to do this, we leverage all available medical literature, research and dozens of subject matter experts and clinicians who are the top in their field for expertise and quality. Many of these experts are members of our Clinical Advisory Board and add to our ever-growing, rich “knowledge graph”.

Next, we apply actual, real-live physician data to the model to predict which real-live physicians have a “sweet spot” for treating a specific diagnosis, for example: benign inter-cranial hypertension, lupus, long Q-T interval, pituitary tumor, scleroderma. Whoa! We’ll stop right there.

Herein lies the current problem: most people, including some physicians, wouldn’t even know what type of specialist or sub-specialist a patient would need to see for some of these diagnoses (hint: they would need more information and might be referred to multiple specialists).

This is a major undertaking! But we have accepted the challenge as our mission, that is, to provide AI-assisted physician recommendations that meet our members’ needs including: proximity, in-network (coverage) and appointment availability, among others. Furthermore, we are steering the right patients to physicians that are qualified and want to treat these conditions, more than others, as they have developed a passion and expertise in these specific areas of medicine.


Healthcare is Messy
I deliberately use the term “real-live” because we are talking about people: a patient’s health and a physician’s practice. Human health, healthcare data and provider reviews are, frankly, a messy business.

Until the advent of modern data science, and the availability of voluminous physician-level data, no one could conceive of how to fix the problem of matching the right physician with the right patient. These advances enable us to provide a predictive model having many factors that weigh into the quality of a provider. We’ll continue to develop these models well into the future, but we are missing the mark if we don’t understand the patient better.

This is where Emotional AI enters the picture and where SuperB’s passion lies. If you are really interested in Emotional AI, you can use this link to get his book on Amazon: Can machines understand how we feel?: A design for Emotionally Intelligent Systems

Feel free to reach out to us if you would like to learn more about how we are scaling to reach millions of consumers using AI-assisted healthcare navigation. In the meantime, please view the video below from the H20 conference about the future of our service.

This is the first blog post of many to come that will focus on conversations about AI-assisted navigation as it applies to provider quality transparency in healthcare.