Machine Based Learning – Part III of A Brave New Health Care
This is a mini-series about technologies that are radically changing healthcare. In each part I preview how a technology came about, how it is being used and how it will evolve, illustrated by my conversations with some of the health care entrepreneurs bringing this technology to life.
Part III – Machine Based Learning
Guest: Gary Velasquez – CEO of Cogitativo
In 1773 Charles Messier first described the famous Whirlpool Galaxy in a simple hand sketch. Centuries later, we can now see intricacies of the galaxy’s spiral arms studded with majestic blue purple and red with the Hubble Telescope.
What if we can dig deep into healthcare behaviors with the same granularity? What if a hospital can save millions in spending by identifying one doctor’s behavior in using expensive instruments in an OR? Or if an insurance company can identify stage 4 cancer patients who are having trouble using their benefits? Even better yet, what if we can compare cost and quality variability for say, a Total Knee Replacement surgery?
A group of highly talented individuals, predominantly Mathematics doctorates, are building a “Hubble Telescope for healthcare behaviors”.
The idea is to collect datasets and feed it into a customized algorithm that can use real-time updates to learn patterns, predict gaps and recommend a more efficient healthcare. Pretty neat, huh?
“I spent the last 30 years trying to come up with the perfect if/then statement for healthcare that would make things snap together. The epiphany I had 3 years ago was that healthcare is not a rules-based system, it’s all about the behaviors of everyone involved in care delivery.”
Gary founded Cogitativo (co-hee-tee-tivo), to analyze healthcare as a set of behaviors rather than a set of rules. They use a mixture of domain expertise and machine based learning, to quantify healthcare behaviors and predict them in the future.
Healthcare As a Set of Behaviors
When the patient comes and presents with their symptoms, the physician has a specific intent for the care plan, or a signal. The problem is that clinical and administrative behavioral “noise”, takes this signal off course.
“It all started with the idea of finding a signal in a bunch of noise, just like how they used the radar in WWII to differentiate between a flock of seagulls and German bombers.”
Inspired by the guys behind DARPA’s signal processing, Gary found himself thinking how could we quantify the behaviors? They took the administrative and clinical physician order systems (CPOE), linked it to EMRs, and link that to coding and billing.
“I can then tell you exactly what is going on, and almost every time there is a human involved, there is a variability linked to behavior.” Gary explained “ So I can tell you exactly what’s happening, what’s causing that variability, who did it and why they did it.”
During a study Gary’s team was doing for a big hospital in California, they went into one OR room and talked to the surgeon. They realized that this OR was set up differently than all the others.
“The doctor had to use his tools differently, and that cost the hospital around several million dollars in reimbursement.” Gary recounted.
The Surprising & Ridiculous Variability of Healthcare
One of the biggest problems in US healthcare is the lack of transparency and the ability of large research hospitals to mark up their prices several folds of what services cost.
“Feeding systems with data, can eventually make these discrepancies visible and allow people to identify what is going on.” Gary explained.
Let’s say I hurt my knee tomorrow, get arthroscopic surgery and I’m billed 20,000 bucks. I would assume that the price change across hospitals would be a normal distribution graph, looking like the Seattle space needle. I would also think that the price varies with quality of care, just like any other industry. In reality, neither is true.
When you look at the data, it can be tri-modal or quad-modal looking like the tip tops of a bunch of mountains. There is nothing normal about this distribution. The same could probably be said about the quality of care.
Accountable Care, you said?:
Though 1.2 Million newly insured people now have access to care, there is confusion and a lack of guidance on how to use these new benefits. Gary thinks evolving and self-learning data systems can help identify those who need help.
“What Obama did was take the healthcare snow globe, which everyone understood as static, and he shook it up.” Gary said, as he spun an imaginary sphere in the air. “The more insights we can give to the stakeholders that can bring these closer, the better off Americans will be. It’s not good having them seek care without guidance.”
The Cogitativo team identified 10,000 newly insured patients with stage IV cancer that are going around California, with no clue about how or where to receive the care they just got access to!
“Why would a prostate cancer patient, living in LA who go to MD Anderson or Cleveland Clinic rather than UCLA or USC?! I mean, they’re lost! They don’t have someone guiding them to the delivery system.” Gary exclaimed as he jumped to the edge of his seat.
The things is, it isn’t the payer’s fault either. With an unprecedented 60,000 new members through Covered California, there was an inevitable large amount of variability. Things have to be done in a different way, but the whole system is still figuring it out.
The good news is that there are ways to apply predictive engines to decrease just that.
Where do we go from here?
The overused and abused term “Big Data” is nothing novel in the field of Healthcare. The data itself has been pretty big for at least 30-40 years, but there was no affordable way of extracting it from feet-long papers and disc drives with platters.
Now the technology isn’t just here, but it is also cheaper than ever before. These powerful predictive data-driven machines can be the link between our healthcare behaviors and how we use our other technologies to change them.
We would have the luxury of asking ourselves: What are the technologies we can bring to this group of people? Can we use Telemedicine or Home-Based care? These are all avenues we can explore after identifying the problem and need by machine-based learning.
After that turning point in Gary’s career he went back to the executives he worked with over 30 years and explained the existence of a Hubble Telescope for health care behaviors.
“You have a moral imperative” he told them, “You can’t come out like the CEO of GM and say I couldn’t know. The capability of knowing…is here.”
“So the questions I pose to you is, what are you going to do about it?”
Other articles in series:
Omar Shaker completed medical school in Egypt, followed by internships in the US. He soon left primary care for the world of digital health, moving to San Francisco to work on his own projects. These posts represent his reflections on a series of interviews he conducted with some of the more exciting entrepreneurs working in digital health today. Omar can be reached at firstname.lastname@example.org.