The mashup approach: How healthcare can save billions on AI and machine learning
Healthcare is at a two-tined fork: One strip leads to repeating the same mistakes others have already made while the more enlightened rail learns from those instead.
That avenue is not a foregone conclusion. Hospitals’ development and implementation of emerging technologies under the artificial intelligence, cognitive computing and machine learning rubric is nascent enough that there’s time to choose which tine to take.
Taking the well-trodden path will neither be easy nor exactly inexpensive, so why pick it?
There’s a lot to be gained — and piles of money to be saved — by learning from those who have gone before.
AI: Dotcom boom then bust all over again?
Artificial intelligence appears to be unfolding along the same lines as the dotcom boom and eventual bust while hospitals and non-healthcare entities alike deploy sexy new technologies for technology’s sake, instead of solving actual business problems.
That means healthcare as an industry is effectively poised to waste considerable sums of money making the same mistakes other sectors have already made with AI, machine learning and cognitive computing.
“Billions of dollars will be lost learning these lessons,” said Leonard D’Avolio, CEO of machine learning company Cyft. “So many of these have already been learned. The question is can we pass them on to people to help avoid the same mistakes?”
That starts with knowing where to look. Retail and banking are the usual suspects, but looking at just those industries oversimplifies the challenge in healthcare because their data is clean and the desired outcomes are relatively straightforward, said John Showalter, MD, Chief Product Officer of Jvion, a clinical cognitive science specialist.
“We’re going to need mashups. Many of the people who believe AI can really benefit humanity have picked an industry and they’re trying to overlay that onto healthcare,” Showalter said. “To get best practices we’re going to need to do mashups.”
Where to pluck best practices
Here’s the hard part: There is no single set of readily available and easily applicable AI best practices hospitals can plug-in, implement the emerging technologies and simply move forward. Instead, hospital IT shops and healthcare technologists can learn a lot from the automotive, airline, business process management, and search sectors.
The automotive industry, for instance, holds many parallels to healthcare. Dale Sanders, President of Technology at analytics vendor Health Catalyst, pointed as chief among those to massive cultural changes, the adoption of large-scale systems including ERP software that was disappointing in many ways (EHRs anyone?) and collaborative standards development to effectively compete on innovation and brand.
“Digitization of automobiles, digitization of the processes of building and maintaining automobiles, preventive maintenance technology for cars vs. for humans, there’s also that sort of thing,” Sanders said. “CRM concepts, tailored advertising according to the consumer profile. Same thing applies to patient engagement.”
Read more Innovation Pulse columns from Healthcare IT News.
Likewise, airlines managed to digitize themselves while still flying people all over the globe.
“I'd argue, and I have argued this for many years, the single most important industry for healthcare to import ideas and technologies from is the business process management industry,” said Charles Webster, MD, President of EHR Workflow, Inc., who has a Masters in Intelligent Systems.
Webster added that BPM vendors have learned to execute workflows in the real world where differences between desired and actual behavior can systematically drive improvements.
“This is the key to avoiding analysis paralysis and standing up systems quickly, something BPM vendors excel at compared to traditional health IT software development practices,” Webster added.
Showalter said that in thinking about which industries to emulate, he became surprised that search is not more frequently held up as a prime example that healthcare should emulate.
“In the search industry massive amounts of data have to be processed to identify the most relevant sites, songs, videos, documents, based on unclear inputs,” Showalter said. “This is similar to the need to look at an entire medical record and identify the most relevant factors based on limited and often dirty inputs.”
Showalter also agreed with Sanders that there are a number of parallels with automotive, notably that busy clinicians can at times be akin to distracted drivers much like distracted driving raises the danger of accidents, clinicians who are otherwise occupied are more likely to make medical errors; both can be fatal.
“Automotive safety has focused on incorporating alerts into the ‘workflow’ of driving. Collision alerts flash directly in the line of sight and only when there is significant risk. Lane crossing alerts beep with increasing frequency,” Showalter said.
What’s more, car manufacturers have different degrees of alerts. Rear lights firing up when a car slows down are fairly sensitive, as are lane warnings, but airbags are another matter and they can’t just deploy every time a driver hits the brakes or they’d cause accidents instead of avoiding injury.
“A mashup of the AI in search and automotive safety would result in the ability to cull through mountains of EHR data, identify the most relevant features, predict risk early enough to take action to avoid an error and integrate it within existing workflows,” Showalter said.
Mistakes hospitals must avoid
There’s that word again: workflow. It’s simultaneously one of the most significant recurring problem areas and among the biggest opportunities ahead for hospitals.
Showalter said one of the clearest mistakes to avoid is not integrating AI and machine learning within existing workflows. “We do this all the time,” he added. “We’re probably wasting billions on it now.”
The ‘I’ve done this and run analysis on it but nobody else knows,’ approach is another misstep to avoid. Take sepsis, for instance. When an intelligent system identifies a patient as high-risk, then the workflow needs to be tuned to inform the right clinicians at appropriate times and deliver only the information they can actually use right then.
“Integrating with workflow is about more than EHRs,” Showalter said. “Pager systems, phones, and other communications tools can be the best workflow.”
Webster added that the BPM industry has made past errors that mostly, in retrospect, were limitations of early software iterations but has since addressed those — and has essentially already invented the workflow wheel of sorts for AI, machine learning and other intelligent systems.
“Lots of health IT vendors are creating their own proprietary crappy workflow engines, instead of using good ones from the BPM industry,” Webster added. “Some current health IT vendors are trying to add BPM-like capacities to their legacy software, but this is like trying to add a foundation to a house or a hull to a ship.”
Another blunder is to not fully grasp the available technologies relative to business problems that need to be solved.
“How can any quality improvement professional, clinician, pop health manager, how can anyone make reasonable judgments and decisions if they don’t understand at a practical level what are the tools good for, what are they not good for?” Cyft’s D’Avolio said. “What are the right problems to solve?”
What has to happen
Healthcare is not blazing a new trail here. Amazon, Facebook, Google, Pandora, Spotify have mastered how to incorporate learning and new ways of conducting business into everything they do.
“In healthcare we have a tendency to be insular,” D’Avolio said. “We accept this notion that healthcare is so different it cannot learn from other fields but that’s not true when it comes to best practices of making AI work.”
The time has come to cut through the hype around artificial intelligence, machine learning and cognitive computing, figure out what works and what does not, institute already existing best practices, and answer a short list of critical questions.
“You have to do two things: Digitization of patients and digitization of the process, such as clinical encounters, diagnostic testing, billing,” Health Catalyst’s Sanders said. “For the most part, we have not digitized the patient and we've digitized the process of a clinical encounter to drop a bill, not the process of ensuring high outcomes and low costs.”
Indeed, also missing are the critical success factors that clearly highlight the answers to these greatest opportunities.
“We can learn from other industries,” D’Avolio said. “It will be the people who can see through the hype and learn the right ways to put technologies to use who are going to capitalize.”
How AI is driving forward-looking healthcare orgs.