How AI is reshaping patient engagement for consumerism in the post-COVID-19 future

Introduction
In work that started before COVID-19 spread to the U.S. and has accelerated since, two health care organizations discussed how each is advancing patient engagement strategies with artificial intelligence (AI).

Providence and Whiterabbit, in fact, are separately applying Amazon Prime-like thinking and tactics to the consumer experience. And while many providers are focusing on COVID-19 right now, for Providence and Whiterabbit (in conjunction with the RadNet radiology network), this work is simultaneously preparing for a future in which health care systems, much like the more advanced retail and financial services sectors, will interact with patients in a modern fashion.

This article, based on the Health Evolution AI-enabled Patient Engagement virtual gathering in which Martin and Mathur shared insights this week, will explore the following:

• • Why patient engagement needs retail methodologies
  • Where to begin AI-enabled patient engagement
  • Key elements of AI in patient engagement
  • Post-pandemic challenges and opportunities

Why patient engagement needs retail methodologies
Engaging patients is difficult for many reasons. Regulatory issues that inhibit data sharing to a certain extent, client-server and legacy architecture, and a lack of common practices in other sectors.  

“If you look at health care from the lens of the rest of our lives, it’s almost Soviet era,” Mathur said. “It’s archaic in terms of the way that a patient deals with the system.”

Rakesh added that retail notions such as same store sales growth, consumer experience and even simply understanding the total available market will be important to patient engagement in the future, which is coming quickly because of COVID-19’s impact.

Register for an on-demand recording of this webinar:

“At Amazon, we knew from early days that the most important person you serve is someone who has actually put something into a shopping cart at least once because they reached out and they believe in the brand,” Mathur said.

Providence’s Martin pointed to another organizing principle he learned at Amazon: Customers don’t care how complicated an organization’s business is and shouldn’t have to, so enterprises need to shield consumers from that complexity.

“The way Amazon ships products is very complex. But from the consumer standpoint, it’s very easy,” Martin said. “So we at Providence have spent a lot of time trying to simplify the experience for our consumers.”

Providence has obscured that complexity by building an integration layer so customers do not have to interface with the idiosyncrasies of back-end systems, such as EHRs. Instead, Providence patients interact with tools including Twistle for secure messaging and remote monitoring, the Xealth platform that delivers digital health prescriptions in the form of content and services, and its Grace chatbot, which Providence deployed two years ago and recently fine-tuned to screen people for COVID-19.

“You have to buffer the old technology with the new technology,” Martin said.

Where to begin AI-enabled patient engagement
When Martin joined Providence six years ago, he was surprised that the health system did not have the ability to acquire new patients online. Rather, people only became established patients by coming in for a clinic visit.

“That was one of the first things we solved for,” Martin said.

When considering where it could drive substantial overall impact moving forward, Providence elected primary care because it is the business function where it encounters the most patients.

Challenges to anticipate when undertaking AI-enabled patient engagement strategies are not altogether different from other industries, Martin said, but they are more extreme because of two key issues: the customer experience is worse and providers often see patients only once or twice a year.

“Anybody familiar with the internet and engagement generally knows that’s not a great statistic,” Martin said. “It needs to be 12 to 14 times a year on average.”

Interested in AI and ML? Register here to receive a copy of our CEO Guide to AI and ML.

Likening what RadNet is doing to Amazon Prime, Mathur explained that the intent is to entice people to shop either more frequently or for more products or services when they do shop.

“In radiology, our patients come to one clinic for an MRI, another clinic for a CT scan, someplace else for an X-ray. Mammography actually plays well in a consumer world, because a woman has choices,” Mathur said. “So we use the two levers of more modalities at one center and come more frequently for things that require compliance to a standard. And that by itself has dramatic impact on same store sales growth.”

Key elements of AI-enabled patient engagement
Important AI considerations relative to patient engagement include data, interoperability, privacy and patient consent.

• • Data. Two aspects of data to consider: Applying information to individuals and continuously collecting more data to make AI systems smarter. Whiterabbit and RadNet, using anonymized mammography data from Washington University in St. Louis, amassed millions of images to feed into its AI systems under the oversight of an internal review board. “We’re making it possible for images to be uploaded to our cloud. And on the app, the patient would get their images from visits across however many years they’ve been coming, as an image on their phone which would be lower resolution than the full high-density image,” Mathur said. He added the organizations need to protect sensitive data in transmission, including when its app delivers images, such as an ultrasound or a 12-year old’s fracture, that patients are likely to share with friends and family.
  • Interoperability. Martin outlined three principles of interoperability: invest to make sure data is used safely and securely for the benefit of the community, build what you need not what you think you will need in the future and establish service-based architectures.
    1. Invest to make sure data is used safely and securely for the benefit of the community. Martin cited one specific use case of interoperability wherein the system is working with Collective Medical Technologies to identify patients presenting at multiple emergency departments shopping for opioids. “We’ve been able to take that thin veneer of information that’s transported between health systems and widen the pipe to where it’s more useful,” Martin said.
    2. Build what you need not what you think you will need in the future. “We’ve seen a lot of HIE initiatives fail under their own weight over time, because you’d probably have to anticipate all possible future usage and it becomes this huge waterfall project,” Martin said.
    3. Establish services-based architectures. The model means opening platforms in a secure way so that only trusted parties accessing data under a BAA, for instance, or under a secure agreement can access the data. Martin suggested building APIs and treating those as a service that is going to have a certain set of service level agreements in place. “This is something we learned at Amazon and it’s one of the reasons Amazon was able to scale the way that it has.”
  • Privacy. Data sharing can pose considerable challenges because it is hard to anonymize data in compliance with HIPAA and requires manual inspection to ensure all the pieces are working correctly.
  • “We need to adhere to both the spirit and the letter of the law. So, let’s not get cutesy about, ‘well, there’s this technicality of the regulation that we can get around,’” Martin said. “It’s pretty clear to most reasonable people what patients expect in terms of their confidentiality.”
  • Patient consent. Explaining consent to patients about AI and machine learning can be difficult. Mathur said that RadNet patients are informed of AI in every bed or setting where it is being used. “It’s not only AI, it’s a tool that radiologists use to make their jobs faster and more effective at a lower cost.”

Post-pandemic challenges and opportunities
The nation’s response to COVID-19 has included sweeping regulatory and policy changes relaxing HIPAA, enabling clinicians to practice medicine across state borders, reimbursing for telehealth services. Many, if not most, providers have reacted by moving quickly to increase the use of readily available consumer apps and digital technologies to treat patients remotely and keep those who do not need care out of hospitals and treatment facilities.

“Right now, since we launched Zoom for 7,000 providers very, very quickly, everybody is celebrating that,” Martin said. “But the problem is that their volumes are also probably anywhere from 50 percent to 80 percent of what they are normally.”

Clinicians have more patience for imperfect technologies, AI and otherwise, than would be expected at 100% volumes and that is not likely to be true for long.  

“It has to be a much more seamless process,” Martin said. “And you need to deal with a mixed model in the future where some clinic visits are in person and some are virtual.”

The hybrid model will both become necessary and create new revenue and population health opportunities. Engaging patients multiple times throughout the year and between sick care visits presents opportunities to discuss a patient’s health and work to improve that person’s overall health, which Martin said fits nicely with Providence’s population health strategy and ability to take on risk.

“Engagement in health care has two benefits,” Martin said. “One is you really start to build a sticky relationship with your patients. There’s also a big economic opportunity in that if you’re entering into risk-based contracts, if you help build a healthier population, your economics are much, much better as a provider.”  

Conclusion
During the Health Evolution AI-enabled Patient Engagement virtual gathering, Martin and Mathur noted that the problems each is using AI to address are not entirely different than the issues executives at consumer-facing organizations are up against.

COVID-19, however, accelerated the need for and use of AI in ways that have various health care organizations pressing ahead more quickly than they otherwise would have to better understand patients and to increase engagement frequency. That requires focusing on data, interoperability, privacy and patient consent.

AI-enabled patient engagement will create challenges as well as opportunities. But they are surmountable obstacles and can be overcome by taking an iterative approach of identifying the problem to be solved, breaking it down into components, and working on each of those.