Skeptics about the future of artificial intelligence (AI) in healthcare are running out of reasons to be doubtful.
AI, in fact, is already revolutionizing several areas of medicine. Researchers rely on AI to help them select candidates for clinical trials, algorithms are proving to be about as accurate as radiologists in diagnosing lung cancer, and experts from Oxford and Yale believe that all surgical work could be conducted by AI by the year 2053.
Clearly, when it comes to clinical work, AI’s time has come. But for healthcare leaders, what may be less clear is how AI can improve upon another critical aspect of the healthcare industry—the patient experience.
That’s because the questions surrounding patient experience don’t easily lend themselves to the analysis by a machine. These problems are fuzzier, more subjective. They’re rooted in a patient’s feelings, which, unlike lab values or diagnostic data, can be challenging to capture and quantify.
However, innovations in data collection and sophisticated models of machine learning are proving out AI’s potential in these subjective domains, too. Health systems will soon be able to use these technologies to dramatically refine their understanding of their patients and to offer a care experience precisely catered to their needs.
Three examples of such cutting-edge AI systems are described below. One is already in use by forward-thinking organizations today; the others will be part of the AI-driven patient-experience revolution to come.
Out today: Natural Language Processing
AI solutions are hungry for data. To train their algorithms, they need to process immense amounts of information. However, until relatively recently, the quantity—and quality—of patient feedback data didn’t measure up to AI’s requirements.
That changed with the advent of two inventions: open-ended real-time feedback and comments on social media. Together, these brought organizations enormous volumes of verbal input from patients. With such a wealth of information, AI tools like Natural Language Processing (NLP) can achieve a remarkable level of customer insight.
NLP uses a type of algorithm called Naive Bayes to “read” patients’ verbal comments, and then classify them based on the probability that they express a certain intention. For instance, most email filters use NLP to flag emails containing the word “money,” because they are highly likely to be spam.
In healthcare, NLP offers considerably more value. A type of NLP process called sentiment analysis can tease out invaluable qualitative information from patient comments, including how patients felt about their encounters. If the word “forever” appears, for example, that might hint at frustration with wait times; the word “rude” probably indicates a service problem.
Humans can understand these comments too, of course. But with the sheer volume of comments coming in, a human staff has no hope of reading them all. That’s why tech-savvy organizations rely on NLP to automate their comment processing. It enables them to spot dissatisfaction much sooner, and craft solutions with much greater precision.
One such NLP-ready organization is Hackensack Meridian Health, in Edison, New Jersey. Bridget Alston, Hackensack’s Director of Patient Experience, says this qualitative information is crucial to how the organization constructs its care experiences:
“We don’t just utilize the quantitative data, we go into the qualitative analysis and relate it to what our data is saying,” she says. “[With NLP] we’re able to take that information on a regular basis, and know what our patients are saying—and try to intervene—much sooner than we normally would.”
This is the real value an NLP solution brings to an organization: it gives leaders a clear and immediate sense of what their patients are trying to express, so they can make smarter decisions about their care.
In development now: predictive analytics for patient experience
Predictive analytics is one of AI’s most promising developments. The term is shorthand for any process that uses historical data to make predictions about the future. Its most famous use-cases are in the finance industry: banks routinely use AI-enabled predictive analytics to assess a borrower’s credit-worthiness.
With a little imagination, it’s not hard to see why such tools would be useful in healthcare. Indeed, predictive analytics might be healthcare AI’s new frontier. Right now, payers are tentatively embracing it, using it to process patient case data and discern the likelihood of a change in health status sometime down the road.
This deployment of AI raises some complicated ethical questions. Concerns about data bias and flawed algorithmic decisions make some providers uneasy. But health systems should not ignore the potential benefits of bringing predictive analytics into their operations, especially when it comes to patient experience. Health systems can use predictive analytics to anticipate what their patients will need, instead of merely reacting as new health concerns arise.
Well-designed analytics engines will combine patient health information (EHR-derived data points like lab values, diagnoses, and treatments administered) and patient behaviors (such as online engagement, appointment setting, cancellations, satisfaction scores, compliance, and follow-up contacts) to bring new clarity on what patients want from their providers.
Health systems will know, for example, not just when a new health concern is likely to arise, but also when and how the patient would prefer to make an appointment to address it, what kinds of services the patient will most likely desire, what kinds of interactions are likely to increase the patient’s compliance with follow-up instructions, and more.
Today, predictive-analytics products can’t yet offer that level of sophistication. But the technology’s evolving fast. Researchers and entrepreneurs are working continuously to bring such AI products to the market. It won’t be long before an experience-focused analytics product is made available to health systems.
Coming soon: personalized engagement engines
As complicated as predictive analytics are, personalized engagement engines demand a much higher order of complexity.
Think of Netflix’s recommendation system. The streaming service takes in what it knows about viewers, and then automatically offers suggestions for what they might like to see next.
On its surface, this might not seem so complex. But what’s remarkable about processes like this—collectively called cluster behavior prediction systems—is that they’re a type of “unsupervised machine learning,” meaning, they function with no assistance from human programmers. Instead, they absorb trillions of raw data points from consumers and interdependently identify patterns that they can use to group consumers by their preferences. That’s how Netflix knows that 30-year-old male viewers who watched Die Hard would probably also enjoy Terminator 2.
Now, Netflix’s algorithms undoubtedly use an incredible number of consumer data points. (It can be disquieting to contemplate just how much these companies know about us.) But for such a system to be useful in healthcare, it would require exponentially more.
These data points aren’t just limited to clinical information, or to patient interactions with health systems. They’ll also include socioeconomic status, demographic data, ZIP codes, fitness-center attendance rates, family status, biofeedback data from wearable tech (like the Fitbit and Apple Watch), grocery and restaurant food consumption, and then some. The array of potentially useful data points is truly staggering.
But imagine the potential. With a refined personalized engagement engine, it’s possible that health systems will have a newfound ability to tailor their services for patients.
They won’t just know what patients will need—they’ll also know where and to whom to send it, in order to maximize a given patient’s happiness with the encounter. They’ll be able to automatically customize their communications, tweaking grammar and vocabulary to ensure that patients understand care instructions. They’ll be able to anticipate certain complaints and pre-emptively recover service. They’ll even be able to present patient-specific preference information to clinicians, intelligently filtered and prioritized, at precisely the moment they’ll need it.
In short, this technology could offer health systems a way to construct a concierge care experience, customized for every patient who comes in the door. It would begin an era of mass personalization.
While such an idea sounds like science fiction, researchers are already at work making such technologies a reality. “Cognitive aide” technologies are already making an impact on clinical care. It’s only a matter of time before they augment service decisions as well.
Care will always be personal
But note carefully—that augments. Not replace.
As robust as AI technologies can become, they will never be a substitute for human judgment. No machine could ever supplant the intimacy and importance of a provider’s relationship with a patient.
On the one hand, this is a comforting thought. It means there will always be a role for humans to play in healthcare. On the other hand, this is also an exhortation. AI technologies can offer direction, but it’s providers who will need to assess and execute on what AI-driven solutions uncover. Mass personalization will require more of us, not less, as we strive to create the kinds of experiences that our patients deserve.
Steve Jackson is the President of NRC Health, a provider of in-depth customer intelligence in healthcare. As President, Steve oversees company strategy and NRC’s portfolio of solutions that bring Human Understanding™ to healthcare. Prior to joining NRC, he held roles of increasing responsibility at Vocera Communications, The Advisory Board Company, Neoforma, and Stockamp & Associates. Steve graduated with honors from the University of California, Davis.
