For many people, the words “artificial intelligence” (AI) or “machine learning” (ML) immediately conjure images of a complex plot in a science fiction movie like I, Robot or Ex Machina. As fascinating as Hollywood’s depiction is, the evolution of AI is much more subtle, and the potential utility is much closer to reality than you may realize. In fact, you’ve probably already been benefiting from either ML or AI without even realizing it. As digital personal assistants like Apple’s Siri, Microsoft’s Cortana and Amazon’s Alexa adapt to your preferences and context, understanding natural language, there are varying degrees of each involved.
The utility of this technology also greatly varies based on industry. So what about healthcare? The potential to benefit patients and/or caregivers is massive. The healthcare landscape continues to evolve. There’s consolidation among health systems, the formation of accountable care organizations (ACOs) and integrated delivery networks (IDNs) that are striving to hit evolving targets for both cost and quality, and empowered patients who are taking more active roles in maintaining their health. Can ML and/or AI play a role in managing the needs of this changing landscape for either individual patients or the health of an entire population?
We’ve identified three areas in which ML/AI could potentially be of great benefit to healthcare marketing and drive outcomes in a population-health model.
1. Integrating and rapid processing of vast amounts of data
Unlike manually tracking your steps, exercises, sleep patterns or using any other latest wearable device, machine learning has the ability to consume virtually unlimited amounts of detailed data and not only integrate it, but also adjust alerts to a patient, caregiver or HCP based on recent behaviors. The most relevant pieces of information can be identified, while ignoring or discarding irrelevant data to focus on the most important variables. If the amount of data processed isn’t impressive enough, the speed at which machine learning can consume and identify relevant data makes the ability to act in real time a reality.
For example, take the partnership between IBM Watson Health and the medical device company Medtronic. The two have partnered to create a prototype of a smartphone app that could someday predict the onset of dangerously low blood sugar in people with diabetes. Data was gleaned from insulin pumps and glucose monitors, and by recognizing patterns over time that also include contextual data, Medtronic believes it can predict hypoglycemia up to three hours in advance, which is early enough for a person with diabetes to prevent a potentially dangerous event. And from a business perspective, think about the benefit to a health plan or an ACO that’s striving to hit their cost benchmark from the previous year if this could prevent a costly ER visit.
From a marketing perspective, what if Medtronic were able to plug the data and patterns directly into the patient portal, so that the patient’s physician could provide customized care or adjust treatment decisions based on real-life experiences of the patient? It’s likely that physicians, health systems and health plans would be very interested in the added benefit for their patients.
2. Invisible action systems
There are a lot of factors in a patient’s care that can affect outcomes. Compliance to a particular treatment may be one of the most frustrating and seemingly insurmountable forces for healthcare providers in a patient’s treatment plan. In addition, much of non-compliance occurs among well-intentioned, capable patients. But, often, there are many other things going on in patients’ lives, and adding technology that can often be frustrating, time-consuming and non-intuitive can either compound the problem or simply not get used.
The benefit of machine learning, especially when it can integrate data from health information sources, is that it can serve as the invisible workhorse in health management. Future technology will provide the greatest benefit when it factors in the human experience. Technology that requires an extra step to use is a burden on the patient and will inevitably join the vast number of other underused “technological advances.”
Devices that take advantage of machine learning that can work in the background — invisible to the patient — will have the greatest utility. Systems that can act based on machine learning and with “always on” and “push capabilities” can make alerts or patient communication more dynamic and customized to a specific situation. In the Medtronic example above, if the app recognizes a potential hypoglycemic event coming up, it can alert the patient and potentially intercept the behavior to remedy the situation.
Conditions in which symptoms like hypoglycemia are often invisible until a serious problem exists offer promising opportunities to incorporate AI into the treatment plan. For instance, think about anticoagulants. Getting patients to remember to take their statins (or now their PCSK9 inhibitor) is often an uphill battle when the patient doesn’t feel any different from day to day. Because these patients often feel healthy, the potential for a future cardiac event is an abstract idea in their minds. What if the device could remind the patient to take their medication, but also offer contextual messaging — such as reminding the patient of the silent build-up of arterial plaque over time — to increase the likelihood of compliance? In addition, what if by tracking nutrition and exercise, the app could alert the user to negative behaviors that could result in a cardiac event? Again, the high cost of a hospitalization from a cardiac event would be of high interest to health systems to find solutions to prevent them.
3. Supplementing human-to-human connection
One particularly promising use for artificial intelligence is in supplementing human-to-human connections. This doesn’t mean replacing meaningful personal relationships. But what if artificial intelligence can fill in where there are current gaps in a patient’s social connection?
One study that analyzed the relationship between social connections and health outcomes found that, based on their “social isolation index,” patients who had a higher degree of social isolation had on average 24% higher hospitalization costs than patients with similar conditions but more social connections. (make a quote) This becomes even more important for chronic conditions where there is a higher comorbidity with often-present conditions like depression.
Solutions could involve “affective computing,” in which computers can identify human emotions through facial expression or vocal tone recognition. Being able to detect a patient’s psychological state would be an extremely valuable variable in adjusting that patient’s treatment plan. It could even evolve current platforms like Microsoft’s Cortana or Apple’s Siri to the next level as a “virtual caregiver.” The specific dynamics for patients will vary greatly by disease state, so further investigation into the individual patient experience is vital to determining the best use of AI in healthcare solutions.
The Difference Between Meaningful Advancement and Fad
The use of artificial intelligence and machine learning is still relatively new in healthcare, but the potential benefits to the healthcare system in driving meaningful outcomes and potentially avoiding costly events like hospitalizations are huge. Finding the right use will be the difference between a meaningful advancement and a fad. As we continue to push the boundaries of what is possible, we need to clearly understand the challenge we’re trying to solve for and, ultimately, the experience it will provide for real human beings.