Powered by AI: Technology, digital health, and a new era of care
A lot of the buzzwords commonly heard around the tech crowds are now gradually working their way into common parlance in industries where technology is disrupting the status quo. So it’s no surprise that these days, it seems like the majority of companies emerging across all industries claim to be “powered by AI.”
But, what exactly is artificial intelligence?
Formally speaking, artificial intelligence (AI) is a branch of computer science that works to develop machines that are capable of functioning like an intelligent human — meaning they have (or will eventually have) the ability to reason, discover meaning, generalize, or learn from past experience. AI can be considered to be a sponge — absorbing as much data that is thrown at it. In fact, the more data the machine or computer is provided with, the more intelligent it can become. Today AI and accompanying analytics are being used by companies, regardless of industry, to explore everything from customer experience to financial forecasts.
In health care, AI and related technologies (like machine learning) can be used to gain insight into a patient’s medical history, present state, and potential future outcomes. In fact, as part of its 2018 Consumer Survey on Digital Health report, Accenture found 20% of respondents were already using an AI-powered health care service.
In 2018 and beyond, digital health companies supporting patients or providers will leverage AI to provide specific, personalized insights based on an individual’s current health status, lifestyle, behavior, and genetic information. Companies and health systems can draw on patient histories, epidemiology statistics, (plenty of) images, video, location data, and physician comments as well, a feat that would be impossible to scale via human insight alone. The wave of innovation is just getting started, as over 100 digital health startups are leveraging AI as part of their value proposition.
The ability to know more, sooner.
Arguably the area of greatest unlocked value for AI to address in health care involves keeping people healthy through preventative care. Instead of waiting for a person’s health to deteriorate significantly, continuous collection of data can help identify potential anomalies early on, allowing (ideally, personalized) interventions to be prescribed.
Data, data, data, different values and numbers, some correlate with each other, others not. How do we make sense of it? Sure, we can try to validate relationships we already know might exist, but what if we’re missing important information in the data simply because we can’t analyze available information quickly enough?
Due to a super-human ability to digest and analyze enormous amounts of information, AI can detect patterns and make predictions about future outcomes based on large quantity of variables. As we will discuss below, data from wearables, fitness trackers, and other devices yields enormous data sets — AI can analyze and learn from these continuous data sources, providing insightful care suggestions, and helping to create customized care plans for each patient.
AI has the potential to provide tremendous value in terms of chronic disease management, helping attenuate the number of people who are diagnosed with conditions like heart disease and COPD, while at the same time maintaining (and hopefully improving) the health of those who are already experiencing clinical-grade symptoms.
Care efficiency, monetary and otherwise.
AI has the potential to affect a massive amount of change in health care treatment from a monetary perspective — in terms of pinpointing what might constitute the most efficacious treatment at the outset, as well as in terms of addressing the patient’s health issues before they deteriorate, relieving them of the trial and error cycle often associated with finding the correct treatment regime in various disease areas.
Chatbots + Robots
Many of us already have “intelligent assistants” — Siri, Cortana, Alexa, or Google Home — in our homes. Though we may still be adjusting to how much easier they make certain aspects of our lives, we certainly have moments of “I wish Siri or Alexa could do this” — the thing is, they likely will soon be able to. These virtual assistants are still very immature in terms of their “intelligence” and, as such, currently have highly limited capabilities.
For instance, today’s AI virtual assistants may be able to answer questions you ask by searching the internet (e.g. “Hey Google, how many calories are in this brownie?”) but in the future, they will have more robust knowledge of your individual diet, exercise, and medications, among other variables.
Conversational agents like Ada, which bills itself as a “personal health companion and telemedicine app” will become more prevalent. They will be able to offer day-to-day advice on maintaining a healthy lifestyle, as well as the ability to triage medical concerns by providing guidance on whether or not a health care professional should be consulted, for example.
Robotics and in-home AI systems offer another avenue for AI-impacted digital health. Robots are able to help patients with independent living and even encourage (clinically) healthy people to live in a health-conscious manner. Robots can be used to assist with tasks such as medication reminders, clearing walkways or cleaning up spills, and reaching for objects (e.g. for people with limited mobility). They also have the potential to play a huge role as a companion, which is already happening with robot pets, providing emotional support to patients and people who may be vulnerable to isolation.
In the past we’ve relied almost entirely on patient self-report to their physician to provide the context that might guide the patient’s diagnosis and treatment. Fortunately (though I use this term loosely), to most of us, our smartphones are the central hub to our lives — they allow us to communicate with others in a multitude of different ways (voice, text, picture, video) and track our behaviour (perhaps more than many of us would like to know) as well as provide us with a means of entertainment. Ultimately, what this means is that our smartphones are continuously aggregating an incredible amount of data every second we have them with us (or even in their absence — e.g. if you have a paired fitness device). Our smartphones can already be used to track everything from mood to movement.
A host of conversational agents have emerged to help provide people with better care from the comfort of their own home (and smartphone) as well as to enable earlier detection of subclinical symptoms. Wysa, a chatbot by Indian startup, Touchkin, uses passive sensor data from the smartphone to track changes in the user’s behaviour. In addition, Your.Md and HealthTap both using natural language processing (NLP, the part of AI that has to do with language ) to understand symptoms being described and guide the user towards a conclusion about what the problem might be.
Voice is proving to be a particularly novel means of assessing health. Vocal patterns have been shown to provide a potential indicator of anxiety. Other studies have supported AI’s ability to diagnose PTSD in veterans. Voice analysis technology may be of particular value for mental health afflictions, where passive data capture might be the only avenue available for analysis if the individual is hesitant to seek help or are unaware that they might benefit from doing so.
Voice analysis certainly isn’t limited to identifying mental or emotional affect. Beyond Verbal has used its technology to discern vocal traits tied to coronary heart disease that cannot be picked up by the human ear.
Passive sensors mean that we don’t even need to actively track a lot of what we’re doing — instead, actions and events can be inferred by simply carrying our phone around with us or chatting with friends. Sonde, for example, uses tone of voice to detect anomalies that may indicate physical or mental illness. Passively collected data on hand movements may even provide insight into Parkinson’s disease. In another Parkinson’s application, CloudUPDRS, a smartphone app, uses the phone’s gyroscope to analyze and quantify tremors, patterns in gait, and performance in a standard “finger tapping” test. Using an AI algorithm, the app is able to differentiate between actual tremors and noise or “bad data” (e.g. if the phone was dropped).
Wearables + Devices
In addition to the data continuously gathered from our smartphones, wearables and fitness trackers help gather physiological data or more advanced activity that isn’t capturable through smartphones alone. Take Boltt, for example. Boltt is leveraging AI to provide wearers with actionable feedback based on data it aggregates on diet and exercise (and even connected shoes).
Another area where AI may prove to be especially influential is in terms of supporting positive, clinical health behaviours. Beyond documenting movements and encouraging people to exercise, for example, AI-based applications can be used to improve medication adherence. Respiro, for example, is a respiratory disease management platform that is able to track inhaler use and treatment effectiveness for people with asthma.
Though not consumer-facing, AI has already dramatically enhanced our diagnostic abilities. Recently, Google and Verily — Google-parent Alphabet’s life sciences unit — published a study exploring how AI can be used to identify patients’ risk of suffering from a cardiovascular event, such as a heart attack or stroke. Google’s AI uses non-invasive retinal scans to infer the patient’s age, sex, blood pressure, and smoking status. Also, less than a month ago, the FDA approved IDx-DR, an artificially intelligent diagnostic software program that is used to detect diabetic retinopathy, and doesn’t require a specialist to confirm the computer’s diagnosis.
Replacement for humans?
It’s not hard to see that when it comes to aggregating and analyzing large amounts of data, AI has a step-up on us in their ability to find relationships and patterns that are amiss to us. Beyond that, AI is tireless. Human brains are fallible, to the extent that they are susceptible to cognitive and memory biases — which can affect our decision making without us even knowing so.
Health care professionals and patients alike are susceptible to the recency bias, for example, whereby it is easier to recall the last case or event and its outcome — which can impede our ability to objectively consider the details of the situation at hand. In other instances, we may believe that we’re not susceptible to an event happening to us, or our physician associates our gender, behaviour, or even family members to an outcome as being improbable. It is in these cases — where an overreliance on context and what is available in our memory — that AI can help guide us. Importantly, it’s not quite as black and white as it appears. As humans, we’re responsible for building predictive models and providing data sets for machines to learn off of — so we must be conscious of the impact of these choices and try to be as impartial as possible.
Another way in which AI is better suited than humans is when it comes to tasks that must be repeated (and are thus ripe for some form of technology-based automation). AI algorithms continue to learn, regardless of how many times it’s looked at scans or hours it’s worked that day. Meanwhile, when exposed to repetitive tasks, humans are susceptible to lapses in attention, in addition to fatigue, stress, and emotions that can all affect performance and the interpretation of information. For these reasons, there has been a lot of discussion around the role of computers and AI in medical disciplines like radiology, where a high volume of scans must be reviewed on a regular basis.
Our expectations, and our fears.
People have greater expectations for targeted, timely answers to their needs and requests. In health care, as in other industries, as the technology becomes simpler, smarter and more intuitive, the customer experience will improve.
But is AI’s role in health care worth the hype? AI has been lending a helping hand in medicine since the 1970s, so why now?
For one, the level and speed at which AI has infiltrated other industries has left health care consumers wanting for comparable experiences when it comes to managing their own health and wellbeing. Patients are frustrated with long wait times for appointments, crowded waiting rooms, and inconclusive, broad recommendations. For reasons inherent to health care delivery, such as the deep importance of data privacy, it is perhaps unsurprising that the industry has struggle to incorporate technology in its workflows to the same extent as others. Moreover, until recently, the consumer health care market was minimal without wearables, fitness trackers, and even smartphones. However, because health care is so pertinent to each and every one of our lives, it means the insights that we can yield will be all that more impactful. To this extent, in an interview with GE Healthcare, Dr. Mark Michalski, Executive Director, MGH & BWH Center for Clinical Data Science, commented, “… [t]he power of Artificial Intelligence technology is that there’s not a boundary in its capability. The more data we can show it, the better it performs. And that flexibility is really what’s so exciting.”
But is it true “Intelligence”?
The fact of the matter is, in virtually all cases, AI can and will not substitute for clinical judgment. What we’ve seen here, however, is that AI enables health care professionals to dive deeper into information that may provide crucial guidance in terms of diagnosis and treatment options, as well as in maintaining health through preventative care.
Medicine falls far beyond the realm of data processing and statistical inferences, though this is certainly a foundational component. Consider empathy. Patients rarely experience a health event on their own — family and friends, often providing invaluable care themselves, must also be considered when it comes to what care is provided and how it is provided. Further, patients themselves have preferences and beliefs which can significantly diminish the impact of objective data and need to be factored into care-related decisions.
Efficiency Gains Needed.
A few years ago, a WHO report forecasted that the world will be short of 12.9 million health-care workers by 2035. Moreover, as societies around the globe encounter aging populations, often requiring more health care ‘touches’, we need to find more efficient ways to provide (quality, timely) health care.
The role AI currently plays in health care and digital health is that of presenting insightful, personalized options much faster and much cheaper than would otherwise be possible (in addition to providing options that we may not have otherwise thought about considering). This shifts the role of the health care provider towards one where they can focus on the most important elements of patient physician interaction. Some have predicted that health care professionals, with the help of AI virtual assistants, will be able to see more than 5 times as many patients as they do today, all while providing better outcomes.
Dr. Rasu Shrestha, Chief Innovation Officer at the University of Pittsburgh Medical Center, considers AI in its current state, and in health care overall, as Augmented Intelligence rather than Artificial Intelligence, insofar as it enhances the ability of health care professionals to address many of the fundamental issues underlying their original rational in pursuing a career in health care, by focusing on the patient, and less so on administrative tasks or caught up in logistical challenges.
Moving forward, intelligently.
AI in health care has the potential to tackle many tasks that are currently handled by humans, but with the potential to do them as well (or better), faster, and in a more cost-effective manner. As technology becomes more embedded in health care, and with an increasing number of interactions, we can expect the consumer (patient) experience to see improvements. Importantly, the value of AI in health care, as anywhere else, is dependent on the quality and amount of data available. Trends in digital health uptake and efforts to increase use in underserved markets do not raise immediate concerns in this area; however, this is an ongoing issue to watch.
In terms of further defining the role in AI in health care, while AI may have talents we can only dream of aspiring to, ultimately, we are responsible for its decisions. Relevant individuals or parties must hold accountability if the decisions made through AI are executed, whether it affects a single patient or population health, more generally.
Creating cross-functional working groups like The Collaborative Listening Summit developed in response to Google’s DeepMind Health pilots in the UK, help open up the conversation on operating practices and societal impact to individuals far beyond company shareholders. These types of meetings, conferences, and organizations are crucial to making sure different perspectives are heard and considered in the development of technology.
Finally, as both AI-enhanced tech and the researchers that work with it become more experienced and versed in its potential uses, we can expect that AI-fueled health care will gradually move towards becoming a part of a new standard of care for patients.