It’s hard to miss the advertisements for virtual assistants from Amazon, Google, and Apple (all powered by AI), and TV commercials promoting AI capabilities for Microsoft and IBM. The hype about AI is real.
Whenever technology is hyped up, I’m always a little skeptical about whether it’s producing the outcomes to justify the hype. Therefore, let’s explore whether or not AI is producing real, tangible differences in healthcare.
What AI is not
Let’s start by clarifying a few common misconceptions about the technology. First, AI is not machines becoming self-aware. It is not Skynet from Terminator or V.I.K.I. from I, Robot. It is not a parkour robot that will take over the world, although that robot may one day save lives on the battle field or during natural disasters. In other words, do not confuse real-life AI with entertainment’s version of AI. The AI being used in production today is not that advanced. I recently wrote an article comparing my newborn to some of today’s most advanced AI algorithms and toddlers are even more advanced than today’s AI.
The days of AI being able to hop on the internet and learn something new like Baymax does in the animated movie Big Hero 6 are far down the road. Between now and then there are a great deal of hurdles to overcome. In the movie Baymax uses information from the internet to teach itself about emotional pain. While the lack of data preparation would be a hurdle today the bigger issue is the lack of data validation. The validity of data is one of the key reasons why AI is not widely trusted in the medical community. In addition to data validity, clinicians question the lack of model expandability and the potential for bias. All of which are very valid concerns and must be addressed if AI is going to be trusted with our most valuable commodity: human life.
What AI is
AI is the tip of the spear of the analytic journey and its role is to predict, automate, and optimize. However, there are a number of data preparation steps that must be completed before AI is applied. Data preparation consumes the vast majority of time, resources, and cost associated with an AI project.
AI is considered “intelligent” because it has the ability to learn, but that doesn’t mean it possesses intelligence on par with humans. Far from it. AI requires vast amounts of data to learn very simple and specific tasks. A more apt comparison is comparing today’s AI capabilities to a smart, well-trained canine. We use canine’s strengths like their superior sense of smell to the benefit of or in service to humans. The animals go through specific and repetitious training that results in a targeted and defined outcome. Similarly, machines have a superior ability to process, calculate, and analyze data as compared to humans. Therefore, we use specific and repetitious training to achieve targeted and defined outcomes such as finding anomalies in vast amounts of data, matching key words, and identifying patterns.
The key similarity between canines and today’s AI is that the training results in a very specific outcome. To illustrate this point, think of all the different types of scents canines are trained to detect – low blood sugar, explosives, drugs, scent of people, etc. Even though the foundational function of detecting a smell remains the same, canines are only trained to detect a single scent. Similarly, today’s AI is is most effective when targeted toward accomplishing a single task. Humans, on the other hand, are able to build on the knowledge they’ve learned and apply it to new situations to effectively teach themselves.
Where AI fits in Healthcare today
Organizations invest in technology due to one or more of three primary drivers: 1) grow revenues 2) reduce costs 3) increase customer satisfaction. AI’s versatility allows it to hit one, two, or all three of these drivers in a single solution which is why it’s such a hot topic.
Every major healthcare provider or insurance organization in the United States is already using AI or is exploring how ways to use AI. Some of the uses are behind the scenes in research and development and may never reach production. For this article, I’m going to focus on the areas where AI is delivering quantifiable results.
Consumer Experience
The key to a good customer service experience is speed and accuracy of information. When the consumer has to wait on hold, repeat themselves multiple times to virtual agent, and re-validate their account to a live person after previously validating with a virtual agent customer satisfaction suffers. Chatbots and virtual assistants like Amazon’s Alexa, Google Home, and Apple’s Siri have raised our expectations for virtual agents. Their natural language processing capabilities make interacting with them easy and intuitive. As a result, upgrading virtual agents to meet these new consumer expectations is one of the top priorities of healthcare organizations I work with. According to the latest third party analyst ratings Watson sits head and shoulders above the rest. IBM’s commitment to the hybrid cloud infrastructure allows organizations to utilize the power of Watson without compromising compliance or security. Additionally, IBM has announced that Watson is now available on third party cloud providers such as Microsoft’s Azure, AWS, and Google Cloud.
Healthcare organizations are beginning to prioritize the accuracy and relevance of information available to consumers through search engines and third party websites. Information like each location’s hours of operation, days physicians are available at a particular location, whether a physician is accepting new patients, and which insurance carriers they accept should be readily available and always up-to-date. When this information is incorrect it not only affects consumers but it can also affect billing. Updating this information manually can be very time consuming and inefficient as some search engines may not populate the changes for days, weeks, or even months. The company leading the way in forging the Digital Knowledge Management space is Yext. Yext’s Healthcare Knowledge Engine is a digital brand management platform allowing healthcare organizations to better manage the digital portion of the consumer-to-patient journey. The company recently announced Yext Brain which they describe as “your AI strategy’s central nervous system.” Yext is a company on the rise and I expect big things from them as they continue to carve out their market.
As a side note, the rise of the consumer in healthcare is one of the most interesting and relevant developments of today’s healthcare landscape. It signifies a shift in the power dynamic between the care provider and the care receiver. As value-based care initiatives and the role the individual plays in their care continues to develop and evolve so too will this dynamic.
Analytics & Business Intelligence
With AI as the tip of the spear of the analytic journey it’s only natural that the analytics and BI space has been an early adopter of AI technology. AI’s ability to predict, automate, and optimize while continuously learning and evolving is a natural fit for tasks like flagging billing and coding errors, clinical decision support, financial and supply chain management, facility management, and NLP for clinical notetaking. Those are just a few examples of where AI fits in this rapidly evolving and growing space.
There are very innovative companies delivering real outcomes in this space. One company that has caught my attention is Health Catalyst. They recently ascended to unicorn status (achieving a company valuation of $1 billion). What’s been even more impressive are the outcomes the company has delivered for their clients along the way. Their website boasts an impressive 178 success stories and Health Catalyst CEO Dan Burton stated to Forbes that last year alone they recorded over 250 projects that resulted in “measurable clinical, financial or operational improvements.”
Care Management
Care management is one of the most exciting uses for AI technology because of the ability to produce clear, measurable outcomes that positively affect quality of life. AI is capable of deriving a deeper level of insight than traditional analytics. Utilizing AI to augment human intelligence in this way can produce a continuous loop of care development, oversight, and growth.
A few examples of how AI is being used in care management are to actively monitor patients with chronic diseases and alerting care providers to pertinent changes in condition in real-time, to alert pharmacists to medication conflicts, to match patients with treatments that prove most effective, proactively identify sepsis, preventing 30-day hospital readmissions, and to monitor the mother and fetus during pregnancy. AI use in this space is growing rapidly because the benefits and ROI are often easily quantifiable.
A company excelling in this space through the use of AI is HealthEC. Recently ranked first on the 2019 “Best in KLAS” report, HealthEC excels at collecting ever-increasing amounts of data from disparate sources, analyzing them, and helping organizations across the care continuum improve their operations. They boast the ability to access 100 percent of all available electronic health data, structured and unstructured. Their use of AI allows them to deliver enhanced insights across the entire healthcare landscape and connect payers, providers, patients, labs, and hospitals through a single platform.
Diagnostic & Imaging
One of the areas that AI technology has matured significantly is image recognition. This capability is a natural fit in dermatology and radiology. A recent study proved that AI has the capability of diagnosing skin cancer more accurately from a picture than a dermatologist. This does not mean that we don’t need dermatologists, but rather that AI can be used as a front-line diagnosis tool to help catch skin cancer earlier because of the convenience of using a smartphone to snap and upload a picture. The convenience factor of this technology has a real chance to make a significant impact in the early detection rates of skin cancer. When skin cancer is found and removed early it is almost always curable.
In radiology and imaging AI use is growing rapidly. While there is great optimism among radiologists that AI will be able to provide more substantive value at scale, there are hurdles that need to be addressed before that can happen. Marrying the technology to current workflows and building trust in the data used to train the AI algorithms are examples of hurdles that need to be addressed but it’s clear that AI uses in radiology are only going to increase.
Final Verdict
Even though AI is currently being over hyped by the marketing teams of many large technology companies, AI uses in healthcare are real. AI is producing legitimate business outcomes that produce a quantifiable ROI, which is table stakes for any new technology to gain a foothold in an industry. Furthermore, the current push for value-based care makes it necessary for healthcare organizations to derive deeper and more meaningful insights from the mounds of data they have available to them. As companies like IBM continue to develop and refine software tools designed to improve visibility, control, and explainability of AI models confidence and usage of AI will only increase.