The Buzz at HIMSS18: Artificial Intelligence

It is that time of year when 40,000 healthcare IT professionals convene at HIMSS. Population health management, big data, interoperability, value-based care, and patient engagement were some of the dominant themes in past years. This year it appears that everyone is talking about Artificial Intelligence (AI). HIMSS is devoting a full day to exploring the potential disruptive capabilities of AI and Machine Learning, including a keynote from former Alphabet chairman Eric Schmidt. We thought it might be helpful to provide an overview on AI so you’ll be well equipped to interpret the jargon on the expo floor this week.  

AI, Machine Learning (ML), and Deep Learning are all terms often used interchangeably to describe computer systems that are capable of performing tasks that once only humans could accomplish. However, it is worth noting a few important distinctions:

• Artificial Intelligence has been heralded as the potential savior of mankind since the mid-1950’s, when the Dartmouth Workshop birthed the term. In theory, AI has the ability to perform specific tasks as well as, or better than, humans. True AI is often measured by The Turing Test, which evaluates if a human can determine whether he or she is interacting with a machine or a human. There are very few platforms that approach this level, which is why many technologists instead prefer to identify solutions based on the definitions below.  

• Machine Learning, a subset of AI, is defined as technology that runs algorithms that parses data and, through pattern recognition and rule-based logic, makes a determination. As the name would imply, ML requires that the program continuously improve, which makes it a more sophisticated application than data analytics programming.

• Deep Learning, a subset of machine learning and the forefront of artificial intelligence, mimics functions the brain performs to “learn” from millions of iterations in order to make determinations that are often more accurate than humans. This includes coupling algorithms with structured data to read medical images and detect differences – whether that is identifying cancer on imaging scans or YouTube videos with cats. The algorithms become increasingly accurate as they ingest more images and receive a feedback loop. Alphabet’s DeepMind is probably the most notable example of this solution.  

• Cognitive Computing is a term coined by IBM as it attempted to describe its approach to AI with the introduction of its Watson development. Cognitive computers offer a synthesis not just of information sources but of influences, contexts, and insights. The machines are capable of “remembering” previous interactions in order to suggest the best possible answer, rather than the right answer. The most notable example is clinical decision support, which is used to supplement humans – not imitate human behavior. IBM Watson Health is often widely discussed as a leader in this area.

• Natural Language Processing (NLP) forms the foundation for many cognitive computing tasks. These programs ingest source material, in its many and varied forms (e.g., medical literature, notes, audio dictation) and convert into a workable format. This task is even more challenging in healthcare, where large volumes of unstructured source files are littered with acronyms and abbreviations. Alexa, Siri, and Cortana are such examples of voice NLP. Clinithink and Deep6 AI are two start-ups that are focused on converting unstructured EHR data for clinical trials into structured data and recommendations.

• Semantic Computing is the understanding of how different elements of data relate to one another and use these relationships to draw conclusions. For example, Google Translate is heavily reliant on this technology in order to distinguish between the meanings of similar words (particularly between different languages).

Adoption gaps and other limitations:

• Patient preference: Patients still have a preference for human relationships. In terms of clinical care, we believe that in the immediate future advanced computing will assist medical professionals – not displace them. For instance, AI / ML applications could automate clinician administrative tasks (e.g., voice assistants for note taking, quickly sifting through unstructured patient data to generate clinical insights, EHR data entry, coding, billing, etc.), which enables them to spend more time with patients. Health Catalyst, Qventus, Recondo, Jvion, Cyft, and HealthPointe Solutions are promising companies that help health systems and payers analyze operations and automate administrative functions such as PHM, risk assessment, and RCM.

• Oversight: It will be imperative that medical professionals use AI / ML applications as tools as opposed to solutions. We have all witnessed the robo-trading flash crashes, but in healthcare the analogous safety issues are potentially life-threatening, such as overmedication or misdiagnoses.

• Privacy and security fears: Individuals such as Elon Musk, Stephen Hawking, Bill Gates, and Eric Schmidt have suggested that AI could one day lead to machines putting their own interests above humans.

• Data and programming: AI / ML rely on the aptitude of their programmers, where training is in its infancy. Furthermore, these solutions get better as they are “fed” more data. The tailwind is that healthcare data is expected to double in the next 10 years. The headwind is that healthcare data is often not “clean” or longitudinal. The early success for companies such as VisualDx on ML for enhancing diagnosis through imaging, where computers have exceeded panels of specialists, is due in large part to the industry’s data inputs that have been digitalized, standardized, and set-up for interoperability.     

Frost & Sullivan is estimating that the market for healthcare AI products will grow at a 40% CAGR and reach $6.7B by 2021 and Accenture is estimating that AI applications could potentially save $150B for the U.S. healthcare economy in 2026. In anticipation of this opportunity, there were 50+ healthcare AI / ML deals with $700M+ invested in 2017. This does not include AI / ML investments made by large corporates such as IBM, Alphabet, Amazon, or Microsoft. It also does not include considerable investments in areas that have stated healthcare applications, but are approaching AI / ML across industries such as Ayasdi and H20.ai.

At TripleTree and TT Capital Partners a number of our clients and investments have been experimenting with ML capabilities in their solutions to create new efficiencies including Nucleus Health and Remedy Partners.     
What do you think about AI and all of its variations? Is success around the corner or are we still a few years away? We look forward to hopefully seeing you this week at HIMSS in Las Vegas!