THANK YOU FOR SUBSCRIBING
Artificial Intelligence (AI) is capturing the imagination of healthcare providers, earning it star-status as the “next big thing.” Who can blame people for dreaming big, especially with the prospect of smart machines unlocking more secrets to the genetic code, precision engineering our surgical capabilities, and even doing away with passwords to our data systems? Consumer adoption adds fuel to the excitement. We have Alexa, Siri and Cortana in our cars and homes, doing everything from playing our favorite songs to turning on the lights.
It seems we are on the verge of a paradigm shift.
Or is it the verge of an AI winter, where the hype bubble bursts and progress stalls for a decade or more?
"The best path for embracing AI is to consider where integration makes sense right now and test its effectiveness over time"
With all that is being promised about the potential of AI to overhaul how health care is delivered, I believe healthcare technologists and data officers must manage expectations for providers, policy makers, and healthcare decision makers.
As illustrated by the Gartner Hype Cycle Curve, AI adoption in healthcare will come in waves of excitement, disappointment, and eventual productivity. There will be highs and lows as the implementation of AI in specific use cases races to keep up with hype and expectations of the technology paradigm as a whole.
Heading into 2018, we’re poised to see a trough in the hype cycle as excitement over the vision and promise of AI, which has caught the attention of healthcare leaders and decision-makers, comes to terms with political and operational realities.
AI solutions aren’t perfect. Out-of-the-box solutions evolve and become better over time. AI works like a human brain in connecting units of information based on former experiences. The strength of a connection depends on the volume of lessons that inform the system. The difference is that while a human brain can learn meaningful lessons from one or two examples, current AI solutions require hundreds of thousands—or even millions— of examples to be carefully curated and fed in. They also require computing power to process all of the examples. This operational reality is not always a popular message when it comes to asking for budget and resources for AI deployments, a well-informed and reasoned leadership is essential as the AI bandwagon gains momentum.
There is also the nature of the health care industry to consider. Whereas technology companies can go full-bore with the latest innovations, healthcare is inherently more conservative. This is often for good reason. People seek care when they are most vulnerable, entrusting their lives to us. This reality, combined with a host of regulatory issues, ultimately slows down experimentation. Add to this scenario, the complex human and organizational systems underlying most health care challenges, and it is no surprise that progress in adapting AI hasn’t been faster. This is not likely to change in 2018.
The best path for embracing AI is to consider where integration makes sense right now and test its effectiveness over time. One use that makes sense is improving the patchwork of hospital systems that collect vitals, aggregate data, and alert clinicians.
Early detection of abnormal vital signs is crucial to prevent problems such as airway obstruction, cardiac arrest, and bleeding from rapidly becoming fatal conditions. Yet, despite our current automation, many clinical systems still require a nurse to physically enter the vitals data into the electronic health record once per shift. This is not exactly high-tech. We should be able to deploy IoT solutions to build an interoperable platform that acts as a clearing-house for various vendor devices, serving up integrated data for AI systems to consume and precisely indicating which patients need care first.
Once discharged, patients’ vital signs are largely unmonitored. Although this is typically safe for most patients, there are some cases (e.g. heart failure) where consistent monitoring and understanding of the patients’ status is critical for better long-term outcomes. An AI system built to anticipate when a patient’s condition is about to deteriorate could prevent a readmission or even save a life.
Population health is another area where AI implementations can make in-roads in the current health care landscape. With population health, clinicians take a broad look at groups of people and work to make their lives healthier and less reliant on acute care. It is an essential practice of value-based medicine; one that more health systems seek to perfect. At its core, good population health management relies on vast amounts of data. This is where AI comes into play, identifying patients at higher risk, recognizing patterns, and changes in health status that can indicate potential illness, and monitoring the effectiveness of treatments. To date, there have been good examples of using AI to understand better specific adverse events such as the risk of an emergency department visit or predicting inpatient admission in the next several months. For the future, it is clear that there is a lot of promise here for care teams making more accurate decisions at a faster rate, due to which patients ultimately remain healthier.
At our health system, we are exploring process mining, which entails studying the data collected from typical business procedures and delving deeper into the digital footprint so we can better understand workflow patterns. In addition, we are starting to develop a diagnosis engine for assessing the risk of all conditions and identifying patients earlier in the disease state. All of this is exciting, but the full integration of AI into our operations is in its infancy.
Proponents of AI technology are best served by aligning with business leaders and presenting a solid value proposition based on the results of trial projects. Finding the right use cases, with the right sponsor and the right plan to demonstrate value will determine whether the infrastructure delivers on its promise or heads into winter.