The emergence of artificial intelligence (AI) as the technology of our time is generating excitement and anxiety across our society and economy. People are excited about what new frontiers AI opens up for us as a specie and what threats it poses to our livelihood and even in our existence. This is not a new phenomenon. Examining the literature from the past eras reveals the same hopes and fears when other transformational technologies first appeared. We are talking electricity, aviation, telecom, and the internet. As has been the case many times before, AI will neither solve all of our problems nor eliminate most jobs.
Applications of AI in healthcare promise to be vast and deep. AI is well suited to sift through large amounts of data and extract insights. No other entity that we deal with is made up of more data than the human body. Most of the secrets to our bodies and how it functions are still eluding us. Human body is made up of billions of genes, trillions of bacteria and viruses in our guts (microbiome,) and billions of other molecular. How these entities interact to give us health or disease is, for the most part, beyond our comprehension at this time. While we have been studying these concepts for centuries, our understanding is still at its primitive stages. AI promises to change that. Why? Because AI is well-suited to study large amounts of data with complex relationships and tease out the key insights. This will accelerate our understanding of diseases and finding or creating treatments for them.
Another area that will be ripe for AI is in the provision of healthcare. We have an understanding of how to manage chronic diseases proactively to prevent complications and thus improve quality of life and lower cost of care. What is keeping us from doing so? This type of care requires us to do many things that are not part of healthcare: Monitoring of patients on an ongoing basis, managing patients outside of the healthcare setting, and keeping them engaged. Given the shortage of resources in healthcare for the reactive care we’re providing today, we can forget about taking on such a labor- intensive new model. What if we could hand off many of these activities to technology? AI can most certainly take on much of these activities and do them well.
All of this sounds exciting and very promising. Yet, although health AI has been around for some 10 years, most clinicians and patients have yet to feel its impact or even see it in action in their daily lives. Each use case has to be carefully studied and implemented. This evaluation will need to start with the benefits to the buyers and whether the effort required to buy and operationalize the technology will be worth it. If medical centers are dealing with significant shortage of staff and difficult economics post-pandemic, they will not prioritize buying technologies that will not address core problems. Limited budgets means prioritization and that means shopping for technologies that can automate labor-intensive functions and generate revenue or lower staffing costs. Great example of this are AI virtual assistants that can start augmenting the nursing staff in managing hospital patients. These AI assistants can passively monitor patient data and only involve the staff when something needs to be done.
Another key consideration before moving ahead with an AI project is the technical issues. The current state of data in healthcare is fragmented and chaotic. Much of the data is in unstructured and narrative format. This means that the AI solutions that need this data may not have easy access to the type and quality of data needed in the real-world. If that is the case, then the output of these solutions will fall short of expectations. An example of this would be clinical decision support tools that need complete and accurate patient data. However, patient data resides in many different databases and interfacing with all of those is very challenging. This, so far, has meant projects that did not continue long-term and an eventual return to the old way of doing things. A careful analysis of the state of data in the real-world is critical before moving ahead with AI projects. Many of the dream use cases for health AI will take time to fully deliver value because we first have to solve the data issues.
Another key consideration for digital products is that for the users to stick with them long-term, they will need to fit with existing workflows or lifestyles. Many health apps see initial adoption only for the user to ultimately return to their old habits. This is in spite of the fact that they do provide some benefits and value. The issue is that if a product requires the clinicians or consumers to perform extra activities or makes the management of their work or life more challenging, they will not stick with it. Health AI is not exception. Nobody cares if a product uses AI on its backend. What they care about is getting a benefit without having to be burdened with extra effort. As such, careful analysis of the existing workflows and designing for ease of use and providing effortless benefit is critical. This has been lacking in many of the products launched to date and the result is predictable every time.
All of this adds up to the conclusion that there are significant business, technical, and clinical barriers that need to be overcome to provide the benefits of health AI. None of these are insurmountable and I’m confident that over time we will make the necessary adjustments to our data collection methods and workflows to take advantage of AI’s capabilities. However, in the short-term, use cases that solve for these issues will be the winners. Ensure a clear benefit to the buyer, availability of the data in the real- world, and fit with existing workflows and you are on your way to health AI stardom. This type of rigorous analysis before you spend years building an AI solution will prevent the fate of most products to date that have not seen meaningful adoption.
