My colleague, Jerome Pagani, Ph.D., and I have written extensively about how new entrants to the health space (hello CVS, Amazon, Walmart, and hundreds of garage-born startups) are attempting to squeeze into the former impenetrable zone between legacy healthcare systems and their patients. We see this disintermediation when hyper-focused apps and services are offered for specific diseases such as diabetes and asthma. We see it when sophisticated scheduling tools make it incredibly easy to book and carry out a virtual visit with a physician on the other side of the state or country. And we see it when companies start manufacturing their own generic medicines that are often cheaper than a traditional insurance company’s copay. After taking care of patients for decades or even centuries, how will hospitals and healthcare systems respond?
Naturally, hospitals have reacted in all the ways they can. Where it makes sense, they partner with the new entrants either in a conventional vendor-client relationship or sometimes put their money where their proverbial mouth is and become a part owner. Larger and well-funded entities have the resources to create their own tech or service offerings, thereby effectively competing directly with Silicon Valley. This can be quite successful, but most healthcare systems have concluded that running a small software company within their walls is far enough outside their core competencies that it’s not viable in the long term.
New artificial intelligence (AI) tools that have been taking the world by storm in the last six months might be just what the doctor ordered to help hospitals and healthcare systems compete to keep their patients in the family. Large language models like ChatGPT have shown great ability to parse large amounts of information, manipulate it in useful ways, and communicate results as well as – or better than – most humans. Not only are these AIs good at generating text, but they have shown the ability to write software and create technical interfaces by “reading” through jargon-filled documents that stump all but the most sophisticated human data exchange experts.
The ultimate goal of these technologies, at least from the perspective of the healthcare provider, is to offer omnichannel communication to their customers (aka patients), both current and prospective. If someone wants to schedule an appointment by calling and speaking to a human, no problem. If someone wants to use a generic app that’s not branded to a specific system (think ZocDoc), no problem. If a prospective patient wants to look for and schedule appointments on the healthcare system’s website from a laptop, no problem. The look and feel might be different, but the outcome should be the same.
How can an AI help? From a sophisticated chatbot on the hospital’s web homepage, patients can enter free text requests and be routed to the correct part of the website to achieve their goals, such as scheduling a procedure or appointment, finding a physician who speaks their language, or getting more information about symptoms that they might be experiencing. Improved engagement can be achieved by making it easy to do the tasks that patients typically do without help from experts or hospital staff.
As mentioned previously, an AI can power charge a lone developer or interface engineer by allowing them to concentrate on the creative or unique aspects of their work, leaving the more straightforward dotting of I’s and crossing of T’s to the machines. For example, GitHub’s Copilot uses AI to suggest code and functions based on input from the programmer, either by starting to write the desired code or by writing a natural language comment describing what needs to be done.
Let's say a developer is working on a project to integrate patient data from multiple sources using FHIR. This task involves writing code that adheres to FHIR standards and uses the appropriate resources and data elements. Writing this code can be time-consuming and complex, especially for developers who are not familiar with the API. With AI tools, the developer can enter a few keywords related to the integration, such as "FHIR patient demographics," and the AI will suggest lines of code that use the appropriate FHIR resources and data elements.
Today, generative AI has proven that it can communicate complex information in just the right way to very specific audiences. Need to ensure that a patient understands why it’s so important for them to follow up with their PCP after seeing the outside specialist? AI can help with that. Not only will it reference details that will likely resonate with the patient (e.g., physician names, reasons for the referrals, etc.), but it can write the message at an appropriate level.
Is AI the silver bullet to cure what ails healthcare in the U.S.? Decidedly not. There is no single fix that will help make what we call a healthcare system into … well, a system. But AIs can move us in the right direction, potentially neutralizing a big chunk of the differences between the haves and the have-nots. AIs can remove some of the necessary yet dreary work that clinicians today must perform, freeing them up to be humans helping humans.