IBM launched Watson Health in 2015 with plans to use its core artificial intelligence platform to help clinicians analyze stockpiles of data and transform cancer treatment. They recently sold many parts of it to a private equity firm at a loss and without transforming healthcare. IBM Watson focused on large, difficult initiatives that led to the hype surrounding their AI. And, rather than building the AI iteratively with back office tasks, they went big with clinical goals, like real-time cancer treatment recommendations.
While I think AI has a bright future and has some wonderful applications in current use, the hype surrounding AI “transforming” care is early and the brakes need to be tapped.
I am not “anti-AI” and believe it has a bright future in health care. In fact, my organization has several AI tools in use today by frontline clinicians in play today. We have a very successful telestroke program that utilizes AI to identify stroke patients whose condition may be amenable to procedural intervention.
Another use of AI involves converting a securely recorded office visit into a structured physician note. This has the potential to improve patient satisfaction, reduce physician burnout and improve documentation and has been well received by clinicians thus far. Getting those frontline clinicians away from the computer and engaged again with the patient is exciting.
In my opinion, there are a few topics surrounding AI that still need to be addressed before it can radically change healthcare.
- We need AI governance best practices. While not the sexiest topic, it is essential to set up standardized data intake, evaluation, and review processes. When you have multiple stakeholders starting AI projects without these operational items in place, it becomes very difficult.
- The “black box” of AI. Before we can claim transformation, quality questions need to be resolved. What kind of oversight is there in the approval process and anytime the algorithm is “updated”? Recognizing that some algorithms are generated via proprietary methods, I think we need a disinterested third party to validate the safety and efficacy of these algorithms to help ensure that they are effective and do not introduce bias. Without this quality oversight, buy-in will be difficult over the long term with clinicians. Appropriate trust needs to be addressed. One promising update was the recent formation of the Artificial Intelligence Industry Innovation Coalition (AI3C) to provide recommendations on how to accelerate adoption by gathering use cases and best practices.
- AI should not add a new workflow for a clinician, but instead should add value to an existing workflow or patient encounter. Physicians will not consistently adopt AI that requires them to step outside of their already overburdened workflows and documentation needs.
Given the current state, I would recommend focusing on AI applications that provide actionable data that can be used in solving a priority problem like improving patient access or lowering documentation requirements for clinicians. Evolve your governance and operational processes, look for best practices and keep an eye on the industry to prepare for a future of increasing AI use and dare I say, transformation?
Brett Oliver, MD, is CMIO at Baptist Health System KY & IN, as well as serving as a member of the ONC Core Data for Interoperability Taskforce and HITAC.