The Yankee legend and baseball Hall of Famer Yogi Berra famously once said, "It's like déjà vu all over again!" It's one of his most famous Yogi-isms, which he said after watching Yankee teammates Mickey Mantle and Roger Maris hit back-to-back home runs during the 1961 baseball season (this was the season that Maris broke Babe Ruth's longstanding record for most home runs hit during a single season). The phrase is now a colloquialism used when someone is experiencing a situation that feels like an exact repetition of something that happened in the past.
Eric Poon, Andrew Rosenberg, Adam Landman, and Tejal Gandhi published an article this past year that asked whether we are experiencing a "déjà vu" moment with the implementation of artificial intelligence (AI) in health care (see "Déjà Vu? How Might Lessons Learned from Electronic Health Record Implementation Apply to Artificial Intelligence?"). Poon, Rosenberg, and Landman are the Chief Information Officers from Duke University Health System, Michigan Medicine, and Mass General Brigham, respectively, while Gandhi is the Chief Safety and Transformation Officer at Press Gainey. They offer several recommendations on how we should be applying lessons learned from the widespread adoption and implementation of electronic health records (EHRs) to the implementation of AI in health care today.
The authors cite statistics showing that the widespread adoption of EHRs have contributed to an estimated $150 billion per year in health information technology spending in the U.S. alone. That would certainly be okay if EHR implementation was associated with significant improvements in quality of care, patient safety, and clinician productivity, though the results here have been mixed at best. In addition, the EHR (perhaps unfairly) is often cited as a major cause of burnout amongst health care workers. Suffice it to say, there are likely some lessons that we can learn from EHR implementation that are translatable to adoption of AI.
The authors first begin with discussing the key differences between the era of EHR implementation and now. First, the HITECH Act of 2009 provided substantial financial incentives for hospitals and providers to adopt EHRs and included delayed penalties for failing to do so. These incentives and delayed penalties pushed health care organizations to adopt EHRs in a short period of time. While some executive leaders are mandating AI adoption and use, to date at least, there have been no federal mandates or incentive programs to do so. Rather than following the EHR "big bang" approach to widescale adoption, AI implementation could occur in a more phased and flexible manner. That being said, while EHR implementation occurred in a top-down fashion, the bottom-up adoption of AI could outpace what organizations are actually ready to implement, particularly in the absence of an overall roadmap to help guide AI implementation.
Second, the overall computer literacy of the health care workforce is quite different today compared to when EHRs were first rolled out. Technology, including AI, already plays a major role in our personal and professional lives. For example, smartphones are ubiquitous now - they were much less so back when EHRs were being implemented. In addition, many health care workers are likely already using AI in some form at home (and in work). However, while the workforce is likely already facile and comfortable with technology, the levels of burnout amongst health care workers is high enough that they will have diminished capacity and interest to take on additional work. There is also significant concerns that AI could replace humans when it comes to certain clinical tasks, leading to fears about some clinical jobs being eliminated (I've discussed this topic to some extent recently - see my post, "Is AI going to replace us?").
Third, as I alluded to earlier, the pace of change in regards to AI implementation is likely going to be much faster. EHR technology evolved slowly over decades, and up until the HITECH Act, implementation was relatively slow and infrequent. AI technology, in contrast, is proliferating at a much faster pace and has been particularly fueled by interest in industries outside of health care.
Poon, Rosenberg, Landman, and Gandhi discuss five key lessons learned based on their collective (and extensive) experience and the available literature on EHR and AI implementation:
1. Respect the Human Element
Processes and workflows need to be designed in advance for optimal efficiency and use. Automating a flawed process will only perpetuate inefficiencies and lead to frustration. A key lesson from EHR implementation comes from the human factors field - the distinction between work-as-imagined or work-as-prescribed versus work-as-done. AI systems should be designed so that they are intuitive, user-friendly, and tailored to the specific needs of the end users.
2. Build Strong Organizational Governance
Particularly during the early days of EHR implementation, governance was loose and decentralized. Robust governance around AI implementation will be crucial so that we do not end up with tools with overlapping and redundant capabilities that interoperate poorly with each other. The "problem to be solved" should be identified beforehand, rather than bringing in a technology first and finding a problem to be solved later.
3. Adapt Leadership and Culture to Enable Change
Retired Marine Corps General and former Secretary of Defense James Mattis once wrote, "A leader's role is problem solving. If you don't like problems, stay out of leadership." I would paraphrase General Mattis slightly, and say that a leader's role is about leading and managing change. If you don't like change, stay out of leadership! Upfront investments to prepare both leaders and teams to deal effectively with the rapid pace of change involving AI technology will pay huge dividends in the long-run.
4. Ready the Workforce for an AI-Enabled Future
Many health care organizations failed to invest in computer literacy and typing skills when they made the shift from paper-based medical records to electronic ones. We should not repeat that mistake. Health care organizations should invest in AI and digital literacy in order to adequately prepare the workforce and fully leverage the benefits of AI.
5. Avoid Short-term Hype and Build for the Long Term
Remember that it's a marathon, not a sprint. AI is here to stay. Health care organizations should commit to measuring the impact of AI implementation. If an AI tool is not working, these same organizations should be prepared to either modify the tool or move on from it.


