Global Chief Medical Officer | Former President @ American Medical Association (2023-2024) | Board Member & Advisor
An AI alert is not a patient outcome. At Newsweek's AI Health Summit in New York, I was encouraged by how often the conversation moved beyond what the technology can do to what *actually* changes for patients. As a physician that is the question I care about most. Finding an abnormality is only the beginning. Does the right clinician see it? Is someone responsible for the next step? Does the patient receive appropriate care sooner? Without those connections, more detection can simply create more unfinished work. The discussions reinforced something central to my work at Aidoc: clinical AI implementation needs to be a care delivery effort, with technology supporting it. That means involving physicians, nurses & operational teams from the start, designing reliable handoffs, and measuring whether care actually improves not just whether a tool gets used. Administrative efficiency matters. But we should be just as ambitious about preventing harm, shortening delays in diagnosis & treatment and giving clinicians more time to care for people. I was especially encouraged by the discussion about sharing implementation lessons across health systems. We should not all have to rediscover the same workflow challenges independently! Sharing what works, what fails & how we evaluate performance should be part of the work particularly to help extend the benefits to rural and less resourced settings. I left energized by the possibilities but also reminded of our responsibility to distinguish enthusiasm from evidence. The goal is not a hospital with more AI. It is a health system where fewer patients fall through the cracks. Thank you to Newsweek and the clinicians, health system leaders & innovators who brought such practical perspectives to the conversation. #PatientSafety #ClinicalAI #HealthcareInnovation
This is a crucial point. Bridging the gap between AI detection and effective patient care requires robust systems and clear accountability. It's about designing workflows that ensure AI insights translate into real-world impact.
This is the difference between detecting a signal and delivering an outcome. Every AI alert needs a named owner, response SLA, escalation path, and closed-loop confirmation. Otherwise, faster detection only creates a larger queue of unfinished work.
YES. The measure of successful AI in healthcare shouldn’t be “Are we using it?” It should be “Is care actually better because of it?” And I’d add one more question: What is it doing to the HUMAN BEINGS delivering that care? In my work with physicians, I see just how much they’re already carrying. The best technology shouldn’t become one more thing clinicians have to manage, learn, click, monitor or work around. It should help remove friction. Give time back. Make collaboration easier. And ultimately allow physicians and care teams to spend MORE of themselves on the part technology can’t replace — caring for the human being in front of them. Better outcomes for patients AND a better way of delivering care for the people responsible for those outcomes. That’s a healthcare AI win I can get behind. 💛
This is the exact structural math missing from most clinical AI deployments. An AI tool can flag an anomaly, but if the backstage handoffs between the clinical and operational teams are undocumented, the alert simply becomes another point of friction for an already exhausted staff. You cannot automate a broken floor. Technology provides the detection, but the operational architecture dictates the outcome. Excellent diagnostic on the reality of care delivery.
The difficult part often starts after the alert. If there isn’t a clear owner, reliable handoff, and defined next step, even a very accurate system can end up creating another notification without really changing what happens to the patient.
I love the patient centered approach. I am trying to change the narrative by calling it the patient-doctor relationship instead of the other way around. The patient should come first, especially as we navigate the complexities of AI having entered the healthcare space.
Spot on, Jesse. Measuring whether care actually improves, not just whether a tool gets used, is the standard clinical AI should be held to. Framing implementation at Aidoc as a care delivery effort first is exactly the right call!
William S.
1w
What a great experience. Thanks for sharing Jesse.
This is the piece that gets missed on every go live I have run. An alert only helps if someone is clearly responsible for the next step, and when that handoff is not designed on purpose it just becomes one more thing for a nurse to chase down. The systems that actually stuck were the ones where we measured whether care changed, not whether the dashboard showed high usage.