Peter Safar helped create modern resuscitation medicine. He was a pioneer of cardiopulmonary resuscitation, founded the first multidisciplinary critical care medicine training program in the U.S., and trained the physician widely recognized as the first pediatric intensivist. Some readers may recognize his name from the television series “The Pitt,” which highlighted his role in training Pittsburgh’s pioneering Freedom House paramedics. His contributions to standardized resuscitation helped revolutionize systems of care for the critically ill by turning emergency response from a series of isolated acts into a coordinated discipline built around physiology, training, teamwork, and meaningful neurologic recovery.
That legacy gave the 23rd Annual Safar Symposium a fitting question: What would it mean to build an intelligent health system? Over two days, speakers addressed AI, physiologic monitoring, data infrastructure, clinical decision support, simulation, and team performance. The strongest message was not that new technology will solve the hardest problems in resuscitation. It was that better outcomes will depend on whether we can connect new tools to sound science, thoughtful implementation, and the realities of bedside care.
Vincent Liu extended Safar’s legacy by asking us to view the health system itself as a patient. A health system can be observed, measured, diagnosed, treated, and reassessed. It can also fail to learn. Liu described the practical requirements for responsible implementation at scale, including workflow design, governance, continuous evaluation, and trust. That distinction matters because a model can accurately predict deterioration and still fail to improve care. It may identify risk without clarifying what should happen next. It may work in one hospital but not another. It may produce more alerts without producing better decisions. The promise of AI is real, but so is the risk of mistaking prediction for progress.
Romain Pirracchio described sepsis care as a sequence of decisions rather than a single prediction problem. Patients change over time, and treatment should respond to those changes. His work on dynamic treatment strategies asks a more useful question than whether a patient is high risk. It asks which action is most appropriate for this patient at this point in the course of illness. Adam Ferguson focused on the foundations required to make those decisions trustworthy through data stewardship in neurotrauma research. His work aims to make complex data findable, accessible, interoperable, and reusable so that discoveries can be reproduced and translated across institutions. Together, these talks underscored that intelligence depends as much on the quality of the system beneath the model as on the model itself.
Several presentations showed how these ideas are reaching the bedside. Mohammed Shaik presented work on intracranial pressure responsivity in children with TBI. Rather than examining intracranial pressure as an isolated number, his approach evaluates how it changes in relation to blood pressure, carbon dioxide, and heart rate. These relationships may offer a clearer view of cerebral autoregulation and intracranial compliance as injury evolves. Hadi Daher described a smart blood pressure cuff designed to estimate cardiac output and elevated venous pressure in addition to blood pressure. The goal is to gather advanced hemodynamic information through a familiar and noninvasive device. Eddie Pérez Claudio presented work combining clinical knowledge with causal structure learning to develop simpler models of acute brain dysfunction in critically ill children. The aim is not only to predict risk, but to build models that are more interpretable, more robust, and less dependent on large numbers of loosely related variables.
The second day broadened the discussion from technology to human performance. Aaron Calhoun described the International Liaison Council on Simulation and its effort to establish a more rigorous global research agenda for simulation. That work recognizes that resuscitation depends on teams functioning well under pressure. Better devices and better algorithms will matter little if clinicians are not trained to use them effectively.
The opportunities in resuscitation research are remarkable. We can measure physiology with greater precision, learn from larger and more diverse datasets, and support decisions in ways that were not previously possible. The threats are equally real. Weak data, poorly designed studies, premature deployment, automation bias, and fragmented implementation could waste those opportunities or cause harm. Progress will require restraint as well as ambition.
The Safar Symposium succeeded because it took both sides seriously. It was hopeful without being naive and sophisticated without losing sight of the patient. The meeting showed a field that is not merely impressed by new tools, but increasingly prepared to ask whether they are useful, trustworthy, and worthy of adoption.
Resuscitation science has always been an expression of practical hope. We study cardiac arrest, brain injury, shock, and trauma because we do not accept that the worst moment of a person’s life must determine the rest of it. The tools are changing, but the purpose remains the same. We are still trying to see more clearly, act more wisely, and give more people the chance to recover.
What should define an “intelligent” health system? Share in the comments.
