Beyond the Hype: The Pragmatic Integration of Intraoperative Artificial Intelligence in Cardiac Surgery Authors: J. Luti Nonu, Julie W. Doberne Published: 2026 Underneath the day-to-day practice of modern medicine lies an undercurrent of big data, in which massive volumes of data are being generated continuously at a pace that frequently outstrips our capacity for storage, analysis, and meaningful clinical output.1 A prime example lies in the modern cardiovascular operating room (OR). During a single procedure, continuous, high-velocity data streams from hemodynamic monitoring, cardiopulmonary bypass (CPB) parameters, anesthesia machine and ventilator outputs, live intraoperative echocardiography, and rapid-cycle laboratory and coagulation monitoring. The recent publication in this journal, “Rethinking cardiopulmonary bypass management in the digital health era” by El Dsouki et al,2 provides a compelling proposition for how we might transform this overwhelming data stream into predictive foresight. The authors rightly challenge us to imagine a future where artificial intelligence (AI) and digital health tools elevate perfusion from a reactive task to a proactive science. In their manuscript, El Dsouki et al challenge the current standard of CPB care to elevate into a much more individual, real-time tailored delivery of care through the integration of existing data streams, augmented through modern data analysis techniques such as machine learning. One particularly provocative prospect is the digital twin, a computational replica of the patient undergoing cardiac surgery, to facilitate tailored delivery of the patient’s perfusion throughout an operation. To fully realize this vision, we must first acknowledge the current frontier of digital cardiovascular management. Fully automated digital twins for CPB remain largely theoretical, but the mathematical groundwork is being aggressively laid in adjacent modalities. For example, a 2026 preprint demonstrated a highly successful in silico digital twin framework for extracorporeal membrane oxygenation (ECMO).3 This breakthrough proves that modeling complex, multivariable extracorporeal physiology in a simulated environment is mathematically viable. Alongside emerging algorithms for CPB metabolic strategies and automated temperature management, the building blocks for the digital OR exist.4,5 However, as we look to translate the conceptual vision of “Rethinking Cardiopulmonary Bypass” and the mathematical success of in silico ECMO models into everyday CPB management, we must ground our enthusiasm. The true hurdle is no longer just refining the predictive power of neural networks. The challenge is determining how to practically, economically, and safely integrate them into the deeply human, highly complex workflow of cardiac surgery. While the technological promise of predictive modeling on bypass is profound, the conversation often ignores the immense infrastructural burden required for its execution. The cost of integrating AI is not merely the price of a software license; it encompasses massive ongoing investments in localized computing power, secure cloud infrastructure, and the dedicated personnel required to maintain the data pipeline. Furthermore, deploying these models reveals a glaring systemic barrier: the transition from fragmented data gathering to unified digital ontologies. Historically, surgical and perfusion records have been siloed across disparate electronic health records or locked in physical paper charts or PDFs. Translating this nonuniversal data into the harmonized, universally tabulated formats required for real-time machine learning is a massive, costly undertaking. Because of this steep infrastructural and financial reality, healthcare economics dictates that these tools cannot immediately become a blanket standard of care. Implementing a resource-intensive digital twin for a routine, elective procedure in a healthy patient offers diminishing returns. Instead, these advanced tools must be targeted at our highest-risk populations—the frail, the elderly, and those with severe comorbidities such as preexisting heart failure.6 It is precisely in these high-risk cohorts, where the physiological margin is thinnest, that the high cost of algorithmic monitoring is justified by preventing a cascade of devastating, expensive postoperative complications. Bridging the Gap: From Prediction to Protocol Once a predictive model is economically integrated, its clinical utility hinges on clinicians’ ability to act on its warnings. In complex procedures, such as robotic-assisted valve surgeries or aortic reconstructions, a surgeon’s options to simply Abort, Speed up, or Alter the case are severely limited by anatomy, clinical indication, and mechanical setup. Knowing that an organ is failing is insufficient if the surgical team’s hands are tied. Therefore, specific implementations of AI in the OR must be appropriately interpretable and actionable. A digital "red alert" for an event like impending acute kidney injury is one such example. We must look to foundational research to build these algorithmic protocols. Lankadeva et al7 have elegantly demonstrated in experimental models of CPB that standard bypass flows often fail to adequately perfuse the renal medulla, leading to severe hypoxia and subsequent acute kidney injury. As highlighted in subsequent commentary by Joles,8 these findings underscore the clinical necessity of dynamically adjusting CPB flow or arterial pressure to maintain medullary oxygenation during surgery. Furthermore, recent work by Furukawa et al9 proves that this hypoxia can cause structural damage persisting for weeks post CPB, underscoring the critical need for immediate intraoperative intervention. Future clinical research must pivot from merely validating an AI’s predictive accuracy to pairing that prediction with specific, granular clinical responses. We need algorithmic frameworks that dictate actionable steps based on proven physiology. Instead of a vague warning, a CPB digital twin must trigger a definitive protocol: “Renal medullary hypoxia predicted in 4 minutes. Initiate targeted intervention: increase pump flow dynamics to >2.4 L/min/m2, elevate targeted mean arterial pressure to >75 mmHg via metaraminol infusion, and optimize hematocrit.” Prediction without a specific, actionable protocol is just sophisticated noise. The Evolution of the Care Team Finally, as we introduce these high-velocity data streams into the OR, we risk crippling the surgical team with cognitive overload. However, instead of dismissing these systems due to “alert fatigue,” we must recognize that alarms are only annoying when they lack actionable context. To avoid overwhelming the perfusionist, the anesthesiologist, and the surgeon, the user interface and human factors engineering of these systems are just as critical as the neural networks underlying them. These tools must be intentionally designed as true decision-support systems. This represents a necessary evolution in the partnership between technology and the clinician. The machine's role is to leverage its computational power to identify the subtle, multivariable physiological drift that escapes the human eye. The human clinician’s role remains paramount: contextualizing that drift within the broader operative reality and executing the tactical response. Conclusion The future of cardiac surgery does not belong to AI alone, nor does it belong to surgical teams that resist digital integration. As the authors of “Rethinking Cardiopulmonary Bypass” astutely point out, the era of digital health is already upon us. Success belongs to the programs that figure out how to operationalize these tools effectively. By directing our infrastructural resources toward high-risk patients, pairing predictive alerts with highly specific physiological protocols, and respecting the cognitive load of the operative team, we can bridge the gap between promise and reality. Potential Competing Interests The authors report no competing interests.