Response to “Reflection on the Integration of Artificial Intelligence in Anaesthesiology: Beyond Algorithmic Performance”

Creative Commons License

Dost B., Turan E. İ., AYDIN M. E., Ahiskalioglu A., Narayanan M., YILMAZ R., ...Daha Fazla

Turkish Journal of Anaesthesiology and Reanimation, cilt.54, sa.2, ss.148-149, 2026 (ESCI, Scopus, TRDizin)

Özet

We sincerely thank the authors for their thoughtful and insightful letter1 regarding our review.2 We appreciate their careful reading of our work and constructive reflections on the evolving role of artificial intelligence (AI) in anaesthesiology. We fully agree that a critical challenge lies in bridging the gap between the algorithmic performance and real-world clinical benefits. As highlighted in both our review and the authors’ letter, high discriminative accuracy does not necessarily translate to improved patient-centred outcomes. The example of the hypotension prediction index (HPI) is particularly illustrative: despite promising predictive metrics, the relatively low positive predictive value in clinical settings raises valid concerns regarding alarm fatigue, cognitive overload, and the potential for inappropriate therapeutic interventions.3 We concur that future investigations must increasingly prioritise outcome-driven, pragmatic clinical trials rather than surrogate technical endpoints alone. Despite these concerns, the evidentiary landscape surrounding HPI continues to evolve and extend beyond early validation cohorts. The recently published randomised controlled trial protocol by Mulder et al.4, which compared a conventional mean arterial pressure alarm strategy with HPI-guided management in moderate- to high-risk non-cardiac surgical patients within a non- inferiority framework, directly examined whether predictive waveform analytics provide incremental clinical value beyond traditional threshold-based monitoring. In parallel, emerging data from critically ill populations, such as the prospective study by Khwannimit et al.5 evaluating HPI performance in patients with septic shock in the intensive care unit, demonstrate both the potential and present limitations of the algorithm in high-risk settings. Taken together, these investigations highlight a broader physiological reality: haemodynamic instability is complex and dynamic and is shaped by multiple interacting variables. No single measurement can fully represent this complexity of the disease. Therefore, HPI should not be interpreted as a replacement for conventional monitoring but rather as a contributory element within an evolving multimodal decision- support framework that integrates pressure trends, waveform characteristics, clinical context, and physician judgment.