S013: ARTIFICIAL INTELLIGENCE IN INTRAOPERATIVE BLOOD PRESSURE MANAGEMENT: A SYSTEMATIC REVIEW & FUTURE DIRECTIONS
Eric M Teichner, BA1; Natasha Doshi, BS2; Benjamin Kelser, BA3; Nina Schur, MMS, BA2; Brian Dinerman, MD4
1Sidney Kimmel Medical College at Thomas Jefferson University, Philadelphia, PA; 2Lake Erie College of Osteopathic Medicine, Bradenton, FL; 3Case Western Reserve University School of Medicine, Cleveland, OH; 4University at Buffalo Jacobs School of Medicine and Biomedical Sciences, Department of Urology, Buffalo, NY, USA
Introduction/Background: Intraoperative hypotension (IOH) is a common and serious perioperative event associated with increased risks of myocardial injury, acute kidney injury (AKI), stroke, and mortality. Current IOH management remains reactive, with interventions initiated only after hypotension has already occurred. Given the well-documented complications of IOH, there is growing interest in artificial intelligence (AI)-driven predictive models to shift toward proactive hemodynamic management. The Hypotension Prediction Index (HPI) is a commercially available AI algorithm designed to anticipate IOH before onset, allowing anesthesiologists to intervene preemptively rather than reactively. This systematic review evaluates the effectiveness of HPI-based management compared to standard care across multiple surgical settings.
Methods: A systematic search of PubMed, Scopus, and Web of Science was conducted to identify studies evaluating AI models for IOH prediction. Studies focusing on machine learning and deep learning models were included. Data were extracted on study design, sample size, AI model type, intraoperative outcomes (mean arterial pressure [MAP], time-weighted averages of MAP), postoperative complications (AKI, mortality), and predictive accuracy (area under the receiver operating characteristic curve [AUROC]).
Results: Across multiple studies, HPI-guided management consistently reduced IOH severity and duration. In a randomized trial of 60 patients, HPI significantly reduced time-weighted average MAP < 65 mmHg (0.02 vs. 0.37, P < 0.001) and maintained a higher median intraoperative MAP compared to standard care (87.5 vs. 77.9 mmHg, P < 0.001). In a multi-center study on open abdominal aortic aneurysm repair, HPI-based management successfully limited IOH to <10% of surgical time, achieving a time-weighted average MAP < 65 mmHg of 0.26 mmHg and MAP < 50 mmHg of 0.00 mmHg.
A systematic review and meta-analysis of 43 studies (22 evaluating HPI and 21 examining alternative AI models) found that HPI significantly reduced IOH duration (standardized mean difference -0.70, P < 0.001) and demonstrated excellent predictive performance with an AUROC of 0.89 (95% CI: 0.88–0.92). By comparison, non-HPI AI models achieved an AUROC of 0.79 (95% CI: 0.74–0.83). A pilot study on patients undergoing cytoreductive surgery with hyperthermic intraperitoneal chemotherapy demonstrated persistent IOH, but HPI monitoring provided early alerts for proactive hemodynamic adjustments.
Despite these intraoperative benefits, postoperative outcomes remain inconsistent. A multicenter randomized trial of 917 patients undergoing moderate-to-high-risk elective abdominal surgery found no significant reduction in the incidence of postoperative AKI (6.1% vs. 7.0%, P = 0.66), overall complications (31.9% vs. 29.7%, P = 0.52), or 30-day mortality (1.1% vs. 0.9%, P = 0.66) with HPI guidance compared to standard care.
Discussion/Conclusion: HPI-based AI prediction systems offer significant intraoperative advantages, particularly in reducing IOH duration and severity. However, postoperative benefits such as AKI prevention, complication reduction, and mortality improvement remain unproven. The superior performance of deep learning models suggests that further refinement and integration of AI-driven hemodynamic monitoring may enhance intraoperative decision-making. Future research should focus on expanding real-world validation, integrating noninvasive monitoring, and developing closed-loop systems for fully automated hemodynamic optimization.