Work

ECG-based deep learning estimation of VO2 prognosticates outcomes in colorectal surgery

Year: 2026

Type: article

Abstract: Cardiorespiratory fitness is a key determinant of perioperative risk and informs decisions regarding operative candidacy, surgical approach, and postoperative care. In routine practice, this is often assessed pragmatically through functional capacity—such as the ability to climb two flights of stairs—or through clinician- or performance status, particularly in oncologic care.1,2 While these approaches are valuable, cardiopulmonary exercise testing (CPET) provides an objective quantification of functional capacity be measuring peak VO2. Unfortunately, routine CPET use for risk stratification is not pragmatic as CPET is resource intensive and not widely accessible. Existing risk models rely largely on comorbidities and may incompletely capture this dimension of physiologic reserve. The 12-lead electrocardiogram (ECG) is nearly universal in preoperative evaluation and contains latent physiologic information not routinely used for risk stratification. Recent advances in deep learning enable extraction of clinically meaningful signals from ECG data, including features related to functional capacity.3–5 In this study, we apply an ECG-derived fitness metric to a colorectal cancer surgical population to test a pragmatic hypothesis: that a single preoperative ECG can provide an objective, scalable surrogate of functional capacity and stratify perioperative risk and outcomes. We therefore evaluate the association between ECG-derived fitness and postoperative complications, critical care utilization, and survival.

Source: European Journal of Preventive Cardiology

DOI: 10.1093/eurjpc/zwag268