← Back to Digital Poster Abstracts
DP34: DEVELOPMENT AND PROSPECTIVE VALIDATION OF A PREDICTIVE RISK MODEL FOR POST-CARDIOPULMONARY BYPASS VASOPLEGIA
Ryan J Lorenzo, DO, MBA; Imani Thornton, MD
Westside Regional Medical Center
Background: Vasoplegia after cardiopulmonary bypass (CPB) is a common and clinically significant complication in cardiac surgery. It is characterized by distributive shock physiology with low systemic vascular resistance despite preserved or elevated cardiac output, resulting in high vasopressor requirements. Patients who develop vasoplegia often require prolonged ICU care and experience increased morbidity and mortality. Management is typically reactive. A tool that identifies high-risk patients before or during surgery could support earlier and more deliberate treatment.
Objective: To develop a clinically practical model that predicts the risk of post-CPB vasoplegia using routinely available perioperative variables and to prospectively evaluate its performance in a future cohort of cardiac surgical patients.
Model Development: Vasoplegia will be defined using criteria commonly applied in contemporary cardiac surgery literature: persistent hypotension within 24 hours of CPB requiring norepinephrine ≥0.1 µg/kg/min (or equivalent vasopressor dose) to maintain mean arterial pressure ≥65 mmHg, in the presence of a cardiac index ≥2.2 L/min/m² and low systemic vascular resistance (<800 dyn·s/cm5), without evidence of ongoing hemorrhage or cardiogenic shock.
Candidate predictors will be selected based on prior studies and will include preoperative variables (age, comorbidities, renal function, left ventricular ejection fraction, and medication exposure such as ACE inhibitors or ARBs) as well as intraoperative factors (CPB duration, cross-clamp time, nadir temperature, transfusion volume, and laboratory values).
After data preparation, the dataset will be divided into a development group and a validation group. The model will be constructed using logistic regression with regularization, an approach that reduces the influence of weaker or redundant variables. This helps prevent overfitting and improves performance when applied to new patients. Model settings will be selected using cross-validation within the development dataset to promote stability.
Performance will be assessed by discrimination (how well the model distinguishes between patients who do and do not develop vasoplegia) and calibration (how closely predicted risk aligns with observed outcomes). Predictors retained in the final model will be reviewed for physiologic plausibility and clinical interpretability. If appropriate, a simplified point-based risk score will be derived to facilitate bedside use.
Prospective Validation: The finalized model will then be applied prospectively to a subsequent cohort of adult patients undergoing CPB. Predicted risk estimates generated before or during surgery will be compared with observed postoperative outcomes. Discrimination and calibration will be reassessed to determine real-world performance, and recalibration will be performed if necessary.
Conclusion: This project aims to shift vasoplegia management toward earlier risk identification through development of a transparent predictive model followed by prospective validation in cardiac surgery patients.