Sharad Patel, MD
Attending Physician
Cooper University Hospital, United States
Disclosure information not submitted.
Title: Serum Albumin as a Predictor of ICU Mortality in Patients Undergoing Cardiac Surgery
Introduction: Serum albumin (SA) is a commonly used data point to risk stratify patients undergoing cardiac surgery. Intraoperative SA concentration decreases up to 50-70% due to bypass priming via hemodilution [2-4]. Albumin has been previously demonstrated to be an in-vivo inhibitor of Angiotensin-Converting Enzyme (ACE) [5]. A decrease in SA increases coronary vascular resistance and decreases coronary blood flow [5]. Low preoperative SA levels correlate with increased postoperative complications and overall mortality [6]. Lower SA levels are associated with a higher risk of myocardial damage [1]. We aim to discern the ICU mortality predictive strength of SA in patients undergoing cardiac surgery with additional analysis to find specific risk thresholds.
Methods: Data was extracted from the eICU Collaborative Research Database. Patients undergoing cardiac surgery who were greater than 18 years old were included. Our final dataset included 6578 patients. Mortality was the dependent variable for our prediction models. Exploratory data analysis was performed using Python software. Recursive feature elimination was used to identify the 10 features most predictive of mortality. We used a gradient boosting decision tree algorithm variant called XGBoost as our model. We used AUC as the metric for model performance. Feature importance was assessed via SHAP Python libraries. SHAP values reflect the relative contribution of the feature in the reduction of prediction error. To better characterize SA cutoff values below which mortality rises, we created a Partial Dependence Plot (PDP). PDP demonstrates the marginal predictive contribution of a feature based upon the feature value.
Results: A SHAP plot was created comparing SA, creatinine, age, diabetes, platelet count, INR, lactate, and hematocrit as predictors of mortality. Of these data points, SA was the strongest predictor of mortality (Figure 1, AUC = 0.86). A PDP comparing SA and mortality demonstrated a sharp increase in mortality when SA drops to below 3.5g/dL (Figure 2).
Conclusion: Our analysis of cardiac surgery patients demonstrated that pre-operative SA was a stronger predictor of mortality than other conventional metrics. Mortality significantly increases with SA under 3.5 g/dL indicating that more studies should be conducted targeting this specific threshold.