Development and External Validation of a Machine Learning Model Based on Preoperative Nutritional Status for Predicting Acute Kidney Injury After Coronary Artery Bypass Grafting

Why It Matters

Integrating objective nutritional metrics into AI‑driven risk models improves early AKI detection, potentially reducing morbidity and costs in cardiac surgery.

Key Takeaways

• PNI outperforms CONUT and GNRI for AKI prediction
• GBM model achieved 0.905 AUC in external validation
• SHAP identified PNI, LVEF, CPB as top risk drivers
• Web calculator enables bedside AKI risk estimation
• Malnutrition is independent AKI risk factor in CABG patients

Pulse Analysis

The relationship between malnutrition and renal complications has long been suspected, yet few studies have quantified this link in the context of coronary artery bypass grafting. By applying three validated nutritional screening tools—CONUT, GNRI, and especially the Prognostic Nutritional Index—the researchers demonstrated that lower PNI scores correlate with a markedly higher incidence of postoperative AKI. This finding underscores the value of incorporating comprehensive nutritional assessments into pre‑operative evaluations, moving beyond traditional markers like serum albumin or body mass index that lack specificity.

Beyond statistical associations, the study leveraged advanced machine‑learning techniques to translate nutritional data into actionable clinical insight. Six algorithms were compared, and the gradient boosting machine emerged as the top performer, maintaining an AUC above 0.90 across internal and external validation sets. Explainable AI methods, particularly SHAP values, revealed that PNI, left ventricular ejection fraction, and cardiopulmonary bypass duration drive the model’s predictions. Such transparency addresses a common barrier to AI adoption in surgery, allowing clinicians to understand and trust the risk estimates generated.

The practical outcome—a publicly accessible RShiny calculator—bridges research and bedside care. Surgeons can input a patient’s PNI, cardiac function, and operative variables to receive an individualized AKI probability, facilitating targeted interventions like optimized fluid management or early nephrology consultation. By highlighting nutrition as a modifiable risk factor, the work invites broader peri‑operative strategies, including pre‑habilitation and dietary optimization, to mitigate AKI risk and improve overall surgical outcomes.