Photo Credit: Vm

The following is a summary of “Contrast-Induced Acute Kidney Injury (CI-AKI) in Lower Limb Percutaneous Transluminal Angioplasty: A Machine Learning Approach for Preoperative Risk Prediction” published in the March 2025 issue of Annals of Vascular Surgery by Lim et al.

Contrast-induced acute kidney injury (CI-AKI) is a well-recognized complication of lower limb percutaneous transluminal angioplasty (PTA), yet existing risk prediction models are primarily derived from percutaneous coronary intervention cohorts and rely on both preoperative and perioperative variables. These models often exclude critical preoperative factors, such as inflammatory markers and medication history, and none leverage machine learning techniques. This study aimed to develop a robust preoperative risk prediction model for CI-AKI in lower limb patients with PTA using machine learning and compare its performance against conventional logistic regression methods. A retrospective cohort of 456 patients who underwent isolated lower limb PTA between 2015 and 2019 was analyzed. 

Exclusion criteria included patients younger than 21 years, those with a preoperative estimated glomerular filtration rate (eGFR) below 15 mL/min/1.73 m² (as defined by the Modification of Diet in Renal Disease [MDRD] equation), and individuals without valid preoperative or postoperative serum creatinine (SCr) measurements. Multiple predictive models were constructed, including traditional logistic regression and machine learning approaches such as Logistic Regression with ElasticNet penalty, Random Forests, Gradient Boosting Machines, K-Nearest Neighbors (KNN), Support Vector Machines, and Multi-Layer Perceptron networks. Model performance was assessed using five-fold cross-validation with hyperparameter tuning via grid search. The predictive ability of each model was evaluated based on the area under the receiver operating characteristic curve (AUROC), the area under the precision-recall curve (AUPRC), F1 score, sensitivity, and specificity. Feature importance was analyzed using SHAP plots. 

Among the models tested, K-Nearest Neighbors demonstrated the highest predictive performance, achieving an AUROC of 0.914 and an AUPRC of 0.809. SHAP analysis identified key predictive variables, including MDRD eGFR, hemoglobin levels, and inflammatory markers such as neutrophil-to-lymphocyte ratio and red cell distribution width. These findings suggest that preoperative inflammatory and hematologic parameters play a crucial role in CI-AKI risk stratification. The development of a machine learning-based model that exclusively utilizes preoperative variables offers a practical tool for risk assessment prior to lower limb PTA. 

Such a model has the potential to enhance preoperative patient counseling by surgeons and anesthesiologists and facilitate the identification of high-risk individuals who may benefit from intensified monitoring or prophylactic interventions. Future research should validate this model in larger, prospective cohorts and explore its integration into clinical decision-making frameworks to optimize perioperative risk management for CI-AKI in patients with vascular intervention.

Source: annalsofvascularsurgery.com/article/S0890-5096(25)00098-6/abstract