Iron deficiency and risk of acute kidney injury after TAVI: a machine learning analysis.

Angelica Cersosimo, Crina-Ioana Radulescu, Marianna Adamo, Chiara Camastra, Catalina-Andreea Parasca, Silviu Dumitrascu, Dan Deleanu, Robert Troncea, Vlad A Iliescu, Marco Metra, Ovidiu Chioncel

Journal: Minerva medica 2026;117(3):103-114

PMID: 42312725

View on PubMed

Abstract

BACKGROUND

Transcatheter aortic valve implantation (TAVI) has become the preferred therapeutic option for patients with severe aortic stenosis (SAS) at high or intermediate surgical risk. Despite its minimally invasive nature, TAVI is frequently associated with post-procedural complications, among which acute kidney injury (AKI) represents a major clinical concern. Early identification of patients at increased risk for AKI remains a critical unmet need.

METHODS

This retrospective, observational, single-center study enrolled consecutive patients with severe aortic stenosis who underwent TAVI between December 2022 and December 2023. The primary endpoint was to identify independent predictors of acute kidney injury (AKI) within 48 hours after TAVI. Secondary endpoint was to identify independent predictors of hospitalization for heart failure (HHF) during follow-up. Predictive modeling was conducted using the eXtreme Gradient Boosting (XGBoost) algorithm, with interpretability enhanced via SHapley Additive exPlanations (SHAP).

RESULTS

A total of 116 patients were included (mean age 76±6.5 years; 57% male). Comorbidities were common, including heart failure (41%), coronary artery disease (32%), atrial fibrillation (41%), diabetes mellitus (29%), and chronic kidney disease (34%). Median EuroSCORE II was 2.1 (IQR 1.5-3.9). Median ferritin was 99.5 ng/mL (IQR 40-129), mean hemoglobin 12.3±1.7 g/dL and mean eGFR was 69.9±24.9 mL/min/1.73 m². After TAVI procedure, AKI occurred in 9% of patients within 48 h. According to the SHAP analysis, ferritin and TSAT emerged as the two most influential features in AKI prediction. The machine learning model achieved high sensitivity (96%) and an F1-score of 0.98, suggesting strong clinical utility for early risk stratification.

CONCLUSIONS

In patients undergoing TAVI, iron deficiency, as reflected by low ferritin and TSAT levels, emerged as independent predictors of early post-procedural AKI.

Address: Institute of Cardiology, Department of Medical and Surgical Specialities, Radiological Sciences, and Public Health, Spedali Civili di Brescia, University of Brescia, Brescia, Italy.; Department of Experimental and Clinical Medicine, Magna Graecia University, Catanzaro, Italy.; Emergency Institute for Cardiovascular Diseases "Prof Dr CC Iliescu", Bucharest, Romania - [email protected].; Carol Davila University of Medicine and Pharmacy, Bucharest, Romania.; Institute of Cardiology, Department of Medical and Surgical Specialities, Radiological Sciences, and Public Health, Spedali Civili di Brescia, University of Brescia, Brescia, Italy.; Cardiothoracic Department, Spedali Civili di Brescia, Brescia, Italy.; Department of Medical and Surgical Sciences, Neuroscience Research Center, Magna Graecia University, Catanzaro, Italy.; Emergency Institute for Cardiovascular Diseases "Prof Dr CC Iliescu", Bucharest, Romania.; Interventional Cardiology Department, Monza Ares Centers, Bucharest, Romania.; Emergency Clinical Hospital, Bucharest, Romania.; Emergency Institute for Cardiovascular Diseases "Prof Dr CC Iliescu", Bucharest, Romania.; Carol Davila University of Medicine and Pharmacy, Bucharest, Romania.; Institute of Cardiology, Department of Medical and Surgical Specialities, Radiological Sciences, and Public Health, Spedali Civili di Brescia, University of Brescia, Brescia, Italy.

DOI: 10.23736/S0026-4806.26.09884-8

MeSH Terms: Humans, Acute Kidney Injury, Retrospective Studies, Aortic Valve Stenosis, Female, Transcatheter Aortic Valve Replacement, Male, Aged, Postoperative Complications, Machine Learning, Aged, 80 and over, Risk Factors, Anemia, Iron-Deficiency, Predictive Learning Models, Boosting Machine Learning Algorithms, Heart Failure, Risk Assessment