Understanding Chronic Kidney Disease After Cardiac Surgery: An AI-Driven Approach
Chronic kidney disease (CKD) is a significant long-term complication that can arise following cardiac surgery. This condition is associated with increased morbidity and mortality, making early detection essential for improving patient outcomes. Often, CKD progresses silently, remaining undetected until it reaches an advanced stage. In response to this challenge, researchers have developed and evaluated an explainable artificial intelligence (XAI)-based model designed to identify patients at high risk of developing CKD after undergoing cardiac surgery.
Methods and Analysis
The development of this innovative model involved the extraction of over 200 clinical variables from cardiac surgery patients treated at Odense University Hospital in Denmark, spanning the years 2000 to 2022. This data, obtained from the Western Denmark Heart Registry, was combined with biochemical data from regional laboratory systems. Importantly, patients who had preoperative renal impairment or missing data required for CKD determination were excluded from the study.
To ensure a balanced dataset, researchers considered age, gender, the decade of surgery, and CKD occurrence. The dataset was then divided into training, validation, and test samples. The team utilized a XAI algorithm known as QLattice, which employs symbolic regression to generate predictive models. An external evaluation of the model was conducted using data from cardiac surgery patients at Aarhus University Hospital, Denmark, covering the period from 2008 to 2024. The model’s performance was evaluated using receiver operating characteristics, focusing on the area under the curve (AUC), and calibration plots.
Results
The study analyzed data from 11,156 patients, revealing that the unadjusted incidence of de novo CKD cases was 13% at three years post-surgery and increased to 18% at five years. Notably, 47% of all CKD cases developed within three years of discharge. Key predictors for CKD development included baseline estimated glomerular filtration rate, perioperative creatinine increase, age, and gender. The model demonstrated strong performance with an AUC of 0.86 and exhibited good calibration. During external evaluation on 9,479 patients, the model maintained an AUC of 0.88 with comparable calibration following intercept recalibration.
Conclusion
The development and evaluation of this XAI-based model mark a significant advancement in predicting CKD risk among cardiac surgery patients. With its robust performance and readiness for clinical implementation, the model offers an opportunity for improved interdisciplinary tracking of kidney function post-surgery. This tool stands to enhance patient care by enabling earlier interventions and better management of CKD risk factors.
For further details on this study, please refer to the original publication Here.
“`
