Predicting Acute Kidney Injury After Cardiac Surgery Using a Simpler Model.

Area of Science:

• Nephrology
• Cardiovascular Surgery
• Medical Informatics

Background:

• Acute kidney injury (AKI) and the need for renal replacement therapy (RRT) are significant complications following cardiac surgery.
• Accurate prediction models are crucial for risk stratification and timely intervention in these high-risk patients.
• Existing models may lack clinical applicability due to complexity or limited validation.

Purpose of the Study:

• To develop and validate simple, clinically applicable models for predicting AKI and RRT in cardiac surgery patients.
• To assess the predictive performance of these models using preoperative and immediate postoperative data.
• To compare the performance of parsimonious models with more comprehensive approaches.

Main Methods:

• Retrospective, multi-institutional analysis of 22,731 cardiac surgery patients (September 2016 - December 2018).
• Development and validation sets (75%/25%) utilizing preoperative and immediate postoperative data.
• Stepwise logistic regression employed to create parsimonious models; model performance evaluated using AU-ROC.

Main Results:

• Incidences of AKI (any stage) and RRT were 25.6% and 2.1%, respectively.
• Developed models showed moderate discrimination for AKI (AU-ROC 0.67-0.69) but good discrimination for RRT (AU-ROC 0.77-0.80).
• Parsimonious models (3-5 variables) performed comparably to more complex models (21-26 variables).

Conclusions:

• Simple, clinically applicable models for predicting RRT in cardiac surgery patients were successfully developed.
• While AKI prediction accuracy was limited, RRT prediction demonstrated good performance with parsimonious models.
• These models can aid in clinical and research-based risk stratification for kidney complications post-cardiac surgery.