An XGBoost machine learning model significantly outperformed conventional logistic regression in predicting acute kidney injury within 48 hours post-PCI (AUC 0.84 vs 0.73; p<0.001).
Observational (n=4,128)
Does an Extreme Gradient Boosting (XGBoost) machine learning model improve the prediction of Acute Kidney Injury within 48 hours in critically ill patients undergoing PCI compared to multivariable logistic regression?
A machine learning model using XGBoost significantly outperformed traditional logistic regression in predicting acute kidney injury within 48 hours after PCI in critically ill patients.
Effect estimate: AUC (95% CI 0.81-0.87)
Absolute Event Rate: 0.84% vs 0.73%
p-value: p=<0.001
Background: Acute Kidney Injury (AKI) is a significant complication following Percutaneous Coronary Intervention (PCI), particularly in critically ill populations. Traditional risk stratification often relies on linear assumptions that may underestimate risk in complex patients. We sought to develop and validate a machine learning (ML) model using the MIMIC-IV critical care database to predict post-PCI AKI with greater accuracy than conventional logistic regression. Methods: We performed a retrospective study utilizing the MIMIC-IV v2. 2 database. Patients 18 years undergoing PCI (2008-2019) were identified using ICD/CPT procedure codes. The primary outcome was AKI within 48 hours, defined by KDIGO criteria. We extracted 38 features, including admission vitals, laboratories, and comorbidities. The dataset was split into training (80%) and testing (20%) sets. An Extreme Gradient Boosting (XGBoost) model was trained with 5-fold cross-validation and compared to multivariable logistic regression (LR) using the Area Under the Receiver Operating Characteristic curve (AUC). Results: The cohort included 4, 128 patients (mean age 68. 4/ pm12. 1 years; 31. 2%female). The overall AKI incidence was 14. 3% (n=590). In the independent test set, the XGBoost model achieved an AUC of 0. 84 (95% CI: 0. 81-0. 87), significantly outperforming the logistic regression model (AUC 0. 73; p<0. 001). The ML model demonstrated superior sensitivity (76% vs. 62%) at matched specificity levels. Feature importance analysis identified admission lactate, BUN, and RDW as top predictors, distinct from static risk factors like age. Conclusions: In this large analysis of critically ill patients undergoing PCI, a gradient-boosting ML model significantly outperformed conventional statistical methods in predicting acute kidney injury. Integrating such models into electronic health records could facilitate automated risk stratification and trigger targeted preventive strategies for high-risk patients.
Masab Mansoor (Wed,) conducted a observational in Acute Kidney Injury following Percutaneous Coronary Intervention (n=4,128). Extreme Gradient Boosting (XGBoost) machine learning model vs. Multivariable logistic regression was evaluated on Acute Kidney Injury (AKI) within 48 hours, defined by KDIGO criteria (AUC, 95% CI 0.81-0.87, p=<0.001). An XGBoost machine learning model significantly outperformed conventional logistic regression in predicting acute kidney injury within 48 hours post-PCI (AUC 0.84 vs 0.73; p<0.001).