Abstract
Background: Acute kidney injury in infective endocarditis presents significant management challenges in intensive care unit (ICU). We explored the role of mean blood pressure(MBP) at the time of ICU admission predicting kidney function outcomes in endocarditis patients using deep learning model, Generative Adversarial Nets for inference of Individualized Treatment Effects (GANITE).
Methods: This study utilized data from the Medical Information Mart for Intensive Care III database. Patients with infective endocarditis admitted to intensive care unit were included in this study. A machine learning model was developed to predict the kidney function deterioration. SHapley Additive exPlanations (SHAP) were used to understand how variables affect kidney function. Moreover, the GANITE model, a causal inference deep learning model, was used to determine the effect of blood pressure to kidney function.
Results. A total of 484 patients were included in the analysis, among whom 85(17.6%) experienced kidney deterioration. Light gradient boosting machine, extreme gradient boosting, and the ensemble model showed area under the receiver operating characteristics of 0.790, 0.772, and 0.785, respectively, on the test data, all achieving an accuracy of 0.828. SHAP value plots revealed that higher blood pressure predicted a lower likelihood of kidney deterioration. Analysis using the GANITE model revealed that maintaining MBP≥65mmHg resulted in a decrease in the probability of kidney deterioration by 12.9%.
Conclusions: In patients with infective endocarditis in ICU, the maintenance of MBP≥65mmHg prevented the future kidney function deterioration after ICU admission.