Multimodal convolutional neural networks for the prediction of acute kidney injury in the intensive care

R. van Slobbe, D. Herrmannova, D. J. Boeke, Elia Lima-Walton, Ameen Abu-Hanna, Iacopo Vagliano

Abstract: Highlights•Research on whether clinical notes and concepts from knowledge bases can improve predictive performance in the ICU is limited•We investigate combining multiple data modalities (clinical variables, notes, and concepts) to predict AKI in the ICU•We investigate model performance when relying vs not relying on Serum Creatinine (SCr) and urine output, used to define AKI•Using clinical notes and concepts with clinical variables improved performance when not using SCr and urine output•The models that used only clinical variables transported well to the external-validation population