A ML and DL models to predict the Readmission or Death after Discharge from ICU

<bold>Background:</bold> Despite initial recovery from critical illness patients requiring intensive care unit (ICU) admission, many patients remain risk of subsequent deterioration or death. And unplanned Readmission to the ICU has been associated with worsening of the original disease process, higher hospital costs, and increased in hospital mortality. The purpose of this study was to identify the risk factors and assess the performance of ML and DL models that predict readmission or death within 7 days after discharge from the ICU. <bold>Methods:</bold> We defined as a readmission to the ICU or unexpected death within 7 days after discharge from the ICU to the ward. This study introduces a clinical decision support system and verifies it for predicting real-time readmission or death occurrences within the ICU by using MIMIC-IV. We develop various machine learning (ML) and deep learning (DL) models for predicting the readmission or death in real time using the MIMIC-IV. To address the challenge of missing values in time series, we propose a novel recurrent neural network model, GRU-D++. <bold>Results:</bold> There were a total of 3398 cases in the death or readmission within 7 days after ICU discharge, and 46,863 cases were successfully discharged. Among the ML models, the RF model showed an accuracy of AUROC 0.799. The fine-tuned GRU-D++ model(DL model) demonstrated superior performances, achieving an AUROC of 0.800. The proposed GRU-D++, designed to consider temporal information and missing values, stands out for its predictive accuracy. <bold>Conclusion:</bold> Our findings suggest that this model could be a tool to early detect and treat patient deterioration after ICU discharge and prevent ICU readmission.