Time Series Multi-task Learning for Prognosis of MICU and SICU

The prognostic assessment of an ICU patient involves assessing the severity of their condition, interventions, and length of ICU stay. Over the past 30 years, researchers have proposed numerous predictive models and severity assessment scales for ICU patients in specific regions, including APACHE II and SAPS II. However, most existing methods rely heavily on curve fitting which do not account for misclassifications caused by false negatives and positives. Specificity and sensitivity must be provided as an indicator of model performance. The primary aim in this study is to develop a machine-learning model to formulate a prognosis for MICU and SICU patients by using data from the MIMIC-IV for training. The predictive models developed in this study facilitate the prediction of mortality and other outcomes across various treatment regimens.