Purpose-The present paper attempts to present the research that has been made on prediction models using deep learning methods with data retrieved from mimic III database and to identify challenges and possible areas for future research.
Methodology-A literature research was conducted for articles related to MIMIC III and prediction models related to the database published from 2016 to 2021. Also, reviews and papers related to neural networks, machine learning, data mining and implementation and usage of electronic health records (EHR) in ICU were investigated to support findings from mimic III papers.
Findings-Prediction algorithms can be very useful in ICU units. Although some algorithms, such as InSight are specialized in specific diseases, others such as XGBOOST and recurrent neural networks can be used in a broader area, presenting quite accurate results.
Originality-Usually, reviews categorize research on MIMIC database per disease or per the desired outcome, such as the prediction of length of stay and the final outcome. The current study categorizes the research based on the tools, prediction models, and algorithms used. This way, it is possible to understand better how each method performs to various conditions and desired outcomes.