The Extreme Gradient Boost (XGB) model achieved the highest AU-ROC scores for predicting both hospital and ICU readmissions across different time windows compared to MLP and Logistic Regression.
The MiPy framework provides an open-source, standardized pipeline for benchmarking machine learning models predicting hospital and ICU readmissions using the MIMIC-IV database, enhancing reproducibility and comparability.
Abstract Avoidable and unplanned readmissions to hospital wards, especially the Intensive Care Unit, have significant implications for the patients’ health and poses additional economic burdens on the health system. If patients who are at risk of readmission are identified early and their risks are mitigated, these complications can be avoided. Machine Learning has been a valuable tool for automatic identification and prediction of various health conditions and situations, including unplanned readmissions. This is made possible through processing large collections of clinical data to build predictive models. However, the clinical data from which these models are built is highly confidential, which has restricted the ability of researchers to provide their data to the wider community, hence limiting reproducibility and comparability between results. The MIMIC databases are large, publicly available clinical datasets, which make reproducibility and comparability feasible. To maximise the benefit the research community derives from this invaluable resource, we developed MiPy, an open source standardised framework for preparing, building, and evaluating machine learning models for predicting both hospital and ICU readmission, on the MIMIC-IV database. The primary aim of this work is to enhance reproducibility and comparability of research in the field.
Partovi et al. (Wed,) conducted a other in Unplanned Hospital and ICU Readmission (n=382,278). Machine Learning Models (XGB, MLP, LR) was evaluated on Area Under the Receiver Operating Characteristic Curve (AU-ROC) for readmission prediction. The Extreme Gradient Boost (XGB) model achieved the highest AU-ROC scores for predicting both hospital and ICU readmissions across different time windows compared to MLP and Logistic Regression.