Sharing is caring: A systematic review of publicly available intensive care data sets

Multiple Intensive Care Unit (ICU) databases have been publicly released to advance data driven intensive care medicine. However, these public ICU data sets are prone to changes, updates and new releases. Therefore, the goal of this review is to provide clinicians and data scientists with a state-of-the-art overview and guide for choosing the relevant ICU data sets for their respective research questions.

A systematic search was carried out in PubMed, PhysioNet, Arxiv, MedRxiv and BioRxiv to identify all publicly available intensive care data sets of adult patients. After data extraction of database characteristics, a qualitative synthesis of results was carried out.

882 publications were identified. After screening, 7 publicly available ICU databases were included for analysis: AmsterdamUMCdb, eICU Collaborative Research Database (eICU-CRD), High Time Resolution ICU Dataset (HiRID), Medical Information Mart for Intensive Care (MIMIC)-IV, Northwestern ICU (NWICU), Salzburg Intensive Care database (SICdb) and Zigong Fourth People's Hospital (ZFPH) database of patients with infections. A qualitative synthesis showed notable differences in number of patients, usage of organ support, admission types and frequency of measurements.

Each public ICU data set differs due to differences in medical practice, information technologies and approach to legal restrictions. This systematic review provides clinicians and data scientists with an overview of available public ICU data sets and their characteristics.