Dear Editor,

In the past decade, data-driven research utilizing advanced statistical methodology such as complex modeling or artificial intelligence (AI) has increasingly gained popularity in the field of intensive care medicine. One catalyst for these research endeavors has been the introduction of the first publicly accessible intensive care database, MIMIC-II, in 2011 [1]. MIMIC-II with its subsequent updates has facilitated over 1000 publications in various medical fields. Subsequently, these single-center datasets have been complemented by multicenter, United States (US) datasets (eICU-CRD), and most recently by two European single-center databases (HiRID, Amsterdam UMCdb) [2, 3]. Overall, 243,429 patient entries are currently available in the 4 mentioned datasets. However, the granularity of the data is comparatively poor (with exemption of the HiRID), complicating the utilization of AI applications.

Hence, we present the Salzburg Intensive Care database (SICdb), a publicly available, highly granular, once-per-minute critical care dataset [4]. SICdb (1.0.4) contains 27,386 admissions from 4 different intensive care units (ICUs) at 1 single tertiary care institution of the Department of Anesthesiology and Intensive Care Medicine at the Salzburger Landesklinik (SALK) and Paracelsus Medical University (PMU) between 2013 and 2021. SICdb provides highly granulated data at the minute level and is preprocessed hourly. Raw signals were collected utilizing the MetaVision (iMDSoft, Tel Aviv, Israel) patient data management (PDMS) software. Additionally, exports from ORBIS (Dedalus Healthcare GmbH, Bonn, Germany), containing information on admission, discharge, and ICD–10 codes, were included. The workflow to generate the dataset is shown in Fig. 1. The export process strictly adhered to the ACID (Atomicity, Consistency, Isolation, Durability) properties of database transactions. SICdb has been fully approved by the ethics commission of the Land Salzburg, Austria (EK Nr: 1115/2021).

Fig. 1
figure 1

Workflow in SICdb. Data are provided as a MSSQL database and several Excel file exports. After analyzing the data structure, an export software to access MV (MetaVision (iMDSoft, Tel Aviv, Israel)) data has been created. After selecting all eligible cases, primary de-identification procedures were performed. All assignments are saved in a publicly not available lookup database for further validation and merging with other datasets. To ensure data quality, several validation steps are performed. Main validation procedure is cross-comparison between MetaVision and the KIS ORBIS (ORBIS (Dedalus Healthcare GmbH, Bonn, Germany)). The internal validation process additionally guaranteed that all electronically available data were exported and that no data, with exemption of protected health information (PHI), was omitted. Time-related data other than the admission year are removed. In a second stage, de-identified raw data are processed to the final database. The optional usable SICdb Server Software provides export data readable for various statistical software solutions

Full size image

All data are pseudo-anonymized as defined by the European General Data Protection Regulation. The de-identification strategy also complies with the US regulations for health data.

Overall, SICdb contains more than 1.2 billion data points. The mean length of stay in SICdb is 3.50 [SD 6.49] days. Twenty-six thousand four hundred and forty-six [96.55%] patients were alive at ICU discharge and twenty-five thousand two hundred and fifty-two [92.2%] were alive at hospital discharge, respectively. For 24,243 [88.5%] admissions, 1-year out-of-hospital survival data are available. Nineteen thousand six hundred and eighty-nine [81.21%] were still alive 1 year after hospital discharge. The main admission reason to the ICU was postoperative treatment. The mean Simplified Acute Physiology Score III (SAPS III) is 44.67 [SD 14.81]. Two thousand four hundred and six [8.79%] patients were mechanically ventilated > 24 h and one thousand and twenty-seven [3.75%] patients required renal replacement therapy (RRT).

A detailed description of the data, access credentials, and documentation on the removal of all protected health information (PHI), can be found at https://www.sicdb.com. To access and download SICdb, the online repository for physiological signals PhysioNet (https://physionet.org/content/sicdb/1.0.5/) was used [5]. It is mandatory to cite this paper in subsequent analyses of the database.

In conclusion, SICdb is among the largest ICU datasets currently available worldwide. SICdb is designed for continuous enhancement, developed, and growth, including consistently evolving patient cases. SICdb’s exceptional high granularity, coupled with its up-to-date case data, holds substantial value for future research, particularly in terms of potential AI applications.