
SepsisExp: A Dataset of Patient Timelines with Expert Sepsis Labels
===================================================================

SepsisExp is a dataset based on data collected at the University Medial 
Centre in Mannheim, Germany (UMM). Timelines of 42 features were extracted 
from the Intellispace Critical Care and Anesthesia (ICCA) system by Philips 
(Eindhoven, Netherlands), and 1 demographic feature was extracted from the 
HIS system by SAP (Walldorf, Germany). Expert sepsis labels were collected 
using a electronic questionnaire implemented on a tablet computer.


Terms of Use
------------
This data is available under a [Creative Commons Attribution 4.0 
International (CC BY 4.0)](http://creativecommons.org/licenses/by/4.0/). 

Please cite [1] if you use this data in your work.


Data
----
Each partition file contains tab-seperated values (tsv) with the first line 
containing the name of each column. Clinical measurements are z-scored.

The four partitions of the dataset were constructed such that the ratio of 
sepsis and non-sepsis patients stays similar. The following table lists 
distribution of hospital stay times and sepsis onset in hours for sepsis and 
non-sepsis patients across the four partitions:

Partition   Patients (Non-/Sepsis)  Median/Min/Max[h]   Median/Min/Max[h]
-------------------------------------------------------------------------
A           319 (245/74)            56.5/15.5/2253.0    159.0/50.0/1394.0
B           319 (245/74)            55.5/15.5/1218.5    162.0/49.0/972.5
C           319 (245/74)            56.0/15.5/1618.5    161.0/51.0/1257.0
D           318 (244/74)            56.0/15.5/831.5     161.5/48.5/1269.5
-------------------------------------------------------------------------
Total       1275 (979/296) 


Version History
---------------
v1: released August 1st, 2022. Initial version.


References
----------
[1] Schamoni, Shigehiko and Hagmann, Michael and Riezler, Stefan (2022): 
    Ensembling Neural Networks for Improved Prediction and Privacy in Early 
    Diagnosis of Sepsis. Proceedings of Machine Learning Research, 182, PMLR, 
    Durham, NC, USA.


