This dataset comprises vital signs, lab tests, administered fluid boluses and vasopressors for 3,910 patients with acute hypotension in the intensive care unit.
You can access our synthetic sepsis dataset through PhysioNet.
What you should know about this dataset
The following tables contain a list of variables in the synthetic data together with descriptive statistics.
|wdt_ID||Variable name||Data Type||Unit||Numeric Statistics|
|1||Mean Arterial Pressure (MAP)||numeric||mmHg||Median: 65.34 (Q1: 59.30, Q3: 71.19)|
|2||Diastolic Blood Pressure (BP)||numeric||mmHg||Median: 54.33 (Q1: 48.37, Q3: 60.26)|
|3||Systolic BP||numeric||mmHg||Median: 113.21 (Q1: 104.23, Q3: 121.60)|
|4||Urine||numeric||mL||Median: 106.21 (Q1: 68.92, Q3: 164.23)|
|5||Alanine Transaminase (ALT)||numeric||IU/L||Median: 32.55 (Q1: 24.59, Q3: 46.09)|
|6||Aspartate Aminotransferase (AST)||numeric||IU/L||Median: 46.82 (Q1: 35.81, Q3: 67.75)|
|7||Partial Pressure of Oxygen (PO2)||numeric||mmHg||Median: 103.02 (Q1: 91.34, Q3: 114.66)|
|8||Lactic Acid||numeric||mmol/L||Median: 1.50 (Q1: 1.29, Q3: 1.80)|
|9||Serum Creatinine||numeric||mg/dL||Median: 1.11 (Q1: 0.83, Q3: 1.62)|
|10||Fluid Boluses||categorical||mL||4 Classes (0, 250): 97.32%; (250, 500): 0.28%; (500, 1000) : 1.46% ≥ 1000 : 0.94%|
|11||Vasopressors||categorical||mcg/kg/min||4 Classes 0: 84.14%; (0, 8.4): 8.34%; (8.4, 20.28): 3.68%; ≥ 20.28: 3.83%|
|12||Fraction of Inspired Oxygen (FiO2)||categorical||-||10 Classes ≤ 0.2: 0.00%; 0.2: 0.54%; 0.3: 2.84%; 0.4: 10.85%; 0.5: 63.30%; 0.6: 8.58%; 0.7: 1.32%; 0.8: 0.20%; 0.9: 2.63%; 1.0: 9.75%|
|13||Glasgow Coma Scale (GCS)||categorical||-||13 Classes 3: 6.61%; 4: 2.16%; 5: 0.00%; 6: 3.00%; 7: 4.77%; 8: 0.00%; 9: 2.22%; 10: 4.32%; 11: 2.46%; 12: 3.56%; 13: 1.00%; 14: 9.80%; 15: 60.09%|
|14||Urine Data Missing (Urine (M))||binary||True/False||False: 63.07%, True: 36.93%|
|15||FiO2 Data Missing (FiO2 (M))||binary||True/False||False: 98.50%, True: 1.50%|
|16||ALT or AST Data Missing (ALT/AST (M))||binary||True/False||False: 92.49%, True: 7.51%|
|17||GCS Data Missing (CGS (M))||binary||True/False||False: 81.49%, True: 18.51%|
|18||PO2 Data Missing (PO2 (M))||binary||True/False||False: 97.56%, True: 2.44%|
|19||Lactic Acid Data Missing (PO2 (M))||binary||True/False||False: 96.98%, True: 3.02%|
|20||Serum Creatinine Data Missing (Serum Creatinine (M))||binary||True/False||False: 95.26%, True: 4.74%|
Comparison of marginal distributions
Comparison of marginal distributions between the real and synthetic data for continuous, categorical and binary variables.
Numeric variable comparisons
E.g. MAP, BP, Systolic BP, Urine, ALT, AST, PO2, Lactic Acid, Serum Creatinine
Categorical variable comparisons
E.g. Fluid Boluses, Vasopressors, FiO2, GCS
Binary variable comparisons
E.g. Urine (M), ALT/AST (M), FiO2 (M), CGS (M), PO2 (M), Lactic Acid (M), Serum Creatinine (M)
Comparison of correlations
The matrices below show:
1) Static correlations between pairs of variables for all patients at any time, in the real and synthetic data.
2) Dynamic correlations between variables in the real and synthetic data. Time series for each patient were linearly decomposed into trends (indicating a general upward or downward slope) and cycles (indicating periodic patterns). Correlations were then computed between trends and cycles for all patients.
Health Gym data can be downloaded either directly from this website or using the Python API.
Note: This section is still in progress, an API will be available soon
Install the health gym python package through pip by running the following in your terminal.
pip install healthgym
To download and access the data from within python.
import healthgym as hg hypotension_data = hg.datasets.Hypotension(root: 'path/to/data/', download = True)
All datasets support the following parameters.
root (string) # Root directory of dataset where dataset exists or will be saved to if download is set to True. download (bool, optional) # If true, downloads the dataset from the internet and puts it in root directory. If dataset is already downloaded, it is not downloaded again.
Download the dataset
Direct download as CSV file or view GitHub repository
Be part of the community
here is how to reach out.
Join our community on GitHub or contact us
|1||Gottesman O, Futoma J, Liu Y, Parbhoo S, Celi L, Brunskill E, Doshi-Velez F. Interpretable Off-Policy Evaluation in Reinforcement Learning by Highlighting Influential Transitions. Proceedings of the 37th International Conference on Machine Learning, PMLR. 2020;119:3658-3667.|