COVID-19 External Data Sets

Collection of COVID-19 data available to researchers maintained by other MGB departments

Pulmonary X-Ray Severity Score

Includes patients with one or more frontal chest radiographs (chest X-Ray) obtained at MGH in the month of April.

The Pulmonary X-Ray Severity (PXS) Score is a quantitative assessment generated by a Machine Learning algorithm from frontal chest radiographs (chest x-ray). PXS scores are currently available for 17K patients with one or more chest x-rays obtained at MGH in the months of March through August 2020. It is expected that final cohort will include most patients with one or more frontal chest x-rays at MGH & BWH.

Data Stewards

If you have any questions, please contact Primary Owner: Dr. Jayashree Kalpathy-Cramer, Secondary Owner: Dr. Matthew Li.

What Data are Available

Data consists of the PXS score. The PXS score is a deep learning-based algorithm that takes chest radiograph DICOM images as inputs and outputs a measure of pulmonary consolidative disease severity. The source data includes frontal chest radiograph DICOMs from patients with COVID-19.

New as of March 2021: The DICOM images associated with these PXS scores are also available for download or easy viewing.

The PXS score is a continuous value that ranges from approximately 1 to 15, where larger scores are indicative of more severe lung disease. Using arbitrary categorical terms for consolidative lung disease severity, PXS < 2.5 can be considered normal/minimal, 2.5 ≤ PXS < 5 mild, 5 ≤ PXS < 9 moderate, and 9 > PXS severe. There is some variance in the PXS metric (+1 point on the scale), thus within subject changes of < 1 point are not likely to represent a significant change. The score is applicable for frontal chest radiographs (AP and PA views are acceptable).

Although the PXS score deep learning algorithm is tuned on radiographs from COVID-19 patients, the score is NOT specific for COVID-19, which has non-specific lung findings. Rather, the algorithm extracts a measure of severity in extent and density of lung opacities on chest radiographs (which can be seen in non-COVID-19 pneumonia, pulmonary edema, ARDS, etc.) PXS scores can be used as an input to algorithmic analysis of patient severity, outcome prediction, etc. in large cohort studies or subgroup analyses.

Note: In the original PXS data release (up to May 30, 2020), in the minority of chest radiographs with >1 raw frontal image acquired had PXS scores averaged for the frontal images to allow for a 1:1 relationship between each chest radiograph study accession and PXS score. In this updated release (August 24, 2020), the PXS scores for each individual frontal image are provided instead of an average. Users of this data can decide how to deal with this difference.

Data is updated on an ad hoc basis. Anticipated updates include the addition of BWH data.

To view the current data dictionary, please view our new data dictionary.

How to Access

The Pulmonary X-Ray Severity data set resides in the COVID-19 Data Mart. For more information on how to access, see the COVID-19 Data Mart housed in the Data Enclave and scroll to How to Access the Tool.

Follow the link COVID-19 Research Tools to return back to the COVID-19 Research Tools page.

Additional Resources

Data Sharing Summary - Detailed methodology & citation information
Finding the PXS Score - Instructions for locating data within the Enclave
Radiology Data Dictionary in Collibra
MGH COVID Registry

Includes an expansive cohort of confirmed SARS-CoV-2 infected patients hospitalized at MGH.

The MGH COVID Registry includes an expansive cohort of confirmed SARS-CoV-2 infected patients hospitalized at MGH who presented for care between March 8, 2020 and the end of the first surge on June 3, 2020. Extensive manual chart reviews were completed by trained reviewers including physicians, pharmacists, research nurses, and clinical research coordinators. 

Data Stewards

If you have any questions, please contact COVIDRegistrySupport@partners.org.

What data are available?

Data are based primarily on manual chart reviews and supplemented with data extractions from electronic health records (EHRs) via the Partners Enterprise Data Warehouse (EDW) and include:

  1. Clinical and demographic histories including comorbid conditions, body mass index, smoking status, prior medications, Covid-19 symptoms and known SARS-CoV-2 epidemiologic risk factors at presentation to care;
  2. Laboratory measures measured daily in-hospital on more than 30 biomarkers;
  3. Medications administered during hospitalization;
  4. Complications measured at 14-days and 28-days after presentation to care, including but not limited to ARDS, liver dysfunction, acute renal failure, cardiac arrhythmia, stroke and cardiomyopathy; 
  5. Outcomes measured at 14-days and 28-days after presentation to care, including ICU admission, intubation, death and NIAID ordinal outcome scale.

All prior data releases are archived and available.

How to Access

The MGH COVID Registry is only available to MGH researchers at this time. Before applying for access, please review the RISC Dashboard: MGH COVID Registry, as well as the requirements for requesting access. When ready, please follow the link to Enter your access request. Once your access request is submitted, the 1-4 day turnaround for granting access is dependent on data owner approval. If you are a Project Staff Member, your Project Leader will be required to approve your access. 

Follow the link COVID-19 Research Tools to return back to the COVID-19 Research Tools page.