An overview of the National COVID-19 Chest Imaging Database: data quality and cohort analysis

Abstract

The National COVID-19 Chest Imaging Database (NCCID) is a centralised database containing chest X-rays, chest Computed Tomography (CT) scans and cardiac Magnetic Resonance Images (MRI) from patients across the UK, jointly established by NHSX, the British Society of Thoracic Imaging (BSTI), Royal Surrey NHS Foundation Trust (RSNFT) and Faculty. The objective of the initiative is to support a better understanding of the coronavirus SARS-CoV-2 disease (COVID-19) and development of machine learning (ML) technologies that will improve care for patients hospitalised with a severe COVID-19 infection. The NCCID is now accumulating data from 20 NHS Trusts and Health Boards across England and Wales, with a total contribution of approximately 25,000 imaging studies in the training set (at time of writing) and is actively being used as a research tool by several organisations. This paper introduces the training dataset, including a snapshot analysis performed by NHSX covering: the completeness of clinical data, the availability of image data for the various use-cases (diagnosis, prognosis and longitudinal risk) and potential model confounders within the imaging data. The aim is to inform both existing and potential data users of the NCCID's suitability for developing diagnostic/prognostic models. In addition, a cohort analysis was performed to measure the representativeness of the NCCID to the wider COVID-19 affected population. Three major aspects were included: geographic, demographic and temporal coverage, revealing good alignment in some categories, e.g., sex and identifying areas for improvements to data collection methods, particularly with respect to geographic coverage. All analyses and discussions are focused on the implications for building ML tools that will generalise well to the clinical use cases.

An overview of the National COVID-19 Chest Imaging Database: data quality and cohort analysis

“…Snapshots of the code and copies of the forms for data access are available from the GigaScience GigaDB repository [ 53 ].…”