Welcome to the COVID-19 Subgroup Discovery and Exploration Tool (COVID-19 SDE Tool)

The COVID-19 infectious disease has led since December 2019 to a worldwide pandemic which is still under control measures. Researchers worldwide are making huge efforts aiming to a comprehensive understanding of the COVID-19 and related healthcare treatments. This work shows the preliminary results of a Machine Learning (ML) approach to identify subgroups of COVID-19 patients based on their symptoms and comorbidities, aiming to a better understanding of variability of severity patterns. In this work, we particularly address the variability (or heterogeneity) between distinct sources populating the research repositories, given the potential impact that this variability may have in data science and the generalization of its results.

What you can do:

  1. Explore the resultant COVID-19 subgroups through dynamic scatter plots which allow comparing subgroups with patient outcomes and data sources.
  2. Study the clinical variability between subgroups in terms of reported symptoms, comorbidities, demographics and outcomes through dynamic barplots and tables.
  3. Quantify the information variability between data sources and over time to support development of reliable, generalizable data science (under-development function).

The tool results are automatically generated from the input dataset. Consequently, the COVID-19 SDE Tool can be easily adapted to other COVID-19 datasets.

Developed by the Biomedical Data Science lab, ITACA Institute, Universitat Politècnica de València, Spain. If you are interested in collaborating in this work please contact us.

nCov-2019 data

The first results of COVID-19 subgroups by symptoms and comorbidities are provided using the open nCov-2019 dataset . The nCov-2019 dataset comprises a collection of publicly available information on worldwide cases confirmed during the ongoing nCoV-2019 outbreak.

Materials: We analyzed the raw nCov-2019 dataset release at 2020-05-11. We included those cases were at least one symptom and an outcome were available. Then, we fixed duplicates and homogenized values in outcomes, comorbidities and symptoms. We mapped the latter to ICD-10 terms. The final sample included 170 cases.

Methods: We applied a Multiple Correspondence Analysis 3-dimensional embedding of symptoms and outcomes and a hierarchical clustering. The proper number of clusters for both age-independent and age group analyses were selected by supervised inspection of group consistency.

Results: We found clinically meaningful patient subgroups based on symptoms and comorbidities for specific age groups and age-independent analyses. However, the two most prevalent source countries were divided into separate subgroups with different manifestations of severity.

For further details read our publication:

Carlos Sáez, Nekane Romero, J Alberto Conejero, Juan M García-Gómez. Potential limitations in COVID-19 machine learning due to data source variability: a case study in the nCov2019 dataset. Journal of the American Medical Informatics Association. Accepted manuscript. doi: 10.1093/jamia/ocaa258

Code available for replication in our COVID-19 Subgroup Discovery tool GitHub repository .