Related Experiment Video
Updated: Jul 24, 2025
08:48
Dynamic Monitoring of Seroconversion using a Multianalyte Immunobead Assay for Covid-19
Published on: February 16, 2022
2.9K
Consistent comparison of symptom-based methods for COVID-19 infection detection
Jesús Rufino1, Juan Marcos Ramírez1, Jose Aguilar2
1IMDEA Networks Institute, 28918, Madrid, Spain.
International Journal of Medical Informatics
|July 2, 2023
Summary
This study compared COVID-19 detection methods using self-reported symptoms. Tree-based machine learning models showed the highest performance, with runny nose and muscle aches being key indicators.
Area of Science:
- Epidemiology
- Machine Learning
- Public Health
Background:
- Global pandemic necessitated rapid COVID-19 detection methods.
- Self-reported symptom-based diagnostics were developed for resource management.
- Previous evaluations used varied datasets and methodologies.
Purpose of the Study:
- To comprehensively compare COVID-19 detection methods using self-reported data.
- To evaluate methods on the University of Maryland Global COVID-19 Trends and Impact Survey (UMD-CTIS) dataset.
- To assess performance across different countries and time periods.
Main Methods:
- Implemented rule-based, logistic regression, and tree-based machine learning models.
- Utilized UMD-CTIS data from six countries, including symptom and antigen test results.
- Evaluated methods using F1-score, sensitivity, specificity, and precision, with an explainability analysis.
Main Results:
- Fifteen methods were evaluated across six countries and two periods.
- Tree-based models achieved the highest F1-scores (45.07% - 73.72%).
- Stuffy/runny nose and aches/muscle pain were consistently relevant symptoms across models.
Conclusions:
- Homogeneous data evaluation provides a consistent comparison of detection methods.
- Explainability analysis aids in identifying key symptoms for COVID-19 detection.
- Self-reported data limitations mean these methods cannot replace clinical diagnosis.
