[Paper Review] Triaging moderate COVID-19 and other viral pneumonias from routine blood tests

Forrest Sheng Bao, Youbiao He+10 more|arXiv (Cornell University)|May 13, 2020

COVID-19 diagnosis using AI23 references19 citations

TL;DR

This study proposes a machine learning approach to triage moderate COVID-19 and non-COVID-19 viral pneumonias using routine blood tests, leveraging random forests and support vector machines (SVMs) on data from 208 COVID-19 and 86 non-COVID-19 patients, achieving an SVM-based classifier with 84% accuracy, 88% sensitivity, 80% specificity, and 92% precision.

ABSTRACT

The COVID-19 is sweeping the world with deadly consequences. Its contagious nature and clinical similarity to other pneumonias make separating subjects contracted with COVID-19 and non-COVID-19 viral pneumonia a priority and a challenge. However, COVID-19 testing has been greatly limited by the availability and cost of existing methods, even in developed countries like the US. Intrigued by the wide availability of routine blood tests, we propose to leverage them for COVID-19 testing using the power of machine learning. Two proven-robust machine learning model families, random forests (RFs) and support vector machines (SVMs), are employed to tackle the challenge. Trained on blood data from 208 moderate COVID-19 subjects and 86 subjects with non-COVID-19 moderate viral pneumonia, the best result is obtained in an SVM-based classifier with an accuracy of 84%, a sensitivity of 88%, a specificity of 80%, and a precision of 92%. The results are found explainable from both machine learning and medical perspectives. A privacy-protected web portal is set up to help medical personnel in their practice and the trained models are released for developers to further build other applications. We hope our results can help the world fight this pandemic and welcome clinical verification of our approach on larger populations.

Motivation & Objective

• Address the challenge of distinguishing moderate COVID-19 from other viral pneumonias due to clinical similarity and limited testing availability.
• Overcome constraints in widespread SARS-CoV-2 testing by utilizing routinely collected blood test data.
• Develop a machine learning model that enables early, low-cost triage of moderate pneumonia cases using accessible clinical data.
• Ensure model interpretability and practical deployment through a privacy-protected web portal and open-sourced models.
• Support healthcare providers in clinical decision-making during pandemic surges when testing capacity is strained.

Proposed method

• Utilize routine blood test data from 208 moderate COVID-19 and 86 non-COVID-19 viral pneumonia patients as input features.
• Apply two robust machine learning families—random forests (RFs) and support vector machines (SVMs)—for classification.
• Train models to differentiate between moderate COVID-19 and non-COVID-19 viral pneumonia based on blood biomarkers.
• Optimize model performance using standard hyperparameter tuning and cross-validation techniques.
• Implement a privacy-protected web portal to enable real-time clinical deployment of the trained models.
• Release the trained models openly to encourage further development and integration into clinical and research applications.

Experimental results

Research questions

• RQ1 Can routine blood test data reliably distinguish moderate COVID-19 from non-COVID-19 viral pneumonias?
• RQ2 How do different machine learning models (e.g., RFs vs. SVMs) perform in classifying these conditions using blood biomarkers?
• RQ3 To what extent are the model predictions interpretable from both machine learning and clinical perspectives?
• RQ4 Can a deployed system using this approach support triage in clinical settings with limited testing capacity?
• RQ5 What is the performance of the best-performing model in terms of accuracy, sensitivity, specificity, and precision?

Key findings

• The SVM-based classifier achieved the highest performance with 84% accuracy, 88% sensitivity, 80% specificity, and 92% precision.
• The model's predictions were found to be interpretable from both machine learning and clinical perspectives, enhancing trust and usability.
• The best-performing model demonstrated strong generalization on the test set, indicating robustness to variation in blood test values.
• The results suggest that routine blood tests contain sufficient discriminative information for triaging moderate viral pneumonias.
• A privacy-protected web portal was successfully deployed to support clinical use, and the trained models were released for public reuse.
• The study provides a scalable, low-cost alternative to PCR testing for initial triage in resource-limited or high-demand settings.