Detection of COVID-19 by Machine Learning Using Routine Laboratory Tests.

Area of Science:

• Biomedical Informatics
• Computational Biology
• Clinical Diagnostics

Background:

• Accurate and timely diagnosis of coronavirus disease 2019 (COVID-19) is crucial for effective patient management and public health.
• Routine laboratory tests offer a readily available data source for diagnostic support.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for assisting in COVID-19 diagnosis using routine laboratory test results.
• To assess the performance of different ML algorithms and data combinations for COVID-19 detection.

Main Methods:

• Developed ML models using laboratory data from 1,391 patients, including clinical chemistry (CC) and complete blood count (CBC) parameters.
• Employed four ML algorithms: random forest (RF), gradient boosting (XGBoost), support vector machine (SVM), and logistic regression.
• Validated models on internal test and external datasets from Brazil, using SARS-CoV-2 RT-PCR as the gold standard.

Main Results:

• Model accuracies ranged from 74% to 91%.
• The RF model using CC and CBC analytes achieved 85.3% accuracy internally.
• The SVM model using CC and CBC parameters showed the highest performance on the external validation dataset with 91.18% accuracy and 100% sensitivity.

Conclusions:

• Machine learning models utilizing routine laboratory data can serve as effective clinical decision support tools for COVID-19 diagnosis.
• These models can augment physicians' clinical judgment, potentially improving diagnostic efficiency and accuracy.