AJCC2203 - CME/CMLE - Detection of COVID-19 by Machine Learning Using Routine Laboratory Tests-Detection of COVID-19 by Machine Learning Using Routine Laboratory Tests

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

Pdf Summary

A study was conducted to utilize machine learning models with routine laboratory test results for diagnosing COVID-19. Various ML algorithms were tested on clinical chemistry and blood count parameters, achieving accuracies from 74% to 91%. The best model reached 91.18% accuracy on external validation. ML models showed potential as clinical decision support tools, especially with combined blood count and laboratory parameters. The Random Forest model performed best in the study dataset, while the Support Vector Machine excelled in external validation. Eosinophils were identified as crucial in distinguishing COVID-19 cases. ML models demonstrated improved sensitivity in severe cases and were effective in differentiating COVID-19 cases in outpatient and hospital settings. Challenges like false-negative rRT-PCR results and preanalytical errors were highlighted, stressing the need for reliable diagnostic tools. The study recommended further validation of ML models on larger datasets, different SARS-CoV-2 variants, and vaccinated populations for enhanced accuracy and reliability in COVID-19 diagnosis.

Keywords

machine learning models

COVID-19 diagnosis

clinical decision support tools

Random Forest model

Support Vector Machine

blood count parameters

laboratory test results

diagnostic tools

ML algorithms

SARS-CoV-2 variants