ER-CoV

ER-CoV applies a machine learning classifier to routine blood-examination data to triage emergency-room patients for likely SARS-CoV-2 infection.

Key Features:

• Machine learning classifier: Uses a machine learning classifier trained on routine blood-test data to predict SARS-CoV-2 infection status.
• High specificity and NPV: Reported average specificity of 85.98% and negative predictive value (NPV) of 94.92%.
• Sensitivity, PPV, and AUC: Reported average sensitivity of 70.25%, positive predictive value (PPV) of 44.96%, and area under the curve (AUC) of 86.78%.
• Dataset and inputs: Analysis focused on 599 subjects with complete data across 16 common blood exams drawn from an initial cohort of 5,644 patients; 81 subjects were confirmed positive by RT-PCR for SARS-CoV-2.
• Training objective: Model was trained to predict COVID-19 negativity with a priority on achieving high specificity and NPV.
• Reduced testing burden: Estimated potential to reduce the number of SARS-CoV-2 tests conducted in emergency settings by at least 90%.
• Error analysis: Approximately 28% of false negatives were estimated to have been hospitalized regardless of the prediction.
• Patient management optimization: Enables immediate segregation of patients predicted as positive to reduce in-hospital transmission risk.

Scientific Applications:

• Efficient triage: Rapidly identify likely negative cases to prioritize confirmatory RT-PCR testing for higher-risk patients.
• Resource allocation: Reduce unnecessary SARS-CoV-2 testing and hospital admissions to aid allocation of limited medical resources.
• Infection control: Support rapid separation of suspected positive patients to mitigate nosocomial spread of SARS-CoV-2.

Methodology:

Developed via a case-control quantitative study using data from 5,644 patients, with analysis on 599 subjects with complete data across 16 common blood exams (81 RT-PCR positives), and the AI model trained to predict COVID-19 negativity prioritizing high specificity and NPV.