Development and External Validation of a Machine Learning Tool to Rule Out COVID-19 Among Adults in the Emergency...

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Diagnostic Accuracy

Background:

• Conventional COVID-19 diagnosis via RT-PCR is time-consuming and expensive.
• SARS-CoV-2 may present detectable patterns in routine blood test results.
• Machine learning can analyze these patterns to potentially expedite COVID-19 diagnosis.

Purpose of the Study:

• To develop and validate a machine learning model for ruling out COVID-19 in emergency departments.
• The model utilizes only routine blood test results for adults.
• Focus on accelerating diagnosis and reducing healthcare costs.

Main Methods:

• Utilized clinical data from 66 US hospitals (pre-pandemic and pandemic periods).
• Trained model on 2,183 PCR-confirmed COVID-19 cases and 10,000 prepandemic controls.
• Externally validated on 1,020 cases and 171,734 controls using AUROC, sensitivity, specificity, and NPV.

Main Results:

• Achieved an Area Under the Receiver Operating Characteristic (AUROC) curve of 0.91 for both training and external validation.
• At a risk score cutoff of 2.0, the model demonstrated 92.6% sensitivity and 59.9% specificity.
• Negative Predictive Values (NPVs) were high, reaching 97% at 20% prevalence.

Conclusions:

• A machine learning model integrating routine laboratory data shows high accuracy in ruling out COVID-19.
• This model may help optimize the selective use of PCR-based testing.
• Potential to improve efficiency in emergency department workflows.