Computing SARS-CoV-2 Infection Risk From Symptoms, Imaging, and Test Data: Diagnostic Model Development.

Area of Science:

• Medical Diagnostics
• Computational Biology
• Infectious Disease Research

Background:

• Accurate diagnosis of SARS-CoV-2 infection is challenging, impacting patient care across different settings.
• Existing diagnostic methods may not always provide sufficient clarity for timely clinical decisions.

Purpose of the Study:

• To develop and validate a personalized diagnostic model for assisting clinicians in diagnosing COVID-19.
• To compare the performance of various models, including machine learning and Bayesian inference, against established benchmarks.

Main Methods:

• Integrated patient symptoms and test results using machine learning and Bayesian inference to quantify SARS-CoV-2 infection risk.
• Trained models on simulated patient data and validated them with real-world patient cases at a major medical center.
• Compared diagnostic performance of Bayesian networks, distance metric learning, and ensemble models.

Main Results:

• Machine learning and Bayesian models demonstrated moderate to good performance in discriminating between SARS-CoV-2 infection and other diagnoses (sensitivity 81.6%-84.2%, specificity 58.8%-70.6%).
• The Bayesian inference network, when integrated with imaging and laboratory data, showed sensitivity to clinical evaluation choices.
• The study included 55 patients, with 69% testing positive for SARS-CoV-2 via RT-PCR.

Conclusions:

• Decision support models incorporating symptoms and test results can aid clinicians in diagnosing SARS-CoV-2 infection.
• Personalized diagnostic models show promise for improving COVID-19 diagnosis in clinical practice.