Data-driven testing program improves detection of COVID-19 cases and reduces community transmission.

Area of Science:

• Epidemiology
• Machine Learning
• Public Health

Background:

• COVID-19 continues to pose a global health challenge due to new SARS-CoV-2 variants and uneven vaccine distribution.
• Effective testing strategies are crucial for controlling viral spread in community settings.

Purpose of the Study:

• To evaluate a data-driven COVID-19 testing program at a university.
• To assess the efficacy of machine learning models in identifying students at elevated risk for testing.
• To compare the speed of testing for model-identified contacts versus manually traced contacts.

Main Methods:

• Implemented a testing program using two simple, interpretable machine learning models to predict high-risk students.
• Conducted 20,862 COVID-19 tests.
• Compared testing turnaround times between model-assisted and manual contact tracing.

Main Results:

• The data-driven program identified a higher positivity rate (0.53%) compared to general surveillance (0.37%).
• Students identified by models were tested significantly faster (0.94 days) than those from manual tracing (1.92 days).
• The program detected a notable increase in positive cases among students within five days of initial testing.

Conclusions:

• Data-driven, machine learning-based COVID-19 testing programs can enhance early detection of positive cases.
• These strategies can expedite contact tracing and reduce transmission in university settings.
• Similar approaches are adaptable for other organizations to manage infectious disease outbreaks.