Predictive triage for testing may improve control of a COVID-19 epidemic while reducing testing requirements

Area of Science:

• Epidemiology
• Public Health
• Machine Learning

Background:

• COVID-19 pandemic mitigation relied heavily on widespread testing.
• Scaling testing capacity faces workforce and infrastructure challenges.
• Delays in sampling and testing impede timely interventions.

Purpose of the Study:

• To enhance pre-test triage using an ensemble model based on self-reported information.
• To improve the efficiency of COVID-19 testing resource allocation.
• To assess the epidemiological impact of an ensemble triage tool.

Main Methods:

• An XGBoost classifier was trained on social and health data from 38,180 students in Leuven, Belgium.
• The model predicted individual COVID-19 infection risk to recommend isolation, testing, or release.
• In silico simulations evaluated the epidemiological impact of the ensemble triage tool.

Main Results:

• The predictive model achieved a ROC AUC of [Formula: see text], with performance varying across retraining windows.
• Simulations showed the ensemble triage system could control infection surges, reducing the effective reproduction number below 1.0.
• Testing requirements were reduced by [Formula: see text], with model predictions influenced by contact number, testing reason, and symptom onset.

Conclusions:

• Pre-test triage using ensemble models can efficiently allocate testing resources.
• Timely implementation and isolation compliance can help rapidly control infection surges.
• Future research should explore deep learning models and validation in diverse settings and for other pathogens.