Model-based assessment of sampling protocols for infectious disease genomic surveillance

TL;DR

This study evaluates how different sampling strategies impact genomic surveillance of COVID-19, demonstrating that adaptive sampling can detect new variants significantly earlier than constant sampling, especially at low sequencing rates.

Contribution

It introduces a model-based framework comparing adaptive and constant sampling strategies, highlighting the benefits of adaptive sampling in early variant detection.

Findings

Adaptive sampling detects variants up to five weeks earlier.

Adaptive sampling reduces detection delays and estimation errors.

Lower sequencing rates combined with adaptive sampling are cost-effective.

Abstract

Genomic surveillance of infectious diseases allows monitoring circulating and emerging variants and quantifying their epidemic potential. However, due to the high costs associated with genomic sequencing, only a limited number of samples can be analysed. Thus, it is critical to understand how sampling impacts the information generated. Here, we combine a compartmental model for the spread of COVID-19 (distinguishing several SARS-CoV-2 variants) with different sampling strategies to assess their impact on genomic surveillance. In particular, we compare adaptive sampling, i.e., dynamically reallocating resources between screening at points of entry and inside communities, and constant sampling, i.e., assigning fixed resources to the two locations. We show that adaptive sampling uncovers new variants up to five weeks earlier than constant sampling, significantly reducing detection delays and estimation errors. This advantage is most prominent at low sequencing rates. Although increasing the sequencing rate has a similar effect, the marginal benefits of doing so may not always justify the associated costs. Consequently, it is convenient for countries with comparatively few resources to operate at lower sequencing rates, thereby profiting the most from adaptive sampling. Finally, our methodology can be readily adapted to study undersampling in other dynamical systems.