Data modelling recipes for SARS-CoV-2 wastewater-based epidemiology

TL;DR

This paper proposes a coherent, simple framework for data modelling in SARS-CoV-2 wastewater-based epidemiology, emphasizing preprocessing, normalization, smoothing, and basic regression techniques for effective analysis and short-term forecasting.

Contribution

It introduces a robust, easy-to-apply data modelling framework focusing on preprocessing and simple models, moving away from complex machine learning approaches.

Findings

Normalization by biomarkers is crucial.

Weekly downsampling suffices for data regularity.

Simple models provide accurate 7-day forecasts.

Abstract

Wastewater based epidemiology is recognized as one of the monitoring pillars, providing essential information for pandemic management. Central in the methodology are data modelling concepts for both communicating the monitoring results but also for analysis of the signal. It is due to the fast development of the field that a range of modelling concepts are used but without a coherent framework. This paper provides for such a framework, focusing on robust and simple concepts readily applicable, rather than applying latest findings from e.g., machine learning. It is demonstrated that data preprocessing, most important normalization by means of biomarkers and equal temporal spacing of the scattered data, is crucial. In terms of the latter, downsampling to a weekly spaced series is sufficient. Also, data smoothing turned out to be essential, not only for communication of the signal dynamics but likewise for regressions, nowcasting and forecasting. Correlation of the signal with epidemic indicators require multivariate regression as the signal alone cannot explain the dynamics but simple linear regression proofed to be a suitable tool for compensation. It was also demonstrated that short term prediction (7 days) is accurate with simple models (exponential smoothing or autoregressive models) but forecast accuracy deteriorates fast for longer periods.