Wastewater based epidemiology-leveraging wastewater to mitigate the biases of the traditional data sources

Abstract

Wastewater-based epidemiology saw a surge in interest as a method to carry COVID-19 surveillance with reduced logistical and economical burden. The possibility to monitor entire communities using a single sample, measuring the concentration of SARS-CoV-2 RNA, is extremely enticing. In this thesis, our goal is to verify the feasibility of such an approach for the Italian region of Tuscany. Wastewater samples were collected in seven different wastewater treatment plants in different provinces. One of them was treated as a pilot to hone the sampling and measuring methodology during 2020-2021. For the other six, measurements were taken later, during 2022-2023. We first show that we can successfully estimate the number of daily newly infected as provided by clinical tests using exclusively wastewater data and a linear model. We devise a pipeline that lessens the impact of noise sources from our collected data, fit the model, show the link between the two data sources and demonstrate that wastewater is capable of nowcasting. We then notice that applying the same methodology to the data we collected later did not provide us with a working relationship between the two time series. We managed to identify the noise sources as the inaccuracy of clinical tests and the shift in viral subvariant prevalence. We tried to apply this knowledge to machine learning models in order to re-establish the lost connection between wastewater and clinical data. We were able to train several linear models that can offer a retrospective of the 2022-2023 trends. However, the nowcasting capability was still out of our reach. We also briefly tried to design and fit a compartmental model to describe the pandemic trends and show that wastewater can help clinical tests by covering its biases and offering an overall more robust reproduction number estimation when the two are paired. Finding the appropriate values for the parameters proved to be extremely challenging, and the correspondence between simulated and observed trends was not satisfactory. We conclude this work offering our perspective on how a wastewater-based surveillance system could be implemented in Tuscany, highlighting the pitfalls encountered in this first sampling campaign and how to tackle them.