A phenomenological neural network powered by the National Wastewater Surveillance System for estimation of silent COVID-19 infections

Abstract

Although wastewater-based epidemiology (WBE) has emerged as an inexpensive and non-intrusive method in contrast to clinical testing to track public health at community levels, there is a lack of structured interpretative criteria to translate the SARS-CoV-2 concentrations in wastewater to COVID-19 infection cases. The difficulties lie in the uncertainties of the amount of virus shed by an infected individual to wastewater as documented in clinical studies. This situation is even worse considering the existence of a population of silent infections and many other confounding factors. In this research, a quantitative framework of a phenomenological neural network (PNN) was developed to compute silent infections. The PNN was trained using the WBE data from the National Wastewater Surveillance System (NWSS) – a program launched by the CDC of the United States in 2020. It is found that the PNN excelled with superior interpretability and reduced overfitting. A big-data perspective on virus shedding by an infected population revealed more deterministic virus-shedding dynamics compared to the clinical studies perspective on virus shedding by an infected individual. With such characteristics employed as the theoretical basis for the estimation of the silent infections, a ratio of silent to reported infections was found to be 5.7 as the national median during the studied period. The study also noted the influence of temperature, sewershed population, and per-capita flow rates on the computation of silent infections. It is expected that the proposed framework in this work would facilitate public health actions guided by the SARS-CoV-2 concentrations in wastewater. In case of a new wave emergence or a new virus disease outbreak like COVID-19, the PNN powered by the NWSS would outline consolidated and systematic information that would enable rapid deployment of public health actions.