Unlocking the Potential of Public Datasets: Wastewater-Based Epidemiological Forecasting During COVID-19

Zhicheng Zhang, Sonja Neumeister, Angel Desai, Maimuna S. Majumder, Fei Fang

Keywords: COVID-19, Disease Surveillance, Wastewater-Based Epidemiology, Time-Series Forecasting

TL;DR: This study demonstrates the effectiveness of using publicly accessible wastewater data and reported COVID-19 case counts for epidemiological forecasting using various statistical and machine learning models.

Abstract: The COVID-19 pandemic has emphasized the necessity for effective tools to monitor and predict epidemiological trends. Traditional approaches to disease surveillance possess certain limitations, leading to the emergence of wastewater-based epidemiology (WBE) as a complementary approach. WBE has demonstrated a strong correlation with traditional epidemiological indicators (e.g., number of clinical cases and hospitalization), which makes it a valuable asset in informing public health decision-making processes. Despite the promising prospects of WBE, it faces certain challenges, including restricted data accessibility, geographical bias in data coverage, high data noise levels, and significant data distribution shifts. In this study, we examine the feasibility of utilizing exclusively two publicly available data, specifically aggregated wastewater data and reported case counts, for epidemiological forecasting in the COVID-19 pandemic. We incorporate a variety of statistical and machine learning models in an attempt to address the inherent volatility and bias of the data. We further introduce the usage of the segmentation method during the evaluation phase as a better evaluation metric. Our empirical results show that, even with limited data, performing epidemiological forecasting is possible, and its performance is comparable with methods that use more diverse data sources, suggesting its potential for broader health applications. Additionally, we utilize the insights from results on the length of the forecasting horizon to provide practical guidelines regarding real-world prediction.