Comparing Rates of Change in SARS-CoV-2 Wastewater Load and Clinical Cases in 19 Sewersheds Across Four Major Metropolitan Areas in the United States

Abstract

There is no standard approach to interoperate the multiple SARS-CoV-2 wastewater surveillance data sets generated during the pandemic. We tested several data processing approaches on wastewater surveillance data sets generated from 19 sewersheds across four major metropolitan areas in the United States from May 2020 through October 2021. First, we explored the effect of different data processing techniques on the correlation between SARS-CoV-2 wastewater RNA load and clinical case counts and found that locally weighted smoothing (LOESS) smoothing applied to multivariate imputation by chain equations (MICE)-imputed wastewater viral load led to the strongest correlations in 16 out of 19 sewersheds (84%). Next, we calculated the rate of change (RC) in wastewater viral load and in clinical cases and found that imputing missing viral load data on a 28-day window produced the strongest correlation (Spearman's ρ = 0.63). Furthermore, we determined an average sensitivity threshold of 2.4 new COVID-19 cases per day resulted in a significant RC in wastewater, but sensitivity varied with the laboratory method used. Our retrospective analysis using RC highlighted certain methodological insights, reduced site-specific impacts, and estimated a wastewater sensitivity threshold�supporting the use of relative, rather than absolute, measures of SARS-CoV-2 wastewater data for more interoperable data sets.

Relationship between SARS-CoV-2 in wastewater and clinical data from five wastewater sheds

“…This confirmed the presence of a poor correlation between wastewater and clinical data values. In agreement with other studies ( Al-Faliti et al, 2022 ; Fitzgerald et al, 2021 ; Rusiñol et al, 2021 ), the large size WRRFs have a better correlation (Spearman's Rho = 0.64) between the viral loads and the daily clinical cases than the medium and small size WRRFs. The use of PMMoV normalization decreased Spearman coefficients for WRRFs 2, 3 and 4 but increases were observed for WRRF1 and WRRF5.…”

Estimating Rates of Change to Interpret Quantitative Wastewater Surveillance of Disease Trends

“…18,32 To address the temporal sparsity of wastewater data, Al-Faliti et al used daily reported cases to impute wastewater viral loads on unsampled days. 20 Daily rates of change were estimated using rolling linear models applied over 21- or 28-day subsets of the imputed daily wastewater viral loads. Following the approach of Al-Faliti et al, we constructed five complete daily viral load datasets using the mice package to implement multivariate imputation using chained equations (MICE) with random forest models.…”

“…These models use log-transformed viral load (log10 gene copies/day) as the response variable and date as the predictor variable. [20][21][22] When fit to three observations of weekly wastewater samples or five observations of twice-weekly samples, the regression coefficient corresponds to the slope of the trend-the average daily change in viral load-over the preceding ~15 days. The estimated rate of change can also be expressed as percent daily change (PDC), enabling more direct comparison with trends in other metrics.…”

“…18 Rate of change estimates from both approaches have been compared with population-based metrics (e.g., reported cases) but, to the best of our knowledge, the estimation performance and trend classification accuracy of either approach has not yet been evaluated. 18,20…”

“…Useful information about the direction and speed at which community infection trends are changing may therefore be inferred solely on the basis of wastewater measurements by estimating the slope of the wastewater trend at specific times, where the sign and magnitude of the slope provide the direction and rate of change, respectively. 20…”