Virginia Wastewater Surveillance: Can SARS-CoV-2 Once-Weekly Sampling Predict Imminent Rises in Community COVID-19 Disease Burden?

Abstract

While wastewater surveillance (WWS) has been a tool arguably used since the days of John Snow, its use has increased sharply – and far more publicly – since the beginning of the COVID-19 pandemic. WWS typically involves sample collection of community sewage at the wastewater treatment plant (WWTP), where it is then analyzed for viral particle counts reflecting the entire WWTP's service area (i.e., sewershed). This dataset is independent of other public health data, helping contextualize community disease burden. This is especially important in the absence of reliable case reporting. However, with WWS being so new, many public health practitioners struggle to understand and apply its nuance.

There is a need to provide easy, actionable interpretation for SARS-CoV-2 WWS in Virginia. To support the Virginia Department of Health (VDH), this PhD dissertation explored the SARS-CoV-2 WWS VDH Sentinel Monitoring Program (SMP) data and correlated with community COVID-19 burden indicators of cases and hospitalizations. I then used Yang's (2017) framework for Early Warning Systems to build an alert system, utilizing Shewhart's (1931) Simple Control Charts with moving ranges as the base.

The chosen variable for the alert system was by-site change in viral load (VL), with a dependent testing variable of by-site change in weekly cases. All data was assessed by individual site (n=24), as inter-site variability was extensive. Baselines for establishing normal site variance for VL and case change were determined by finding periods when case change and SARS-CoV-2 variant shifts were minimized. Once baselines were determined, the standard deviation (SD) of change was calculated for each site, with a multiplier of 3 to set the Upper Control Limit (UCL) of acceptable variance. The alert system indicated which changes of VL and cases were sufficiently out of normal variance by assigning flags. The flags for both VL and cases were compared to each other, and performance metrics calculated.

When using the SD multiplier threshold of 3 for both changing VL and cases (daily cases for next 7 days), the alert flags matched each other 88% of the time. If calculated for case flags of that week or following week (accounting for possible VL leading indicator effects), matching increased to 92%. For every positive VL alert, 49% had an associated positive case alert that week; 52% for the next 1-2 weeks. For every negative VL alert, 94% had an associated negative case alert that week; 97% for the next 1-2 weeks. Sensitivity was calculated at 39% for the same week, or 52% for the next 1-2 weeks. Specificity was calculated at 94% for the same week, or 99% for the next 1-2 weeks. These metrics were explored using other thresholds, which did improve certain performance indicators (e.g., sensitivity), but usually at the cost of other indicators. Major limitations include nuances of the datasets themselves, especially with the case counts, which were subject to changing reporting standards and methods throughout the pandemic.

This dissertation provides a proposed and evaluated alert system for WWS programs, with specific guidance and tips in Chapters 11-12 for readers. It follows up with a mock scenario retrospective test using the Omicron spike of December 2021. The methods and principles used here can be used by public health practitioners for SARS-CoV-2, as well as potentially other WWS targets.