AI Insight
This study evaluated wastewater-based surveillance (WBS) as a tool for detecting outbreak onset and epidemic peaks of SARS-CoV-2, using county-level data from 281 U.S. counties collected between 2021 and 2024. The authors developed a classification-based framework and found that a Bayesian exponential growth model outperformed a standard reproductive number (Rt) method for outbreak detection, achieving sensitivity of 0.82 and positive predictive value of 0.64. Peak detection also performed well, with sensitivity of 0.84 and PPV of 0.70, and wastewater signals preceded case-defined outbreak onset by 4 to 6 days, though they showed minimal lead time over epidemic peaks.
Why it matters
Early and reliable detection of outbreak onset through wastewater monitoring could give public health authorities several days of advance warning before clinical cases are confirmed, enabling faster resource deployment and intervention. The reusable evaluation framework introduced here could also standardize how wastewater and other surveillance signals are assessed for operational public health use beyond COVID-19.
⚠️ Preprint – Noch nicht peer-reviewed
Dieser Artikel wurde noch nicht von unabhängigen Experten begutachtet. Die Ergebnisse sind vorläufig und sollten mit Vorsicht interpretiert werden.
Wastewater-based surveillance (WBS) is increasingly used to monitor infectious disease dynamics, yet most evaluations focus on correlation or forecasting – neither of which directly assesses whether wastewater signals can identify the epidemiological events most relevant to public health decision-making. We argue that outbreak onset and epidemic peak detection are the operationally critical use cases of WBS, requiring a fundamentally different evaluation framework. We introduce a classification-based framework that treats WBS as an event-detection problem, defining outbreaks and peaks as discrete events, establishing detection intervals to account for timing uncertainty, and incorporating censoring and data completeness criteria for valid comparisons against imperfect clinical reference outcomes. Within this framework, we apply a Bayesian exponential growth model for outbreak detection – benchmarked against a standard reproductive number (Rt)-based method – and a rule-based algorithm for peak detection, evaluating performance via sensitivity and positive predictive value (PPV). Applied to county-level SARS-CoV-2 wastewater data from 281 U.S. counties (Biobot, 2021-2024), the exponential growth approach substantially outperforms the Rt-based baseline: sensitivity 0.82 and PPV 0.64 versus sensitivity 0.58 and PPV 0.19 for the best-performing Rt variant. Peak detection achieves sensitivity 0.84 and PPV 0.70 at the county level. Both peak and outbreak detection achieve strong and consistent performance against hospitalizations and deaths at the state level. Spatial aggregation yields a statistically significant improvement in peak detection PPV against a curated reference standard ($p < 0.001$), while outbreak detection improvements under aggregation are directionally consistent but not statistically significant. Wastewater leads case-defined outbreaks by 4-6 days but minimally leads epidemic peaks, consistent with wastewater approximating prevalence rather than incidence. These findings demonstrate that wastewater signals can reliably detect outbreak onset and epidemic peaks across spatial scales and clinical outcomes, and that the choice of detection method matters substantially in practice. The classification framework developed here provides a reusable and principled tool for evaluating any surveillance signal as an event-detection system, with direct relevance to how WBS is actually used in public health decision-making.