Enterovirus D68 in United States wastewater: a longitudinal surveillance study integrating climatic, demographic, and clinical data

Summary

Background

Enterovirus D68 (EV-D68) can cause severe respiratory illness and acute flaccid myelitis (AFM), but limited clinical testing and sparse surveillance data hinder the ability to address epidemiological questions about its circulation. This study aimed to characterize EV-D68 seasonality across the U.S. using wastewater data, evaluate the influence of climatic and sociodemographic factors, and compare trends with symptom-based clinical indicators.

Methods

We analyzed 43,876 samples collected from 147 wastewater treatment plants across 40 U.S. states between July 2023 and July 2025 to quantify EV-D68 RNA in wastewater solids and estimate the seasonal peak and season duration. We evaluated climatic and sociodemographic drivers of variation and compared wastewater trends with symptom-based clinical diagnoses for AFM, wheezing, and enterovirus retrieved from Epic Cosmos, a nationwide electronic health record-based research network, representing 4.5 billion encounters during the same study period.

Findings

We observed a biennial EV-D68 pattern with a national peak in September 2024 and an extended 20-month period of detection in California. Seasonal peaks occurred 28–31 days earlier in regions with 5 °C higher temperatures or dew points. Season duration was longer by 7–11 weeks in dense, urban catchments with more childcare facilities, crowded households, hospitals, and nursing homes. Wastewater concentrations correlated positively with enterovirus diagnoses (Spearman ρ = 0.34, p = 0.01) and negatively with wheezing in adults ≥65 years (ρ = −0.49, p < 0.0001).

Interpretation

Wastewater surveillance can generate epidemiological metrics for EV-D68 without clinical surveillance data, identify where and when activity increases, and reveal the environmental and demographic factors driving these patterns. Wastewater surveillance offers high-resolution data for pathogens with diagnostic testing constraints and provides information that can strengthen epidemiological modeling and support preparedness for future disease waves.

Funding

Sergey Brin Family Foundation.

Research in context.

Evidence before this study

Wastewater surveillance has significantly expanded worldwide since the SARS-CoV-2 pandemic, and many programs now monitor additional pathogens. Beyond detecting circulation, wastewater data provide dense, continuous, population-level measurements that can be used to describe seasonal patterns and support epidemiological modeling. This approach has the potential to offer insights into the circulation dynamics of viruses for which clinical surveillance is limited, such as enterovirus D68 (EV-D68). We searched PubMed for articles published from database inception to Jan 1, 2026, without language restrictions, using the search terms (“Enterovirus D68” OR “EV-D68”) AND (seasonality OR climate OR humidity OR temperature OR demographic) and (“Enterovirus D68” OR “EV-D68”) AND wastewater. This search yielded 149 studies. Existing literature included analyses of EV-D68 seasonality based on clinical or historical non-polio enterovirus datasets and wastewater studies, mainly assessing correlations between EV-D68 RNA and clinical case counts in a limited number of locations only. However, no studies have evaluated whether wastewater data can independently generate epidemiological metrics, such as the timing and duration of circulation, or identify climatic and sociodemographic factors driving these patterns.

Added value of this study

This study shows that wastewater surveillance can independently generate epidemiological metrics for enterovirus D68 (EV-D68) and track its spread across large geographic areas without reliance on clinical testing. Using more than 43,000 wastewater solids samples from 147 treatment plants across the U.S., we estimated when and how long EV-D68 circulated and identified the climatic and sociodemographic factors shaping these patterns. Wastewater data revealed a clear latitudinal and climate-related gradient, with earlier circulation in warm, humid regions. We show that season duration was influenced by population density, urbanicity, childcare density, household crowding, and number of nursing homes and hospitals. While environmental conditions determined when circulation began, demographic structure and social mixing shaped how long it persisted. We identified a prolonged, nearly two-year EV-D68 season in California, suggesting a possible disruption of the expected biennial pattern. By comparing wastewater signals with clinical diagnoses from Epic Cosmos, we highlight the limitations of symptom-based clinical data for EV-D68 and demonstrate that wastewater provides a more complete and timely view of viral circulation. Together, these findings establish wastewater surveillance as a powerful tool for characterizing seasonal and geographic patterns of pathogens with sparse or unreliable clinical surveillance data and for identifying the environmental and demographic factors that drive viral activity.

Implications of all the available evidence

This study shows that wastewater surveillance can generate the high-resolution epidemiological metrics needed to understand when circulation begins, how long it lasts, and which environmental or sociodemographic factors shape these patterns for pathogens with limited clinical testing, such as EV-D68. By providing dense, population-level spatiotemporal data that are independent of healthcare-seeking behavior, wastewater monitoring can identify when and where viral activity is increasing, guide the timing of diagnostic testing, and support targeted interventions in high-risk settings such as childcare facilities. Wastewater-derived estimates of timing and duration also supply the detailed inputs required to improve modeling of future circulation. As wastewater networks expand globally, this approach offers a practical way to fill gaps in clinical surveillance and deepen our understanding of the epidemiology of viruses that are rarely detected through routine diagnostic testing.

Introduction

Enterovirus D68 (EV-D68) is a non-enveloped picornavirus transmitted primarily via the respiratory route, for which no vaccine is available.¹ Symptoms range from mild upper respiratory illness to severe lower respiratory disease with dyspnea, wheezing, hypoxia and, especially in children, can involve central nervous system complications (CNS) such as acute flaccid myelitis (AFM).¹ AFM cases have already been shown to increase after biennial EV-D68 outbreaks, and since 2014, the U.S. Centers for Disease Control and Prevention documented 779 confirmed cases of AFM as of Aug 5, 2025.2, 3, 4

Diagnosing and tracking EV-D68 remains challenging because clinical testing is not a standard practice in most laboratories, and the virus is typically identifiable in nasopharyngeal specimens only during the first week after CNS symptom onset.¹ Population-level wastewater surveillance could facilitate the detection of EV-D68 circulation at scale, as viral RNA is occasionally detected in stool.^5,6 Studies have shown that wastewater EV-D68 RNA tracks trends in laboratory-confirmed cases in California and internationally.7, 8, 9 However, it remains unclear whether wastewater signals reflect overall population-level EV-D68 clinical activity as approximated by clinical diagnoses such as AFM, wheezing, and enterovirus encounters.

Furthermore, the drivers of EV-D68 seasonality and its biennial pattern remain uncertain. Enterovirus circulation in the U.S. has been linked to climatic factors.10, 11, 12 In temperate regions, enteroviruses tend to peak in summer and early autumn with flatter, earlier seasons in the south and sharper, later peaks in the north.^11,13 A latitudinal gradient has been reported specifically for EV-D68.² Demographic factors, such as school-term timing and birth rate, have also been associated with outbreak timing and duration.^10,11 However, these inferences largely rely on clinical data that are limited, delayed, or missing for EV-D68. Wastewater surveillance can generate epidemiological metrics independent of clinical testing and allow systematic evaluation of climatic and sociodemographic determinants of circulation. Identifying these drivers could improve predictions of EV-D68 activity and support timely public health responses, such as scheduling testing appropriately and implementing targeted measures to reduce transmission.

We analyzed wastewater from 147 U.S. wastewater treatment plants (WWTPs) collected between July 2023 and July 2025 to quantify EV-D68 concentrations and characterize spatiotemporal patterns by estimating the EV-D68 center of season and season duration in each location. We used environmental and sociodemographic variables to examine spatiotemporal differences. Finally, we retrieved diagnostic data from Epic Cosmos, a dataset created by a community of Epic health systems representing more than 300 million patient records from more than 1800 hospitals and 41,000 clinics from all 50 U.S. States and D.C., for the same study period (July 2023–July 2025), to compare wastewater trends with clinical diagnoses for AFM, wheezing, and enterovirus, which were used as proxies for EV-D68 clinical cases.

Methods

Wastewater concentration and temporal metrics of EV-D68

We collected wastewater solids samples one to seven times weekly (median = 3) from each of 147 WWTPs across 40 states in the United States from July 2023 to July 2025 (Appendix Table S1, Figure S1) as part of routine wastewater surveillance conducted through the WastewaterSCAN program. We quantified EV-D68 RNA in each sample using droplet digital RT-PCR following previously peer-reviewed methods (Appendix Section 1).^7,14 Wastewater solids were analyzed because enteroviruses have been shown to preferentially partition to solids, and previous work has confirmed that EV-D68 RNA measured in the solids fraction of wastewater correlates well to EV-D68 infections in the contributing community.¹⁵

We derived two variables from the EV-D68 RNA concentration in wastewater solids to describe the temporal dynamics of EV-D68 across the country: (1) the center of season, defined as the week corresponding to the midpoint of the seasonal activity detected from wastewater, and (2) the season duration, defined as the number of weeks around this midpoint during which seasonal activity was detected from wastewater (Appendix Section 2). We estimated both variables separately for each WWTP using all individual daily sample measurements. We then estimated both variables at the state and national levels. We first averaged the daily measurements for each WWTP over a week, then used this weekly average to compute a population-weighted weekly average at the state or national level, considering only the weekly means of the WWTPs within each geographical jurisdiction (Appendix Section 3). To estimate both variables at all different geographical scales, we first normalized EV-D68 RNA concentrations to those of pepper mild mottle virus (PMMoV) RNA to account for variations in fecal load and to correct for differences in viral RNA isolation efficiency.^16,17 Second, we applied a five-day centered trimmed moving average to daily PMMoV-normalized concentrations (referred to hereafter as the smoothed, PMMoV-normalized EV-D68 concentration) to reduce high-frequency noise (Appendix Section 3, Figure S2). A five-day window was selected to balance noise reduction with temporal resolution given the high sampling frequency, consistent with previous wastewater surveillance analyses.¹⁸ The center of season was defined identically across all spatial scales, whereas the operational definition of season duration differed between WWTP or state and national analyses to account for differences in signal aggregation (Appendix Section 2).

Environmental variables

We retrieved daily measurements of average temperature, dew point, relative humidity, precipitation, wind speed, and surface pressure from the National Oceanic and Atmospheric Administration (NOAA) via a custom Python script (accessed September 3, 2025). For each WWTP, we obtained environmental data from the nearest weather station; in total, 117 stations were used, each representing one to five WWTPs (Appendix Table S2).

Sociodemographic variables

We derived the proportions of children ≤5 years, adults ≥65 years, and crowded households (with more than one person per room) from the 2023 5-year American Community Survey (ACS) at the census tract level.¹⁹ These age groups were selected because EV-D68 severe outcomes are most commonly reported in young children and older adults.¹ Because sewershed and census tract boundaries were not aligned, we aggregated tract-level counts to the WWTP level by calculating the proportion of each tract intersecting a given sewershed using ArcGIS Pro (version 3.1.1), as previously described.²⁰ We adjusted counts based on these proportions and summed them to produce sewershed-level values (Appendix Table S3). We calculated the birth rate (number of births per capita) at the county level using 2024 US Census Bureau Population Estimates.¹⁹ Similarly, we aggregated county-level birth counts and population denominators to the WWTP level using the proportion of each county intersecting a given sewershed and adjusting the counts accordingly.

We retrieved the number of childcare establishments in each county (NAICS code 624410) from the 2023 County Business Patterns data.²⁰ We obtained the land area in square kilometers from the 2022 US Census Bureau county boundaries (5-m resolution).²¹ We calculated the density of childcare establishments in each county and assigned the density value of the county containing the largest share of the sewershed to each sewershed area. Similarly, population density was defined as the number of residents per square kilometer in the predominant county of each sewershed, using 2022 ACS population estimates.²²

We calculated urbanicity as the proportion of each sewershed intersecting urban-designated areas, based on 2020 US Census Bureau Urban-Rural Classification data.²³

We retrieved the number of hospitals, nursing homes, and airports in sewersheds as described in a previous study.²⁴

Clinical diagnosis variables

We obtained clinical diagnostic records from Epic Cosmos (https://cosmos.epic.com, accessed September 10, 2025) from July 2023 to July 2025 for the entire nation and for each state separately. Epic Cosmos is a nationwide electronic health record-based research network representing over 4.5 billion clinical encounters during the study period. We analyzed four diagnostic categories: acute flaccid myelitis (AFM) (ICD-10-CM G04.82), wheezing in patients aged ≤5 years and those aged ≥65 years (ICD-10-CM R06.2), and enterovirus diagnoses (ICD-10-CM B34.1, B97.10, B97.19) as proxies for population-level EV-D68 clinical activity, since EV-D68 specific laboratory-confirmed case data are not available at the spatial and temporal resolution required for our analysis. For both state and national levels and for each week, we calculated a diagnosis proportion by dividing the number of encounters with the diagnosis by the total encounters in that jurisdiction for that week. We retrieved and analyzed AFM data monthly at the state level because weekly state counts were often small and masked. On Epic Cosmos, counts of ten or fewer are set to “zero” to reduce the risk of reidentification, resulting in a “zero” value for the calculated rate during a specific period. We calculated the monthly AFM proportions as the monthly AFM count divided by the total encounters in that state for that month.

Statistical analyses

All statistical analyses and visualizations were conducted in R version 4.5.1 using RStudio version 2025.09.0.

We analyzed temporal metrics (EV-D68 center of season and season duration) at the WWTP level using weighted least-squares regression models (Appendix Sections 3 and 4i). Regression weights for all models accounted for variability in EV-D68 concentration and detection frequency across WWTPs and were derived from cumulative smoothed, PMMoV-normalized EV-D68 RNA concentrations. This weighting approach reduces the influence of WWTPs with sparse or consistently low EV-D68 detection and places greater emphasis on sites with stronger and more stable signals.²⁵

We evaluated the association between geographic location (latitude and longitude), environmental variables, and EV-D68 temporal metrics using weighted least-squares regressions (Appendix Section 4i–iii). We analyzed each variable in a separate model to avoid multicollinearity, as environmental variables were correlated with one another as well as latitude and longitude (Appendix Figure S3). We fit two models for each environmental variable: one including the variable alone, and a second including the variable as well as latitude and longitude (Appendix Section 4iv–v). A third, geography-only baseline model (latitude and longitude only) was used for comparison. We summarized model performance using the weighted coefficient of determination (R_w²), and we evaluated the added explanatory value of each predictor by comparing the differences in R_w² relative to the baseline model.

We assessed the relationship between sociodemographic variables and EV-D68 center of season and season duration using the non-parametric Kruskal–Wallis test. The proportion of crowded households, population density, childcare density, the proportion of children aged ≤5 years, adults aged ≥65, and the birth rate were divided into tertiles. Urbanicity, the presence of airports, and the number of hospitals and nursing homes were categorized as dichotomous variables: airports as present or absent; hospitals and nursing homes as below or above their respective medians. Pairwise comparisons were performed using Wilcoxon rank-sum tests with Bonferroni adjustment.

We assessed correlations between clinical diagnoses from Epic Cosmos and wastewater EV-D68 RNA concentration and using PMMoV-normalized values that were smoothed with the five-day moving-average employed elsewhere in the study. We computed Spearman rank correlations at national and state levels using weekly or monthly data, depending on diagnosis frequency (Appendix Section 5).

Role of the funding source

The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the manuscript.

Results

Spatiotemporal analysis of EV-D68

We detected EV-D68 RNA in 24% (n = 10,451) of the 43,876 wastewater solids samples analyzed, with concentrations ranging from 580 to 4.8 × 10⁶ gene copies per gram dry weight (gc/g) (median = 8.1 × 10³ gc/g). Detection frequency varied by WWTP, ranging from 0.7% (2/297) to 82% (568/693) (Appendix Table S4). We observed an expected biennial EV-D68 outbreak in 2024, with only 5% (n = 501) detection before February 2024 and 29% (n = 9950) afterward (Fig. 1a).

Fig. 1.

EV-D68 RNA in wastewater at national and state levels, 2023–2025. Smoothed, PMMoV-normalized EV-D68 RNA concentration for (a) United States, (b) California, and (c) Pennsylvania. The shaded band denotes the EV-D68 activity duration; the vertical line marks the center of the EV-D68 season. For the national series (a), duration was defined as the period when concentrations exceeded a threshold equal to a baseline plus three standard deviations, where the baseline and standard deviation were estimated from the quiet tail (lowest tertile) of the data distribution. For state series (b–c), duration was defined as the continuous run containing the center of season during which values were detectable, with a minimum of at least two consecutive weeks with EV-D68 RNA concentration above 0. Dates with day-of-month annotate the start and end of the duration and the season center in each state. ∗EV-D68 RNA concentration refers to the smoothed, PMMoV-normalized concentration, calculated using a five-day centered trimmed moving average. For each five-day window, the highest and lowest daily values were excluded, and the remaining three were averaged.

We estimated the EV-D68 center of season in the U.S. as September 2024, with season duration lasting 66 weeks: from February 12, 2024, to May 18, 2025. At the state level, center of season ranged from early June to November 2024, occurring earlier in southern and eastern states (June to August), and progressively later in northern and western regions (September to November) (Fig. 2a, Appendix Figure S4). Most states exhibited relatively short season duration (below five months), while a few states, notably California and Florida, showed prolonged duration (above 13 months) (Fig. 2b). Season duration in California spanned from November 27, 2023, to July 31, 2025, with a center of season on November 11, 2024 (Fig. 1b). Pennsylvania, representative of most states, experienced a shorter season duration from July 8 to December 8, 2024, centered on September 23, 2024, aligning with the national center of season (Fig. 1c).

Fig. 2.

State-level timing and duration of EV-D68 activity in U.S. wastewater, with WWTP-level geographic gradients. (a) Map shows, for each state, the calendar date of the EV-D68 seasonal center estimated from the state-aggregated weekly series (PMMoV-normalized, 5-day–trimmed), after first aggregating across WWTPs within state and week. Colors run from earlier (Jun ’24) to later (Nov ’24). (b) Map shows the length of the detectable activity window (displayed in months) around the seasonal center of the state, computed on the same state-aggregated weekly series by identifying consecutive weeks with detectable signal (runs of at least two weeks above the non-detect threshold) and converting weeks to months. Gray states do not have wastewater data available. Alaska and Hawaii are shown out of scale for layout. Panels c–f are weighted bivariate regressions displaying gradients between longitude or latitude and EV-D68 center of season (C and d) or activity duration (e and f). Each point is a WWTP within the United States (n = 146) Clinton, Iowa had W = 0 (no detections after normalisation and trimming) and therefore was excluded. Point size denotes the plant weight WW, defined as the sum across weeks of PMMoV-normalised, 5-day–trimmed EV-D68 concentrations, with non-detects set to 0. Larger circles indicate greater cumulative signal and therefore greater influence on the weighted least-squares fits. Lines show weighted least-squares fits with 95% CIs.

Weighted linear regression analyses indicated that the EV-D68 center of season occurred significantly later in the north than in the south (6.5 days per degree latitude; 95% CI 5.1–7.9; p < 0.0001; R_w² = 0.37) and toward the west (slope −1.8 days per degree longitude; 95% CI −2.3 to −1.4; p < 0.0001; R_w² = 0.31) (Fig. 2c and d). EV-D68 season duration shortened with increasing latitude (−4.6 days per degree; 95% CI −7.4 to −1.7; p < 0.0001; R_w² = 0.07) and decreased towards the east (−1.8 days per degree; 95% CI −2.7 to −1.0; p < 0.0001; R_w² = 0.11) (Fig. 2e and f).

Climatic patterns of EV-D68

Higher temperatures and increased atmospheric moisture (dew point) were associated with earlier seasonal centers (Fig. 3). After accounting for site latitude and longitude, a 5 °C increase in weekly temperature was associated with an earlier seasonal center by 28.1 days (95% CI −33.7 to −22.6; p < 0.0001), and a 5 °C rise in weekly dew point correlated with a 31.3-day earlier seasonal center (95% CI −3790 to −25.6; p < 0.0001) (Table 1). Latitude and longitude alone explained 59% of the weighted variance, which increased to 76% with temperature and to 78% with dew point. Temperature and dew point are highly correlated (Appendix Figure S3), so their adjusted estimates reflect the same underlying warm, moist conditions and should not be considered additive. Surface pressure also showed an independent association, with later peaks at higher pressure (+15.6 days per 5 hPa; 95% CI 9.3–21.8; p < 0.0001; R_w² = 0.06). EVD-68 timing was not associated with relative humidity, precipitation, and wind speed. Analyses revealed no association between EV-D68 season length and environmental factors (Appendix Figure S5, Table S5).

Fig. 3.

EV-D68 center of season as a function of mean environmental conditions across WWTPs (n = 146). For each plant, the x-axis is the mean of the indicated variable (daily values aggregated to weekly means over the EV-D68 activity window); the y-axis is the calendar date of the season center. Lines are univariate weighted least-squares fits with 95% CIs; weights are proportional to each plant's cumulative EV-D68 signal, with non-detects set to 0. Point sizes are proportional to the weights and the weighted R² is shown on each panel.

Table 1.

Effects of weekly environmental variables on the calendar date of the local EV-D68 seasonal center in wastewater.

		Bivariate			Adjusted for geography
Variable	Step	Effect in days [CIs]	p-value	Rw2	Effect in days [CIs]	p-value	Rw2 (Lat + Lon)	Rw2 (Lat + Lon + X)	Increase in Rw2 from adding X
Temperature (°C)	per 5 °C	−34.1 [−37.7, −30.5]	p < 0.0001	0.71	−28.1 [−33.7, −22.6]	p < 0.0001	0.59	0.76	0.17
Dew point (°C)	per 5 °C	−33.7 [−37.0, −30.3]	p < 0.0001	0.73	−31.3 [−37.0, −25.6]	p < 0.0001	0.59	0.78	0.18
Relative humidity (%)	Per 10% RH	−11.6 [−21.6, −1.7]	0.02	0.04	−1.2 [−7.8, 5.5]	0.73	0.59	0.6	0
Precipitation (mm)	per 5 mm	−8.1 [−17.8, 1.6]	0.10	0.02	4.6 [−1.9, 11.2]	0.17	0.59	0.6	0.01
Wind speed (km/h)	per 1 km/h	−1.7 [−3.7, 0.4]	0.11	0.02	−0.9 [−2.2, 0.4]	0.19	0.59	0.6	0
Pressure (hPa)	Per 5 hPa	22.3 [12.7, 31.8]	p < 0.0001	0.13	15.6 [9.3, 21.8]	p < 0.0001	0.59	0.65	0.06

Effects are expressed in days, with negative values denoting earlier timing and positive values denoting later timing. Effects were estimated using weighted least-squares models fitted first without geographic covariates (bivariate) and then with geographic adjustment for latitude and longitude (Lat + Lon + X). For interpretability, effects correspond to fixed increments (temperature and dew point 5 °C; relative humidity 10%; precipitation 5 mm; wind speed 1 km/h; pressure 5 hPa). Weighted R_w² is shown for the bivariate model and for the geography-only baseline (Lat + Lon) and the geography-adjusted model (Lat + Lon + X); “Increase in R_w² from adding X” gives the gain (R_w²) from adding the climate variable to geography. Models used plant-level weights proportional to cumulative EV-D68 signal after trimming (non-detects set to zero). Brackets denote 95% confidence intervals (CIs).

Sociodemographic patterns of EV-D68

Higher childcare density, household crowding, number of hospitals, number of nursing homes, population density, and urbanicity were each associated with a longer EV-D68 season duration (Fig. 4, Appendix Table S6). EV-D68 season duration was notably longer in catchments with greater urbanicity (median [IQR] 20 [13–28] weeks for >50% urban and 12 [4–18] weeks for ≤50%; p < 0.0001) and higher population density (median 24 [15–29] weeks in the highest tertile and 13 weeks [5–21] in the lowest; p < 0.0001). Childcare establishment density showed a strong gradient, with the longest duration in the highest tertile (median 25 [19–31] weeks) compared to the middle (15 [7–22] weeks) and lowest (14 [7.8–23] weeks) tertiles (p < 0.0001 for both comparisons). The proportion of crowded households was similarly associated with duration (median 25 [16–31] weeks in the highest tertile vs 15 [6.8–21] weeks in the lowest; p < 0.0001). Catchments above the dataset median for hospitals (>2 per catchment) and nursing homes (>8 per catchment) also experienced longer seasons (median 22 [16–29] and 15 [8.5–23] weeks for hospitals, p = 0.01; 23.5 [15–31] and 15 [8–20.8] weeks for nursing homes, p < 0.0001). A modest association was observed with the proportion of adults aged ≥65 years, with the highest median duration in the middle tertile (21 [16–29] weeks) compared to the lowest (16 [10–23] weeks, p = 0.03). No associations were found for the proportion of children aged ≤5 years, birth rate, or the presence of an airport. Variable distributions and pairwise correlations are shown in Appendix Figure S6, Figure S7. Sociodemographic variables were not meaningfully associated with the center of season, except for a moderate association indicating that lower childcare density corresponded to a later center of season (Appendix Figure S8, Table S7).

Fig. 4.

Duration of EV-D68 circulation by wastewater treatment plant (WWTP)-level characteristics. Each panel shows the distribution of EV-D68 circulation duration (in weeks) across categories of a single WWTP–level determinant. Points represent individual WWTPs and boxplots show the median, interquartile range (IQR), and whiskers at 1.5 × IQR. Airport presence is dichotomized as absent and present. Hospital and nursing home coverage are split using the dataset medians: 2 hospitals and 8 nursing homes within each WWTP catchment area, respectively. Urbanicity is defined by the proportion of the WWTP catchment area classified as urban (≤50% vs >50%). The proportions of children aged ≤5 years, adults aged ≥65 years, crowded households, birth rate (per capita), childcare density (per km²), and population density (per km²) are grouped into within-study tertiles (“low”, “middle”, “high”). Sample sizes (n) shown in each panel indicate the number of WWTPs included in each comparison group. Kruskal–Wallis test was used to assess overall differences, and pairwise comparisons used Wilcoxon rank-sum tests with Bonferroni adjustment. Horizontal connector bars indicate statistically significant differences; asterisks denote adjusted p-values (∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001).

Comparison with clinical diagnoses

Nationally, wastewater EV-D68 RNA concentrations did not track proportions of clinical diagnoses used as proxies for EV-D68 clinical cases over the study period (Fig. 5a–d). Wastewater showed a pronounced seasonal increase in autumn 2024, whereas clinical diagnoses were comparatively constant or diffuse. Enterovirus diagnoses displayed a weak positive correlation with wastewater (Spearman ρ = 0.34, p = 0.01), while wheezing in adults aged ≥65 years correlated negatively (ρ = −0.49, p < 0.0001) (Appendix Figure S9). No association was observed for wheezing in children ≤5 years or for AFM. Correlation patterns varied across states (Appendix Figure S10, Table S8). AFM analyses were limited by masked monthly counts when diagnoses were ≤10, yielding few paired months per state and no significant associations (Fig. 5e). Nevertheless, positive correlations were observed in Ohio (ρ = 0.73, number of pairs months (n) = 8, p = 0.31), Tennessee (ρ = 0.36, n = 10, p = 1), and Georgia (ρ = 0.21, n = 10, p = 1), whereas negative but non-significant correlations appeared where paired months were few, such as in New York and Wisconsin (n = 4). Among the diagnoses investigated, enterovirus showed the broadest evidence of positive association across states, with at least weak positive correlations in 18 states (45%) (Fig. 5f). Even so, strong positive and significant correlations were detected only in Kentucky (ρ = 0.72, n = 27, p = 0.01), North Carolina (ρ = 0.65, n = 28, p = 0.01), and Georgia (ρ = 0.59, n = 31, p = 0.02). Correlations between wastewater EV-D68 RNA concentrations and wheezing in children ≤5 years were heterogeneous and not significant overall, aligning with the national result (Fig. 5g). In contrast, wheezing in adults ≥65 years tended to correlate negatively with wastewater, particularly in Georgia, Florida, and Texas (Fig. 5h).

Fig. 5.

National and within-state correlations between wastewater EV-D68 RNA concentrations and clinical diagnoses. (a–d) National time-series comparing wastewater EV-D68 RNA concentrations (black lines, right axis) with proportions of clinical diagnoses (bars, left axis) for wheezing in children ≤5 years (A), wheezing in adults ≥65 years (b), enterovirus-specific encounters (c), and acute flaccid myelitis (AFM) (d). (e–h) State-level Spearman correlations between wastewater EV-D68 and AFM (e), enterovirus-specific encounters (f), wheezing ≤5 years (g), and wheezing ≥65 years (h). Points represent correlation coefficients, with colors denoting effect size and significance (adjusted p < 0.05, Bonferroni). Correlations at the state level were performed using data filtered to the EV-D68 seasonal duration defined by wastewater and extended ±2 weeks for clinical diagnoses, and ±2 months for AFM.

Discussion

In this nationwide analysis of wastewater solids from 147 WWTPs across 40 U.S. states, we identified a biennial pattern in EV-D68 circulation, with limited activity in 2023 and 2025, and widespread detections during summer and fall 2024, consistent with previous clinical investigations.^11,26 When we compared wastewater patterns with symptom-based clinical diagnoses used as population-level proxies for EV-D68 clinical activity from Epic Cosmos, such as wheezing, enterovirus infection, and AFM cases, the correlations were weak or absent. This limited concordance is expected given the population-level nature of wastewater measurements and the reliance on nonspecific, symptom-based clinical proxies for comparison, as these metrics capture fundamentally different underlying populations. Wastewater RNA concentrations reflect aggregate viral shedding from the entire contributing population, including asymptomatic and mildly symptomatic infections, whereas clinical diagnoses represent a small, care-seeking subset of individuals. Most EV-D68 infections cause mild or no symptoms and therefore rarely prompt clinical testing, and EV-D68 specific diagnostics are not routinely performed. As a result, the clinical indicators available in routine surveillance are nonspecific and provide an incomplete representation of true EV-D68 circulation.⁷ Accordingly, wheezing and general enterovirus diagnoses did not consistently match wastewater concentrations, and AFM counts were often masked to zero, limiting the ability to analyze temporal associations. In a few states, including Ohio, Tennessee, and Georgia, wastewater concentrations and AFM cases increased at the same time, but these patterns could not be reliably assessed at the national level because of clinical data masking and sparsity. These findings highlight the structural limitations of clinical datasets for EV-D68, where infrequent molecular testing and reliance on symptom-based diagnoses limit epidemiological interpretation.

We observed that EV-D68 wastewater concentration reflects climatic patterns. Temperature and dew point influence the center of season of EV-D68, which tends to circulate earlier in warm, moist environments. Similar observations were made in the past through modeling clinical case data of enteroviruses.10, 11, 12 Dew point increases with both temperature and humidity and therefore serves as an effective indicator of warm and humid atmospheric conditions. Summer viruses such as EV-D68 remain stable at high relative humidity, and warm, moist air preserves airway physiology and mucosal environments that support viral stability and efficient replication.27, 28, 29 Results indicate that EV-D68 begins circulating earlier in areas with consistently warm, humid conditions. Consistent with this, the EV-D68 center of season ranged from early summer to late autumn 2024, occurring earlier in the south and east, aligning with previous analyses of clinical non-polio enteroviruses and EV-D68 cases, suggesting broadly conserved seasonal patterns despite post-pandemic changes.^2,11 This concordance shows that wastewater measurements can reliably reveal the spatiotemporal variation of EV-D68 circulation, even when clinical data are scarce. Expanding on this, a similar approach could be taken for many pathogens that do not have accurate clinical testing.

The EV-D68 center of season was largely unrelated to sociodemographic variables, underscoring that environmental factors primarily determine when circulation begins. In contrast, season duration was not associated with environmental variables but reflected demographic structure and social mixing. Prolonged EV-D68 season duration occurred in areas with higher population density, greater urbanicity, more crowded households, and higher densities of childcare facilities, hospitals, and nursing homes. However, we did not observe a prolonged duration with a lower birthrate as previously reported.¹¹ Age structure was not associated with duration, even though children and older adults often present with more severe symptoms.¹ This suggests that prolonged community circulation is maintained through overall contact density rather than age-specific susceptibility. Higher childcare density in particular could promote frequent close–contact interactions among young children, a key group for transmission and amplification.³⁰ Once EV-D68 becomes established, transmission can persist through frequent interpersonal contacts and ongoing reintroduction in dense or highly connected communities. Usually, for many viruses, studying these patterns is difficult because clinical data are limited, but wastewater surveillance enables direct measurement of these epidemiological parameters.

In California, EV-D68 was detected continuously from late 2023 to mid-2025, suggesting infections occurred continuously for nearly two years instead of the expected brief, biennial pattern. Prior modeling studies predicted a large and prolonged EV-D68 outbreak in California in 2016 and a much smaller one in 2018, suggesting that the unusually extended 2024 season may signal a loss of the regular biennial rhythm.² Such deviations are epidemiologically important, and for viruses with limited clinical surveillance, they might go undetected, highlighting the power of wastewater data as a complementary source to investigate these patterns.

Limitations of our study include the lack of large-scale, high-resolution laboratory-confirmed EV-D68 case data, which limited clinical comparisons. As a result, symptom-based diagnoses from Epic Cosmos were used as proxies for population-level EV-D68 clinical activity. These proxies are nonspecific and constrain concordance with wastewater signals. Second, we quantified EV-D68 RNA without strain-level typing. This prevented assessment of whether lineage-specific dynamics contributed to observed patterns or clinical proxies. Third, estimates of epidemiological metrics rely on analytic choices required for data aggregation. Choices such as smoothing, scale-specific season-duration definitions, and regression weighting may influence effect sizes and interpretation, even when applied consistently.

Beyond study-specific constraints, wastewater surveillance has inherent limitations when used to infer epidemiological patterns. For public health decision-making, clinical symptoms and disease severity remain important, and wastewater cannot distinguish symptomatic from asymptomatic infections or indicate how severe infections are. Wastewater signals suggest that all infections have the same clinical impact, so individual testing remains necessary to assess healthcare burden. Interpretation may also be biased when viruses have highly variable shedding patterns, as large differences in shedding between individuals or over the course of an infection can weaken the connection between wastewater concentrations and true infection rates. Although recent work suggests that wastewater-based estimates can remain robust despite differential shedding, future studies should include dedicated shedding investigations to improve the interpretation of wastewater concentrations.³¹ Furthermore, analyzing epidemiological factors is challenging in settings with limited environmental variability or where sociodemographic data are unavailable. Lastly, although EV-D68 wastewater surveillance has already been implemented on a near-national scale in the U.S.,¹⁴ further expansion will rely on factors such as coordinating sampling and reporting, expanding to more catchments, and balancing between geographic coverage and sampling frequency within fixed resources. Implementation may be more challenging in decentralized wastewater settings, where targeted surveillance at institutions like schools, hospitals, or universities might be more feasible.

Many important epidemiological questions benefit from dense, continuous data streams that are often unavailable for EV-D68 and other pathogens. Wastewater provides these data consistently across space and time, enabling the reconstruction of circulation patterns even when clinical datasets are sparse, masked, or nonspecific. Future work should extend the use of wastewater data beyond demonstrating concordance with clinical trends toward systematically characterizing epidemiological parameters. These can have practical implications: understanding where prolonged circulation is likely can help guide the timing of clinical testing and the deployment of preventive measures where sustained transmission may occur. As wastewater networks continue to expand globally, they offer a scalable, high-resolution data source for studying pathogens with limited or no clinical testing, allowing for spatiotemporal analyses, outbreak prediction, and assessments of how climate change might influence viral dynamics.

Contributors

S.C.: conceptualization, data curation, formal analysis, investigation, methodology, validation, visualization, writing–original draft, and writing–review & editing. A.Z.: conceptualization, data curation, methodology, and writing–review & editing. A.B.B.: conceptualization, funding acquisition, project administration, resources, supervision, validation, writing–review & editing.

Data sharing statement

All data used in this study, along with the Python scripts and R code for statistical analyses, are publicly available at the Stanford Digital Repository: https://purl.stanford.edu/xz608ws5349.

Declaration of interests

AB has received funding from the Sergey Brin Family Foundation for this work. All other authors declare no competing interests.