Association between Combined Sewer Overflow Events and Gastrointestinal Illness in Massachusetts Municipalities with and without River-Sourced Drinking Water, 2014–2019

In $>700$ municipalities in the United States, combined sewer systems (CSS) collect residential and nonresidential sewage, industrial waste, and stormwater that flow together to a wastewater treatment facility. CSS discharge untreated combined wastewater into rivers and lakes when the volume of water in the common collection pipe exceeds treatment capacity, potentially posing risks to the health of communities and ecosystems downstream of discharge points. Discharge events—called combined sewer overflow (CSO) events—are driven by heavy precipitation and runoff from snowmelt.^1,2 Observed and predicted increases in the intensity and frequency of precipitation events due to climate change in the Northeast and Midwest, where most CSS in the United States are located, underscore the importance of understanding and addressing the health risks associated with CSO events.^3–5

CSO events introduce viral, bacterial, and protozoan pathogens, as well as chemical and physical contaminants, to surface waterbodies.^1,2,6–14 Compared with dry weather conditions, CSO events can lead to one to two orders of magnitude increases in the concentration of waterborne pathogens and fecal indicator organisms in receiving waterbodies.^2,6,15 Exposure to sewage-associated pathogens can lead to gastrointestinal, respiratory, and soft tissue infections among recreational and drinking water users.^16–21 The association between fecal contamination of surface waters and gastrointestinal illness is well documented in settings with modern wastewater infrastructure,^22–25 as is the relationship between precipitation, which is the primary driver of CSO events, and gastrointestinal illness.^26–28 However, relatively little research has been conducted to determine the direct association between CSO events and health outcomes.

The few studies that have investigated the relationship between CSO events and health outcomes suggest a link between discharge events and acute gastrointestinal illness (AGI). Research has demonstrated elevated rates of AGI following large-volume CSO events among residents of Atlanta, Georgia,²⁹ and increased odds of AGI among children living near CSO outfalls in Cincinnati, Ohio, after any CSO event.³⁰ Studies of recreational use of CSO-impacted waterbodies suggest that the risk of AGI among swimmers can increase up to five times after a CSO event compared with when no CSO event occurred³¹ and that the likelihood that recreators will develop AGI can vary widely by activity and level of water contact.⁹ CSO events in drinking water sources may also pose a risk to community health. A study in the Milwaukee, Wisconsin, metropolitan area found a higher rate of pediatric cases of AGI following partially treated sewage discharge events in a drinking water source.³² In Massachusetts, increased risk of AGI was observed following extreme precipitation events in a region with a CSO-impaired drinking water source, whereas the risk of AGI did not change following similar precipitation events in other regions of the state.³³ Importantly, neither of the studies involving a CSO-impacted drinking water source used CSO event data in their analyses but, instead, respectively used partially treated sewage discharge events or precipitation data to characterize exposure.^32,33 In all studies, the association between discharge events and reported AGI has a lag of 2 to 8 days (d), which could be due to pathogen travel time in the waterbody, timing of recreational activity, drinking water treatment and distribution, pathogen infection and incubation, or decision to seek care.^29,30,32,33

There remain key gaps in our understanding of the association between CSO events and health, including the robustness of the association, relevant exposure pathways, and the role of precipitation events. Accordingly, we sought to assess the association between CSO events and emergency department (ED) visits for AGI in the Massachusetts municipalities that border the Merrimack River, a drinking water and recreational resource for surrounding communities. We evaluate whether the association between CSO events and AGI differs by municipal drinking water source and after adjusting for precipitation events. We hypothesize that a) upstream CSO events increase risk of AGI in downstream municipalities after adjusting for precipitation, and b) the risk of AGI in municipalities that exclusively source their drinking water from the Merrimack River is more strongly associated with upstream CSO discharge than in municipalities with other drinking water sources. This study builds on the work by Jagai et al.—led by one of the coauthors of the present analysis—that was conducted in the same region of Massachusetts³³ by leveraging a self-matched study design, incorporating CSO discharge volume data in the exposure classification, stratifying municipalities by drinking water source, and analyzing outcome data from more recent years. To our knowledge, this is the first study to directly investigate the role of drinking water source in the association between CSO event discharge volume and AGI.

In this analysis of the association between CSO events and ED visits for AGI, we observed higher risk of AGI in the 4 d following extreme CSO events compared with days with no CSO events. Results were robust to adjustment for potential confounding by precipitation. The magnitude of the associations observed in the present analysis were similar to or larger than those found in previous studies, although methodological differences and study populations make direct comparisons difficult. Using a similar self-matched study design, Brokamp et al. found a 16% higher odds of AGI after any CSO event among children living within $500\mathrm{m}$ of a CSO outfall [$\text{odds ratio}=$ 1.16 (95% CI: 1.04, 1.30)].³⁰ This study differed in notable ways from our study, including the spatial approach to exposure classification, pediatric study population, and the binary classification of CSO events, but the magnitude of the association was similar to the CRR we observed following 95th percentile CSO events. On the other hand, Miller et al. did not use a self-matched study design and found a smaller but more precise association between large CSO events and AGI after adjusting for precipitation among Atlanta residents of all ages [$\text{rate ratio}=$ 1.09 (95% CI: 1.03, 1.14)].²⁹ The definition of a large CSO event used by Miller et al. was $\ge 56.6\text{\hspace{0.17em}Mgal}$ of discharge, which falls in the range between 95th ($27.3\text{\hspace{0.17em}Mgal}$) and 99th percentile ($78.7\text{\hspace{0.17em}Mgal}$) CSO discharge events in our study. In an earlier analysis in the Merrimack Valley region of Massachusetts that used extreme precipitation as a proxy for CSO events, Jagai et al. found a 13% increase in cumulative risk of AGI following extreme precipitation events [equivalent to $50.0\text{\hspace{0.17em}mm}$, $\text{CRR}=$1.13 (95% CI: 1.00, 1.28)].³³ This result is remarkably similar to the association observed in this analysis between 95th percentile precipitation events (equivalent to 31.5 mm) and the cumulative risk of AGI over 4 d observed in our study. Overall, we found that the strength of the associations observed in this study between 95th percentile CSO or precipitation events and AGI were fairly consistent with previous findings but that the CRR of AGI following 99th percentile CSO events was notably higher than what has been reported in similar studies.

The magnitude of cumulative CSO discharge volume was an important factor in the association between CSO events and AGI, with the largest associations observed following $\ge 95$th and $\ge 99$th percentile CSO events. Our findings suggest that a nonlinear dose–response relationship exists between upstream CSO discharge volume and AGI, a finding consistent with previous studies.^29,33 Miller et al. similarly found that only large CSO events ($\ge 75$th percentile CSO events by volume, equivalent to $\ge 56.5\text{\hspace{0.17em}Mgal}$) were significantly associated with AGI in the week following an event, but models estimating the association between any CSO event and AGI resulted in no association.²⁹ Jagai et al. found that only the most extreme precipitation events (99th percentile, equivalent to 50.0 mm) were significantly associated with increased cumulative risk of AGI; lower magnitude, nonstatistically significant associations were found for lower percentile precipitation events.³³ These results are particularly relevant given that the frequency and intensity of extreme precipitation events (and extreme CSO events as a result) are expected to increase due to climate change in the Northeast and Midwest, where most CSS in the United States are located.³

We hypothesized that CSO events would be more strongly associated with risk of AGI among municipalities whose drinking water is sourced from the Merrimack River vs. municipalities that use other sources of water. However, in stratified analyses we found that the risk of AGI associated with CSO events was similar regardless of drinking water source. Although the adjusted estimates following extreme CSO events (95th and 99th percentile by volume) were less pronounced among municipalities using the Merrimack as their drinking water source compared with those with other sources, the overlap in the CIs between the associations for the two groups suggests that these estimates are not statistically significantly different. These results suggest that contaminated municipal drinking water may not be the only exposure pathway linking CSO events with AGI risk. The association between extreme precipitation and AGI risk was slightly higher among municipalities sourcing drinking water from the Merrimack River, but the estimates remain statistically indistinguishable owing to their overlapping CIs. Although the differences between associations by strata are modest at best, it is possible that heavy precipitation could directly influence the risk of AGI among municipalities with surface water sources independent of CSO events. A number of studies have found an association between extreme precipitation and gastrointestinal illness in areas served by surface water sources without CSS,^26,27 and a study in New Jersey found positive associations between rainfall and AGI among areas served by surface water but not groundwater sources.⁴⁹ Heavy precipitation can impact surface water sources by increasing turbidity and reducing water treatment efficacy, transporting pathogens from environmental reservoirs to surface water via runoff, and resuspending pathogens from riverine sediments.⁵⁰

The evidence in the literature is inconclusive on the potential for drinking water contamination after CSO events. Although it is conceptually possible for treated drinking water to be contaminated by CSO discharge,^51,52 studies of health risks that incorporated drinking water source contamination following discharge events either did not include CSO event data in the exposure^32,33 or found that controlling for CSO events as a binary covariate did not alter the association between precipitation and AGI.⁴⁵ Interviews of workers at drinking water treatment utilities located within $1.6\text{\hspace{0.17em}km}$ downstream of CSO outfalls suggest that many drinking water utilities adjust treatment based on wet weather, adding further complexity to any assessments of the association between CSO events and health outcomes in municipalities with CSO-impacted drinking water sources.¹ We are unaware of an accounting of how many municipalities have a drinking water source that is impacted by CSO discharge, so it is unclear how widespread this issue might be, although the results of this analysis suggest that drinking water contamination may not be the only important exposure pathway in a CSO-impaired river system. Understanding the critical exposure pathway(s) following CSO discharge events is an important direction for future work that has direct implications for infrastructure investment decisions and policies designed to protect public health under future climate scenarios.

The findings of this study suggest that CSO events and AGI were more strongly associated across the 4 d following an event compared with a longer lag period of 7 or 14 d. These findings are consistent with previous studies that found significant associations between sewage discharge events and AGI over lag periods from 2 to 8 d.^29,30,32,33 A 4-d lag period is also consistent with the incubation times of some of the pathogens found in waterways contaminated with sewage,^1,9,53 including noroviruses (12–48 h), rotavirus (1–3 d), Shigella (12 h–7 d), enterotoxigenic Escherichia coli (12–72 h), and Cryptosporidium (2–10 d).⁵³ Viral and protozoan waterborne pathogens can be infectious at low doses, persistent in environmental media, and less effectively removed by conventional drinking water treatment compared with bacteria.^54,55 Another consideration with a short lag time is the time between CSO event and exposure, which could differ based on exposure pathway. It is possible that people exposed primarily through recreational activity or aerosolized pathogens could ingest contaminated water within hours after a CSO event, although our analysis was not spatially refined enough to assess the possibility of exposure through the latter pathway.

This study has a number of potential limitations. First, the analysis is limited by a lack of individual-level information on exposure, including rates of consumption or quality of drinking water, amount or timing of contact with recreational waters, and recreational surface water quality experienced by individuals within the study population. The lack of individual exposure data limits our ability to gain insights about specific exposure pathways. Moreover, the lack of information about local water contamination and potential local responses to CSO events (e.g., adjustments to drinking water treatment processes) may have led to substantial exposure misclassification and likely biased results toward the null hypothesis of no association. Second, the AGI case data were deidentified, precluding identification of patients that may have been treated for AGI more than once. Our inability to account for the correlation among repeated episodes within an individual may have resulted in CIs that are perhaps too narrow. Third, people seeking care for AGI in the ED likely represent the most severe cases of gastrointestinal illness, given that $\le 10\%$ of all cases of AGI come to medical attention in acute care settings.⁵⁶ Thus, it is unclear whether these results are generalizable to milder episodes of AGI or to other communities across the country.

This study also has some notable strengths. First, the case time-series design minimizes potential confounding by differences between municipalities and effectively controls for temporal trends. Second, by using daily measures of CSO discharge volume instead of approaches taken in other studies to measure CSO exposure, such as precipitation as a proxy for CSO events,³³ modeling CSO events as binary exposures,³⁰ or aggregating to a weekly time step,²⁹ this study more directly evaluates the relationship between CSO discharge and health. Our use of daily CSO and precipitation data allowed for a refined temporal analysis in the days following an exposure event, and CSO volume data provided the opportunity to assess potential dose–response relationships between CSO discharge and AGI. We also considered the spatial relationship between CSS municipalities and downstream municipalities in our approach to assigning exposure. Although there are limitations to the characterization of exposure in this study introduced by considering upstream discharge, this approach is representative of how CSO discharge moves through the area and accumulates in downstream stretches of the river. The PRISM dataset provided spatial resolution of precipitation estimates that may account for heterogeneity in rainfall that is not captured by a regional weather station, the latter of which has been the source for precipitation data in other studies.^29,33 Finally, reliance on the Merrimack River as a drinking water source downstream of CSO outfalls provided an opportunity to compare associations between municipalities that do and do not get their drinking water from this CSO-impacted source. To our knowledge, only three studies report the association between sewage-related discharge events and health outcomes in regions with modern wastewater infrastructure where drinking water intakes exist downstream of outfalls.^32,33,45 In those studies, the exposure was either extreme rainfall as a proxy for CSO events,³³ undertreated sewage discharge events,³² or precipitation with CSO discharge included as a covariate.⁴⁵

Our findings suggest that extreme CSO events increase the risk of ED visits for AGI in downstream municipalities in the 4 d following an event and that a nonlinear dose–response relationship exists between CSO discharge and ED visits for AGI. The largest CSO events are associated with AGI across drinking water sources and concurrent precipitation levels, although we observed differences in the timing and magnitude of associations based on drinking water source. Although our study is ecological in design, our findings indicate that CSS may negatively impact public health. As precipitation events intensify as a result of climate change, large-volume CSO events may become more common and the risk of AGI in downstream municipalities may increase. Future work needed to inform CSO mitigation and improve climate change resiliency includes research clarifying the importance of different exposure pathways in the association between CSO events and AGI, expansion of this type of analysis to a regional or national scale to evaluate the robustness of the results across systems, and estimation of attributable cases of AGI following expected CSO events under future climate scenarios.

The authors acknowledge staff at US Environmental Protection Agency Region 1 and the City of Manchester for their help locating combined sewer overflow discharge data. We are also grateful to Kevin Brander of the Massachusetts Department of Environmental Protection for sharing his expertise and knowledge of the study system and to Aine Studdert-Kennedy for her assistance with data entry.

This project was supported by an Early Stage Urban Research Award from the Boston University Initiative on Cities (to B.M.H., W.H-B, and J.A.). B.M.H. was partially supported by National Institute of Environmental Health Sciences (NIEHS) grant T32ES014562 and a National Science Foundation Research Traineeship (NRT) grant to Boston University (DGE 1735087).

Dr. Wellenius serves as a consultant to the Health Effects Institute (Boston, MA) and Google, LLC (Mountain View, CA). All other authors declare they have nothing to disclose.