Heavy Precipitation, Drinking Water Source, And Acute ... - PLOS

Discussion

Our study brought together several sources of data to address our central hypothesis, that heavy precipitation is associated with increased incidence of AGI in Philadelphia, in the days following the precipitation event. In addition to the hypothesis we tested, we postulated that any associated increase occurred via waterborne transmission. We found an association between heavy precipitation and AGI incidence in Philadelphia that was primarily limited to the spring season. Heavy precipitation was consistently associated with AGI in spring, whereas AGI increases with spring streamflow were not statistically significant. In our consideration of up to 28 days lag, we found significant increases in AGI during spring that peaked from 8 to 16 days following a heavy precipitation event.

A mechanistic pathway involving waterborne transmission of AGI begins with heavy precipitation causing higher than normal concentrations of pathogens to enter source waters, either with stormwater runoff or from CSOs. Stream bottom resuspension of sediment is also likely to play a role [17]. Philadelphia residents may then be exposed through direct contact with river water/sewage or by ingestion of contaminated drinking water, although the plausibility and importance of these pathways have not been adequately investigated. Philadelphia’s combined sewer system covers almost two-thirds of the sewer service area and discharges an estimated 16 billion gallons of overflow effluent per year [18]; however, the city’s 164 combined sewer outfalls are located downstream of river water intakes for the drinking water supply [19] (Fig 1). Nevertheless, Philadelphia CSOs remain a concern for the possibility of direct contact with contaminants, and CSOs in upstream municipalities are a relevant concern for contamination of Philadelphia’s drinking water sources.

As an example, we consider the plausibility of exposure pathways for Cryptosporidium, a frequent cause of waterborne gastroenteritis. Cryptosporidiosis has been associated with heavy precipitation in multiple studies, presumably due to contamination of water supplies with runoff and flooding [3, 13]. In an investigation of Delaware River water at Trenton, NJ, Cryptosporidium concentration was increased following precipitation [17], and was even more strongly correlated with other variables such as streamflow and turbidity.

Recreation in untreated rivers and lakes is a known cause of disease outbreaks [20], and studies of sporadic cryptosporidiosis have also identified freshwater swimming as a risk factor [21–22]. Recreational contact with local waters may have contributed to the increases in AGI we observed following heavy precipitation; however, the association was present only during the spring season when recreational contact is likely limited due to cold river water temperatures. Furthermore, we there was no association with heavy precipitation during the summer, when recreational contact is expected to be high. An additional exposure pathway involving direct contact with sewage may occur with heavy precipitation causing sewer lines to back-up into basements, but again, there is no reason that this scenario would be more likely to occur with heavy precipitation during spring. It is possible that seasonal differences in the relationship between heavy precipitation and AGI could result, not from differences in contact with water, but from differences in the presence of pathogens between seasons. For example, Cryptosporidium concentration was highest during winter and spring in the Delaware River survey at Trenton, NJ [17], coinciding with the highest river flow and turbidity, and consistent with the association we observed in the spring season.

Waterborne illness via ingestion of contaminated drinking water could occur if pathogen input from source waters evades treatment. Cryptosporidium is commonly detected at low levels in source waters for Philadelphia’s water supply [8], and was also frequently detected in a survey of filtered drinking water supplies [23]. Cryptosporidium can be resistant to drinking water disinfection with chlorine [24–25], particularly during its spore phase in which oocysts are encased by a thick wall. Outbreaks of cryptosporidiosis through drinking water have occurred primarily with suboptimal treatment; however, rates of sporadic cryptosporidiosis with treated water supplies are unknown [24]. Epidemiologic studies of the relationship between turbidity of finished water from public systems and AGI incidence rather consistently find a positive relationship [14], which may indicate risks due to residual microbial contamination of treated drinking water. Another pathway to contamination of drinking water by pathogens is within the distribution system, through cross-contamination of water distribution pipes by sanitary sewer pipes, which can occur under conditions of low water pressure or when there is a water main break; however, it is not clear that cross-contamination would be more likely to occur following heavy precipitation.

The highest AGI increases and most consistent results in our study were found with the exposure metric of precipitation falling within the local watersheds. These results were robust to all sensitivity analyses and adjustments, and demonstrated exposure-response relationships. Precipitation measured at the Philadelphia airport (PHL) was also associated with increased AGI incidence, but with less consistency across analyses. AGI increases observed with streamflow were not consistently elevated during the spring season, and any observed increases in spring were diminished with adjustment for 4-week cumulative precipitation. Interpretation of the differences between exposures is speculative. We created the watershed precipitation exposure metric to represent precipitation that would directly impact the Delaware- and Schuylkill River source waters for Philadelphia’s drinking water supply, via runoff into the rivers. In contrast, precipitation at PHL is measured at one watershed location below the PWD treatment plant intakes. The stronger and more consistent findings with watershed precipitation may indicate a relevant exposure pathway through contamination of source waters, and subsequently, the water supply. However, inconsistent associations with streamflow during spring suggest that runoff causing high streamflow is not an important part of the pathway. These findings may indicate that the input of pathogens to source waters is more important than the amount of runoff, itself. Observed AGI increases with precipitation and not streamflow might also indicate importance of an exposure pathway involving direct contact with contaminated stormwater in the city, such as through backed-up sewers, that can occur independently of streamflow levels. Further characterization of exposure pathways is needed to more fully understand the utility of particular exposure metrics for predicting AGI risks with heavy precipitation.

AGI increases associated with precipitation and streamflow variables in our study peaked at different lags following a heavy precipitation event, from 8 to 16 days. These lags represent the amount of time from the heavy precipitation event until presentation of a patient with AGI to the emergency department. For an exposure pathway involving drinking water, the lag time incorporates the time from river contamination until the drinking water reaches the consumer’s tap, incubation periods for common microbiological infections, and any delay in seeking medical care for AGI. Given the ecological nature of the time-series study design, the lag time represents the average lag within the area studied. We suspect that a lag time based on direct contact with contaminated stormwater would be shorter than for a pathway involving ingestion of contaminated drinking water, because of the timing of exposure after a heavy precipitation event. However, it is difficult to infer particular mechanisms based on lag times seen in our results, given disparate lags between exposure variables and between the entire year and the spring season. We are nevertheless assured that the peaks we observed are fairly consistent with those reported in similar research [3,14].

The AGI metric in our study, based on emergency department visits, is subject to lack of sensitivity, as typically only severe AGI cases will seek emergency care. Undercounting is common in timeseries studies that use patient visit counts as the dependent variable; this undercounting reduces the study power but will not cause bias as long as the variation in identified AGI cases is representative of the variation of the underlying true daily counts. It is also likely that persons with lower socioeconomic status use emergency care more frequently than the rest of the population (e.g., persons who do not have health insurance), and therefore these population groups are overrepresented in the AGI counts. This overcounting of a subpopulation would not affect our results unless there was a systematic difference in the association between precipitation and AGI among this subgroup (e.g., if this subgroup was more susceptible to waterborne AGI), compared to the rest of the population–and even then, it would not necessarily dismiss the overall association.

The syndromic surveillance system we used to identify AGI is designed for rapid identification of spikes in illness; this system has not been checked for accuracy of disease classification. Furthermore, the grouping of diarrhea and vomiting syndromes lacks detailed etiology. Nevertheless, syndromic surveillance data for gastroenteritis had a strong relationship with norovirus outbreaks in a US national database [26], indicating its utility for tracking major infectious illnesses. Furthermore, syndromic surveillance data has been used in other disease research; a study conducted in New York City reported that daily counts of the diarrhea syndrome (identified using the same algorithm as in Philadelphia) were associated with daily average turbidity of the drinking water supply [27]; interesting with comparison to our results, the association was only present during the spring months. There is a well-known peak of highly communicable viral agents in late winter/early spring, such as norovirus and rotavirus [26], and this clear seasonal pattern was apparent in the syndromic surveillance data we utilized in our study (Fig 2A). Cryptosporidiosis, in contrast, occurs predominantly in summer [28]. In our time-series analysis, we adjusted for predominant seasonal patterns of AGI within the calendar year, as well as adjusting for season (4 discrete seasons, as defined in our study). These adjustments appeared to effectively control for seasonal confounding of the relationship between heavy precipitation and AGI, as judged by a plot of residuals across the days of the study period (S1 Fig).

Our findings follow research conducted in Philadelphia by Schwartz et al. in the 1990s [6–7], in which turbidity of finished drinking water was associated with increased incidence of AGI in children and the elderly. We did not study turbidity directly, as we were not granted access to these data by the PWD (personal communication, PWD). Instead, we studied heavy precipitation as part of the same hypothesized mechanistic pathway as turbidity, although turbidity is a surrogate for microbial contamination whereas precipitation is a cause of microbial contamination. Our results suggest that despite improvements to Philadelphia’s drinking water supply since the Schwartz studies were conducted, such as notable reductions in turbidity since the 1990s [8], there may still be conditions under which detectable increases in AGI are caused via waterborne transmission.

We found an association between heavy precipitation events and AGI incidence in Philadelphia, limited to the spring season. While it is not possible to confirm the exposure pathway or implicated pathogens from our data, these details could be addressed in future research. Targeted sampling strategies to identify specific pathogens following heavy precipitation events during the spring season may provide evidence designating water as the medium of exposure and could help clarify the mechanistic pathway (e.g., source water, finished water, CSOs). Development of mitigation strategies will only be possible with further information.