A brief overview of the evolution of National Primary Drinking Water Regulations and the Unregulated Contaminant Monitoring Rule

Prior to the 1970s, nationwide studies revealed that water systems had been struggling with inadequate facilities, poor operating procedures, and uneven management of public water supplies in communities of all sizes, leading to extensive water quality problems across the country [15,16]. The Safe Drinking Water Act (SDWA), passed in 1974, gave EPA authority to regulate drinking water contaminants and delegate enforcement and implementation of regulations to states, territories, and tribes. Major amendments to the SDWA occurred in 1986 and 1996 to ensure access to safe drinking water to protect public health.

Through their statutory authority, under the SDWA, EPA sets national drinking water standards by determining whether three main conditions are met: “(1) The contaminant may have an adverse effect on human health, (2) The contaminant is known to occur or there is substantial likelihood that the contaminant will occur in public water systems with a frequency and at levels of public health concern, (3) in the sole judgement of the Administrator, regulation of the contaminant presents a meaningful opportunity for health risk reduction for persons served by PWS” [17]. Once a National Public Drinking Water Regulation (NPDWR) is in effect, enforcement and implementation of the regulation is primarily the responsibility of states [16].

The EPA evaluates the Contaminant Candidate List (CCL), a list of contaminants that are unregulated but may warrant regulation by EPA and are known or expected to occur at public water systems, and other priority contaminants every five years to determine which contaminants should be monitored under the UCMR based off a multi-step prioritization [18]. UCMR occurrence data is used to advise future regulatory standards. All public water systems (PWSs) serving 10,000 or more people are required to monitor under the UCMR; for PWSs serving less than 10,000 people, 800 representative PWSs, including tribal water systems, are chosen to monitor [19].

UCMR 2 monitored contaminants from 2008 to 2010. Of the contaminants monitored are two insecticides, five flame retardants, three explosives, three parent acetanilides, six acetanilide degradants, and six nitrosamines [20] (Fig 1). UCMR 3 monitored contaminants from 2013 to 2015. Of the contaminants monitored are seven volatile organic compounds, one synthetic organic compound, six metals, one oxyhalide anion, six per fluorinated compounds, seven hormones, and two viruses [21] (Fig 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Contaminants monitored under UCMR 2 and UCMR 3 categorized by contaminant type.

https://doi.org/10.1371/journal.pwat.0000208.g001

Since the SDWA’s inception, there have been over 90 NPDWRs established through the SDWA and the overall drinking water quality in the United States has improved [17] as evidenced by decreases in infectious waterborne outbreaks [22]. However, the nation still faces challenges to ensure safe drinking water for all. Community water systems across the country are often under-resourced and lack adequate funding. Previous research suggests that these overburdened water systems may incur monitoring and reporting violations to avoid health-based violations that are often accompanied by stricter penalties such as requiring expensive treatment techniques [23]. While many community water systems are stuck in the cycle of not having the resources or funding to comply with NPDWRs and then faced with costly penalties they cannot afford, a general distrust of drinking water has grown in the United States. Reliance on bottled water has increased since the Flint, Michigan water crisis [24], most notably among Black and Hispanic populations [25]. Rebuilding trust requires more direct examinations of potential health effects and home water quality that can both identify associations of concern that require interventions and provide consumer peace of mind when identify the absence of adverse health outcomes.

Contaminants of interest

This section will review the human and/or fetal health impacts of unregulated drinking water contaminants that were identified in a water system service area in Virginia that monitored under UCMR 2 or 3 (Table 1). Only one contaminant, N-Nitroso-dimethylamine (NDMA), was identified in samples collected under UCMR 2, and 11 contaminants were identified in samples collected under UCMR 3.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Summary of literature reviewed for each contaminant and their corresponding minimum reporting level (MRL) under the UCMR.

https://doi.org/10.1371/journal.pwat.0000208.t001

UCMR data

Data for UCMR 2 and 3 were acquired from the EPA’s National Contaminant Occurrence Database (NCOD) [45]. Downloaded data included: PWS ID and name, PWS type, counties and cities served, contaminant name, and sample collection date. As water system service area boundaries for Virginia are not publicly available [46], estimated service areas (n = 662) used in previous work by Marcillo et al. (2021) [47] and Young et al. (2023) [5] were applied. In brief, the approach developed by Marcillo et al (2021) [47] assigned populations to community water systems (CWSs) based on population served and proximity to the system (Fig 2). To determine proximity to CWS, the closest zip codes to a CWS were assigned to a corresponding water system until the population served by that system was fully assigned. It is important to recognize that water service areas do not always develop according to population proximity and can be driven by local socio-economic factors and politics [48], so these service areas should be viewed as an approximation. However, it is also worth noting that most counties in Virginia comprise multiple separately managed community water systems which can vary in service and quality. Therefore, the use of estimated service area boundaries, as opposed to county boundaries, enables a finer scale examination that potentially limits exposure misclassification. Only CWSs that had estimated service area data available and that sampled under UCMR 2 and 3 were considered for the analysis. UCMR 2 and UCMR 3 were analyzed separately; a total of 336 CWSs sampled under UCMR 2 and 346 CWSs sampled under UCMR 3 (Fig 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. This study includes 346 CWSs, represented as service areas, that sampled under UCMR 2 and UCMR 3.

https://doi.org/10.1371/journal.pwat.0000208.g002

Birth outcome data

Virginia birth records (n = 567,342) for the years covered by the UCMR 2 (2008, 2009, 2010) and the UMCR 3 (2013, 2014, and 2015) were acquired from the Virginia Department of Health (VDH) and accessed for research purposes from 23, February 2020 to 1, August 2021. During analysis, authors had access to information that could identify individual participants (e.g., mother’s address) stored on a secure university computer; the Virginia Department of Health IRB waived the requirement for individual informed consent for use of birth records. All analyses were performed according to protocols to protect confidentiality as approved by both Virginia Tech IRB (VT IRB #17–1190) and the Virginia Department of Health IRB (VDH IRB #40221). Birth record data included: mother’s self-reported information (including race, education, age, Hispanic origin, tobacco use, parity, and address) child information collected at birth (sex, gestational length, weight, plurality) and payment method (private insurance, Medicaid, self-pay).

Data processing

Monitoring frequencies can vary across systems depending on their size and source water type, and contaminant type [49]. Given the variation in monitoring frequency, analysis was conducted on an annual basis. Exposure period was classified via majority gestational year (MGY). Birth records were assigned to the year in which the majority of gestation occurred using gestational age and birth date to determine the midpoint of the gestational period [50,51]. For example, a birth that occurs in February would be matched with contaminant occurrence data in the previous year when most exposure during pregnancy would have occurred, rather than the year in which the birth occurred. Three outcomes of interest were calculated based on data acquired from the birth records: low birth weight (LBW) = <2500 grams, preterm birth (PTB) = <37 weeks of pregnancy, and term low birth weight (tLBW) = <2500 grams and >37 weeks of pregnancy [52,53], with each outcome classified as a binary variable (i.e. present/absent). The study was limited to singleton births, with all plural births removed due to their increased risk of preterm birth and low birth weight [54]. Birth records with any of the following were removed from the study: P.O box address, unknown birth weight, unknown gestational age, and incomplete address (Fig 3), resulting in a final sample size of 435,449 births. Esri’s 2013 StreetMap dataset was used to geocode birth records.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Data processing for individual birth records in Virginia.

Note birth records were geocoded to an estimated service area boundary that sampled under UCMR 2 and UCMR 3 (n = 346).

https://doi.org/10.1371/journal.pwat.0000208.g003

Contaminant occurrence was aggregated at the estimated service area level for each year and binarily classified into [x] contaminant was not identified at a level above or equal to the MRL (0) or [x] contaminant was identified one or more times at a level above or equal to the MRL (1), where [x] contaminant represents one unregulated drinking water contaminant. Birth records were then joined to corresponding service areas for analysis. Contaminants that did not occur at one of the 346 estimated service areas in Virginia that were sampled under UCMR 2 and 3 were not considered.

Statistical analyses

Logistic regression models were built (or used) to determine if a relationship exists between unregulated drinking water contaminants and adverse birth outcomes within service areas in Virginia. Each birth outcome (PTB, LBW, and tLBW) was examined separately with each individual contaminant. Odds ratios and 95% confidence intervals (CI) were calculated to understand the relationship between the outcome of interest and unregulated contaminants. The following covariates were entered into the regression model: water system service area; majority gestational year; infant’s sex; parity/birth order; mother’s race, age, and education; if mother is of Hispanic origin; tobacco use during pregnancy (yes/no); and payment method. All analyses were completed using JMP statistical software.

Birth outcomes

PTB, LBW, and tLBW were examined in relation to individual unregulated contaminants (Table 4). NDMA was the only contaminant significantly positively associated with PTB (OR = 1.08; 95% CI: 1.02–1.14). Molybdenum (OR of 0.92; 95% CI: 0.87–0.97) and vanadium (OR of 0.96; 95% CI: 0.92–1.00) were significantly negatively associated with LBW. Molybdenum was also significantly negatively associated with tLBW (OR of 0.90; 95% OR: 0.82–0.99), though HCFC-22 was significantly positively associated with tLBW (OR 1.18; 95% OR: 1.01–1.37).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Odds ratios (CI 95%) for associations between UCMR contaminants and PTB, LBW, and tLBW.

Adjusted for water system service area, mother’s race, mother’s age, mother’s education, if mother is of Hispanic origin, parity, majority gestational year, infant’s sex, payment method, and tobacco use during pregnancy.

https://doi.org/10.1371/journal.pwat.0000208.t004

Spatial distribution of contaminant occurrence

Contaminants significantly associated with birth outcomes in Virginia (i.e. NDMA, molybdenum, HCFC-22, and vanadium) were detected in different regions of the states (Fig 4). NDMA, which was only targeted under UCMR 2, was detected in more urban central and coastal areas: in 2008, it was identified in CWSs north of Virginia Beach and in 2009 and 2010, it was identified near Virginia Beach and Richmond. Molybdenum was detected in community water systems near Richmond, Roanoke, Virginia Beach, and Fairfax in 2014 and 2015 (Fig 4). HCFC-22 was detected only in 2014 in CWSs near Richmond, Roanoke, and northern Virginia (Fig 4). Vanadium was notably the most widespread throughout the state, as it was detected in systems near Richmond, Roanoke, Radford, and in northern Virginia in 2014 and 2015.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4.

(A) NDMA occurrence in CWS service areas within Virginia under UCMR 2. (B) HCFC 22 occurrence in CWS service areas within Virginia under UCMR 3. (C) Molybdenum occurrence in CWS service areas within Virginia under UCMR 3. (D) Vanadium occurrence in CWS service areas within Virginia under UCMR 3. Source: U.S. Census Bureau (shapefile: cb_2018_us_state_5m)

https://doi.org/10.1371/journal.pwat.0000208.g004

Limitations and future research

Assumptions were necessary to conduct this cross-sectional study. As mentioned previously, service areas were approximated [47] in order to refine exposure classification at a finer than county scale. The absence of publicly available service area data is recognized as a limitation in examinations of potential trends in community water exposures [46]. In addition, this work pre-supposes that mothers are consuming the water supplied to their homes, though recent work suggests that the consumption of bottled water and/or consumer treatment of home water prior to consumption is increasing, particularly in minority communities [25]. The detection of significant relationships despite these limiting assumptions is notable.

Addresses of birth records were geocoded, presenting locational error at the individual level. As opposed to addresses in a city, there is often a higher geocoding error when locating addresses in a rural area. Although 88% of Virginia’s population lives in an urban area [58], it is worth noting that there is evidence of elevated incidence of adverse birth outcomes in rural areas [59].

Additional limitations of this study include assignment of births to majority gestational year, available covariates, and binomial classification of contaminants. Given that contaminant occurrence was aggregated yearly, majority gestational year (MGY) was estimated to define the exposure period. Contaminants were analyzed with birth records that corresponded to the MGY. Results of the study could be skewed if the infant was exposed during the minority gestational year (e.g., exposed in December and born in July). Future research could benefit from examining drinking water contaminants quarterly, rather than annually, particularly in areas with high adverse birth outcome incidence. It is worth noting that during the study period, race was recorded differently on original birth records. Prior to 2013, child’s race was used in lieu of mother’s race as mother’s race was not recorded on birth records until 2013. Additionally, ethnicity was reported as whether a mother was of Hispanic origin or not. Combinations of reported ethnicity and race were not considered, nor were races other than Black, white, or the combined category of other self-reported races due to low numbers in individual categories. Finally, since individual statistical tests were completed for each contaminant, the possibility of spurious associations is increased.

The EPA requires all PWSs serving over 10,000 people and a subset of 800 PWSs serving less than 10,000 people to monitor for contaminants under the UCMR [18]. This study only considered contaminant info for water systems that were required to monitor under UCMR 2 and UCMR 3. Since monitoring frequency varies by water system size, contaminant occurrence was represented as binomial variables (yes = the contaminant occurred at least once at or above the MRL; no = the contaminant did not occur at or above the MRL). Future studies could benefit from looking at contaminant concentration and frequency of occurrence. As two of the contaminants with positively significant associations, molybdenum and vanadium, are geogenic in origin, it is important to recognize that the trends presented here with regard to contaminant occurrence are not necessarily generalizable to other states or regions with dissimilar hydrogeology; application of the demonstrated research framework, in particular in examining relations observed in Virginia, is highly recommended. Furthermore, these results suggest the need for further evaluation to determine if there is a potential risk for adverse birth outcomes when unregulated drinking water contaminants occur, especially considering frequency of occurrence at a finer spatial scale, which would allow the researcher to determine if additional drinking water regulations should be implemented locally.