Water disinfection, crucial for public health, inadvertently produces disinfection by-products (DBPs). Over 600 DBPs have been identified, posing widespread human exposure risks through drinking water, inhalation, and dermal contact. Long-term exposure is linked to increased bladder cancer risk and potential reproductive effects. Current knowledge primarily focuses on regulated DBPs (e.g., THMs, bromate), while epidemiological research on non-regulated DBPs is limited due to insufficient monitoring data. Regulated DBPs represent a small fraction of total halogenated DBPs and may not be the primary toxicity drivers. Trichloroacetic acid (TCAA) is often used as a biomarker, but knowledge about other urinary HAAs is limited. A comprehensive understanding necessitates evaluating various DBP classes beyond THMs. This study aimed to estimate exposure to a wide range of DBPs in Barcelona, Spain, by describing their occurrence in tap and bottled water, developing statistical models to predict non-regulated DBPs based on routinely monitored parameters, evaluating the impact of domestic filters, and exploring the use of urinary DBPs as biomarkers. This is essential for accurate exposure assessment in epidemiological studies investigating the health effects of non-regulated DBPs.

The literature extensively documents the presence and health risks associated with DBPs, particularly regulated ones like THMs and HAAs. Studies have shown links between DBP exposure and bladder cancer, with less consistent evidence for reproductive effects. However, the focus has mainly been on THMs and to a lesser extent HAAs, with limited research on the broader range of DBPs present in drinking water. TCAA has gained attention as a potential biomarker due to its correlation with ingested TCAA from drinking water. The need to evaluate a wider range of DBP classes and the lack of adequate long-term exposure biomarkers necessitate alternative exposure assessment methods, including predictive modeling based on routinely monitored parameters. Previous studies have developed THM predictive models, but such models for non-regulated DBPs remain largely unexplored.

This study involved collecting tap water (n=42), bottled water (n=10), and filtered tap water (activated carbon, n=6; reverse osmosis, n=5) samples from Barcelona, Spain. Urine samples (n=39) were also collected from participants. Samples were analyzed for 11 HAAs, 4 THMs, 4 HANs, 2 haloketones, chlorate, chlorite, and trichloronitromethane using LC-MS/MS and GC-MS/MS techniques. Participant information (water intake, demographics) was gathered via questionnaires. Multivariate linear regression and machine learning (super learner) models were developed to predict non-regulated DBPs based on THMs, considering various transformations of variables to optimize model performance. Five-fold cross-validation was used to assess model accuracy. The impact of domestic filters was evaluated using paired t-tests, and Spearman's correlation coefficients were used to assess relationships between DBPs and physicochemical parameters. Urinary creatinine was used to adjust for urinary concentration, and the correlation between creatinine-adjusted urinary TCAA and ingested TCAA was assessed.

Chlorate, THMs, HAAs, and HANs were detected in nearly all tap water samples. Median concentrations were 214, 42, 18, and 3.2 µg/L, respectively. Multivariate linear regression models showed good predictive ability (R² = 0.8-0.9) for several non-regulated DBPs when using THMs as predictors. Activated carbon filters reduced DBP concentrations by 27-80%, while reverse osmosis filters reduced them by ≥98%. Chlorate was detected in some bottled water samples (median 13.0 µg/L). Creatinine-adjusted TCAA was the most frequent HAA detected in urine (69.2%), showing a moderate correlation with estimated drinking water intake (r = 0.48). Brominated DBPs were more prevalent than chlorinated ones, consistent with higher bromide concentrations in the source water. Significant correlations were observed between various DBPs, particularly those with similar halogen substituents. Conductivity showed strong correlations with several DBPs and physicochemical parameters.

The findings demonstrate the presence of a wide range of DBPs in Barcelona's drinking water, highlighting the potential for significant exposure. The development of accurate predictive models for non-regulated DBPs based on routinely monitored parameters like THMs is a significant contribution, enabling more comprehensive exposure assessment in epidemiological studies. The effectiveness of different domestic water filters in reducing DBP concentrations was also established, with reverse osmosis showing superior performance. The moderate correlation between urinary TCAA and drinking water intake provides further support for using urinary TCAA as a biomarker, although limitations related to short half-life and home-only water intake should be considered. The prevalence of brominated DBPs emphasizes the need to control bromide levels in source water to minimize the formation of more toxic compounds. The strong correlations between various DBPs suggest the potential of using total THM levels as indicators for other DBPs, but further investigation is necessary.

This study provides valuable insights into DBP occurrence and exposure assessment. The developed predictive models, while needing validation in larger datasets and diverse settings, offer a promising approach for estimating exposure to non-regulated DBPs using existing monitoring data. The effectiveness of domestic filters and the utility of urinary TCAA as a biomarker further enhance exposure assessment capabilities. Future research should focus on validating these models, exploring additional biomarkers, and investigating the long-term health consequences of exposure to the full range of DBPs.

The study's relatively small sample size might limit the generalizability of findings, particularly concerning the predictive models. Self-reported water consumption data may introduce measurement error. The use of spot urine samples provides a snapshot in time and might not fully reflect long-term exposure. The study focused on home water consumption and might not capture total personal exposure. Recruitment using social media may have introduced selection bias, potentially limiting the representativeness of the study population.