A new approach to a legacy concern: Evaluating machine-learned Bayesian networks to predict childhood lead exposure risk from community water systems

Improved Decision Making for Water Lead Testing in U.S. Child Care Facilities Using Machine-Learned Bayesian Networks

“…In contrast, centers served by water systems using surface water supplies showed a 12% average decrease in risk. This result is consistent with prior work highlighting groundwater reliance as an important risk factor for children’s water lead exposure in households, schools, and child care settings, ,,, likely due to a lack of, or poorly optimized, corrosion control in small and private water systems that often rely on groundwater (Figure S14).…”

“…24 • Surrounding demographic characteristics. Given research showing that low-income and minority children are disproportionately exposed to lead, 12,17,25 each facility was matched with the socioeconomic and demographic characteristics of its Census block group using 2020 American Community Survey data, 26 including the proportion of the population identifying as non-White, the proportion with a high school degree or higher, and the median household income. The final data set used for developing the models included 29 potential predictor variables and is openly available for download.…”

“…Research has shown, however, that lead occurrence in drinking water in schools and child care settings can be highly variable from one tap or building to the next, − thus making it notoriously difficult to predict. Water lead risk is commonly determined according to simple binary heuristics or risk factors such as building age despite evidence that lead occurrence in drinking water is a complex function of infrastructural, operational, chemical, historical, socioeconomic, and demographic factors that influence both the presence of lead in premise plumbing components and corrosion risk. ,− To optimize resources and protect children’s health, public health practitioners, policy makers, educators, and community water system managers need improved methods for predicting water lead risk in schools and child care settings that balance this complexity with the ease of implementation that simple heuristics provide.…”

Full Lead Service Line Replacement: A Case Study of Equity in Environmental Remediation

“…Ideally such a model would employ premise-level data for all variables, which in turn would require access to Census Bureau data on household characteristics by street address. The study team did not have such access and therefore used area-based measures of income and racial composition, as have many previous studies of lead exposure [41][42][43][44][45][46]. Under the approach adopted for this study, area-level demographic information estimates a given household's probability of having a particular income profile and racial/ethnic identity.…”

Predictive modeling of indoor dust lead concentrations: Sources, risks, and benefits of intervention

“…Predictive modeling of indoor dust Pb concentrations in general has been sparse (Dietrich et al, 2022). A growing number of predictive models for Pb have appeared for different environmental media, such as soil (e.g., Obeng-Gyasi et al, 2021;Schwarz et al, 2013), BLLs and water infrastructure (e.g., Gibson et al, 2020;Mulhern et al, 2022), and even predictive models for BLLs based on spatial and spatiotemporal data (e.g., Potash et al, 2020). However, many predictive models are complex and require extensive datasets with multiple variables for input.…”