Predictors of pesticide levels in carpet dust collected from child care centers in Northern California, USA

The study addresses the research question of what behavioral and environmental factors predict pesticide concentrations and loadings in carpet dust within early care and education (ECE) centers. Young children spend substantial time in out-of-home care and are particularly vulnerable to chemical exposures during critical developmental windows. Prior work has documented pesticides in ECE facilities and linked early-life pesticide exposure to respiratory, neurological, and oncologic outcomes. Dust is a key exposure pathway for children, and carpet dust reflects longer-term contamination. In California, regulatory efforts (Healthy Schools Act; restrictions on agricultural applications near schoolsites) aim to reduce exposures and promote integrated pest management (IPM). The authors hypothesized that proximity to agricultural pesticide applications, onsite pesticide storage, fewer IPM practices, older building age, recent professional pesticide applications, placement of area carpets on carpeted flooring, and fewer observed pests would be associated with higher dust pesticide concentrations and loadings.

Prior studies found pesticides commonly detectable in ECE and residential environments, including chlorpyrifos, diazinon, and permethrin, with associations between early-life pesticide exposure and adverse respiratory and neurodevelopmental outcomes. Meta-analytic evidence links chronic early-life indoor insecticide exposure with increased leukemia and lymphoma risk. Age-related differences in pesticide metabolism and neurotoxic susceptibility have been modeled. U.S. national and regional ECE studies have detected multiple pesticides in dust, but quantitative predictors in ECE settings are understudied. Residential studies identify determinants such as proximity to agricultural fields, amounts applied nearby, housekeeping practices, flooring, and exterminator use. California policies (Healthy Schools Act; application restrictions near schoolsites) provide context, with reporting of professional structural applications and promotion of IPM, but their effectiveness for ECE indoor contamination requires evaluation.

Design and setting: Baseline data from the UCSF Healthy Children & Environments Study (HCES), a randomized controlled trial of an IPM intervention, were used. Fifty-one licensed ECE centers from four Northern California counties (two San Francisco Bay Area and two San Joaquin Valley) were sampled across three periods: Nov 2017–Jan 2018, Aug–Nov 2018, and Sep–Nov 2019. Centers in Valley vs. Bay Area were matched on geography, demographics, and agricultural pesticide use. IRB approval obtained; director consent provided. Data collection: Director interviews captured demographics, facility characteristics (including building age), pest problems, pesticide use and IPM practices, and cleaning routines. Two observational checklists were completed per center: the validated Integrated Pest Management (IPM) Checklist for ECE Programs (73 items; eight subscales) with additional items on doormats, flooring/carpet types, and onsite pesticide products; and the Health and Safety Checklist. Dust sampling: Using a high-volume surface sampler (HVS3), trained staff collected dust from the primary classroom carpet used for circle/nap time (area 1–2 m²), recording area sampled, weather, and GPS. Equipment was cleaned between uses to prevent cross-contamination. Samples were sieved (150 µm), the fine fraction weighed, and an aliquot (up to 1.0 g) Soxhlet-extracted with dichloromethane:hexane (1:1) for 18 h with labeled surrogates, followed by florisil cleanup and GC/MS analysis (Agilent 6890N/5973, SIM mode) for 14 analytes: bifenthrin, chlorfenapyr, chlorpyrifos, cyfluthrin, cypermethrin, dacthal, deltamethrin, diazinon, esfenvalerate, fipronil, lambda-cyhalothrin, permethrin (cis-, trans-), and piperonyl butoxide. Internal standards were chlorpyrifos-d10 and trans-permethrin-13C6. QA/QC included duplicates, matrix spikes, and solvent blanks. Concentrations (ng/g) and loadings (ng/m²) were reported; detection limits provided in supplements. Pesticide Use Report (PUR) data and geospatial exposure: DPR PUR data provided agricultural applications (2015–2019) geocoded to one-square-mile PLSS sections, and professional structural applications at schoolsites. For each center, agricultural use density (kg/km²) within 3 km during the 365 days before dust sampling was estimated using area-weighted methods. Structural applications by licensed PMPs at the center in the prior year were obtained via DPR. Statistical analysis: Descriptive statistics summarized detection frequencies, concentrations, and loadings. Further analyses focused on analytes with detection frequency >75%: bifenthrin, chlorpyrifos, fipronil, and permethrin (cis+trans summed). Values <LOD were imputed as LOD/√2; concentrations and loadings were natural-log transformed. Spearman correlations assessed associations between pesticide levels and continuous predictors. Multivariable Tobit regression models (left-censored at LOD) estimated associations of predictors with log-transformed concentrations and loadings. Predictors included: agricultural use density of the specific active ingredient within 3 km (kg/km²), DPR-reported professional structural application in the past year (binary), observation of a product containing the active ingredient onsite (binary), IPM Checklist score (Yes responses/eligible items), number of pest types observed (none, one, two or more), geographic region (San Joaquin Valley vs. Bay Area), and for loading models, sampled carpet placement (area rug on hard surface [referent], area rug on carpeted floor, carpeted base flooring). Building year was excluded from multivariable models due to collinearity with IPM score. Stata 15 was used. Regression coefficients were converted to percent change: (exp(β)−1)×100.

- Detection: All centers had at least one detectable pesticide. Most frequently detected: cis-permethrin 98%, trans-permethrin 98%, bifenthrin 94%, fipronil 94%, chlorpyrifos 88%. Diazinon was not detected. - Concentration/loadings ranges (selected): Mean concentrations: bifenthrin 1667.9 ng/g; fipronil 166.6 ng/g; cis-permethrin 1467.7 ng/g; trans-permethrin 1508.4 ng/g; chlorpyrifos 5.7 ng/g. Mean loadings: bifenthrin 4908.7 ng/m²; fipronil 983.4 ng/m²; permethrin isomers ~5000 ng/m² each; chlorpyrifos 35.4 ng/m². - PUR context: Most centers were within 3 km of agricultural applications in the prior year. Chlorpyrifos had the highest agricultural use density near centers. DPR-reported structural applications at centers included 18 active ingredients; bifenthrin comprised the largest share (36%). - Correlations (Spearman, log-transformed, LOD-imputed): • Bifenthrin dust vs. agricultural bifenthrin use within 3 km: r=0.38 (concentration), r=0.44 (loading), p<0.01. • Fipronil dust vs. number of DPR-reported fipronil PMP applications: r=0.30 for both concentration and loading, p<0.05. • IPM Checklist score vs. chlorpyrifos: r=−0.28 (concentration), p<0.05; similar negative trend for loading. - Multivariable Tobit models (percent change, 95% CI): • Region (San Joaquin Valley vs. Bay Area) predicted higher levels: bifenthrin concentration +1166% (274%, 4185%); bifenthrin loading +3457% (733%, 15086%); chlorpyrifos loading +236% (43%, 691%); fipronil loading +362% (20%, 1682%); permethrin loading +567% (112%, 2001%). • Higher IPM Checklist score associated with lower loadings: chlorpyrifos −6% per unit (−10%, −2%); permethrin −8% per unit (−14%, −1%). For chlorpyrifos concentration: −4% per unit (−7%, −1%). • Carpet placement predicted chlorpyrifos loading vs. area rug on hard surface (referent): area rug on carpeted floor −57% (−81%, −5%); carpeted base flooring −89% (−98%, −50%). • Structural applications and onsite product observation were generally not significant predictors across models, except fipronil showed positive, imprecise associations with PMP applications. - Overall, predictors more strongly related to loadings than concentrations.

Findings indicate that both environmental (geography, nearby agricultural applications) and program-level factors (IPM practices, carpet placement) influence pesticide contamination in ECE carpet dust. Location in the San Joaquin Valley was the strongest consistent predictor of higher loadings across frequently detected pesticides, reflecting regional agricultural activity and broader environmental burdens. Agricultural bifenthrin use within 3 km correlated with indoor bifenthrin in dust, aligning with residential literature on proximity and dust contamination. DPR-reported professional fipronil applications correlated with fipronil in dust, suggesting contributions from structural pest control. Higher IPM scores were associated with lower chlorpyrifos and permethrin loadings, implying that comprehensive IPM can reduce both current-use and legacy pesticide burdens indoors, potentially via better building maintenance, cleaning, and source control practices. Carpet placement influenced chlorpyrifos loadings, with area rugs on hard floors having higher loadings, possibly due to dust entrainment from hard surfaces. No consistent associations were observed for pests observed onsite, observed stored pesticide products, or facility age, suggesting these measures may be less sensitive indicators or confounded by other factors. Comparisons with earlier ECE and residential studies showed similar detection patterns, with declines in organophosphate pesticides like chlorpyrifos consistent with regulatory changes and market trends, and high frequencies of pyrethroids reflecting current usage. Given that loadings better reflect potential exposure than concentrations, the stronger associations for loadings underscore the relevance of cleaning practices, flooring configurations, and regional sources to children’s exposure risk in ECE settings.

This study demonstrates that pesticides are widespread in ECE classroom carpet dust and that both modifiable practices and non-modifiable contextual factors influence contamination. The San Joaquin Valley location and nearby agricultural use were associated with substantially higher pesticide loadings, while higher IPM implementation and certain carpet placements were linked to lower loadings. Results support promoting IPM policies and practices in ECE programs to mitigate exposures, including to legacy pesticides, and highlight the need to consider regional agricultural influences when assessing and managing risks. Future research should refine exposure models by incorporating longer temporal windows for persistent pesticides, more granular cleaning and housekeeping metrics, additional flooring and ventilation characteristics, and evaluations of policy measures such as agricultural application buffers around schoolsites.

- The cross-sectional baseline design limits causal inference and temporal resolution; persistent pesticides like chlorpyrifos may require longer exposure windows to detect associations with agricultural use. - Limited variability in certain practices (e.g., most centers had doormats and at least annual carpet deep cleaning) constrained evaluation of housekeeping determinants; detailed cleaning frequency and methods were not fully captured. - Building year was excluded from multivariable models due to collinearity with IPM score, potentially omitting an independent building-age effect. - Sample size (51 centers) limited power and precision for some predictors, leading to wide confidence intervals. - Potential exposure misclassification from PUR spatial aggregation (PLSS sections) and buffer-based estimates; structural application reporting may be incomplete for some products/uses. - Left-censoring at detection limits addressed by Tobit models, but measurement error and imputation (LOD/√2) may influence estimates.