T RANSPORTE DE LA MODALIDAD POLIFÓNICA

First, the PIN Study is a prospective birth cohort design, which allows for establishment of temporality of the cause-effect relation. This study design allowed for prospective collection of a large body of information to support our analyses, and was not vulnerable to recall biases characteristic of cross-sectional and retrospective designs. Additionally, investigators and mothers were blinded to their exposure status throughout assessment of covariates and outcomes, which eliminates the possibility of mothers knowingly altering their responses as a result of their exposure status. However, as with any non-randomized observational study, our investigation may be vulnerable to confounding. In our statistical analyses, we sought to control for such confounding by conditioning associations on covariates identified as potential

confounders on the basis of a directed acyclic graph; however, the possibility of residual

confounding or confounding by unmeasured factors remains a possibility. Also, the prospective design and ensuing follow-up that occurred over a period of years introduced loss-to-follow-up, which may have introduced selection bias and reduced generalizability; however, we

conditioned on a variety of factors potentially related to selection bias, which would be expected to reduce such bias.

Second, the PIN Study population is a relatively socioeconomically homogenous population of women from central North Carolina. This bodes well for the internal validity of our analyses, as they will be less vulnerable to confounding by these phenomena (an ideally internally valid study population will be identical aside from exposure status); however, such homogeneity limits our generalizability to other populations that may not share the

characteristics of the PIN Study population.

Third, our exposure assessment was based on maternal urinary OPE metabolite concentrations measured during pregnancy. Urinary OPE metabolite concentrations are the current gold-standard of OPE exposure assessment, due to their ability to capture exposure from a wide range of sources and their relative objectivity. Our analysis used four metabolites, allowing us to examine exposure to a broad array of OPEs, including a novel metabolite of TCIPP (BCIPHIPP) that has never been assessed in relation to developmental outcomes. Our analysis of OPE metabolites in maternal urine used samples that were collected during the second and third trimester, a period of development in which the fetus is particularly sensitive to harm from exogenous agents. On the other hand, our exposure assessment was limited to a single spot urine collection, and may be vulnerable to misclassification due to intra-individual variability. Depending on the time period of interest, available evidence suggests that intraclass correlation coefficients for OPEs range from 0.3 to 0.6, indicating low to moderate

reproducibility. Additionally, some of the OPE metabolites we measured, specifically DPHP, may be non-specific measures of their parent compounds (i.e., TPHP), and therefore we must be cautious in attributing apparent effects of DPHP exposure to TPHP exposure. Additionally, OPE exposures appear to demonstrate seasonal trends; because we only used a single metabolite measurement to estimate exposure, such seasonal trends may have introduced exposure misclassification.

variety of cognitive and behavioral endpoints. Assessment of such a broad array of developmental endpoints allows for identification of potentially sensitive endpoints and

corroboration of results with past work. However, these instruments are imperfect measures of ambiguous qualities, and must be interpreted with care.

Our statistical analyses were supported by a large body of available covariate data and powerful statistical computing software. Our use of directed acyclic graphs to identify

appropriate adjustment sets is also novel, and allows for more logically sound analyses. Additionally, our use of the augmented product term method proposed by Buckley et al. allows estimation of sex-specific effects in an effectively stratified population (which is less vulnerable to bias than a single exposure-sex product term model), while also obtaining p-values for sex- by-exposure interactions while controlling for potential sex-by-covariate confounding. However, our sample size was modest, which resulted in reduced precision of effect estimates, potentially low statistical power, and limited useable covariate specifications, which potentially reduced our means to control for confounding. Additionally, the large number of statistical tests we

performed increased the likelihood of false-positive results; however, given this is an exploratory analysis, we do not feel compelled to “correct” for the potentially inflated Type I error [209].

In summary, the results of our study contributed to the body of evidence pertaining to cognitive and behavioral effects of early life OPE exposures. The primary strengths of our study are its prospective cohort design, utilization of relatively unbiased biomarkers of OPE exposure to estimate exposure during the sensitive prenatal period, utilization of a variety of useful covariate information, and assessment of a broad array of developmental endpoints. Our study helps characterize cognitive and behavioral effects of early life exposure to these ubiquitous compounds, and this information may contribute to decision making by regulators, industry, and consumers regarding the use of these compounds.