Accommodating detection limits of multiple exposures in environmental mixture analyses: an overview of statistical approaches.

BACKGROUND: Identifying the impact of environmental mixtures on human health is an important topic. However, such studies face challenges when exposure measurements lie below limit of detection (LOD). While various approaches for accommodating a single exposure subject to LOD have been used, their impact on mixture analysis has not been thoroughly investigated. Our study aims to understand the impact of five popular LOD accommodation approaches on mixture analysis results with multiple exposures subject to LOD, including omitting subjects with any exposures below LOD (complete case analysis); single imputations by LOD/ 2 , and by estimates from a censored accelerated failure time (AFT) model; and multiple imputation (MI) with or without truncation based on LOD. METHODS: In extensive simulation studies with high-dimensional and highly correlated exposures and a continuous health outcome, we examined the performance of each LOD approach on three mixture analysis methods: elastic net regression, weighted quantile sum regression (WQS) and Bayesian kernel machine regression (BKMR). We further analyzed data from the National Health and Nutrition Examination Survey (NHANES) on how persistent organic pollutants (POPs) influenced leukocyte telomere length (LTL). RESULTS: Complete case analysis was inefficient and could result in severe bias for some mixture methods. Imputation by LOD/ 2 showed unstable performance across mixture methods. Conventional MI was associated with consistent mild biases, which can be reduced by using a truncated distribution for imputation. Estimating censored values by AFT models had a minimal impact on the results. In the NHANES analysis, imputation by LOD/ 2 , truncated MI and censored AFT models performed similarly, with a positive overall effect of POPs on LTL while PCB126, PCB169 and furan 2,3,4,7,8-pncdf being the most important exposures. CONCLUSIONS: Our study favored using truncated MI and censored AFT models to accommodate values below LOD for the stability of downstream mixture analysis.

Prenatal exposure to air pollutant mixtures and birthweight in the upstate KIDS cohort.

BACKGROUND: Single-pollutant models have linked prenatal PM2.5 exposure to lower birthweight. However, analyzing air pollutant mixtures better captures pollutant interactions and total effects. Unfortunately, strong correlations between pollutants restrict traditional methods. OBJECTIVES: We explored the association between exposure to a mixture of air pollutants during different gestational age windows of pregnancy and birthweight. METHODS: We included 4,635 mother-infant dyads from a New York State birth cohort born 2008-2010. Air pollution data were sourced from the EPA's Community Multiscale Air Quality model and matched to the census tract centroid of each maternal home address. Birthweight and gestational age were extracted from vital records. We applied linear regression to study the association between prenatal exposure to PM2.5, PM10, NOX, SO2, and CO and birthweight during six sensitive windows. We then utilized Bayesian kernel machine regression to examine the non-linear effects and interactions within this five-pollutant mixture. Final models adjusted for maternal socio-demographics, infant characteristics, and seasonality. RESULTS: Single-pollutant linear regression models indicated that most pollutants were associated with a decrement in birthweight, specifically during the two-week window before birth. An interquartile range increase in PM2.5 exposure (IQR: 3.3 µg/m3) from the median during this window correlated with a 34 g decrement in birthweight (95 % CI: -54, -14), followed by SO2 (IQR: 2.0 ppb; ß: -31), PM10 (IQR: 4.6 µg/m3; ß: -29), CO (IQR: 60.8 ppb; ß: -27), and NOX (IQR: 7.9 ppb; ß: -26). Multi-pollutant BKMR models revealed that PM2.5, NOX, and CO exposure were negatively and non-linearly linked with birthweight. As the five-pollutant mixture increased, birthweight decreased until the median level of exposure. DISCUSSION: Prenatal exposure to air pollutants, notably PM2.5, during the final two weeks of pregnancy may negatively impact birthweight. The non-linear relationships between air pollution and birthweight highlight the importance of studying pollutant mixtures and their interactions.

Variable selection for discrete survival model with frailty in presence of left truncation and right censoring: Studying association of environmental toxicants on time-to-pregnancy.

Understanding the association between mixtures of environmental toxicants and time-to-pregnancy (TTP) is an important scientific question as sufficient evidence has emerged about the impact of individual toxicants on reproductive health and that individuals are exposed to a whole host of toxicants rather than an individual toxicant. Assessing mixtures of chemical effects on TTP poses significant statistical challenges, namely (i) TTP being a discrete survival outcome, typically subject to left truncation and right censoring, (ii) chemical exposures being strongly correlated, (iii) appropriate transformation to account for some lipid-binding chemicals, (iv) non-linear effects of some chemicals, and (v) high percentage of concentration below the limit of detection (LOD) for some chemicals. We propose a discrete frailty modeling framework (named Discnet) that allows selection of correlated covariates while appropriately addressing the methodological issues mentioned above. Discnet is shown to have better and stable false negative and false positive rates compared to alternative methods in various simulation settings. We did a detailed analysis of the pre-conception endocrine disrupting chemicals and TTP from the LIFE study and found that older females, female exposure to cotinine (smoking), DDT conferred a delay in getting pregnant, which was consistent across various approaches to account for LOD as well as non-linear associations.

Gestational age at term delivery and children's neurocognitive development.

BACKGROUND: Preterm birth is associated with lower neurocognitive performance. However, whether children's neurodevelopment improves with longer gestations within the full-term range (37-41 weeks) is unclear. Given the high rate of obstetric intervention in the USA, it is critical to determine whether long-term outcomes differ for children delivered at each week of term. METHODS: This secondary analysis included 39 199 live-born singleton children of women who were admitted to the hospital in spontaneous labour from the US Collaborative Perinatal Project (1959-76). At each week of term gestation, we evaluated development at 8 months using the Bayley Scales of Infant Development, 4 years using the Stanford-Binet IQ (SBIQ) domains and 7 years using the Wechsler Intelligence Scales for Children (WISC) and Wide-Range Achievement Tests (WRAT). RESULTS: Children's neurocognitive performance improved with each week of gestation from 37 weeks, peaking at 40 or 41 weeks. Relative to those delivered at 40 weeks, children had lower neurocognitive scores at 37 and 38 weeks for all assessments except SBIQ and WISC Performance IQ. Children delivered at 39 weeks had lower Bayley Mental (ß = -1.18; confidence interval -1.77, -0.58) and Psychomotor (ß = -1.18; confidence interval -1.90, -0.46) scores. Results were similar for within-family analyses comparing siblings, with the addition of lower WRAT scores at 39 weeks. CONCLUSIONS: The improvement in development scores across assessment periods indicates that each week up to 40 or 41 weeks of gestation is important for short- and long-term cognitive development, suggesting 40-41 weeks may be the ideal delivery window for optimal neurodevelopmental outcomes.