Prenatal Metal Concentrations and Childhood Cardiometabolic Risk Using Bayesian Kernel Machine Regression to Assess Mixture and Interaction Effects

Machine Learning Within Studies of Early-Life Environmental Exposures and Child Health: Review of the Current Literature and Discussion of Next Steps

PURPOSE OF REVIEW

The goal of this article is to review the use of machine learning (ML) within studies of environmental exposures and children's health, identify common themes across studies, and provide recommendations to advance their use in research and practice.

RECENT FINDINGS

We identified 42 articles reporting upon the use of ML within studies of environmental exposures and children's health between 2017 and 2019. The common themes among the articles were analysis of mixture data, exposure prediction, disease prediction and forecasting, analysis of complex data, and causal inference. With the increasing complexity of environmental health data, we anticipate greater use of ML to address the challenges that cannot be handled by traditional analytics. In order for these methods to beneficially impact public health, the ML techniques we use need to be appropriate for our study questions, rigorously evaluated and reported in a way that can be critically assessed by the scientific community.

Prenatal Environmental Metal Exposure and Preterm Birth: A Scoping Review

Preterm birth (PTB) and its complications are the leading causes of under-five year old child deaths, accounting worldwide for an estimated one million deaths annually. The etiology of PTB is complex and multifactorial. Exposures to environmental metals or metalloids are pervasive and prenatal exposures to them are considered important in the etiology of PTB. We conducted a scoping review to determine the extent of prenatal exposures to four metals/metalloids (lead, mercury, cadmium and arsenic) and their association with PTB. We reviewed original research studies published in PubMed, Embase, the Cochrane Library, Scopus, POPLINE and the WHO regional indexes from 2000 to 2019; 36 articles were retained for full text review. We documented a higher incidence of PTB with lead and cadmium exposures. The findings for mercury and arsenic exposures were inconclusive. Metal-induced oxidative stress in the placenta, epigenetic modification, inflammation, and endocrine disruptions are the most common pathways through which heavy metals and metalloids affect placental functions leading to PTB. Most of the studies were from the high-income countries, reflecting the need for additional data from low-middle-income countries, where PTB rates are higher and prenatal exposure to metals are likely to be just as high, if not higher.

Prenatal metal mixtures and child blood pressure in the Rhea mother-child cohort in Greece

BACKGROUND

Child blood pressure (BP) is predictive of future cardiovascular risk. Prenatal exposure to metals has been associated with higher BP in childhood, but most studies have evaluated elements individually and measured BP at a single time point. We investigated impacts of prenatal metal mixture exposures on longitudinal changes in BP during childhood and elevated BP at 11 years of age.

METHODS

The current study included 176 mother-child pairs from the Rhea Study in Heraklion, Greece and focused on eight elements (antimony, arsenic, cadmium, cobalt, lead, magnesium, molybdenum, selenium) measured in maternal urine samples collected during pregnancy (median gestational age at collection: 12 weeks). BP was measured at approximately 4, 6, and 11 years of age. Covariate-adjusted Bayesian Varying Coefficient Kernel Machine Regression and Bayesian Kernel Machine Regression (BKMR) were used to evaluate metal mixture impacts on baseline and longitudinal changes in BP (from ages 4 to 11) and the development of elevated BP at age 11, respectively. BKMR results were compared using static versus percentile-based cutoffs to define elevated BP.

RESULTS

Molybdenum and lead were the mixture components most consistently associated with BP. J-shaped relationships were observed between molybdenum and both systolic and diastolic BP at age 4. Similar associations were identified for both molybdenum and lead in relation to elevated BP at age 11. For molybdenum concentrations above the inflection points (~ 40-80 μg/L), positive associations with BP at age 4 were stronger at high levels of lead. Lead was positively associated with BP measures at age 4, but only at high levels of molybdenum. Potential interactions between molybdenum and lead were also identified for BP at age 11, but were sensitive to the cutoffs used to define elevated BP.

CONCLUSIONS

Prenatal exposure to high levels of molybdenum and lead, particularly in combination, may contribute to higher BP at age 4. These early effects appear to persist throughout childhood, contributing to elevated BP in adolescence. Future studies are needed to identify the major sources of molybdenum and lead in this population.

Integrated measures of lead and manganese exposure improve estimation of their joint effects on cognition in Italian school-age children

Every day humans are exposed to mixtures of chemicals, such as lead (Pb) and manganese (Mn). An underappreciated aspect of studying the health effects of mixtures is the role that the exposure biomarker media (blood, hair, etc.) may play in estimating the effects of the mixture. Different biomarker media represent different aspects of each chemical's toxicokinetics, thus no single medium can fully capture the toxicokinetic profile for all the chemicals in a mixture. A potential solution to this problem is to combine exposure data across different media to derive integrated estimates of each chemical's internal concentration. This concept, formalized as a multi-media biomarker (MMB) has proven effective for estimating the health impacts of Pb exposure, but may also be useful to estimate mixture effects, such as the joint effects of metals like Pb and Mn, while factoring in how the association changes based upon the biomarker media. Levels of Pb and Mn were quantified in five media: blood, hair, nails, urine, and saliva in the Public Health Impact of Metals Exposure (PHIME) project, a study of Italian adolescents aged 10-14 years. MMBs were derived for both metals using weighted quantile sum (WQS) regression across the five media. Age-adjusted Wechsler Intelligence Scale for Children (WISC) IQ scores, measured at the same time as the exposure measures, were the primary outcome and models were adjusted for sex and socioeconomic status. The levels Pb and Mn were relatively low, with median blood Pb of 1.27 (IQR: 0.84) μg/dL and median blood Mn of 1.09 (IQR: 0.45) μg/dL. Quartile increases in a Pb-Mn combination predicted decreased Full Scale IQ of 1.9 points (95% CI: 0.3, 3.5) when Pb and Mn exposure levels were estimated using MMBs, while individual regressions for each metal were not associated with Full Scale IQ. Additionally, a quartile increase in the WQS index of Pb and Mn, measured using MMBs, were associated with reductions in Verbal IQ by 2.8 points (1.0, 4.5). Weights that determine the contributions of the metals to the joint effect highlighted that the contribution of the Pb-Mn was 72-28% for Full Scale IQ and 42-58% for Verbal IQ. We found that the joint effects of Pb and Mn are strongly affected by the medium used to measure exposure and that the joint effects of the Pb and Mn MMBs on cognition were the stronger than any individual biomarker. Thus, increase power and accuracy for measuring mixture effects compared to individual biomarkers. As the number of chemicals in mixtures increases, appropriate biomarker selection will become increasingly important and MMBs are a natural way to reduce bias in such analyses.

Association of Urinary and Blood Concentrations of Heavy Metals with Measures of Bone Mineral Density Loss: a Data Mining Approach with the Results from the National Health and Nutrition Examination Survey

Osteoporosis and its consequence of fragility fracture represent a major public health problem. Human exposure to heavy metals has received considerable attention over the last decades. However, little is known about the influence of co-exposure to multiple heavy metals on bone density. The present study aimed to examine the association between exposure to metals and bone mineral density (BMD) loss. Blood and urine concentrations of 20 chemical elements were selected from 3 cycles (2005-2010) NHANES (National Health and Nutrition Examination Survey), in which we included white women over 50 years of age and previously selected for BMD testing (N = 1892). The bone loss group was defined as participants having T-score < - 1.0, and the normal group was defined as participants having T-score ≥ - 1.0. We developed classification models based on support vector machines capable of determining which factors could best predict BMD loss. The model which included the five-best features-selected from the random forest were age, body mass index, urinary concentration of arsenic (As), cadmium (Cd), and tungsten (W), which have achieved high scores for accuracy (92.18%), sensitivity (90.50%), and specificity (93.35%). These data demonstrate the importance of these factors and metals to the classification since they alone were capable of generating a classification model with a high prediction of accuracy without requiring the other variables. In summary, our findings provide insight into the important, yet overlooked impact that arsenic, cadmium, and tungsten have on overall bone health.

Prenatal and Early Childhood Exposure to Lead and Repeated Measures of Metabolic Syndrome Risk Indicators From Childhood to Preadolescence

Background: Exposure to lead (Pb) during the early life stages has been associated with the development of metabolic syndrome (MetS). Longitudinal studies of Pb exposure in critical developmental windows in children are limited. Methods: Our study included 601 mother-child dyads from the PROGRESS (Programming Research in Obesity, Growth, Environment and Social Stressors) birth cohort. Blood lead levels (BLLs) were assessed during the second and third gestational trimesters, in cord blood at delivery, and at ages 1, 2, and 4 years. Bone lead levels in the patella and tibia were assessed at 1 month postpartum and evaluated in separate models. To account for cumulative exposure (prenatal, postnatal, and cumulative), we dichotomized the BLLs at each stage visit and determined the following: "higher" if a BLL was at least once above the median (HPb) and "lower" if all BLLs were below the median (LPb). We analyzed fasting glucose, HbA1c, triglycerides (TGs), total cholesterol (TC), high-density lipoprotein cholesterol (cHDL), low-density lipoprotein cholesterol (cLDL), body mass index, waist circumference (WC), body fat percentage, and systolic (SBP) and diastolic blood pressure (DBP) at two study visits between 6 and 12 years of age and created cutoff points based on the clinical guidelines for each indicator. Mixed effects models were used to analyze each outcome longitudinally for each BLL score, adjusting for child's sex, size for gestational age, child's age, maternal parity, mother's age, and socioeconomic status. Results: We observed associations for HPb exposure and TC in all stages (OR = 0.53, 95%CI = 0.32-0.86) and postnatally (OR = 0.59, 95%CI = 0.36-0.94) and for prenatal HPb and TGs (OR = 0.65, 95%CI = 0.44-0.95). HPb at all stages was associated with WC (OR = 0.27, 95%CI = 0.08-0.86), BMI (OR = 0.33, 95%CI = 0.11-0.99), SBP (OR = 0.53, 95%CI = 0.32-0.85), and DBP (OR = 0.57, 95%CI = 0.34-0.95). Pb levels in the patella were associated with cHDL (OR = 1.03, 95%CI = 1.00-1.07) and those in the tibia with TGs (OR = 0.95, 95%CI = 0.91-0.99). Conclusion: Early life exposure to Pb may alter early indicators of MetS. A follow-up of these children will allow for more definition on the impact of longer-term exposures.

The joint effect of ambient air pollution and agricultural pesticide exposures on lung function among children with asthma

BACKGROUND

Ambient environmental pollutants have been shown to adversely affect respiratory health in susceptible populations. However, the role of simultaneous exposure to multiple diverse environmental pollutants is poorly understood.

OBJECTIVE

We applied a multidomain, multipollutant approach to assess the association between pediatric lung function measures and selected ambient air pollutants and pesticides.

METHODS

Using data from the US EPA and California Pesticide Use Registry, we reconstructed three months prior exposure to ambient air pollutants ((ozone (O₃), nitrogen dioxide (NO₂), particulate matter with a median aerodynamic diameter < 2.5 μm (PM_2.5) and <10 μm (PM₁₀)) and pesticides (organophosphates (OP), carbamates (C) and methyl bromide (MeBr)) for 153 children with mild intermittent or mild persistent asthma from the San Joaquin Valley of California, USA. We implemented Bayesian kernel machine regression (BKMR) to estimate the association between simultaneous exposures to air pollutants and pesticides and lung function measures (forced expiratory volume in 1 s (FEV₁), forced vital capacity (FVC), and forced expiratory flow between 25% and 75% of vital capacity (FEF_25-75)).

RESULTS

In BKMR analysis, the overall effect of mixtures (pollutants and pesticides) was associated with reduced FEV₁ and FVC, particularly when all the environmental exposures were above their 60th percentile. For example, the effect of the overall mixture at the 70th percentile (compared to the median) was a -0.12SD (-50 mL, 95% CI: -180 mL, 90 mL) change in the FEV₁ and a -0.18SD (-90 mL, 95% CI: -240 mL, 60 mL) change in the FVC. However, 95% credible intervals around all of the joint effect estimates contained the null value.

CONCLUSION

At this agricultural-urban interface, we observed results from multipollutant analyses, suggestive of adverse effects on some pediatric lung function measures following a cumulative increase in ambient air pollutants and agricultural pesticides. Given the uncertainty in effect estimates, this approach should be explored in larger studies.

A Quantile-Based g-Computation Approach to Addressing the Effects of Exposure Mixtures

BACKGROUND

Exposure mixtures frequently occur in data across many domains, particularly in the fields of environmental and nutritional epidemiology. Various strategies have arisen to answer questions about exposure mixtures, including methods such as weighted quantile sum (WQS) regression that estimate a joint effect of the mixture components.

OBJECTIVES

We demonstrate a new approach to estimating the joint effects of a mixture: quantile g-computation. This approach combines the inferential simplicity of WQS regression with the flexibility of g-computation, a method of causal effect estimation. We use simulations to examine whether quantile g-computation and WQS regression can accurately and precisely estimate the effects of mixtures in a variety of common scenarios.

METHODS

We examine the bias, confidence interval (CI) coverage, and bias-variance tradeoff of quantile g-computation and WQS regression and how these quantities are impacted by the presence of noncausal exposures, exposure correlation, unmeasured confounding, and nonlinearity of exposure effects.

RESULTS

Quantile g-computation, unlike WQS regression, allows inference on mixture effects that is unbiased with appropriate CI coverage at sample sizes typically encountered in epidemiologic studies and when the assumptions of WQS regression are not met. Further, WQS regression can magnify bias from unmeasured confounding that might occur if important components of the mixture are omitted from the analysis.

DISCUSSION

Unlike inferential approaches that examine the effects of individual exposures while holding other exposures constant, methods like quantile g-computation that can estimate the effect of a mixture are essential for understanding the effects of potential public health actions that act on exposure sources. Our approach may serve to help bridge gaps between epidemiologic analysis and interventions such as regulations on industrial emissions or mining processes, dietary changes, or consumer behavioral changes that act on multiple exposures simultaneously. https://doi.org/10.1289/EHP5838.

Environmental mixtures and children's health: identifying appropriate statistical approaches

PURPOSE OF REVIEW

Biomonitoring studies have shown that children are constantly exposed to complex patterns of chemical and nonchemical exposures. Here, we briefly summarize the rationale for studying multiple exposures, also called mixture, in relation to child health and key statistical approaches that can be used. We discuss advantages over traditional methods, limitations and appropriateness of the context.

RECENT FINDINGS

New approaches allow pediatric researchers to answer increasingly complex questions related to environmental mixtures. We present methods to identify the most relevant exposures among a high-multitude of variables, via shrinkage and variable selection techniques, and identify the overall mixture effect, via Weighted Quantile Sum and Bayesian Kernel Machine regressions. We then describe novel extensions that handle high-dimensional exposure data and allow identification of critical exposure windows.

SUMMARY

Recent advances in statistics and machine learning enable researchers to identify important mixture components, estimate joint mixture effects and pinpoint critical windows of exposure. Despite many advantages over single chemical approaches, measurement error and biases may be amplified in mixtures research, requiring careful study planning and design. Future research requires increased collaboration between epidemiologists, statisticians and data scientists, and further integration with causal inference methods.

Serum Microelements in Early Pregnancy and their Risk of Large-for-Gestational Age Birth Weight

Excessive birth weight has serious perinatal consequences, and it “programs” long-term health. Mother’s nutritional status can be an important element in fetal “programming”; microelements such as selenium (Se), zinc (Zn), copper (Cu), and iron (Fe) are involved in many metabolic processes. However, there are no studies assessing the relationship of the microelements in the peri-conceptual period with the risk of excessive birth weight. We performed a nested case control study of serum microelements’ levels in the 10–14th week of pregnancy and assessed the risk of large-for-gestational age (LGA) newborns using the data from a prospective cohort of pregnant women recruited in 2015–2016 in Poznań, Poland. Mothers delivering LGA newborns (n = 66) were examined with matched mothers delivering appropriate-for-gestational age (AGA) newborns (n = 264). Microelements’ levels were quantified using mass spectrometry. The odds ratios of LGA (and 95% confidence intervals) were calculated by multivariate logistic regression. In the whole group, women with the lowest quartile of Se had a 3 times higher LGA risk compared with women in the highest Se quartile (AOR = 3.00; p = 0.013). Importantly, the result was sustained in the subgroup of women with the normal pre-pregnancy BMI (AOR = 4.79; p = 0.033) and in women with a male fetus (AOR = 6.28; p = 0.004), but it was not sustained in women with a female fetus. There were no statistical associations between Zn, Cu, and Fe levels and LGA. Our study provides some preliminary evidence for the relationships between lower serum Se levels in early pregnancy and a higher risk of large-for-gestational age birth weight. Appropriate Se intake in the periconceptual period may be important for optimal fetal growth.

Associations of plasma metal concentrations with the decline in kidney function: A longitudinal study of Chinese adults

Metals are widespread pollutants in the environment which have been reported to be associated with kidney dysfunction in many existing epidemiological studies. However, most of the studies are cross-sectional design and mainly focus on several toxic metals including arsenic, lead and cadmium. Therefore, we conducted this prospective study within the Dongfeng-Tongji cohort to evaluate the associations of plasma multiple metals with the decline in kidney function among Chinese middle-aged and elderly. In total, 1434 participants free of chronic diseases at baseline were included in analysis. We measured baseline plasma concentrations of 23 metals and calculated estimated glomerular filtration rate (eGFR) using the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation based on serum creatinine, age, sex and ethnicity. Bonferroni correction was used for multiple testing to reduce the probability of a type I error. Principal component analysis was conducted to evaluate the combined effect of multiple metal co-exposure. Most of the plasma metal concentrations were within the literature reported reference values, whereas the concentration of lead and nickel exceeded the guideline value. We found that plasma concentrations of aluminum, arsenic, barium, lead, molybdenum, rubidium, strontium, vanadium and zinc were significantly associated with the decline in kidney function measured by annual eGFR decline, rapid renal function decline (defined as an annual decline in eGFR ≥ 5 mL/min/1.73 m²) or incident eGFR < 60 mL/min/1.73 m², with the adjusted beta coefficients (95% CI) for annual eGFR decline 0.50 (0.30, 0.69), 0.98 (0.74, 1.23), 0.56 (0.32, 0.79), 0.21 (0.03, 0.39), 0.35 (0.16, 0.54), 0.94 (0.71, 1.17), 0.37 (0.15, 0.60), 0.78 (0.54, 1.02), and 0.74 (0.57, 0.91), respectively. The metals exposures were linked with increased risks of impaired kidney function. Associations of principal components representing these metals with the decline in kidney function were significant and suggest a possible additional health risk by co-exposure. Participants engaged in manufacturing had higher plasma levels of several metals compared with those who had been involved in management- or administration-related work. Our findings suggest that exposure to multiple metals contribute to the decline in kidney function among the middle-aged and elderly. Co-exposure to multiple metals may have synergetic effect on the kidney function. Further studies are warranted to confirm our findings and clarify the potential mechanisms.

Associations of a mixture of urinary phthalate metabolites with blood lipid traits: A repeated-measures pilot study

Evidence is available about the associations of phthalates or their metabolites with blood lipids, however, the mixture effects of multiple phthalate metabolites on blood lipid traits remain largely unknown. In this pilot study, 106 individuals at three age groups of <18, 18- and ≥60 years were recruited from the residents (n = 1240) who were randomly selected from two communities in Wuhan city, China. The participants completed the questionnaire survey and physical examination as well as provided urine samples in the winter of 2014 and the summer of 2015. We measured urinary levels of nine phthalate metabolites using a high-performance liquid chromatography-tandem mass spectrometry. We estimated the associations of individual phthalate metabolite with blood lipid traits by linear mixed effect (LME) models, and assessed the overall association of the mixture of nine phthalate metabolites with blood lipid traits using Bayesian kernel machine regression (BKMR) models. LME models revealed the negative association of urinary mono-2-ethylhexyl phthalate (MEHP) with total cholesterol (TC) as well as of urinary mono-benzyl phthalate or urinary MEHP with low density lipoprotein cholesterol (LDL-C). BKMR models revealed the negative overall association of the mixture of nine phthalate metabolites with TC or LDL-C, and DEHP metabolites (especially MEHP) had a greater contribution to TC or LDL-C levels than non-DEHP metabolites. The findings indicated the negative overall association of the mixture of nine phthalate metabolites with TC or LDL-C. Among nine phthalate metabolites, MEHP was the most important component for the changes of TC or LDL-C levels, implying that phthalates exposure may disrupt lipid metabolism in the body.

Prenatal Lead Exposure, Type 2 Diabetes, and Cardiometabolic Risk Factors in Mexican Children at Age 10-18 Years

CONTEXT

Several cross-sectional studies have assessed the association of lead exposure with type 2 diabetes and cardiometabolic risk factors in adults; however, studies of such associations in childhood are rare.

OBJECTIVE

We assessed the prospective associations of prenatal exposure to lead with type 2 diabetes and cardiometabolic risk factors in children.

DESIGN

The Early Life Exposure in Mexico to Environmental Toxicants is a birth cohort study of pregnant women and their offspring.

SETTING

Public hospitals in Mexico City.

PATIENTS OR OTHER PARTICIPANTS

Women were recruited during pregnancy; their offspring were recruited for a follow-up visit at age 10 to 18 years (n = 369).

MAIN OUTCOME MEASURES

We measured fasting serum markers of type 2 diabetes and cardiometabolic risk factors in children, including fasting glucose, insulin, and lipids. The index of insulin resistance was calculated.

RESULTS

The geometric mean of maternal blood lead levels (BLLs) during pregnancy was 4.3 µg/dL (95% confidence interval [CI]): 4.0-4.6 µg/dL) in the entire sample. In boys, those with maternal BLLs ≥ 5 µg/dL (compared with those with BLLs < 5 µg/dL) had significantly lower z scores for total cholesterol (β = -0.41, 95% CI: -0.71, -0.12), high-density lipoprotein cholesterol (β = -0.32, 95% CI: -0.59, -0.05), and low-density lipoprotein cholesterol (β = -0.52, 95% CI: -0.81, -0.22), adjusting for covariates. No associations were detected in girls.

CONCLUSIONS

In our study, we found that higher prenatal exposure to lead was associated with lower levels of cholesterol in children following a sex-specific pattern. Further studies with a larger sample size that examine whether sex is a potential modifier are needed to confirm our findings.

Exposure to 17 trace metals in pregnancy and associations with urinary oxidative stress biomarkers

BACKGROUND

Exposure to some toxic metals, such as lead and cadmium, has been associated with increased oxidative stress. However less is known about other metals and metal mixtures, especially in pregnant women who are a vulnerable population.

METHODS

To study the relationship between exposure to trace metals and oxidative stress, we analyzed a panel of 17 metals and two oxidative stress biomarkers (8-isoprostane and 8-hydroxydeoxyguanosine [8-OHdG]) in urine samples collected at ~26 weeks gestation from pregnant women in Boston (n = 380). We used linear regression models to calculate percent differences and 95% confidence intervals (CI) in oxidative stress markers for an interquartile range (IQR) increase in each urinary metal with adjustment for other metals. In addition, we applied principal components analysis (PCA) and Bayesian kernel machine regression (BKMR), to examine cumulative effects (within correlated groups of exposures as well as overall) and interactions.

RESULTS

We estimated 109% (95% CI: 47, 198) higher 8-isoprostane and 71% (95% CI: 45, 102) higher 8-OHdG with an IQR increase in urinary selenium (Se). We also estimated higher 8-isoprostane (47%, 95% CI: 20.5, 79.4) and 8-OHdG (15.3%, 95% CI: 5.09, 26.5) in association with urinary copper (Cu). In our PCA, we observed higher 8-isoprostane levels in association with the "essential" PC (highly loaded by Cu, Se, and Zinc). In BKMR analyses, we also estimated higher levels of both oxidative stress biomarkers with increasing Se and Cu as well as increasing levels of both oxidative stress biomarkers in association with cumulative concentrations of urinary trace metals.

CONCLUSION

We observed higher 8-isoprostane and 8-OHdG levels in association with urinary trace metals and elements, particularly Se and Cu, in linear models and using mixtures approaches. Additionally, increasing cumulative exposure to urinary trace metals was associated with higher levels of both oxidative stress biomarkers. The beneficial effects of these compounds should be carefully questioned.

Exploring the associations of serum concentrations of PCBs, PCDDs, and PCDFs with walking speed in the U.S. general population: Beyond standard linear models

Studies have shown that persistent organic pollutants (POPs) can have various health effects. However, little is known about the effects of multiple chemicals with possible common sources of exposure on walking speed, a proxy index reflecting lower limb neuromuscular function and physical function. We simultaneously applied multiple linear and nonlinear statistical models to explore the complex exposure-response relationship between a mixture of 22 selected POPs and walking speed. A total of 14 polychlorinated biphenyls (PCBs), 3 polychlorinated dibenzo-p-dioxins (PCDDs), and 5 polychlorinated dibenzofurans (PCDFs) were measured in the serum of participants in the National Health and Nutrition Examination Survey (NHANES) from 1999 to 2002. Walking speed was measured during a physical examination. Linear regression (LR), least absolute shrinkage and selection operator (LASSO), and group LASSO were used to evaluate the linearity of mixtures, while restricted cubic spline (RCS) regression, random forest (RF), and Bayesian kernel machine regression (BKMR) models were used to evaluate the nonlinearity of mixtures. Potential confounders were adjusted in the above models. A total of 436 subjects were included in our final analysis. The results of the LR model did not identify any POP exposure that was significantly associated with walking speed. The LASSO results revealed an inverse association of one PCDD congener and two PCDF congeners with walking speed, while the group LASSO analysis identified PCDFs at the exposure level and at the group level. In the RCS analysis, two PCB congeners presented significant overall associations with walking speed. The PCB congener PCB194 showed statistically significant effects on the outcome (P = 0.01) when a permutation-based RF was used. The BKMR analysis suggested that PCBs and PCDFs (probabilities = 0.887 and 0.909, respectively) are potentially associated with walking speed. Complex statistical models, such as RCS regression, RF and BKMR models, can detect the nonlinear and nonadditive relationships between PCBs and walking speed, while LASSO and group LASSO can identify only the linear relationships between PCDFs and walking speed. Fully considering the influence of collinearity in each method during modelling can increase the comprehensiveness and reliability of conclusions in studies of multiple chemicals.