Study population
This study is based from the HELIX project, which includes six existing population-based birth cohorts: Born in Bradford (BiB, UK)13, Étude des Déterminants pré et postnatals du développement et de la santé de l’Enfant (EDEN, France)14, Infancia y Medio Ambiente (INMA, Spain)15, Kaunas Cohort (KANC, Lithuania)16, The Norwegian Mother, Father and Child Cohort Study (MoBa, Norway)17, and Mother-Child Cohort (RHEA, Greece)18. Around 32000 mothers were recruited during pregnancy (2003-2009), from which 1301 mother-child pairs were followed-up when the child was 6-11 years old (2014-2015). Standardized protocols were used to collect biological samples and questionnaires, conduct health examinations and characterise a large range of exposures.
Health data: cardiometabolic, respiratory/allergy and mental health
Fifteen health parameters were considered for this study, covering the cardiometabolic, respiratory and mental health, as listed in Table 1 (see more details in eMethods 1). The cardiometabolic parameters considered were the child blood pressure (diastolic and systolic), the waist circumference, lipids (high-density lipoprotein (HDL) cholesterol and triglycerides) and insulin levels. The first two parameters were measured by medical staff, and the last two were obtained through blood and serum, respectively. The respiratory and allergic-related health was assessed by spirometry (Forced Expiratory Volume in one second (FEV1)) and by a questionnaire adapted from the International Study on Asthma and Allergy in Childhood (ISAAC)19 including doctor-diagnosed asthma, food allergies and eczema, as well as rhinitis symptoms3,7. The cognitive and behavioural parameters considered were the measured fluid intelligence (Raven Colour Progressive Matrix™), an index regarding symptoms of Attention Disabilities and Hyperactivity Disorders (ADHD) (Conner’s rating scales of 27 items) and an internalizing and an externalizing score (99-item Child Behaviour Checklist (CBCL)6,8.
From the whole HELIX population (n=1,301), at least one health parameter was missing for 11.5% (n=150) of children regarding cardiometabolic parameters, for 32.4% (n=294) of children regarding respiratory and allergic parameters (mostly due to FEV1), and for 0.8% (n=23) of children regarding mental parameters. Complete cases were used, leading to the inclusion of 870 mother-child pairs.
Table 1 - List of health parameters studied
Cardiometabolic health
|
Respiratory health and allergies
|
Mental health and cognition
|
- Lipids (HDL cholesterol, triglycerides)
- Blood pressure (diastolic, systolic)
- Circumference of the waist
- Insulin levels
|
- Lung function (FEV1 % pred)
- Asthma
- Food allergies
- Eczema
- Rhinitis
|
- ADHD index (Conners)
- Internalizing and externalizing indexes (CBCL)
- Test of fluid intelligence (Raven)
|
Abbreviations HDL: High Density Lipoprotein, FEV1: Forced Expiratory Volume in 1 second, ADHD: Attention Deficit Hyperactivity Disorders, CBCL: Child Behaviour Checklist.
Characterisation of the exposome
A wide range of environmental exposures was assessed in each mother-child pair, covering 21 families of exposures, with 53 prenatal and 105 postnatal exposures, as detailed in Table 2 (see also previous Helix papers12,20,21). Briefly, outdoor exposures were assessed based on remote and spatial sensing data from a geographical information system (see eMethods 2). Factors regarding the lifestyle were collected by questionnaire and included smoking habits of the mother, food intakes, the social environment (pregnancy and childhood), physical activity, sleep and the presence of pets (childhood) (see eMethods 3). Biomarkers of chemical compounds were measured through biological samples (mostly serum and urine, as detailed in eTable 1) during pregnancy and childhood (see eMethods 4).
Table 2 – List of all exposures measured and studied
Type of exposure
|
Exposures
|
Number of exposures Pregnancy
|
Number of exposures Childhood
|
|
URBAN EXPOSOME
|
|
|
Outdoor air pollution
|
NO2, PM10. PM2.5, PM2.5 absorbance
|
4
|
7
|
Indoor air pollution
|
NO2, PM2.5, PM absorbance, Benzene, TEX
|
0
|
5
|
Meteorology
|
Temperature, humidity, pressure, UV-vit D
|
3
|
3
|
Surrounding natural spaces
|
NDVI, presence of a major greenspace and bluespace
|
3
|
6
|
Built environment
|
Population density, building density, street connectivity, accessibility, facility richness, walkability, land use index
|
7
|
16
|
Road traffic
|
Traffic load on all roads and nearest road, traffic density on nearest road, inverse distance to nearest road
|
3
|
6
|
Water DBPs
|
THMs, brominated THMs, chloroform
|
3
|
0
|
Total of urban exposures for each period
|
23
|
43
|
|
LIFESTYLE
|
|
|
Tobacco
|
Active and passive smoking (pregnancy), child exposure to smoke (ETS) and parental smoking (childhood)
|
2
|
2
|
Diet
|
Food intakes (15 groups), food habits, supplements in folic acid, KIDMED score
|
3
|
20
|
Sleep
|
Average sleep
|
0
|
1
|
Physical activity
|
Moderate/vigorous activity, sedentary time
|
0
|
2
|
Allergens
|
Pet
|
0
|
1
|
Socio-economic
|
Family Affluence Score (FAS), family contact, participation in organizations, house crowding
|
1
|
3
|
Total of lifestyle exposures for each period
|
6
|
29
|
|
CHEMICAL EXPOSOME
|
|
|
Perfluoroalkyl substances (PFASs)
|
PFHxS, PFOS, PFOA, PFNA, PFUnDA
|
5
|
4
|
Brominated compounds (PBDEs)
|
PBDE 47
|
0
|
1
|
Metal and essential elements
|
Hg, Cd, Pb, Cs, Cu, Mn, Co, Mo, Se, Tl
|
1
|
10
|
Phthalates
|
MEP, MiBP, MnBP, MBzP, DEHP b, DiNP c
|
6
|
6
|
Phenols
|
MEPA, ETPA, PRPA, BUPA, BPA, OXBE, TCS
|
5
|
6
|
Organochlorine pesticides (OCs)
|
PCBa, DDE, HCB
|
3
|
3
|
Organophosphate (OP) metabolites
|
DMP, DMTP, DEP
|
3
|
2
|
Tobacco
|
Cotinine
|
1
|
1
|
Total of chemical exposures for each period
|
24
|
33
|
a Pregnancy: PCB138 + PCB153 + PCB 180. Childhood: PCB118 + PCB 138 + PCB153 + PCB170 + PCB180.
b DEHP: molar sum of MEHP, MEHHP, MEOHP and MECPP.
c DiNP: molar sum of oxo-MiNP and oh-MiNP.
BPA: Bisphenol-A, BUPA: Butyl-paraben, Cd: Cadmium, Co: cobalt, Cu: copper, DDE: Dichlorodiphenyldichloroethylene, DEHP: Di EthylhexylPhthalate, DEP: Diethyl phosphate, DMP: Dimethyl phosphate, DMTP: Dimethyl thiophosphate, ETPA: Ethyl-paraben, HCB: Hexachlorobenzene, Hg: Mercury, KIDMED: Mediterranean diet in children, MBzP: Mono benzyl phthalate, DiNP: Diisononyl phthalate, MEP: Monoethyl phthalate, MEPA: Methyl-paraben, Mo: Molybdenum, MiBP: Mono-iso butyl phthalate, NDVI: Normalized difference vegetation index, NO2: Nitrogen dioxide, OCs: organochlorine compound, OP: organophosphate pesticide, OXBE: Oxybenzone, Pb: Plomb, PBDE: Polybrominated diphenyl ether, PCB: Polychlorobiphenyls, PFASs: per- and polyfluoroalkyl substance, PFHxS: Perfluorohexane sulfonate, PFNA: Perfluorononanoate, PFOA: Perfluorooctanoate, PFOS: Perfluorooctane sulfonate, PFUNDA: Perfluoroundecanoate, PM: Particulate matter, PRPA: Propyl-paraben, Se: Selenium, Tl: Thalium, TRCS: Triclosan, UV-Vit. D, Vitamin-D dose from ultraviolet.
Covariates
Covariates used for the prenatal analyses included cohort, child age and sex, maternal age, highest parental education (primary, secondary or higher education), parental country of birth (none, one or both parents born in the cohort country), pre-pregnancy body mass index (BMI) and season of birth (winter, spring, summer or autumn). Regarding postnatal analyses, breastfeeding duration (<11 weeks, 11-35 weeks, >35 weeks) was added to the set of covariates.
Creation of the general health score
The general health score averaged three sub-scores, each representing a specific health domain (cardiometabolic, respiratory/allergy and mental health). Beforehand, continuous health parameters were transformed in z-scores, using Generalize Additive Model for Location, Scale and Shape (GAMLSS)22 to standardize on covariates (mostly age and sex, see eTables 2 and 3) and approach normality. As used previously in the Helix population, the cardiometabolic sub-score was defined as (-z waist circumference) + (- z insulin) + (z HDL cholesterol – z triglycerides)/2 + (-z systolic BP – z diastolic BP)/223,24. Following the approach of Eisenmann25 the respiratory/allergy sub-score and the mental sub-score were defined as the first principal component of a multiple factorial analysis. All of the three sub-scores were built such that a higher score means the child is in better health (see eMethods 5 and eTable 4 and 5). The three sub-scores were scaled and aggregated into a single general health score by taking their mean. By construct, the general health score is low for children with conjointly low-to-moderate cardiometabolic, respiratory/allergy and mental health in children, as well as for children highly affected in one health domain while no or moderately affected for the other two.
Strategy of analysis for the exposome-health association
For all exposures and covariates, the optimal transformation to approach normality was applied (see eTable 6). Imputation of the missing values was done, using the method of chained equations26 (see more details in eMethods 6). It generated 20 imputed datasets, used in the statistical analyses with the Rubin’s rule. After imputation, continuous exposures were centred and standardized by the interquartile range (IQR).
The exposure-general health score association study was performed separately for the prenatal and postnatal exposures using the Least Absolute Shrinkage and Selection Operator (LASSO) as the main analysis. This penalized regression model considers all exposures simultaneously and selects the best predictors of the outcome. Optimization of the penalizing parameter 27 was performed by minimizing the mean cross-validated error on each of the 20 imputed dataset. The exposures selected for at least 50% of LASSO models (10 imputed datasets out of the 20) were used as the final set of exposures28. The main models consisted in two multivariable linear regressions (one prenatal and one postnatal) considering all the selected exposures, after removing all exposures with p-value higher than 10%. More details on the strategy of analysis can be found in eMethods 7.
As secondary analyses, an exposome-wide association study (ExWAS) was conducted. It considered each exposure in separate linear regression models29 and corrected for multiple hypothesis testing (adapted from Li30). Moreover, some specific hypotheses were tested: 1) For organochlorine compound (OCs), the associations found were stratified on the terciles of the BMI because OCs are known to accumulate in fat; 2) the final multivariable models were stratified on sex to address a potential gender-specific association; 3) the final multivariable models were stratified on cohort to address the robustness of the findings to the multicentre study design. In addition, a sensitivity analysis to assess the robustness to extreme values was conducted by fitting the multivariable model after excluding the 2% lowest and 2% highest values for the general health score (n=836).
Significance level was defined as 5% for all statistical tests. Analyses were done with R version 4.2.1, using the packages mice, gamlss, FactoMineR, psych and glmnet. The main steps in the analysis are summarized in Figure 1.