Study design and population.
A mother–child cohort study was set up in four Spanish areas following a common protocol to constitute the INMA [Infancia y Medio Ambiente (Environment and Childhood)] Project (INMA 2011
). Study subjects in Asturias, Gipuzkoa, Sabadell, and Valencia were recruited at the 12th week of gestation and followed until delivery. A previous mother–child hospital-based cohort in Granada also was incorporated into the INMA Project [for a detailed description of the study areas, see Supplemental Material (http://dx.doi.org/10.1289/ehp.1002425
)]. Eligibility criteria for enrollment were maternal age ≥ 16 years, singleton pregnancy, planning to deliver at the study hospitals, being able to communicate in either of the official languages, and not having followed an assisted reproduction program (Guxens et al. 2011
). The study sample was representative of the target population in terms of prenatal care attendance in the public health system (used by > 80% of the pregnant women); however, the educational level was higher than the target population average. From 45% to 98% of the eligible pregnant women agreed to participate, and enrollment periods ranged from October 2000 in Granada to February 2008 in Gipuzkoa (Guxens et al. 2011
). Numbers of recruited subjects at week 12 of gestation were 494 in Asturias, 638 in Gipuzkoa, 657 in Sabadell, and 827 in Valencia. At week 32, 485 women in Asturias, 618 in Gipuzkoa, 628 in Sabadell, and 794 in Valencia were interviewed. There were 668 eligible subjects at gestation in Granada. From the initial sample at recruitment, 485 women (98%) in Asturias, 611 (96%) in Gipuzkoa, 620 (95%) in Sabadell, 787 (95%) in Valencia, and 502 (75%) in Granada were followed until delivery and confirmed informed consent for themselves and their children to participate. The study was approved by the ethical committees of the participating centers, and all subjects gave written consent at enrollment and delivery. Only living newborns were included in the study.
Water use during pregnancy.
The interview at week 32 of gestation included questions on water use during pregnancy: source of drinking water (municipal, bottled, private well, other) at home and outside the home, use of a home water filter, changes in water ingestion since getting pregnant, and frequency and duration of showering, bathing, and swimming pool attendance (indoors and outdoors during winter and summer). Tap water ingestion was ascertained at weeks 12 and 32 using a food frequency questionnaire that queried intake of tap water and beverages made with tap water (nine categories of 250-mL glass consumption: never or < 1/month; 1–3/month; 1, 2–4, or 5–6/week; 1, 2–3, 4–5, or ≥ 6/day). A continuous variable (liters per day) was computed using the midpoint of each category. Overall, 10% of women reported that they changed the type of water used for drinking or cooking water since getting pregnant (range, 6% in Sabadell to 14% in Asturias). The mean ± SD intake was 0.44 ± 0.6 L/day in week 12 and 0.46 ± 0.67 L/day in week 32. The percentages of women reporting different categories of tap water intake between the two reporting periods ranged from 15% (Sabadell) to 54% (Gipuzkoa). Ingestion of water-based fluids (coffee, herbal drinks, and soup) and source of water for cooking were also obtained but not further used because THM levels in food are modified from levels in tap water (Huang and Batterman 2009
). In Granada, water use during pregnancy was collected retrospectively from 132 women in 2008, 6–8 years after delivery, using the same questions as for the rest of the cohort.
Chlorine was the main disinfectant used for drinking water in all the study areas. Levels of THMs were ascertained based on sampling campaigns and regulatory data from local authorities and water companies. The sampling strategy did not consider individual pregnancy periods but attempted to represent the period between the minimum and maximum conception dates of study subjects, except for the Granada cohort. Measurements were conducted at different time points: 2004–2008 (Asturias), 2006–2008 (Gipuzkoa), 2004–2006 (Granada), 2004–2006 (Sabadell), and 2004–2005 (Valencia). Sampling locations were defined to be geographically representative of the study areas, and water samples were collected from taps with no filtration or other treatments that could affect THM concentration. THMs were determined in 183 samples in Asturias (18 from our own sampling and 165 from regulatory measurements), 421 in Gipuzkoa (own sampling), 128 in Granada (79 own sampling, 49 regulatory), 198 in Sabadell (148 own sampling, 50 regulatory), and 162 in Valencia (own sampling). Water samples in swimming pools were collected from municipalities that accounted for the top 70% or more of the study population within each cohort (13 municipalities). For details on the sampling campaigns and the THM analyses, see Supplemental Material (http://dx.doi.org/10.1289/ehp.1002425
Comparison of mean THM concentrations based on regulatory surveys and our own measurements did not show significant differences (p
-value from t
-test > 0.10), and data from both sources were used. Separate models for each area were conducted to predict chloroform, bromodichloromethane, dibromochloromethane, bromoform, and total THM and to assign a concentration to the distribution system of the municipality where women resided. For the modeling procedure and tested variables and for details of the final prediction models, see Supplemental Material, and (http://dx.doi.org/10.1289/ehp.1002425
). Final models predicted average monthly THMs levels from conception until delivery in each participant’s residential water supply. Estimates of residential THM level were calculated for 455 subjects in Asturias, 592 in Gipuzkoa, 572 in Sabadell, 727 in Valencia, and 199 in Granada. Estimation of THM levels was not possible for all pregnancies followed to delivery because of missing THM data in some municipalities, missing or incomplete address, or missing gestational age.
Characteristics of the study population.
Water use during pregnancy in the study population.
The modeled residential THM level was multiplied by daily personal water use and uptake factors from the literature, to derive an estimate of daily THMs concentration in the bloodstream (Whitaker et al. 2003
), as described in Supplemental Material, (http://dx.doi.org/10.1289/ehp.1002425
). Chloroform and brominated THM were analyzed separately because toxic properties differ among species, particularly brominated versus chlorinated species. A 90% reduction in ingestion was applied if a home filter was used (Egorov et al. 2003
; Weinberg et al. 2006
). We averaged the 12- and 32-week tap water intakes to compute the ingested THM. Average THM uptake in the first, second, and third trimester and the whole pregnancy were calculated. Bathing and showering uptakes were added, and total household uptake was calculated by adding ingestion, showering, and bathing. Because of missing data in water use variables, residential THM uptake were calculated for 2,386 subjects (425 in Asturias, 576 in Gipuzkoa, 560 in Sabadell, 720 in Valencia, and 105 in Granada). To estimate swimming pool uptake, study area–specific THM averages were calculated for indoor and outdoor pools. Second, personal attendance at indoor and outdoor pools was multiplied by the area-THM average. To obtain an overall swimming pool attendance index, indoor and outdoor uptakes were added.
Estimated change in birth weight (g) expressed as β-coefficients (95% CIs) from a linear regressiona for a 10% increase in THM uptake (μg/day) in the second trimester.
Birth weight was recorded by trained midwives at delivery. Gestational age was calculated from the date of the last menstrual period (LMP) reported at recruitment and was confirmed using ultrasound examination in week 12 of gestation. If gestational age based on reported LMP and ultrasound differed by ≥ 7 days (12% of newborns), duration of gestation was recalculated using a formula based on crown–rump length from an early ultrasound measurement (Westerway et al. 2000
). Final gestational age ranged between 23.4 and 42.3 weeks. Birth weights < 10th percentile for gestational age and sex according to national growth curves (Carrascosa et al. 2004
) were defined as small for gestational age (SGA). Deliveries before 37 weeks of gestation were defined as preterm births. Birth weight < 2,500 g was defined as LBW (World Health Organization 1995
Variables potentially influencing the birth outcomes based on previous knowledge were considered. Maternal age, height, prepregnancy weight, education, marital status, parity, and country of origin and paternal weight were collected at the week 12 interview. Smoking during pregnancy was recorded at the week 32 interview. Maternal weight gain during pregnancy was computed as the rate of weight gain during the second and third trimester in kilograms per week (Rasmussen and Yaktine 2009
), adjusted for gestational age at the last available weight measure to correct for possible heteroskedasticity and nonlinearity of the rate (Dietz et al. 2006
). Maternal social class was coded from the longest-held job during the pregnancy, using the four-digit Spanish classification (Instituto Nacional de Estadística 1994), which is closely related to the International Standard Classification of Occupations (ISCO 88): Those in social class I are managers of companies with ≥ 10 employees, senior technical staff, higher-level professionals; II, managers of companies with < 10 employees, intermediate-level professionals; III, administrative and financial management supporting personnel, other self-employed professionals, supervisors of manual workers, other skilled nonmanual workers; IV, skilled and partly skilled manual workers; V, unskilled manual workers. LMP date was used as conception date.
Statistical analysis. Chloroform and brominated THM uptakes were log transformed to normalize the distribution. Because logarithm of zero values in tap water ingestion and swimming pool attendance from bottled water consumers or nonswimmers led to invalid transformed variables, these were imputed arbitrarily half the area-specific lowest value for ingestion and swimming, respectively. We evaluated the association between birth weight and log THM uptake by linear regression adjusting for gestational age and other potential confounders. Fractional polynomials were applied to identify the best transformation of gestational age in the birth weight regression models, because fetal weight gain is not constant over pregnancy. Statistically significant covariates (p-value < 0.05) and variables resulting in a ≥ 10% change in the ®-coefficient for log THM were retained in the models. Logistic regressions were used to estimate odds ratios (ORs) for dichotomous outcomes adjusting for potential confounders, and generalized additive models were used to evaluate the shape of the dose–response curve.
Coefficients from the regression models were multiplied by the logarithm of 1.1 to derive an effect estimate for a 10% increase in exposure. Analyses were stratified by region, and comparable cohorts in terms of design and water data collection (Asturias, Gipuzkoa, Sabadell, and Valencia) were combined. Previous evidence suggests that the vulnerable window for exposure could be the second trimester (Hoffman et al. 2008a
; Lewis et al. 2006
; Porter et al. 2005
) or third trimester (Hinckley et al. 2005
; Wright et al. 2004
). We used exposure in the second trimester in the main models to maximize the sample size, because some pregnancies did not reach the third trimester. We estimated the effect of exposure during the first and third trimesters and overall exposure during pregnancy in alternative models.
In sensitivity analyses, we weighted models for the Valencia and Granada cohorts based on the geographical variability, following the method suggested by Waller et al. (2001)
. Point estimates and confidence intervals (CIs) did not change substantially (results not shown). A meta-analysis was conducted to compare pooled estimates for the individual study areas with the overall analyses adjusted for cohort. A sensitivity analysis excluding observations with residuals > 3 SDs and < 3 SDs showed no differences, so all observations were included. Subjects who changed residence between weeks 12 and 32 of gestation (5% overall, 3% in Asturias, 3% in Gipuzkoa, 5% in Sabadell, 7% in Valencia) were excluded from the analyses to minimize exposure misclassification. Subjects excluded for changing residence and for missing values in covariates led to final models with smaller numbers compared with the numbers of exposure estimates.