Environmental exposures and fetal growth: the Haifa pregnancy cohort study

Maccabi Healthcare Services (MHS), the second largest Israeli health maintenance organization (HMO) provides primary care services to two million beneficiaries (~25% of the Israeli population). Since 1998, information on all members’ interactions with MHS has been collected, including diagnoses, visits to primary and secondary care physicians, visits to outpatient clinics, hospitalizations, laboratory tests, and prescribed and purchased medications. The HPCS cohort potentially includes ~750,000 newborns born between 1998 and 2017 to MHS members.

We will estimate daily exposures to air pollutants, ambient temperature and greenness (NDVI), using satellite based data and models. PM2.5/10 estimations will be based on novel hybrid satellite based spatiotemporal resolved models developed by us [15‐18] for predicting PM10_, PM2.5 and temperature (Fig. 1). Our PM models produce high temporal (daily) and spatial (1 square km) resolution PM2.5/10. The exposure model was rigorously validated using standard validation methods such as “ten-fold” out of sample cross validation techniques. We will estimate daily 1 km PM_2.5 concentration levels for all grid cells in the study including the challenging to model Haifa region. We will take into account the specific geo-topography by using very high resolution elevation and meteorological variables (such as wind) in model calibrations in addition to running several sensitivity analysis taking into account topography barriers. To allow even greater spatial variability at higher resolutions, we will use small-scale spatial predictors defined around each monitoring station, regressing them on the residuals of the third stage model, by use of machine learning techniques. This generates additional daily 200 m “local PM” estimates which are the difference of the estimated local pollution from the average 1 km PM2.5 concentrations at a very fine resolution (200 × 200m). We also developed similar novel satellite based (MODIS) models to estimate air temperature (Ta) at a very fine spatial resolution across Israel [17]. We developed spatiotemporally resolved models which allow us to predict three daily parameters: Ta Max (day time), 24 h mean, and Ta Min (night time) on a fine 1 km grid across the state of Israel. We used and compared both the Aqua and Terra MODIS satellites. We used linear mixed effect models, IDW (inverse distance weighted) interpolations and thin plate splines (using a smooth nonparametric function of longitude and latitude) to first calibrate between Ts and Ta in those locations where we have available data for both and used that calibration to fill in neighboring cells without surface monitors or missing Ts. Out-of-sample ten-fold cross validation (CV) was used to quantify the accuracy of our predictions. Our model performance was excellent for both days with and without available Ts observations for both Aqua and Terra (CV Aqua R2 results for min 0.966, mean 0.986, and max 0.967; CV Terra R2 results for min 0.965, mean 0.987, and max 0.968). The model allows us to generate daily min, mean and max Ta which can be reliably used with high accuracy in Epidemiology studies.

Exposure to green and natural areas around each address will be estimated using a satellite-image based vegetation index. Chlorophyll in plants strongly absorbs visible light (0.4–0.7 μm) for use in photosynthesis, while leaves strongly reflect near-infrared light (0.7–1.1 μm). The Normalized Difference Vegetation Index (NDVI) calculates the ratio of the difference between the near-infrared region and red reflectance to the sum of these two measures, and ranges from −1.0 to 1.0, with larger values indicating higher levels of vegetative density. We will then use GIS analysis to link the NDVI data to each address as a measure of exposure to “green” surroundings as commonly done previously. High-resolution satellite data can be used to accurately assess artificial LAN as it provides precise data at high resolutions (~0.6 km2). LAN data on ambient outdoor night-time illumination for the study area and corresponding years will be obtained from two sources: the radiance-calibrated satellite images of the US Defense Meteorological Satellite Program (DMSP) and the new Visible Infrared Imaging Radiometer Suite Day-Night-Band (VIIRS-DNB) program with higher resolutions than DMSP.

Our MHS database, together with additional data from the Ministry of Health includes demographic data of all members, including newborns, their mothers and fathers, data on birth weight, gestational age at delivery and congenital malformations. Also, we will account for pregnancy complications such as pre-eclampsia, gestational diabetes and hypertension and parental variables, such as maternal weight, BMI, height, age, gravidity, parity, smoking habits and socio-economic parameters as potential confounders. We will evaluate associations between growth outcomes and environmental exposures as mentioned above, for the entire population of pregnant women treated by MHS.

For the field study, we will recruit 150 pregnant women, members of MHS, with singleton pregnancies as early as possible in pregnancy, at the first visit to gynecologist by week 8 (calculated from the date of the last menstrual period). Inclusion criteria will include nonsmoking women, 18–40 years of age, who are free from chronic diseases such as diabetes and hypertension, who plan to give birth in a hospital and are residents of urban localities in the Haifa bay area and the Rehovot area located in the Central District of Israel. The Rehovot area was chosen based on lower expected exposure to air pollution originating from industrial sources while comparable in socio-demographic characteristics and access to healthcare in a population residing in urban areas of similar magnitude, with similar proportion of MHS. Exclusion criteria will include pregnancy achieved with the aid of fertility treatments, twin or multiple pregnancies and women exposed to second hand smoking at their residency. Recruitment will take place at MHS pregnancy follow-up clinics through a targeted campaign facilitated by the Maccabi clinical research unit and guided by health promotion and health and risk communication experts within our research staff.

Urine, serum and blood samples will be obtained from the 150 pregnant women three times, in each trimester of pregnancy. Samples are planned to be examined for: (1) VOCs including blood VOCs, serum aldehydes, urinary VOCs metabolites, aromatic diamines (and creatinine); (2) Elements using ICP-MS sensitive methods 3) Hydroxy polycyclic aromatic hydrocarbons using HPLC-MS/MS sensitive methods [19] and 4) cotinine using sensitive GC-MS method. In addition, the stored bio-samples will serve as infrastructure for measuring further exposures in the future, including non-targeted approaches.

Indoor air will be sampled at residencies of the 150 pregnant women living in Haifa and Rehovot at the same time biological samples will be collected. A sampler will be placed in the main activity room (other than kitchen). Sampling will include measurements of PM2.5 and nitrogen dioxide (NO2), for two consecutive days. Women will complete a questionnaire providing information about the type, age, and size of the home, as well as information about indoor sources that may impact air quality measurements, such as cigarettes, wood stoves, and candles. Participants will keep a log of open window status and heating, ventilation and air conditioning (HVAC) use during the sampling period. In addition, women will carry a personal air monitor that will sample personal exposure to PM2.5 and NO2. Participants will complete a daily time-activity questionnaire regarding time spent indoors, outdoors or in transit and additional data regarding life-style, occupational and dietary exposures. Anthropometric measurements will be taken. A research nurse and an assistant will visit the homes, while minimizing participant burden. Whenever possible, we will use similar questionnaires and bio-samples collection protocols as part of the Israeli Birth Cohorts Consortium. Harmonization will allow comparison and data pooling with other studies conducted in Israel [20], which is especially important in a small country with limited resources.

Also, we will assign ambient exposure on our well-established exposure assessment models and the geocoded location of the participants’ residence. This will yield spatiotemporally resolved robust ambient estimates.

The Helsinki committee of Assuta Medical Center approved the study protocol. For the retrospective study, the committee granted exemption from informed consent since it includes de-identified electronic medical records. For the field study, informed consent will be obtained from all women upon recruitment.

Data analysis will be conducted using IBM SPSS STATISTICS software, version 24.0 and STATA software, version 14.0. The distribution of the continuous variables will be checked by the skewness using the Cox test (coefficient of skewness divided by standard error of skewness) as well as examination of the mean–median difference and the Q-Q plot. Skewed data will be log-transformed. Chemical measures in the biomonitoring, will be described by medians, inter-quartiles range, geometric means and maximum values for analyte concentrations. Correlations between exposures will be assessed using Spearman correlations tests, overall and by sub-groups. We will compare exposures by geographical regions, first by comparing means and proportions and later by using univariate and multivariate quantile or logistic regression models. Associations between environmental exposures and fetal growth will be analyzed using advanced statistical methods, used by our group in other studies [20, 21].

We will use linear mixed-effect regression models to estimate associations between environmental exposures during different time windows and growth continuous outcomes (birth weight and head circumference), and logistic mixed-effect regression models to estimate associations with preterm birth (< 37 weeks), low birth weight (LBW) (< 2500 g) or microcephaly. Specifically, we will fit the following models:

where BW_ij/PT _ij /LBW _ij is the response (birth weight/logistic outcomes- preterm, LBW etc) for the ith subject in small statistical area j, α and u _j are the fixed and random (area specific) intercepts, respectively, PM_i, X_1i, etc. denote the set of covariates of interest used in the model, e_ij is the error term and finally, σ _u ^{2 i} is the variance of the tract random effects, and e _jj ~ N[0,σ _e ² ].

Gestational duration and prematurity will be also analyzed using a Cox proportional hazards model, with gestational age as the basic time scale and adjusted for the same factors as the anthropometric measurements. Linearity of the relationship between the outcomes under study and PM_2.5 and additional exposures will be explored using smooth functions, such as splines. We will further examine interactions between PM_2.5 and other potential predictors such as high ambient temperatures.