Advertisement

Environmental Science and Pollution Research

, Volume 26, Issue 4, pp 3578–3592 | Cite as

Spatial identification of environmental health hazards potentially associated with adverse birth outcomes

  • Alina Svechkina
  • Boris A. PortnovEmail author
Research Article
  • 167 Downloads

Abstract

Reduced birth weight (RBW) and reduced head circumference (RHC) are adverse birth outcomes (ABOs), often linked to environmental exposures. However, spatial identification of specific health hazards, associated with these ABOs, is not always straightforward due to presence of multiple health hazards and sources of air pollution in urban areas. In this study, we test a novel empirical approach to the spatial identification of environmental health hazards potentially associated with the observed RHC and RBW patterns. The proposed approach is implemented as a systematic search, according to which alternative candidate locations are ranked based on the strength of association with the observed birth outcome patterns. For empirical validation, we apply this approach to the Haifa Bay Area (HBA) in Israel, which is characterized by multiple health hazards and numerous sources of air pollution. We identified a spot in the local industrial zone as the main risk source associated with the observed RHC and RBW patterns. Multivariate regressions, controlling for personal, neighborhood, and geographic factors, revealed that the relative risks of RHC and RBW tend to decline, other things being equal, as a function of distance from the identified industrial spot. We recommend the proposed identification approach as a preliminary risk assessment tool for environmental health studies, in which detailed information on specific sources of air pollution and air pollution dispersion patterns is unavailable due to limited reporting or insufficient monitoring.

Keywords

Adverse birth outcomes (ABOs) Reduced birth weight (RBW) Reduced head circumference (RHC) Air pollution Distance gradient method (DGM) Wind adjustment Environmental hazards Haifa Bay Area (HBA) Israel 

Introduction

Reduced birth weight (RBW) and reduced head circumference (RHC) are leading causes of infant mortality, considered to be responsible for about 30% of all neonatal deaths and postnatal deaths under the age of five (WHO 2012, 2016; UNICEF 2016; Isayama et al. 2012; Lau et al. 2013; Lundgren et al. 2014; Bhutta et al. 2002; Maisonet et al. 2004).

RBW and RHC are frequently linked with such factors as multiple pregnancy, inherited medical conditions, poor nutrition, maternal smoking and alcohol use, and exposure to environmental chemicals (Blencowe et al. 2012; Morken et al. 2012; Heaman et al. 2013; Pedersen et al. 2016; Eliyahu et al. 2002; Chiavarini et al. 2012; Wen et al. 2004).

The relationship between air pollution and Adverse birth outcome (ABO) incidence is fairly well established (Bell et al. 2007; Morello-Frosch et al. 2010; Llop et al. 2010; Ballester et al. 2010; Ferguson et al. 2013; Backes et al. 2013; Stacy et al. 2015). In addition to direct areal exposure, people may also come in contact with contaminants after they are deposited in soils or accumulated in other media, such as groundwater or food (Khan et al. 2008; Liu et al. 2013; Mahmood and Malik 2014; Li et al. 2014). However, linking the observed ABO patterns with specific environmental health hazards is not always straightforward.

Several methods are used in empirical studies to identify potential environmental health hazards in urban areas. The first group of methods is based on measuring harmful substances at the receptor sites and comparing them with emissions from different pollution sources (Cooper and Watson 1980; Stohl 1995; Hopke 2003; Salvador et al. 2004; Begum et al. 2004; Kim and Hopke 2004; Xie et al. 2006; Cesari et al. 2014; Banerjee et al. 2015; Zhang et al. 2015). However, this identification approach requires a considerable amount of information and field measurements, which are not always available to researchers (Cooper and Watson 1980; Xie et al. 2006; Wang et al. 2013; Cesari et al. 2014; Zhang et al. 2015).

An alternative approach to the spatial identification of potential environmental hazards in urban areas is based on the distance gradient method (DGM) that uses aerial proximities from potential air pollution sources, such as heavy roads and chemical and power plants, as proxies for unknown exposures. If the observed disease rates decline as the distance from a particular hazard increase, a conclusion is reached that proximity to the hazard in question is associated with morbidity in surrounding areas (see inter alia Moore and Carpenter 1999; McKenzie et al. 2012; Sermage-Faure et al. 2012; Paz et al. 2009; Zusman et al. 2012).

In the rest of this paper, we extend the DGM approach to the spatial identification of a priori unknown health hazards associated with ABOs. The main motivation for the proposed approach is that several potential sources of exposure may exist simultaneously in urban areas, some known and some unknown, and the strength of association between each of them and the observed health outcomes may not always be straightforward (Svechkina et al. 2017; Svechkina and Portnov 2017).

Materials and methods

The study area

The Haifa Bay Area (HBA), data for which we use in the present study, consists of the city of Haifa and its suburbs—Krayot, Tirat Carmel, Nesher, and Kiryat Tivon (see Fig. 1). The metropolitan area is located on the Mediterranean coast and is the third largest metropolitan area in Israel after Tel Aviv and Jerusalem. Its total population is about 600,000 residents, 81% of whom are Jews, 11% are Arabs, and 8% are other ethnic minorities (ICBS 2016).
Fig. 1

The map of the study area, showing the location of main industrial facilities (1–5), thoroughfare roads, and air quality monitoring stations (AQMSs). AQMSs, wind roses for which are used in the analysis, are marked by red circles

HBA is a major center of heavy industry, including pharmaceutical and chemical processing plants, oil refineries, and a small power plant (see Appendix Table 3). A seaport and a local airport are also located in the metropolitan area’s proper. Because of industrial activity and heavy traffic, large concentrations of emissions are released into the air, potentially linked to elevated morbidity (MH 2014).

At present, the network of air quality monitoring in HBA includes 18 stationary stations (IMEP 2016). Most of these stations monitor gaseous substances and particulates (SO2, NOx, O3, PM10, PM2.5), and only 5, recently established, stations monitor the levels of volatile organic compounds (VOCs). The uneven geographic distribution of the monitoring stations and differences in the list of air pollutants each of them measures reduce possibilities for accurate exposure assessment, thus necessitating the use of alternative risk assessment methods, such as that discussed in this paper.

Birth data sources

Medical literature uses two different approaches for assessing ABOs: dichotomous (that is, normal vs. RBW/RHC) classifications (Bell et al. 2007; Llop et al. 2010; Stacy et al. 2015; Hannam et al. 2014) and z-scores or other continuous ABO measurements (Morello-Frosch et al. 2010; Ballester et al. 2010; Ferguson et al. 2013; Stacy et al. 2015; Dadvand et al. 2013; Fleischer et al. 2014). In our study, we adopted the later approach, which is widely used in medical literature (ibid.).

We obtained data for the present study from the Mother and Child health care clinic’s (MCHC) computerized database. The database was established by the Israel Ministry of Health in early 2000, underwent several revisions, and has become operational in 2014. From July 2014 on, each child born in the study area is registered in the MCHC database within 2 weeks after the birth. The database contains information on children’s physical development after the birth, vaccination schedule, implementation of different screening programs, etc.

For the year 2015 (that is, the latest year for which validated records are available), the database contained 8487 records of the newborns in HBA without congenital disorders, out of 9093 children born in HBA in 2015 overall, thus indicating more than 93% of the dataset completeness. All records in the database are based on measurements implemented by the MCHC nurses performed according to the international standards for measuring newborns’ weight, length, and head circumference (IHME 2006), which are mandatory in Israel since 2006.

We excluded from the analysis non-singleton births (407 cases) and all the newborns, whose birth measurements were missing and those who were measured for the first time in the MCHC clinic more than 14 days after the birth (418 records). We also excluded from the analysis obvious outliers and erroneous records, such as birth weight records of less than 500 g and more than 5500 g, children with head circumference at birth of less than 20 cm and more than 40 cm, and all the deliveries with recorded gestation age of less than 24 weeks and more than 44 weeks (166 cases).

In addition to individual records of the newborns (such as gender and birth weight), we also retrieved from the database street addresses of the families and several maternal characteristics, including mothers’ age, educational level, country of birth, nationality, welfare support (if any), and the total number of siblings in the family.

The total number of cases available for the analysis after these subtractions was 7460. Out of this number, some 5638 records (75.2%) contained accurate street addresses and could be geocoded.1 In addition, some 1137 cases (15.1%) were geocoded using 5-digit zip codes of the place of residence and the rest of the newborns (721 cases (9.7%)) were positioned on the map using coordinates of the MCHC in which they were registered. A statistical comparison between all the records in the database and records with geocoded addresses detected no significant differences in terms of birth weight and head circumferences of the newborns (p > 0.2).

In addition to street addresses of the families, the newborn’s head circumference and birth weight were retrieved from the database. We also retrieved several maternal attributes (such as age, education, country of birth, nationality, and the number of children in the family) and several neonatal characteristics (gender, gestational age, and the month of birth).

Identification methodology

Assuming that a specific indicator of prenatal development (PD), IPD (such as RBW or RHC), observed in the ith point of space (\( {I}_{{\mathrm{PD}}_i} \)), depends on the distance from a potential health hazard, j; the relationship between\( {I}_{{\mathrm{PD}}_i} \) and distji can be expressed as a linear function, reflecting a monotonic decline in \( {I}_{{\mathrm{PD}}_i} \)as a function of distji, adjusted for potential confounders:
$$ {I}_{{\mathrm{PD}}_i}={b}_0+{b}_1\bullet \overset{\sim }{dist_{ji}}+{b}_2\bullet BIO+{b}_3\bullet \mathrm{BAG}+{b}_4\bullet \mathrm{SEC}+{b}_5\bullet \mathrm{GEO}+{\varepsilon}_i $$
(1)

where \( \overset{\sim }{{\mathrm{dist}}_{ji}}= \)the distance between j and i, adjusted by wind frequency from point j to point i (see Appendix 2); BIO = the vector of birth characteristics, observed at the ith point of space, including gender, month of birth, and gestation age of the new born; BAG = the vector of maternal characteristics, including mother’s age, ethnicity, and the number of previous births; SEC = the vector of maternal socioeconomic characteristics, including residential conditions, education, and income; GEO = the vector of geographic attributes of i, including topography and proximities to different environmental features, such as seashore, and main roads; and εi is the random error term.

As long as specific sources of exposure (e.g., roads and industrial facilities) are a priori known, the calculation of the strength of association between \( {I}_{{\mathrm{PD}}_i} \)and proximity to the specific source of exposure can be measured as a standardized coefficient of regression (beta), or the coefficient of determination, \( {R}_{ji}^2 \).

However, if actual sources of exposure associated with \( {I}_{{\mathrm{PD}}_i} \) are unknown, alternative locations, js, can be assessed, one by one, as potential exposure sources and then ranked by their “probability” to be sources of exposure (Pji):
$$ {P}_{ji}\to {R}_{ji}^2\left(\ \overset{\sim }{dist_{ji}},\mathrm{BIO},\mathrm{BAG},\mathrm{SEC},\mathrm{GEO}\right),\forall {b}_1<0. $$
(2)

In this case, the interpretation of (2) is as follows: values of \( {R}_{ji}^2 \) close to 1 (when bi is negative) indicate a high “probability” that exposure originating from point j is associated with morbidity observed in i, while values of \( {R}_{ji}^2 \) close to 1 (when b1 is positive) indicate a negative effect with distance, and values of \( {R}_{ji}^2 \) close to zero will point out at the absence of any significant association between \( {I}_{\mathrm{P}{\mathrm{D}}_i} \)and \( \overset{\sim }{\mathrm{dist}} \)ji (Svechkina et al. 2017; Svechkina and Portnov 2017).

Research variables and data analysis

We investigated two indicators of prenatal development: birth weight (grams) and head circumference (cm). Preterm births (PTBs), defined by the duration of pregnancy of less than 37 weeks (365 cases or about 5%), were represented in the analysis by a dichotomous variable, to separate preterm and normal-term cases.

We performed analysis in several phases. First, we calculated wind-weighted proximities of maternal residences to potential sources of exposure (see “Geo-statistical analysis” section) and then used these distances as the main explanatory variable in the multivariate regression analysis. In addition to proximities, our analysis covered several explanatory (control) variables, including birth characteristics (gender, month of birth, gestation age); maternal background (country of birth, nationality, age, number of previous births); socioeconomic status of the residential neighborhood (population density and average income); geographic location of the residence (the side of the Carmel Mountain, either facing the industries or the Mediterranean, and elevation above the sea level); and tobacco smoking rates in the neighborhood, calculated using smoking prevalence rates in 2008, reported in Baron-Epel et al. (2010) and obtained using representative telephone surveys.

As well established, there is a strong relationship between birth weight and maternal socioeconomic status (Spencer et al. 1999; Bacci et al. 2016). In the absence of family-specific measures in the database, we assigned the values available for small census areas (SCAs) to all residential locations in the same neighborhood, using several variables to capture this effect: ownership structure (percent of apartment owners) and socioeconomic status of small census areas (SCAs). To control for ethnicity, known to influence birth outcomes (Urquia et al. 2010), we also included in the analysis the maternal country of birth (Israel/not Israel) and nationality (Jewish vs. non-Jewish). Maternal age and the month of delivery were also added to the list of explanatory variables, as factors known to influence birth outcomes (Luo et al. 2006; Sorbye et al. 2016; Hannam et al. 2014). Descriptive statistics of the research variables are reported in Appendix Table 1.

Geo-statistical analysis

At the next step, we generated a layer of 1000 evenly distributed points representing the locations of potential environmental hazards (j or “source” points) and calculated Euclidian distances (distji), from each residential location (i) to each “source” point (j). Because simple Euclidian distances may not be an accurate proximity matrix, considering wind frequency and direction, we adjusted these distances by applying the wind frequency transformation described in Appendix 2. For wind adjustments, we used two wind roses, generated out of data, obtained from the “Igud Arim” and “Kiryat Tiv’on” air quality monitoring stations, located inside and southeast of the industrial zone, respectively (see Fig. 1). The first wind rose (Fig. 2a) was used for model estimation, whereas the second one (Fig. 2b) was used for a model sensitivity test, as further detailed in the “Results” section.
Fig. 2

Wind roses for the “Igud Arim” (a) and “Kyrat Tiv’on” (b) AQMSs. See Fig. 1 for the AQMSs’ locations

After the wind-adjusted distances were calculated, we introduced them (jointly with geographic, maternal background, and socioeconomic confounders) into multivariate regression models as explanatory variables for the observed indicators of prenatal development.

Next, we ran separate regressions for each “source” point. Observations in these models were represented by residential locations of the newborns (7460 locations), factored in by wind-weighted distances to a given “source” point, and by other attributes of residential locations, including elevation above the sea level and seashore proximity, as well as by maternal characteristics of the newborn and socioeconomic confounders. This way, for each of the two indicators of prenatal development covered by the study—that is, head circumference and newborns’ birth weight—1000 separate regressions were calculated, one regression model for each potential “risk source” point. Based on these regressions, each “source” point was assigned R2 of the regression model estimated for it. Since some 2000 separate regression models were estimated, we report in the paper, for brevity’s sake, only the best-performing models for each type of the ABOs studied (see “Results” section).

Next, we interpolated the coefficients of determination, to create continuous “probability” surfaces, differentiating between areas with high and low R2 values. To this end, we used the kriging interpolation, performed in the ArcGISTM 10.X Software (ESRI 2015). Kriging is an interpolation method, based on statistical models that include autocorrelation. As a result, kriging not only has a capability of producing a prediction surface but also provides some measures of the certainty or accuracy of the predictions (ESRI 2015). Concurrently, to visualize the general spatial patterns of head circumference and birth weight of the newborns, kernel interpolation with barriers was used. This interpolation technique helps to account for barriers within the settlement area, such as boundaries of residential neighborhoods, and does not generate interpolation estimates for unpopulated areas, such as industrial zones and open areas.

As well established in the literature, air pollution from industrial sources with high smokestacks moves with a wind to more remote areas and may have a maximum impact beyond a certain distance from the pollution source (Seinfeld and Pandis 2006). To account for this phenomenon, we investigated two possible functional forms of relationships between adverse birth outcomes and health hazards’ proximity: a monotonic linear function for close-to-the-ground sources of exposure (such as thoroughfare roads and oil storages) and a non-monotonic, quadratic function, assumed to fit better industrial facilities with high smokestacks.

To monitor the degree of multicollinearity between the explanatory variables, we used the variance inflation factor (VIF) and did not introduce simultaneously into the models the variables for which VIF exceeded a predefined threshold (VIF > 3.0) (Kutner 2004). Although different combinations of control variables were tested, only the best-performing models are reported in the following discussion, for brevity’s sake. The analysis was performed in the SPSS 25.0TM software (IBM 2016).

Results

General trends

Figure 3 features residential locations of the children born in the study area in 2015. As the figure shows, the residential locations of the newborns are scattered fairly evenly across the entire study area.
Fig. 3

Residential locations of children born in 2015 and covered by the analysis. The study cohort includes full-term singleton births (N = 7095) and preterm singleton births (N = 365), registered in the M&C clinic not later than 14 days after the birth

Figure 4 illustrates spatial patterns of head circumference and birth weight of the newborns. In this figure, orange and red colors indicate areas where average measurements of the newborns are lower than the region-wide averages (34.4 cm for head circumference and 3273 g for the birth weight, respectively), while blue and light blue colors indicate areas in which the measurements tend to exceed those observed region-wide.
Fig. 4

Geographic patterns of selected indicators of perinatal development. a Head circumference (cm). b Birth weight (grams). Individual-level data are interpolated using kernel interpolation with barriers (see text for explanations)

Figure 4 shows several spots, in which birth measurements tend to be lower than the average (region-wide) values. Most of these spots surround the main industrial zone. Another hotspot of low measurements located in the southeastern part of the study area, in the town of Kiryat Tivon (see Fig. 4a and b).

Regression models

Table 1 reports the results of controlled regressions estimated for distances from the “best-performing” source locations, that is, locations identified out of 1000 potential “source” locations considered in the analysis (see “Materials and methods” section). As previously noted (see “Identification methodology” section), the proposed approach ranks all potential “candidate” points, according to the degree of potential association of each of them with the observed morbidity events. Therefore, the result of such an assessment is not a single point but all candidate points ranked by regression fit estimates. However, for brevity’s sake, Table 1 reports only the regression results for the “best-performing” points, marked by small black stars in Fig. 5.
Table 1

Factors effecting selected birth outcomes in the study area (method, multivariate OLS regression)

Predictors

Model 1a

Model 2a

Betab

95% CI for Bc

p value

Betab

95% CI for Bc

p value

Lower

Upper

Lower

Upper

Linear distance from the identified industrial spot, km

− 0.147

− 0.153

− 0.035

0.002

− 0.090

0.344

3.514

0.040

Squared distance from the identified industrial spot

0.167

− 0.760

− 0.644

< 0.001

0.104

− 38.268

− 0.924

< 0.001

Number of cases

7460

7460

Moran’s I index

0.001ns

0.001ns

R 2

0.254

0.363

F

416.246***

703.725***

R2 changed

0.003

0.002

F-test of R2 change

9.663***

5.440***

The models reported in the table are estimated for the distances to the “candidate” point with the highest probability of being potential risk sources for the observed ABOs, that is, risk source location, distances to which help to improve the models’ fits most significantly (see text for explanations)

***Indicates a 0.01 significance levels, respectively; bstandardized regression coefficient; ct statistic and its significance value; dcompared to bivariate models, including linear and squared distance terms only

Model 1, multivariate regression with linear and quadratic wind-adjusted distance terms; dependent variable, birth head circumference (cm); controlled for gestation age (number of days), newborns’ gender (1 male, 2 female), the total number of siblings (the number of children in the family), mother’s nationality (1, Jewish; 0, not Jewish), socioeconomic status of the neighborhood, and a dichotomous variable separating preterm and normal-term babies (1 full-term, 0 preterm)

Model 2, multivariate regression with linear and quadratic wind-adjusted distance terms; dependent variable, birth weight (grams); controlled for gestation age (number of days), newborns’ gender (1 male, 2 female), the total number of siblings (the number of children in the family), mother’s nationality (1, Jewish; 0, not Jewish), socioeconomic status of the neighborhood, and a dichotomous variable separating preterm and normal-term babies (1 full-term, 0 preterm)

Fig. 5

Risk source assessment for the observed reduction in the newborns’ head circumference (a, c) and birth weight (b, d), estimated by controlled linear (a, b) and square-term (c, d) regressions with wind-weighted distances. The maps feature the values of the coefficient of determination (\( {R}_{ij}^2 \)), estimated for different locations in the study area (see text for explanations) and interpolated using kriging technique. The wind adjustments are based on the wind rose from the “Igud Arim” AQMS, marked by small black triangles in the maps. Small stars mark the location, distances to which are used in controlled regressions reported in Table 1

In Table 1, models, incorporating distances to these locations, are reported separately for head circumference (HC) and birth weight (BW) and incorporate linear and quadratic wind-weighted distance terms. As models show, linear and quadratic distances emerged as statistically significant predictors of both HC and BW (dist, β = − 0.147, p < 0.01; dist2, β = 0.167, p < 0.01 for HC; and dist, β = − 0.090, p < 0.05; dist2, β = 0.104, p < 0.01 for BW).

In the initial stages of the analysis, we tested for the spatial autocorrelation of OLS regression residuals using Morans I spatial dependence test (Anselin et al. 2010). Results of the test, reported in the Table 1, do not provide evidence for the presence of significant spatial autocorrelation (Morans I < 0.01, p > 0.1), which would have necessitated the use of spatial dependence models.

Figure 5 features probability surfaces based on the determination coefficients (\( {R}_{ji}^2 \)), obtained from multivariate regression models, and estimated separately for HC (Fig. 5a and c) and BW (Fig. 5b and d). Concurrently, Fig. 5a and b shows risk source probability surfaces based on determination coefficients obtained from controlled linear regressions, while Fig. 5c and d shows risk source probability surfaces based on determination coefficients obtained from controlled square-term multivariate regression models. These maps were generated by applying kriging interpolation (see “Geo-statistical analysis” section).

Figure 6 reports the results of a sensitivity test of changes in the analyzed indicators of prenatal development as a function of distance to the identified industrial spot (based on the regression equation for the best-performing point). As Fig. 6 shows, adding distances from the points in this locus is associated, ceteris paribus, with a reduction in HC by about 0.5 cm and a reduction in BW by about 60 g, when average birth performance measures of children born in the “maximum impact zone” at the distance of 3–4 km from that risk source locus and in more remote areas are mutually compared (see Fig. 6a). This spot is located in the central part of the industrial zone, near industrial facility 5 (see Fig. 1).
Fig. 6

The sensitivity test of changes in the analyzed indicators of prenatal development as a function of distance to the identified industrial spot. a Full-term cases. b Preterm cases. The estimates are based on models 1 and 2 (see Table 1). During calculations, the values of all variables (except distances) were set to predefined values or to the averages recorded in the study population: gestation age (274 days (full-term) and 244 days (preterm)); gender, male; total number of siblings (2.04 and 1.83 for full-term and preterm babies, respectively); mother’s nationality, Jewish; socioeconomic status (0.29 and 0.3, respectively); mother’s age, 31.02 and 30.8 years, respectively

Concurrently, preterm newborns, whose residential addresses are located in the “maximum impact” zone, are estimated to have a 0.6–0.7-cm reduction in HC, and a 190-g reduction in BW (Fig. 6b), in addition to the reduction in prenatal measurements in comparison to full-term babies (Marcdante and Kliegman 2015).

One comment is important: although the observed increase in the model fit, attributed to the inclusion of industrial proximity variables, is relatively small (0.2–0.3%), it appears to be statistically significant, as indicated by F-test of R2 change (HC, F = 9.663, p < 0.01 and BW, F = 5.440, p < 0.01; see Table 1).

The model sensitivity test

Figure 7 reports results of a sensitivity test, we ran to determine whether our results are critically sensitive to the wind adjustment of distances we performed. For this test, we used the wind rose obtained for the “Kiryat Tiv’on” AQMS (see Fig. 7b), located 11.2 km southeast from the “Igud Arim” AQMS, data for which were originally used in the analysis (the AQMSs in question are marked by small black triangles in Figs. 1, 5, and 7). As Fig. 7 shows, the re-identified risk source spot moved slightly closer to industrial facility 5, but, in general, its configuration and location appear to be nearly identical to those initially detected (see Fig. 5b and d), thus indicating that our results are essentially robust.
Fig. 7

The sensitivity test of the observed reduction in the newborns’ head circumference (a) and birth weight (b) to changes in the recorded wind source. The maps feature the values of the coefficient of determination (R2), estimated for different locations in the study area (see text for explanations) and interpolated using kriging technique. The wind adjustments are based on the wind rose from the “Kyrat Tiv’on” AQMS, marked by small black triangles in the maps

Discussion and conclusions

In the present study, we test a novel empirical approach to the spatial identification of environmental health hazards potentially associated with the observed adverse birth outcome patterns. The underlying assumption behind this identification approach is that people living in a close proximity to an environmental hazard tend to be more exposed than those living at a distance from that source (Dummer et al. 2003; Yang et al. 2004; Hansen 2005; Ahern et al. 2011; Castello et al. 2013; Adela et al. 2013; Barnett et al. 2011; Bertin et al. 2015; Gehring et al. 2011, 2014). Initially, this methodology was developed and tested by Svechkina et al. (2017) using cancer morbidity data. According to this method, alternative “candidate” locations are assessed in sequence and then ranked based on the strength associated with the observed ABOs.

In the present study, we applied this approach to the indicators of the prenatal development, observed in the Haifa Bay Area (HBA) in Israel, characterized by multiple sources of air pollution. Using the proposed approach, we identified a spot in the local industrial zone, which hosts several “candidate” points with the highest probability of association with the observed ABO patterns. These points appear to cluster in the area, in which local petrochemical industries are located (see Appendix Table 3 and Fig. 1).

This result is fairly consistent with results of other empirical studies, indicating the association between proximity to the petrochemical industries and adverse birth outcomes (Lin et al. 2001; Yang et al. 2004; Hansen 2005; Castello et al. 2013, Dandvand et al. 2014; Stacy et al. 2015). Thus, according to Yang et al. (2004), the prevalence of delivery of preterm infants was found to be significantly higher in residential locations near the oil refinery plants than in more remote areas (OR = 1.14, 95% CI = 1.01–1.28). According to another study conducted by Dadvand et al. (2013), maternal exposure to particulate air pollution, associated with proximity to petrochemical industries, was found to be associated with RBW. Similar results are reported by Castello et al. (2013) for Spain, who found that residential proximity to mining, biocides, and animal waste management plants is significantly associated with RBW prevalence.

Our study provides further support for the fact that air pollution from petrochemical industry can affect pregnancy outcome. In particular, multivariate regressions, controlling for personal, socioeconomic, and geographical factors, revealed that the observed rates of RHC and RBW generally decline as a function of distance from the identified hot spot, which is coincide with a local industrial locus. It should be noted, however, that previous studies analyzed proximity to known sources of environmental exposure, while in our study, the location of potential risk sources was not identified a priori, but was determined by the systematic search approach among a range of potential risk source locations.

Considering potential differences in the special distribution of air pollutants, we investigated two functional forms of relationships between birth outcomes and proximity to environmental health hazard: a monotonic linear function, considered more applicable to close-to-the-ground sources of exposure, such as thoroughfare roads and oil storages, and a non-monotonic, quadratic function, considered to fit better industrial facilities with high smokestacks.

Although the observed increase in the model fit, attributed to wind adjustment of the distance, is relatively small (0.2–0.3%), it is statistically significant, as indicated by F-test of R2 change (p < 0.01). In addition, even such small differences detected in the entire study cohort effectively mean that the entire development curve of the newborns in the study area is shifted downwards due to environmental exposure, and this shift, albeit relatively minor, may become critical for future postnatal development (Marcdante and Kliegman 2015).

This study contributes to the existing body of literature by providing an additional verification of the risk source methodology originally developed by Svechkina et al. (2017) for the analysis of cancer incidence. The practical importance of this methodology is that it makes it possible to investigate the association between potential health hazards, even in cases in which the exact location of such hazards is not a priori known to the researchers. In the present study, we applied this methodology to individual birth outcome data, instead of generalized cancer morbidity risk surfaces used in the former study. The results of the study demonstrate the risk source identification method performs effectively and may thus be recommended for epidemiological studies which goal is to identify potential sources of exposure to which the observed morbidity might be related.

It should be noted, however, that the present study is a semi-ecological analysis, which includes both individual-level variables as well as explanatory variables measured at the group level (such as smoking and socioeconomic status) or as distance gradients, A well-known limitation of such studies is that they cannot attribute causality to the relationship revealed but can only indicate a possible health risk effect (Greenland 2001; Loney and Nagelkerke 2014).

Footnotes

  1. 1.

    Geocoding is a geo-statistical process of converting street addresses into X and Y coordinates suitable for mapping (Rushton et al. 2006).

  2. 2.

    The average annual wind frequency from point j to point i was calculated based on the angle between the points. The angles were calculated using the proximity toolset in ArcMap 10.4TM (ESRI, 2015).

Notes

Acknowledgements

The authors express their gratitude to the members of the study’s steering committee of the Israel Ministry of Health, specifically to Dr. Jonathan Dubnov, Ms. Batia Madjar, and Ms. Riki Shemer for consultations, quality control of birth records, and initial processing of data for this research. Our gratitude is also due to Mr. Shahar Fertig for his valuable help with database preparation.

Funding

The first author thanks the Israel Ministry of Absorption and the Rieger Foundation-Jewish National Fund Program for Environmental Studies for their financial support of this study.

Compliance with ethical standards

Competing interests

The authors declare that they have no competing interests.

Ethical considerations

The study was approved by the Helsinki committee of the Ministry of Health (MoH 084-2016) and the Ethical Board of University of Haifa (394/15).

References

  1. Adela CA, Río I, García-Pérez J et al (2013) Adverse birth outcomes in the vicinity of industrial installations in Spain 2004–2008. Environ Sci Pollut Res 20(7):4933–4946CrossRefGoogle Scholar
  2. Ahern M, Mullett M, Mackay K, Hamilton C (2011) Residence in coal-mining areas and low-birth-weight outcomes. Matern Child Health J 15(7):974–979CrossRefGoogle Scholar
  3. Anselin L, Syabri I, Kho YGD (2010) an introduction to spatial data analysis. Geogr Anal 38:15–22Google Scholar
  4. Bacci S, Bartolucci F, Minelli L, Chiavarini M (2016) Preterm birth: analysis of longitudinal data on siblings based on random-effects logit models. Frontiers in public health: population, reproductive and sexual health 4:278CrossRefGoogle Scholar
  5. Backes CH, Nelin T, Gorr MW, Wold LE (2013) Early life exposure to air pollution: how bad is it? Toxicol Lett 216(1):47–53CrossRefGoogle Scholar
  6. Ballester et al (2010) Air pollution exposure during pregnancy and reduced birth size: a prospective birth cohort study in Valencia, Spain. Environ Health 9(6)Google Scholar
  7. Banerjee T, Murari V, Kumar M, Raju MP (2015) Source apportionment of airborne particulates through receptor modeling: Indian scenario. Atmos Res 164(165):167–187CrossRefGoogle Scholar
  8. Barnett AG, Plonka K, Seow WK, Wilson LA, Hansen C (2011) Increased traffic exposure and negative birth outcomes: a perspective cohort in Australia. Environ Health 10:26CrossRefGoogle Scholar
  9. Baron-Epel O, Keinan-Boker L, Weinstein R, Shohat T (2010) Persistent high rates of smoking among Israeli Arab males with concomitant decrease among Jews. Isr Med Assoc J: IMAJ 12(12):732–737Google Scholar
  10. Begum BA, Kimb E, Biswasa SK, Hopke PK (2004) Investigation of sources of atmospheric aerosol at urban and semi-urban areas in Bangladesh. Atmos Environ 38:3025–3038CrossRefGoogle Scholar
  11. Bell ML, Ebisu K, Belanger K (2007) Ambient air pollution and low birth weight in Connecticut and Massachusetts. Environ Health Perspect 115:1118–1123CrossRefGoogle Scholar
  12. Bertin M, Chevrier C, Serrano T, Monfort C, Rouget F, Cordier S, Viel JF (2015) Association between prenatal exposure to traffic-related air pollution and preterm birth in the PELAGIE mother-child cohort, Brittany, France. Does the urban-rural context matter? Environ Res 142:17–24CrossRefGoogle Scholar
  13. Bhutta AT, Cleves MA, Casey PH, Cradock MM, Anand KJS (2002) Cognitive and behavioral outcomes of school-aged children who were born preterm: a meta-analysis. JAMA 288(6):728–737CrossRefGoogle Scholar
  14. Blencowe H, Cousens S, Oestergaard MZ et al (2012) National, regional, and worldwide estimates of preterm birth rates in the year 2010 with time trends since 1990 for selected countries: a systematic analysis and implications. Lancet 379(9832):2162–2172CrossRefGoogle Scholar
  15. Castello A, Rio I, Fernandez-Novarro P, Waller LA, Clennon JA et al (2013) Adverse birth outcomes in the vicinity of industrial installations in Spain 2004-2008. Environ Sci Pollut Res Int 20:4933–4946CrossRefGoogle Scholar
  16. Cesari R, Paradisi P, Allegrini P (2014) A trajectory statistical method for the identification of sources associated with concentration peak events. Int J Environ Pollut 55:94–103CrossRefGoogle Scholar
  17. Chen X, Ye J (2015) When the wind blows: spatial spillover effects of urban air pollution. environment for development discussion: paper series. EFD DP 15–15Google Scholar
  18. Chiavarini M, Bartolucci F, Gili A, Pieroni L, Minelli L (2012) Effects of individual and social factors on preterm birth and low birth weight: empirical evidence from regional data in Italy. Int J Public Health 57(2):261–268CrossRefGoogle Scholar
  19. Cooper JA, Watson JG (1980) Receptor oriented methods of Air Particulate Source Apportionment. J Air Pollut Control Assoc:1116–1124Google Scholar
  20. Dadvand et al (2013) Maternal exposure to particulate air pollution and term birth weight: a multi-country evaluation of effect and heterogeneity. Environ Health Perspect 121(3):367–373CrossRefGoogle Scholar
  21. Dadvand P, Ostro B, Figueras F et al (2014) Residential proximity to major roads and term low birth weight: the roles of air pollution, heat, noise and road-adjacent trees. Epidemiology 25(4):518–525CrossRefGoogle Scholar
  22. Dore AJ, Vieno M, Fournier N, Weston KJ, Sutton MA (2006) Development of a new wind-rose for the British Isles using radiosonde data, and application to an atmospheric transport model. Q J R Meteorol Soc 132:2769–2784Google Scholar
  23. Dummer TJ, Dickinson HO, Parker L (2003) Adverse pregnancy outcomes around incinerators and crematoriums in Cumbria, North West England, 1956-93. J Epidemiol Community Health 57(6):456–461CrossRefGoogle Scholar
  24. Eliyahu S, Weiner E, Nachum Z, Shalev E (2002) Epidemiologic risk factors for preterm delivery. The Israel Medical Association Journal: IMAJ 4(12):1115–1117Google Scholar
  25. ESRI: ArcGIS desktop help 10.2. http://webhelp.esri.com (2015) Accessed 20 January 2016
  26. Ferguson KK, O’Neill MS, Meeker JD (2013) Environmental contaminant exposures and preterm birth: a comprehensive review. Journal of Toxicology and Environmental Health, Part B: Critical Reviews 16(2):69–113CrossRefGoogle Scholar
  27. Fleischer NL et al (2014) Outdoor air pollution, preterm birth, and low birth weight: analysis of the World Health Organization global survey on maternal and perinatal health. Environ Health Perspect 122(4):425–430CrossRefGoogle Scholar
  28. Gehring U, Wijga AH, Fischer P et al (2011) Traffic-related air pollution, preterm birth and term birth weight in the PIAMA birth cohort study. Environ Res 111(1):125–135CrossRefGoogle Scholar
  29. Gehring U, Tamburic L, Sbihi H, Davies HW, Brauer M (2014) Impact of noise and air pollution on pregnancy outcomes. Epidemiology 25(3):351–358CrossRefGoogle Scholar
  30. Greenland S (2001) Ecologic versus individual-level sources of bias in ecologic estimates of contextual health effects. Int J Epidemiol 30(6):1343–1350CrossRefGoogle Scholar
  31. Hannam K, MacNamee R, Baker P et al (2014) Air pollution exposure and adverse pregnancy outcomes in a large UK birth cohort: use of a novel spatio-temporal modeling technique. Scand J Work Environ Health 40(5):518–530CrossRefGoogle Scholar
  32. Hansen CA (2005) Investigating the effect of maternal exposure to ambient air pollution during pregnancy on birth outcomes: challenges related to exposure assessment methods and data analysis. Australasian Epidemiologist 12(2):11–15Google Scholar
  33. Heaman M, Kingston D, Chalmers B, Sauve R, Lee L, Young D (2013) Risk factors for preterm birth and small-for-gestational-age births among Canadian women. Pediatric and Perinatal Epidemiology 27(1):54–61CrossRefGoogle Scholar
  34. Hopke PK (2003) Recent developments in receptor modeling. J Chemom 17:255–265CrossRefGoogle Scholar
  35. IBM: SPSS Statistics desktop help 22. http://www.ibm.com; Accessed 1 May 2016.
  36. Institute of Health Metrics and Evaluation (2006) www.healthdata.org. Accessed 1 May 2016.
  37. Isayama T, Lee SK, Mori R et al (2012) Comparison of mortality and morbidity of very low birth weight infants between Canada and Japan. Paediatrics 130(4):965CrossRefGoogle Scholar
  38. Israel Central Bureau of Statistics (ICBS) (2016) Statistical abstract of Israel 2015: population, by district, sub districtand religion. http://www.cbs.gov.il/. Accessed 1 May 2016
  39. Israel Ministry of Environmental Protection (IMEP). Map of the air monitoring stations. http://www.sviva.gov.il (2016). Accessed 21 Apr 2016.
  40. Khan S, Cao Q, Zheng YM, Huang YZ, Zhu YG (2008) Health risks of heavy metals in contaminated soils and food crops irrigated with wastewater in Beijing. China Environmental Pollution 152(3):686–692CrossRefGoogle Scholar
  41. Kim E, Hopke PK (2004) Comparison between conditional probability function and nonparametric regression for fine particle source directions. Atmos Environ 38:4667–4673CrossRefGoogle Scholar
  42. Kutner MH, Nachtsheim CJ, Neter, J. Applied Linear Regression Models. McGraw-Hill Irwin, 2004.Google Scholar
  43. Lau C, Ambalavanan N, Chakraborty H, Wingate MS, Carlo WA (2013) Extremely low birth weight and infant mortality rates in the United States. Paediatrics 131(5)Google Scholar
  44. Li Z, Ma Z, Kuijp v d TJ, Yuan Z (2014) Huang L. A review of soil heavy metal pollution from mines in China: Pollution and health risk assessment. Sci Total Environ 468(469):843–853CrossRefGoogle Scholar
  45. Lin MC, Yu HS, Tsai SS et al (2001) Adverse pregnancy outcome in a petrochemical polluted area in Taiwan. Journal of Toxicology and Environmental Health. Part A: Current Issues 63(8):565–574Google Scholar
  46. Liu X, Song Q, Tang Y, Li W, Xu J, Wu J, Wang F, Brookes PC (2013) Human health risk assessment of heavy metals in soil–vegetable system: a multi-medium analysis. Sci Total Environ 463(464):530–540CrossRefGoogle Scholar
  47. Llop S, Ballester F, Estarlich M, Esplugues A, Rebagliato M, Iñiguez C (2010) Preterm birth and exposure to air pollutants during pregnancy. Environ Res 110(8):778–785CrossRefGoogle Scholar
  48. Loney T, Nagelkerke NJ (2014) The individualistic fallacy, ecological studies and instrumental variables: a causal interpretation. Emerg Themes in Epidemiol 11:18CrossRefGoogle Scholar
  49. Lundgren P et al (2014) Low birth weight is a risk factor for severe retinopathy of prematurity depending on gestational age. PLoS One 9(10)Google Scholar
  50. Luo ZC, Wilkins R, Kramer MS (2006) Effect of neighborhood income and maternal education on birth outcomes: a population-based stud. CMAJ 174(10):1415–1420CrossRefGoogle Scholar
  51. Mahmood A, Malik RN (2014) Human health risk assessment of heavy metals via consumption of contaminated vegetables collected from different irrigation sources in Lahore, Pakistan. Arabian J Chem 7(1):91–99CrossRefGoogle Scholar
  52. Maisonet M, Correa A, Misra D (2004) Jaakkolaf JJK. A review of the literature on the effects of ambient air pollution on foetal growth. Environ Res 95:106–115CrossRefGoogle Scholar
  53. Marcdante KJ, Kliegman RM (2015) Nelsons essentials of pediatrics. Philadelphia, PA, 784Google Scholar
  54. McKenzie LM, Ruixin G, Witter RZ, Savitz DA, Newman LS, et al. (2014) Birth outcomes and maternal residential proximity to natural gas development in rural Colorado. Environ Health Perspect 412Google Scholar
  55. MH (2014) Environmental health in Israel 2014. Ministry of health of Israel. http://www.health.gov.il/publicationsfiles/bsv_sviva2014e.pdf. Accessed 20 Jan 2015
  56. Moore DA, Carpenter TE (1999) Spatial analytical methods and geographic information systems: use in health research and epidemiology. Epidemiologic Reviews 21(2):143–161Google Scholar
  57. Morello-Frosch R, Jesdale BM, Sadd JL, Pastor M (2010) Ambient air pollution exposure and full-term birth weight in California. Environ Health 9(44):1–13Google Scholar
  58. Morken NH (2012) Preterm birth: new data on a global health priority. Lancet 379(9832):2128–2130CrossRefGoogle Scholar
  59. Paz S, Linn S, Portnov BA, Lazimi A, Futerman B, Barchana M (2009) Non-Hodgkin Lymphoma (NHL) linkage with residence near heavy roads - a case study from Haifa Bay. Israel Health and Place 15:636–641CrossRefGoogle Scholar
  60. Pedersen M, Gehring U, Beelen R et al (2016) Elemental constituents of particulate matter and newborn’s size in eight European cohorts. Environ Health Perspect 124(1):141–150CrossRefGoogle Scholar
  61. Rushton G, Armstrong MP, Gittler J, Greene BR, Pavlik CE, West MM, Zimmerman DL (2006) Geocoding in cancer research: a review. Am J Prev Med 30(2):16–24CrossRefGoogle Scholar
  62. Salvador P, Artinano B, Alonso DG, Querol X, Alastuey A (2004) Identification and characterization of sources of PM10 in Madrid (Spain) by statistical methods. Atmos Environ 38:435–447CrossRefGoogle Scholar
  63. Seinfeld JH, Pandis SN (2006) Atmospheric Chemistry and Physics. Hoboken; NJ. WileyGoogle Scholar
  64. Sermage-Faure C, Laurier D, Goujon-Bellec S, Chartier M, Guyot-Goubin A, Rudant J, Hemon D, Clave J (2012) Childhood leukemia around French nuclear power plant. The Geocap study, 2002–2007. Int J Cancer 131(5):E769–E780CrossRefGoogle Scholar
  65. Sorbye IK, Wanigaratne S, Urquia M (2016) Variations in gestational length and preterm delivery by race, ethnicity and migration. Best Pract Res Clin Obstet & Gynaecol 32:60–68CrossRefGoogle Scholar
  66. Spencer N, Bambang S, Logan S, Gill L (1999) Socioeconomic status and birth weight: comparison of an area-based measure with the Registrar General’s social class. J Epidemiol Community Health 53:495–498CrossRefGoogle Scholar
  67. Stacy SL, Brink LL, Larkin JC, Sadovsky Y, Goldstein BD, Pitt BR, Talbott EO (2015) Perinatal outcomes and unconventional natural gas in Southwest Pennsylvania. PLoS One 10(6):1–15CrossRefGoogle Scholar
  68. Stohl A (1996) Trajectory statistics-a new method to establish source-receptor relationships of air pollutants and its application to the transport of particulate sulphate in Europe. Atmos Environ 30(4):579–587Google Scholar
  69. Svechkina A, Portnov BA (2017) A new approach to spatial identification of potential health hazards associated with childhood asthma. Sci Total Environ 595:413–424Google Scholar
  70. Svechkina A, Zusman M, Rybnikov N, Portnov BA (2017) Spatial identification of potential health hazards: a systematic areal search approach. Int J Health Geogr 16(5):1–13Google Scholar
  71. UNICEF data: Monitoring the situation of children and women (2016) http://data.unicef.org/nutrition/low-birthweight.html#sthash.tfLramHH.dpuf. Accessed 20 June 2016.
  72. Urquia ML et al (2010) International migration and adverse birth outcomes: role of ethnicity, region of origin and destination. J Epidemiol Community Health 64:243–251CrossRefGoogle Scholar
  73. Wang C, Zhou X, Chen R, Duan X, Kuang X, Kan H (2013) Estimation of the effects of ambient air pollution on life expectancy of urban residents in China. Atmos Environ 80:347–351Google Scholar
  74. Wen SW, Smith G, Yang Q, Walker M (2004) Epidemiology of preterm birth and neonatal outcome. Semin Fetal Neonatal Med 9(6):429–435CrossRefGoogle Scholar
  75. WHO (2012) Born too soon: the global action report on preterm birth: World Health Organization report. http://www.who.int/pmnch/media/news/2012/ preterm_birth_report/en/. Accessed 11 May 2016
  76. WHO (2016) Children: reducing mortality: World Health Organization fact sheet. http://www.who.int/mediacentre/factsheets/fs178/en/. Accessed 1 May 2016
  77. Xie Y, Berkowitz CM (2006) The use of positive matrix factorization with conditional probability functions in air quality studies: an application to hydrocarbon emissions in Houston. Texas Atmospheric Environment 40:3070–3091CrossRefGoogle Scholar
  78. Xu X, Akhtar US (2010) Identification of potential regional sources of atmospheric total gaseous mercury in Windsor, Ontario, Canada using hybrid receptor modeling. Atmos Chem Phys 10:7073–7083Google Scholar
  79. Yang CY, Chang CC, Chuang HY, Ho CK, Wu TN, Chang PY (2004) Increased risk of preterm delivery among people living near the three oil refineries in Taiwan. Environ Int 30(3):337–342Google Scholar
  80. Zhang ZY, Wong MS, Lee KH (2015) Estimation of potential source regions of PM2.5 in Beijing using backward trajectories. Atmospheric Pollution 6:173–177CrossRefGoogle Scholar
  81. Zoë L, Fleming ZL, Monks PS, Manning AJ (2012) Review: untangling the influence of air-mass history in interpreting observed atmospheric composition. Atmos Res 104–105:1–39Google Scholar
  82. Zusman M, Dubnov J, Barchana M, Portnov BA (2012) Residential proximity of petroleum storage tanks and associated cancer risks: Double Kernel Density approach vs. zonal estimates. Sci Total Environ 441:265–276CrossRefGoogle Scholar

Copyright information

© Springer-Verlag GmbH Germany, part of Springer Nature 2018

Authors and Affiliations

  1. 1.Department of Natural Resources and Environmental Management, Faculty of ManagementUniversity of HaifaHaifaIsrael

Personalised recommendations