Environmental and socio-economic determinants of infant mortality in Poland: an ecological study

Research design and data source

An ecological design was applied to study the relationships between infant mortality and environmental and socio-economic determinants at an aggregate level. We used data from the 66 sub-regions of Poland, defined as NUTS-3 (Nomenclature of Units for Territorial Statistics), according to the recommendations of the European Union [31]. Information on infant mortality rates, and indices of urbanization, occupation, occupational conditions, environmental pollution and socio-economic status in the years 2005–2011 were obtained from the Central Statistical Office of Poland. A detailed list of the original predictors is presented in Additional file 1: Table S1.

Data analysis

A total of 23 independent variables measured at the sub-region level were used in the analysis. As most variables had a statistically significant deviation from the normal distribution, they were transformed before analysis. For dealing with skewness and simultaneously accounting for unique properties of variables that include a zero value, the inverse hyperbolic sine transformation [32] was used according to the formula:

The θ parameter was chosen, so that the distribution of transformed variables was as similar as possible to the normal distribution.

As correlation analysis revealed strong correlations within some groups of independent variables, to avoid the multicollinearity problem in regression modeling, factor analysis was used. Factor analysis places together closely related items to form a smaller number of “latent variables” (factors) explaining the maximum amount of common variance in a correlation matrix. Each factor consists of all items, but the importance of individual variables corresponds to factor loadings. Therefore, individual factors are scores that represent different dimensions of source data. To identify factors, Principal Component Analysis with orthogonal rotation (equamax) on factor loadings was used [33]. This allowed a reduction of the large number of variables to a smaller number of mutually uncorrelated factors [34]. Finally, four factors were extracted that explained 77.3 % of the cumulative variance of the original dataset. Factors were given descriptive names based on their intercorrelation structure. For this purpose, variables with absolute factor loadings above 0.6 were considered noteworthy (Additional file 2: Table S2). This criterion allowed for unequivocal naming, as none of the variables were highly correlated with more than one factor (the variable “cold microclimate” was not substantially correlated with any factor). The factors were assigned the following names: “poor working environment” (explained 23.2 % of variance), “urbanization and employment in the service sector” (explained 20.8 % of variance), “industrial pollution” (explained 19.7 % of variance), “economic wealth” (explained 13.6 % of variance). As extracted factors were composed of a set of variables measured in different units, the factors themselves have no particular unit; rather, they are standardized scores (with mean = 0 and standard deviation (SD) = 1). Factors should be interpreted as hidden characteristics underlying the observed variables used to create them and the factor loadings describe which variables (and in which direction) are represented the most by a given factor [35].

To assess the relationship between infant mortality and the four extracted factors, the Spearman correlation was calculated. This simple analysis was then extended with a more complex multivariate regression model with the infant mortality ratio as the dependent variable and all four factors as independent ones, taking into consideration the presence of the correlated data (time-repeated measurements of 7 years of observation). We employed the Generalized Estimating Equations modeling approach [36] assuming an exchangeable structure for the working correlation matrix. The results are presented as the expected change of infant mortality rate per 100,000 live births for an increase in a given factor of one unit which was equal to 1 SD.

To describe the average annual change of the infant mortality rate during the studied period (2005–2011), a simple regression model was used in each sub-region separately and the outcome was presented graphically on a map. Statistical analyses were conducted with IBM® SPSS® Statistics for Windows, Version 20.0-IBM Corp. Armonk, NY, USA.

Strengths

As far as we are aware, this is the first study to measure the combined effect of working environment and industrial pollution on infant mortality in Poland. Using the unique analysis of complex relationships between environmental factors and infant mortality, we were able to achieve a high statistical power after including data from 7 years of observation in the 66 sub-regions. Factor analysis enabled the approximate determination of hidden phenomena that manifest themselves in different ways via many individual indicators. This multivariate correlation method is well-suited for revealing underlying patterns or structure among variables showing high degrees of inter-correlation, as in the case of environmental risk variables supposedly influencing health status [37]. Our approach enabled the identification of four independent environmental characteristics out of 23 variables. Two of them appeared to be significantly related to the infant mortality.

Limitations

The most important limitation is the ecological study design, which does not allow either firm conclusions to be drawn in terms of relationships on an individual level (ecological fallacy) or on causal relationships. Second, the study allowed for limited control of confounding factors, as the appropriate information was not available from public statistics. For example, in Poland 10 % of pregnant women experienced alcohol consumption at least once during pregnancy. The proportion of alcohol consumers was higher among more educated women and women living in informal relationships [38]. Information on the distribution of alcohol consumption among pregnant women by sub-region was not available, but there is evidence of regional variation in alcohol consumption in Poland. The amount of alcohol consumption during pregnancy varied from 0.06 units to 1.19 units per week. Also, information on smoking prevalence by sub-region was not available; however, there is evidence of regional variation in smoking rates among women in general (26.5–64.9 %) and pregnant women in particular (2.7–17.2 %) [38]. Third, our basic data were collected via the national routine statistics system, which put the methods of data collection and data quality beyond our control. Some information obtained from the Central Statistical Office was not comprehensive. For example, data on occupational hazards included only data on workplaces employing more than nine persons [39]. Fourth, in our analysis we focused on environmental factors, but characteristics of the health status of mothers and their access to health services were not controlled directly. However, the latter factors are strongly correlated with economic wealth and urbanization [7, 40].

Our results could be partially explained by the confounding effect of disparities in access to health services by place of residence (urban/rural) [41]. There is evidence that in some regions high risk pregnancies are not identified early enough in prenatal care and there are problems in referring such cases to specialists. Also, there are issues in terms of the equipment of intensive care units for premature infants [42, 43]. Moreover, there is evidence concerning health and life-style problems in pregnant women, including anemia in about 30 %, hypertension in about 6 %, diabetes in about 5 %, smoking in about 7.8 % and alcohol consumption in about 10 %. Nevertheless, it is worth stressing that the vast majority of pregnant women (98 %) are covered by prenatal care in the first trimester [44]. Also, there are observations on positive changes in the diet of the pregnant women, i.e. an increased intake of fresh fruit and vegetables, as well as dairy products, in over 60 % of women [38].

Furthermore, the relationship between the extracted factors and infant mortality might be modified by the timing of the infant death. Addressing this problem was not possible at the population level because information on neonatal and post-neonatal mortality is not available from routine statistics.

Plausibility

Causes of health conditions leading to infant death are not fully understood [18]. It is likely that other, potentially modifiable environmental factors could play an important role. Generally, our results are in accordance with results from other populations. Low birth weight, which increases the risk of infant mortality, is related to poor working environment [15, 45]. Moreover, exposure to harmful substances at work (e.g., lead, mercury, arsenic, benzene, phthalates) is associated with birth defects or low birth weight [46–48]. However, poor working conditions do not necessarily imply a direct threat to the development of the fetus in mothers exposed, as this relationship has also been shown with the exposure of fathers [6, 49, 50]. Poor occupational environments may act as a marker of other traits, e.g., socio-economic position, which affects the development of the fetus through lifestyle behaviors, psychosocial stress or access to health care [7, 51]. In another study, work in harmful conditions combined with smoking and drinking was associated with the risk of fetal and neonatal death [22].

In our analysis, infant mortality was related to the extracted factors which combined characteristics of industrial pollution. Harmful effects of exposure to air pollution during pregnancy have been observed in many studies [16–18]. Prenatal exposure to air pollution can influence fetal growth due to oxidative stress and inflammation, which may lead to lower birth weight, impaired lung development, early changes in immune development and higher infant mortality. This risk is often observed in populations with low socio-economic status [52]. Higher health risks associated with living and working in areas of heavy industry, e.g., coal mining, metallurgy, smelting, petrochemical, have been found to be related to the release of heavy metals, sulfur compounds, and polycyclic aromatic hydrocarbons and to the presence of fine particulate matter PM_2.5 and PM₁₀, which worsens the quality of water and air [53, 54]. Relationships between women of reproductive age living near hazardous waste landfills and risks of birth defects, low birth weight, prematurity have been described in many studies [14, 20]. In the EUROHAZCON study in European populations, authors found that living during pregnancy within 3 kilometers of a landfill, as well as water pollution, significantly increased the risk of chromosomal and non-chromosomal birth defects [19].

In our study, the combined effect of “urbanization and employment in the service sector” was not related to infant mortality, which may be related to the impact of employment structure, (i.e. unequal distribution by category of employment according to the division in Table 1). The results of other studies are controversial. Higher infant mortality has been associated with living in urban areas in some studies [55, 56], but in others the opposite effect has been demonstrated [57, 58]. Our results could be explained partially by the increase in the employment of rural residents in urban settings, which has occurred recently in Poland, especially in younger generations [59]. The impact of other, not assessed, confounders, such as health behaviors, is also possible.

The factor “economic wealth” was not associated with infant mortality. This finding is in accordance with the fact that Poland is situated at the beginning of the flat part of the Preston curve [60, 61], which means that further economic growth would have relatively small impact on population health. Kim et al. found that associations between macroeconomic characteristics and infant mortality are ambiguous in developed countries [7]. This suggests that the frequently used measure of economic position (GDP per capita) may not be the best marker of the complex characteristics which affect infant mortality.

Conclusion

The study findings suggested that infant mortality was associated with industrial environmental conditions at the population level. The combined effect of the population characteristics, which are indicated by factors such as “poor working environment” and “industrial pollution”, was related to higher infant mortality and might contribute to explaining the geographical variation in that mortality within a country. On the other side, the effect of the population characteristics, which are indicated by factors such as “urbanization and employment in the service sector” and “economic wealth”, was small and not statistically significant. Strategies to improve working conditions and reduction of industrial pollution might contribute to a decrease in infant mortality in Poland.