Elsevier

Environment International

Volume 60, October 2013, Pages 23-30
Environment International

Does consideration of larger study areas yield more accurate estimates of air pollution health effects? An illustration of the bias-variance trade-off in air pollution epidemiology

https://doi.org/10.1016/j.envint.2013.07.005Get rights and content

Highlights

  • Air pollution models with fine spatial resolution were developed in a large area.

  • Rural and urban areas differed for factors related to health and air pollution levels.

  • Simulations indicated that adjustment for area could limit residual confounding.

  • This study shows that potential for bias increases when large areas are considered.

  • It provides an example of the bias-variance trade-off known in epidemiology.

Abstract

Background

Spatially-resolved air pollution models can be developed in large areas. The resulting increased exposure contrasts and population size offer opportunities to better characterize the effect of atmospheric pollutants on respiratory health. However the heterogeneity of these areas may also enhance the potential for confounding. We aimed to discuss some analytical approaches to handle this trade-off.

Methods

We modeled NO2 and PM10 concentrations at the home addresses of 1082 pregnant mothers from EDEN cohort living in and around urban areas, using ADMS dispersion model. Simulations were performed to identify the best strategy to limit confounding by unmeasured factors varying with area type. We examined the relation between modeled concentrations and respiratory health in infants using regression models with and without adjustment or interaction terms with area type.

Results

Simulations indicated that adjustment for area limited the bias due to unmeasured confounders varying with area at the costs of a slight decrease in statistical power. In our cohort, rural and urban areas differed for air pollution levels and for many factors associated with respiratory health and exposure. Area tended to modify effect measures of air pollution on respiratory health.

Conclusions

Increasing the size of the study area also increases the potential for residual confounding. Our simulations suggest that adjusting for type of area is a good option to limit residual confounding due to area-associated factors without restricting the area size. Other statistical approaches developed in the field of spatial epidemiology are an alternative to control for poorly-measured spatially-varying confounders.

Introduction

The respiratory system is vulnerable to exposure to airborne toxicants during development (Baïz et al., 2011, Bateson and Schwartz, 2008, Bråbäck and Forsberg, 2009, Clark et al., 2010, Latzin et al., 2009, Latzin et al., 2011, Mauad et al., 2008, Pinkerton and Joad, 2000, Pope, 1989, Sram et al., 2005, Woodruff et al., 2008, Wu et al., 2009). Early life exposure to ambient air pollution has been associated with cord blood immunologic parameters (Baïz et al., 2011, Latzin et al., 2011), decreased lung function in newborns (Latzin et al., 2009), asthma and asthma-related symptoms in the first year of life (Aguilera et al., 2013, Andersen et al., 2008, Bateson and Schwartz, 2008, Bråbäck and Forsberg, 2009, Brauer et al., 2002, Clark et al., 2010, Ebisu et al., 2011, Esplugues et al., 2011, Gehring et al., 2002, Gouveia and Fletcher, 2000, Karr et al., 2009, Morgenstern et al., 2007, Nordling et al., 2008, Orazzo et al., 2009, Pino et al., 2004, Pope, 1989, Ryan et al., 2005). Early studies relied on permanent monitoring stations to assess exposure (Andersen et al., 2008, Bateson and Schwartz, 2008, Bråbäck and Forsberg, 2009, Gouveia and Fletcher, 2000, Latzin et al., 2009, Latzin et al., 2011, Orazzo et al., 2009, Pino et al., 2004, Pope, 1989, Sram et al., 2005), an approach with a limited spatial resolution. Over the last decade, models with improved spatial resolution such as dispersion or land-use regression (LUR) models (Hoek et al., 2008) have been increasingly applied (Aguilera et al., 2013, Bråbäck and Forsberg, 2009, Brauer et al., 2002, Clark et al., 2010, Ebisu et al., 2011, Esplugues et al., 2011, Gehring et al., 2002, Karr et al., 2009, Morgenstern et al., 2007, Nordling et al., 2008, Ryan et al., 2005). Such approaches allow consideration of large study areas as a whole, including city centers, surrounding suburban and sometimes rural areas. Considering such large areas entails a potential increase in exposure contrasts and study size, which is a priori desirable in terms of statistical power.

However, the greater population heterogeneity may also increase the potential for confounding, in particular due to spatially-varying factors.

Indeed, not only do ambient air pollution levels vary largely over space, but also possibly other parameters such as exposure to infectious factors, to sources of allergens (e.g. pet ownership, dairy farms, pollen), health-care access and distribution of personal characteristics such as socioeconomic status, smoking and diet (Wright and Subramanian, 2007). Residual confounding may occur if these asthma risk factors are unevenly distributed across types of areas (rural and urbanized areas) and are not assessed, or are assessed with measured error or poorly taken into account in regression models. The alternative between studying small homogeneous areas with fewer subjects and expanding area size together with population heterogeneity and possibly resulting bias can be seen as an illustration of the classical epidemiological trade-off between bias and variance.

The present study aimed to document this bias-variance trade-off related to confounding, size and heterogeneity of study area that may occur in studies of air pollution health effects relying on exposure models that are applied to a large study area. We did not consider the impact on exposure misclassification of using models with a fine spatial resolution, compared e.g., to approaches resting on permanent air quality monitoring stations (Gulliver et al., 2011, Lepeule et al., 2010, Marshall, 2008). We illustrated this trade-off in a study of ambient air pollution effects on respiratory health in the first year of life using simulations and real data. First, we conducted a simulation study to estimate the bias and statistical power in studies with area-dependent unmeasured confounders, to illustrate the efficiency of various analytical approaches. Secondly, we assessed the association between air pollution levels and infant respiratory health in a mother–child cohort study using several analytical approaches to limit residual confounding.

Section snippets

Study population

This study relies on mother–child pairs from the EDEN mother–child cohort. The EDEN mother–child cohort study is a population-based prospective study on pre- and early postnatal nutritional, environmental and social determinants of the children's development and health (Baïz et al., 2011, Drouillet et al., 2009, Hampel et al., 2011, Slama et al., 2009). Between 2003 and 2006, pregnant women expecting singletons at the University Hospitals in Poitiers and Nancy, two middle-sized cities in West

Study population and prevalence of asthma-related outcomes

Approximately half (N = 1082) of EDEN mother–child study population lived in the modeling area of our air dispersion model, had follow-up data and could be included in this study (Fig. 1).

Excluded mothers more often lived in rural areas during pregnancy (Fig. 2) were less likely to breastfeed; their children less frequently attended day-care facilities with other children (Table S1, Supplementary material). The prevalence of recurrent wheeze was 10.5% among included and 7.1% among excluded

Discussion

We characterized the associations between air pollution levels estimated using a fine scale dispersion model and respiratory health in the first year of life within a mother–child cohort conducted in two middle-sized cities. A dispersion model enabled the inclusion of a relatively large area and allowed taking into account local sources of pollution and temporal variations. The areas, including city-centers, urban areas and rural areas appeared to be quite heterogeneous for many factors

Acknowledgment

We thank the midwife research assistants (L. Douhaud, S. Bedel, B. Lortholary, S. Gabriel, M. Rogeon, and M. Malinbaum) for data collection and to P. Lavoine for checking, coding, and data entry.

The EDEN study group includes M.A. Charles, M. de Agostini, A. Forhan, B. Heude, P. Ducimetière, M. Kaminski, M.J. Saurel-Cubizolles, P. Dargent, X. Fritel, B. Larroque, N. Lelong, L. Marchand, C. Nabet, I. Annesi-Maesano, R. Slama, V. Goua, G. Magnin, R. Hankard, O. Thiebaugeorges, M. Schweitzer, B.

References (44)

  • Z.J. Andersen et al.

    Ambient air pollution triggers wheezing symptoms in infants

    Thorax

    (2008)
  • N. Baïz et al.

    Maternal exposure to air pollution before and during pregnancy related to changes in newborn's cord blood lymphocyte subpopulations. The EDEN Study Cohort

    BMC Pregnancy Childbirth

    (2011)
  • T.F. Bateson et al.

    Children's response to air pollutants

    J Toxicol Environ Health A

    (2008)
  • L. Bråbäck et al.

    Does traffic exhaust contribute to the development of asthma and allergic sensitization in children: findings from recent cohort studies

    Environ Health

    (2009)
  • M. Brauer et al.

    Air pollution from traffic and the development of respiratory infections and asthmatic and allergic symptoms in children

    Am J Respir Crit Care Med

    (2002)
  • D.J. Carruthers et al.

    Use and validation of ADMS-Urban in contrasting urban and industrial locations

    Int J Environ Pollut

    (2000)
  • CERC

    ADMS-Urban: urban air quality management system. User guide

    (2001)
  • B. Clarisse et al.

    Bronchial obstructive phenotypes in the first year of life among Paris birth cohort infants

    Pediatr Allergy Immunol

    (2009)
  • N.A. Clark et al.

    Effect of early life exposure to air pollution on development of childhood asthma

    Environ Health Perspect

    (2010)
  • P. Drouillet et al.

    Maternal fatty acid intake and fetal growth: evidence for an association in overweight women. The ‘EDEN Mother–Child’ cohort (study of pre- and early postnatal determinants of the child's development and health)

    Br J Nutr

    (2009)
  • U. Gehring et al.

    Traffic-related air pollution and respiratory health during the first 2 yrs of life

    Eur Respir J

    (2002)
  • U. Gehring et al.

    Traffic-related air pollution and the development of asthma and allergies during the first 8 years of life

    Am J Respir Crit Care Med

    (2010)
  • Cited by (15)

    • Indoor home environments of Danish children and the socioeconomic position and health of their parents: A descriptive study

      2022, Environment International
      Citation Excerpt :

      While we cannot exclude the possibility that the relationship between mould and affective disorders could be due to reverse causation or a bidirectional relationship, as mould has been associated with induced cognitive effects (Harding et al., 2020), a causal relationship with modified behaviours among those with affective disorders as a mediating path may also explain our findings. Cats and dogs were more common in the DNBC (51 %) than in other European birth cohorts, such as PIAMA (48%), MAAS (37%), GINI/LISA North (20%), BAMSE (15%) and GINI/LISA South (10%) (Aguilera et al., 2013; Jones et al., 2011; Mölter et al., 2015; Pedersen et al., 2013; Riaz et al., 2018). Pet ownership, of dogs and cats, or other pets, was consistently associated with lower socioeconomic position and parental diabetes, cardiometabolic disease, or affective disorders, whereas the proportion of pet ownership was lower among those with parental asthma or allergies and atopy.

    • A review on the role of dispersion and receptor models in asthma research

      2021, Environmental Pollution
      Citation Excerpt :

      For instance, in Modig and Forsberg (2007), the population that reported constant lightweight vehicle traffic outside their homes had higher OR (up to 1.38) of reporting asthma symptoms and for heavyweight vehicles OR 2.38. These results corroborate with Pedersen et al. (2013) work in which the area of exposure addressed by the ADMS-Urban DM also appeared to change the OR value. On the other hand, two other studies point to gender differences.

    • 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?

      2015, Environmental Research
      Citation Excerpt :

      Most studies investigating the impact of air pollution on PTB have been conducted in urban areas, although rural areas are characterized by different levels and different mixtures of air pollutants. Population characteristics, including health behaviors (diet, smoking status), socioeconomic (occupation, educational level) and psychosocial factors (stress, social support), and environmental and social characteristics of places (neighborhood deprivation, health care access, traffic noise, natural spaces) may also vary along the urban–rural continuum and increase confounding or mediate bias (Pedersen et al., 2013; Malmqvist et al., 2011). Furthermore, a growing body of evidence indicates that living near green spaces may improve birth outcomes (Nieuwenhuijsen et al., 2013, 2014; Dadvand et al., 2012a, 2012b; Hystad et al., 2014; Markevych et al., 2014; Agay-Shay et al., 2014) and more specifically decrease the risk of PTB (Hystad et al., 2014).

    View all citing articles on Scopus
    View full text