Article Text

Download PDFPDF

Self report and GIS based modelling as indicators of air pollution exposure: is there a gold standard?
  1. F Forastiere1,
  2. C Galassi2
  1. 1Department of Epidemiology, Rome E Health Authority, Rome, Italy
  2. 2Cancer Epidemiology Unit, CeRMS and Center for Cancer Prevention, University of Turin, Italy
  1. Correspondence to:
 Dr F Forastiere
 Department of Epidemiology, Rome E Health Authority, Via Santa Costanza, 53, 00198 Rome;

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Commentary on the paper by Heinrich et al (see page 517)

Exposure to traffic generated pollutants, especially among people living along busy roads, has been associated with increased risk of respiratory disorders among children and adults as well as with overall mortality.1–5 Simple exposure indicators (self report, distance from pollutant sources, traffic density) and more complex and integrated models that take into account demographic factors and land use by means of geographic information system (GIS) based technology6–8 have been employed in epidemiological studies.

As has happened in the past in occupational epidemiology, environmental epidemiologists are trying to develop more accurate methods for assessing air pollution exposure.9 As a result, comparing the different methods available is of great interest. In this issue, Heinrich et al have compared parental report of traffic intensity near homes (cars, trucks, buses, and mopeds on the street of residence and other nearby streets) with a combination of air pollution measures (fine particle mass, filter absorbance, and nitrogen dioxide) and GIS based modelling in two different European locations (the Netherlands and Munich, Germany).10 They found that predicted exposure estimates for air pollutants increased with self reported traffic level in Munich and in urban areas of the Netherlands. However, agreement rates between the two methods were relatively low for all three pollutants studied, and subjective assessments consistently overestimated fine particulate matter and nitrogen dioxide, in both locations and particularly in rural areas. The authors underline the limited validity of self reported traffic intensity measures.

From the results of this study should we conclude that subjective measures are bad and should not be used any more? Are the “objective” GIS based modelling methods the gold standard to recommend? Let us look at the main points.

The best way to learn about a specific fact, at least in theory, is to ask the most informed subject—that is, the person who is experiencing the exposure at issue, especially when his judgement can integrate several aspects like exposure intensity, distance from the source, and protective and remedial measures. Several well designed case-control studies in the occupational setting have used self reported exposures to toxic substances for this purpose.11 Beyond real exposure, however, self report is influenced by several other factors that may vary from one subject to another: individual perception, annoyance, socioeconomic and cultural background, and perhaps environmental “worry”. The propensity to report more illnesses and symptoms as a result of proximity to a potential hazard, in the absence of a measurable biological effect, has been named “awareness bias”.12 Moffat and colleagues12 showed the influence of such awareness bias in two epidemiological studies conducted in communities exposed to emissions from heavy industry. The influence of the disease status on the report of exposure has not been described, to our knowledge, in air pollution studies. The possibility of this bias makes the use of self report problematic, however, unless there is special care taken to standardise information collection, to use simple and easy to understand terms, to employ relative or “objective” benchmarks, and to evaluate repeatability of the questionnaire.

The need to apply individual exposure estimates to air pollutants to large study populations has motivated the use of GIS in which geographic data are combined with concentration measurements to estimate exposures for subjects of large populations.6,7 In the original methodological study reported by Brauer and colleagues,8 and described here by Heinrich et al, 40 measurement sites within each of the two study areas provided annual average pollutant concentrations. Regression models predicting these concentrations were developed on the basis of population density and traffic intensity within buffer zones up to 1000 metres. The variability explained from the regression models using the 40 sites were 73% and 56% for PM2.5, respectively for the two locations, and 81% and 67%, for filter absorbance (a marker of diesel exhaust).8 Air pollutant concentrations at the home address of the study subjects were estimated from these models. The performances of the models were good enough to render the method very attractive for exposure assessment. However, it is also evident that the “predicted exposure” for any individual subject in the study is affected by error and cannot be considered “the gold standard”. Moreover, as the authors are well aware, GIS data cannot account for local scale differences when the grid size available is large because traffic related air pollutants exhibit substantial variability at distances of 50 m or less from major roads. Finally, exposures to some primary pollutants emitted from vehicular traffic may be overlooked with this method.

The past debate over exposure assessment in occupational epidemiology has some analogies.9 Work histories and job titles are of limited value in many instances and it may be difficult to obtain unbiased and valid estimation of occupational exposures in large studies relying on questionnaires. Alternative methods have been proposed, such as job exposure matrices (JEMs) and expert assessment. In particular, JEMs have been developed to study cancer, asthma, and, more recently, ergonomic problems. Although there is a large variability in validity and reliability estimates, it seems that self report tends to be a more sensitive method, whereas JEM seems to be more specific. Industrial hygienists, chemists, or engineers may be employed to estimate exposures of workers, with better performance than simple self report. However, there is a tendency towards integrating the methods available. In particular, self report can provide important information for expert judgement. With more complex and integrated exposure assessments a higher degree of reliability is obtained.

In conclusion, the results presented by Heinrich et al are very welcome, although they are not surprising, given the inherent limitations of the compared methods. Since no single method can be considered the “gold standard”, it is clear that there is still room for further development in exposure assessment to air pollutants, possibly by integrating the information available from different sources, including the subjects under study. Such development will lead to more sophisticated approaches but perhaps, as in the occupational setting,13 to indications that simple indicators may be sufficient to obtain a valid study.


We thank Margaret Becker for her editorial help.

Commentary on the paper by Heinrich et al (see page 517)



  • Competing interests: none declared

Linked Articles