American Journal of Respiratory and Critical Care Medicine

To the Editor:

In a recent pooled analysis of case-control studies, Olsson and coworkers concluded that they found a consistent association between occupational exposure to diesel motor exhaust (DME) and increased risk of lung cancer (1). The authors have pointed out it is unlikely that this association can be explained by bias or confounding. However, in a preliminary report of this pooled analysis it has been shown that additional adjustment for education halved the estimate of the excess relative risk (2). Unfortunately, the effect of educational attainment has not been further discussed in the final article.

Using meta-regression models for the individual studies pooled in the above-mentioned paper (from Table 1 and Figure 1), it appears that the response rate among controls has a significant impact on the corresponding risk estimate (Figure 1). Assuming a complete response among population controls under this model, the lung cancer risk of people exposed to DME is no longer significantly elevated. What could be the reason for such a relationship?


Educational AttainmentExposure Prevalence (%)Distribution of Controls (%)
Without any formal vocational training24.710.0
Finished vocational training13.068.6
University degree3.721.4

Based on data from Reference 6.

It is well known that the lower the response rate, the higher the likelihood of a biased sample. However, little is known about the magnitude of this type of bias. In a survey of the working population on qualification and working conditions in Germany, conducted in 2005/2006 by the Federal Institute for Vocational Education and Training (BIBB) and the Federal Institute for Occupational Safety and Health (BAuA), data from 20,000 individuals of the German active labor force were ascertained by telephone interview (CAPI). To attain a sample comparable to the German population according to the microcensus, a special drawing mechanism was implemented to account for the expected higher nonresponse of lower-educated workers (manual workers) (3). The overall response rate was reported to be 44%. The detailed data from the field report (4) enable one to calculate crude estimates of the response rates in the different subgroups: among manual workers, the response rate was calculated to be about 25%, whereas among higher-educated employees, it was about 50% (our own calculations).

Assessing the potential bias in risk estimates due to a biased sample of controls, we should take into account that educational attainment is strongly related to the prevalence of exposure to most occupational hazardous substances. In the original analyses of two German studies that are included in the pooled analysis, it was shown that controls without any formal vocational training were exposed to DME more than four times more often than were controls with a university degree (Table 1). Therefore, if manual workers are underrepresented in the sample of controls, it follows that the exposure prevalence in the control group underestimates the true exposure prevalence in the reference population. Consequently, the risk associated with a certain exposure is overestimated.

Decreasing response rates, especially among population controls, seem to increasingly affect epidemiological research. In a recent comment on this topic, it was recommended that population controls be compared with census data to ensure that the sample of controls truly represents the reference population (5).

Comparing study data with census data for the two German studies mentioned above, it again appears that subjects with lower educational attainment are underrepresented in the control group (Table 2).


Educational AttainmentObserved Distribution of Controls (%)Expected Distribution of Controls* (%)
Without any formal vocational training10.016.9
Finished vocational training68.671.5
University degree21.411.6

*Based on German microcensus 1991.

Therefore, adjustment for educational attainment seems to be a useful and in some cases even required approach to reduce the impact of selection bias in occupational population-based case-control studies.

Moreover, a comparison between control group and census data in terms of years of formal education and/or educational attainment could help to describe the degree of representativeness of the control group.

1. Olsson AC, Gustavsson P, Kromhout H, Peters S, Vermeulen R, Brüske I, Pesch B, Siemiatycki J, Pintos J, Brüning T, et al.. Exposure to diesel motor exhaust and lung cancer risk in a pooled analysis from case-control studies in Europe and Canada. Am J Respir Crit Care Med 2011;183:941948.
2. Straif K, Olsson AC, Gustavsson P, Kromhout H, Peters S, Vermeulen R, Brüske I, Pesch B, Brüning T, Kendzia B, et al.. Diesel motor exhaust and lung cancer risk in SYNERGY: pooled analysis of case-control studies on the joint effects of occupational carcinogens in the development of lung cancer [abstract]. Occup Environ Med 2010;60:A25.
3. Rohrbach-Schmidt. 2009 BIBB-BAuA Surveys of the Working Population on Qualification and Working Conditions in Germany. BIBB-FDZ Daten- und Methodenberichte 1/2009 (accessed September 15, 2011). Available from:
4. Hartmann J. BIBB/BAUA-Erwerbstätigenbefragung 2005/2006 - Feldbericht, tns-Infratest München, 2006 (accessed September 15, 2011). Available from:
5. Sneyd MJ, Cox B. Commentary: Decreasing response rates require investigators to quantify and report the impact of selection bias in case-control studies. Int J Epidemiol 2011;40:13551357.
6. Möhner M, Pohlabeln H, Jöckel KH, Wichmann HE. Lungenkrebsrisiko durch berufliche Exposition: Weiterführende Ansätze zur Modellierung und Adjustierung. In: , Jöckel KH, Brüske-Hohlfeld I, Wichmann HE, editors. Lungenkrebsrisiko durch berufliche Exposition. Landsberg: ecomed; 1998. pp. 240251.

Author disclosures


No related items
American Journal of Respiratory and Critical Care Medicine

Click to see any corrections or updates and to confirm this is the authentic version of record