CONTRASTING EVIDENCE WHEN USING HOSPITAL OR POPULATION CONTROLS: THE EXAMPLE OF THE ASSOCIATION BETWEEN EXPOSURE TO GASOLINE AND DIESEL EXHAUST, AND LUNG CANCER

Abstract

ISEE-407 Introduction: Much of the evidence for an association between exposure to gasoline and diesel exhausts and lung cancer comes from occupational case-control studies. The choice of controls is critical in providing valid estimates. There is no such thing, however, as a perfect control group. Although hospital controls may be less subject to reporting bias and show higher response rates than population controls, diseases in the control pool may themselves be associated with the exposures of interest, leading to biased estimates. Do the two types of control groups tend to give similar answers? Methods: In the 1980s, we conducted a large population-based case-control study to assess the role of occupational exposures and circumstances on cancer incidence in Montreal. Incident cases from all area hospitals were ascertained. We focus here on the results from an in-depth analysis of our data on gasoline and diesel exhausts and lung cancer. For this analysis, we used 857 lung cancer cases and two distinct control series: one consisted of 533 controls from the general population, and the other, comprising 1349 patients with cancers at sites other than the lung. All subjects were interviewed to obtain a detailed job history and relevant data on potential confounders. A team of chemists and hygienists translated each job into a list of potential exposures. Analyses were carried out for exposure to gasoline and diesel exhausts, as well as for occupations presumed to have entailed exposure to those agents. Results: The odds ratios (OR) for lung cancer associated with nonsubstantial and substantial exposure to gasoline exhaust were 0.9 and 0.9, respectively, using either population or cancer controls. However, for diesel engine emissions, the two control groups yielded somewhat different estimates. Using population controls, the OR was 1.1 [95% confidence intervals (95% CI): 0.7-1.7] for nonsubstantial exposure, and 1.6 (95% CI: 0.9-2.8) for substantial exposure. Using cancer controls, the corresponding values were 1.0 (95% CI: 0.7-1.4) and 1.0 (95% CI: 0.7-1.5). Discussion: Few studies offer the opportunity to contrast results obtained with several control groups. Although they may be difficult to reconcile, our results underline the challenge in obtaining both valid occupational exposure information and representative control subjects.