Assessment of control selection bias in a hospital-based case-control study of upper aero-digestive tract cancers.

Background: Most of the evidence identifying determinants of UADT cancer risk comes from hospital-based case-control studies. However, such studies are thought to be influenced by selection bias given the extent of association of tobacco smoking and alcohol drinking with other diseases in the general population. We assessed the extent of selection bias due to inclusion of controls with tobacco or alcohol related diseases in a hospital-based case-control study of upper aero-digestive tract (UADT) cancers in Brazil.

Methods: Risk factor information was obtained from 784 cases and 1568 non-cancer controls. Causes of hospitalization for controls were scored according to their likelihood of association with tobacco and alcohol. Odds ratios (OR) were calculated by conditional logistic regression with and without exclusion of controls with tobacco or alcohol related diseases.

Results: The OR for smoking slightly increased after exclusion of controls with tobacco-related diseases, from 15.3 (95% confidence interval [CI]: 9.7, 24.2) among heaviest smokers (pack-years > 91) to 18.6 (95%CI: 11.4, 30.2). The OR for alcohol consumption remained unchanged after exclusion of controls with alcohol related diseases. Associations of low magnitude for some dietary variables (citric fruits, carotene-rich foods, spicy foods, maté tea) and wood stove use were not appreciably affected by the exclusion of controls.

Conclusion: The bias introduced by controls with tobacco and alcohol related diseases is small, when these exposures are investigated as determinants, and almost negligible when considered as confounders of the association between other factors and UADT cancers.