BACKGROUND: Uncontrolled confounding by personal exposures like smoking can limit the inferential power of occupational cohort studies. We developed and demonstrate a refinement of an existing type of sensitivity analysis, indirect adjustment, for evaluating the potential magnitude of confounding by alcohol and tobacco. Results of a large retrospective cohort study of laryngeal cancer and exposure to metalworking fluids (MWF) are used to illustrate the methods. METHODS: Data on smoking and drinking habits representative of the study cohort were obtained from a sample of US manufacturing workers from the 1977 National Health Interview Survey (NHIS). Two different mechanisms were assumed to affect the distribution of confounding factors between MWF exposure groups: socially determined and chance differences. Chance variation was investigated with Monte Carlo sampling from the NHIS survey distribution of smoking and drinking. An upper bound on systematic differences in smoking and drinking was set by assuming that differences between exposure groups within the same unionized blue collar workforce were very unlikely to be larger than differences between blue and white collar manufacturing workers in the NHIS data. RESULTS: Under plausibly large differences in smoking and drinking habits among MWF exposure groups occurring by either mechanism, the exposure-risk association was unlikely to have been over- or under-estimated by as much as 20%. CONCLUSIONS: When comparing exposure groups within the same working population, it is unlikely that either systematic or chance differences in smoking and drinking habits will cause as much as a 20% change in the relative risk in large studies. While this study focused on an occupational exposure and laryngeal cancer, there are many situations in which epidemiologists are concerned that unmeasured 'lifestyle factors' may differ among exposure groups, and it would appear that the likely confounding effect of such differences will often be modest.
BACKGROUND: Uncontrolled confounding by personal exposures like smoking can limit the inferential power of occupational cohort studies. We developed and demonstrate a refinement of an existing type of sensitivity analysis, indirect adjustment, for evaluating the potential magnitude of confounding by alcohol and tobacco. Results of a large retrospective cohort study of laryngeal cancer and exposure to metalworking fluids (MWF) are used to illustrate the methods. METHODS: Data on smoking and drinking habits representative of the study cohort were obtained from a sample of US manufacturing workers from the 1977 National Health Interview Survey (NHIS). Two different mechanisms were assumed to affect the distribution of confounding factors between MWF exposure groups: socially determined and chance differences. Chance variation was investigated with Monte Carlo sampling from the NHIS survey distribution of smoking and drinking. An upper bound on systematic differences in smoking and drinking was set by assuming that differences between exposure groups within the same unionized blue collar workforce were very unlikely to be larger than differences between blue and white collar manufacturing workers in the NHIS data. RESULTS: Under plausibly large differences in smoking and drinking habits among MWF exposure groups occurring by either mechanism, the exposure-risk association was unlikely to have been over- or under-estimated by as much as 20%. CONCLUSIONS: When comparing exposure groups within the same working population, it is unlikely that either systematic or chance differences in smoking and drinking habits will cause as much as a 20% change in the relative risk in large studies. While this study focused on an occupational exposure and laryngeal cancer, there are many situations in which epidemiologists are concerned that unmeasured 'lifestyle factors' may differ among exposure groups, and it would appear that the likely confounding effect of such differences will often be modest.
Authors: David B Richardson; Dominique Laurier; Mary K Schubauer-Berigan; Eric Tchetgen Tchetgen; Stephen R Cole Journal: Am J Epidemiol Date: 2014-09-21 Impact factor: 4.897
Authors: Melissa C Friesen; Paul A Demers; John J Spinelli; Ellen A Eisen; Maria F Lorenzi; Nhu D Le Journal: Am J Epidemiol Date: 2010-08-11 Impact factor: 4.897
Authors: Elizabeth M Allen; Bruce H Alexander; Richard F MacLehose; Heather H Nelson; Gurumurthy Ramachandran; Jeffrey H Mandel Journal: Ann Epidemiol Date: 2015-08-29 Impact factor: 3.797
Authors: James C Helmkamp; Jennifer E Lincoln; John Sestito; Eric Wood; Jan Birdsey; Max Kiefer Journal: Am J Ind Med Date: 2012-12-19 Impact factor: 2.214
Authors: Eva Andersson; Bodil Persson; Ing-Liss Bryngelsson; Anders Magnuson; Håkan Westberg Journal: Int Arch Occup Environ Health Date: 2009-07-28 Impact factor: 3.015