Bias from matching on age at death or censor in nested case-control studies.
Hein-MJ; Deddens-JA; Schubauer-Berigan-MK
Am J Epidemiol 2008 Jun; 167(11)(Suppl):S112
Background: incidence density sampling in nested case-control studies frequently matches controls to cases based on attained age because age is one of the most important risk factors for mortality. Additional matching criteria such as sex, race or year of birth have been used in order to reduce confounding. Recently it was suggested that an additional matching criterion be age at death or censor with eligible controls having an age at death or censor within a specified number of years of the case's age at death. Methods: we used simulated occupational cohorts to evaluate the potential for this additional matching criterion to bias the estimated hazard ratio (HR) from Cox proportional hazards regression. An additional objective was to investigate issues related to exposure lagging; therefore, we compared estimated HRs from analyses including and excluding lagged out workers. Results: the use of the additional matching criterion resulted in downwardly biased HR estimates. In these simulations, when risk was related to a lagged cumulative exposure, the estimated HRs from analyses including lagged out workers (assigned zero lagged cumulative exposure) were generally similarly or less biased than estimated HRs from analyses excluding lagged out workers. Conclusion: incidence density sampling with matching based on attained age plus age at death or censor introduces bias and is not recommended for nested case-control studies. The observed bias was not unexpected since information used to select controls must be known at the time of selection. Since there can be no way of predicting the age of death or censor for the controls, this information cannot be used for selecting controls in incidence density sampling.
Workplace-studies; Statistical-analysis; Sample-preparation; Sampling; Age-groups; Control-methods
American Journal of Epidemiology; 41st Annual Meeting Society for Epidemiologic Research Chicago, Illinois, June 24-27, 2008