Residual confounding in follow-up studies with time dependent doses

Follow-up studies of populations with exposures accumulated over a period of years are subject to major confounding by time and age. The longer the period that doses accumulate, the greater the potential for bias due to confounding. Examples of sex- and cause-specific maximum likelihood estimates of the mortality rate ratio are compared for radiation workers while stratifying by median age and time period, trichotomizing on age and time, and using 10- and 5-year age and time strata. Point estimates were 5.47 for dichotomized strata, 2.74 for trichotomized strata, 2.57 for 10-year strata, and 3.72 for 5-year strata. Stratification by induction time also influenced residual confounding. Estimates dichotomized on age and time were 3.32, 4.64, and 6.47 for 2, 5, and 10 years induction, respectively. When 5-year age and time strata were used, the estimates for 2, 5, and 10 years induction times, respectively, were 2.63, 2.08, and 3.72. These findings indicate that residual confounding for age and time can be a major source of bias in follow-up studies where doses are accumulated over a long periods of time, inclusion of varying induction times may introduce additional residual confounding.