Objectives We considered a problem common to occupational epidemiology, where cumulative exposure is the true dose metric for disease but investigators are only able to measure duration of exposure.

Methods and Results We considered the problem from the theoretical perspective and explored our results in simulations of an occupational cohort of medium size. The duration of exposure is related to cumulative exposure by measurement error with some properties of Berkson-type error. This arises because cumulative exposure = duration*intensity and can be re-written as true = observed*error, with error term having distribution of average long-term exposure intensity for a worker. When duration and intensity are independent, the theory predicts that fitting duration instead of cumulative exposure will not inflate probability of type-I error under the null hypothesis. However, when there is an association between cumulative exposure and the outcome, loss of power to detect an association is expected. In practice, data do not always conform to assumptions made in the theoretical study. We confirmed these predictions in a simulation study for a cohort of 1000 workers with rare outcome in unexposed and with varying correlation of intensity and duration. We first analysed the data using logistic regression models including metrics of exposure as continuous variables. We then explored the situation where exposure groups are formed using quartiles of observed exposure metrics among “cases” and odds in the highest quartile are compared to the lowest. Patterns observed in both analyses were consistent with those expected from theory.

Conclusions Epidemiologists should be more confident in interpreting positive results that arise from use of duration of exposure in lieu of true dose metrics when it is cumulative exposure because type-I error remains at nominal values. The interpretation of null associations remains difficult due to loss of power.