Bias from matching on age at death or censor in nested case-control studies

Nested case-control studies frequently use incidence-density sampling based on attained age when matching controls to cases. A recently suggested additional matching criterion is age at death, with eligible controls having an age at death or censor within a specified number of years of the case's age at death. We simulated occupational cohorts with time-dependent exposures to evaluate whether adding this criterion introduces bias, and we investigated alternative methods of treating workers with zero exposure because of latency assumptions (ie, "lagged-out").

We used simulated cohorts to consider null, positive, and negative exposure effects and lag periods of 0 and 10 years. Risk sets were constructed using incidence-density sampling with matching on attained age alone or attained age plus age at death. We estimated exposure effects using conditional logistic regression for unlagged and 10-year lagged cumulative exposure. Lagged-out workers were either excluded or included and assigned zero exposure.

Effect estimates were generally unbiased when controls were selected by matching on attained age alone. However, the estimates were downwardly biased under the additional matching criterion. When risk was related to a lagged cumulative exposure, estimates including lagged-out workers were similarly or less biased than those excluding lagged-out workers.

In these simulations, incidence-density sampling with matching on attained age plus age at death introduced bias. This is because sampled controls were younger at first exposure, with higher cumulative exposure compared with controls selected by matching on attained age alone. Incidence-density sampling with matching on attained age alone (and including lagged-out workers) did not introduce bias.