Common referent versus shifting referent methods when using case-control data to examine patterns of incidence across multiple exposure variables

In case-control studies, when odds ratios estimate rate ratios or risk ratios, trends in odds ratios can reflect trends in incidence in the underlying population.

In this report, we use an empirical example and hypothetical data to compare two methods ("common referent" and "shifting referent") for calculating odds ratios from case-control data across multiple exposure variables. With common referent methods, a single independent variable is constructed from the joint distribution of all exposure variables. With shifting referent methods, each stratum-specific trend has its own referent.

Using examples involving a single ordinal "trend" variable and a dichotomous "stratification" variable, we show that common referent odds ratios will reflect the pattern of incidence in the underlying population, whereas odds ratios calculated using shifting referent methods will not.