Log-linear models for cancer risk among migrants.

Epidemiological studies of migrants have played an important role in separating the environmental and genetic components of cancer aetiology. The statistical analyses of these studies have typically involved calculating age- and sex-adjusted rates by country of birth. We describe methods which permit the effect of duration of residence in the host country to be estimated after adjusting for other temporal effects such as age, and period or cohort. The methods, which are based on log-linear modelling, can also be applied in a case-control manner if appropriate denominator populations are unavailable. Examples are given in which the more traditional methods of analysis may have yielded misleading results.