Objective: To investigate properties of population attributable risk (PAR) estimates, when its components come from different sources. Examples include situations where one requires local estimates of PAR for a study subset (e.g., in states or counties within a national study) or if one wishes to apply the findings of an epidemiologic study to another population.
Study design and setting: A framework for estimating local PAR values is developed, and then illustrated using synthetic and empirical data.
Results: A general expression for the variance of a local PAR estimate is formulated. It involves three components, reflecting (1) the variance of the disease relative risk associated with exposure to a risk factor, (2) the variance of the exposure prevalence (P), and (3) their covariance. The effects of variable stratum sizes, case-control sample size ratios, and variation in exposure P are illustrated by some synthetic scenarios, and with data from an international case-control study of heart disease.
Conclusion: The precision of local PAR estimates can be considerably improved by incorporating external data, as opposed to limiting the calculation to data only from the local population. In some cases, variation in local PAR estimates largely reflects uncertainty in the local estimate of exposure P.