Bias correction of estimates of familial risk from population-based cohort studies

In addition to guiding molecular epidemiology investigations, estimates of the increased risk of disease in relatives of affected persons are also important for screening and counselling decisions. Since precise estimation of such familial risks (FRs) requires large sample sizes, many of the estimates in common use have been obtained from historical electronic records of disease in entire populations, where the relatives of affected and unaffected persons are compared. These estimates may be biased due to failure to identify relatives as affected if they are diagnosed before the start-up date of disease registration.

This article presents a method for correcting the bias in FR estimates from such misclassification of family history, using a simple formula that depends on the prevalence and sensitivity of the observed family history. The sensitivity is estimated by using the R package poplab to create realistic populations of related individuals and then imposing the start-up effect of disease registration.

For a range of FRs, the truncation of family history is demonstrated to result in non-differential misclassification, and sensitivity that has little or no dependence on the FR. The bias is most pronounced for high FRs and for registers with a short life span, and increases with the age of the study cohort. In all the situations studied, the bias-corrected estimates are in excellent agreement with the true values.

In summary, our method can correct the inevitable bias in FRs induced by using electronic population data, and is a feasible alternative to the use of validation samples.