Age-at-interview bias in anticipation studies: computer simulations and an example with panic disorder

The phenomenon of anticipation has received considerable interest recently, especially among researchers investigating the genetics of psychiatric disorders. Anticipation can involve an earlier age at onset, greater severity, and/or a higher number of affected individuals in successive generations within a family. There is some controversy concerning detection of age-at-onset anticipation, due to problems of sampling bias, which may account for the phenomenon by preferentially sampling either lateronset parents or earlier-onset children. One source of bias that has not been explicitly investigated is differential age at interview between parent and child, such that parents have passed through more of the risk period than their offspring. We conducted a computer simulation study of affected parent-child pairs to determine whether, for realistic age-at-onset and age-at-interview distributions, this source of bias is a serious one. Our results show that the timing of diagnostic assessment can strongly affect the ascertainment of parent-child pairs, to produce a severely biased sample exhibiting apparent anticipation. Under realistic assumptions, an investigator may face a greatly increased risk of false positives (i.e. detecting anticipation when none exists). For example, a nominal 5% significance level may correspond to true p values as high as 50% or even approaching 100%. We conclude with an application to existing data on panic disorder.