19841 recently presented a segregation analysis of 200 families of breast cancer patients published some 40 years ago. Unfortunately, in their discussion of work done by others Go et al, 1983; Cleton et al, 19831, they made some statements that could be misunderstood. It is our purpose here to clarify the principles and issues involved.
When pedigrees are selected for linkage analysis because many cases of disease are present, an appropriate genetic model is first obtained by segregation analysis of those pedigrees, either individually or in clusters with similar epidemiologic characteristics. In any segregration analysis estimates of susceptibility gene frequencies based on high-risk families alone (without additional prevalence or incidence information) will not accurately reflect the gene frequencies in the population at large from which the families were selected. Thus Williams and Anderson [1984] used an approximate ascertainment correction and incidence data reported by Clemmesen [1976] to obtain gene frequencies relevant for that population. Elston et a1 [1981] and Go et a1 [1983], on the other hand, did not attempt any ascertainment correction in their analysis of a few large breast cancer pedigrees, so the resulting gene frequency estimates are relevant only to the set of individuals marrying into the particular pedigrees they analyzed. In linkage analysis, one should employ the best-fitting segregation model for the high-risk pedigrees while allowing for a low population frequency of the susceptibility allele. This is what King et al [1983] did (in fact, choosing frequencies of .01 and .001) in their linkage analysis of these pedigrees, verifying that the choice of estimates, provided that they were small, did not critically affect the resulting lod scores.
Whereas the failure to provide for an ascertainment correction has a large effect on gene frequency estimates, which must be carefully allowed for in any subsequent linkage analysis, experience to date suggests that when large pedigrees are involved its effect on the estimation of other parameters of the genetic model is much smaller; certainly in the case of rare dominant diseases the effect would appear to be negligible