Re: Risks of cancer and families.

We have noted with interest the editorial ( 1 ) that Marvin Zelen wrote on our article ( 2 ) . We do not think that the referred “ length-biased sampling, ” i.e., the sampling of families based on their size, particularly in case – control studies ( 3 ) , is relevant to our study. It is true that large families contributed more case patients to the analysis than did small families. In the complete Swedish Family Cancer Database, 29.1% of the families had three or more offspring; the proportion was 45.6% for families with at least one affected offspring. However, our study was based on the whole population, not on samples thereof, and the relative risks were estimated between the fi rst diagnosis and any subsequent cancer in the families. The design would be liable to bias only if cancer risks in small families were different from those in large families, which is not the case for any neoplasm we have studied ( 4 ) . To prove our point, we adjusted the data additionally for family size (families with 2, 3, 4, 5, or ≥ 6 offspring); the other adjustments were as in the original study, and they included age, sex, and calendar year. We recalculated all relative risks, and the results show that the adjustment for family size has no effect. As an example, Table 1 shows the estimated relative risks of invasive cancer for siblings of the probands, with and without consideration of family size. Controlling for family size resulted in very small differences between the two estimates. This can be best assessed for “ any site ” : Two relative risks differed less than 0.8%; the other four were