Effects of selective return on estimates of heritability for body mass index in the National Heart, Lung, and Blood Institute Twin Study.

In the National Heart, Lung, and Blood Institute Twin Study, body mass index (BMI) was studied at military induction and at three subsequent examinations spanning five decades in a cohort of white, male World War II veterans. At military induction (1940s) and again at the first clinical examination of this study (1969-1973), there was close agreement of three commonly used estimates of heritability (range 0.72 to 0.80), and no evidence of a difference in total variance of BMI between the zygosities. However, at the last two examinations (1980s), the total variance in dizygotic (DZ) twins was significantly greater than that of monozygotic (MZ) twins (P less than 0.01) and these same heritability estimates varied widely. The among-pair estimate of heritability fell to unrealistic negative values, the within-pair estimate rose to values of 1.0 or greater, and the intraclass correlation coefficient estimate was slightly lower than in the entire cohort at baseline. The cause of the unequal zygosity total variance appears to have been nonparticipation at later examinations of MZ twins with extreme values of BMI, with no evidence of a similar selection process influencing DZ twins. This selection process biased the three estimates of heritability, making it difficult to determine which estimate is the most appropriate. Despite these biases, it remains clear that genetic factors contribute substantially to BMI in this population.