School Seminars and Colloquia

Smoking, Twins and The Heritability of Binary Traits

Statistics Seminar

by Graham Byrnes, School of Population Health, Melbourne


Institution: University of Melbourne
Date: Thu 2nd August 2007
Time: 12:00 PM
Location: Room 213, Richard Berry Building, Uni of Melbourne

Abstract: There is currently a concerted international effort to find genetic risk factors for many common diseases, due to the convergence of several technologies that make this feasible (although still very expensive). However before investing the several million dollars required to carry out a genome-wide scan, it’s as well to check for some evidence that the disease in question has a genetic component. A classical approach to this problem is a twin study, where the correlation between identical (MZ) and fraternal (DZ) twins is compared.


However the results are not as easily interpreted as some have suggested. For example, there have been many studies in twins of the heritability of either initiation of smoking, or of regular tobacco use. Estimates of 75% heritability in males have resulted from large studies (ie Lessov-Schlegger et al, Int J Epidemiol 2006). This has been used to argue for a genome-wide scan for a tobacco-use gene (eg Vink et al, Behav. Genet. 2006) or suggest “a role for genetics…. to identify novel targets for treatment” (Tyndale, Ann Med. 2003). Yet in a study of twins reared apart and together, Kendler et al (Arch Gen Psychiatry 2000) estimate that “In women born before 1925… resemblance was environmental in origin… For women born after 1940, heritability of [regular tobacco use] was similar to that seen in men (63%).” This could be interpreted to mean that heritability is high only if the environment does not vary! Moreover, such heritability estimates depend on untestable modelling assumptions.


This talk will provide an overview of the use of variance components models to estimate heritability of normally distributed quantitative traits from twin studies, then discuss the difficulties of extending the methodology to binary traits. We examine data on smoking behaviour in a longitudinal cohort of twins. Rather than attempt to estimate heritability, we developed a method of estimating the odds of greater than expected concordance of smoking behaviour due to various covariates. These included zygosity (identical vs fraternal twins), but also frequency of contact, age and similarity of smoking behaviour of friends. Zygosity was found to be a predictor of concordance, but only during teenage years. Similarity of smoking behaviour of friends was far stronger and consistent across survey waves, which suggests that social factors may be the most important target for smoking prevention.

For More Information: Dr. Owen Jones O.Jones@ms.unimelb.edu.au