Can the causal contingent common-pathway model of substance-use initiation and disorder be applied to genotyped samples of unrelated participants?: Results of a Monte Carlo simulation

When applied to the classical twin design, the causal contingent common-pathway (CCC) model (Kendler et al. 1999, Psychological Medicine 29; Kendler 2001, Arch Gen Psychiatry 58) estimates genetic and environmental... [ view full abstract ]

When applied to the classical twin design, the causal contingent common-pathway (CCC) model (Kendler et al. 1999, Psychological Medicine 29; Kendler 2001, Arch Gen Psychiatry 58) estimates genetic and environmental contributions to risk of substance-use initiation and substance-use disorder, taking into account the fact that initiation is a necessary precondition for substance-use disorder. Theoretically, the CCC model could also be applied to genotyped samples of unrelated participants, via genomic-relatedness-matrix restricted maximum likelihood (GREML).  To assess whether such an analysis is realistic in practice, we present results from Monte Carlo simulations in which the CCC model is fitted in OpenMx to simulated genotypic and phenotypic datasets.  In order to avoid calculating intractably high-dimensional multivariate-normal probability integrals, these simulations treated initiation as though it were a continuous variable (even though it was in fact a dichotomous threshold trait) and transformed its heritability estimates from the observed scale to the liability scale (Dempster & Lerner 1950, Genetics 35). Results clearly showed that this approximation made the causal path from initiation to exposure virtually empirically unidentified, even in very large samples of simulated twin pairs.  Alternatives approaches were discussed, such as initiation variables with more than one level of initiation, and imputation of latent liability for initiation (Hayeck et al. 2015, AJHG 96; Weissbrod et al. 2015, Nature Methods 12).