To optimize the power for genetic analyses, we constructed secondary continuous phenotypes based on the probability of assignment to each of the three latent classes. As a result, information describing each latent class provided an interpretation of the three classes while the probability of being in each class was employed as the phenotype in analysis. This approach avoids multiple pairwise comparisons of the three classes in analyses, avoids misclassification inherent in assigning individuals to a particular latent class, maximizes the number of subjects, and was shown previously to improve power to detect genetic association (Bureau et al., 2011).
