Comparison of Methods for Identifying Phenotype Subgroups Using Categorical Features Data With Application to Autism Spectrum Disorder

Abstract

We evaluate the performance of the Dirichlet process mixture (DPM) and the latent class model (LCM) in identifying autism phenotype subgroups based on categorical autism spectrum disorder (ASD) diagnostic features from the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition Text Revision. A simulation study is designed to mimic the diagnostic features in the ASD dataset in order to evaluate the LCM and DPM methods in this context. Likelihood based information criteria and DPM partitioning are used to identify the best fitting models. The Rand statistic is used to compare the performance of the methods in recovering simulated phenotype subgroups. Our results indicate excellent recovery of the simulated subgroup structure for both methods. The LCM performs slightly better than DPM when the correct number of latent subgroups is selected a priori. The DPM method utilizes a maximum a posteriori (MAP) criterion to estimate the number of classes, and yielded results in fair agreement with the LCM method. Comparison of model fit indices in identifying the best fitting LCM showed that adjusted Bayesian information criteria (ABIC) picks the correct number of classes over 90% of the time. Thus, when diagnostic features are categorical and there is some prior information regarding the number of latent classes, LCM in conjunction with ABIC is preferred.

Figure A.5

Heat Map for the five class DPM. The level of dark color indicates the severity of the classes where (least affected is less dark and (highly affected is the darkest. The number of people assigned to the cluster is proportional to the size of the heat map box

Figure 1

Information criteria and deviance for the latent class analysis of the four and five class simulated datasets. Y-axis is value of the IC and x-axis is the number of estimated latent classes.

Figure 2

Distribution of the estimated number of classes by Information criteria for the four and five class simulated datasets. The X axis shows the counts of the estimated number of clusters as a function of the true simulated number of clusters (Y-axis) for different types of information criteria (legend).

Figure 3

Box and whisker plots of the Rand indices between the estimated and true (simulated) cluster partitions, computed for each LCM and DPM analysis across the four and five class simulated datasets. Higher values indicate better classification.

Figure 4

Heat Map for the five class LCM. The level of dark color indicates the severity of the classes where (least affected is less dark and (highly affected is the darkest. The number of people assigned to the cluster is proportional to the size of the heat map box