Using Two-Step Cluster Analysis and Latent Class Cluster Analysis to Classify the Cognitive Heterogeneity of Cross-Diagnostic Psychiatric Inpatients

Abstract

The heterogeneity of cognitive profiles among psychiatric patients has been reported to carry significant clinical information. However, how to best characterize such cognitive heterogeneity is still a matter of debate. Despite being well suited for clinical data, cluster analysis techniques, like the Two-Step and the Latent Class, received little to no attention in the literature. The present study aimed to test the validity of the cluster solutions obtained with Two-Step and Latent Class cluster analysis on the cognitive profile of a cross-diagnostic sample of 387 psychiatric inpatients. Two-Step and Latent Class cluster analysis produced similar and reliable solutions. The overall results reported that it is possible to group all psychiatric inpatients into Low and High Cognitive Profiles, with a higher degree of cognitive heterogeneity in schizophrenia and bipolar disorder patients than in depressive disorders and personality disorder patients.

Keywords: cluster analyses; cognitive functioning; latent class cluster analysis; psychiatric inpatients; two-step cluster analysis.

FIGURE 1

Indexes of fit changes obtained from Latent Class cluster analysis and Two-Step cluster analysis for solutions ranging from one to four clusters. The panels show the change in information criterion (left) or entropy (right) between two close clusters’ solutions (e.g., 2vs1 shows two-cluster solution minus one-cluster solution). LCA, Latent Class cluster analysis; TwoStep, Two-Step cluster analysis; BIC, Bayesian information criterion; AIC, Akaike information criterion.

FIGURE 2

Contribution of the single cognitive tests to the clustering solution as reported from the Two-Step (top) and Latent Class cluster analysis (bottom). The top panel shows the index of relative importance of each cognitive test as identified by the Two-Step cluster analysis. The panel on the bottom shows the conditional item response probabilities for the two clusters identified by the Latent Class cluster analysis. Performance class of score below (z score < –1.3) average (z score between –1.3 and +1.3) and above (z score > 1.3) the normative sample. MCST, Modified Wisconsin Card Sorting Test; CPM-47, Raven’s Colored Progressive Matrix; AM, attentional matrices; ToL, tower of London—Drexel University test; STROOP, Stroop Word Interference Test.

FIGURE 3

Cluster assignment according to the Two-Step clustering solution as a function of the predicted probability of cluster membership of the Latent Class clustering solution, on the cross-diagnostic sample and within each diagnosis. The left panel represents the clustering solutions obtained on the cross-diagnostic sample. The panel on the right represents the clustering solutions obtained within each diagnosis. SZ, Schizophrenia Spectrum and Other Psychotic Disorders; BD, Bipolar and Related Disorders; DD, Depressive Disorders; PD, Personality Disorders.