3.1. Do sub-groups of patients exist that significantly differ in valence and arousal ratings?
Visual inspection of the dendogram indicated that a two-cluster solution was reasonable, as 2 distinct clusters could be identified (see ). The selection of the two-cluster solution was supported by analyses indicating good agreement between Ward’s method and the K-Means clustering (kappa = 0.79, p < 0.001). Agreement was less satisfactory for three and four cluster solutions, which also did not have enough cases to be considered meaningful. Support for a 2-cluster solution was also found after clusters were plotted in multidimensional space, as separation among cluster centroids was clear for the two-cluster solution, but not as clear for the three and four cluster solutions. These multiple methods thus converged to indicate that a 2-cluster solution was optimal.
Means and standard errors were calculated for each group to determine the meaning of clusters obtained. Mean valence and arousal ratings of the IAPS stimuli are plotted in . As can be seen in , cluster 1 (n = 30) reflected a group who was relatively normal with regard to momentary affective experience, i.e., valence and arousal ratings made by Group 1 for both pleasant, neutral, and unpleasant images were nearly identical to ratings made by the CN group. Cluster 1 therefore appears to represent a group of schizophrenia patients with relatively normal emotional experience. Cluster 2 (n = 19) reflected a group of patients who differed from controls with regard to momentary valence and arousal ratings, and who reported that negative stimuli were more unpleasant and arousing than controls and other patients. No differences were noted for this cluster with regard to rating the valence or arousal of pleasant stimuli.
Means and Standard Errors for IAPS Valence and Arousal Ratings for Patient Clusters
3.1.3. Are emotional experience sub-groups valid?
When valence and arousal ratings were entered in discriminant function analysis, results indicated that the 2 cluster solution (see ) was adequately separated in discriminant function space. An iterative partitioning method was also applied to examine cluster stability. Results of this iterative method indicated that 95.9% of cases were correctly classified (Wilks’ Lambda = 0.36, p < .001). These results indicate that there is little overlap in valence and arousal ratings made among the patient clusters identified. Thus, results are supportive of the notion that a sub-group of patients exist who experience emotion atypically, even though the majority of patients experience emotion similarly to controls.
Patient 2 Cluster Solution Plotted in Discriminant Function Space
3.2. How do sub-groups differ in their momentary valence-arousal ratings?
Recent studies have suggested that affective disturbance in SZ might not reflect a simple hedonic impairment, but rather a tendency to experience positive stimuli as being aversive (Cohen & Minor, 2010
; Kring & Moran, 2008
; Tremeau et al., 2010). To assess whether this was true of the subjects in our study, we assessed how frequently individuals in the two SZ subgroups and CN subjects reported an emotional response that was unusual (e.g., reported a negative emotional responses to a positive stimulus, or a positive emotional response to a negative stimulus). For this analysis, we assessed the frequency with which subjects reported an emotional response that was consistent with the “normative” response: specifically, we calculated the proportion of items subjects rated that were consistent with normative values for the stimuli based on categories of positive (ratings of 1-3), neutral (ratings of 4-6) and negative (ratings of 7-9) stimuli. Analyses were conducted separately for positive and negative stimuli using the Kruskal-Wallis H test because the % of items rated in each of the categories is not independent.
Statistical analyses focused on the consistency between normative ratings and subject ratings. For “normatively positive” stimuli, results indicated that the 3 groups significantly differed in the % of positive stimuli rated in the unpleasant range of the valence scale (χ2 = 11.88, p = 0.003), but not for the percentage of items rated in the pleasant range or neutral ranges. Post hoc Scheffe tests indicated that cluster 2 rated a significantly greater proportion of positive items in the unpleasant range of the valence scale than either the CN group (p = .007) or cluster 1 (p = .001) (see ).
Means Percentage of Positive items rated in Pleasant, Unpleasant, and Neutral Ranges of the Valence Scale
For “normatively negative” stimuli, groups significantly differed in the % of negative stimuli rated in the unpleasant (χ2 = 8.63, p = 0.013) and neutral ranges (χ2 = 14.07, p = 0.001), but not in the pleasant range. Post hoc Scheffe contrasts indicated that cluster 2 rated a higher proportion of negative items as falling in the unpleasant range and fewer negative items as falling in the neutral range than CN and the emotionally normal patient group (see ).
Means Percentage of Negative items rated in Pleasant, Unpleasant, and Neutral Ranges of the Valence Scale
3.4. Do patient clusters differ on relevant demographic and clinical characteristics?
A series of one-way ANOVAs and Chi-Square analyses were conducted to examine differences in clinical and demographic variables for the two patient subgroups. As shown in , analyses indicated that the 2 patient clusters did not significantly differ on demographic variables; however, there was a trend indicating a lower WASI full-scale IQ in cluster 2.
Demographics for Patient Clusters
3.5. Do patient clusters differ on measures of symptom severity?
Separate MANOVAs were conducted to examine the effects of group membership on PANSS symptom severity and Chapman Scale Anhedonia. In the first set of MANOVAs, PANSS positive, negative, and general scale factors served as dependent variables. In the second set of MANOVAs, Chapman scale physical and social anhedonia served as dependent variables. One-way ANOVAs were also calculated to determine whether the 2 patient clusters differed on the PANSS total score and Chapman total anhedonia score.
MANOVA indicated a significant main effect of group with PANSS scores serving as the dependent variable, F (1, 47) = 3.45, p = .02 (η2 = .24), signifying differences in symptom severity between the 2 patient clusters. Significant between-subjects effects were observed for the PANSS negative symptom factor (F = 8.39; p = .006), with individuals in cluster 2 demonstrating higher levels of negative symptoms than individuals in cluster 1. However, no significant difference was found between groups for the PANSS positive and general symptom factors ().
Means and Standard Deviations for Symptom Measures and Functional Outcome Among Patient Clusters
MANOVA also indicated that the 2 patient clusters significantly differed on Chapman Scale Physical Anhedonia, F (1, 46) = 4.23, p = .02, with patients in cluster 2 indicating significantly higher levels of Physical Anhedonia than patients in cluster 1. The two groups did not significantly differ on Hamilton Depression scale severity. Thus, the two patient clusters appear to differ primarily on negative symptoms, with cluster 2 evidencing a greater severity of negative symptoms.
3.6. Do patient clusters differ on functional outcome?
MANOVA was conducted to examine the effects of group membership on functional outcome. HCQL Instrumental Role, Interpersonal Relations, Intrapsychic Foundations, and Common Objects and Activities scores served as dependent variables. Mean HCQL scores are presented in . MANOVA indicated that the overall effect of group was significant for HCQL functional outcome, F (4, 44) = 2.72, p = .04 (η2 = .24). Significant individual effects were found for the Instrumental Role Functioning (p = .004) and Common Objects and activities (p = .009) subscales. The Intrapsychic Foundations subscale approached significance (p = .087). As can be seen in , cluster 1) displayed significantly better overall functional outcome than cluster 2.