T46. EXPLORING COGNITIVE HETEROGENEITY: A COMBINATION OF TWO POPULAR DATA-DRIVEN MODELS. (9th April 2019)

• Record Type: Journal Article
• Authors: Carruthers, Sean; Gurvich, Caroline; Neill, Erica; Sumner, Philip; Tan, Eric; Thomas, Elizabeth; Rheenen, Tamsyn Van; Rossell, Susan
• Is Part Of: Schizophrenia bulletin. Volume 45(2019)Supplement 2
• Page Start: S221
• Page End: S222
• Publication Date: 2019-04-09
• Subjects: Schizophrenia -- Periodicals; Schizophrenia -- Research -- Periodicals; 616.898005
• DOI: 10.1093/schbul/sbz019.326
• Languages: English
• ISSNs: 0586-7614

Abstract

Abstract: Background: Considerable cognitive heterogeneity exists within the schizophrenia spectrum disorder (SSD) population. Data-driven methodologies are becoming increasingly popular in the investigative pursuit of characterising the variability of cognitive impairments in SSD. A common data-driven method compares the relationship between estimates of premorbid and current intellectual function to identify IQ trajectory subgroups within SSD samples. An alternative method involves classifying SSD participants based on their current neurocognitive performance. To-date, researchers have used either putative IQ trajectories or current neurocognitive function to identify cognitive subgroups within the SSD population, however the overlap in classifications applied using the two models has yet to be investigated in a single sample. The aim of the present study was to compare the classifications applied using two common data-driven techniques and establish if a combination of the two models provides any additional discriminant ability. Methods: Estimates of premorbid and current intellectual functioning and the non-social cognitive domain scores from the MATRICS consensus cognitive battery (MCCB) from 137 SSD participants were entered into two separate hierarchical cluster analyses with k¬-means optimization. A third analysis was performed, entering estimated premorbid IQ and the MCCB cognitive domains into the cluster analysis. A series of discriminate function analyses (DFA) were performed to compare the strength of each model. Variables were standardized to the performance on 293 healthy controls. Results: In line with the literature, entering estimates of premorbid and current IQ into the cluster analysis resulted in three clusters emerging, reflecting putative preserved (PIQ), deteriorated (DIQ) and compromised (CIQ) IQ trajectories. Using current cognitive performance, three clusters of high, intermediate and impaired cognitive function emerged. The results of the DFA indicated that the two models exhibited similar overall classification accuracy and discriminatory ability, with both methods providing good separation of subgroups. The combined model resulted in four clusters emerging, a preserved and a compromised ability subgroup emerged, in addition to two deteriorated subgroups. One subgroup was characterised by a meaningful deterioration from premorbid levels to an intermediate level of current neurocognitive capacity; with the remaining subgroup characterised by a large deterioration from premorbid function to a comprised level of current neurocognitive capacity. The results of the DFA indicated that the combined models exhibited better overall classification accuracy and discriminatory ability compared to each model individually. Irrespective of the model employed, emergent subgroups did not significantly differ in clinical severity. Discussion: In comparing two commonly used data-driven models used to characterise cognitive heterogeneity in SSD, it was found that a combination of the two provided the best discriminatory strength. It appears that each model lacks the strength to accurately classify the proportion of SSD individuals who fall within the two anchoring subgroups of intact and impaired cognitive function. By combining estimates of premorbid intellect with multiple indices of current neurocognitive function two subgroups emerged that represent an intermediate and large degree of deterioration from premorbid function. Future data-drive studies should consider both estimated premorbid intellectual function and current, multidimensional neurocognition.