Semantic Memory and Lexical Availability in Parkinson’s Disease: A Statistical Learning Study

Abstract

Parkinson’s disease (PD) is a neurodegenerative disorder that causes a progressive impairment in motor and cognitive functions. Although semantic fluency deficits have been described in PD, more specific semantic memory (SM) and lexical availability (LA) domains have not been previously addressed. Here, we aimed to characterize the cognitive performance of PD patients in a set of SM and LA measures and determine the smallest set of neuropsychological (lexical, semantic, or executive) variables that most accurately classify groups. Thirty early-stage non-demented PD patients (age 35–75, 10 females) and thirty healthy controls (age 36–76, 12 females) were assessed via general cognitive, SM [three subtests of the CaGi battery including living (i.e., elephant) and non-living things (i.e., fork)], and LA (eliciting words from 10 semantic categories related to everyday life) measures. Results showed that PD patients performed lower than controls in two SM global scores (picture naming and naming in response to an oral description). This impairment was particularly pronounced in the non-living things subscale. Also, the number of words in the LA measure was inferior in PD patients than controls, in both larger and smaller semantic fields, showing a more inadequate recall strategy. Notably, the classification algorithms indicated that the SM task had high classification accuracy. In particular, the denomination of non-living things had a classification accuracy of ∼80%. These results suggest that frontostriatal deterioration in PD leads to search strategy deficits in SF and the potential disruption in semantic categorization. These findings are consistent with the embodied view of cognition.

Neurocognitive correlates of semantic memory navigation in Parkinson’s disease

“…PD patients’ responses were characterized by higher concreteness and imageability. This mirrors results from statistical learning analyses showing that both variables contribute to discriminating PD patients from HCs 9 . Given that increased concreteness and imageability involve reduced cognitive demands 27 – 29 , 56 – 59 , such findings suggest that patients favor easily accessible units during semantic memory navigation.…”

“…First, our sample size was moderate. Although robust findings have been reported in relevant studies with similar numbers of participants 9 , 16 , 17 , 20 , 23 , 24 , 34 , 37 , 75 , 101 , 102 , replications with larger groups would be desirable. Second, our stimuli comprised concrete concepts only, precluding insights on other critical categories, such as action verbs 2 , 3 .…”

“…Core motor symptoms in early stages of PD are typically accompanied by cognitive dysfunctions 8 , including semantic anomalies 3 , 9 . These range from difficulties with concept association 10 , retrieval 11 , and comprehension 4 , 10 , 12 , 13 to word finding and definition deficits 14 .…”

“…Highly concrete and imageable concepts (e.g., ‘table’, as opposed to ‘freedom’) involve faster responses 27 as well as reduced electrophysiological 28 and hemodynamic 29 brain modulations, pointing to lower cognitive effort. In PD research, concreteness and imageability values capture subtle differences between patients and HCs 9 . Interestingly, abstract concept processing is affected in cognitively impaired PD patients but often spared in cognitively preserved ones 3 , 11 , 12 , revealing a link with overall cognitive status.…”

Ingredients for Responsible Machine Learning: A Commented Review of The Hitchhiker’s Guide to Responsible Machine Learning

“…Note that it is indeed possible to combine classification algorithms in order to inform a final model. For example, Cardona et al [ 16 ] used the Boruta and 1R algorithms for selecting variables to be used in a logistic regression model. In the case of numeric dependent variables, techniques such as distributional regression trees and forests [ 74 ] and transformation forests [ 44 ] could be used (these are implemented in the disttree and trtf R packages, respectively).…”