The relevance of Spearman's g for epilepsy

Abstract

Whilst the concept of a general mental factor known as 'g' has been of longstanding interest, for unknown reasons, it has never been interrogated in epilepsy despite the 100+ year empirical history of the neuropsychology of epilepsy. This investigation seeks to identify g within a comprehensive neuropsychological data set and compare participants with temporal lobe epilepsy to controls, characterize the discriminatory power of g compared with domain-specific cognitive metrics, explore the association of g with clinical epilepsy and sociodemographic variables and identify the structural and network properties associated with g in epilepsy. Participants included 110 temporal lobe epilepsy patients and 79 healthy controls between the ages of 19 and 60. Participants underwent neuropsychological assessment, clinical interview and structural and functional imaging. Cognitive data were subjected to factor analysis to identify g and compare the group of patients with control participants. The relative power of g compared with domain-specific tests was interrogated, clinical and sociodemographic variables were examined for their relationship with g, and structural and functional images were assessed using traditional regional volumetrics, cortical surface features and network analytics. Findings indicate (i) significantly (P < 0.005) lower g in patients compared with controls; (ii) g is at least as powerful as individual cognitive domain-specific metrics and other analytic approaches to discriminating patients from control participants; (iii) lower g was associated with earlier age of onset and medication use, greater number of antiseizure medications and longer epilepsy duration (Ps < 0.04); and lower parental and personal education and greater neighbourhood deprivation (Ps < 0.012); and (iv) amongst patients, lower g was linked to decreased total intracranial volume (P = 0.019), age and intracranial volume adjusted total tissue volume (P = 0.019) and age and intracranial volume adjusted total corpus callosum volume (P = 0.012)-particularly posterior, mid-posterior and anterior (Ps < 0.022) regions. Cortical vertex analyses showed lower g to be associated specifically with decreased gyrification in bilateral medial orbitofrontal regions. Network analysis of resting-state data with focus on the participation coefficient showed g to be associated with the superior parietal network. Spearman's g is reduced in patients, has considerable discriminatory power compared with domain-specific metrics and is linked to a multiplex of factors related to brain (size, connectivity and frontoparietal networks), environment (familial and personal education and neighbourhood disadvantage) and disease (epilepsy onset, treatment and duration). Greater attention to contemporary models of human cognition is warranted in order to advance the neuropsychology of epilepsy.

Keywords: brain morphology; graph theory; neuropsychology; resting-state fMRI.

Graphical Abstract

Figure 1

Correlation plot and cognitive testing between groups. (A) Correlation plot between cognitive tests used to generate g and the results of the LDA—a machine learning tool to find a linear combination of variables that maximal separates TLE from controls. The numbers are Pearson correlations. In the left box are the correlation between g and LDA results. In the right box are the weights of each cognitive test for g and LDA. (B) Differences in cognitive tests between TLE and controls. (C) Differences in cognitive tests between TLE and controls from the NIH cognitive toolbox before and after adjustment for g factor; g is also present for comparison. Permutation tests equivalent to a Student's t-test were performed: *P < 0.05, **P < 0.01 and ***P < 0.001. LDA, linear discriminant analysis; WASI, Wechsler Abbreviated Scale of Intelligence; VOC, vocabulary; JOLO, Judgement of Line Orientation; SemntlFl, semantic fluency; RAVLT, Rey auditory verbal learning test; BNT, Boston Naming Test; Flank, Flanker inhibitory control and attention; WMEM, working memory; Peg, pegboard; SQMEM, sequence memory; Pspeed, processing speed; VOCAB, picture vocabulary; ORAL, oral reading recognition; DGRIP, grip strength of dominant hand; NDGRIP, grip strength of non-dominant hand; adj, adjusted for g; NIH, National Institutes of Health; TLE, temporal lobe epilepsy.

Figure 2

Association of g with ADI. Student's t-test was performed for controls (n = 79) (t_1,187=−2.401, P = 0.02) and TLE participants (n = 110) (t_1,187=−2.134, P = 0.038) to evaluate the significant relationship between decreasing neighbourhood disadvantage and increasing g. ADI, Area Disadvantage Index; TLE, temporal lobe epilepsy.

Figure 3

g score correlation with Local Gyrification Index on TLE (n = 101) controlling for age. (A) Left hemisphere. (B) Right hemisphere. Significant clusters of correlation were found in bilateral medial orbitofrontal and right lateral occipital gyrus. To correct for multiple comparisons, a Monte Carlo simulation was implemented with cluster-forming threshold set to P = 0.05.

Figure 4

Global GT measures. Normalized transitivity (left), normalized global efficiency (middle) and modularity index (Q) (right) in healthy controls (squares) (n = 50) and TLE participants (circles) (n = 101) across density levels. Student's t-test was performed; results are significantly different at each density level for normalized transitivity and Q (t_1,149 > 4, P < 0.001) and for density levels from 14 to 26% for normalized global efficiency after Bonferroni correction (t_1,149 < −4, P < 0.001). TLE, temporal lobe epilepsy.

Figure 5

Average PC across parcellations. Controls (n = 50) are the squares and temporal lobe epilepsy participants (n = 101) are the circles. ANOVA was performed between groups; *significantly different between groups after false discovery rate correction (F_1,149 > 5, P < 0.05). Abbreviations can be found in Supplementary Table 1. TLE, temporal lobe epilepsy.