Methods

2.2.1 Participants

This analysis involved 104 participants from the CATFieLD study, including 49 patients diagnosed with probable LBD (26 DLB and 23 PDD), 33 with probable AD, and 22 age- matched healthy controls (HC).

DLB and PDD patients were combined into one LBD group a priori for this analysis as previous studies have shown similar attentional and executive impairment in DLB and PDD (Ballard et al., 2002a; Firbank et al., 2016) and similar patterns of brain structural alterations (Burton et al., 2004).

2.2.2 Modified Attention Network Test

The CATFieLD study used a modified version of the ANT (Cromarty, 2016; Firbank et al., 2016) based on the version described by Fan et al. (2007). The main rationale for adapting the ANT was to make it suitable for older adults and dementia patients. This was achieved by increasing the size of the stimuli to account for participants with poor visual acuity and by adjusting the timings to account for slower cognitive processing speed in older adults.

Furthermore, an additional level of executive conflict complexity was added to create a more graded task difficulty (see below).

The computerised task was programmed by Dr Michael Firbank using the Cogent toolbox in Matlab (http://www.vislab.ucl.ac.uk/cogent_2000.php). Participants completed between 3 and 14 runs of the task (median=8), each run consisting of 36 trials. Throughout the task a central fixation cross and three boxes were presented on a screen (Figure 2.1). At the beginning of each trial, one of three possible cues (no, neutral, or spatial cue) was presented for 200 ms. In the no cue condition, the boxes remained unchanged. During the neutral cue condition, the central box flashed and during the presentation of a spatial cue, one of the boxes either above or below the central box flashed (to indicate the box in which a subsequent target would appear). The disappearance of the cue was followed by a target comprising four arrowheads in a row, either in the box above or below the central fixation. The time between the

disappearance of the cue and the appearance of the target was one of the following exponentially distributed times: 700, 770, 850, 960, 1080, 1240, 1430, 1660, 1940, 2300, 2700, 3200 ms. The target stimuli could be congruent or incongruent; congruent targets were arrowheads which were all pointing in the same direction (left or right), whereas for

incongruent targets one arrowhead was pointing in the opposite direction. In the incongruent condition, the incongruent arrowhead appeared either on the end of the row (easy

33

Figure 2.1. Depiction of a single trial of the modified Attention Network Test.

incongruent) or as one of the two arrowheads in the middle (hard incongruent). The target was presented on screen until the participant made a response by squeezing a right or left hand air pressure bulb to indicate the direction in which the majority of arrowheads were facing, or until 3000 ms had elapsed. The intertrial interval was one of the following: 4300, 4500, 4750, 5000, 5350, 5700, 6100, 6400, 6800, 7200, 7700, 8300 ms, with each time occurring three times during each run in random order.

During each run, the six trial types were presented in a predetermined counterbalanced order; each cue appeared 12 times and there were 18 congruent and 18 incongruent target trials (9 easy incongruent and 9 hard incongruent). All trials from runs with less than 2/3 correct responses were excluded from further analysis as performance below this was not different from chance (Firbank et al., 2016).

2.2.3 Analysis of ANT effects

Mean RTs for each cue and target condition were calculated in Matlab (R2015b), using only the trials in which the participants gave correct responses in alignment with previous studies. The ANT effects were calculated as defined by Fan et al. (2002):

34 Alerting effect = no cue - neutral cue

Orienting effect = neutral cue - spatial cue

Executive conflict effect = incongruent target - congruent target

The alerting effect is therefore a measure of the extent to which response speed is facilitated by the presence of a warning, indicating that a response is imminently required. The orienting effect is the extent to which responses are further facilitated when the actual spatial location of the oncoming target is cued, rather than a simple warning that a response is imminent. Finally, the executive conflict effect pools all types of cued conditions and examines the impairing/interfering effect of having conflicting information regarding the target stimuli (in terms of the direction in which each of the arrowheads are pointing), compared to the

facilitative effect of having target stimuli which are all pointing in the same direction. To calculate the alerting and orienting effects, mean RTs from congruent and incongruent trials were averaged. Similarly, the executive conflict effect was calculated by averaging mean RTs across the cue conditions. For the purpose of this thesis, easy and hard incongruent conditions were not analysed separately. Error rates were also determined for each task condition by dividing the number of incorrect and missed response trials by the number of recorded trials for each participant.

2.2.4 MR imaging and analysis

Structural MR images were acquired with a 3 T Philips Intera Achieva scanner with a magnetisation prepared rapid gradient echo (MPRAGE) sequence, sagittal acquisition, echo time 4.6 ms, repetition time (TR) 8.3 ms, inversion time 1250 ms, flip angle = 8°, SENSE factor = 2, and in-plane field of view 240 x 240 mm2 _{with slice thickness 1.0 mm, yielding a}

voxel size of 1.0 x 1.0 x 1.0 mm3.

A VBM analysis was performed in SPM12 (Statistical Parametric Mapping,

www.fil.ion.ucl.ac.uk/spm/) to assess voxel-wise correlations between the ANT results and mean RT and grey matter and white matter volume. Images were first segmented into grey matter, white matter, and CSF. The segmented grey and white matter images were then co- registered and normalised to MNI space using SPM’s DARTEL algorithm (Ashburner, 2007) and modulated. Finally, images were smoothed with an 8 mm full width at half maximum Gaussian kernel.

2.2.5 Statistics

Statistical analyses were carried out in IBM SPSS Statistics version 22 and R version 3.5.1 (http://www.R-project.org/). For the mean RT data for each cue and target condition, a

35

repeated measures (cue x target) analysis of variance (ANOVA) was conducted with a between-subject factor of group (HC, AD, LBD). Subsequently, separate repeated measures (cue x target) ANOVAs were conducted for each group; post- hoc pairwise comparisons were used to calculate RT differences between the cue and target conditions, thus determining the significance of the ANT effects. The magnitude of the ANT effects was compared between groups using univariate ANOVAs; the dependent variable being the ANT effect with group as a fixed factor. For each ANOVA analysis Mauchly's sphericity test was used and F-values were adjusted accordingly. Post-hoc pairwise comparisons were corrected for multiple

comparisons using Bonferroni correction. The same analysis was repeated for the error rates. Additionally, to control for the effect of overall processing speed and to test whether between- group differences in overall processing speed influenced the analyses of the ANT effects, I repeated all analyses using normalised RTs; for each participant, the mean RT for each condition was divided by the participant’s overall mean RT in alignment with previous studies (Faust and Balota, 1997; Fernandez-Duque and Black, 2006).

Spearman’s correlations were calculated to investigate associations between the behavioural data (overall mean RT and the ANT effects) and clinical variables in the dementia groups. In the LBD group, correlations were calculated for cognitive fluctuation scores (Mayo and CAF total and subscores); supplementary analyses were performed for measures of overall

cognition (MMSE, CAMCOG), the UPDRS motor subscale, and the NPI hallucination scale. In the AD group, correlations were calculated for MMSE and CAMCOG. P-values were false discovery rate (FDR)-corrected for multiple comparisons.

Correlations between the ANT behavioural data and grey and white matter volume were assessed using a general linear model (GLM) in SPM12. The GLM combined all three ANT effects (alerting, orienting, and executive) as variables of interest in one design matrix and a separate model was used for mean RT. Covariates of no interest for age, sex, total intracranial volume, and UPDRS motor scores (in LBD) were included. An explicit mask was estimated to restrict the statistical analysis to voxels which represented grey and white matter,

respectively (Ridgway et al., 2009). Significant results are reported at a voxel-level p- value<0.001. Additionally, the minimum cluster size for a multiple comparison corrected threshold of p<0.05 was determined by Monte Carlo simulations using the REST software (www.restfmri.net).

To study the possible influence of dopaminergic medication on the ANT effects in the LBD group, the repeated-measures (cue x target) ANOVA was repeated including a covariate for levodopa equivalent daily dose (LEDD) (Tomlinson et al., 2010). This was tested for both raw and normalised RT.

36