Abnormalities of saccadic eye movements in dementia due to Alzheimer’s disease and mild cognitive impairment

Antisaccade latency

The latencies for all correct saccades in the antisaccade are shown in Figure 1A. A one-way ANOVA on the mean antisaccade latencies revealed that there was a significant main effect of participant group (F(3,238)=13.541;p<.005; η²⁼.146). AD (N=65; mean=404; SD=86; 95% CI=383-427) generated significantly longer latencies than the CP group (N=91; mean=338ms; SD=84; 95% CI=321 -355; t(154)=4.801;p<.0005; d=.77) and the naMCIs (N=46; mean=363ms; SD=62; 95% CI=346-381; t(109)=2.78;p=.006; d=.53). However, antisaccade latencies did not differ between the aMCI (N=40; mean=419ms; SD=82; 95% CI=394-444) and the AD group (t(103)=.857;p=.394; d=.17). The mean latencies of aMCI group were significantly longer than the naMCI group (t(84)=3.607;p=.001; d=.79), and the CP group (t(129)=5.116;p<.0005; d=.90). The latencies for the naMCI did not differ from the CP group significantly (t(135)=1.785;p=.077; d=.31). Because age was not balanced between the groups we next performed an analysis of covariance. The ANCOVA, with a between-subjects factor: group (AD, aMCI, naMCI, CP); and a covariate: age, revealed a significant main effect of age (F(1,230)=32.115;p<.0005; η²⁼.123), and group (F(3,230)=6.175;p<.0005; η²⁼.075). However, the corrected model, with age as a covariate was also significant (F(4,230)=18.087;p<.0005; η²⁼.239), confirming that the group effect remained after age was partialled out. To provide further verification that age did not affect results we selected a subsample of age-matched participants. This age-matched sample was achieved by excluding the oldest AD (N=4) and aMCI (N=3) and the youngest CPs (N=22) and naMCI (N=9) participants from the sample. An ANOVA confirmed that the groups were matched across age (F(3,201)=.800;p=.495; η²⁼.012). The antisaccade latency findings were replicated in this age-matched subsample. ANOVA revealed a significant main effect of participant group for mean antisaccade latencies (F(3,201)=7.263;p<.0005; η²⁼.098). AD (N=62; mean=402ms; SD=88; 95% CI=381-425) revealed increased latencies in comparison to the CP group (N=69; mean=351ms; SD=88; 95% CI=331-374; t(129)=3.304;p=.001; d=.58), and the naMCI group (N=37; mean=363; SD=64; 95% CI=343-383; t(97)=2.311;p=.023; d=.47). Again, the aMCI (N=37; mean=418; SD=83; 95% CI=392-445) did not differ from AD (t(97)=.910;p=.365; d=.19). The aMCI group revealed significantly longer latencies compared to the naMCI group (t(72)=3.160;p=.002; d=.75), and the CP group (t(104)=3.811;p<.0005; d=.75). The antisaccade latencies for the naMCI group did not differ from CP group (t(104)=.759;p=.449; d=.15).

Antisaccade uncorrected errors

Figure 2B shows the mean frequency (%) total number of trials with uncorrected errors in each group for the antisaccade experiment. A logistic regression [18] was fitted to the proportion of uncorrected errors, with weights proportional to the number of AST completed per participant and a dispersion parameter to account for potential higher than anticipated variation in the number of uncorrected errors. Estimated log-odds regression coefficients, presented in Table 1, identify that the proportion of uncorrected errors are significantly lower than 50% for all groups, and thereby revealing that there was a significant effect on participant group for the antisaccade errors (F(3,245)=10.914; p<.0005). The distribution of the standardised Pearson residuals did not visually appear to satisfy the assumption of the normal distribution. However, a one-way ANOVA test suggest that the residual contain no further dependence on group (F(3,245)=.108; p=.956) and likewise from a non-parametric analysis using the Kruskal Wallis H Test (H(3)=3.073; p=.381).

As expected, AD (N=68; mean proportion=26; SD=29; 95% CI=19-33) generated increased errors in comparison to the CP group (N=91; mean proportion=10; SD=11; 95% CI=8-13; χ2(1)=18.459; p<.0005). Critically, the aMCI group (N=42; mean proportion=30; SD=30; 95% CI=21-39) generated a higher proportion of antisaccade errors compared to the naMCI group (N=46; mean proportion=12; SD=11; 95% CI=9-16; χ2(1)=11.774; p=.001) and the CP group (χ2(1)=21.806; p<.0005). However, there was not a significant difference in the errors of the aMCI and the AD group (χ2(1)=.665; p=.415) and also between the naMCI and CP groups (χ2(1)=.588; p=.446). The difference between the AD and naMCI was shown to be significant (χ2(1)=8.792; p=.003).

An extension to the model that included participant age revealed that the group main effect remained significant (F(3,235)=8.173; p<.0005), whilst age was not (F(1,235)=.167; p=.684). To provide further verification that age was not a confounding factor, an age-matched sample was entered in the analysis. The results, presented in Table 1, were essentially replicated. There was a significant main effect of participant group for antisaccade errors (F(3,207)=9.295; p<.0005). For antisaccade errors, the AD (N=64; mean proportion=26; SD=28; 95% CI=19-32) revealed an increased error rate compared to the CP group (N=69; mean proportion=10; SD=10; 95% CI=7-12; χ2(1)=16.945; p<.0005) and the naMCI group (N=37; mean proportion=13; SD=12; 95% CI=9-16; χ2(1)=7.183; p=.007). Again, critically, the aMCI group (N=39; mean proportion=30; SD=31; 95% CI=20-40) revealed a significantly higher frequency of errors compared to the naMCI group (χ2(1)=8.327; p=0.004) and the CP group (χ2(1)=17.635; p<.0005). The aMCI group did not differ from the AD group (χ2(1)=.258; p=.612 and nor did the naMCI group from the CP group (χ2(1)=.976; p=0.323).

Further, to confirm whether AST error rate was associated with memory decline we performed a series of correlations between AST failure rate and FCSRT free recall score. Overall there was a negative association between AST error rate and FCSRT free recall (r(180)=-.480;p<.0005) which indicates that an increase in AST errors was associated with poorer free recall on the FCSRT task. This association was also observed when considering the participant groups separately: a significant negative association was found between AST error rate and FCSRT free recall within AD patients (r(42)=-.430;p=.004), aMCI (r(39)=-.439;p=.004), and marginally for naMCI (r(44)=-.288;p=.052). The association was not significant in control participants (r(49)=-.166;p=.243). The results indicate that AST inhibitory control is associated with memory decline in patient groups.

To clarify the spatial-temporal characteristics of the extracted signals in the AST, heatmaps of the gaze positions of each participant group are displayed in Figure 2. The heatmaps that the density and length of the “comet” (yellow) tail is shortest for the CP and naMCI groups; and longest for the AD and aMCI groups. This reflects the findings reported above, that the AD and aMCI generated a high proportion of errors in the AST which are uncorrected. Fewer of these uncorrected errors are seen in the traces of the control and naMCI group. In summary, three salient patterns emerge: (1) There were a number of reflexive saccades to the target location (which should have been inhibited) that increased from the level in CP participants, to the naMCI patients, to the aMCI patients, and finally showing the highest level in the AD patients. (2) Similarly, the time cost to correct the mistakes increases from the CP participants, to the naMCI patients, to the aMCI patients, and finally to the AD patients. Given 1500 milliseconds since target onset, the naMCI patients and the CP participants could correct almost all the mistakes, while the AD and aMCI patients have more uncorrected mistakes. (3) The CP participants revealed a higher proportion of correct antisaccades and corrected errors in comparison to the patient groups (AD, aMCI, naMCI).

Participants

Participants were men and women between the ages of 55 and 90, with at least 11 years of education and fluent English-speakers. Of these, 68 were people with dementia due to AD, 42 had a diagnosis of aMCI, and 47 had a diagnosis of naMCI. We also included 92 age-matched cognitively healthy people to act as control participants (see Table 2). Control participants were recruited from the local community or were the spouse/partner of the AD or MCI participants. All participants were white British or European. Participants with MCI or dementia due to AD were recruited through local memory clinics in the National Health Service (NHS) and had received a clinical diagnosis following a full assessment with a dementia specialist. All those in the dementia group met clinical criteria for dementia due to AD, as per NINCDS-ADRDA criteria [23]. Those with a diagnosis of MCI met the following criteria [24]: (1) subjective complaints of memory decline (reported by the person themselves or an informant); (2) objective memory or other cognitive impairment (considered when scores on standard cognitive tests were >1.5 SDs below age/education adjusted norms) with or without deficits in other cognitive domains; (3) intact daily-life activities. To subtype the MCI group further into aMCI and naMCI, we applied the [24] criteria (see below). Participants with intact cognition (as assessed by a series of cognitive tasks; see below) were recruited from the local community. Participants were not eligible for the study if they had a previous history of head trauma, stroke, cardiovascular disease, active or past alcohol or substance misuse or dependence, or any physical or mental condition severe enough to interfere with their ability to participate in the study. Those with a global or specific learning disability were also not eligible to participate in the study. All participants had the capacity to consent to participation in the study and signed informed consent. Ethics’ Committee approval was granted by Lancaster University and NHS Health Research Authority.

Stimuli and tasks

Eye-tracking was assessed using the AST. Cognition was assessed the Montreal Cognitive Assessment (MoCA [25]), the Free and Cued Selective Reminding Test with Immediate Recall (FCSRT-IR [26]), and the digit and spatial span [27,28].

Apparatus

We used an EyeLink Desktop 1000 eye-tracker (SR Research) with sampling at 500 Hz. Participants sat 55 cm away from the monitor (60 Hz). Their dominant eye was determined using the Miles test [29] and tracked accordingly. Experiment Builder software (SR Research) was used to control the stimulus events during the eye-tracking tasks.

Antisaccade task (AST)

Each antisaccade trial was preceded by a 1 second instruction screen stating that the participant should look toward the target. A central fixation target central fixation was displayed in white on a black background. This was displayed for one second. There was a 200ms blank interval before the appearance of the saccade distractor. The saccade distractor (in red) was then presented in a random order 4 degrees away from where the fixation target had been either on the left or right side for 2 seconds. Participants were asked to fixate at the central point then generate the saccade to the opposite position of the screen as soon as the distractor appeared. There were a total of 24 antisaccade trials in AST.

Montreal cognitive assessment (MoCA)

The MoCA is a brief screening tool for Alzheimer’s dementia that includes the domains of attention and concentration, executive functions, memory, language, visuoconstructional skills, conceptual thinking, calculations, and orientation. The test generates a maximum score of 30; a scores of 26 or more is considered normal.

Free Cued Selective Reminding Task Immediate Recall (FCSRT-IR)

The International Working Group on Alzheimer’s disease [30] recommended the free and cued selective reminding test (FCSRT [26]) to assess memory, as it showed high sensitivity and specificity in the differentiation of dementia due to AD from healthy controls, and aMCI from naMCI (see [12]). Participants were asked to memorise line drawings of a set of easily recognised objects (e.g., grapes) which belonged to unique category cues (e.g., fruit). 16 items to be learned were presented four at a time on a card, one picture in each quadrant. Participants were asked to search each card, then point to and then name aloud each item (e.g., grapes) after its cue (fruit) was verbally presented. The card was then removed, and the participant immediately performed a cued recall of just those four items on the basis of the category cue. Participants were reminded of any items that were not recalled by presenting the cue and the item together (e.g., the vehicle was a train). This procedure was repeated for all 16 pictures across the four cards. Participants were then given a test phase that consisted of three recall trials. Each test was preceded by a 20 second counting distractor task. Each recall trial consisted of two parts; after two minutes of free recall, category cues were provided for items not retrieved during the free recall phase. If subjects failed to retrieve the item when provided with the category cue, they were reminded by the presentation of both the cue and the item together. A measure of free recall and a measure of cued recall was obtained by calculating the correct responses (both out of a total of 48). This task has high validity as it has been used extensively in informing MCI and AD diagnoses [24]. [24] recommends that scores equal to or below 27 on the free recall score indicate aMCI, whilst scores of 28 and over indicate naMCI.

Digit span

In the digit span task [27,28], which assesses phonological memory function, participants were verbally presented with strings of single digit numbers. Once the string had been presented, the participant was asked to recall the digits in the correct serial order. The number of digits presented gradually increased during the course of the experiment, starting with two and going up to a maximum of nine. Two trials were presented at each level and a participant must get both correct in order to progress to the next level of the task. Participants then completed a reverse digit span task where participants need to reverse the order of the presented sequence. Again, the number of digits presented gradually increased from two to eight. Two trials were presented at each level with both correct needed in order to progress. A total digit span score was then calculated out of 30 by summing the correct trials together.

Spatial span

In the spatial span task [27,28], participants were presented with an array of nine squares. Squares in the array were selected one at a time by the experimenter. At the end of the sequence, it was the task of the participant to indicate the locations of the selected squares in the correct serial order. The number of squares selected in sequence increased over the course of the experiment from two up to a maximum of nine. Two trials were presented at each level and a participant had to get both correct in order to progress. A reverse version of the task was also performed with the participant indicating the sequence in the backwards order. A measure of total spatial span score was calculated out of a total of 32.

Data processing

The raw eye tracking data exported from the EyeLink DataViewer software were analysed offline in bespoke software [31] with the following features: Noise and spikes were filtered by removing all the frames where the velocity signal was greater than 1,500 deg/s or the acceleration signal was greater than 100,000 deg²/sec. All the fixations and saccadic events were detected by the EyeLink parser. All the saccades extracted for each trial as well as a range of spatial and temporal properties measured for each saccade were then stored in a table. Microsaccades with amplitude less than 0.7 deg were filtered from the data. The latency of the saccade was measured from the onset of the saccade to the target onset. Only the saccades made within the time window 80-700 ms after target onset (see Figure 3) were included to avoid ‘anticipatory saccades’ i.e. saccades which are initiated prior to the presentation of the distractor.