Overall, the pattern of group differences reflected the group allocations, showing greater CRS scores in the ADHD and Autism+ADHD groups and greater SCQ scores in the Autism and Autism+ADHD groups. The clinical groups had lower IQ than the neurotypical group; however, this difference was statistically significant only between NT and Autism + ADHD group (see Table 1).
Table 1 Sample characteristics for Study 1
|
Neurotypical (n = 30)
|
Autism (n = 18)
|
ADHD (n = 23)
|
Autism + ADHD (n = 32)
|
Group Comparisons (p-value)
|
Demographics
|
|
|
|
|
|
Age
|
129.63 (29.29)
|
130.89 (25.05)
|
127.87 (27.14)
|
130.06 (18.36)
|
Ns (pw>.1)
|
Gender M:F
|
17:13
|
11:7
|
15:8
|
24:8
|
Ns (pw>.1)
|
WASI Full-scale IQ
|
116.2 (13.34)
|
104.61 (15.64)
|
108.61 (11.67)
|
102.06 (19.29)
|
pw= .006a
|
SCQ
|
|
|
|
|
|
Total
|
3.79 (3.71)
|
19.11 (5.98)
|
15.17 (6.96)
|
21.16 (6.23)
|
pw<.001b,c
|
SCQ Social
|
1.25 (1.5)
|
7.56 (3.34)
|
4.91 (3.26)
|
7.68 (3.47)
|
pw<.001b,c
|
SCQ Comm
|
1.82 (1.49)
|
5.61 (2.3)
|
4.61 (1.99)
|
6.39 (2.33)
|
pw<.001b,c
|
SCQ RRB
|
0.5 (1.1)
|
4.56 (2.2)
|
4.04 (2.51)
|
5.42 (2.76)
|
pw<.001b
|
CPRS
|
|
|
|
|
|
Global Index
|
51.82 (13.45)
|
79.44 (12.59)
|
87.87 (4.25)
|
87.13 (5.32)
|
pw<.001b
|
Inattention
|
50.57 (9.75)
|
77 (12.48)
|
86.78 (6.64)
|
85.09 (6.41)
|
pw<.001b, d
|
Hyperactivity
|
52.32 (12.93)
|
76.44 (13.68)
|
87.83 (3.9)
|
87.38 (5.56)
|
pw<.001b,e
|
Data shown for all measures except Gender are mean with standard deviation in parentheses. Data for gender are n male:female. WASI: Wechsler Abbreviated Scale of Intelligence; CPRS: Conners Parent Rating Scale (values shown are mean T-scores); SCQ: Social Communication Questionnaire |
p-values in the table refer to the significance value of the main ANOVA, comparing the 4 groups on respective demographic characteristics; multiple comparisons for these variables are Bonferroni-corrected. pw refers to the p value of Welch’s F test carried out where homogeneity of variances assumption was violated; for these variables, post-hoc comparisons are corrected using Games-Howell method. |
aNT>Autism + ADHD, bNT<Autism, ADHD, Autism + ADHD; cADHD< Autism + ADHD; dAutism<ADHD, eAutism< ADHD, Autism + ADHD |
Number of fixations (control variable measuring task engagement)
First, we analysed participants’ number of fixations to the screen to ensure that all participants were attentive to the task at all levels of Condition. The between-subjects factor of Autism interacted significantly with Condition: F (2, 198) = 3.03, p = .05, ƞ2p = .03. However, follow up pairwise comparisons comparing groups (Autism-Present, Autism-Absent) within each condition yielded no significant differences (all p>.1) (descriptive statistics provided in Additional Files AF1). Main effects of Autism and ADHD were not significant: Autism: F (1, 99) = .008, p = .93, ƞ2p = .00; ADHD: F (1,99) = .009, p = .92, ƞ2p = .00.
Rate of change in look durations
We predicted that all participants would show reduced look durations over time to the repeating stimulus (indexed by a negative slope) and increased look durations over time to the changing stimulus (indexed by a positive slope). There was a main effect of Stimulus (F (1, 99) = 52.78, p = .000, ƞ2p = .35). As predicted, this was driven by a significantly more positive slope for the changing stimulus (Mean ± S.E. = 40.04 ± 4.84) as compared to the repeating stimulus (Mean ± S.E. = -10.84 ± 3.68). There was also a main effect of Autism (F (1, 99) = 4.74, p = .032, ƞ2p = .046). This was driven by those without Autism (neurotypical and ADHD-only: Mean ± S.E. = 20.03 ± 3.42) showing steeper slopes than those with Autism (Autism-only and Autism+ADHD: Mean ± S.E. = 9.17 ± 3.63).
There was an interaction between Condition and Stimulus (F (1.87, 185.25) = 8.74, p < .001, ƞ2p = .08) driven by a significant main effect of Stimulus for the Non-Social Simple (Mean difference Repeating vs Changing = -82.38 ± 11.16, p < .001) and Social (Mean difference = -53.74 ± 9.93, p < .001) conditions, which was non-significant in the Non-Social Complex condition (Mean difference= -16.51 ± 13.18, p= .213). This two-way interaction was moderated by a 4-way interaction between Condition*Stimulus*Autism*ADHD: F (1.87, 185.25) = 3.82, p = .026, ƞ2p = .037. We broke this interaction down by running two repeated-measures ANOVAs, separately within each level of Autism and within each level of ADHD. At each level of Autism (Absent, Present), the three-way Condition*Stimulus*ADHD interaction was not significant: Autism-Absent: F (2, 102) = 1.49, p = .23, ƞ2p = .028; Autism-Present: F (1.78, 85.55) = 2.39, p = .103, ƞ2p = .047. The equivalent analysis at each level of the ADHD factor showed that the three-way Condition*Stimulus*Autism interaction was not significant at ‘ADHD-Present’: F (2, 106) = 1.18, p = .308, ƞ2p = .022; but, in the groups without ADHD (that is in the neurotypical (NT) and Autism-only groups), there was a three-way interaction of Condition*Stimulus*Autism (F (2, 92) = 4.375, p = .015, ƞ2p = .087). Follow-up comparisons were conducted to test the Condition*Stimulus interaction in each of these groups (NT, Autism-only). These analyses showed a significant main effect of Stimulus in Neurotypical children (p < .0001, ƞ2p = .447), with shorter looks to repeating stimuli (Mean ± S.E. = -9.03 ± 5.5) and longer looks to changing stimuli (Mean ± S.E.= 46.49 ± 7.74) over time across conditions (see Figure 2a); the Condition*Stimulus interaction was not statistically significant in this group (F (2, 58) = .29, p = .75). On the other hand, the Condition*Stimulus interaction was significant in the Autism-only group (F (2, 34) = 5.50, p = .009, ƞ2p = .24) with shorter look durations over time to the repeating stimulus and longer look durations over time to the changing stimulus in the Non-Social Simple (repeating vs changing Mean ± S.E.: -31.39 ± 7.03 vs 54.64 ± 16.48) and Social conditions (repeating vs changing Mean ± S.E.: -8.68 ± 9.53 vs 33.77 ± 12.52) but a numerical difference in the opposite direction in the Non-Social Complex condition which did not reach statistical significance (repeating vs changing Mean ± S.E.: 27.79 ± 23.96 vs -19.88 ± 20.41) (as shown in Figure 2b).
Figure 2a. The main effect of Stimulus in Neurotypical participants.
Figure 2a Legend: Bars show the mean (±1 standard error) coefficient of the slope for the rate of change in look durations over trials (plotted on the y-axis). These data are split by stimulus type and condition. Asterisks denote statistical significance: *p<.05, **p<.01, ***p<.001. The interaction between Condition*Stimulus is non-significant but shown here for the purpose of visualization of differences from the Autism-only group shown in Figure 2b.
Figure 2b. Condition*Stimulus Interaction in the Autism-Only Group
Figure 2b Legend: Bars show the mean (±1 standard error) coefficient of the slope for the rate of change in look durations over trials (plotted on the y-axis). These data are split by stimulus type and condition. Asterisks denote statistical significance: *p<.05, **p<.01, ***p<.001
Correlations with SCQ
Bootstrapped bivariate correlations were computed between number of fixations to repeating and background stimuli (across conditions) and rate of change of attention to the repeating and changing stimuli in the non-social complex condition) and the SCQ subscales of social, communication and RRB symptoms. A greater reduction in look durations to the changing stimulus over time in the Non-Social Complex condition was associated with higher SCQ Social symptoms (r= -.198, p= .05, [-.365, -.032]) (See Figure 3), suggesting that those with higher symptom severity on this scale showed a bias against attending to the changing stimulus over time, in this condition. To evaluate the role of IQ, we computed partial correlations between SCQ Social symptoms and Rate of change of attention to the changing stimulus in the Non-Social Complex Condition, whilst controlling IQ. The correlation became nonsignificant (r = -.161, p = .112, [-.326, -.007]).
Given the finding of flatter slopes for the rate of change in look durations overall in autistic individuals as compared to non-autistic individuals in our sample, we also ran a correlation between IQ and the average rate of change of look durations over time with data collapsed across conditions and stimuli. The correlation was not statistically significant (r = -.111, p= .264, [-.282, .079]).
Figure 3. Relationship between SCQ-Social scores and Rate of change measure in Non-Social Complex condition
Figure 3 Legend: Scatterplot of scores on Social Communication Questionnaire (SCQ) Reciprocal Social Interaction Subscale (plotted on the x-axis) with the coefficient of the slope for the rate of change in look durations over trials to the Non-Social Complex Changing Stimulus (plotted on the y-axis) for participants with and without Autism (represented by orange and blue dots respectively. Dotted orange and blue lines represents the trend lines for the participants with and without Autism respectively.
Summary and Discussion of Study 1
In this study, we set out to identify whether differences in attention to repeating versus changing information in autism are present across stimulus contexts, suggesting a bias away from novelty towards repetition and predictability; or if they are dependent upon stimulus complexity, indicating slower information processing which is exacerbated when stimuli are complex. Further, we investigated whether this attention profile was specific to children with autism or not. In doing so, we included a group of children with ADHD who are more likely to have a novelty bias, reflecting an information foraging style that is opposite to those with autism. Finally, we also included a group of children with co-occurring autism and ADHD to investigate what type of profile of information foraging biases they show.
Analysis of the rate of change in look durations to the repeating versus changing stimuli revealed that autistic participants (with or without ADHD) showed flatter slopes of change in look durations to repeating and changing stimuli across conditions of stimulus complexity, suggesting that they were slower to shift attention, possibly due to taking longer to process information. Further, autistic children (without co-occurring ADHD) showed a neurotypical profile of reduced attention over time to the repeating stimulus and increased attention over time to the novel stimulus in the Non-Social Simple (shapes) and Social conditions. However, they did not show this effect in the Non-Social Complex (clocks) condition, in which they showed prolonged attention to the repeating over the changing stimulus. This is a reversal of the neurotypical effect and indicates that autistic children are not just defined by reduced habituation to a repeating stimulus but, when presented with visually complex stimuli, they show a bias towards repetition and away from novelty. This effect is more complex than we predicted as it suggests both slower information processing, reflected in flatter slopes to the repeating and changing stimuli compared with neurotypical participants with a preservation of the changing>repeating pattern, to Social and Non-Social Simple stimuli, in combination with a bias for repetition over novelty (reflected in a reversal of the changing>repeating effect) but only to Non-Social Complex stimuli. This is an important effect, which suggests that differences in attentional biases against novelty might partly be driven by a response to stimulus complexity such that greater complexity elicits this bias towards sameness and predictability, away from novelty. Importantly, prolonged attention over time to the repeating stimulus in the Non-Social Complex condition could be reflective of slower habituation or preference for repetition; however, in this condition only, the direction of change over time reversed such that autistic participants did not shift their attention preferentially towards the novel stimulus, showing a bias against novelty.
Our findings are partly in line with evidence that autistic individuals show attentional biases in favour of exploring known over unknown information (24-26), but build on this by suggesting that stimulus complexity might play a role in eliciting such biases. In support of this, there is other literature which suggests that reduced attention to social information and information foraging biases against novelty might emerge in contexts of higher complexity (48, 49). Interestingly, although this effect of a bias towards repetition did not occur in the Social condition, the effect in the Non-Social Complex condition was associated with social impairments in our sample, such that those with more parent-reported social interaction difficulties showed an atypical bias away from the changing stimulus in the Non-Social Complex condition. It is interesting that the autistic sample showed a neurotypical profile in the Social condition, albeit with flatter slopes for look durations than the NT group. One possibility is that the social stimuli used here were not complex enough; further work is needed to determine whether more socially complex stimuli (for example multimodal stimuli combining faces with speech) would also elicit the effect found here in the Non-Social Complex clocks condition.
ADHD was not related to any predicted effects. Further, while autistic participants (with or without ADHD) showed flatter slopes of rate of change in attention to both stimuli overall, only those with autism without ADHD showed an additional bias against novelty when stimuli were particularly complex. This suggests that the co-occurring presence of ADHD benefited those with autism, protecting them from biases against novelty in the Non-Social Simple and Social conditions, possibly through a compensatory effect of an opposing bias towards novelty, as suggested by Gliga et al. (2015), who reported that the infants at elevated likelihood of both autism and ADHD did not show exploitative biases. However, in our study, given that ADHD was not a main effect in these analyses, we cannot call this an additive effect because we did not find evidence of opposing biases being nulled in the comorbid group.
Importantly, in our sample, there were differences in IQ between clinical and neurotypical groups. We found that while IQ was partly associated with the above effect, it did not explain completely the relationship between SCQ scores and differences in looking to changing stimuli in the Non-Social Complex condition (the partial correlation did not reach statistical significance but the correlation was still present and indicated an effect size of similar magnitude). Further, the autistic participants with co-occurring ADHD had lower IQ than those without; yet the pattern of differences was specific to autistic children without co-occurring ADHD. Therefore, while IQ might contribute to these differences in processing and habituating to more complex stimuli, from our data it appears that IQ does not fully explain these differences.
To summarize, Study 1 found that autistic participants (with and without ADHD) exhibited a slower rate of change in look durations over time as evidenced by flatter slopes, possibly due to slower processing of information. Autistic children (without ADHD) showed a profile of prolonged attention to repetition and reduced attention to the changing stimulus over time, but only in the Non-Social Complex condition. Biases against exploration of new information in complex conditions were associated with higher social impairments in our sample, across autistic and non-autistic participants.