Trajectories of imitation skills in preschoolers with Autism Spectrum Disorders

Participants

The participants in this study were recruited as part of the Geneva Autism Cohort (Franchini & al., 2016, 2017, 2018; Kojovic & al., 2019; Robain & al., 2020; Sperdin & al., 2018). The University’s Institutional Review Board approved this study. All families provided written informed consent to participate. For the current study, a total of 220 participants were included, 177 with ASD and 43 typically developing children (TD) (see Table 1 and Table S1 in Supplementary Materials). Among these, 126 were followed longitudinally, yielding a total of 396 time-points (78 subjects had 2, 46 had 3, and 2 participants had 4 time-points), each time-point was separated by 12 months, except for two children for whom the two time-points are separated by 24 months (see Figure 1). The inclusion criteria for ASD group was a clinical diagnosis of autism according to the DSM-5 criteria (American Psychiatric Association, 2013). The clinical diagnosis was confirmed by the scores exceeding the cut-off for autism at the ADOS-G (Lord et al., 2000) or ADOS-2 (Lord & al., 2012) (see Measures for more details). Children from the TD group had no developmental concerns, no first or second-degree family member diagnosed with autism, and were all administered the ADOS diagnostic test to exclude the presence of ASD symptoms (see Table 1). Participants were aged between 2.0 and 5.0 years, and the two groups did not differ by age (U = 351, p = 0.429) nor sex (p = 0.156).

• Download figure
• Open in new tab

Figure 1.

Data distribution available for each group from our longitudinal study. Circles (red = ASD, blue = TD) denote visits/time-points, repeated visits of the same subject are connected with a dotted line.

Imitation

For the measure of imitation skills we used the Visuo-Motor Imitation scale of the PEP-3 (Schopler & al., 2005) (for a detailed description of this test please refer to Supplementary Materials). This scale includes 10 items of imitation, assessed on a 3 point scale (0 = absence of imitation, 1 = equivoque, partial or prompted imitation, 2 = complete imitation of the skill). Raw scores range between 0 and 20, which allows enough variability to highlight the heterogeneity of imitation profiles.

The scale comprises the imitation of familiar actions and imitation with explicit instruction. Vivanti & Hamilton (2014) distinguished two types of imitation that we find in this scale : true imitation (copying both the mean and the goal of the action) and stimulus enhancement (perform a specific action on a stimulus after having been made attentive to this stimulus by another person). For instance, one item of the PEP-3 is to organize a fake birthday party to explore imitation of daily actions like eating etc. (stimulus enhancement). For another one, the examiner touches her/his nose and says “Now it’s your turn” (true imitation of meaningless actions).

Level of symptoms of autism

Each participant was assessed with the Autism Diagnosis Observation Schedule, 2nd edition (Lord et al., 2012) or Autism Diagnosis Observation Schedule, Generic (ADOS-G; Lord et al., 2000). For subjects who underwent the earlier ADOS version (ADOS-G) assessment, the scores were transformed according to the revised ADOS algorithm (Gotham et al., 2009) to ensure comparability with ADOS-2. The ADOS Calibrated Severity Score (ADOS-CCS; Gotham et al., 2009), a measure of symptom severity relatively independent of language levels and age, was then used for all subjects included in this study.

Mullen Scale of Early Learning

Developmental functioning of all children included in the present study was assessed with the Psychoeducational Profile, 3rd edition (Schopler et al., 2005). Nevertheless, as our aim was to test the relation between the imitation skills (VMI subscale of PEP-3) and the developmental functioning, we decided to use another measure than the PEP-3 to estimate the level of developmental functioning. We thus used the Mullen Scale of Early Learning-MSEL (Mullen, 1995), which was available for 133 out of the 177 children with ASD included in this study (this measure was indeed added at a later point in our cohort protocol and was thus missing for the data collected earlier).

The MSEL scale is developed for the assessment of typically developing children consists of 5 domains with items organized according to developmental stages and their level of difficulty : Gross Motor (GM), Visual Reception (VR), Fine Motor (FM), Receptive Language (RL) and Expressive Language (EL). This assessment yields a global composite score (Early Learning Composite ; ELC) by adding the four cognitive subscales (VR, FM, RL, EL), and has a mean of 100 and standard deviation of 15. For the global developmental functioning in this study, we considered the standard composite score (ELC). For the measure of individual domains (Receptive and Expressive language), we used the measure of age equivalents (AE). We opted for the AE scores as the standard T scores of the MSEL subscales show poor sensitivity in the lower functioning end of the continuum (floor of 20), and an important number of our participants had this lowest score.

Between-group analyses (ASD vs TD)

Between-group analyses (ASD vs TD)

Group comparison of imitation skills

Using a cross-sectional sample (N = 220, 177 children with ASD, 43 with TD, age range: 2.0 -5.0 years old), we were able to show that the two groups differed significantly in their imitations skills (U = 1270, p < 0.001). Indeed, children with ASD had markedly lower imitation skills than typically developing children (see Figure 2).

• Download figure
• Open in new tab

Figure 2.

Group comparison of imitation skills obtained from Visuo-Motor Imitation scale of PEP-3 (raw scores).

Imitation trajectories over time

Using all available visits for our cross-sectional sample (total of 396 time-points, N=220, n^ASD=177, n^TD=43) we tested the between-group differences in trajectories of imitation. As shown in Figure 3, the group trajectories in imitation skills estimated using mixed-models were significantly different (group effect : p < .001 ; interaction : p < .001). The best fitted model was of the 2nd order, meaning that the relation of the imitation with age was quadratic. Indeed, the TD children showed good imitation skills at two years old with only a slight improvement thereafter. In contrast, at the group level, children with ASD present markedly low imitation skills at a young age with substantial improvement between the ages of two and five (see Figure 3). However, we also observe that trajectories of children with ASD are very heterogeneous and this point is addressed in our later analyses.

• Download figure
• Open in new tab

Figure 3.

Imitation trajectories over time in the ASD and TD groups. The trajectory on the group level (solid lines, red=ASD, blue=TD) is obtained using mixed-effect models (Mancini & al., 2019; Mutlu & al., 2013). The 95% confidence interval of estimated group-level trajectory is represented in colored bands. Repeated time-points of the same subjects are connected with dotted lines.

Relation of imitation with level of symptoms and developmental domains at baseline

Within our ASD group for whom the Mullen Scales of Early Learning was available (n = 133, 113 males / 20 females, age range: 2.0 -4.8 years old), imitation skills were moderately negatively related to the level of symptoms (r^s = -0.583, p < .001), as the children who show the most of symptoms are also those with more imitation difficulties (see Figure 4). In addition, imitation skills were strongly positively correlated with the composite score of cognitive skills at baseline, as well as communication skills (Receptive and Expressive Language) (ELC : r^s = 0.671, p < .001, RL : r^s = 0.778, p < .001, EL : r^s = 0.738, p < .001), that is, children with lower imitation skills were also those with lower skills in these domains (see Figure 5).

• Download figure
• Open in new tab

Figure 4.

Spearman correlation between imitation and level of symptoms within the ASD group (Visuo-Motor Imitation scale of PEP-3 and Calibrated Severity Score of ADOS-G or 2)

• Download figure
• Open in new tab

Figure 5.

Spearman correlations between baseline imitation skills (raw scores of Visuo-Motor Imitation scale of PEP-3) and (a) Early Learning Composite of MSEL, (b) Receptive language age equivalent scores and (c) Expressive language age equivalent scores.

Imitation at baseline as a predictor of the autistic symptoms and developmental changes

Imitation skills at baseline were predictive of the gains in communication skills (Receptive and Expressive Language) a year later in our preschoolers with ASD (RL : r^s = 0.371, p = .001, EL : r^s = 0.601, p < .001). Indeed, children with better imitation skills at baseline showed a greater improvement in these domains a year later (see Figure 6).This improvement was the most evident in the domain of expressive language. Further correlations between imitations skills and the composite cognitive score or the level of symptoms did not survive after correction for multiple comparisons (considering a stringent significance level at p < .0125 using Bonferroni correction).

• Download figure
• Open in new tab

Figure 6.

Spearman partial correlations between imitation skills (raw scores of the Visuo-Motor Imitation scale of PEP-3) and a one year gain in the ASD group for receptive language (a) and expressive language (b) (difference in age equivalent scores of the Mullen Scale of Early Learning).

Imitation trajectories within the ASD group

The first step of K-means clustering on the baseline raw imitation scores yielded two optimal clusters (silhouette score of 0.828). The first cluster (sub-group) of ASD children (named ASD1, n = 46) (see Figure 7) had less difficulties with imitation at baseline and followed a trajectory highly similar to the one seen in our TD group (Figure 3) while the other cluster started significantly lower at baseline.

• Download figure
• Open in new tab

Figure 7.

Imitation trajectories over time within the ASD group (n = 102). The trajectories at the group level (solid lines black=ASD1, n_ASD1 = 46, red=ASD2, n_ASD2 = 27, blue=ASD3, n_ASD3 = 29) are obtained using mixed-effect models (Mancini & al., 2019; Mutlu & al., 2013). The 95% confidence interval of estimated group-level trajectory is represented in colored bands. Repeated time-points of the same subjects are connected with dotted lines. The distribution in the sub-groups was defined by the K-means clustering method in two steps. First clustering step yielded two clusters based on imitation skills raw scores at first time-point. We retained the higher cluster (ASD1), and proceeded to the second clustering step performed on the slopes of imitation skills, yielding ASD2 and ASD3 subgroups.

To better understand developmental trajectory in children who start lower on imitation skills, we then applied a second K-means clustering on the developmental slopes on imitation scores in our remaining group of ASD children. These children belonged to the cluster that showed less imitation skills at baseline (Total -ASD1 = 56 participants). This second clustering step yielded the two groups which we denote ASD2 and ASD3 (n^ASD2 = 27, n^ASD3 = 29). The trajectories of the two clusters from the second clustering step were significantly different (see Figure 7), one subgroup (ASD2) had a deficit on imitation skills at two years old, but showed a great improvement until five years old while the other subgroup in ASD children (ASD3) showed many difficulties in imitation at two years old and these difficulties continue over the time (group effect : p < .001 ; interaction : p < .001).

Finally, we wanted to understand which clinical characteristics were related to the distinct imitation trajectories in our sub-samples ASD2 and ASD3. In particular, we tested the baseline differences between the two groups (ASD2 and ASD3) with regards to the level of symptoms, cognitive and language skills, type of intervention and social orientation measured with an eye-tracking task (Franchini et al., 2017) (see Table S2 in Supplementary Materials). We found no significant baseline differences between the ASD2 and the ASD3 according to all tested clinical variables. However, by analyzing these two groups’ trajectories in terms of many other aspects (communication skills, cognitive skills and adaptive behaviors), we observed that the subgroup ASD2 showed an overall improvement in all domains. This pattern suggests that imitation improvement in this group indicates an overall enhancement rather than the specific link of imitation skills progress and a particular developmental measure.

Developmental trajectory of imitation in ASD and TD

Besides showing lower baseline imitative skills, we found that the acquisition of imitation skills in children with ASD did not follow the same trajectory as in TD children. In ASD children, we observed the steep progress between the age of 2 and 3.5 years, highlighting the importance of this early sensitive period for learning new skills (Dawson, 2008; Guralnick, 1997; Johnson & Munakata, 2005; Lovaas, 1987; Vivanti et al., 2016). While most TD children reach the ceiling on this measure around the age of 5 years, the ASD group showed a lower average skill level at this age. Previous studies have suggested that the imitation deficits in ASD are more specific than broad (Rogers & Williams, 2006), concerning for example, more meaningless gestures than actions on the objects. Nevertheless, our results speak in favor of a more broad impairment as most of the actions measured by the imitation scale we used are rather simple one-step actions on objects and relatively few concern more abstract meaningless gestures. These difficulties in imitation of simple actions with objects can have deleterious effects on learning (Ingersoll, 2008; Ingersoll & Meyer, 2011; Poon et al., 2012; Villalobos et al., 2005). Indeed, our results showed that imitation skills were related to composite cognitive skills and communication skills (receptive and expressive language) in toddlers with ASD. As proposed by several authors (e.g. Wadsworth & al., 2017) a child with major imitative difficulties shows delay in the acquisition of new skills, especially because he/she does not use this channel to learn from her/his parents, siblings and peers (e.g. Howe & al., 2018; Zmyj & Seehagen, 2013).

In addition, confirming previous studies (McDuffie et al., 2007; Rogers et al., 2003; Zachor et al., 2010), our results showed a strong correlation between imitation skills and the level of symptoms measured by the ADOS-2 (Lord & al., 2012), whereby the children with more difficulties in imitation had also more symptoms of ASD at baseline. Thus, imitation skills were clinically informative of the severity of the level of autistic symptoms. This result corroborates findings claiming the importance of imitation in the acquisition of early tools of communication and its contribution to autistic symptoms (Hanika & Boyer, 2019; Ingersoll, 2008; Toth et al., 2006; Villalobos et al., 2005). Indeed, difficulties of engagement in reciprocal imitation early on may greatly limit social learning opportunities, lower the experiences of sharing and joint attention, important for socio-communicative development (Franchini et al., 2017). Moreover, taking part in imitation activities prepares the first basis of communication turns. Thus from the earliest moments, the reciprocal imitation of vocalisations and facial expressions in face-to-face interactions paves the way for the development of a more complex social interchange (Trevarthen et al., 1999, as cited by Ingersoll, 2008).

Imitation as a predictor of developmental domains in children with ASD

While previous studies demonstrated that imitation predicts future cognitive skills (Hurley & Chater, 2005; Rogers & Pennington, 1991; Strid et al., 2006; Vivanti et al., 2013), the correlation that we observed in our sample did not hold after Bonferroni correction. The same was true for the relationship between imitation and change in the severity of autistics symptoms. Nevertheless, the baseline imitation skills were predictive of communication development one year later. Especially, gains in expressive language were greater in children with better imitation skills at baseline. The link between imitation and language has already been explored in previous studies (Heimann & al., 2016; Stone & al., 1997; Stone & Yoder, 2001; Toth & al., 2006). Indeed, infants start by imitating vocalizations, exchanges of vocalizations, then later words and sentences (e.g. Jones, 2007). Some studies found that verbal imitation (Smith et al., 2007) but also body movements imitation (Stone et al., 1997) are predictors of expressive language development.

Although children with ASD at a group level showed important gains in imitation skills, between two and five years old, the early difficulties in this domain presented at the age of two could still have an important influence on learning later on. Imitation is important not only for children’s current achievements, but also for their future skills. Learning to imitate can then be considered as one of the primary goals in early intervention programs for children with ASD (Rogers & Dawson, 2010), which can then be used to build on this skill for future acquisitions.

Imitation trajectories within the ASD group

Given the important heterogeneity of the ASD phenotype (Wiggins et al., 2012), we explored the extent to which there might be different developmental trajectories of imitation within our sample of children with ASD. Although some authors have studied the trajectories of different types of imitation in children with ASD (Sanefuji & Yamamoto, 2014), to the best of our knowledge, no study to date tried to parse the heterogeneity of the early imitation development in ASD. Here, using a clustering approach, we observed three separate imitation trajectories in children with ASD. Similar to previous studies (Charman et al., 2003; Stone et al., 1997; Stone & Yoder, 2001; Vivanti et al., 2013), we found that good early imitation skills are linked to better outcomes, while early difficulties predict impairment in developmental domains. Nevertheless, our study showed that some children with ASD have major difficulties in imitation at 2 years old, but show a great improvement until 5 years of age. As mentioned previously, we were not able to find a characteristic that would differentiate these two groups at baseline and predict whether a child with imitation difficulties at baseline will follow one trajectory or the other. However, our results support that imitation improvement is an indicator of an overall enhancement, without a clear causal relationship with the used measures. More investigation should be done to understand the mechanisms through which a child’s development is set on one trajectory or the other. The absence of evidence for specific differences between the two groups with a divergent imitation trajectory is interpreted with caution. Indeed, imitation is often used to incite and encourage the child to do the exercises in assessments such as the one used in this study (MSEL; Mullen, 1995), specifically when there are difficulties of language comprehension, as it is frequently the case in young children with ASD (Kwok et al., 2015). Thus, a low score on domains of cognitive or language skills could, in some cases, be better explained by maladaptive behaviors than by real difficulties in these areas (Robain et al., 2020).

Perspectives and limitations

As a limitation to this study, we would like to emphasize that the imitation skills measured in the context of our study are elementary skills, and concern mainly actions with objects. Thus, it would be particularly interesting to refine the analyses using a scale adapted for a larger age range, with a broader variety of activities, allowing differentiation of different imitation types as described by Vivanti & Hamilton (2014).

There are many avenues to pursue to better understand the mechanisms underlying imitation in young children with ASD. Indeed, as discussed, it would be particularly interesting and valuable to find an early predictor of the developmental trajectory in children with early imitation difficulties. We showed that the children with an imitation deficit at 2 years will either maintain their difficulties in a rather global way in the years to come, or show significant improvement in imitation but also in other areas in general. It would then be a question of finding which characteristics of the behavioral phenotype, or the interventions received allowed the positive evolution in some cases. One of the avenues to follow in this sense could be to explore earlier milestones of imitation than one measured in our study. Indeed, imitation is very linked to other skills such as social orientation, joint attention and social reciprocity as demonstrated by some authors (Girardot et al., 2009; McDuffie et al., 2007). Despite our unsuccessful attempt to distinguish our two groups in terms of social orientation using an eye-tacking task (Franchini et al., 2017), more investigations could be done in this sense with a younger sample.