A single-arm pre–post design was used for this proof-of-principle pilot study to assess signals of change in autistic children and their caregivers after 12, 1-h NDBI-informed caregiver coaching sessions implemented by ECD practitioners. The study was a collaboration between the Centre for Autism Research in Africa at the University of Cape Town and the Duke Centre for Autism and Brain Development at Duke University. Caregiver–child dyads were recruited from Western Cape Provincial Department of Education waiting lists for autism special education services. In South Africa, public special education schools are the third level of school placement in the education system, intended for learners with high-support needs (Nthibeli et al., 2022).

As outlined in the pilot study by Rieder et al. (2023), inclusion criteria for autistic children and their caregivers were as follows: (1) the child was between 18 and 72 months of age; (2) the child’s family primary language was isiXhosa, isiZulu, Afrikaans or English; (3) the participant’s self-declared race was black person, coloured person or Indian person; (4) the child lived within an area served by the recruitment sites and was on the Provincial Department of Education waiting list; (5) the child met the DSM-5 criteria for autism spectrum disorder informed by the Autism Diagnostic Observation Schedule Second Edition (ADOS-2) administered by research-reliable clinicians (the child entered the study with diagnosis of autism, which was confirmed in the study); and (6) the caregiver was ≥ 18 years. Exclusion criteria for caregiver–child dyads were as follows: the child had (1) significant sensory or motor impairments, major physical abnormalities, the presence of a neurological disorder of known aetiology (e.g. Fragile X syndrome), history of serious head injury and/or neurological disease, or (2) the caregiver–child dyads were unable to attend assessments and the 12 coaching sessions. All families lived within a metropolitan area.

After consenting to participate, the dyads were video-recorded during their baseline assessment, using a standardised CCI protocol. The CCI consisted of two consecutive 6-min free play sessions referred to as ‘Part one’ and ‘Part two’. In Part one, the caregiver was instructed to remain seated, while their child interacted with play materials in the room. In Part two, the caregiver was instructed to interact with their child in a way they would at home. After the caregiver–child dyads had completed 12 coaching sessions, the dyads attended a follow-up assessment. During the follow-up assessment, the standardised CCI protocol was repeated. The video recordings of the CCI were uploaded to Duke Box (a Health Insurance Portability and Accountability Act-compliant transfer and storage platform hosted by Duke University). All CCI occurred in the same room (see Figure 1 for material set-up for the CCI protocol).

Sociodemographic characteristics of the child (age, self-declared race by caregiver, gender and home language) and caregiver (age, relationship to child, level of education, employment status and household income) were assessed. Autism-related behaviours were assessed using the ADOS-2, a semi-structured play-based assessment administered by a research-reliable clinician (Lord et al., 2012).

The JERI is a behavioural coding system used to characterise joint engagement, communication dynamics and shared topics between young children and their caregivers (Adamson et al., 2004, 2019). Sixteen JERI items were included in this study that capture descriptions of child joint engagement and social communication, caregiver’s ability to support and facilitate child activity and dyadic interaction (see Table 1). Joint Engagement Rating Inventory items are coded on a 7-point Likert scale with descriptive anchors at ratings of 1, 4 and 7 from videos of CCI (see Table 1). Data were coded by two research-reliable JERI raters who were blinded to whether CCI recordings were pre- or post-intervention. As per coding convention, a master rater was randomly assigned CCI videos for double coding to determine inter-rater reliability with the research-reliable JERI raters.

Data analysis was conducted using IBM SPSS Statistics (version 27) (IBM Corp, 2020). Descriptive statistics were used to summarise caregiver and child demographic variables using median with interquartile range, range or frequency with percentages, as appropriate. As per convention, JERI inter-rater reliability was assessed using weighted Cohen’s kappa, observer estimated accuracy and percent agreement within one scale point. Weighted Kappa values were interpreted as follows: 0–0.20 indicated no agreement, 0.21–0.39 minimal agreement, 0.40–0.59 weak agreement, 0.60–0.79 moderate agreement, 0.80–0.90 strong agreement and above 0.90 as almost perfect agreement (Cohen, 1968; McHugh, 2012; Sim & Wright, 2005). The observer estimated accuracy determined whether weighted Kappa values were adequate, with the goal of 80% or greater. The inter-rater reliability goal for percentage agreement within one scale point was 80%.

The Wilcoxon signed-rank test for paired samples was used to compare change in child behaviours, caregiver behaviours and dyadic engagement across the two time points (pre- versus post-intervention) because we did not assume a normally distributed dataset given the small sample size and heterogeneity. Given the small sample size in this proof-of-principle study, visual summaries of each JERI rating (pre- versus post-intervention) were also generated to show how participant scores changed over time. Line graphs were created that represented the individual participant scores with box plots indicating medians and ranges of scores across the group.

This study was reviewed and approved by both ethical review boards of the University of Cape Town Human Research Ethics Committee (reference number: 301/2015, 468/2019) and Duke University Institutional Review Board (reference number: Pro00103045, Pro00064533).

Full caregiver–child participant characteristics and participants’ flow are outlined in the pilot study by Rieder and colleagues (Rieder et al., 2023), with key relevant details summarised next. Twelve dyads consented to participate in the study and 10 completed the coaching intervention (one withdrew from the study after three coaching sessions, and one was lost to follow-up after all coaching sessions were completed). Median child age at baseline was 53 months. Eleven of the 12 children were male. To inform DSM-5 autism diagnosis, an ADOS-2 Module 1 was completed with 11 children and the ADOS Module 2 with the remaining child. Six of the primary caregivers were mothers, four were fathers and two were grandmothers. Five caregivers completed Grade 12, three obtained a post-Grade 12 diploma and three completed tertiary education. Nine caregivers were married. In terms of financial status, one caregiver reported they were ‘struggling’, six reported they were ‘just getting by’ and five reported they were ‘doing okay’. Ten caregivers reported speaking English to their children at home and two caregivers reported using another language (but did not specify which one).

Inter-rater reliability for each JERI item is outlined in Table 2. Weighted Kappa values for 13 of the 16 JERI items ranged from 0.62 to 1, indicating moderate to strong agreement between raters. The observer estimated accuracy of weighted Kappa values, ranged between 83% and 99%. Joint Engagement Rating Inventory items with moderate to strong agreement included unengaged, object engagement, joint engagement, symbol-infused joint engagement, responsiveness, expressive language level and use, attention to caregiver, following in on child’s focus, affect, communicative temptations and fluency and connectedness. Percentage agreements within one scale point ranged between 75% and 100% across JERI items.

Joint Engagement Rating Inventory pre–post results for caregiver, child, child activity and dyadic item categories are summarised below. Wilcoxon signed-rank tests are presented in Table 3. Visual summaries of individual participant scores and box plots are shown in Figure 2 to Figure 5.

Overall, across caregiver JERI items, no significant pre- versus post-intervention differences were detected (see Table 3). However, visual summaries indicate variation in participant-level responses (see Figure 2a–e). In scaffolding, four caregivers scored higher, three caregivers scored the same and three caregivers scored lower post-intervention (Figure 2a). For following in on child’s focus five caregivers scored higher, two caregivers scored the same and three caregivers scored lower post-intervention (Figure 2b). In communicative temptations, six caregivers scored higher, two caregivers scored the same and two caregivers scored lower post-intervention (Figure 2c). For language facilitation, six caregivers scored higher, three caregivers scored the same rating at both time points, and one caregiver scored lower post-intervention (Figure 2d). Finally, caregiver affect results indicated that six caregivers scored the same ratings at both time points and four caregivers scored higher post-intervention (Figure 2e).

Across child JERI items, one significant pre- versus post-intervention difference was detected for symbol-infused joint engagement (z = −2.46, p = 0.01) (Table 3). Visual summaries indicate individual variation in responses (Figure 3). For unengaged, two children scored higher and the remaining eight children scored the same post-intervention (Figure 3a). In object engagement, two children scored higher, four children scored the same and four children scored lower post-intervention (Figure 3b). For joint engagement and supported joint engagement, two children scored higher, five children scored the same ratings at both time points and three children scored lower post-intervention (Figure 3c and Figure 3d). In coordinated joint engagement, one child scored higher, eight children scored the same and one child scored lower post-intervention (Figure 3e). Finally, for symbol-infused joint engagement, where a significant change post-intervention was detected, seven children scored higher post-intervention and three children scored the same ratings at both time points (Figure 3f).

Across child activity JERI items, one significant pre- versus post-intervention difference was detected for expressive language level and use (z = −2.16, p = 0.03) (Table 3), with visual summaries indicating individual-level variations across items (Figure 4). For child responsiveness to partner’s communication, three children scored higher post-intervention, four children scored the same at both time points and three children scored lower post-intervention (Figure 4a). For expressive language level and use, six children scored higher, three children scored the same rating at both time points and one child scored a lower post-intervention (Figure 4b). Finally, for attention to caregiver, three children scored higher, three scored the same at both time points and four scored lower post-intervention (Figure 4c).

Overall, across dyadic JERI items, no significant pre- versus post-intervention differences were detected (Table 3). Visual summaries indicate variation in dyad responses for fluency and connectedness, with five caregiver–child dyads scoring higher, four scoring the same at both time points and one scoring lower post-intervention (Figure 5a). For shared routines and rituals, visual summaries did not indicate individual score variation, with all dyads scoring the same at both time points (Figure 5b).