The Early Development in NF1 (EDEN) study is a UK-based prospective longitudinal cohort study investigating early development in infants and children with NF1. The behavioural measures used in our study were part of a more comprehensive experimental protocol used for the EDEN study, and our data formed a proportion of the results obtained from EDEN. Our previous paper also describes the methods used in the EDEN study [15] .
Recruitment
Participants were enrolled through regional genetic centres and NF1 charities. Rolling recruitment was conducted between 2016 and 2019. Participants in the TD group were enrolled from a volunteer database for the Studying Autism and ADHD in at Risk Siblings (STAARS) study at the Centre for Brain and Cognitive Development, Birkbeck, University of London. These children had typical development and had not been diagnosed with a developmental disorder. The sample size calculation derived from our previous work comparing infants at high likelihood of developing autism to controls (e.g. n = 17, η2 = 0.17; n = 19, η2 = 0.16) [22, 23]. However it is important to note that this was based on detecting EEG biomarker differences rather than the behavioural measures utilised in this study.
Inclusion criteria for the NF1 cohort included (a) infant under 14 months of age at the time of recruitment (b) NF1 diagnosed via testing of cord blood samples or clinical diagnosis.
Inclusion criteria for the TD group included (a) infant under 14 months of age at the time of recruitment (b) no first-degree relatives with known genetic conditions, autism or ADHD (c) no parent-reported developmental issues in the child (d) full-term birth (gestational age at least 36 weeks).
Exclusion criteria for both groups included (a) conditions which might make it difficult for the infant to participate, such as physical complications of NF1 (b) significant hearing or visual impairments (c) significant prematurity (d) parents with significant learning difficulties or who were unable to give informed consent.
To offer maximum flexibility for participants, recruitment was offered up to the age of 14 months. Retention was variable across visits, which meant that the sample size varied at different assessment time points. However, participants could rejoin for later assessments if they were unable to attend at a particular timepoint.
35 children with NF1 and 29 TD participants were enrolled. 8 NF1 participants and 15 TD participants completed all 5 visits, 12 NF1 participants and 9 TD participants completed 4 visits, 9 NF1 and 2 TD participants completed 3 visits, 3 NF1 and 0 TD participants completed 2 visits, and 3 NF1 and 3 TD participants completed 1 visit. Further information on study numbers and attrition is outlined in the supplementary material (Supplementary Fig. 1).
Testing
Participants were assessed at 5, 10, 14, 24 and 36 months of age. The study assessments took place at the Division of Psychology and Mental Health, University of Manchester, and the Centre for Brain and Cognitive Development, Birkbeck, University of London. The NF1 participants at 5, 10, 14 and 24 months were tested at Birkbeck, and the University of Manchester at 36 months.
Prior written informed consent was obtained from the parent. Testing took place if the child was physically well and content. Assessments were carried out over 2 days for infants at 5, 10 and 14 months, to account for breaks and sleep schedules, and over one full day for the older participants at 24 and 36 months. Participant families were provided with reimbursement for expenses for travel, food and overnight stay in a hotel if required. A £20 gift card was offered as an incentive for each visit completion.
Measures
Table 1 summarises the measures used at each time point.
Maternal education
Maternal education was collected as part of a larger questionnaire ascertaining demographic factors. It was classified as either primary, secondary, undergraduate or postgraduate (1,2,3 or 4) (Table 2). We focused on maternal, rather than paternal, education level due to evidence suggesting that among core domains of socio-economic status (employment, income and education), maternal education is most strongly associated with a child’s cognitive development [24]. Maternal education has been shown to be significantly associated with trajectories of cognitive and adaptive functioning at 5, 10 and 14 months in this population [15] .
Cognitive and adaptive behavioural skills
The Mullen Scales of Early Learning (MSEL) [25], a standardised assessment for children aged up to 68 months, was used to assess cognitive functioning at all five time points. Five domains were assessed, including Visual Reception, Fine Motor, Receptive Language, Expressive Language skills (all measured as T-scores) and an Early Learning Composite (ELC) (Standard Score). T-scores range from 20 to 80 and the ELC standard score range from 49 to 155.
The Vineland Adaptive Behavior Scales (VABS) Third Edition [26], a parent-report questionnaire, was used to assess adaptive behavioural skills at all five time points. The standard scores of five domains were assessed, including Communication, Daily Living Skills, Socialisation, Motor skills and an Adaptive Behavior Composite score. Standard scores range from 20 to 160.
ADHD traits
The Child Behavior Checklist (CBCL) for ages 1.5–5 [27], a parent-report questionnaire, was used to assess ADHD traits (inattention/hyperactivity) at 24 and 36 month time points. T-scores were used for the DSM-orientated Attention Deficit Hyperactivity problems scale. T-scores of 70 are in the clinically significant range, and 65–69 are considered borderline [28] .
Autism traits
The Autism Diagnostic Observation Schedule (ADOS-2), a semi-structured assessment of social communication, social interaction and imaginative play for individuals suspected to have autism [29, 30], was utilised at 24 and 36 months. Based on the expressive language ability of the participants, either the Toddler module, Module 1 or Module 2 of the ADOS-2 were used at 24 and 36 months (Table 3).
All three ADOS-2 modules provide a score for the domains of Social Affect and Restricted and Repetitive Behavior, followed by a total score. For the Toddler module, separate algorithms are based on age and language ability. For our study, children at 24 months who produced fewer than 5 words during the ADOS-2 received the non-verbal 21–30 months algorithm, and children who produced 5 or more words received the verbal 21–30 months algorithm. Total scores were classified into ‘levels of concern’ for autism: no concern, mild-to-moderate concern (a score of 10 + for the non-verbal algorithm and 8 + for the verbal algorithm) or moderate-to-severe concern (a score of 14 + for the non-verbal algorithm and 12 + for the verbal algorithm). Luyster et al. suggest that at least 95% of children with Autism Spectrum Disorder and no more than 10% of typically developing children would fall into the two groups suggesting clinical concern on the ADOS Toddler module (mild-to-moderate and moderate-to-severe). This gives an instrumental sensitivity of at least 95% and a specificity of more than 90% [30] .
For Module 1, children with some language who gain a score of 8 + receive a classification of autism spectrum, and a score of 12 + gives a classification of autism. For children with few to no words, a score of 11 + gives a classification of autism spectrum and a score of 16 + gives a classification of autism. For Module 2, children less than 5 years of age who receive a total score of 7 + receive a classification of autism spectrum and a score of 10 + gives a classification of autism. Comparison scores can also be calculated to indicate level of autism-related symptoms, although analysis of this data was beyond the scope of this paper.
Coding was carried out from videos, with an inter-rater reliability of 79.1% for the NF1 cohort.
The ADOS-2 was administered from 24 months of age. Although the ADOS-2 Toddler module can be used for children from 12 months of age, Luyster et al. (2009) recognised in their development of the instrument that their final sample would include very few children in the autism group at this lower cutoff due to the frequency of developmental delay in children with autism [30] .
During 2020–2022, the Covid-19 pandemic required some assessments to be carried out virtually, as a result of social distancing legislation. For some participants, the Brief Observation of Symptoms of Autism for Minimally Verbal children (BOSA-MV) was utilised. This is an observational measure designed to be administered remotely [31]. Based on the expressive language ability of the participants, either the Toddler module or Module 1 of the BOSA-MV were used at 24 and 36 months (Table 3).
Both BOSA-MV modules provide a score for the domains of Impairment in Social Communication and Social Interaction, and Restricted and Repetitive Behaviors, followed by a total score. This gives a range of concern for autism of little to no concern, mild-to-moderate concern and moderate-to-severe concern. Dow et al. recommend a cut-off of 6 for Autism Spectrum Disorder for the BOSA-MV toddler module (corresponding with the moderate-to-severe concern category) and a score of 5 as a cut-off for the BOSA-MV Module 1 (corresponding with the mild-to-moderate or moderate-to-severe category) [31]. For the Toddler module, this gives an instrumental sensitivity of 96% and a specificity of 83%. For Module 1, this provides a sensitivity of 91% and a specificity of 100% (although the authors acknowledge that their non-autism sample was small when developing the BOSA-MV) [31].
Table 3 outlines the participant numbers at each timepoint who were administered ADOS-2 versus BOSA-MV.
The Autism Diagnostic Interview-Revised (ADI-R), an investigator-based semi-structured interview for parents [32], was carried out at 36 months. Scoring is based on two algorithms, depending on whether the subject is verbal or non-verbal. Four subscales are produced: A—qualitative abnormalities in reciprocal social interaction, B—qualitative abnormalities in communication, C—restricted, repetitive and stereotyped patterns of behaviour and D—abnormalities of development evident at or before 36 months. Each subscale has a cut-off for autism (A = 10, B = 8 if verbal and 7 if non-verbal, C = 3 and D = 1) [32] .
The ADI-R was utilised at the 36 month time point. Psychometric analyses have determined that for children over 36 months of age, the algorithms differentiate children with autism from those with non-spectrum disorders with a high sensitivity and specificity of over 90% [33] .
The ADOS-2/BOSA and ADI-R assessors were not blind to the participant’s group (NF1 versus TD children), however videos and interviews were double coded and this second coder was blind as to the participant’s condition.
Classification of autism
In our paper, the following thresholds are used on the ADOS-2 and BOSA-MV to determine autism traits at 24 months of age:
OR
OR
OR
Participants were assigned a research instrumental classification of autism at 36 months of age if they met threshold on either the ADOS-2 or BOSA-MV, in addition to meeting threshold on the ADI-R:
OR
OR
ALONG WITH
-
meeting threshold for subscale A and coming within one point of B, or meeting threshold for B and coming within one point of A on the ADI-R, as suggested by Risi et al. [34].
Risi et al. provide a rationale for combining the ADI-R with the ADOS-2 at 36 months of age, giving a sensitivity of 61.1% for autism detection and a specificity of 87.7% for the combination of the ADOS-2 with ADI-R criteria for A and B as outlined above [34]. To our knowledge, there have been no sensitivity and specificity estimates for the combination of BOSA-MV and ADI-R, due to the relative recency of the BOSA-MV.
A research classification of autism was not given at 24 months, as the ADI-R was not utilised at 24 months in this study, and best diagnostic practice combines parent report with objective assessments.
Statistical analyses
Statistical analyses were performed using IBM SPSS Statistics 28.0.0.0. Linear mixed modelling was used to analyse the change in cognition (MSEL), adaptive behaviour (VABS) and ADHD traits (CBCL DSM-ADHD subscale) over time. For each subscale, overall group differences were modelled using fixed effects (group, timepoint and sex) and random effects (ID – individual variation). Maternal education was included as a co-variate within the model [15] .
In all models, sex was non-significant (Table 4). Age in days was not included in the model as a fixed effect, as this had already been corrected for by using age-corrected T scores (MSEL and CBCL) and age-corrected standard scores (VABS). Post hoc T-tests were carried out to further explore group differences on the MSEL and VABS at each timepoint.
Missing data was imputed using the maximum likelihood option. Earlier time points were imputed for children that joined later in the study and data from subsequent missed sessions were also imputed. 29% of the NF1 dataset was imputed and 17% of the TD dataset was imputed (Supplementary Fig. 1).
A p value of below 0.05 was determined to be significant for the MSEL and VABS. For post-hoc tests, a Bonferroni corrected p value of below 0.01 was determined to be significant. This was based on the use of T-tests at 5 time points for each measure (0.05/5).
Pearson Chi-squared tests were carried out for proportion of participants meeting autism threshold on the ADOS-2/BOSA/ADI-R. Mann–Whitney non-parametric analyses were used to compare ADI-R subscale means due to non-normality.
Add Comment