Participants
Neurotypical (N = 151) and neurodiverse (N = 111) children and adolescents were recruited by the Healthy Brain Network (HBN), an initiative to create an inventory of biological markers of mental health disorders in the developing brain [24]. In order to meet inclusion criteria, one must identify as a male or female person between the ages of five and 21, with parents (or a caregiver) who are capable of providing verbal and written consent. Youth between five and 18 years of age must provide verbal assent, which is the clear expression of agreement to participate. Children with ASD were diagnosed by a clinician using the Autism Diagnostic Observation Schedule (ADOS-2) and the Kiddie Schedule for Affective Disorders and Schizophrenia-Children’s Version (K-SADS). Youth with no diagnosis resulting from the K-SADS were classified as TD. Exclusion parameters included acute safety concerns, cognitive or behavioural impairments that may interfere with the child’s participation (e.g., inability to fully understand or communicate responses to questionnaire items, as discerned by the examiner and informed by clinical expertise), or medical issues that may confound brain scan results. Additionally, all individuals taking stimulants were required to document the medication taken on the day of their participation given that stimulant use may influence one’s performance on cognitive and behavioural measures.
Both TD and children and adolescents with ASD completed testing over the course of two years. Measures of behaviour, family structure, stress and trauma, substance use, and language were collected, as well as physiological and diagnostic assessments. The data used in the present study were obtained from the Child and Mind Institute, Healthy Brain Network (https://data.healthybrainnetwork.org/main.php). The current work utilized 262 observations comprised of data from TD children and adolescents as well as those with a primary clinical diagnosis of ASD. Ethical approval was obtained by the Chesapeake Institutional Review Board (IRB). Written and verbal informed consent was collected from adults prior to data collection, as well as verbal assent in participants 17 and under. The research was conducted in accordance with the Declaration of Helsinki.
Materials and measures
Demographics
Through self and parent-report questionnaires, demographic information was collected including biological sex, age, and parental relationship status [24].
Socioeconomic status
The Barratt Simplified Measure of Social Status (BSMSS) was administered to measure socioeconomic status (SES) [25]. Marital and employment status, educational accomplishments, and occupational prestige are used to orient one’s SES. The BSMSS is strictly ordinal for the purpose of clustering participants into like-groups. It is important to note that the BSMSS does not indicate one’s social class (e.g., middle-class), and reliability statistics cannot be applied. For example, level of education choices are less than grade seven (score = 3), less than ninth grade (score = 6), less than 11th grade (score = 9), high school graduate (score = 12), at least one year of college (score = 15), college education (score = 18), and graduate degree (score = 21). Depending on the sample scores obtained, a mean value will be determined within the sample. This mean value will be used to group participants into similar educational attainment categories.
Diagnostic assessments
The K-SADS is a semi-structured interview used to measure current and past symptoms of anxiety, mood, psychotic, and disruptive behavioural disorders in children and adolescents ages six to 18 [26]. Questions such as “Have you been having any worries lately?” and “Did you look forward to doing the things you used to enjoy?” are asked by the examiner. Questions are designed to elicit responses that indicate the presence of depressive disorders, anxiety disorders, eating disorders, conduct disorders, substance use disorders, and other mental health concerns. Items are scored from zero (no information) to three (feels the queried symptom most of the day more days than not). Interviews were conducted with both participants and parents in order to determine or rule out a clinical diagnosis. Children and adolescents suspected of having ASD symptoms were referred for further diagnostic assessment using the Autism Spectrum Screening Questionnaire (ASSQ) and the ADOS-2.
The ASSQ is among the most widely used tools to identify children and adolescents ages six to 17 who may have an ASD [27]. In this instrument, 27 items are completed by parents and/or teachers of youth displaying symptoms that are characteristic of ASD [26]. Participants are asked to state no (0), somewhat (1), or yes (2) to each question with regard to their child or student. The ASSQ contains questions such as “this child has markedly unusual posture” and “this child accumulates facts on certain subjects but does not really understand the meaning” [27]. Parents and/or teachers are asked to report based on how this child differs from other children or adolescents of the same age [27]. Total scale scores range from zero to 54, with higher scores indicating the presence of more severe ASD symptoms. Research has demonstrated that there is a 90% positivity rate of an ASD diagnosis among those who score 13 or above [27, 28]. The ASSQ demonstrates excellent test–retest reliability (r = 0.90, p < 0.001) and good inter-rater reliability (r = 0.79, p < 0.001) between parents and teachers. Additionally, this assessment has been found to have 91% sensitivity, and 86% specificity, indicating that the ASSQ produces few false negatives (sensitivity) and few false positives (specificity).
Children and adolescents suspected of having an ASD were evaluated using the ADOS-2. The ADOS-2 is a play and activity-based assessment that allows for the real-time observation of ASD symptoms and behaviours [29]. The test is designed for youth ages 12 months into adulthood who do not have significant sensory or motor impairments. An appropriate module is selected based upon the participant’s age, language, and level of development. The ADOS-2 is comprised of various activities, including a construction task, make-believe play, joint interactive play, a demonstration task, a description of a picture, and understanding of friends, relationships, and marriage. Activities are completed and scored using a coding system that examines the presence or absence of abnormality on a given task. The ADOS-2 is considered the “gold standard” in the diagnosis of ASD.
Autism symptom severity
The Social Communication Questionnaire (SCQ) developed by Rutter and colleagues (2003) is used to assess communication skills among youth with and without ASD [30]. Forty items are completed by the primary caregiver of the child or adolescent aged four or older. Caregivers are asked questions such as “does she/he ever have any interests that preoccupy her/him and might seem off to other people (e.g., traffic lights, drainpipes, etc.)?” or “does she/he nod her/his head to indicate yes?”. Items are answered using a yes or no response system, whereby no = zero and yes = one. The first instrument item regarding the use of language is not scored, and is used to determine whether the abnormal language section is applicable to the specific child. As such, total scale scores range from zero to 39 (if the abnormal language section is completed) or zero to 32 (if not). A total score is obtained based upon the sum of all items, and a cut-off score of 15 is suggested to indicate those with more severe ASD symptoms who may need further clinical evaluation. Research has found the SCQ to have strong internal consistency (α = 0.80), and test–retest reliability ranging from r = 0.87 to 0.96 (p < 0.0001) [31].
The Social Responsiveness Scale (SRS-2)-School Age is comprised of 65 items to identify the presence and severity of social impairment related to ASD [32, 33]. The SRS-2 is completed by parents or teachers of youth ages four to 18, and asks responders to consider statements concerning social awareness, social cognition, social communication, restricted interests and repetitive behaviour, social motivation, and social interaction. The SRS-2 includes statements such as “expressions on his or her face don’t match what he or she is saying” and “my child knows when he or she is talking too loud or making too much noise”. Each item is scored on a Likert scale ranging from not true (0) to almost always true (3) and scores are summed for each subscale as well as a total scale score. Higher scores indicate greater severity of social skill deficits, with T scores of 76 or higher representing severe deficits and those under 60 falling within the typical range. The SRS-2 demonstrates strong rest-retest reliability (r = 0.80 to 0.95, p < 0.001) and interrater reliability (r = 0.75 to 0.77, p < 0.001). Total internal consistency is excellent, at α = 0.95.
The Repetitive Behaviors Scale-Revised (RBS-R) provides a quantitative, continuous measure of repetitive behaviours that are often attributed to ASD [34, 35]. Stereotypic behaviour, self-injurious behaviour, compulsive behaviour, ritualistic/sameness behaviour, and restricted interests are measured using 43 items. There is also an additional item that asks respondents to rate the overall severity of behaviours on a range of zero to 100. The RBS-R is completed by parents of participants ages six to 17, and includes items such as “my child flaps hands, wiggles or flicks fingers, claps hands, waves or shakes hands or arms” and “my child hits or bangs head or other body part on the table, floor or other surface”. Items are rated on a 4-point Likert scale (0 = behaviour does not occur, 1 = behaviour occurs and is a mild problem, 2 = behaviour occurs and is a moderate problem, and 3 = behaviour occurs and is a severe problem) and responses should be based upon behaviour within the past month [36]. Scores for each subscale are totaled to determine the area of greatest concern. Subscale internal consistency is good, ranging from r = 0.73 to 0.95 (P < 0.001) [36, 37]. Both participants with and without ASD completed ASD diagnostic assessments.
Anxiety symptoms
The Screen for Child Anxiety Related Disorders-Parent Report (SCARED-P) contains 41 items assessing a child’s (aged eight to 18) anxiety symptoms within the last three months [38] The SCARED-P uses a 3-point Likert scale (0 = not true, 1 = somewhat true, and 2 = very true) to examine five subcategories of anxiety including generalized, separation, social, panic/somatic, and school avoidance. A sample-item from the separation anxiety subscale is My child worries about sleeping alone. Possible scores range from zero to 82, with a total score of 25 or more indicating the likely presence of an anxiety disorder. The SCARED-P demonstrates high reliability across the literature (average α = .95), even among ASD populations [38].
Emotional and behavioural problems
The Strengths and Difficulties Questionnaire (SDQ) is a 25-question Likert assessment of behavioural and psychosocial concerns in children ages two to 17 [39]. This questionnaire is used to assess internalizing (e.g., internalizing, depression) and externalizing behaviours (e.g., aggression). Five subscales consisting of emotional symptoms, conduct problems, hyperactivity/inattention, peer relationship problems, and prosocial behaviour are included in the questionnaire. The SDQ-internalizing subscale (SDQ-I) consists of both the emotional and peer relationship problem subscales. The emotional problems subscale measures depression, worry, fear, nervousness, and somatic symptoms. In the present study, the SDQ-I was used to assess both anxiety and depressive symptoms. Example statements from the emotional problems subscale include I worry a lot and I am often unhappy, depressed or tearful. A sample statement from the peer relationship subscale is Other children or young people pick on me. Questions are answered using a 3-point system (0 = not true, 1 = somewhat true, and 2 = certainly true). Both children and parents completed this inventory. SDQ-I scores can be calculated by adding up the scores for the emotional and peer relationship problem subscales, resulting in a score range between zero and 20. An SDQ-I score of 20 indicates severe emotional and relational concerns. Each of the SDQ-I subsections, emotional and peer relationship problems, are scored from zero to 10. A score of six or higher on the emotional problems subscale indicates very high difficulties, while a score of four or higher on the peer relationship problems subscale indicates severe social deficits. The SDQ demonstrates good reliability (α = .75).
In addition to the SDQ, the Child Behavior Checklist (CBCL) was completed by the guardians or teachers of all participants to further assess specific behavioural and emotional concerns. Two versions of the CBCL were utilized in the present study, namely the preschool version and the version for youth ages six to 18 (CBCL/6-18) [40]. The CBCL-preschool version was completed by guardians of children up to five, and includes 99 items regarding aggressive behaviour, attention problems, somatic complaints, sleep issues, anxious and/or depressed behaviour, oppositional defiant problems, attention deficit/hyperactivity issues, and pervasive developmental concerns. Adults completing the CBCL-preschool version were asked questions such as whether or not this child “avoids looking others in the eye”, is “cruel to animals”, and “is disobedient”. Similarly, the CBCL/6-18 asks 113 questions such as whether or not this child or adolescent “can’t sit still, is restless or hyperactive”, “fears going to school”, and “screams a lot”. The subscales measured on the CBCL/6-18 are similar to the preschool version, including anxious, depressed, and/or withdrawn behaviour, somatic complaints, thought problems, attention issues, social concerns, rule breaking behaviour, aggressive behaviour, and other significant problems of note. Both versions of the CBCL are scored using a three-point Likert scale ranging from zero (not true) to two (very true). Subscale scores are totalled along with a total scale score online, where higher scores indicate more frequent and severe symptoms. All versions of the CBCL demonstrate excellent total scale internal consistency (α = .97), with strong reliability on internal (α = .90) and external (α = .94) subscales [41]. All subscales have shown strong reliability, ranging from α = .71 to .89 for both test versions [42].
Tolerance of stress
The Distress Tolerance Scale (DTS) measures one’s capacity to experience and withstand negative psychological states [43]. The DTS consists of 15-items that explore four subscales. First, the absorption subscale measures the amount of attention one shifts to negative emotions and includes items such as “when I feel distressed or upset, I cannot help but concentrate on how bad the distress actually feels”. The appraisal subscale uses statements such as “my feelings of distress or being upset are not acceptable” to measure one’s subjective appraisal of distress. Statements such as “I’ll do anything to avoid feeling distressed or upset” make up the regulation subscale, which assesses one’s efforts to alleviate distress. Finally, the tolerance scale measures one’s perceived ability to tolerate emotional distress and includes statements such as “I can’t handle feeling distressed or upset”. Items are rated on a five-point Likert scale ranging from one (strongly disagree) to five (strongly agree). In addition to subscale scores, a total scale score is calculated by averaging each subscale mean. Higher scores indicate a greater ability to tolerate distress. The DTS exhibits strong test-retest reliability (r = .82 to .85, p < .001) over a six month period and displays excellent internal consistency (α = .93) [43, 44].
ADHD symptoms
The Strengths and Weaknesses Assessment of Normal Behavior (SWAN) Rating Scale for ADHD consists of 18 items used to assess both the presence and severity of ADHD symptoms [45]. The SWAN is completed by a caregiver, teacher, and/or physician on behalf youth ages six to 17. Adults are asked to answer each item with respect to how the youth in question compares to their peers of the same age over the past month. Questions around sustained attention, ability to remain seated and focused, and turn-taking are included in the assessment. Items are scored on a 7-point Likert scale (-3 = far above average, -2 = above average, -1 = slightly above average, 0 = average, 1 = slightly below average, 2 = below average, and 3 = far below average), with weaknesses scored positively and strengths scored negatively. An inattention average and hyperactivity average score are calculated by totalling each item score and dividing by the number of items in each subscale. An additional SWAN total average score is calculated based on each of the two subscales. The more attention or activity problems displayed by the child, the lower their total SWAN score. The SWAN demonstrates excellent total scale internal consistency (α = .88 to .95) with subscale internal reliability ranging from r = .72 to .90 [4, 46].
Parental practices
Parents completed the Alabama Parenting Questionnaire (APQ) [47].
The APQ consists of 42-Likert scale items examining parental involvement, positive parenting, poor monitoring/supervision, inconsistent discipline, and corporal punishment. Additional discipline practices including reasoning, ignoring, loss of privileges, time-out, and extra work are also assessed. Participants answer each question on a scale system (1 = never, 2 = almost never, 3 = sometimes, 4 = often, and 5 = always) with regards to how often each scenario occurs in their household. An example scenario from the inconsistent discipline subscale is Your child is not punished when they have done something wrong. All items are summed to obtain a total scale score ranging from 42 to 210. Total scores that are one standard deviation or greater above or below the mean are considered concerning. For example, parents who score low on involvement and monitoring, and high on inconsistent discipline and corporal punishment would be flagged as concerning over parents who receive average scores on these measures (e.g., average total score = 126). Good reliability across the literature is demonstrated with the APQ (α = 0.82).
Parental stress
The Parenting Stress Index (PSI-IV) is composed of 36-items assessing stress levels in parents of children between one month and 12 years old [48]. There are three domains, including parent distress (PD), parent-child dysfunctional interaction (P-CDI), and difficult child (DC). Within the PD domain are seven subscales, including competence, isolation, attachment, health, role restriction, depression, and spouse/parenting partner relationship. An example from the PD is I feel trapped by my responsibilities as a parent. The P-CDI contains questions regarding the parent/child relationship, the parent’s perception of the child, and whether or not their expectations have been met in parenthood. A sample P-CDI statement is When I do things for my child I get the feeling that my efforts are not appreciated very much. The DC domain contains six subscales, examining distractibility/hyperactivity, adaptability, reinforces parent, demandingness, mood, and acceptability. A statement from the DC domain is My child makes more demands of me than most children. The PSI-IV is a Likert scale (1 = strongly disagree, 2 = disagree, 3 = not sure, 4 = agree, and 5 = strongly agree), and each value can be tallied to get the total-stress score for the scale. Total stress scores range from 36 to 180, with scores of 36 to 100 representing typical stress, and scores 100 or higher as clinically significant distress. The PSI-IV demonstrates excellent reliability (α = 98).
Cognitive ability
The Wechsler Intelligence Scale for Children-Fifth Edition (WISC-V) was used to assess cognitive ability in the children [49, 50]. Subtests covering a range of abilities, including verbal comprehension, visual spatial skills, and processing speed, were completed by all children and adolescents ages six to 16. Average scores range from 90 to 109, with exceptionally low scores falling below 79 and high above 120. The WISC-V is often used to identify intellectual exceptionalities among school-age children, such as giftedness or IDs. Test scores have been repeatedly validated as useful for identification, placement, and resource allocation. Furthermore, the WISC-V is supported by strong split-half reliability (r = 0.96, p < 0.001), subtest reliability (r = 0.80 to 0.94, p < 0.001), and test–retest reliability (r = 0.71 to 0.90, p < 001).
Procedure
Email and poster advertisements were distributed among community members, educators, care providers, and parents in the New York City area [51]. Data collection for the study began in 2015. Advertisements stressed the value of participation for families whose children could benefit from learning accommodations at school, such as an individualized education program (IEP). All participants were screened over the phone by HBN researchers prior to answering questionnaires to ensure inclusion criteria was met.(i.e., between the ages of five and 21, with parents (or a caregiver) who are capable of providing verbal and written consent). Information regarding psychiatric and medical history was collected during the screening call, including the use of stimulant medication. Those participants enrolled in the study following the screening call were administered a semi-structured diagnostic interview by an HBN licensed clinician. Appropriate follow-up measures were completed when necessary by participants and/or their parents depending on the child’s age. For example, if the diagnostic interview yielded a suspected language disorder, the Clinical Evaluation of Language Fundamentals (CELF) would be completed by the participant in question. Participants completed all language and intelligence testing with supervision from a qualified clinician. Self-administered questionnaires, such as the CBCL and the APQ, were completed using an online patient portal system called NextGen. All assessments were completed over the course of four visits. Participants were compensated for their time.
Analysis
Statistical analyses were performed using R (Version 4.2.0) software [52]. To address the first aim to identify anxiety profiles, an unsupervised machine learning algorithm was used to generate a self-organizing map (SOM) [53]. A SOM is an unsupervised artificial neural network used to map and cluster large data sets, and has been used previously to characterize children with ASD [54,55,56]. Importantly, the SOM’s algorithm attempts to learn about the underlying structure of data itself, rather than which data corresponds to predefined groups, thereby allowing for the development of unique data-driven subgroups. This approach may be used to tease apart different anxiety subtypes that are embedded within clinically-recognized NDDs, thereby accounting for the heterogeneity of anxiety symptom presentation among children and adolescents with ASD.
In the present study, the SOM contained a node (or neuron) representing the unique anxiety profile of each participant within a two dimensional plane. Initial data points, referred to as input data, were used to generate the first several nodes within the model. This served as the framework for the model, within which the remainder of the data points were statistically compared. The nodes with the closest weight vector to the input data were selected by the SOM as the best-matching unit. This process allowed for all data points (nodes) to be appropriately categorized in the model. The statistical model was trained using 262 observations and a learning rate of α = 0.05, which is a standard SOM value [53]. The number of resulting clusters was informed through visual inspection of elbow graphs prior to SOM development.
Scores from the ASSQ, SCQ, SRS, SWAN, APQ, SCARED-P, and SDQ were used to create a five-by-five hexagonal topology [53, 57] resulting in three clusters. A hexagonal method is used to preserve topographical distances between nodes and reduce distortions from mapping. Participant age, sex, family SES, intelligence, and parental age were statistically controlled for throughout all analysis procedures by including them as covariates in a separate linear regression model for each variable of interest (that is, the ASSQ, SCQ, SRS, SWAN, APQ, SCARED-P, and SDQ) and extracting the residuals from each model to carry forward as the adjusted values. In the final model, a node representing the unique anxiety profile of each participant was mapped in a two-dimensional framework. Each cluster was formulated irrespective of diagnosis. Nodes that fell within a similar location were representative of participants with a similar anxiety profile [53]. For example, if two participants with ASD both display high levels of social anxiety and low separation anxiety, it was expected that their respective nodes would fall in close proximity within the model. The goal was to determine if diagnosis was predictive of cluster (or group) membership, regardless of the broad spectrum of characteristics displayed within each group. For example, the anxiety profile of a child with high anxiety may be characteristically distinct from those with low anxiety.
To address the second aim of identifying anxiety factors that determined cluster membership, recursive feature elimination (RFE) was conducted using all measures of anxiety, consisting of the five subscale scores and the overall score from the SCARED-P, as well as the nine subscale scores from the SDQ. Using RFE, the anxiety measures that contributed the most to the differences between cluster membership were identified. RFE is a statistical process in which the key features that contribute to the SOM model are isolated [58]. RFE scores and ranks features by permutation importance and removes those features with limited input to the model. This process is repeated until one feature with the largest contribution to the model remains. Permutation importance considers a variable of great importance only with respect to improving the predictive accuracy of the model.
Add Comment