The Adapted ADOS: A new module set for the assessment of minimally verbal adolescents and adults.

Abstract

Few measures are appropriate to assess autism symptoms in minimally verbal adolescents and adults. The Autism Diagnostic Observation Schedule, 2^nd Edition (ADOS-2, Lord et al. 2012) Modules 1 and 2 were designed and validated with children whose spoken language ranges from few to no words to phrase speech. This study describes the development and initial validation of the Adapted-ADOS (A-ADOS), which includes tasks, materials and behavioral codes modified to be suitable for assessing older minimally verbal individuals. A-ADOS algorithms exhibit comparable sensitivity and improved specificity relative to ADOS-2 Modules 1 and 2. Although further validation is needed, the A-ADOS will facilitate research to further understanding of minimally verbal adults and symptom trajectories across the lifespan.

There is increasing attention to understanding autism spectrum disorder (ASD) in adulthood. Specific calls by stakeholders and scientists have been made for research to include individuals across the range of abilities, including those who are minimally verbal (Tager-Flusberg & Kasari, 2013; Jack & Pelphrey, 2017; Interagency Autism Coordinating Committee, 2017). At this time, however, few measures are appropriate to assess ASD symptoms in older minimally verbal children, adolescents or adults (Kasari, Brady, Lord, & Tager-Flusberg, 2013). The Autism Diagnostic Observation Schedule, 2^nd Edition (ADOS-2; Lord et al. 2012) includes Modules 1 and 2 for use with children with limited language; however, the materials and tasks were not designed for adolescents and adults (Berument et al., 2005). Moreover, studies to derive the algorithms only included individuals up to age 12 (Gotham, Risi, Pickles, & Lord, 2007; Gotham et al., 2008), leaving the interpretation of these modules with older individuals unclear. In a study utilizing the ADOS-2 in adults with intellectual disability, Sappok and colleagues (2013) reported high sensitivity (95%) and low specificity (50%) in a sample of 34 adults with ASD and 20 adults with non-ASD diagnoses. Their study, however, included adults of varying levels of ID that spanned the four ADOS-2 modules and did not report diagnostic validity for each module individually.

Longitudinal studies of parent-report data suggest that young adults may demonstrate fewer symptoms compared to childhood (e.g., Bal et al., 2019; McGovern & Sigman, 2005; Shattuck et al., 2007), suggesting a need to develop tools designed to assess the specific characteristics of ASD in adulthood. While the Module 4 is available for assessing verbally fluent adults, adaptation of the ADOS modules 1 and 2 are needed to advance understanding of the manifestation of ASD symptoms in adults with limited spoken language. Several longitudinal studies that have followed individuals from childhood into adulthood have used the ADOS to track ASD symptoms. Thus, such adaptations will facilitate studies of symptom trajectories that can advance our understanding of stability and change across the range of ability levels. This adaptation could also be useful in clinical settings to assess older individuals and facilitate appropriate access to care as they transition to adult service systems, as well as those seeking assessment of social-communication strengths and challenges to inform treatment plans.

This paper describes the Adapted ADOS Module 1 (A-ADOS-M1) and Module 2 (A-ADOS-M2), designed for use with minimally verbal adolescents and adults. The Adapted ADOS (A-ADOS) retains the original spirit of ADOS-2 tasks, but includes new materials and tasks selected to be developmentally appropriate and of interest to older individuals. A comparison is provided in Table 1. As with other modules, the focus remains on providing a naturalistic social interaction that is standardized through the administration of tasks using carefully selected materials and a hierarchy of “presses” that provide varying levels of structure. Particular attention was paid to the appropriateness of materials for adolescents and adults. Some ADOS-2 tasks and items geared toward infants or toddlers were replaced with new activities and objects specifically selected to maximize older individuals’ interest and motivation to communicate. Other tasks were modified by substituting new materials and/or modified administration (see Table 1). Many changes were inspired by a recognition of differences in expectations for older children and adults (e.g., no longer requiring a parent/caregiver present in the room; activities and presses that are mindful of personal space). Similar to other ADOS-2 modules, the A-ADOS is intended for use with individuals without significant sensory (e.g., blindness, deafness) or motor impairments (e.g., able to walk independently). Compared to ADOS-2 Modules 1 and 2, however, A-ADOS activities are conducted largely at the table, except for two activities (Anticipation of a Routine with Objects and Bubble Guns) that are conducted standing up. While use of the A-ADOS with individuals with significant motor impairments was not part of the initial validation, this may be an area warranting further consideration in future studies.

Table 1.

Comparison of ADOS-2 and A-ADOS activities

ADOS-2			A-ADOS
Module 1	Module 2		Module 1	Module 2	Changes
	Construction Task		Construction Task	Construction Task	Both modules; new puzzle; Module 2 task is interactive*
Response to Name	Response to Name		Response to Name	Response to Name	Parent presses removed
	Make-Believe Play		Interactive Soccer Game	Interactive Soccer Game	Removed toys; new Interactive soccer game
Anticipation of a Social Routine			Masks, Fives	Fives	New Masks task for M1; Fives task for both modules
Funct. & Symb. Imit.			Funct. & Symb. Imit.		Larger materials
Routine With Objects	Routine With Objects		Routine With Objects	Routine With Objects	New materials; M2 is interactive
Free Play	Break		Break	Break	Incorporate M3/M4 materials and new materials
Resp. to Joint Attn	Resp. to Joint Attn		Resp. to Joint Attn	Resp. to Joint Attn	Remove bunny, use break materials
Bubble Play	Bubble Play		Bubble Guns	Bubble Guns	M2 task is interactive
Snack	Snack		Snack	Snack	Examiner has snack; changes to administration
	Description of a Picture			Description of a Picture	New picture options
	Conversation			Conversation	No change
	Demonstration Task			Demonstration Task	No change
	Socio-Emotional Qs			Socio-Emotional Qs	Simplified wording
	Tell Story from Book			Tell Story from Book	New book options

Note.

^*Interactive indicates that opportunities for turn-taking are specified in the administration of this activity

Consistent with other ADOS-2 modules, the examiner should take detailed notes during administration and score the A-ADOS behavioral codes immediately afterwards. Many items have a similar focus to existing ADOS-2 items but have been calibrated to capture variation and maximize sensitivity and specificity in minimally verbal adults. A subset of items determined to best differentiate ASD from non-ASD conditions are then mapped on to an algorithm to yield a classification of ASD or non-ASD that may be used in conjunction with other instruments to make clinical or research diagnoses and describe ASD symptoms. It is essential that examiners be experienced and comfortable working with older, minimally verbal individuals. It is also particularly important to obtain an estimate of nonverbal IQ (NVIQ) prior to beginning the A-ADOS, as nonverbal cognitive abilities tend to vary widely in the minimally verbal population with ASD (Bal et al., 2016). Failure to recognize significant discrepancies (e.g., a nonverbal adult whose NVIQ is in the borderline to average range) may result in a markedly uncomfortable assessment for the participant and missed opportunities for observing social-communicative strengths and challenges. As such, the validity of the A-ADOS assumes the clinical competencies required to navigate the needs of working with older individuals of varying language and cognitive abilities in order to maintain administration fidelity and reliable scoring.

Methods

Participants

The sample included a total of 207 cases from 175 unique participants. Each case was defined as having an A-ADOS administered at a different time point and corresponding best estimate clinical diagnosis; 28 participants provided data for two cases and 2 participants provided data for 3. The majority of participants were research referrals: 71 were participants in the Early Diagnosis study (Lord et al., 2006), 96 were recruited in Michigan, New York and California for studies focused on older minimally verbal individuals with ASD and the remaining 8 were clinic-referrals to a specialty ASD clinic in Michigan. Approximately 73% of participants were white and 87% non-hispanic. All participants were 10 years of age or older. Additional descriptives are provided in Table 2.

Table 2.

Participant characteristics

	Module 1 – No Words				Module 1 – Some Words				Module 2
	ASD		Non-ASD		ASD		Non-ASD		ASD		Non-ASD
N	65		14		48		10		48		22
Age	18.36	(4.36)	20.86	(2.85)	16.90	(4.40)	18.96	(3.48)	18.87	(3.20)	20.33	(5.15)
	10.00–32.33		17.08–26.58		10.08–32.08		12.67–23.92		10.75–24.08		10.17–33.00
VIQ	10.40	(7.60)	14.57	(15.91)	24.32	(7.79)	20.60	(9.70)	39.22	(17.35)	31.68	(9.08)
	1.00–32.00		3.00–65.00		12.00–42.00		7.00–42.00		16.00–94.00		13.00–47.00
NVIQ	18.92	(9.77)	15.75	(11.24)	48.85	(22.18)	29.10	(15.36)	53.63	(23.36)	34.32	(12.94)
	5.00–50.00		2.00–36.00		14.00–111.00		14.00–58.00		20.00–106.00		12.00–64.00
VMA	20.10	(14.30)	28.54	(23.36)	44.83	(14.60)	36.67	(13.01)	71.30	(23.60)	65.64	(19.03)
	2.00–72.00		7.00–94.00		17.00–78.00		15.00–52.00		34.00–132.00		28.00–99.00
NVMA	36.33	(16.54)	32.75	(25.44)	77.14	(32.50)	50.56	(20.17)	95.98	(35.91)	64.90	(14.51)
	11.00–90.00		4.00–78.00		28.00–165.00		30.00–178.00		44.00–173.00		26.00–92.00

Note: bold indicates significant mean difference between groups

Of 207 cases, 161 had best estimate clinical diagnoses of ASD and 46 non-ASD. The non-ASD sample included 20 individuals with Down syndrome, 4 with Fragile X, 20 with idiopathic Intellectual Disability, 2 with other medical syndromes.

Measures

Adapted ADOS

All A-ADOSes were administered and scored by a clinical psychologist or trainee (e.g., graduate student or research assistant) who had met standard requirements for research reliability on the ADOS-2 and established reliability on the A-ADOS (80%). Consensus codes based on discussion between scorers were used for analyses. Within this sample, 15 different examiners collected data from the A-ADOS over 9 years (2009–2017).

Cognitive Assessment

Mental ages and IQs were derived from a standard developmental hierarchy of measures including the Wechsler Abbreviated Scale of Intelligence (Wechsler 1999), the Differential Ability Scales (Elliott 1990, 2007), the Mullen Scales of Early Learning (Mullen, 1995), or the Peabody Picture Vocabulary Test (Dunn and Dunn 2007) and Ravens’ Progressive Matrices (Raven, 2003). When using a test outside the standardization age range, separate verbal and nonverbal ratio IQs were computed by averaging the age equivalents for the relevant subtests to yield a mental age and then dividing by chronological age and multiplying by 100 (see Bishop et al., 2015). To minimize deflation of scores, 216 months was used in place of chronological age for all participants 18 years and older.

Procedure

The A-ADOS was conducted as part of a clinical or research evaluation including at least the A-ADOS and a cognitive assessment. Information regarding developmental history was gathered from the Autism Diagnostic Interview – Revised (ADI-R; Rutter, Le Couteur, & Lord, 2003), Social Communication Questionnaire (SCQ; Rutter, Bailey, & Lord, 2003), or intake. Best estimate clinical DSM-IV-TR or DSM-5 (American Psychiatric Association, 2000; American Psychiatric Association, 2013) diagnoses based on all available information were made by a supervising psychologist or an advanced graduate student. Thus, diagnoses were not made independently of the A-ADOS; however, algorithms were not derived until after data were collected. Verbal and/or nonverbal IQ estimates were available for 99% of cases. Procedures related to this project were approved by Institutional Review Boards at all sites.

Design and Analysis

Analyses to derive A-ADOS algorithms followed a similar procedure to what was used to develop the ADOS-2 Toddler Module (Luyster et al., 2009) and derive ADOS-2 (Gotham et al., 2007) and Module 4 revised algorithms (Hus & Lord, 2014). These will be described below.

Test Construction and Pilot Testing

Multiple drafts of the A-ADOS modules were generated and preliminary results were used to inform structural decisions about the measure.

Tasks.

Adaptations made to the Pre-Linguistic ADOS (PL-ADOS; DiLavore, Lord, & Rutter, 1995), a precursor to the ADOS designed for toddlers, were used to inform early changes to make A-ADOS tasks and materials more age-appropriate for use with adults (Berument et al., 2005). Further modifications were informed by pilot testing of tasks and materials. Examiners provided feedback regarding participant interest and engagement, as well as the clinical utility of information gained from behavioral observations made during tasks (e.g., whether the task provided an opportunity to observe a skill or impairment that the examiner felt was clinically significant or would inform treatment recommendations).

Behavioral codes.

Initially, behavioral items from ADOS-2 Modules 1 and 2 were included and data distributions based on preliminary data were used to inform item revisions (i.e., rewriting of codes to improve sensitivity or specificity) or omission of items. Throughout the project, items from Modules 3 and 4 were added and new items were written to capture additional aspects of behavior thought to be of clinical importance or to directly assess the diagnostic utility of new tasks added to the protocol. This process yielded a final set of 27 social-communication and five Restrictive and repetitive behavior (RRB) items for A-ADOS-M1 and 31 social-communication and seven RRB items for A-ADOS-M2. As in the ADOS-2, both protocols also include three “other behavior” codes (Overactivity; Aggression and Disruptive Behavior; Anxiety). Also similar to the ADOS-2, there are two items assessing use of objects; one unique item assessing spontaneous initiation with materials and the other capturing engagement in activities, comparable to the ADOS-2 Toddler and Module 1 code Level of Engagement. Unique to the A-ADOS, codes were added to capture Alternative and Augmentative Communication use (as such modalities are increasingly common and it is of clinically significance if a participant has an alternative method of communication) and presence and level of caregiver involvement (because caregivers were not required to be present). Video recorded administrations of early versions were reviewed and items that had been modified were recoded to reflect the current item structure.

Other design decisions.

As noted above, data for participants 10 years of age or older were included. It is important to note that 10-to-12-year-olds were included in the original ADOS and validation of the ADOS-2 algorithms and that the A-ADOS was designed with older adolescents and adults (13+) in mind. In consideration of research suggesting that a significant minority of minimally verbal children have much higher nonverbal problem-solving skills than expressive language skills (e.g., Bal et al., 2016), it was decided to assess the utility of A-ADOS with older school-age children. Pilot administrations indicated that the appropriateness and clinical utility of the A-ADOS with school-aged children aged 10–12 varied, most likely attributable to developmental level. In most cases, clinicians felt the A-ADOS tasks were “too difficult” for children below 10. Thus, it was decided to include children from age 10 and up in recognition that some older minimally verbal school-age children with higher mental ages would benefit from the adapted assessment.

A-ADOS algorithm development

Social-communication items were selected to create diagnostic algorithms. Specific attention was given to maintaining as much continuity as possible across A-ADOS modules, as well as with other ADOS-2 modules, to facilitate longitudinal analyses and intraindividual comparison over time. Thus, item distributions were examined for three separate groups: 1) A-ADOS-M1-Few-to-No-Words (FNW; fewer than 5 words used during the A-ADOS), 2) A-ADOS-M1-Some Words (SW; at least 5+ words) and 3) A-ADOS-M2. For other modules (i.e., Gotham et al., 2007; Hus & Lord, 2014), social-communication items have been considered “preferred” if fewer than 20% of the ASD group had a ‘0’ and fewer than 20% of the non-ASD group had a ‘2’. Because no A-ADOS-M1-FNW items and few A-ADOS-M2 items met these criteria, cut-offs were relaxed. For A-ADOS-M1-FNW any item for which fewer than 30% of the ASD group scored a ‘0’ was considered, recognizing challenges of specificity considering the low nonverbal mental age (NVMA) in this non-ASD group. For A-ADOS-M2, items for which fewer than 30% of the ASD group scored a ‘0’ and fewer than 30% of the non-ASD group scored a ‘2’ were considered. Selection criteria were not applied to RRB items. Considering the small non-ASD sample sizes, it was of interest to select items for preliminary use that could be refined when larger samples become available to validate or revise the algorithms. Exceptions to the criteria above were also allowed for items that were theoretically important and overlapped with items on algorithms from other modules to promote cross-module comparison. From this pool of preferred items, items were selected to promote comparability within the set of A-ADOS modules, as well as to ADOS-2 algorithms of the comparable module.

Validity assessment of the A-ADOS algorithms

Factor Analysis.

Ordinal probit item response models were conducted in MPlus version 8 (Muthen and Muthen 1998) to account for the ordinal nature of ADOS items. Factor loadings from promax oblique rotations were used to inform creation of subdomains. Root Mean Square Error Approximation (RMSEA) <.08 was used to indicate satisfactory fit (Browne & Cudek, 1993). Confirmatory factor analyses were conducted to verify the goodness-of-fit; Comparative Fit Index (CFI) >.9 was used to indicate good fit (Skrondal & Rabe-Hesketh, 2004).

Diagnostic Validity.

Items were summed to generate a total score. In keeping with ADOS-2 conventions, scores of ‘2’ and ‘3’ were collapsed. Logistic regressions were used to examine the predictive value of subdomain scores. Receiver Operating Curves (Siegel et al., 1989) were computed to inform cut-offs that afforded the best balance of sensitivity and specificity. A-ADOS items were also used to compute ADOS-2 Module 1 and 2 algorithms to allow comparison of classification accuracy with the A-ADOS algorithms. One item, Integration of Gaze and Other Behaviors During Social Overtures, was excluded from later versions of the A-ADOS; Amount of Reciprocal Social Communication was substituted for the purpose of computing the ADOS-2 Module 1 algorithms.

Correlations with Participant Characteristics.

Correlations between item and domain scores and participant characteristics were examined.

Reliability assessment of the A-ADOS algorithms

Internal Consistency.

Correlations between each item and the domain score minus that item were assessed. Each domain’s internal consistency was measured by Cronbach’s alpha for each of the three algorithms.

Inter-rater reliability.

Inter-rater reliability was calculated for 25 assessments rated by 12 unique pairs of coders (either live or via video). Item-level reliability was computed across modules for 34 items that were identical across modules. Seven items unique to A-ADOS-M2 were compared for 6 assessments. Package irr in RStudio v.1.1.463 was used to calculate Fleiss’ weighted kappas for non-unique pairs of raters. Kappas above .75 were considered excellent and between .4 and .74 were considered good (Fleiss, 1986). Percent exact agreement was also computed for each item. Percent agreements were assessed as follows: 70–79% = fair, 80–89% good, 90+% excellent (Cicchetti et al., 1997). Finally, intraclass correlations (ICC) were computed to assess reliability of algorithm SA, RRB and Total scores and exact agreement of the resulting A-ADOS classifications was computed.

Test-Retest Reliability.

A sample of 12 participants were administered the A-ADOS twice within 30 months (1–30, M=14.58 months) to evaluate test-retest reliability. Intraclass correlations (ICC) were computed.

Results

Tasks

Based on scores available for 117 ASD and 20 non-ASD cases, Response to joint attention had limited sensitivity (i.e., only 14% [n=7] of A-ADOS-M1-FNW ASD cases scored a ‘2’; all other ASD and non-ASD cases scored a ‘0’ or ‘1’). Nonetheless, the task and item were included due to the diagnostic significance of response to joint attention in younger children (including its inclusion on the ADOS-2 M1-FNW algorithm) and clinical significance if response to joint attention was impaired in an adult.

Make believe play was not found to yield diagnostically or clinically useful information, as most adults did not have situations in which they encountered action figures or imaginative play items. Considering limited clinical utility, this task was ultimately removed to reduce administration time. However, a table-top soccer routine using figurines and a plastic ball was maintained to provide an opportunity to observe a more concrete, reciprocal interaction that could be relevant to treatment planning (e.g., engagement in leisure activities with a partner). In addition, a Break is provided in both modules to observe how individuals independently occupy themselves when given leisure time. A range of different objects available during the break and at other times to provide opportunities to observe repetitive or stereotyped behaviors.

A-ADOS algorithm development

From the preferred item pool, 12 social-communication items were selected for each algorithm. For A-ADOS-M1-FNW, all items fell within the modified preferred criteria; however, specificity was low with 50% or more of the non-ASD participants scoring a ‘2’ on 6 of the items. For A-ADOS-M1-SW, two items fell just outside of the preferred criteria (Odd Vocalizations 21% ASD scored ‘0’, 20% non-ASD scored ‘2’ and Gestures 8%, 30%, respectively) and were included in the Exploratory factor analysis. For A-ADOS-M2, 4 items met preferred criteria and 6 met modified preferred criteria. Two items outside these ranges, Facial Expressions (2%, 32%) and Eye Contact (17%, 64%) were included to promote comparability with ADOS-2 modules.

Validity assessment of the A-ADOS algorithms

Factor Analysis

As shown in Table 3, a 2-factor solution fit well for both A-ADOS-M1 algorithms, consistent with other modules (Lord et al., 2012; Hus & Lord, 2014). Similar to the ADOS-2, several communication items loaded on to the RRB domain: Frequency of Language or Undirected Vocalization and Odd/Unusual Vocalizations for both A-ADOS-M1 algorithms and Odd/Inappropriate Facial Expressions for A-ADOS-M1-FNW. In addition, the Stereotyped, Repetitive Behaviors item loaded on the Social Affect (SA) domain for both algorithms. Because this item has been included in the RRB domain for all other modules, it was maintained in the same domain for the A-ADOS-M1 algorithms. Confirmatory factor analysis for the two factor solutions shown in Table 3 indicated good fit (CFI: FNW=.90, SW=.99).

Table 3.

A-ADOS algorithm mapping

Domains	Mod 1 FNW (N=79)	Factor Loadings	Mod 1 SW (N=58)	Factor Loadings		Mod 2 (N=70)	Factor Loadings
	Pointing	0.79	Initiation of JA	0.76	Basic Soc-Commb	Initiation of JA	0.81
	Gestures	0.55	Gestures	0.40		Gestures	0.23
	Unusual Eye Contact	0.58	Unusual Eye Contact	0.82		Unusual Eye Contact	0.49
	Facial Expressions	0.69	Facial Expressions	0.53		Facial Expressions	0.49
	Shared Enjoyment	0.73	Shared Enjoyment	0.74		Shared Enjoyment	0.75
Social Affect	Amount Social Overtures	0.87	Amount Social Overtures	0.77	Inter-action Qualityb	Amount Social Overtures	0.78
	Showing	0.77	Showing	0.49		Quality of Social Overtures	0.86
	Social Engagement	0.87	Social Engagement	0.92		Quality of Social Response	0.57
	Frequency of Vocalization	0.58	Amount Social Communication	0.83		Amount Social Communication	0.61
			Quality of Rapport	0.96		Quality of Rapport	0.92
Eigen value	5.42		1.77			4.94, 2.28
Restricted Repetitive Behaviors	Repetitive Interestsa	0.12	Repetitive Interestsa	−0.08		Repetitive Interests	0.49
	Hand Mannerisms	0.45	Hand Mannerisms	0.51		Hand Mannerisms	0.81
	Odd Vocalizations	0.93	Odd Vocalizations	0.91		Unusual Intonation	0.40
	Frequency Undirected Voc.	0.70	Frequency Undirected Voc.	0.85		Stereotyped Language	0.64
	Unusual Sensory Interest	0.44	Unusual Sensory Interest	0.30		Unusual Sensory Interest	0.53
	Unusual Facial Expressions	0.85
Eigen value	2.89		7.45			1.81
RMSEA	0.03		0.04			0.07
Rho	0.08		0.50			0.33, 0.31

Note.

^aloads on Social Affect (0.60 M1NW, 0.40 M1SW) but included in RRB domain for comparability to other modules.

^bModule 2 Exploratory Factor Analysis indicated a 3-factor solution dividing Social and Communication items into two factors. FNW = Few to No Words; SW = Some Words; RMSEA = root mean square error from exploratory factor analysis (values .08 or less indicate a good fit); Rho = correlation between Social Affect & Restricted Repetitive Behaviors factors for Module 1 and between Basic Social-Communication and Restricted Repetitive Behaviors and Interaction Quality and Restricted Repetitive Behaviors for Module 2, respectively. Rho for Basic and Interaction factors = 0.22. JA = Joint Attention; Frequency Undirected Voc. = Frequency of Undirected Vocalizations

Results of the Exploratory Factor Analysis suggested that the 2-factor model did not meet fit criteria for A-ADOS-M2 algorithm (RMSEA=.09). As shown in Table 3, a 3-factor solution emerged. Results approximated results reported in Module 3 (Bishop et al., 2016), with factors reflecting Basic Social-Communication, Interaction Quality and Repetitive Behaviors. Facial Expressions loaded solidly on both the Basic (.49) and RRB (.49) factors, whereas Speech Abnormalities loaded on both Interaction Quality (.43) and RRB (.40). While Gestures had the lowest loadings (.23 on Basic, .17 on RRB), it was maintained because of its inclusion on all other ADOS-2 and A-ADOS algorithms. For comparability to other ADOS-2 modules, Facial Expression and Gestures were maintained on the Basic factor and Speech Abnormalities on the RRB. Confirmatory factor analysis indicated good fit (CFI=.93). Although the three-factor model best fit the data, the Basic and Interaction Quality domains were collapsed to create a Social Affect domain to allow comparisons across modules for the validity and reliability analyses below.

Logistic Regression Check on Weighting Domains

Logistic regression indicated that both domains made independent contributions to best estimate clinical diagnosis (ASD vs. Non-ASD). Results suggested that SA [B=.32, SE=.06, z=12.86, Exp(B)=1.37, CI=1.23, 1.53] and RRB [B=.30, SE=.08, z=32.04, Exp(B)=1.35, CI=1.15, 1.59] made similar contributions to predicting diagnosis.

Sensitivity and Specificity

To maintain consistency with DSM-5 and Module 4, a single ASD cut-off was identified for the overall total (i.e., SA+RRB). As shown in Table 4, the A-ADOS algorithms showed substantial gains in specificity compared to the ADOS-2 algorithm. This came at some cost to sensitivity, though sensitivity remained above 80% for all A-ADOS modules.

Table 4.

Comparison of sensitivity and specificity for ADOS-2 vs. A-ADOS

	N (ASD, NonASD)	ADOS-2		A-ADOS
		Sens	Spe	Sens	Spe
M1-FNW	65, 14	100	28.6	89.2	64.3
M1-FNW (NVMA>18)	61, 8	100	25	90.2	75.0
M1-SW	48, 10	97.9	30	91.7	90.0
M2	48, 22	100	31.8	83.3	86.4

Note. Sens = Sensitivity; Spec = Specificity

Item correlations with participant characteristics

Correlations with age and cognitive abilities generally fell below r=.4 across modules. A-ADOS-M1-SW, Spontaneous Initiation of Joint Attention was moderately correlated with age (r=−.47); no items correlated with VIQ or NVIQ. For A-ADOS-M2, Frequency of Undirected Language or Undirect Vocalizations was moderately correlated with nonverbal IQ (r=.41).

Domain correlation with participant characteristics

A-ADOS-M1-FNW SA and RRB totals were weakly correlated (r=.14), whereas domain totals were moderately correlated for A-ADOS-M1-SW (r=.47) and A-ADOS-M2 (r=.37). For A-ADOS-M2, RRB correlated more strongly with Quality (r=.37) than Basic (r=.27). Correlations with participant characteristics exceeded .3 for the M1-FNW algorithm total (Age r=−.31; VIQ r=−.32). This was driven by a higher negative RRB domain correlation with age (r=−.42 vs. SA=−.07) and higher negative SA domain correlation with VIQ (−.30). For M2, the overall total was correlated with NVIQ (r=.40), NVIQ was positively correlated with the SA (.32; Basic-NVIQ r=.22, Quality-NVIQ r=.35) and RRB subdomains (.37). RRB was also negatively correlated with age (−.31). Correlations with the M1-SW algorithm all fell below .3.

Reliability assessment of the A-ADOS algorithms

Internal Consistency

For A-ADOS-M1-FNW, item-domain correlations ranged from .31 to .70 for SA and .15 to .60 for RRB. Internal consistency was comparable to ADOS-2 modules (Cronbach’s alpha: SA=.82; RRB=.67). For A-ADOS-M1-SW, item-domain correlations ranged from .53 to .82 for SA and .12 to .48 for RRB. Internal consistency was comparable to ADOS-2 (Cronbach’s alpha: SA=.90; RRB=.60). For A-ADOS-M2, item-domain correlations ranged from .33 to .61 for the Social Affect domain except for Gestures (.19) and eye contact (.28) and from .25 to .55 for RRB. Internal consistency was acceptable for both domains (Cronbach’s alpha: SA=.77, RRB=.77), as well as the Basic (.81) and Quality (.72) subdomains.

Inter-rater reliability

Out of 41 items, 19 weighted Kappas were .75 or higher, with the remainder exceeding .47. Mean exact agreement for 34 items across both modules was 83.51%. Two items, Odd Facial Expressions and Unusually Repetitive Interests or Stereotyped Behaviors, fell just below the fair range (percent exact agreement = 68%; weighted Kappa=.47 and .59, respectively).

Across the three algorithms, ICC were .91 for SA (95%CI[.80, .96]), .85 for RRB (95%CI [.68, .93]) and .93 for Total (95%CI[.85, .97]). Interrater agreement for diagnostic classification was 88% across algorithms, ranging from 81.9% (A-ADOS-M1SW) to 100% (A-ADOS-M2).

Test-retest Reliability

For the 12 participants administered the A-ADOS twice within 30 months, test-retest ICCs were comparable to that reported for other modules: .87 for SA [.61, .96], .60 for RRB [.04, .87] and .88 for Total [.64, .96]. Mean absolute differences between the first and second assessments were .42 (SD=2.43) for SA, .25 (SD=2.73) for RRB and .67 (SD=3.08) for the Total score. Ten of the 12 participants were classified consistently across the two evaluations. For one participant, change in classification was attributable to more RRBs observed during the second assessment (the participant’s SA scores were the same on both occasions). The increase in RRBs resulted in an ASD classification (which was consistent with the participant’s ASD diagnosis). For the second participant, change in classification was due to a 2-point decrease in both SA and RRB totals, resulting in a non-ASD classification for the second visit (which was not consistent with his clinical diagnosis).

Discussion

The A-ADOS expands the repertoire of tools available to assess ASD symptoms in minimally verbal adolescents and adults. It includes two modules, with three algorithms that demonstrate acceptable internal consistency and diagnostic validity. Compared to the ADOS-2, algorithms show considerably improved specificity at the cost of slightly reduced sensitivity. Given the small sample sizes, these algorithms should be considered preliminary until further validated in larger, independent samples. As with other ADOS-2 modules, the A-ADOS is intended to be one source of information within a comprehensive assessment.

The A-ADOS was originally conceptualized for adolescents and adults (i.e., 13+). Research suggests, however, that a significant minority of minimally verbal school-aged children with ASD may exhibit considerably higher nonverbal problem-solving skills (Bal et al., 2016). Thus, children as young as 10–12 years were included, recognizing that older school-age children exhibiting discrepant profiles (i.e., nonverbal>verbal) may benefit from an adapted assessment. When used in this age range, particular attention should be paid to nonverbal mental age; for children with lower MA, the standard ADOS-2 modules may be more appropriate. Future studies will be needed to examine the diagnostic validity in younger non-ASD samples, as only 23 ASD and 2 non-ASD participants under 13 were included.

Many older individuals with the level of language impairment that would warrant an ADOS-2 or A-ADOS Module 1 or 2 will already have received a diagnosis. Nonetheless, an algorithm was derived to maximize diagnostic validity with the intention of providing an index for comparisons across time and developmental levels. As with the other ADOS-2 modules, calibrated severity scores will be needed to facilitate such comparisons. This will require larger samples than available in the present study. In addition to fostering longitudinal research and clinical tracking over time, the A-ADOS was carefully designed with the intention of providing valuable information beyond diagnostic classification in both clinical and research contexts. For example, the Break activity in both modules provides opportunities to observe how the individual occupies him or herself during independent leisure time and opportunities to observe repetitive or sensory engagement with a variety of materials. The snack activity, also included in both modules, affords the opportunity to observe how language or communication is used during a common daily activity, as well as to observe manners and sharing. New items coded for amount of requesting and initiation and engagement with materials and activities may provide important clinical information regarding prompt dependence to engage with materials, and how an individual might engage in a novel environment and introduced to a range of different tasks. Results and observations from the Adapted Modules may capture strengths and challenges useful in better understanding the individual and inform development of treatment plans.

The small size of this sample, particularly in the non-ASD Module 1 algorithm groups is a limitation to this study. In addition, though the age range spans from 10–33 years of age, the sample is relatively young with 22% of cases falling between 13–17, 53% 18–21 and only 12% 22+ years. It is also important to acknowledge that the ASD M1-SW and M2 groups had significantly higher nonverbal IQs compared to their non-ASD counterparts. This reflects larger VIQ-NVIQ gaps in the ASD group, consistent with previous literature on minimally verbal children (Bal et al., 2016). These sample differences resulted in higher cut-offs to balance sensitivity and specificity. That individuals with ASD in this sample exhibit more social-communication impairments despite having higher nonverbal abilities than their non-ASD counterparts demonstrates that the A-ADOS is capturing variability in behavior not solely attributable to intellectual disability. Nonetheless, the A-ADOS will need to be further validated in larger, older samples including a range of non-ASD participants.

It is interesting to note that the traditional two-factor solution (Social Affect and Repetitive Behaviors) did not fit the present data for adults using flexible phrase speech (i.e., A-ADOS-M2). Rather, a three-factor solution emerged that included an RRB factor and two social factors: a Basic social-communication and an Interaction Quality. The item loadings were very similar to that reported by Bishop and colleagues (2016) in a mixed diagnosis (ASD and non-ASD) sample of school-aged children with fluent speech (i.e., Module 3). While the small sample size of the Module 2 group limits interpretation of these data, it is also interesting to highlight that Module 2 items and domains show positive correlations with nonverbal IQ. This is in contrast to negative correlations between items and IQ for both A-ADOS-M1 groups, and negative correlations between domains and IQ for the FNW group. Taken together, these findings warrant further investigation to understand the presentation of ASD symptoms in adults with phrase speech. Considering that instrument cut-offs are based on the overall total score (SA + RRB domains) for other modules, collapsing of the domains for the A-ADOS-M2 for the algorithm seems justifiable until the three-factor presentation can be further explored in larger samples.

The present study provides initial support for the validity of the A-ADOS Modules 1 and 2, modified to be more appropriate for use with minimally verbal adolescents and adults with nonverbal mental ages of at least 18 months. It may be appropriate for use with older school age children (10–12-year-olds), though further validation including non-ASD samples in this age range, as well as older adults, is needed. Future studies standardizing the scores to provide a calibrated severity metric will facilitate score comparisons over time in clinical settings and longitudinal research. The A-ADOS provides a standardized context for assessment and expands the repertoire of instruments available to characterize social-communication strengths and impairments in older minimally verbal individuals. It is hoped that the availability of this instrument will foster additional research to further understanding of this understudied population.