Participants
The participants of this study are part of the Simons Simplex Collection (SSC) (12). The SSC is a database that contains behavioural, cognitive, and genetic data from approximately 2,800 individuals meeting the criteria for an autism diagnosis. Probands’ phenotypes were evaluated with a battery of instruments for which descriptions are available on the Simons Foundation Autism Research Initiative website (https://sfari.org). The SSC includes simplex cases only and enrolment to the database is based on referrals from clinical genetic centres, testing laboratories, web-based networks or active online registration. The inclusion criteria for entry into the SSC database required probands to: 1) be between 4 and 18 years of age; 2) meet the criteria for a diagnosis of autism, Asperger syndrome, or autism spectrum; and 3) have a nonverbal mental age of at least 18 months. For the analyses of diagnostic certainty, only individuals who received a diagnosis of autism according to DSM-IV (autistic disorder) criteria were included. For the analyses of correlations between HC and other variables, all individuals with any autism spectrum diagnosis were included. All analyses were conducted separately for those assessed with each ADOS module. Only a relatively small number of individuals were assessed with module 4 (74 in total, of whom 33 were diagnosed with autistic disorder) and this module was therefore not included. The demographics of the participants separated based on language level (ADOS module) are presented in Table 1.
Table 1: Summary statistics of the included participants, separated by the ADOS module used for assessment. The rows “Verbal IQ” and “Nonverbal IQ” show the mean IQ values including ratio as well as deviation scores. The rows “Verbal IQ, deviation only” and “Nonverbal IQ, deviation only” show mean IQ values based only on deviation scores, since ratio scores were not included in the analyses.
|
ASD
|
Autistic disorder
|
|
ADOS module 1
|
ADOS module 2
|
ADOS module 3
|
ADOS module 1
|
ADOS module 2
|
ADOS module 3
|
N
|
429
|
514
|
1337
|
394
|
412
|
705
|
Females (%)
|
66 (15.4)
|
84 (16.3)
|
149 (11.1)
|
56 (14.2)
|
68 (16.5)
|
77 (10.9)
|
Mean age (std)
|
8.0 (3.5)
|
7.2 (3.2)
|
9.7 (3.2)
|
8.0 (3.5)
|
7.5 (3.2)
|
9.9 (3.2)
|
Race
|
65% white, 8% Asian,
7% African American
|
73% white, 6% Asian,
6% African American
|
84% white, 3% Asian,
2% African American
|
65% white, 8% Asian,
7% African American
|
72% white, 5% Asian,
7% African American
|
81% white, 3% Asian,
2% African American
|
Ethnicity
|
15% hispanic
|
15% hispanic
|
9% hispanic
|
16% hispanic
|
15% hispanic
|
11% hispanic
|
Verbal IQ (std)
|
34.5 (19.7)
|
69.4 (22.5)
|
93.7 (20.8)
|
33.7 (19.0)
|
66.7 (22.3)
|
88.3 (20.3)
|
Nonverbal IQ (std)
|
51.2 (21.2)
|
79.8 (21.4)
|
96.0 (18.8)
|
51.1 (21.1)
|
78.0 (21.6)
|
92.6 (18.7)
|
Verbal IQ, deviation only (std)
|
58.8 (15.4) N=118
|
78.7 (16.3) N=390
|
94.6 (19.8) N=1315
|
57.9 (14.9) N=102
|
76.6 (16.2) N=301
|
89.4 (19.3) N=690
|
Nonverbal IQ, deviation only (std)
|
70.9 (17.2) N=169
|
85.3 (16.8) N=441
|
96.4 (18.3) N=1324
|
70.9 (17.2) N=152
|
83.9 (17.1) N=347
|
93.1 (18.0) N=695
|
Highest certainty (%)
|
|
|
|
329 (83.5)
|
302 (73.3)
|
401 (56.9)
|
Father bachelor degree (%)
|
239 (55.7)
|
316 (61.5)
|
801 (59.9)
|
221 (56.1)
|
255 (61.9)
|
407 (57.7)
|
Father some college (%)
|
350 (81.6)
|
442 (86.0)
|
1133 (84.7)
|
322 (81.7)
|
358 (86.9)
|
603 (85.5)
|
Mother bachelor degree (%)
|
237 (55.2)
|
312 (60.7)
|
825 (61.7)
|
216 (54.8)
|
249 (60.4)
|
425 (60.3)
|
Mother some college (%)
|
372 (86.7)
|
466 (90.7)
|
1214 (90.8)
|
338 (85.8)
|
375 (91.0)
|
635 (90.1)
|
Measures
Verbal and non-verbal IQ
The SSC database contains IQ data based on several different instruments, including
the Differential Ability Scales 2nd edition (DAS-II), Wechsler Intelligence Scale for Children 4th edition (WISC-IV) and the Mullen Scales of Early Learning (MSEL). Most of the IQ scores were calculated as deviation IQ, i.e. using an age-adjusted norm data set with a mean of 100 and a standard deviation of 15. A smaller set of the scores were calculated as ratio IQ, i.e. using norm data to estimate an age equivalent and dividing by the chronological age. Since these two methods produce scores that may not be fully comparable (13), we only included scores calculated with the deviation method in our analysis.
Normalized head circumference
HC was measured by using a non-stretchable tape measure, measuring the widest part of the head. Head circumference is strongly associated with factors such as sex, age, height and ancestry. Therefore, a normalized head circumference variable was calculated to have a measure of head size that is not affected by these. This was done as described by Chaste et al. (14) by using the data of all autistic individuals in the SSC to fit a linear model with HC as the dependent variable and height, weight, age, sex and genetic ancestry factors as independent variables. The normalized HC variable was defined as the residuals from the linear model, i.e. the difference between each individual’s measured HC and the expected HC based on the predictor variables.
Certainty variable
The SSC contains a variable describing the certainty of the diagnosing clinician that the child has autism. If the child is deemed to lie somewhere on the autism spectrum, a base level of 5 points is given on the certainty variable. The clinician is then asked to rate how certain he/she is that the child meets the criteria for an autism spectrum diagnosis on a scale from 1 to 5 and this is added to the certainty score. In addition, if the child is also deemed to meet the strict DSM-IV-TR criteria for autism, the clinician must again rate how certain they are of this on a scale from 1 to 5, and this is also added to the certainty score which thus has a maximum value of 15. For children who receive an autism spectrum diagnosis but do not meet the criteria for a DSM-IV-TR autism diagnosis, the certainty variable has a maximum value of 10. The certainty variable is thus not directly comparable between these two groups. In this study, all analyses of diagnostic certainty were limited to those who met the strict DSM-IV-TR autism criteria. Among them, the certainty score was highly skewed, with a large proportion having a maximal certainty score of 15, and a decreasing number of individuals with progressively lower scores. For most of the statistical analyses (see below), the certainty scores were thus converted into a binary categorical variable: a certainty rating was coded as 1 if the score was 15, whereas certainty scores lower than 15 were coded as 0. This resulted in a less skewed variable in which the numbers of individuals with and without the highest possible certainty score were closer to each other.
Autism symptomatology
Total scores and individual item scores from the ADOS were used as measures of symptom severity as well as symptom presentation. The ADOS module used to assess a participant was used as a proxy of language level: module 1 indicates no phrase speech, module 2 indicates phrase but not fluent speech, while module 3 indicates fluent speech.
Vineland and CBCL measures
The Vineland Adaptive Behaviour Scale was used as a measure of adaptive abilities. Externalizing and internalizing composite scales from the Child Behavior Checklist (CBCL) were also included in our analysis.
Statistical analyses
Association between certainty and total autism symptom score
Spearman correlation coefficients were calculated between the total ADOS score and the raw certainty score (from 1 to 15). The correlations were calculated separately for each ADOS module.
Association between certainty and individual ADOS items
Each ADOS item was converted to a binary variable representing the presence/absence of a clear presentation of a given sign. Thus, ADOS item scores of 0 and 1 were coded as 0 (absence), whereas item scores of 2 and 3 were coded as 1 (presence). For each ADOS item, the association between the binary coding and the binary certainty variable was investigated by calculating odds-ratios. Odds-ratios greater than 1 indicate that the presence of a given sign is associated with a higher likelihood of being diagnosed with the highest certainty, whereas odds-ratios less than 1 indicate a negative association where the presence of a sign is associated with a lower likelihood of being diagnosed with the highest certainty. The scipy Python package was used to calculate p-values and 95% confidence intervals for the odds-ratios. The analyses were performed separately for each ADOS module. Within each ADOS module, p-values for each ADOS item were corrected for multiple testing using the Benjamini-Hochberg method (15). For some items, no individuals in one of the two certainty groups had a score of 2 or 3, leading to division by 0 when calculating the odds-ratios. In these cases, the odds ratio was instead estimated using the Haldane-Anscombe correction (16), by adding 0.5 to each of the counts used to calculate the odds-ratio.
Association between certainty and language level
The ADOS module with which an individual was evaluated was used as a proxy for the individual’s language level at the time of assessment. Binomial regression was used to investigate the association between language level and certainty, with the binary certainty variable as the dependent variable and the ADOS module as the independent variable. Since the degree to which a given language level is normal/abnormal depends on age, age was also included as an independent variable, as well as an interaction term between age and the ADOS module. Binomial regression was performed using the ‘glm’ function in R and the statistical significance of the effects was evaluated using the ‘Anova’ function from the car package.
Association between diagnostic certainty and IQ, head circumference, and internalizing, externalizing, and adaptive behaviours
The association between certainty and each of the following variables: verbal IQ, nonverbal IQ, the verbal/nonverbal IQ ratio, internalizing behaviour, externalizing behaviour, adaptive behaviour, and normalized HC, was assessed by investigating the group-level difference for each variable between those who were diagnosed with the highest certainty compared to those who were not. This was performed separately for each ADOS module as language level was expected to influence certainty. Statistical significance of the group differences was evaluated using t-tests.
Association between normalized HC and other variables
Pearson correlation coefficients were calculated between normalized HC and ADOS items as well as the binary diagnostic certainty, verbal IQ, nonverbal IQ and verbal/nonverbal IQ ratio. This was performed separately for each ADOS module. Correlations with certainty included only individuals with a DSM-IV diagnosis of autistic disorder, whereas correlations with the remaining variables included all individuals in the SSC who received any autism spectrum diagnosis.