Intelligence Profile and Diagnosis Model in Children with Autism Spectrum Disorder (ASD)

Friday, May 13, 2016: 5:30 PM-7:00 PM
Hall A (Baltimore Convention Center)
B. Chen1, H. Deng1, X. Zou1, J. He1 and B. Wu2, (1)Children Developmental & Behavioral Center, The Third Affiliated Hospital of Sun Yat-Sen University, Guangzhou, China, (2)School of Mathematics & Computational Science, Sun Yat-Sen University, Guangzhou, China, Guangzhou, China

Differences in intelligence profile for ASD and other developmental disabilities such as ADHD are commonly observed in clinical practice. Some studies indicate that the intellectual-indicators may contribute to diagnostic discrimination. This study establishes a novel computerlearning-model for ASD diagnosis, which is demonstrated to have satisfied performances in discriminating ASD from ADHD individuals.


1. To explore diagnostic prediction accuracy of ASD and ADHD base on individual intelligence profile.

2. To provide informative models for decision making for ASD diagnosis based on performance pattern differences on intelligence measurement.


The study population consisted of 605 children from Chinese Han population, which includes 275 children diagnosed as ASD by clinical-judgments according to DSM5 criteria, 192 with ADHD, 104 with Intellectual-Disabilities, and 36 Typically-Developed individuals. Chinese-Wechsler Intelligence Scale for Children (C-WISC) was administered between 01/2013 to 01/2015 in all subjects. We built a Random-Forest model to make diagnostic classification by randomly taking 2/3 samples as the training dataset and 1/3 as the test, then applied bootstrap strategies to re-sample 100 times. The Random-Forest method is used to build the diagnostic model, where each factor's contribution to diagnostic classification are evaluated and the important factors are selected to build the prediction model. We presented the decision tree for ASD diagnosis using intellectual-indicators.


In those with Full-Scale IQ ≥70 , Arithmetic, Comprehension and Similarities were the most important factors for ASD diagnosis, with the general prediction accuracy for ASD of 0.537 and ADHD 0.74. With the threshold of Arithmetic Standard score (S-score) ≥8.5, plus | VIQ – PIQ | >10, the prediction accuracy for ASD increases to 82% if Picture Arrangement S-score is <8.5 plus Information S-score ≥14. With the Arithmetic S-score < 8.5 and Block Design S-score ≥12, the prediction of ASD diagnosis shows 77% accuracy without requirement of VIQ and PIQ differences.  In those with Full-Scale IQ <70, Comprehension became a primary component in the ASD decision making process, followed by Block Design and Object Assembly. General prediction accuracy is 0.822 to separate ASD from others including ADHD, ID and TD. With | VIQ – PIQ | >5, predicting accuracy for ASD diagnosis could reach to 100% in a total of an 18 case subset, with Comprehension S-score <1.5 and Block Design S-score ≥7.5. In the subgroup without Verbal IQ and Performance IQ differences requirement, predicting accuracy for ASD diagnosis is 93% with Comprehension S-score <1.5 and Block Design S-score ≥6.5.


Our findings suggest that a low Comprehension level shows promising predictive accuracy in low functioning children with ASD, with an individual’s visual learning abilities remaining at a similar level. This may contribute to the effectiveness for rescreening ASD missed-diagnosis in other developmental-disabilities population, especially children who receive a prior diagnosis of intellectual disability. In higher functioning children with ASD, the predictors include normal visual learning skills and cognition, but abilities related to logic or reasoning are poor. Risk factors in other intellectual assessments which may contribute to ASD diagnosis should be explored in further research.