Characterizing the Heterogeneity of Academic Achievement in ASD

Friday, May 12, 2017: 5:00 PM-6:30 PM
Golden Gate Ballroom (Marriott Marquis Hotel)
H. N. Wakeman1, L. Chen2, T. Iuculano3, M. Rosenberg-Lee4 and V. Menon5, (1)University of Colorado - Boulder, Boulder, CO, (2)Psychiatry, Stanford School of Medicine, Palo Alto, CA, (3)Stanford University School of Medicine, Palo Alto, CA, (4)Psychiatry, Stanford University School of Medicine, Palo Alto, CA, (5)Stanford University School of Medicine, Stanford, CA
Background:  While autism spectrum disorders (ASD) are known for heterogeneous symptom presentation (Hu, 2009) and cognitive ability (Lenroot, 2013), the heterogeneity of ASD in academically-relevant domains, like math and reading, is still poorly understood. Although previous studies (Jones, 2009; Wei, 2015) have provided initial evidence on individual differences in academic achievement in ASD, no studies to date have used unbiased, data-driven, and quantitatively-validated approaches, nor included a typically-developing (TD) comparison group.

Objectives:  The current study aims to: (a) characterize heterogeneous patterns of academic performance, namely math and reading skills, in children with ASD, using an unbiased and data-driven approach; (b) determine whether heterogeneous patterns are unique to children with ASD compared to TD peers; and (c) identify cognitive factors contributing to these heterogeneous patterns in ASD.

Methods:  118 children with ASD (ages: 7 – 13 years old; M=9.68; sd=1.51) and 96 age- and IQ-matched TDs (ages: 7-13 years old; M=9.39; sd=1.09) completed two standardized math (Wechsler Individual Achievement Test-II Numerical Operations and Mathematical Reasoning) and reading (WIAT-II Word Reading and Reading Comprehension) assessments. We used hierarchical clustering with complete-linkage criterion based on Euclidean distance (NbClust package in R) to separately cluster the ASD and TD samples by achievement measures. Logistic regression analysis was conducted to investigate cognitive predictors (IQ, verbal, visuo-spatial and executive working memory) of cluster-membership in both groups.

Results:  The NbClust package recommended a two-cluster solution, characterized by one cluster with low and the other with high achievement, in each group (see Figure 1). In the ASD sample, we found a larger difference in math measures (71.54 points), between the Low and High clusters, than that in reading measures (23.35 points, F(1, 116)=104.58, p<.001). Moreover, the difference in math achievement between Low and High clusters was more pronounced in children with ASD than TD children (71.54 vs. 45.84 points, F(1, 210)=24.41, p< .001).

Logistic regression analysis revealed that a) IQ was predictive of cluster-membership in both groups (both p < .01); but (b) distinct components of working memory predicted cluster-membership in ASD vs. TD: in the ASD group, cluster-membership was predicted by verbal (Z = 2.36, p = .018) and executive (Z = 2.44, p = .015) working memory; whereas visuo-spatial working memory was predictive in the TD group (Z = 3.46, p< .001).

Conclusions:  Our study addressed three key questions related to heterogeneity of academic skills in a large group of 7-13 year-old children with ASD. We found that (a) there was a heterogeneous pattern of weakness in math in the low achievement group and relative strength in math in the high achievement group (b) the math weakness in the low achievement group is unique to ASD relative to their TD peers; and (c) cognitive factors such as working memory differentially contribute to heterogeneous patterns of skills in ASD relative to TD children. These findings advance our understanding of heterogeneity in academic achievement in ASD, which may ultimately help us develop appropriate interventions to target and remediate low academic performance in this population.