Temperament As an Early Risk Marker for Autism Spectrum Disorders in a Cohort of the British Autism Study of Infant Siblings (BASIS)
Objectives: (1) Investigating group differences in early temperament at and across multiple time points between HR-ASD (HR siblings subsequently diagnosed with ASD), HR-Atypical (HR siblings not diagnosed with ASD, but following an atypical development), HR-Typical (HR siblings with a typical development), and LR siblings, and (2) Predicting ASD and atypical development at 36 months at an individual level based on temperament during the first two years of life.
Methods: As part of the British Autism Study of Infant Siblings (BASIS), 247 infants (170 HR, 77 LR) participated in a battery of assessments during their first three years of life. Parents completed the Infant Behavior Questionnaire-Revised (IBQ-R) at 8- and 14 month visits, and the Early Childhood Behavior Questionnaire (ECBQ) at the 24 month visit. (1) MANCOVAs were used to investigate whether a risk gradient was present in polynomial group contrasts (HR-ASD > HR- Atypical > HR-Typical > LR) at separate time points. In post-hoc analyses pair wise group contrasts were examined across time by performing two-way mixed ANCOVAs. (2) Machine learning algorithms were used to investigate how temperamental measures at 8, 14 and 24 months related to atypical development, and more specifically ASD, at an individual level. Sex and age at the first visit were included as covariates in all analyses.
Results: (1) Risk gradients were found for SU at 14 months (Contrast Estimate [CE] = 0.40, p = 0.02), NA at 8, 14 and 24 months (CE = -0.46, p = 0.004; CE = -0.38, p = 0.02; CE = -0.69, p < 0.001, respectively), and EC at 14 and 24 months (CE = 0.69, p < 0.001; CE = 0.84, p < 0.001, respectively). Furthermore, differences were found between groups based on both early trajectories (as revealed by interaction effects between group and time) and levels (as shown by main group effects) of temperament. (2) None of the temperamental traits was able to accurately predict ASD at 36 months. For prediction of HR-ASD, the combination of all factors at 24 months provided the most promising classifier (AUC=72%, p=0.02). For prediction of atypical outcome (i.e., HR-ASD plus HR-Atypical), the integration of EC and NA at 24 months provided the most promising classifier, but performance was not significantly different from chance level (AUC=61%, p=0.056).
Conclusions: Our findings indicate that although differences on temperamental traits at a group level can be detected early in infancy, this does not necessarily translate into an acceptably accurate individual classification. Explanations for findings and recommendations for future research will be discussed.