Table 1.

Summary of recent and representative studies predicting future ASD diagnosis through multivariate analysis of early behavioral evaluations. Reported sample sizes are the numbers used for classification and do not necessarily reflect the study’s total sample size.

Reference	Study Participants	Experimental Methods	Key Features	Multivariate Technique	Key Results
Macari et al. (2012)173	13 infants that were later diagnosed with ASD and 71 that were not	Evaluated behaviors at 12 months that would be predictive of ASD diagnosis at 24 months	Seven individual items from the ADOS-Toddler	Classification tree	Classified ASD versus non-ASD with 85% sensitivity and 96% specificity
Chawarska et al. (2014)174	157 high-risk infants that were later diagnosed with ASD and 562 that were not	Assessed behaviors at 18 months that would be predictive of ASD diagnosis at 36 months	Six individual items from the ADOS	CART	Classified ASD versus non-ASD with 83% training accuracy and predicted with 77% validation accuracy
Barbaro and Dissanayake (2017)24	77 children at risk for ASD identified from a community-based sample	At 24 months, assessed ASD status and behavior that would predict retention or loss of ASD diagnosis at 48 months	Four items total from the ADOS and Mullen Scales of Early Learning	Logistic regression	Classified the stable group 96% correctly and the crossover group 44% correctly
Bussu et al. (2018)175	32 high-risk infants that were later diagnosed with ASD and 129 that were not	Examined behavior and developmental measures at 8 and 14 months to predict ASD status at 36 months	Motor scores at 8 months and daily living score at 14 months	Least-squares SVM	Best AUROCs for classifying ASD versus non-ASD at 36 months were 0.65 and 0.71 using the 8-month and 14-month measures, respectively