Leveraging a Unique Dataset to Identify Outcome Predictors in Late Talkers

Late talking in toddlers is not only a common reason for pediatrician concern and referral for developmental evaluation, but it can be a precursor of either persistent language disorder; worsening functional ability such as autism or global delay; or, oddly enough, the opposite: a neurotypical outcome. Yet, predictors of such

dramatically divergent, heterogenous outcomes remain elusive. Literature reviews reveal substantial lack of replication in the field, even among the strongest findings, perhaps because many studies tend to have samples too small to address heterogenous trajectories; use non-comparable ascertainment and recruitment; and engage

limited language, clinical, social, and neurobehavioral assessments. In contrast, our existent sample: (a) is very large and includes N=1,667 toddlers (552 late talkers), (702 ASD) and (413 typical); (b) were all ascertained, recruited and clinically characterized in a uniform procedure by licensed clinical psychologists; (c) is

representative of the spectrum of late talkers and typical toddlers; and (d) were longitudinally phenotyped at toddler (mean age 20 months) and preschool ages (mean age 36 months) using the same language and clinical tests. In our sample of N=552 late talking toddlers defined using a cut-off of expressive language (EL) < -1 SD,

51% had persistent expressive language delays or worsening language outcomes such as ASD or global delay by preschool. Yet, 49% of our late talkers made rapid and substantial expressive language advances, achieving neurotypical levels by preschool. AIM 1 will leverage this unique sample to identify toddler-age precursors

predictive of one of 5 divergent language & clinical outcomes at preschool ages (Transient EL Delay; Persistent EL Delay; Conversion to LD; Conversion to GDD; Conversion to ASD). Nine commonly reported predictors of expressive language outcomes will be analyzed using linear regression specifically: receptive language ability

at intake; expressive vocabulary size at intake; % nouns and shape nouns in vocabulary composition; SES, sex, socialization, mean length of utterance (MLU) and % verbal initiations. Multinomial logistic regression will determine which of the toddler-age variables are most strongly associated with clinical and language outcome

group membership. Change across time for each measure within each outcome group will also be analyzed. AIM 2: Social and language development are inextricably linked, and measures of attention to social speech such as motherese and social images have been shown to be associated with language ability. To go beyond

commonly examined predictors, AIM 2 will leverage previously collected eye tracking (ET) data of auditory and visual social attention in late talking, ASD, and TD toddlers. Using our large TD sample, reference standards for levels of social auditory and social visual attention based on 7 key metrics (e.g., level of attention to motherese

speech) across 2-month age bands will be created. Deviations from these norms will be used to reveal novel predictors of preschool outcomes. Using unsupervised Similarity Network Fusion, multimodality late talker ET- clinical subtypes will be defined, and predictors associated with each subtype identified.