Bayesian Hierarchical Multidimensional Item Response Modeling of Small Sample, Sparse Data for Personalized...

Area of Science:

• Developmental Psychology
• Biostatistics
• Pediatric Health

Background:

• Developmental surveillance tools are crucial for monitoring infant and child development.
• Sparse data from online tools presents challenges in accurate developmental assessment.
• Existing methods may not fully capture the complex, multidimensional nature of early development.

Purpose of the Study:

• To implement a novel multidimensional item response model for developmental profiling.
• To utilize Bayesian hierarchical priors for robust analysis of sparse developmental data.
• To create personalized developmental profiles for early identification of children with developmental delay.

Main Methods:

• A multidimensional item response model within a Bayesian hierarchical framework was applied.
• Latent abilities of children and attributes of 348 developmental milestones (birth to 3 years) were estimated.
• Hierarchical clustering of ability estimates identified child subgroups for profile construction.

Main Results:

• The model successfully constructed developmental profiles from sparse data in a small cohort (N=115).
• Individual profiles integrated latent ability estimates and milestone attributes across six domains (auditory, hands, movement, speech, tactile, vision).
• The approach retained the correlation structure among latent developmental domains, offering a nuanced view.

Conclusions:

• The developed method provides a robust approach to developmental surveillance using online tools.
• Personalized developmental profiles facilitate early identification of developmental delays.
• This framework supports tailored early intervention strategies for at-risk children.