Dataset from a multicenter randomized controlled trial of AI-based digital cognitive training combined with conventional rehabilitation in young children (2–5 years) with global developmental delay (2024–2025)

This dataset originates from a multicenter, parallel-group randomized controlled trial evaluating the effects of AI-based digital cognitive training combined with conventional rehabilitation in young children with global developmental delay (GDD). Participants aged 2–5 years were recruited from multiple clinical centers and randomly assigned to either an intervention group (digital therapeutics plus conventional rehabilitation) or a control group (conventional rehabilitation alone). The intervention period lasted 16 weeks. Outcome measures included developmental age assessed by the C-PEP-3, adaptive functioning assessed by the S-M scale, and developmental quotients measured by the Gesell Developmental Scale (GDS).Data were generated through standardized clinical assessments conducted by trained evaluators following predefined protocols. All assessment procedures were performed at baseline and post-intervention. Digital training data, including session logs and active training time, were automatically recorded by the AI-based system. Raw data were exported from clinical records and digital platforms, followed by data cleaning, coding, and verification procedures. Missing values were handled using multiple imputation under a fully conditional specification framework for the primary statistical analyses; however, the deposited dataset preserves both original and processed data where applicable.The dataset covers the study period from approximately 2024 to 2025, corresponding to the recruitment and intervention phases. Data are at the individual participant level. Each row represents a single participant, and columns include demographic characteristics (e.g., age, sex), group allocation (intervention vs control), baseline and post-intervention outcome measures, and training-related variables (e.g., average active training time, adherence indicators). Continuous variables are reported in their original measurement units (e.g., months for developmental age, scores for assessment scales).The total dataset includes 118 participants in the intention-to-treat (ITT) population and 98 participants in the per-protocol (PP) population. Missing data are present for a subset of participants due to loss to follow-up or incomplete assessments. No systematic patterns of missingness were identified.Data quality was ensured through double-checking procedures and standardized training of assessors. Potential sources of variability include inter-individual differences in behavioral engagement and assessment conditions. Measurement error is expected to be within the known reliability ranges of the respective clinical scales (C-PEP-3, S-M, and GDS), and no abnormal data distributions or outliers outside plausible clinical ranges were observed after data cleaning.The dataset is provided in commonly used formats (Excel). This dataset can be reused for secondary analyses exploring digital therapeutics, neurodevelopmental outcomes, dose–response relationships, and individualized intervention effects in pediatric populations.