BuddyBench: A Privacy-Constrained Multi-Task Benchmark for Pediatric Social-Communication Personalization

BuddyBench introduces a privacy-protected multi-task benchmark dataset combining clinical assessments, learning trajectories, and treatment outcomes for pediatric social-communication research. The dataset integrates two cohorts (189 observational and 86 randomized controlled trial participants) to enable knowledge tracing, clinical prediction, and causal inference while maintaining pediatric data protection standards.

BuddyBench addresses a critical gap in pediatric neurodevelopmental research by creating an integrated benchmark that moves beyond siloed data collection. Existing repositories typically emphasize isolated data types—imaging, genetics, or cross-sectional assessments—but fail to connect behavioral learning patterns with clinical outcomes. This benchmark links granular drill-level learning trajectories with standardized clinical assessments and randomized treatment endpoints, enabling researchers to trace how personalized interventions affect developmental trajectories over time.

The dataset's architecture reflects growing recognition that pediatric healthcare increasingly requires longitudinal, multi-modal data integration. By combining an observational cohort with dense behavioral tracking (ND-03) alongside a randomized controlled trial cohort (ND-02), BuddyBench supports diverse analytical approaches from knowledge tracing algorithms to causal inference methods. The inclusion of BuddyBench-Sim, a synthetic companion dataset, addresses reproducibility concerns while protecting sensitive pediatric records—a crucial consideration for clinical datasets involving minors.

For AI researchers and healthcare technologists, this benchmark enables development of personalization algorithms that predict which interventions benefit which children, advancing precision medicine in neurodevelopmental care. The baseline results demonstrating signal across tasks suggest the dataset contains meaningful patterns suitable for training and evaluating machine learning models. This infrastructure could accelerate research into automated intervention recommendation systems and clinical decision support tools.

Future value depends on adoption within the research community and whether this benchmark spawns similar integrated datasets across other pediatric domains. The privacy-first design sets a template for responsible clinical AI development, potentially influencing how healthcare institutions structure their data collection and sharing practices.

• →BuddyBench integrates learning trajectories, clinical assessments, and treatment outcomes into a unified multi-task benchmark for pediatric social-communication research.
• →The dataset combines 189 observational participants with 86 randomized controlled trial participants, supporting knowledge tracing, prediction, and causal inference tasks.
• →Privacy-constrained design with synthetic companion dataset (BuddyBench-Sim) enables reproducible research while protecting sensitive pediatric clinical records.
• →Baselines demonstrate signal across all tasks, indicating the dataset's readiness for training machine learning models for intervention personalization.
• →The benchmark advances precision medicine by linking behavioral personalization algorithms to measurable clinical outcomes in children.