Abstract:
Background: Early-onset neonatal sepsis (EOS) continues to pose a critical threat to newborns, particularly those born preterm or with very low birth weight. The first 72 hours of life are decisive, yet clinicians must often rely on nonspecific signs and delayed diagnostics. In this context, neonatal intensive care units (NICUs) generate vast, underutilized volumes of clinical data. This project explores how data-driven approach and artificial intelligence (AI) can transform these data streams into timely, explainable support for EOS risk stratification.
Objective: The innovative AI project at Institute of Mother and Child in Warsaw is an interdisciplinary initiative aiming to develop and implement AI-based strategies for early EOS detection. The overarching goal is to empower neonatologists with real-time insights that support earlier and more targeted intervention—moving from reactive treatment to proactive decision-making.
Methods: The project involves the development of a structured clinical data pipeline combining electronic health records, laboratory parameters, and vital signs collected from electronic systems. Using a retrospective cohort, we are currently preparing machine learning models that incorporate temporal patterns in vital signs and laboratory trends (e.g., CRP, WBC). The next steps include internal validation of model performance, and co-design of a prototype alert interface with clinicians to ensure workflow compatibility and explainability. Future phases will include federated learning to expand generalisability across European NICUs while ensuring data privacy.
Results: Data-driven concept for AI system lays the groundwork for safe, clinically grounded implementation of technologies in neonatal sepsis diagnostics. By aligning data science with frontline clinical practice, this work aspires bridging medical data infrastructure, algorithm development, and ethical implementation in paediatric care.
