Project Summary A significant portion of newborns (6-10%) enter neonatal intensive care units (NICUs) in the U.S. and without adequate nutrition, this vulnerable population could suffer from long-term consequences. Total parenteral nutrition (TPN) is crucial for the care of premature infants in NICUs not only for growth but also for improving other health outcomes. However, current practices of TPN administration are expensive, lack scientific support, and have low accessibility. The process is also the most cited error within NICUs, potentially leading to adverse outcomes in infants. The goal of this proposal is to address these drawbacks by developing an FDA-approved artificial intelligence (AI)-driven TPN platform (TPN2.0) derived from our unique data collected over 10 years that link TPN composition to adverse outcomes. The model would also be built on our prior work that predicts adverse outcome risks in newborns based on maternal and child electronic health records. The platform development will be accomplished through two steps. First, we will develop a set of standardized TPN units whose number is manageable for central manufacturing, leveraging the benefits of standardized units, without sacrificing their personalization. This is based on our proven hypothesis that similar TPN compositions can be grouped together without noticeable changes in clinical decisions. Second, we will validate our TPN2.0 platform in real clinical settings to show non-inferiority performance to the best standard of care. In Phase 1, our initial focus is on developing an AI model that groups infants based on their profiles and suggests corresponding TPN compositions. This involves using an advanced transformer model with a clustering layer that takes advantage of the longitudinal nature of lab test values and TPN data. To validate the model, we plan to involve healthcare professionals in a study where they evaluate different solutions. In Phase 2, we aim to further validate this approach across larger multi-site groups of professionals. Additionally, we'll work on software development for submission to regulatory authorities, seeking approval for a Class II SaMD. Our validation studies and regulatory pathways have been guided by consultations with a regulatory consultant. Early results are promising, resulting in the creation of a standardized set of TPN solutions suitable for multi-site testing. Notably, evaluations from experts favored our solution over individual prescriptions. Furthermore, the analysis revealed a lower incidence of certain adverse outcomes with TPN2.0 compared to prescriptions. While not required for regulatory approval, we will also conduct a clinical pilot where actual newborns would receive TPN2.0 units, enabling us to gather performance data for comparison. These findings will support our product and lay the groundwork for potential sales and broader pilot initiatives.