This study tests whether an artificial intelligence (AI) tool can help doctors order total parenteral nutrition (TPN) for babies in the neonatal intensive care unit (NICU). Premature babies often cannot eat by mouth and need nutrition delivered through an IV. Ordering TPN is complex, time-consuming, and mistakes can happen. This study will test an AI tool that suggests TPN formulas to doctors based on each baby's lab values and health information. Doctors can accept, change, or reject the suggestions at any time. The main goal is to measure how often doctors accept the AI suggestions. The study will also track time to complete TPN orders, weight changes, days on TPN, whether lab values stay in normal ranges, provider satisfaction, and baby health outcomes including complications such as lung disease, brain bleeding, infections, and other conditions common in premature babies. Babies admitted to the NICU who need TPN may participate if their doctors agree to use the tool. Each baby will be in the study while they need TPN, typically about 14 days. The AI tool only makes suggestions and does not replace doctor decision-making. All other care remains the same as standard practice.
Our AI-driven TPN (TPN2.0) platform is a combination of AI and a premade set of TPN units. The AI is used to formulate and assign the optimal TPN unit to each infant, given their daily profile and lab test values. It is driven by decades of data, including our published morbidity risks, basic demographics, and routinely collected lab test values. The approach will save staff time and eliminate high errors in the current TPN ordering process. Our pilot will be deployed as a clinical decision support tool that only makes recommendations, and doctors can always override it. As such, this minimally affects the current practice. We aim to enroll 260 neonates in this pilot study. The primary outcome is physician acceptance rate of TPN2.0 recommendations. Secondary outcomes include time to complete TPN orders, change in weight z-score, days on TPN, lab value abnormalities (values outside normal range), provider satisfaction, and a composite morbidity index comprising rates of bronchopulmonary dysplasia, necrotizing enterocolitis, retinopathy of prematurity, respiratory distress syndrome, congenital heart disease, sepsis, anemia, intraventricular hemorrhage, cholestasis, jaundice, pulmonary hemorrhage, pulmonary hypertension, readmission during the study, and mortality.
Study Type
INTERVENTIONAL
Allocation
NON_RANDOMIZED
Purpose
HEALTH_SERVICES_RESEARCH
Masking
NONE
Enrollment
260
An AI-driven clinical decision support (CDS) software integrated with EHR system that provides TPN composition recommendations to NICU providers. The tool uses patient lab values, basic profile (days since birth, weight, gestational age), and physician inputs to suggest TPN components. Providers can accept, modify, or decline if needed. The final prescribing authority remains with the providers. The intervention targets provider workflow efficiency while maintaining precision and equivalent patient outcomes (including labs and long-term adverse outcomes).
Stanford University
Stanford, California, United States
RECRUITINGSystem Acceptance Rate
Percentage of AI-generated recommendations that are accepted or modified by providers for each ingredient of TPN. Measured by retrospective comparison between AI suggestion and actual TPN order submitted.
Time frame: 10 months
Change in Weight Z-Score
Change in weight-for-age z-score from admission to discharge. Calculated from daily weights recorded in EMR. Target is to maintain or improve z-score trajectory in Period 2 compared to Period 1.
Time frame: 10 months
Days on TPN
Number of days the infants have to get TPN
Time frame: 10 months
Composite Clinical Outcome
Outcome Measure: Composite Clinical Outcome Description: Number and proportion of participants who experienced at least one of the following neonatal morbidities during the study period: bronchopulmonary dysplasia (BPD), necrotizing enterocolitis (NEC), retinopathy of prematurity (ROP), respiratory distress syndrome (RDS), congenital heart disease (CHD), sepsis, anemia, intraventricular hemorrhage, cholestasis, jaundice, pulmonary hemorrhage, pulmonary hypertension, hospital readmission during the study period, or mortality. Each morbidity is assessed in the same unit -- as present or absent. Participants experiencing one or more listed morbidities are counted once in the composite outcome.
Time frame: 13 months
Rate of Laboratory Value Abnormalities
Outcome Measure: Abnormal Serum Laboratory Values Description: Number and proportion of participants with ≥1 serum laboratory value outside the age-appropriate normal range during the study period. Laboratory values assessed include sodium, potassium, calcium, magnesium, phosphorus, glucose, triglycerides, aspartate aminotransferase (AST), alanine aminotransferase (ALT), and direct bilirubin. Participants with one or more abnormal values will be counted once in this outcome.
Time frame: 10 months
This platform is for informational purposes only and does not constitute medical advice. Always consult a qualified healthcare professional.