Development and Multicenter Validation of an Explainable System for Neonatal ARDS-Compatible Phenotyping

Neonatal acute respiratory distress syndrome (NARDS) can be difficult to identify consistently because relevant information is distributed across clinical notes, imaging reports, and respiratory support records. This study will develop and validate an explainable system that organizes routinely collected electronic health record information and identifies infants whose clinical findings are compatible with NARDS. The study will involve three neonatal intensive care units in China. Historical records from the lead center will be used for system development. A later period at the lead center and independent records from the two collaborating centers will be used for temporal and geographic validation. After the system has been locked, newly eligible infants will be included in a prospective validation phase without changing their clinical care. System results will be compared with an independent reference assessment by neonatal intensive care specialists. The specialists will classify each infant's first eligible respiratory event as a high-likelihood NARDS-compatible clinical phenotype, a non-compatible clinical phenotype, clinically uncertain, or insufficient information. The study will evaluate sensitivity, specificity, predictive values, review workload, and the ability to identify insufficient information. During prospective validation, system results will remain unavailable to treating clinicians. The system is being evaluated as a research screening and review tool; it is not intended to replace a full clinical diagnosis or to determine treatment.

Inclusion Criteria: 1. Receiving care in the NICU of a participating hospital before 44 weeks of postmenstrual age at the index event. 2. Meeting at least one prespecified candidate screening criterion, including a relevant acute lung injury or NARDS/ARDS expression, respiratory support or support escalation, bilateral or diffuse pulmonary parenchymal abnormalities in an imaging report, or a prespecified acute lung injury trigger. 3. Having records within the prespecified center and calendar-period cohort, or being newly eligible during the approved prospective validation period. 4. Availability of an appropriate legal basis for use of the records, including parental or guardian consent or an ethics-approved waiver when applicable. Exclusion Criteria: 1. Structural lung or airway malformations, complex congenital heart disease, or clearly established cardiac dysfunction or fluid overload that fully explains the respiratory failure and imaging abnormalities. 2. Short-term postoperative mechanical ventilation without evidence of acute pulmonary injury. 3. Duplicate or unresolvable records that prevent reliable identification of the infant or index event.