Multidimensional Dynamic Early Warning System for Neonatal Necrotizing Enterocolitis: A Multicenter Clinical Study

Study design and objectives This is a three-center observational cohort study combining retrospective model development and validation with subsequent prospective validation. The coordinating center is Guangzhou Women's and Children's Medical Center. The participating centers are Northwest Women's and Children's Hospital and the Maternal and Child Health Hospital of Guangxi Zhuang Autonomous Region. The study will evaluate whether genetic susceptibility and longitudinal immune and inflammatory markers improve prediction of necrotizing enterocolitis beyond clinical information alone. Cohorts and separation of development from validation The development cohort will comprise eligible infants from an existing clinical exome sequencing cohort at the coordinating center. Internal validation will use resampling within this development cohort. A separate historical cohort from the coordinating center will be used for temporal validation, provided that its observation period is distinct from that of the development cohort. Independent historical cohorts from the two collaborating hospitals will provide geographic external validation. A subsequent independent cohort at the three hospitals will provide prospective validation after the prediction model and its implementation rules have been finalized. The cohorts will not overlap. All records from the same infant will be assigned to a single cohort, and duplicate records across hospitals will be checked. Historical records will be screened using the same eligibility definitions at each site. Prospective recruitment will be consecutive among eligible infants during the defined recruitment period; no random sampling from the general population is planned. Predictors and biological information Candidate predictors will include perinatal and early postnatal clinical characteristics, eosinophil counts and percentages, the neutrophil-to-lymphocyte ratio, absolute neutrophil counts, platelet counts, white blood cell differential measurements, and prespecified longitudinal features derived from repeated measurements. Genetic predictors will be derived from existing clinical exome sequencing or targeted gene panel results, including candidate gene burden measures and an individual weighted genetic burden score. Existing results will be used whenever available. If additional testing is required and authorized, targeted testing may use residual ethylenediaminetetraacetic acid blood from routine clinical sampling, without additional research-only phlebotomy. Prediction timing and availability of information Predictor values, genetic information and laboratory results must be available before the outcome.The same prespecified timing rules will be applied to infants who do and do not develop the outcome. Development and internal validation Clinical-only, genetic, dynamic inflammatory and integrated prediction models will be developed at the coordinating center. Logistic regression, including least absolute shrinkage and selection operator methods, will be considered for model development and simplification. Landmark-based or other longitudinal methods may be used in accordance with a prespecified statistical analysis plan. Internal validation will assess optimism and overfitting by resampling at the infant level. Predictor selection, transformations, missing-data procedures and model tuning will be repeated within the resampling process rather than performed once using the full dataset before internal validation. Data management and ethics All centers will use harmonized data definitions and collection procedures, with range checks, consistency checks and review of key outcomes. Clinical and genetic information will be coded and held in separate access-controlled systems. Retrospective data will be used under the consent requirements or waivers authorized by the relevant ethics committees. Written permission from a parent or legally authorized guardian will be obtained for prospective participation and any additional genetic analysis as required by the approved protocol. Public reports will present aggregate information and will not identify individual infants.