KNN

사용자ID
사용자명
기관
이메일

Research Project (승인과제목록)

KNN 연구 요약서

Title	Machine Learning-Based Clinical Decision Support for Preterm Premature Rupture of Membranes: Analysis of the Korean Neonatal Network
Author	성세인
작성자	성세인
Background	Preterm premature rupture of membranes (PPROM) presents a significant clinical dilemma regarding optimal timing of delivery, balancing the risks of prematurity against potential complications of prolonged PPROM. Current clinical guidelines provide limited direction on individualized management based on patient-specific factors.
Aim / Hypothesis	To develop and validate a machine learning-based clinical decision support framework for optimizing delivery timing following PPROM. We hypothesize that machine learning models can identify complex interactions between clinical factors (particularly oligohydramnios status, gestational age, and PPROM duration) that significantly impact neonatal outcomes and can be translated into an evidence-based clinical decision framework.
Inclusion Criteria	All infants with birth weight < 1,500 g or gestational age < 32 weeks
Exclusion Criteria	Infants with severe congenital anomaly
Study Design Statistical methods	• Cohort analysis of KNN data • Development and comparison of four machine learning models (Logistic Regression, Gradient Boosting Decision Trees, Explainable Boosting Machines, and CatBoost) • Feature importance analysis using SHAP (SHapley Additive exPlanations) values • Interaction analysis between PPROM duration and key clinical factors • Creation and validation of a clinical decision tree and heat map visualizations for clinical application
Primary Outcomes	Composite adverse neonatal outcome including bronchopulmonary dysplasia, necrotizing enterocolitis, severe intraventricular hemorrhage, retinopathy of prematurity requiring treatment, and death.
Secondary Outcomes and Definitions	• Neonatal mortality (death prior to discharge) • Individual components of the composite outcome • Length of NICU stay • Duration of mechanical ventilation • Bronchopulmonary dysplasia defined as oxygen requirement at 36 weeks postmenstrual age • Necrotizing enterocolitis defined as Bell's stage ≥2 • Severe intraventricular hemorrhage defined as grade III-IV • Retinopathy of prematurity requiring treatment defined as stage ≥3 or requiring laser therapy or anti-VEGF treatment • early and late sepsis
Protocols	We will extract all cases of confirmed PPROM from the KNN database, including maternal characteristics, pregnancy complications, and neonatal outcomes. Machine learning models will be developed using 70% of the data for training and 30% for validation, with hyperparameter optimization performed using cross-validation. Feature importance analysis will identify key predictors of outcomes, with special focus on the interaction between PPROM duration and oligohydramnios status. Based on model findings, we will develop a clinical decision tree that stratifies management recommendations according to key clinical factors. The final decision support framework will be validated by assessing its potential impact on clinical outcomes if it had been applied to the validation dataset.
Funding	None