Predicting early mortality and severe intraventricular hemorrhage in very-low birth weight preterm infants: a nationwide, multicenter study using machine learning

Yun Hsiang Yang, Ts Ting Wang, Yi Han Su, Wei Ying Chu, Wei Ting Lin, Yen Ju Chen, Yu Shan Chang, Yung Chieh Lin, Chyi-Her Lin, Yuh Jyh Lin

Research output: Contribution to journalArticlepeer-review

Abstract

Our aim was to develop a machine learning-based predictor for early mortality and severe intraventricular hemorrhage (IVH) in very-low birth weight (VLBW) preterm infants in Taiwan. We collected retrospective data from VLBW infants, dividing them into two cohorts: one for model development and internal validation (Cohort 1, 2016–2021), and another for external validation (Cohort 2, 2022). Primary outcomes included early mortality, severe IVH, and early poor outcomes (a combination of both). Data preprocessing involved 23 variables, with the top four predictors identified as gestational age, birth body weight, 5-min Apgar score, and endotracheal tube ventilation. Six machine learning algorithms were employed. Among 7471 infants analyzed, the selected predictors consistently performed well across all outcomes. Logistic regression and neural network models showed the highest predictive performance (AUC 0.81–0.90 in both internal and external validation) and were well-calibrated, confirmed by calibration plots and the lowest two mean Brier scores (0.0685 and 0.0691). We developed a robust machine learning-based outcome predictor using only four accessible variables, offering valuable prognostic information for parents and aiding healthcare providers in decision-making.

Original languageEnglish
Article number10833
JournalScientific reports
Volume14
Issue number1
DOIs
Publication statusPublished - 2024 Dec

All Science Journal Classification (ASJC) codes

  • General

Fingerprint

Dive into the research topics of 'Predicting early mortality and severe intraventricular hemorrhage in very-low birth weight preterm infants: a nationwide, multicenter study using machine learning'. Together they form a unique fingerprint.

Cite this