Scholarlink

Background <p>Preterm infants face higher neurodevelopmental risks than term infants due to immature brain development, including delays, cognitive impairments, and poor academic outcomes. Early identification of neurodevelopmental heterogeneity and key perinatal factors is crucial for improving growth and long-term outcomes in preterm infants.</p> Methods <p>This study enrolled preterm infants born in the Department of Neonatology, Tongji Hospital, between July 2022 and March 2024. Clinical data were collected during hospitalisation, with follow-ups at 3, 6, 9, and 12 months corrected age for infants and their primary caregivers. Group-Based Trajectory Modelling (GBTM) identified neurodevelopmental heterogeneity. Six machine learning algorithms, including Random Forest (RF), were utilised to construct prediction models, evaluated by the Receiver Operating Characteristic (ROC) curve and Shapley Additive exPlanations (SHAP) values.</p> Results <p>A total of 509 neonates completing all four follow-ups were included in the final analysis. Two distinct neurodevelopmental trajectory groups were identified: “Catch-up” (46.56%) and “Stable” (53.44%). The “Catch-up” group showed a rapid increase in DQ from 87.12 to 103.36 between 3 and 12 months, while the “Stable” group exhibited a modest rise from 107.12 to 108.99. Among the six machine learning models, the RF model demonstrated optimal predictive performance in both training (AUC = 0.991, Accuracy = 0.896, F1-score = 0.713) and validation sets (AUC = 0.916, Accuracy = 0.827, F1-score = 0.697), and was selected as the final model. Seven key predictors were identified by SHAP importance: gestational age (0.0427), primary caregiver age (0.0425), caregiver experience with preterm infants (0.0423), intracranial haemorrhage (0.0343), hyperbilirubinaemia (0.0275), birth length (0.0262), and caregiver educational level (0.0230). Preterm infants with higher gestational age, absence of intracranial haemorrhage or hyperbilirubinaemia, birth length of 40–45&#xa0;cm, and primary caregivers aged 25–35 years with prior experience and higher caregiver educational levels were more likely to belong to the Stable trajectory.</p> Conclusion <p>Preterm infants exhibit heterogeneous neurodevelopmental trajectories, classified as “Stable” and “Catch-up”, with the latter characterised by early delays followed by catch-up growth. Gestational age, caregiver age, and caregiver experience were the top three factors most strongly associated with trajectory membership.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Group-based trajectory modelling and interpretable machine learning to identify factors associated with neurodevelopmental trajectories in preterm infants

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月23日

查看全文
  • Kun Dai,
  • Xiaomeng Yang,
  • Mei Huang,
  • Xiali Peng,
  • Dangwei Li,
  • Lu Zheng,
  • Suqing Wang,
  • Zhihui Rong

摘要

Background

Preterm infants face higher neurodevelopmental risks than term infants due to immature brain development, including delays, cognitive impairments, and poor academic outcomes. Early identification of neurodevelopmental heterogeneity and key perinatal factors is crucial for improving growth and long-term outcomes in preterm infants.

Methods

This study enrolled preterm infants born in the Department of Neonatology, Tongji Hospital, between July 2022 and March 2024. Clinical data were collected during hospitalisation, with follow-ups at 3, 6, 9, and 12 months corrected age for infants and their primary caregivers. Group-Based Trajectory Modelling (GBTM) identified neurodevelopmental heterogeneity. Six machine learning algorithms, including Random Forest (RF), were utilised to construct prediction models, evaluated by the Receiver Operating Characteristic (ROC) curve and Shapley Additive exPlanations (SHAP) values.

Results

A total of 509 neonates completing all four follow-ups were included in the final analysis. Two distinct neurodevelopmental trajectory groups were identified: “Catch-up” (46.56%) and “Stable” (53.44%). The “Catch-up” group showed a rapid increase in DQ from 87.12 to 103.36 between 3 and 12 months, while the “Stable” group exhibited a modest rise from 107.12 to 108.99. Among the six machine learning models, the RF model demonstrated optimal predictive performance in both training (AUC = 0.991, Accuracy = 0.896, F1-score = 0.713) and validation sets (AUC = 0.916, Accuracy = 0.827, F1-score = 0.697), and was selected as the final model. Seven key predictors were identified by SHAP importance: gestational age (0.0427), primary caregiver age (0.0425), caregiver experience with preterm infants (0.0423), intracranial haemorrhage (0.0343), hyperbilirubinaemia (0.0275), birth length (0.0262), and caregiver educational level (0.0230). Preterm infants with higher gestational age, absence of intracranial haemorrhage or hyperbilirubinaemia, birth length of 40–45 cm, and primary caregivers aged 25–35 years with prior experience and higher caregiver educational levels were more likely to belong to the Stable trajectory.

Conclusion

Preterm infants exhibit heterogeneous neurodevelopmental trajectories, classified as “Stable” and “Catch-up”, with the latter characterised by early delays followed by catch-up growth. Gestational age, caregiver age, and caregiver experience were the top three factors most strongly associated with trajectory membership.