Perinatology.  2022 Sep;33(3):127-135. 10.14734/PN.2022.33.3.127.

Unsupervised Clustering of Late Preterm Infants in Terms of Developmental Outcome

• Kim H
• Song JH
• Kim SS

• KMID: 2533888
• DOI: http://doi.org/10.14734/PN.2022.33.3.127

Abstract

Objective
This study aimed to derive the subtype of late preterm infants (gestational age, 34 to 36 weeks) according to their developmental outcomes.
Methods
We retrospectively investigated the medical records of premature infants who had undergone developmental testing and were discharged from a single regional newborn intensive care center. We used 5 domains (motor, language, cognition, social-emotional, adaptive behavior) of the Korean version of the Bayley scale of infant and toddler development III (K-Bayley III) to group subjects. K-means clustering (KM), hierarchical clustering, and density-based spatial clustering of applications with noise were used. We used the average silhouette index (ASI) and Calinski-Harabasz (C-H) score as evaluation metrics.
Results
KM showed the best performance (ASI, 0.25; C-H score, 58.83) and revealed 3 clusters. Cluster 1 (need observation) showed low normal scores in K-Bayley III Scales, and cluster 2 (excellent development) showed high normal scores. In contrast, cluster 3 (global delay) showed delayed or borderline scores other than the social-emotional scale. Maternal age (P<0.01), number of fetuses (P=0.03), prenatal steroid use P=0.01), pH (P<0.01), and base excess (P=0.03) showed a statistical significance among the 3 clusters.
Conclusion
The authors found 3 phenotypes with distinct developmental outcomes among late preterm infants and discovered variables necessary for their prediction. If the target group, requiring developmental testing, can be screened early by using these predictors, it may be beneficial in improving the developmental prognosis of late preterm infants.

Keyword

Premature infants; Child development; Cluster analysis; Unsupervised machine learning