Identifying clinical phenotypes in extremely low birth weight infants-an unsupervised machine learning approach

Reassessing acquired neonatal intestinal diseases using unsupervised machine learning

BACKGROUND

Acquired neonatal intestinal diseases have an array of overlapping presentations and are often labeled under the dichotomous classification of necrotizing enterocolitis (which is poorly defined) or spontaneous intestinal perforation, hindering more precise diagnosis and research. The objective of this study was to take a fresh look at neonatal intestinal disease classification using unsupervised machine learning.

METHODS

Patients admitted to the University of Florida Shands Neonatal Intensive Care Unit January 2013-September 2019 diagnosed with an intestinal injury, or had imaging findings of portal venous gas, pneumatosis, abdominal free air, or had an abdominal drain placed or exploratory laparotomy during admission were included. Congenital gastroschisis, omphalocele, intestinal atresia, malrotation were excluded. Data was collected via retrospective chart review with subsequent hierarchal, unsupervised clustering analysis.

RESULTS

Five clusters of intestinal injury were identified: Cluster 1 deemed the "Low Mortality" cluster, Cluster 2 deemed the "Mature with Inflammation" cluster, Cluster 3 deemed the "Immature with High Mortality" cluster, Cluster 4 deemed the "Late Injury at Full Feeds" cluster, and Cluster 5 deemed the "Late Injury with High Rate of Intestinal Necrosis" cluster.

CONCLUSION

Unsupervised machine learning can be used to cluster acquired neonatal intestinal injuries. Future study with larger multicenter datasets is needed to further refine and classify types of intestinal diseases.

IMPACT

Unsupervised machine learning can be used to cluster types of acquired neonatal intestinal injury. Five major clusters of acquired neonatal intestinal injury are described, each with unique features. The clusters herein described deserve future, multicenter study to determine more specific early biomarkers and tailored therapeutic interventions to improve outcomes of often devastating neonatal acquired intestinal injuries.

Machine Learning Detects Intraventricular Haemorrhage in Extremely Preterm Infants

OBJECTIVE

To test the potential utility of applying machine learning methods to regional cerebral (rcSO₂) and peripheral oxygen saturation (SpO₂) signals to detect brain injury in extremely preterm infants.

STUDY DESIGN

A subset of infants enrolled in the Management of Hypotension in Preterm infants (HIP) trial were analysed (n = 46). All eligible infants were <28 weeks' gestational age and had continuous rcSO₂ measurements performed over the first 72 h and cranial ultrasounds performed during the first week after birth. SpO₂ data were available for 32 infants. The rcSO₂ and SpO₂ signals were preprocessed, and prolonged relative desaturations (PRDs; data-driven desaturation in the 2-to-15-min range) were extracted. Numerous quantitative features were extracted from the biosignals before and after the exclusion of the PRDs within the signals. PRDs were also evaluated as a stand-alone feature. A machine learning model was used to detect brain injury (intraventricular haemorrhage-IVH grade II-IV) using a leave-one-out cross-validation approach.

RESULTS

The area under the receiver operating characteristic curve (AUC) for the PRD rcSO₂ was 0.846 (95% CI: 0.720-0.948), outperforming the rcSO₂ threshold approach (AUC 0.593 95% CI 0.399-0.775). Neither the clinical model nor any of the SpO₂ models were significantly associated with brain injury.

CONCLUSION

There was a significant association between the data-driven definition of PRDs in rcSO₂ and brain injury. Automated analysis of PRDs of the cerebral NIRS signal in extremely preterm infants may aid in better prediction of IVH compared with a threshold-based approach. Further investigation of the definition of the extracted PRDs and an understanding of the physiology underlying these events are required.

Identifying two distinct subphenotypes of patent ductus arteriosus in preterm infants using machine learning

To use unsupervised machine learning to identify potential subphenotypes of preterm infants with patent ductus arteriosus (PDA). The study was conducted retrospectively at a neonatal intensive care unit in Brazil. Patients with a gestational age < 28 weeks who had undergone at least one echocardiogram within the first two weeks of life and had PDA size > 1.5 or LA/AO ratio > 1.5 were included. Agglomerative hierarchical clustering on principal components was used to divide the data into different clusters based on common characteristics. Two distinct subphenotypes of preterm infants with hemodynamically significant PDA were identified: "inflamed," characterized by high leukocyte, neutrophil, and neutrophil-to-lymphocyte ratio, and "respiratory acidosis," characterized by low pH and high pCO2 levels.    Conclusions: This study suggests that there may be two distinct subphenotypes of preterm infants with hemodynamically significant PDA: "inflamed" and "respiratory acidosis." By dividing the population into different subgroups based on common characteristics, it is possible to get a more nuanced understanding of the effectiveness of PDA interventions. What is Known: • Treatment of PDA in preterm infants has been controversial. • Stratification of preterm infants with PDA into subgroups is important in order to determine the best treatment. What is New: • Unsupervised machine learning was used to identify two subphenotypes of preterm infants with hemodynamically significant PDA. • The 'inflamed' cluster was characterized by higher values of leukocyte, neutrophil, and neutrophil-to-lymphocyte ratio. The 'respiratory acidosis' cluster was characterized by lower pH values and higher pCO2 values.