Identification of severe acute pediatric asthma phenotypes using unsupervised machine learning

Area of Science:

• Pediatric pulmonology
• Clinical informatics
• Biostatistics

Background:

• Targeted management of severe acute pediatric asthma is crucial for improving patient outcomes.
• Current treatment approaches may not adequately address the heterogeneity of severe asthma presentations in children.

Purpose of the Study:

• To identify distinct clinical phenotypes of severe acute pediatric asthma.
• To utilize early hospitalization data (within the first 12 hours) for phenotype identification.

Main Methods:

• Retrospective cohort study (2014-2022) including children aged 2-18 years admitted for asthma.
• Consensus k-means clustering applied to demographics, vital signs, diagnostics, and laboratory data from the first 12 hours of admission.
• Study population of 683 encounters divided into derivation (80%) and validation (20%) sets.

Main Results:

• Two distinct clusters (phenotypes) were identified.
• Cluster 2 (32% of encounters) included older children, more males, Black race, and non-Hispanic ethnicity compared to Cluster 1.
• Cluster 2 showed less improvement in vital signs, different neutrophil/lymphocyte percentages, higher rates of invasive mechanical ventilation, longer hospital stays, and higher mortality.

Conclusions:

• Two distinct clinical phenotypes of severe acute pediatric asthma were identified.
• These phenotypes are characterized by differing clinical features and patient outcomes.
• Findings support the need for phenotype-specific management strategies in severe pediatric asthma.