Pediatric Pneumonia Clinical Phenotypes Identified via Machine Learning

A retrospective cohort of 305 pediatric patients with community-acquired pneumonia from November 2022 to October 2025. Meng Xiao applied unsupervised machine learning to identify clinical phenotypes based on microbiological, inflammatory, and clinical data from bronchoalveolar lavage. The study developed a prediction model for a Mixed-Infection phenotype.

Use Cases

• Identify pediatric pneumonia clinical phenotypes based on microbiological and inflammatory markers.
• Develop early prediction models for specific infection phenotypes like Mixed-Infection.
• Analyze associations between clinical features and hospital stay duration.
• Validate unsupervised clustering methods (k-prototypes) for medical cohort analysis.
• Create visual clinical assessment tools (nomograms) for phenotype identification.

Strengths

• Includes 305 patient records with bronchoalveolar lavage evidence.
• Identifies three distinct clinical phenotypes with specific prevalence percentages (37.7%, 28.2%, 34.1%).
• Reports a prediction model performance metric (AUC=0.917, accuracy=91.3%) on a test set.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data is derived from a bronchoscopy/BAL-selected cohort with more severe disease, potentially limiting generalizability.

Provenance

Source

figshare

Collection Method

Retrospective cohort study using bronchoalveolar lavage fluid, inflammatory markers, and clinical features.

Time Range

November 2022 to October 2025

Freshness

Last updated 2026-05-21 05:42:38; freshness should be verified.