why choose us

300×250 Ad Slot

Research Article: Identification of clinical phenotypes and prediction model for the mixed-infection phenotype of pediatric community-acquired pneumonia based on unsupervised machine learning

Date Published: 2026-05-21

Abstract:
Pediatric community-acquired pneumonia (CAP) exhibits significant clinical heterogeneity. Traditional microbiological classification overlooks host factors, making it challenging to accurately determine prognosis and provide targeted, precise treatment. Based on unsupervised machine learning, this study integrates microbiological, host inflammatory response, and clinical characteristics to phenotype pediatric CAP and develops an early prediction model for the Mixed-Infection phenotype. A retrospective cohort of 305 pediatric patients with CAP who underwent bronchoalveolar lavage (BAL) was included between November 2022 and October 2025. Using microbiological evidence from BAL fluid, inflammatory markers, and clinical features, k-prototypes clustering was applied to identify and classify phenotypes. A decision tree and nomogram were developed to predict the Mixed-Infection phenotype. Three clinical phenotypes were identified through machine learning: Mycoplasma-Dominant (37.7%), characterized by Mycoplasma infection with moderate inflammatory response; Mixed-Infection (28.2%), characterized by multi-pathogen coinfection, the youngest age group, and the most extended hospital stays; and High-Inflammation (34.1%), characterized by elevated CRP and WBC levels. The Mixed-Infection phenotype had the highest proportion of prolonged hospitalization (31.4%). However, this difference did not reach statistical significance ( p =?0.117), suggesting a trend toward higher medical resource utilization that warrants further investigation. A model based on white blood cell count, lactate dehydrogenase, and procalcitonin performed well on the test set (AUC?=?0.917, accuracy?=?91.3%). The nomogram provided a visual clinical assessment tool for early identification of the Mixed-Infection phenotype. This study systematically applied k-prototypes clustering to identify three clinical phenotypes, revealing distinct "pathogen-host" interaction patterns among them. We developed a simple early identification tool for the Mixed-Infection phenotype. However, our findings are derived from a bronchoscopy/BAL-selected cohort with more severe or complex disease, which may limit generalizability to all pediatric CAP patients. While this tool shows significant potential, further validation in larger prospective cohorts is needed to confirm its generalizability and clinical applicability.

Introduction:
Community-acquired pneumonia (CAP) remains one of the most common and severe infectious diseases in children worldwide, contributing substantially to morbidity, healthcare utilization, and mortality in children under five years of age ( 1 ). Despite advances in vaccination and antimicrobial therapy, the clinical management of pediatric CAP remains a fundamental challenge due to substantial heterogeneity in disease presentation, progression, and treatment response. Children infected with the same pathogen can…

Read more

300×250 Ad Slot