Latent Class Analysis to identify clinical profiles among Indigenous
infants with bronchiolitis
Abstract
Abstract Background and Objectives: Better phenotyping of the
heterogenous bronchiolitis syndrome may lead to targeted future
interventions. This study aims to identify severe bronchiolitis profiles
among hospitalised Australian Indigenous infants, a population at
high-risk of bronchiectasis, using Latent Class Analysis (LCA). Methods:
We included prospectively collected clinical, viral and nasopharyngeal
bacteria data from 164 Indigenous infants hospitalised with
bronchiolitis. We undertook multiple correspondence analysis (MCA)
followed by LCA. The best-fitting model for LCA was based on adjusted
Bayesian information criteria and entropy R2. Results: We identified
five clinical profiles. Profile-A’s (23.8% of cohort) phenotype was
previous preterm (90.7%), low birth-weight (89.2%) and
weight-for-length z-score <-1 (82.7% from combining those
with z-score between -1 and -2 and those in the z-score of <-2
group) previous respiratory hospitalisation (39.6%) and bronchiectasis
on chest high-resolution computed tomography scan (35.4%). Profile-B
(25.3%) was characterised by oxygen requirement (100%) and marked
accessory muscle use (45.5%). Infants in profile-C (7.0%) had the most
severe disease, with oxygen requirement and bronchiectasis in 100%,
moderate accessory muscle use (85% vs 0-51.4%) and bacteria detected
(93.1% vs 56.7-72.0%). Profile-D (11.6%) was dominated by rhinovirus
(49.4%), mild accessory muscle use (73.8%) and weight-for-length
z-score <-2 (36.0%). Profile-E (32.2%) included
bronchiectasis (13.8%), RSV (44.0%), rhinovirus (26.3%) and any
bacteria (72%). Conclusions: Using LCA in Indigenous infants with
severe bronchiolitis, we identified 5 clinical profiles with one
distinct profile for bronchiectasis. LCA can characterise distinct
phenotypes for severe bronchiolitis and infants at risk for future
bronchiectasis, which may inform future targeted interventions.