Multiple Correspondence Analysis (MCA)
In the original studies, a total of 104 variables were available for MCA
analysis. Firstly, we removed variables if missing data was
>15%, then further removed variables if there were less
than 20 infants for each categorical result (e.g. present or
absent)23. We then selected variables based on
clinical importance for bronchiolitis severity which included; age
(<12-months), preterm (<37-weeks), low birth weight
(<2.5kg), weight-for-length z-score, exposure to household
smoke or in utero, currently breastfed, lobar collapse/consolidation on
chest x-ray, previous respiratory hospitalisation, sex, remote (defined
as >100km from hospital with paediatric
expertise)7, any antibiotics during hospital,
supplemental oxygen requirement, any co-morbidity, carer-reported cough
or breathing difficulty in last 7-days, any NPS bacteria, bronchiectasis
on HRCT in addition to including as many variables (RSV, HRV, accessory
muscle use and LOS) as Dumas and colleagues included in their
study3. These variables were then entered into the
MCA23 to identify the most relevant variables for LCA
and to decompose the inertia by identifying a small number of mutually
independent dimensions that represent the most important
deviations
from independence24.