Multinomial logistic regression models associated with exacerbation frequency groups during the year following fibreoptic bronchoscopy
To identify parameters that were independently associated with exacerbation frequency during the year following the fibreoptic bronchoscopy, a stepwise multinomial regression model analysis was performed (Table E3). All parameters significantly associated with at least one of the three groups described in Table 3 were initially selected and the following parameters were included in the multinomial regression model analysis: presence of atopic dermatitis (history or current), age at the first episode of wheezing, number of hospitalisations within the year, treatment with LABA at inclusion, the presence of Haemophilus genus in the BALF, the presence ofStreptococcus genus in the BALF, RBM-BSM distance, sub-mucosal fibrosis area, BSM area, and RBM thickness (Table E3). However, parameters with more than 10% missing data (i.e. , Macrophages and PMN in the BALF as well as and density of lymphocytes in bronchial biopsies) were excluded, as previously proposed (15,16). Moreover, when two parameters were strongly associated with each other, only the most discriminating parameter between the groups was included in the analysis. The presence of bacteria, Haemophilus influenzae ,Haemophilus parainfluenza , Streptococcus pneumonia , all in the BALF and post fibreoptic antibiotic treatments were discarded from the model because already represent by Haemophilus orStreptococcus genera in the BALF. Indeed, 97% patients with positive bacteria isolated in the BALF and 95% of those who received a post-fiberoptic antibiotic treatment had at least either aHaemophilus or a Streptococcus genera bacteria. Macrophages/PMN count in the BALF as well as density of lymphocytes in bronchial biopsies were excluded from the model due to excessive missing values (19% and 35%, respectively).
The best model identified atopic dermatitis, age at first episode of wheezing, the presence ofHaemophilus genera in the BALF, the RBM-BSM distance and the BSM area as factors significantly and independently associated with the exacerbation frequency in the year following the bronchoscopy (Table 4 & E3). This model was able to accurately identify the exacerbation frequency group membership in 94.3% of patients (Table E3). In addition, the risk factors independently associated with “Low-Ex” compared to “No-Ex” were the absence of atopic dermatitis, the presence of Haemophilus genus in the BALF and an increased in both the RBM-BSM distance and BSM area (Table 4). The risk factors independently associated with “High-Ex” compared to “No-Ex” were an early age at first episode of wheezing and an increased BSM area (Table 4). The risk factors independently associated with “High-Ex” compared to “Low-Ex” were the presence of atopic dermatitis, an early age at first episode of wheezing, the absence of Haemophilus genus in the BALF, a decrease in the RBM-BSM distance and an increased BSM area (Table 4). This analysis thus identified BSM area as the sole parameter independently and significantly associated with each exacerbation frequency group: the greater the BSM area, the higher the frequency of exacerbations (Table 4).