Identification of bronchiolitis profiles in Italian children through the application of latent class analysis

Bronchiolitis is the primary infection of the lower respiratory tract in children under 2 years of age, imposing a huge clinical burden as one of the commonest causes of hospital admission in this age group [1‐4]. Although it is generally considered a single nosological entity, the clinical presentation could be extremely variable and recent studies have suggested that this condition is characterized by remarkable heterogeneity [5, 6]. Even hospitalized children may show various short-term outcomes (eg. length of hospital stay, risk of recurrence) [7‐9], as well as different mid- and long-term sequelae, such as the risk of developing recurrent wheezing and asthma [10, 11]. Therefore, identifying different disease phenotypes might contribute to develop and implement proper preventive and therapeutic strategies. Most of the studies conducted so far have generally classified children with bronchiolitis on the basis of a limited number of characteristics (eg. type of virus, presence of wheezing, severity score). Of note, a multidimensional approach has proven to be more effective in clarifying the intrinsic heterogeneity of other respiratory disorders such as asthma and wheezing in childhood. In this context, clustering statistical methods are suitable for categorizing a heterogeneous population into subpopulations who share some aspects of a disease [12]. In particular, Latent Class Analysis (LCA) is a “hypothesis-free” approach, which assigns patients to classes based on their homogeneous characteristics, rather than them being arbitrarily assigned to classes by the researchers [13]. This method has been successfully used to detect subgroups of children with wheezing or asthma who share similar characteristics of the disease [14‐17]. Using this statistical approach in multicenter cohorts of US (MARC-30 USA) and Finnish (MARC-30 Finland) children hospitalized with severe bronchiolitis, disease profiles were identified that differed based on the personal history of wheezing/eczema, wheezing during acute infection, severity levels and viral aetiology [18]. More recently, LCA has been applied in a study conducted on a large cohort of US children (MARC-35) hospitalized for bronchiolitis and followed up to the age of three, allowing the identification of three disease profiles significantly different with respect to inflammation and atopy markers, the nasopharyngeal microbiota, and the development of recurrent wheezing at 3 years of life [19].

No similar studies have been performed in the Italian children population so far. Therefore, the aim of the current study was to identify and describe discrete profiles of children hospitalized with bronchiolitis by means of LCA in a retrospective, single centre study conducted in Palermo, Italy, and to compare our findings with those obtained in international cohorts.

The current study identified distinct clinical profiles of children hospitalised for bronchiolitis through the application of a hypothesis-free statistical clustering approach, such as LCA, which allowed us to describe three discrete phenotypes of patients based on gender, age, clinical characteristics at admission.

Clustering statistical methods are useful for identifying profiles that summarize shared aspects of disease within different groups of patients. LCA has been previously used to identify three to four homogeneous subgroups of patients starting from a large series of observed characteristics in multicenter cohorts of hospitalized children with severe bronchiolitis [18]. More recently, the same approach has been applied in a study conducted on a large population of US children hospitalized for bronchiolitis and followed up to the age of 3 years, finding three distinct profiles of children that showed significant differences with respect to markers of inflammation and atopy, nasopharyngeal microbiota, and respiratory outcomes by 3 years of age [19]. To our knowledge, no previous studies have identified classes of children with bronchiolitis within the Italian population. Noteworthy, the bronchiolitis profiles identified in the current study by means of LCA share some similarities with those identified in the aforementioned studies using the same statistical approach.

We identified a group of children (Class 1) predominately composed of female subjects and characterized by early onset (< 6 months of age) of symptoms. In this group, bronchiolitis was mainly caused by RSV (77%); all patients had respiratory distress, and apnoea episodes were present in 11%. These children share similarities with Profile C children identified in the two US cohorts (MARC-30 USA and MARC-35), in which patients showed the lowest age at admission, had the most severe clinical presentation and were mostly infected by RSV (> 80%). Conversely, in the Finnish cohort (MARC-30 Finland), children with a more serious clinical presentation (Profile A) were mostly aged ≥12 months and had an RV infection in 54%. Interestingly, Profile C children in the MARC-35 cohort showed an augmented risk of recurrent wheezing by age 3 years. This result confirms that the severity of bronchiolitis may influence mid-term respiratory outcomes, increasing the risk of recurrent wheezing and asthma later in life [24‐26].

Class 3 included the largest proportion of male subjects (94%) and was mostly composed of children aged > 6 months at the time of admission (68%). 70% of children had cough, 12% showed respiratory distress and none presented with apnoea. Similarly to Profiles D identified in MARC-30 USA and MARC-30 Finland cohorts, this was the least severely ill group, and it differed from other classes by the lowest proportion of children with RSV infection (27%), which was in agreement with Profile A in the MARC-35 cohort. Of note, also these profiles were mainly composed of male subjects and, with the exception of the MARC-30 USA cohort, they were characterized by a higher proportion of children aged > 6 months.

With regard to the statistically different characteristics for the three classes identified in the current study, we found that children in Class 1 were more frequently born near the epidemic season. This is considered a risk factor both for the development of lower respiratory tract infections as well as for hospitalization in subjects with RSV infection, who likely develop more frequently severe symptoms if born during the first half of the epidemic season [27]. Together with the low age at the time of hospitalization (< 6 months) and the higher prevalence of infections by RSV, this may have contributed to the greater severity of clinical presentation observed in children in Class 1 compared to the other two classes. Indeed, evidence suggests that the RSV aetiology is associated with greater severity of the clinical presentation [28‐30] and that a lower age at admission is associated with greater clinical severity of the RSV infection [28, 31].

As a piece of further evidence derived by the comparison of the three classes identified in the current study, duration of exclusive breastfeeding was significantly longer in children in Class 3, which was characterized by the least clinical severity of the disease. Notably, the differences among LCA classes in breastfeeding duration are due to the highest percentage in Class 3 of children who were exclusively breastfed. Indeed, when children who were not exclusively breastfed are excluded similar duration of breastfeeding are found. Thus, duration of breastfeeding can be considered as a proxy of exclusive breastfeeding. Even though the length of exclusive breastfeeding observed in this study was on average lower than 6 months or even 4 months, our finding provides further evidence for a protective role of breastfeeding against the severity of bronchiolitis, probably due to immunomodulatory factors in human milk against the RSV, as suggested by previous studies [32, 33].

The application of a statistical methodology such as LCA allowed us to detect bronchiolitis profiles different in gender, age at the onset of symptoms, severity of the clinical picture. These profiles were obtained directly from the data collected within a heterogeneous sample of children rather than being arbitrarily assigned. The resulting classes show a certain degree of flexibility: each individual can be assigned to different classes with different probabilities and this classification reflects clinical reality more closely, emphasizing the effectiveness of a multidimensional approach in capturing the clinical heterogeneity of the disease. A possible drawback is given by the need to establish a priori the set of variables, which requires a certain degree of subjectivity in data analysis. A further limitation of our study is related to the identification of disease profiles in a cohort of hospitalized children, which are actually a small proportion of all children with bronchiolitis. For a better characterization of the disease, it could be useful to apply the same statistical approach also to children with a milder level of disease, not requiring hospitalization, as they could show different socio-demographic characteristics and risk factors, as well as different clinical characteristics, compared to patients with most severe cases. However, our study allowed us to highlight that even within the group of children with bronchiolitis who require hospitalization there is a certain level of heterogeneity that could influence the diagnostic and therapeutic approach. Furthermore, the choice to study this cohort of children is significant, given that bronchiolitis is the most frequent cause of hospitalization in infants [34] and that patients with more severe disease have a higher risk of developing respiratory sequelae later in life. Lastly, it should be acknowledged as a limitation that viruses could not be identified in more than 50% of the enrolled patients, thus our results should be interpreted with caution with respect to viral aetiology. Indeed, our virus detection rate was poor, likely due to the different diagnostic tests used (immunofluorescence/enzyme immunoassays vs. PCR) as well as to technical problems in collecting and storing biological samples. Nonetheless, our detection rate is quite similar to those reported in other epidemiologic studies in Italy [35, 36].

Describing profiles of a heterogeneous disease such as bronchiolitis could be useful in providing the appropriate clinical approach and in identifying the best available therapeutic choice. The current findings may help to increase the understanding of the phenotypic variability that typically characterizes bronchiolitis, although they need validation in further studies. Namely, prospective longitudinal studies are needed to define bronchiolitis phenotypes and endotypes through an integrated clinical, epidemiological and molecular approach in order to develop personalized clinical management and clarify the physiopathologic basis of mid- and long-term respiratory outcomes [6].