Discussion

By including a large sample of asthma subjects with ever reported or physician-diagnosed asthma and a comprehensive set of parameters, covering demographic/risk factors/triggers, clinical, and pathophysiological aspects into a novel machine learning algorithm, we could derive a four-cluster solution that captured clinically meaningful asthma phenotypes. The derived asthma phenotypes could be distinguished based on age at asthma onset, ranging from those with onset in childhood to those with onset in adulthood. They could also be distinguished on the basis of level of severity, ranging from the childhood atopic mild asthma to the late adulthood more troublesome asthma. The phenotypes could also be differentiated on the basis of several demographic/risk factors/triggers, clinical aspects, symptom profiles, and various measures of inflammation.
WSAS is representative of the adult population of western Sweden; as such our findings have a reliable generalizability to the underlying target population. The population-based sampling also constitutes an advantage over some previous studies that have relied primarily on hospital-based setting in their phenotyping 7,17-19 . With a population-based sample, the whole spectrum of asthma severity can be captured. We selected a comprehensive set of variables for the phenotyping exercise, which ensured that multiple dimensions of asthma were captured, providing an advantage over approaches that utilize fewer sets of variables or that focus primarily on clinical variables. We employed a novel and robust machine learning approach, deep embedded learning, which is particularly adept at managing complex, multi-dimensional data, offering an advantage over conventional clustering approaches. The selection and description of the derived phenotypes represent a hybrid of data science and clinical experience, ensuring that the phenotypes accurately align to both clinical and statistical expectations. Our study, however, may be limited by the absence of certain inflammatory markers, like sputum measures, which are valuable in defining asthma endotypes. In addition, the lack of certain co-morbidities could also mean potential aspects of phenotype characterization were not captured. Nevertheless, the alignment of our findings to previous studies indicates that our approaches were largely valid and reliable.
The first phenotype in our work (Cluster 1) was characterized by substantially older age, late onset and troublesome asthma with increased smoking, which had high symptom and health care use burden, compared to other clusters. It also has a higher proportion of patients that can be classified as having severe asthma based on medication usage. This phenotype overlaps with findings from previous studies among adults 20-25. For instance, a similar phenotype derived by Kaneko et al.22 carried same characteristics as our first phenotype. Kim et al.23also reported a phenotype with high airway obstruction, non-atopy, and older age. The phenotype derived by Loureiro et al. 26was similar to ours by being late onset and severe, uncontrolled asthma, dominated by obese women, and had high eosinophil, neutrophil and monocyte counts. Different characteristics of this phenotype have been described in a similar phenotype derived by other studies, including systemic inflammation 27. late-onset and severe asthma28, high comorbidity burden29, need for more medication 30, 31, and increased cigarette smoking30,32 .
Our second phenotype (Cluster 2) that was characterized by female dominance and early adult-onset asthma with high breathlessness and moderate symptoms, nearly normal lung function, and moderate healthcare use also closely aligns with findings from previous studies22,26,29,33-35. This phenotype closely mirrors the phenotype identified by Dudchenko et al.36, which was notably sensitive to weather as a trigger, while our phenotype had exercise and infections as important triggers. Ilmarinen et al.32 reported a similar phenotype that was described as ‘female asthma’ and marked by near normal lung function but being moderately symptomatic and using health care services. A similar phenotype was described by Kim et al.23 as early adulthood-onset, mild, female asthma, featuring persistent normal lung function and a gentle disease progression in young women.
The phenotype of adult-onset asthma with high inflammation (Cluster 3) also aligns with a phenotype found in previous studies20,28,33,37-39. For example, Bochnek and colleagues20 reported a moderate asthma phenotype with elevated eosinophil levels. However, their study did not address the age of asthma onset. Boudir and colleagues also identified a moderate asthma phenotype characterized by significant bronchodilator reversibility and pronounced respiratory symptoms with a high rate of atopy, consistent with our results. Hsaio and colleagues 33 also described a similar phenotype to our findings, which was further distinguished by a history of smoking.
Similarly, our fourth phenotype of early-onset, mild asthma with atopy (Cluster 4) was frequently reported in previous studies22,23,32,36,37,39-42. In addition to overlapping characteristics of mild disease course, good control status, high atopy, and relatively early onset, Dudchenko et al.36reported high impairment on physical activity that additionally characterize this phenotype. Two studies additionally reported younger age of subjects belonging to this phenotype, which is in line with our observation of young mean age among members of this phenotype. Loza et al.39 additionally reported this phenotype to be associated with low inflammation of high T2 cell pattern.
The first phenotype (Cluster 1), characterized by late onset troublesome asthma, with older age, high rate of smoking, COPD as a comorbidity, and reduced diffusion capacity, may point to presence of emphysematous changes. Clinically, this phenotype may present asthma and COPD co-existing in the same patient 32,43-45. Additionally, compared to the other phenotypes, with high BMI, high proportion of females, more healthcare use, hospital emergencies, and systemic inflammation, this phenotype could also be reflecting the group of severe female obesity-related asthma with mixed inflammation patterns that have been reportedly associated with severe presentation at late age 26,32,33. The presentation of more comorbidities amongst such a group of asthma patients also aligns with the greatest impairment to quality of life that had been associated with such phenotype previously 26.
The second phenotype (Cluster 2) that constituted a group of women with moderate asthma seemed to have better overall health because they had fewer other health problems, had low smoking rates, and generally demonstrated good lung function. This group also showed relatively moderate asthma symptoms, which might be because women tend to notice their symptoms more and seek medical help sooner 46, which is demonstrated by high utilization of emergency service among this group compared to others. Further, low smoking history, low BMI, low count of comorbidities may have influenced the good overall prognosis of this female cluster. Additionally, these women were particularly good at noticing what triggered their asthma, like changes in weather or infections, which may help them avoid these triggers and have fewer symptoms.
The asthma phenotype that typically begins in early adulthood (Cluster 3) was characterized by significant inflammation and moderate symptom severity, with a prominent feature being an elevated FeNO and eosinophil count. These are associated with type 2 immune response39,47. Such group with high eosinophiles and FeNo levels may represent a sensitive treatment group with ICS therapy, however they may be undertreated. Additionally, this phenotype tends to have greater exposure to smoking, which has been linked to increased eosinophilic inflammation 48. This phenotype also exhibited a high occurrence of rhinitis and allergic conditions, such as chronic nasal problems accompanying asthma. Nonetheless, the observation that this group reported the fewest symptoms of drug-induced asthma contradicts this hypothesis.26
The early onset, mild atopic asthma phenotype (Cluster 4) possibly represents the traditional childhood-onset asthma characterized by high allergic sensitization, better asthma control, and low symptom burden. Childhood onset-asthma has greater propensity for remission than asthma starting in adulthood 49. Our data suggests that this phenotype also may have the highest rate of remission as it had the lowest proportion of those who have current asthma as defined by recent symptoms and medication use, indicating that although members of this phenotype developed asthma during childhood, some of them might be transient in part of patients in this cluster.