introduction
Asthma is a heterogeneous disease, and a clear-cut characterization of
its various phenotypes has historically remained daunting for both
clinical practice and research purposes 1-3.
Conventionally, asthma phenotypes have been defined based on timing of
onset, atopic origin, eosinophilic inflammation, and presence of
obesity, to name a few 4. Such phenotypic
characterization has been described as primarily based on clinical
insights and experiences of the attending clinician. However, it has
been suggested that such asthma phenotyping is largely subjective as the
classification may vary from clinician to clinician4,5. Additionally, asthma phenotyping has mostly been
attempted in selected cohorts, example hospital-based asthma patients or
those with severe asthma, with less data from population-representative
samples.
The advancements being made by computational science at elucidating
biological processes have been welcomed in the field of asthma,
particularly in identifying asthma phenotypes 4-6. In
this context, various features of asthma are inputted into algorithms
that learn from unlabelled data, with less artefact bias, to produce
meaningful asthma phenotypes. This data-driven approach is believed to
be more objective and can, with relevant clinical inputs, produce
phenotypes that are clinically meaningful 4,5,7.
Characterizing asthma at a more granular level is in parallel with
efforts towards precision medicine, subsequently enabling prevention and
optimal, tailored management 8.
In this work, by including a broad range of clinical, biological, and
epidemiological parameters that are relevant to asthma, we employed a
novel machine learning approach to identify and describe asthma
phenotypes in an adult representative sample in western Sweden.