Introduction: While reliable in vitro testing for sensitivity to
common aeroallergens has been available for decades, if and how asthma
might be predictably expressed in people matched for comparable multiple
sensitizations is unknown. Objective: Our aim is to develop an
understanding of these relations, which are known as allergic
poly-sensitizations (APS) using a machine learning approach. We
performed an audit of adult urban patients with moderate to severe
asthma who presented to an urban outpatient pulmonary clinic. Methods:
We constructed a database of sensitizations to the 25 aeroallergens in
the zone 1 ImmunoCAP® assay. We used the Scikit-Learn® machine learning
library to perform model-based clustering to identify APS clusters.
Subsequently, clusters were compared for differences in clinical markers
of allergic asthma. Results: The database consisted of 509 patients.
Mixture modeling identified ten clusters of increasing APS of varying
size (n = 1 to 339). There were significant increases in mean serum
immunoglobulin E (p<.001), peripheral blood eosinophil count
(p<.001), and D LCO (p=.02) with increasing
APS. There was a significant decline in mean age at presentation
(p<.001), FEV 1/FVC (p=.01), and FEF
25-75 (p=.002), but not FEV 1 (p=.29),
nor RV/TLC (p=.14) with increasing APS by simple linear regression.
Finally, we identified two apparent divergent paths for the poly-atopic
march, one driven by perennial allergens and the other by seasonal
allergens. Conclusion: We provide the framework for a novel machine
learning understanding and approach to the classification of APS and its
heretofore under-appreciated potential influences on asthma cluster
analyses. To our knowledge, this represents the first attempt to
identify poly-sensitization patterns that have clinical implications.