Eosinophilia predicted small airway obstruction in patients with low sensitization

Eosinophilia was found to be a marker of airway obstruction irrespective of degree of airway sensitization in a cohort of patients with severe allergic asthma.

Using machine learning, researchers identified nine specific clusters of allergic polysensitization in adults with severe allergic asthma living in urban Philadelphia.

The largest cluster — Cluster 1 — represented 36% of cases in the cohort and was characterized having low sensitization (2 or fewer positive sensitizations) and low allergen-specific IgE.

Increasing eosinophilia in the Cluster 1 cohort was a significant predictor of small airway obstruction. While low allergen sensitization alone was not a predictor of airway obstruction in this cohort, lower allergen sensitization along with an eosinophil count of 300/μl or greater was predictive.

The machine learning analysis was presented by colleagues Granit Mavraj, MSc, and Brian Patchett, MSc, Drexel University College of Medicine, Philadelphia, in separate video presentations to the annual meeting of the American College of Allergy, Asthma & Immunology — ACAAI 2020 — held virtually Nov. 13-15.

Mavraj explained that while it is widely assumed a strong correlation exists between higher eosinophil count and greater level of allergic poly-sensitization, this has been difficult to show.

The researchers sought to determine if asthmatic clusters could be identified with machine learning to help address the question by processing large numbers of complex allergy reports containing multiple allergens and wide-ranging allergen-specific IgE levels.

The researchers developed a framework for understanding the clinical implications of allergic poly-sensitization using machine learning in large data sets that can be applied to other ImmunoCAP zones, food allergy, or other allergies measured in specific IgE.

“In the future, we aim to use machine learning to input environmental and other data to develop more advanced prospective clinical markers with the goal of creating more precise diagnostic and therapeutic approaches for the practicing allergist,” Patchett explained in his presentation.

Patchett said machine learning has the potential to simplify complex allergy reports with multiple allergens, each with a specific IgE and understand whether a large group with similar positives and sensitizations represent a cluster with defined clinical characteristics that could be useful in clinical practice.

Mavraj expanded on this idea in his presentation.

“With 25 allergens on each ImmunoCAP, and each allergen displaying a specific IgE that can range from 0.10 to more than 100.00 IU/ml there are 1090 possible allergen profiles for each asthmatic,” Mavraj said, adding that the study included data derived from profiles of 477 patients with severe asthma.

“You can imagine standard progression techniques are insufficient to adequately handle the enormity of these data structures,” Mavraj said. “Machine learning is, at its core, a form of regression analysis that has a variety of applications, including finding distinct clusters within a data structure.”

The complexity of machine learning ranges from simple linear regression to convoluted neural networks.

“A remarkable feature of the modern machine learning approach is the ability to handle large, high dimensional datasets – such is the case with allergy – and find underlying clusters within these datasets,” Mavraj said.

The researchers used the machine learning clustering algorithm Gaussian mixture modeling to identify 9 clusters representing a range of allergic poly-sensitizations among the 477 patients with severe asthma included in the analysis, with the lowest being Cluster 1 and the highest, Cluster 9.

A total of 171 of the 477 patients were in Cluster 1, which was marked by allergen sensitivity to 2 or less of the 25 specific sensitizations and a low allergen-specific IgE count on these positive sensitizations.

This cluster, with the lowest sensitization positivity, was further divided into three eosinophilia groups: <150/μL (low, n=99); ≥150 to <300/μL (medium, n=49) and ≥300/μL (high, n=23 patients).

The mean age among all three Cluster 1 groups ranged from 51 years to 57 years, and mean body mass index ranged from around 31 to 34. Smoking status and smoking histories were similar among the three Cluster 1 cohorts.

While smoking history, BMI and age did not appear to be clinically predictive of airway obstruction in the cohort with low allergen sensitization, worsening eosinophilia did appear to predict small airway obstruction.

“Surprisingly, we could not conclude that the correlation between higher eosinophilia and IgE levels was strong in our cohort,” Mavraj said. “The next step in our project is to apply machine learning to make better use of allergy reports along with other laboratory data to decode polysensitization cohorts for the practicing allergist.”

  1. Eosinophilia was found to be a marker of airway obstruction irrespective of degree of airway sensitization in a cohort of patients with severe allergic asthma.

  2. Using machine learning, researchers identified nine specific clusters of allergic polysensitization among 477 adults with severe allergic asthma living in urban Philadelphia.

Salynn Boyles, Contributing Writer, BreakingMED™

Researchers Granit Mavraj and Brian Patchett reported no disclosures relevant to this research.

Cat ID: 100

Topic ID: 80,100,100,328,192,63,925,327