Machine learning predicts exacerbations in adults with asthma

Key takeaways:

• The model was based on the records of 21,686 adults with asthma.
• The model had a 0.85 area under the curve, 0.7 sensitivity and 0.49 specificity.
• Blood eosinophils were the most important clinical factor.

Researchers designed a model that used machine learning to predict recurrent exacerbations among adults with asthma by analyzing real-world data, according to a letter published in Annals of Allergy, Asthma & Immunology.

The heterogeneity in the results may indicate different pathologic drivers in specific patient groups, Ji-Hyang Lee, MD, PhD, department of allergy and clinical immunology, Asan Medical Center, University of Ulsan College of Medicine, and colleagues wrote in the letter.

The researchers used the electronic health records of 21,686 adults with asthma from Dec. 1, 1994, to Feb. 28, 2019, at Asan Medical Center to build models using regression and tree-based machine learning to predict asthma exacerbations within a year of a 3-month index duration.

Next, the researchers identified 803 patients who experienced an exacerbation within 3 months of the first day they met the working definition of asthma at the hospital. Among these patients, 112 (13.9%; mean age, 56.3 years; 59.7% women) experienced recurring exacerbations within 12 months of the index period.

The least absolute shrinkage and selection operator prediction model obtained a 0.85 (95% CI, 0.81-0.89) area under the curve, a 0.7 (95% CI, 0.63-0.76) sensitivity and a 0.49 (95% CI, 0.31-0.66) specificity, which the researchers called the most important features for predicting exacerbations. Also, the model had a 0.89 (95% CI, 0.85-0.94) positive predicting value and a 0.2 (95% CI, 0.11-0.29) negative predicting value for exacerbations.

Blood eosinophils, hospital utilization and prebronchodilator FEV₁ percent predicted during the index duration were the most important clinical factors, the researchers said, as patients with more frequent hospital visits and lower lung function were more likely to experience exacerbations.

Factors that affected risks for exacerbations varied across subgroups based on sex, age, BMI and blood eosinophils, the researchers continued, although blood eosinophils were highly ranked as a positive predicting factor for both men and women.

The use of leukotriene receptor antagonists, lower lung function based on prebronchodilator FEV₁ and forced vital capacity (FVC), and lower BMI were the primary variables that affected exacerbations in men.

Hospital utilization and laboratory data including hemoglobulin, white blood cell count and platelet count were more closely associated with exacerbations than lung function among women.

The most significant factor for recurrent exacerbations among patients aged 65 years and older was hospitalization, followed by blood eosinophils, coexisting chronic sinusitis, hemoglobulin and platelet count. The number of hospital visits during the index period was positively related to recurrent exacerbations in these patients as well.

Lung function was the primary risk factor among patients aged younger than 65 years, the researchers continued, indicating that poorer lung function and higher numbers of hospital visits overall were associated with higher risks for exacerbations.

Prebronchodilator FEV₁ was the most significant feature among patients with BMI greater than 25 kg/m², followed by blood eosinophils, number of hospital visits and prebronchodilator FVC.

The number of hospital visits was the most significant feature among patients with BMI less than 25 kg/m², followed by blood eosinophils, the use of triple inhalers, BMI and prebronchodilator FVC. Patients using triple inhalers specifically were more likely to experience additional exacerbations, the researchers said.

An additional prediction model found that lung function, platelet count and blood eosinophils were important factors among patients with eosinophil counts of 300 µL or higher.

Hospital visits were the most important factor for patients under that threshold, followed by prebronchodilator FEV₁ and laboratory data including hemoglobulin, eosinophils and platelet count.

Considering the heterogeneity in the priorities of contributors to exacerbations, the researchers said, different key pathologic drivers may impact these exacerbations in specific patient groups.