Article Text
Abstract
Introduction In the UK, asthma deaths are at their highest level this century (1). Increased recognition of at-risk patients is needed. This study phenotyped frequent asthma exacerbators, and used machine learning to predict frequent exacerbators. We hypothesised that frequent exacerbators would have more severe peripheral eosinophilia than infrequent exacerbators.
Methods Patients admitted to Watford General Hospital with an asthma exacerbation between 1st March 2018 – 1st March 2020 were included. Patient data was retrospectively collected from hospital and primary care records. Patients were organised into two groups: “Infrequent Exacerbator” (1 admission in the previous 12 months) and “Frequent Exacerbator” (≥2 admissions in the previous 12 months). Good adherence to inhaled corticosteroids (ICS) was defined as medication possession ratio (MPR) ≥0.8; poor adherence was defined as MPR ≤0.5. Machine learning models were used to predict frequent exacerbators.
Results 200 patients admitted for asthma exacerbations were randomly selected (73% female; mean age 47.8 ± 19.3 years; table 1). Peripheral eosinophilia was uncommon in either group (19% vs 21%). More frequent exacerbators were being treated with high-dose ICS (46.5% vs 23.2%; P < 0.001), and frequent exacerbators used more SABA inhalers (10.9 vs 7.40; P = 0.01) in the year preceding the current admission. Good adherence to ICS was similarly low between both groups (40.0% vs 48.3%). BMI was raised in both groups (34.2 vs 30.9). Logistic regression classifier was the most accurate machine learning model for predicting frequent exacerbators (AUC = 0.80).
Conclusions Patients admitted for asthma are predominately female, obese and non-eosinophilic. Patients who require multiple admissions per year have poorer asthma control at baseline. Machine learning algorithms can predict frequent exacerbators using clinical data available in primary care. Instead of simply increasing the dose of corticosteroids, multidisciplinary management targeting Th2-low inflammation should be considered for these patients, including weight loss regimens, reflux management and macrolide therapy.