Machine Learning-Based Model for Predicting Severe Exacerbations in Adult-Onset Type 2 Inflammatory Asthma

Abstract

INTRODUCTION: Currently, scholars have applied machine learning to the clinical prediction of acute asthma exacerbations. However, given the heterogeneity of inflammatory phenotypes in asthma, it is imperative to develop machine learning models tailored to specific asthma inflammatory phenotypes. The aim of this study was to develop predictive models to identify risk factors for the severe exacerbations in adult-onset type 2 inflammatory asthma, which could help facilitate early diagnosis and intervention, potentially reducing healthcare costs. METHODS: This was a retrospective analysis of patients with acute exacerbations of type 2 inflammatory asthma at Shenzhen People's Hospital from May 2017 to September 2022. Patients were categorized into mild-to-moderate exacerbation (n = 300) and severe exacerbation groups (n = 209). We collected clinical data from all participants, including demographic characteristics, laboratory results, pulmonary function test results, comorbidities, and asthma medication use. We tested four models: decision trees, logistic regression, random forests, and LightGBM. For each model, 80% of the dataset was used for training and 20% was used to validate the models. The area under (AUC) the receiver-operating characteristic (ROC) curve was calculated for each model. RESULTS: Multivariate logistic regression revealed that low ACT scores, low FEV1/FVC ratio, a history of diabetes, high absolute neutrophil count, and a family history of asthma were independent risk factors for severe exacerbations of type 2 inflammatory asthma. LightGBM outperformed all other models, achieving the highest AUC of 0.9344, with sensitivity = 0.8293, specificity = 0.9180, PPV = 0.8718, and NPV = 0.8889. The accuracy stood at 0.8824, with an F1 score of 0.8500. The top 10 clinical variables impacting the prediction outcome in the LightGBM model were ACT score, FEV1/FVC ratio, age, lactate dehydrogenase, FEV1% predicted, fasting blood glucose, angiotensin-converting enzyme, duration of disease, neutrophil-to-lymphocyte ratio, and platelet-to-lymphocyte ratio. Finally, through DCA, the clinical decision-support value of the LightGBM model was confirmed, demonstrating its maximum net benefit for type 2 asthma patients across a threshold probability range of 20%-80%. CONCLUSIONS: We have developed and established a prediction model for severe exacerbations of adult-onset type 2 inflammatory asthma using the LightGBM machine learning approach, which exhibits good predictive performance. This model can aid in the early prediction and prevention of severe exacerbations of adult-onset type 2 inflammatory asthma.

Ähnliche Arbeiten