Related Experiment Video
Updated: Sep 8, 2025
A Reversible, Non-invasive Method for Airway Resistance Measurements and Bronchoalveolar Lavage Fluid Sampling in Mice
Published on: April 13, 2010
Innovative machine learning-based prediction of early airway hyperresponsiveness using baseline pulmonary function
Hua Yang1, Xingru Zhao2, Zhuochang Chen2
1Department of Cardiopulmonary Function, Henan Provincial People's Hospital, Zhengzhou University People's Hospital, Zhengzhou, China.
Background:
The Bronchial Provocation Test (BPT) is the gold standard for diagnosing airway hyperresponsiveness (AHR) in suspected asthma patients but is time-consuming and resource-intensive. This study explores the potential of baseline pulmonary function parameters, particularly small airway indices, in predicting AHR and develops a machine learning-based model to improve screening efficiency and reduce unnecessary BPT referrals.
Methods:
This retrospective study analyzed baseline pulmonary function data and BPT results from Henan Provincial People's Hospital (May to September 2024). Data were randomly split into training (69.8%, n = 289) and validation (30.2%, n = 125) groups using R software (Version 4.4.1). The Least Absolute Shrinkage and Selection Operator (LASSO) was applied to identify the most predictive variables, and 10-fold cross-validation was used to determine the optimal penalty parameter (λ = 0.023) to prevent overfitting. Model fit was evaluated using the Akaike Information Criterion (AIC), and a logistic regression model was constructed along with a nomogram.
Results:
The optimal model (Model C, AIC = 310.44) included FEV1/FVC%, MEF75%, PEF%, and MMEF75-25%, which demonstrated superior discriminative capacity in both the training (AUC = 0.790, cut-off = 0.354, 95% CI: 0.724-0.760) and validation cohorts (AUC = 0.756, cut-off = 0.404, 95% CI: 0.600-0.814). In the validation cohort, multidimensional validation through calibration plots showed a slope of 0.883. The Net Reclassification Improvement (NRI) for Model C compared to other models was 0.169 (vs. Model A), 0.144 (vs. Model B), and 0.158 (vs. Model D). The Integrated Discrimination Improvement (IDI) and Decision Curve Analysis (DCA) indicated that Model C provided superior predictive performance and a significantly higher net benefit compared to the extreme curves. For instance, the 10th randomly selected patient in the validation cohort showed an 89.80% probability of AHR diagnosis, with a well-fitting model.
Conclusion:
This study identifies MEF75%, MMEF75-25%, FEV1/FVC%, and PEF% as effective predictors of early airway hyperresponsiveness in suspected asthma patients. The machine learning-based predictive model demonstrates strong performance and clinical utility, offering potential as a visual tool for early detection and standardized treatment, thereby reducing the risk of symptom exacerbation, lung function decline, and airway remodeling.
