Rapid machine learning model for differentiating asthma and chronic obstructive pulmonary disease using age and blood parameters

• Published in: International journal of medical informatics (Shannon, Ireland) Vol. 205; p. 106114
• Main Authors: Yu, Lin, Hu, Zhaoming, Lin, Zhiwei, Liu, Zhe, Chen, Wenxiang, Wang, Ying, Guan, Chushi, Li, Shanshan, Liu, Shuang, Zhang, Ruixue, Yang, Qianyue, Zhang, Xueyan, Sun, Baoqing
• Format: Journal Article
• Language: English
• Published: Ireland Elsevier B.V 01.01.2026
• Subjects: Adult; Age Factors; Aged; Aged, 80 and over; Algorithms; Asthma; Asthma - blood; Asthma - diagnosis; Blood parameters; Chronic obstructive pulmonary disease; Diagnosis, Differential; Differential diagnosis; Female; Humans; Machine Learning; Male; Middle Aged; Pulmonary Disease, Chronic Obstructive - blood; Pulmonary Disease, Chronic Obstructive - diagnosis; LYMPH; PCT; MCV; P-LCR; BASO; EPO; MONO; RDW-CV; Blood parameters; SVM; AUC; AECOPD; RBC; MPV; NLR; XGBoost; LASSO; PLT; WBC; Chronic obstructive pulmonary disease; Differential diagnosis; ML; HGB; PDW; SHAP; DCA; LightGBM; ROC; AI; EO; RDW-SD; Asthma; FVC; NEUT; MCH; MCHC; Machine learning; HCT; FEV1; COPD
• ISSN: 1386-5056; 1872-8243; 1872-8243
• DOI: 10.1016/j.ijmedinf.2025.106114

Differentiating COPD from asthma in emergency settings is clinically challenging due to overlapping symptoms. Early and accurate diagnosis is critical for initiating appropriate treatment. Traditional spirometry is often impractical in urgent care due to limited cooperation and resources. This study aims to develop a rapid, accessible diagnostic model using routine blood tests to support timely decision-making. We analyzed clinical and laboratory data from 9,038 patients (aged 40–98 years) diagnosed with COPD or asthma at the Datansha Branch of the First Affiliated Hospital of Guangzhou Medical University (2022–2024). The dataset was split into training (6326 cases, 70 %) and internal validation (2712 cases, 30 %) cohorts, with five-fold cross-validation. LASSO regression selected key variables, and the AdaBoostClassifier algorithm built the model. External validation used 2,741 patients from the Yanjiang Branch. Performance was assessed via sensitivity, specificity, accuracy, and area under the curve (AUC). The model identified seven predictors: mean corpuscular hemoglobin concentration (MCHC), age, lymphocyte percentage (LYMPH), hemoglobin content (HGB), plateletcrit (PCT), monocyte percentage (MONO), and eosinophil percentage (EO). It achieved AUCs of 0.890 (training), 0.871 (internal validation), and 0.855 (external validation), with accuracies of 0.810, 0.804, and 0.792, respectively. Calibration and decision curve analyses confirmed strong predictive reliability and clinical utility. An online tool was developed for rapid application. This machine learning model, leveraging age and routine blood parameters, provides a convenient, accurate, and widely applicable solution for differentiating COPD from asthma. Its reliance on standard, rapidly available blood tests support early diagnosis in diverse healthcare settings, enhancing patient outcomes.