Diagnosis of Asthma Based on Routine Blood Biomarkers Using Machine Learning

Abstract

Intelligent medical diagnosis has become common in the era of big data, although this technique has been applied to asthma only in limited contexts. Using routine blood biomarkers to identify asthma patients would make clinical diagnosis easier to implement and would enhance research of key asthma variables through data mining techniques. We used routine blood data from healthy individuals to construct a Mahalanobis space (MS). Then, we calculated Mahalanobis distances of the training routine blood data from 355 asthma patients and 1,480 healthy individuals to ensure the efficiency of MS. Orthogonal arrays and signal-to-noise ratios were used to optimize blood biomarker variables. Receiver operating characteristic (ROC) curve was used to determine the threshold value. Ultimately, we validated the system on 182 individuals based on the threshold value. Out of 35 patients with asthma, MTS correctly classified 94.15% of patients. In addition, 97.20% of 147 healthy individuals were correctly classified. The system isolated 7 routine blood biomarkers. Among these biomarkers, platelet distribution width, mean platelet volume, white blood cell count, eosinophil count, and lymphocyte ratio performed well in asthma diagnosis. In brief, MTS shows promise as an accurate method to identify asthma patients based on 7 vital blood biomarker variables and threshold determined by the ROC curve, thus offering the potential to simplify diagnostic complexity and optimize clinical efficiency.