A FRAMEWORK FOR THE ANALYSIS OF COMORBID CONDITIONS USING INTELLIGENT EXTRACTION OF MULTIPLE FLUID BIOMARKERS

Abstract

Fluid biomarkers extracted from many types of body fluids provide significant information that serve as indicators of the underlying physiological and pathological conditions of the human body. Analysis of multiple fluid biomarkers could help improve the early identification and progression of comorbid conditions to enhance the diagnostic accuracy, which can help in developing patient-specific treatment plans. In this work, an attempt has been made to differentiate the co-occurrence of diabetes, hypertension and cardiovascular disease (comorbid conditions) from non-comorbid using multiple fluid biomarkers. Fluid biomarkers are obtained from a public dataset under comorbid ([Formula: see text]) and non-comorbid ([Formula: see text]) conditions. Five features, such as serum creatinine, serum sodium, platelet count, creatine phosphokinase and ejection fraction, are extracted for further analysis. Machine learning algorithms namely, [Formula: see text]-nearest neighbor and linear discriminant analysis (LDA) are used to classify comorbid and non-comorbid conditions. The results show an increase in platelet count in comorbid subjects. This feature also exhibits significant difference ([Formula: see text]) between both the conditions. This study also uses the random undersampling technique to reduce bias associated with data imbalance. LDA classifier yields a maximum accuracy of 54.30% in classifying these two conditions. Further study can be carried out to improve the accuracy and might be helpful in clinical practice for prediction of comorbid conditions.