Use of Multiple Fluid Biomarkers for Predicting the Co-occurrence of Diabetes and Hypertension Using Machine Learning Approaches

The co-existence of diabetes and hypertension can complicate and affect the management of these diseases. The early detection of these comorbidities can help in developing personalized preventive treatments and thereby, reduce the healthcare burden. The inclusion of readily available fluid biomarkers from different body fluids can be used as diagnostic tools and can facilitate in the designing of treatment strategies. In this work, an attempt has been made using multiple fluid biomarkers to differentiate diabetic from diabetic and hypertensive comorbid (DHC) condition. The fluid biomarkers are obtained from a publicly available dataset for diabetic (N=105) and DHC (N=57) conditions. The features, such as systolic blood pressure, fasting blood glucose, diastolic blood pressure, and total cholesterol are extracted and statistically analyzed. Data balancing technique namely synthetic minority oversampling technique is applied on the minority class to balance the dataset. Machine learning techniques namely, linear discriminant analysis, random forest, K-nearest neighbor, and linear support vector machine are used to perform the classification between the two groups. The results show that systolic blood pressure, diastolic blood pressure, and total cholesterol are elevated in the comorbid condition. These features also exhibit a statistical significance (p<0.001) between the two groups. This study also addresses the data imbalance issue, which is resolved by using an oversampling technique to mitigate the bias resulting from imbalanced data. The LDA classifier achieves a maximum accuracy of 61.2% in distinguishing between the two conditions. Machine learning based approaches may help in the prediction of comorbid conditions. This can act as a guideline for future studies on the progression of diseases and the identification of fluid biomarkers.