Decision-making model for early diagnosis of congestive heart failure using rough set and decision tree approaches

Abstract

The accurate diagnosis of heart failure in emergency room patients is quite important, but can also be quite difficult due to our insufficient understanding of the characteristics of heart failure. The purpose of this study is to design a decision-making model that provides critical factors and knowledge associated with congestive heart failure (CHF) using an approach that makes use of rough sets (RSs) and decision trees. Among 72 laboratory findings, it was determined that two subsets (RBC, EOS, Protein, O2SAT, Pro BNP) in an RS-based model, and one subset (Gender, MCHC, Direct bilirubin, and Pro BNP) in a logistic regression (LR)-based model were indispensable factors for differentiating CHF patients from those with dyspnea, and the risk factor Pro BNP was particularly so. To demonstrate the usefulness of the proposed model, we compared the discriminatory power of decision-making models that utilize RS- and LR-based decision models by conducting 10-fold cross-validation. The experimental results showed that the RS-based decision-making model (accuracy: 97.5%, sensitivity: 97.2%, specificity: 97.7%, positive predictive value: 97.2%, negative predictive value: 97.7%, and area under ROC curve: 97.5%) consistently outperformed the LR-based decision-making model (accuracy: 88.7%, sensitivity: 90.1%, specificity: 87.5%, positive predictive value: 85.3%, negative predictive value: 91.7%, and area under ROC curve: 88.8%). In addition, a pairwise comparison of the ROC curves of the two models showed a statistically significant difference (p < 0.01; 95% CI: 2.63–14.6).