A nonadditive rough set model for long-term clinical efficacy evaluation of chronic diseases in real-world settings

Abstract

The pursuit of clinical effectiveness in real-world settings is at the core of clinical practice progression. In this study, we address a long-term clinical efficacy evaluation decision-making problem with temporal correlation hybrid attribute characteristics. To address this problem, we propose a novel approach that combines a temporal correlation feature rough set model with machine learning techniques and nonadditive measures. Our proposed approach involves several steps. First, over the framework of granular computing, we construct a temporal correlation hybrid information system, the gradient method is employed to characterize the temporal attributes and the similarity between objects is measured using cosine similarity. Second, based on the similarity of gradient and cosine, we construct a composite binary relation of temporal correlation hybrid information, enabling effective classification of this information. Third, we develop a rough set decision model based on the Choquet integral, which describes temporal correlation decision process. We provide the ranking results of decision schemes with temporal correlation features. To demonstrate the practical applications of our approach, we conduct empirical research using an unlabeled dataset consisting of 3094 patients with chronic renal failure (CRF) and 80,139 EHRs from various clinical encounters. These findings offer valuable support for clinical decision-making. Two main innovations are obtained from this study. First, it establishes general theoretical principles and decision-making methods for temporal correlation and hybrid rough sets. Second, it integrates data-driven clinical decision paradigms with traditional medical research paradigms, laying the groundwork for exploring the feasibility of data-driven clinical decision-making in the field.