Using modern information theory to classify and identify high-risk disease groups is one of the research concerns in medical decision-making. The early diagnosis of gout is missing a single indicator, and relying on artificial labeling of disease characteristics is not only costly for decision-making, but also has a high misdiagnosis rate. Aiming at incomplete and attribute-related random large sample data, we propose a three-way clustering algorithm based on neighborhood rough sets, which is used to initially label the data, reduce the rate of misdiagnosis, and improve decision-making efficiency. Firstly, a neighborhood rough set theory in a heterogeneous information system is established. Secondly, the Best-Worst method-based neighborhood rough set attribute reduction model considering attribute correlation is constructed. Thirdly, a neighborhood rough set-based three-way clustering method for heterogeneous information system is proposed. Finally, we use 2,683 random samples and the proposed model to identify and classify potential gout patients in the samples. The results show that the proposed model can be used to mark and cluster potential gout groups in random samples without prior probability and with fuzzy decision rules, which is helpful for clinical decision-making. (C) 2020 Elsevier Inc. All rights reserved.