Diabetes Network Pharmacology Dataset

The Real-World Diabetes Network Pharmacology Dataset (RDNPD 2025) is a large-scale multimodal healthcare dataset developed for diabetes severity prediction, biomedical graph-learning analysis, and federated healthcare intelligence research. The dataset contains 466,605 patient records collected from five distributed healthcare institutions, including tertiary hospitals, diabetes treatment centers, and remote clinical facilities. The dataset was designed to support research on wearable-assisted healthcare monitoring, explainable artificial intelligence, biomedical interaction modeling, and privacy-preserving federated learning systems. The dataset integrates heterogeneous healthcare information obtained from wearable physiological sensing devices, electronic health records, laboratory investigations, medication histories, and biomedical network pharmacology resources. The included demographic and lifestyle attributes consist of age, gender, body mass index, waist circumference, smoking status, and physical activity level. Physiological and diabetes-related biomarkers include HbA1c, fasting glucose, postprandial glucose, insulin level, HOMA-IR, C-peptide, glycemic variability, blood pressure measurements, heart rate, cholesterol indicators, triglycerides, and cardiovascular risk measurements. The dataset further incorporates organ-damage and complication-related indicators including eGFR, creatinine, albuminuria, liver enzyme measurements, neuropathy score, retinopathy score, foot ulcer risk, and cardiovascular risk index. Medication and treatment-related attributes include metformin utilization, insulin therapy, SGLT2 inhibitor usage, GLP1 receptor agonist usage, medication adherence, treatment duration, and medication count. To support biomedical graph intelligence and network pharmacology research, the dataset additionally contains graph-derived healthcare interaction descriptors including drug--target affinity score, pathway enrichment score, inflammation pathway index, oxidative stress score, cytokine activity score, gene--disease association score, protein--protein interaction degree, graph node degree, clustering coefficient, graph attention centrality, patient--drug edge count, drug--protein edge count, and pathway connectivity measurements. The target label of the dataset is Diabetes_Severity_Level}, which consists of five clinically relevant classes including Healthy/Normal, Prediabetes, Mild Diabetes, Moderate Diabetes, and Severe/Complicated Diabetes.