In preventive healthcare, it remains a critical challenge to detect diabetes earlier. Delays in the diagnosis may cause other health issues such as neuropathy, cardiovascular disease and renal dysfunction. By the quality and diversity of the available data, Machine Learning (ML) and Deep Learning (DL) methods are often limited to detecting diabetes. This study introduces a unified approach to integrate two datasets, utilizing robust preprocessing to handle missing values and class imbalance. In our framework, more accurate and generalizable diabetes prediction is provided by constructing a comprehensive patient population for analysis. K-Nearest Neighbours (KNN) is employed to construct a patient similarity graph, where each patient is represented as a node and edges are formed based on feature-space closeness, linking each patient to their k most similar neighbours. This graph structure serves as the input to the Graph Convolutional Network (GCN) for downstream prediction tasks. This enables the model to learn from the individual patient and the relational structure among the patients’ attributes. Across diverse populations, this approach enhances the generalizability of diabetes prediction while improving model interpretability. Our model achieved its best performance on the hold-out test set, attaining an accuracy of 90.66%, specificity of 91.56%, ROC-AUC of 0.955, and recall of 81.18%. Integration of multi-source data with a graph-based method that highlights the potential of geometric deep learning and dataset fusion in healthcare Artificial Intelligence (AI). This work is helpful in future studies on relationship-based modelling to detect chronic diseases like diabetes.
Ahmad et al. (Thu,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: