A hypergraph convolution-based intelligent healthcare platform for aging population management

Wenjie Li, Bing Hou, Xiaoxiao WangFrontiers in Public Health 2025 · DOI

Abstract: The proposed platform significantly improves personalized care for older adults, enhances the efficiency of healthcare resource allocation, and offers a scalable solution for intelligent healthcare services.

Introduction

As the global population ages, healthcare systems face an unprecedented burden from chronic diseases and comorbidities. Traditional healthcare models are predominantly reactive and siloed, often failing to capture the complex interdependencies of patient health. While Graph Convolutional Networks (GCNs) have emerged as a powerful tool for modeling patient data, they are fundamentally limited to "pairwise" relationships—connecting only two data points at a time.

In clinical reality, health is a higher-order problem where a patient’s risk is determined by the simultaneous interplay of physiological, behavioral, and environmental factors. This paper introduces an intelligent healthcare platform utilizing Hypergraph Convolutional Networks (HGCN) to capture these complex associations, achieving high accuracy in disease risk prediction while maintaining robust performance under heavy user loads.

Platform Architecture

The proposed system is designed as a full-stack architecture that bridges the gap between wearable IoT devices and cloud-based management terminals. The core of the system is a virtual community-based platform that integrates medical and aged care services through several specialized subsystems, including rehabilitation care, health information management, and intelligent support.

Figure 1: Architecture diagram of the intelligent healthcare management platform.

As shown in the architecture diagram, the platform facilitates a continuous loop between needs research, service implementation, and evaluation. This structure ensures that the intelligent healthcare management platform remains responsive to the specific requirements of the aging population it serves.

Technical Methodology: Hypergraph Convolution

Hypergraph Construction and Adaptive Learning

Unlike standard graphs, a hypergraph utilizes hyperedges that can connect any number of nodes simultaneously. This allows the model to represent complex group associations, such as a "syndrome" involving multiple physiological markers. To address the common issue of "oversmoothing" in deep learning—where critical individual details like ECG anomalies are lost during data aggregation—the authors implemented a "rooted hierarchical" approach. This uses a learnable diagonal matrix to assign weights to low-level raw signals, ensuring they are preserved even as the model extracts high-level patterns.

Multimodal Data Fusion

The platform integrates three distinct data streams to provide a holistic view of patient health:

  • Physiological: Real-time data from devices like ECG stickers, oximeters, and blood pressure meters.
  • Behavioral: Daily activity metrics, sleep quality, and medication adherence.
  • Environmental: Ambient conditions such as temperature and humidity.
Figure 2: Technology roadmap detailing data integration and the HGCN-based management platform.

The technology roadmap illustrates how these multi-modal, multi-parameter data streams are processed through quality control and privacy protection layers before entering the HGCN-based analytical model to generate personal health portraits and risk assessments.

Experimental Validation

The researchers validated the platform using a combination of large-scale public benchmarks and a real-world pilot study:

  • UCI Machine Learning Repository: Over 100,000 patient records from Heart Disease and Diabetes datasets.
  • PhysioNet: High-fidelity physiological signals from the MIT-BIH Arrhythmia Database.
  • Pilot Study: A two-month study involving 50 participants (aged 65–85) at an elder care facility, utilizing IoT sensors and behavioral logs.

Key Results and Performance

Accuracy and Reliability

The HGCN model achieved a peak accuracy of 89.58% on the UCI dataset and 87.26% in the real-world pilot, consistently outperforming traditional CNN and GCN-based methods. Ablation studies highlighted that while physiological data is the most critical (causing an 8.81% drop in accuracy if removed), the highest predictive performance is only reached when behavioral and environmental contexts are included.

System Scalability

In stress tests, the platform demonstrated remarkable stability, maintaining a 100% request success rate with up to 480 concurrent users. Even under high load, the average transaction execution time remained approximately 0.072 seconds, indicating the platform's readiness for real-time clinical deployment.

Discussion and Future Directions

While the platform offers a sophisticated solution for aging population management, several areas for improvement remain. The current "black box" nature of the AI model necessitates the integration of Explainable AI (XAI) techniques to provide clinicians with the reasoning behind high-risk flags. Furthermore, future iterations will focus on scaling the participant base and enhancing interoperability with existing Electronic Health Record (EHR) systems using standards like HL7 FHIR to ensure seamless integration into the broader healthcare ecosystem.

Summary updated

Abstract

The proposed platform significantly improves personalized care for older adults, enhances the efficiency of healthcare resource allocation, and offers a scalable solution for intelligent healthcare services.
BibTeX
@inproceedings{li2025hypergraph,
  author = {Wenjie Li and Bing Hou and Xiaoxiao Wang},
  title = {{A hypergraph convolution-based intelligent healthcare platform for aging population management}},
  booktitle = {Frontiers in Public Health},
  year = {2025},
  doi = {10.3389/fpubh.2025.1640808},
}
Figure