Scholarlink

This study introduces an optimized joint clustering algorithm to identify hospital groupings based on disease-specific monthly admission patterns using data from the California State Inpatient Database (2008–2011). Unlike the previous two-step method that relied on a predefined disease similarity network, the proposed approach dynamically constructs a meta-disease super network while clustering, enabling simultaneous optimization of disease and hospital networks. The method models 145 disease-specific hospital networks across 152 hospitals, forming a multilayer Network of Networks structure. It uses low-rank approximations and regularization to capture both local (hospital-level) and global (disease-level) similarities. Evaluation on synthetic and real-world data shows improved clustering homogeneity (average of \(61.2\%, SD \pm 2.1\%\) ) over the prior method (average \(52.4\%, SD \pm 5.6\%\) ), with statistically significant gains (p = 0.0038). These clusters showed temporal stability and meaningful clinical groupings, aiding in referral coordination and resource allocation. The method is robust, interpretable, and extensible to other datasets and healthcare systems.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Discovering Linkages Among Multiple Disease Networks by Joint Clustering

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Nouf Albarakati,
  • Hussain Otudi,
  • Rafaa Aljurbua,
  • Zoran Obradovic

摘要

This study introduces an optimized joint clustering algorithm to identify hospital groupings based on disease-specific monthly admission patterns using data from the California State Inpatient Database (2008–2011). Unlike the previous two-step method that relied on a predefined disease similarity network, the proposed approach dynamically constructs a meta-disease super network while clustering, enabling simultaneous optimization of disease and hospital networks. The method models 145 disease-specific hospital networks across 152 hospitals, forming a multilayer Network of Networks structure. It uses low-rank approximations and regularization to capture both local (hospital-level) and global (disease-level) similarities. Evaluation on synthetic and real-world data shows improved clustering homogeneity (average of \(61.2\%, SD \pm 2.1\%\) ) over the prior method (average \(52.4\%, SD \pm 5.6\%\) ), with statistically significant gains (p = 0.0038). These clusters showed temporal stability and meaningful clinical groupings, aiding in referral coordination and resource allocation. The method is robust, interpretable, and extensible to other datasets and healthcare systems.