Missing information imputation for disease-dedicated social networks with heterogeneous auxiliary data. Issue 2 (2nd April 2020)

• Record Type: Journal Article
• Authors: Liu, Xu; He, Jingrui; Min, Wanli; Yang, Hongxia
• Is Part Of: IISE transactions on healthcare systems engineering. Volume 10:Issue 2(2020)
• Page Start: 87
• Page End: 98
• Publication Date: 2020-04-02
• Subjects: Disease-dedicated social network -- heterogeneous learning -- missing value imputation; Biomedical engineering -- Periodicals; Medical informatics -- Periodicals; Medical care -- Periodicals; 610.28
• DOI: 10.1080/24725579.2020.1716115
• Languages: English
• ISSNs: 2472-5579

Abstract

Abstract: Many high impact applications suffer from missing information. For example, disease-dedicated social networks provide additional resources to glimpse into patients' daily life related to disease management. However, due to the voluntary nature of such social networks, the information reported by patients is often incomplete, making the following data analytics tasks particularly challenging. On the other hand, in addition to the target data that we aim to analyze, we may also have other related data at our disposal. For example, to analyze disease-dedicated social networks, auxiliary clinical data (with potentially non-overlapping patients), as well as the users' online social relationship might provide additional information for estimating the missing information. Therefore, the key question we aim to answer in this paper is how we can leverage the heterogeneous auxiliary data for the sake of missing information imputation. To answer this question, we focus on diabetes-dedicated social networks, and we aim to estimate the missing information from patients' self-reported biomarker measurements. In particular, we propose a hypergraph structure to model the relationship among users and user-generated content (posts). Based on the hypergraph structure, we further introduce an optimization framework to estimate the missing biomarker measurements using heterogeneous auxiliary data. To solve the optimization framework, we design iterative algorithms to find the local optimal solution. Experimental results on both synthetic and real data sets (including a data set collected from a diabetes-dedicated social network) demonstrate the effectiveness of the proposed algorithms.