Methods Inf Med 2/2013 © Schattauer 2013 160 Focus Theme – Original Articles Health Webscience Keywords Diabetes, social networks, homophily, assor- tativity, community detection Summary Background: Detecting community struc- tures in complex networks is a problem inter- esting to several domains. In healthcare, dis- covering communities may enhance the quality of web offerings for people with chronic diseases. Understanding the social dynamics and community attachments is key to predicting and influencing interaction and information flow to the right patients. Objectives: The goal of the study is to em- pirically assess the extent to which we can infer meaningful community structures from implicit networks of peer interaction in on- line healthcare forums. Methods: We used datasets from five online diabetes forums to design networks based on peer-interactions. A quality function based on user interaction similarity was used to as- sess the quality of the discovered commu- nities to complement existing homophily measures. Results: Results show that we can infer meaningful communities by observing forum interactions. Closely similar users tended to co-appear in the top communities, suggest- ing the discovered communities are intuitive. The number of years since diagnosis was a significant factor for cohesiveness in some diabetes communities. Conclusion: Network analysis is a tool that can be useful in studying implicit networks that form in healthcare forums. Current analysis informs further work on predicting and influencing interaction, information flow and user interests that could be useful for personalizing medical social media. Correspondence to: Taridzo Chomutare Norwegian Center for Integrated Care and Telemedicine University Hospital of North Norway 9038 Tromsø Norway E-mail: taridzo.chomutare@telemed.no Methods Inf Med 2013; 52: 160–167 doi: 10.3414/ME12-02-0003 received: March 1, 2012 accepted:August 28, 2012 prepublished: February 8, 2013 Inferring Community Structure in Healthcare Forums An Empirical Study T. Chomutare 1 ; E. Årsand 1,3 ; L. Fernandez-Luque 2 ; J. Lauritzen 3 ; G. Hartvigsen 1 1 University hospital of North Norway, Norwegian Center for Integrated Care and Telemedicine, Tromsø, Norway; 2 Northern Research Institute, Tromsø, Norway; 3 University of Tromsø, Department of Computer Science, Tromsø, Norway 1. Introduction In this work, we examine whether interac- tions in healthcare forums can be repre- sented as networks. We examine diabetes networks because it is a disease case that involves several stages of illnesses, and is a heterogenic disease with many sub-factors. Diabetes is also a highly relevant disease to focus on from a patient healthcare and economic perspective. In 2003, 194 million people globally were estimated to have a form of diabetes, which is predicted to in- crease to 333 million in 2025, which consti- tute an increase of 72% [1]. The problem is seen not only in Europe and America, but on a global scale. WHO estimates that, 5 –10% of the national healthcare budget in western countries is used on diabetes, which will increase with the increasing number of diabetes patients globally. In some instances, patients must rely on peers for emotional, empathetic and practi- cal support. Internet forums are one of the most popular social media for self-help. A major distinguishing feature between inter- action in these forums (message boards) and interaction in other social media is that the forums normally do not have explicit rela- tionships, and unlike most social networking websites, relationships in forums are implied. These relationships are encoded in large da- tasets of forum threads-and-comments dy- namics, and in this work we use network analysis to decipher these relationships. Community detection [2] are a group of net- work analysis [3] methods that hold a poten- tial to reveal characteristics that help us identify important peers [4], predict com- munity attachments and influence informa- tion flows and temporal patterns [5, 6]. 1.1 Related Literature The idea of discovering communities from forum interactions is not new; researchers have long been fascinated by the prospect [7, 8]. Different methods have been dis- cussed extensively in the literature, but work by L’Huillier et al. [8] enhanced our understanding of how forum discussions can be analyzed and connected using net- work analysis and text mining. While the work focused on terrorism and a few hundred users, it nonetheless sheds some light. In healthcare, the review by Dunn and Westbrook [9] provides a complete synthesis to date of the relevant network analysis concepts for small scale health- care networks. Network analysis has been used extensively in bioinformatics [10] and For personal or educational use only. No other uses without permission. All rights reserved. Downloaded from www.methods-online.com on 2014-07-21 | IP: 85.137.218.210