A Systematic Literature Review on Health ...

The 4th IEEE International Conference on E-Health and Bioengineering - EHB 2013 Grigore T. Popa University of Medicine and Pharmacy, Iaşi, Romania, November 21-23, 2013

A Systematic Literature Review on Health Recommender Systems Emre Sezgin

Sevgi Özkan

Informatics Institute Middle East Technical University Ankara, Turkey [email protected]

Informatics Institute Middle East Technical University Ankara, Turkey [email protected] On the other side, literature researches also showed that there is no trace or initiation about a review of studies and practices in HRS. It is an acceptable outcome in the field of HRS in which resources are limited. However, it is important to introduce a set of knowledge for researchers who are interested in HRS studies. Thus, here, this paper provides a preliminary literature review study in HRS domain. The literature was reviewed systematically and findings were presented considering the purpose of HRS and methods. In the following sub-sections, RS and HRS will be introduced. Then, methodology of the review, discussion of findings and conclusion sections were outlined.

Abstract—Health Information Systems are becoming an important platform for healthcare services. In this context, Health Recommender Systems (HRS) are presented as complementary tools in decision making processes in healthcare services. Health Recommender Systems increase usability of technologies and reduce information overload in processes. In this paper, a literature review was conducted by following a review procedure. Major approaches in HRS were outlined and findings were discussed. The paper presented current developments in the market, challenges and opportunities regarding to HRS and emerging approaches. It is believed that this study is an illuminating start-up point for HRS literature review. Keywords— Health Recommender Information Systems; Literature review

I.

Systems;

Health

INTRODUCTION

Today, information technologies have led to number of innovations and developments in number of fields. In this context, Recommender systems (RS) have been a cutting edge development in the service industry [1]. In the case of web-based services, RS aims to increase reachability of products and to provide alternatives for potential customers. Many variances of RS have been used in online stores (such as eBayTM and AmazonTM), and it is substantially being adapted by many organizations on the web. However, RS is not limited to marketing products online. On the other side, RS serve to decision support mechanism by providing options (substitutes) to decision makers [2, 3]. In health services, information systems have assisted to optimize decision making processes and to increase effectiveness of communication channels and infrastructures, such as ERP systems. In the health industry, RS has a significant role in terms of assisting decision making processes about individuals’ health. The studies demonstrated that RS have already been employed in health services, as Health Recommender Systems (HRS), for educational, dietary, and activity assistance purposes [4-7]. However, review study of Park et. al. [3] and literature research of scholar databases unveiled that relevant studies are rare in the field. The literature has presented several researches about RS being used in health information services.

A. Recommender Systems Due to “big data” piling up on the web, recommender systems have gained importance with respect to data cleansing and mining. In the early years of 90s, it was identified that information filtering techniques were needed in order to retrieve the information effectively. It was defined as “recommender systems form a specific type of information filtering (IF) technique that attempts to present information items (e.g. movies, music, books, news, images, web pages, etc.) that are likely of interest to the user”. In the literature, there are fundamentally three types of filtering in Recommender Systems [2]: • Collaborative Filtering: It is based on the knowledge collected and composed from users. Example: AmazonTM • Content-based Filtering: It is based on the knowledge aggregated from the users and unit descriptions of historical data. For example: Last.fmTM • Hybrid Filtering: It is a combination of different approaches and techniques, basically combining collaborative and content based filtering. The Appendix A exhibited the filtering types, their definitions, primary points of pros and cons and examples of Recommender Systems. In addition to 3 fundamental filtering types, two emerging types of filtering, knowledge-based and mobile RS, were also presented in the Appendix A. These filtering methods were identified as promising in HRS domain due to their incremental approaches and additional dataset involvement in recommending methods.

978-1-4799-2373-1/13/$31.00 ©2013 IEEE

B. Health Recommender Systems Health Recommender Systems are part of RS being applied in the health industry. It has been used for diagnostic assistance by physicians and for personal health advising tools by users [2]. As the communication platform, Internet has been the main source for users to access health information and recommendations. Fernandez et al. classified the health information being searched on the internet as following [1]: • Image, videos, web blogs, forums, tutorials, etc. • Publications by medical organizations, patients, governments, etc. • Multimedia resources on autopsies, recipes on herbal cures for cancer, etc. Thus, HRS have significant role in filtering information for self-diagnostic searches of users on the web as well as the given categories. In addition to that, HRS have been used by physicians for diagnostic and educational purposes. In this manner, suggesting online health resources (HealthyHarlem), cancer related web sites and educational resources with patient records (MyHealthEducator) can be given as examples for web based diagnostic recommender system use [1]. C. Issues of Recommender Systems Aside from common filtering problems (Sparsity, Cold start and Scalability problems), it is important to point out a major socio-technical issue about RS. Privacy is the major and emerging topic in this context. Using data from multiple sources may raise a question about use of individual private data. This issue was identified [9] as “the combination of weak ties (an unexpected connection that provides unexpected recommendations) and other data sources can be used to uncover identities of users in an anonymized dataset” Thus, it may present an important flaw especially in health information, which constitutes a delicate topic about privacy. II.

LITERATURE REVIEW METHODOLOGY

Since HRS are an emerging and a new field of research, the database research demonstrated that new studies were mostly presented on conferences, instead of top scientific journals. Many papers on the field can be found on journals of conference proceedings. Thus, in this study, research criteria were expanded considering keeping the quality of studies high but covering publications in conference proceedings. The research method was built upon a review protocol in order to review the literature systematically. In the study, Kithchenam’s systematic review procedure was employed [10]. The following steps were pursued: 1. Determining the topic of the research 2. Extraction of the studies from literature considering exclusion and inclusion criteria 3. Evaluation of the quality of the studies 4. Analysis of the data 5. Report of the results

Fig 1. Literature Review process

The process of literature review (Fig. 1) was started with research of leading academic databases (Sciencedirect, IEEE and Scopus) about HRS and its practice. The keywords were composed of “health”, “recommendation”, “system”, “recommender”, “eHealth”. Initial refinement was made considering 2 criteria: Publication year (within 10 years: 2002-2012) and quality of journals (by evaluating impact factor and citation rates). Then, the studies were retrieved and refined by title and abstract basis. In total, 310 papers were retrieved. It was refined to 251 papers by titlebasis elimination and to 35 papers by abstract-basis elimination. In the next phase, 8 papers were found meeting the following criteria of quality: reliability of the source, integrity in the content and providing applicable studies. In the final phase, findings were synthesized and reported. Method and Discussion sections of papers were the main focus area in analysis. In order to acquire information about aim and methodology of the studies, key elements in each section of papers were noted. Socio-technical aspects were held primary point of research rather than pure technical side of studies in order to provide a generic body of knowledge about all of the studies. There have been several limitations while conducting the study. First, the scarcity of academic resources in HRS was the main limitation. In addition to that, the studies were not explicitly providing details about their methods and techniques, and their variety of research approaches disallowed to make classification of HRS studies. Thus, they were presented “as it is” in the following section. III.

FINDINGS AND DISCUSSION

In total, 8 papers were found likely to contribute to HRS review. The list of papers, their aims and methods being employed were given in Appendix B. The review results generically presented that HRS were analyzed in terms of user groups and system design [5, 7, 8, 11], and a set of studies aimed to investigate physical activities and nutrition based recommender systems [4, 6, 12]. In addition to that, two of the studies aimed to outline challenges and opportunities [1, 12]. Electronic health records were also part of HRS in terms of health marketing, personal recommendations and self-examination [5, 7, 11]. Trending domain in HRS was pinpointed as telemedicine [8]. The current studies showed that telemedicine and diagnosis applications were main target of HRS studies in terms of managing health affairs for housebound and mobile patients. In HRS, it should be underlined that semantics is also a challenging topic which is important input in predicting user behavior on the web [13]. Thus, papers put emphasis

TABLE 1 CHALLENGES AND OPPORTUNITIES IN HRS Challenges

Opportunities

• HRS can be a target of cyberattacks due to its significance • Generally RS based on popularity of resources, thus it may be misleading in HRS • Web health applications have not been yet capable of integration in terms of data exchange • Data mining is an issue for user modeling which causes ethical issues in terms of races, gender and sexual orientations.

• HRS require less expertise to operate because of its autonomic operability and collaborativefiltering approaches • Integrating records of personal health data can enhance predictive power of HRS and may solve the cold start problem • Collaborative approach may enable gathering user preferences and knowledge in HRS, which are not very common but can be useful in health education • Autonomic structure can enhance recommendations in terms of consistency, and it improve knowledge gathered from the users

related to HRS use, design and methodologies. In this respect, it was found that emerging HRS studies were also in a rising trend in the literature. However, HRS domain is relatively new, thus it needs time to present mature researches and to improve filtering algorithms. In addition to that, privacy issues constitute a major concern to overcome. In this paper, it was aimed to contribute literature by (1) giving an opinion about the literature of HRS, (2) underlining the studies about HRS, and (3) providing a set of review methods for further studies in the field. It constituted a preliminary study, thus, further studies covering broader set of criteria, as well as academic journals, can be conducted as the next step in literature review of HRS. REFERENCES [1]

on semantics in development phase of HRS. In addition to that, Software-oriented Architecture (SOA) was commonly the basis for development approach [1, 8]. Its modularity and compatibility with web services made SOA favorable in HRS development. However, the main element of HRS lies behind the algorithms. It is crucial to develop the algorithm with high prediction rates in terms of user behavior [11]. From methodological side, it was observed that content based and collaborative based filtering were commonly used, however, hybrid RS and emerging methods in filtering

[2]

(profile-based) were also being applied in increasing manner. It was observed that filtering methods, SOA approaches and linguistic studies were combined in a platform of HRS to create algorithms. Thus, significance of algorithms is at the paramount in terms of health affairs. Considering the privacy concerns in RS as well as the delicacy of human health records, HRS algorithms were restricted by strict limitations in this context. On the other side, the challenges and opportunities of HRS were discussed in several studies [1, 2, 12]. The main challenge was outlined regarding to privacy, and the opportunity was its contribution in diagnostic process. Table 1 presented the challenges and opportunities underlined in the studies.

[6]

IV.

CONCLUSION

In this study, a literature review on Health Recommender Systems was conducted, and the findings were presented. The main conclusion is that HRS are a promising development for healthcare services. The studies demonstrated that HRS have been branched out in different fields of health industry, and HRS applications have been increasingly embedded in the health service systems. Considering the literature, there were number of studies

[3] [4]

[5]

[7] [8]

[9] [10] [11]

[12] [13]

[14]

L. Fernandez-luque, R. Karlsen and L.K. Vognild, "Challenges and Opportunities of using Recommender Systems for Personalized Health Education” Stud. Health Technol. Inform., 150(903), pp. 903-7, 2009. F. Ricci, L. Rokach, B. Shapira and P. B. Kantor, Introduction to Recommender Systems Handbook, pp. 257-297, Springer, Berlin, 2011. D. H. park, H. K. Kim, I. Y. Choi and J. K. Kim, “A literature review and classification of recommender systems research”, J. Expert Syst. Appl., 39 (11), pp.10059-10072, 2012. J. Kim, J. Lee, J. Park and Y. Lee, “Design of Diet Recommendation System for Healthcare Service Based on User Information”, Fourth International Conference on Computer Sciences and Convergence Information Technology, ICCIT '09, pp.516-518, 24-26 Nov 2009. P. Pattaraintakorn, G. Zaverucha and N. Cercone, “Web Based Health Recommender System Using Rough Sets, Survival Analysis and RuleBased Expert Systems”, Proceedings of the 11th International Conference on Rough Sets, Fuzzy Sets, Data Mining and Granular Computing, pp. 491-499,13-16 May 2007. A. Sami, P. Nagatomi, M. Terabe, and K. Hashimoto, “Design of Physical Activity Recommendation System”, Proceedings of IADIS International Conference on e-Learning, pp. 148-152, 22-25 July 2008. M. Wiesner, “Adapting recommender systems to the requirements of personal health record systems”, Proceedings of ACM International Health Informatics Syposium, pp. 410-414, 11-12 November 2010. C. Lee, M. Lee, and D.A. Han, “Framework for Personalized Healthcare Service Recommendation. Health”, Proceedings of 10th International Conference on e-health Networking, Applications and Services, pp. 90-95, 7-9 July 2008. N. Ramakrishnan, B. J. Keller, B. J. Mirza, A. Y. Grama and G. Karypis, “When being Weak is Brave: Privacy Issues in Recommender Systems”, IEEE Internet Comput., 5(6), pp. 54 - 62, 2001. B. Kitchenham, “Procedures for Performing Systematic Reviews”, Technical Report TR/SE-0401, Keele University, NICTA, 2004. M. Lopez-Nores, Y. Blanco-Fernandez, J.J. Pazos-Arias, J. GarciaDuque and M.I. Martin-Vicente, “Enhancing Recommender Systems with Access to Electronic Health Records and Groups of Interest in Social Networks”, Proceedings of 7th International Conference on Signal-Image Technology and Internet-Based Systems, pp. 105-110, 2830 Nov 2011. S. Mika, “Challenges for Nutrition Recommender Systems”. American Journal of Public Health, Proceedings of the 2nd Workshop on Context Aware Intelligent Assistance, pp.25-33, 4 October 2011. T.G. Morrell and L. Kerschberg, “Personal Health Explorer: A Semantic Health Recommendation System”, Proceedings of the 28th International Conference on Data Engineering Workshops, pp. 5559,1-5 April 2012. G. Adomavicius, and A. Tuzhilin, “Toward the next generation of recommender systems: A survey of the state-of-the-art and possible extensions”, IEEE T. Knowl. Data En., 17(6), pp. 734-749, 2005.

APPENDIX A. FILTERING TYPES, DEFINITIONS, PROS & CONS AND EXAMPLES [1, 2, 14] Type Collaborative filtering

Definition Gathering and analyzing information about activities and behaviors, of users and predicting what users is likely to do regarding to their similarity with other users Examining the historical data and current preferences of users and predicting based on characteristics of the items

Pros It is able to recommend complex items

Cons Sparsity, Cold start and scalability issues

Example systems Last.FmTM, amazonTM, facebookTM, myspaceTM, linkedinTM

Kick-off information is not required to be much

Scope of recommendation source is limited

Internet Movie DatabaseTM (IMDB.com)

Hybrid recommender systems

Combining content-based capabilities with collaborative-based approach or unifying the approaches into one unique model

Highly accurate and effective The need of more results in recommendations knowledge engineering than other approaches. Solution to cold start and sparsity issues

NetflixTM

Knowledge-based recommender systems

Gathering knowledge about users and generating approach to provide a recommendation by reasoning about what products can meet user needs In addition to the traditional approaches, mobile RS involve geographic data and enable context sensitive recommendations

No ramp-up problem because recommendations do not depend on user ratings

Suggestion ability is static

The restaurant recommender entree

Effective results on regional based recommendations

Heterogeneous and noisy environment; transplantation, validation and generality problems.

Taxi Routing apps

Content-based filtering

Mobile Recommender Systems

APPENDIX B. FINDINGS OF LITERATURE REVIEW Paper

Aim of the paper

Methods

Ref #

A Framework for Personalized Healthcare Service Recommendation

A personalized healthcare service recommendation framework that considers consumers’ health status to find adequate services for them. Providing accurate, low-cost clinical examination recommendations given patients’ self- reported data

Content based RS; SOA, web based contents (HTML,XML, web portal); DCAP algorithm to get measurable standard health status data

3

Content base RS; Rough sets, survival analysis (reliability) and rule-based expert systems (recommendation rules like AND,OR); Forward chaining:(this approach begins with a set of facts and rules, and tries to find a way of using those rules and facts to deduce a recommendation or a suitable action) Profile based RS

6

Text mining and association mining; Hybrid recommendation system; Semantic-web rule language

4

Collaborative based RS; Building ontology for physical activities- distance; Categories by effort level Agent networks (vital signs data recording)

7

Web Based Health Recommender System Using Rough Sets, Survival Analysis And Rulebased Expert Systems

Adapting Recommender Systems To The Requirements Of Personal Health Record Systems Enhancing Recommender Systems with Access to Electronic Health Records and Groups of Interest in Social Networks Design Of Physical Activity Recommendation Design of Diet Recommendation System for Healthcare Service Based on User Information Challenges for Nutrition Recommender Systems Challenges and Opportunities of Using Recommender Systems for Personalized Health Education

To supply Personal Health Records system users relevant, individually tailored health information by developing HRS with emphasis on semantics A semantics-based recommender system devised to embed selected advertisements in Digital TV programs - deal with the risks that arise from ignorance or commercial interest in social context Leisure time physical activity RS by collecting checkup data of people and recommend basing on similar people exercise Providing a personalization diet recommended service for the users who require the prevention and management for coronary heart disease difficulties and challenges in nutrition recommender systems which make suitable suggestion based on user profile Exploring the usage of RS for Health Education

8

2

No system provided. Development was suggested on user ratings and nutritional needs

5

Hybrid based; SOA; CTHES (the algorithms for personalization are based on the human expert knowledge)

1