A Classification of Healthcare Social Network Analysis Applications

Lamia Benhiba, Asmae Loutfi and Mohammed Abdou Janati Idrissi

TIME team, ENSIAS, Mohamed V University, Rabat, Morocco

Keywords: Network Dynamics, Structural Analysis, Social Network Analysis, Healthcare Organization, E-health,

Healthcare SNA Applications.

Abstract: As the web, social networks and the internet of things permeated our daily life; a new perspective for

understanding the complexity of our interconnectedness has become necessary. One approach that has

predominantly proven useful in discovering hidden relationships, connections and trends of complex

systems through mathematical and graphical techniques is Social Network Analysis (SNA). This approach

has become increasingly appeling for Healthcare in particular as many of this domain’s problems examine

systems with dynamic actors that interact with each other and exhibit emergent complex behaviors.

However, due to their multiplicity, the application of SNA methodologies proves to be a complex and

confusing endeavor. In an attempt to support the effort of applying SNA methodologies on Healthcare

research problems, this paper offers firstly a categorization of SNA methodologies (structural and dynamic

analysis), then inventories Healthcare SNA applications and classifies them into organizational and e-health

related problems. The resulting categorization helps identify the Healthcare research problems most

auspicious for SNA methodologies and should thus provide a guiding material of adequate SNA

methodologies for a given Healthcare research problem.

1 INTRODUCTION

With the emergence of the web, online social

networks, the internet of things etc., we are

increasingly aware of our interconnectedness and its

quantifiability. There is thus a growing realization

that the behavior of a system is shaped by the

interactions among its discrete components

(Bullmore, 2009). Thereby, the study of the

underlying network has become a stepping stone

into understanding complex systems.

Social network analysis (SNA) has gained a lot

of attention from both academia and practitioners of

various domains (from social science (Lewis, 2008),

economics (Krempel, 2002), politics (Klofstad,

2003), fight against crime and terrorism (Paulo,

2013), to neuroscience (Rubinov, 2010) and

epidemiology (Chen, 2007)). SNA offers a new

perspective for analysis and prediction as it focuses

on the interconnectedness between the various

constituents of the system and not on their inherent

characteristics. It relies on Graph theory to express

complex systems as a set of nodes (e.g. persons,

organizations etc.) interconnected through social

relationships (e.g. friendship, collaboration, transfer

of funds, co-occurrence etc.). SNA aims to model,

map, characterize and quantify topological

properties of the network, identify patterns of

relations and recognize the roles of sub-groups and

nodes within it.

With the increasing availability of data and the

advent and development of methods used to (a)

collect, store and (b) visualize network data

(Abraham, 2010), the interest in SNA has grown

massively. Healthcare is among the chief domains

where this particular approach is increasingly

appealing. Many healthcare research problems

examine systems with dynamic actors that interact

with each other and exhibit emergent complex

behavior. This makes these problems an auspicious

application of SNA’s methodologies and techniques.

The rest of the paper is organized as follows:

Section II will introduce SNA and its underlying

principles. It will also present the classification of

the different SNA methodologies used throughout

the literature into two main categories: structural and

dynamics analysis. Section III will particularly focus

on organizational healthcare and e-health SNA

applications and then match them with the two SNA

categories of section II. Section IV will summarize

the results and enumerate different opportunities and

challenges of the application of SNA in the

Benhiba L., Loutﬁ A. and Janati Idrissi M.

A Classiﬁcation of Healthcare Social Network Analysis Applications.

DOI: 10.5220/0006168001470158

In Proceedings of the 10th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2017), pages 147-158

ISBN: 978-989-758-213-4

147

healthcare domains. The last section will conclude

the paper with providing hints on future work.

2 SOCIAL NETWORK ANALYSIS

METHODOLOGIES

SNA is an interdisciplinary descriptive, empirical

discipline that studies networks as a mathematical

representation of complex systems by expressing

them in terms of relationships among actors. SNA

has four features: 1) It is motivated by a structural

intuition based on ties linking social actors, 2) It is

grounded in systematic empirical data, 3) It draws

heavily on graphic imagery, and 4) It relies on the

use of mathematical and/or computational models

(Freeman, 2004). The body of research has used

SNA methodologies in various domains to help

validate theories made about the structure or the

behavior of a social construct or complex system.

These methodologies can be categorized in several

ways. No matter how limited and flawed the effort,

doing so is useful because it guides the first steps

when attempting to answer a specific research

question.

We propose a categorization based on the

purpose of the SNA analysis. A review of seminal

works on SNA {(Wasserman, 1994), (Albert, 2002),

(Barabási, 2002), (Newman, 2003), (Watts, 2004),

(Christakis, 2011), (Scott, 2012), (Blonder, 2012),

(Barabási, 2016)} has rendered two distinct purposes

of network analysis:

▪ Structural Analysis: describes in discrete time

snapshots the topology of the network, the roles

of particular nodes, communities and subgroups

within the network etc.

▪ Dynamics Analysis: studies the changes of the

network’s topology through time (its evolution

and growth, the removal and adding of nodes and

edges, the change in link weight etc.) and

examines the diffusion of processes within the

network.

2.1 Structural Analysis

Structural analysis aims to examine the topology of

the network in order to uncover the overall

properties of the network and its constituents’

characteristics. It offers two perspectives: a micro-

view and a macro-view. The micro or Ego-centric

view focuses on a select actor (ego) and examines its

neighbors (nodes that are connected to it), their

neighbors and so forth. It studies the features of

personal networks. The macro or Socio-centric view,

on the other hand, provides a bird's eye perspective

of the network and helps examine the structural

patterns of the interactions among nodes with the

aim to explain and potentially generalize an

outcome. Studying the structure of a network relies

on a number of measures. Because of their ability to

give an indication on the topology of the network

(random, small world or scale free), the most studied

concepts in contemporary network research are:

degree distribution, clustering and Assortativity.

The degree of a node is the number of links it

has in the network and thus reflects the size of a

node's neighborhood. The average degree has been

used to gauge the cohesion (Kratzer, 2005) or

connectedness on the network level (Shrader, 1989).

The degree distribution is often plotted, using

histograms, to obtain insight into the overall

structure of the network and detect potential heavy-

tailed distributions.

The clustering coefficient represents the

tendency of nodes to form tightly knit groups within

the network. It is measured on the node level and on

the network level (Watts, 1998). The local

Clustering coefficient of a node is used to quantify

the level of transitivity within the network, i.e. the

chance that a node u is connected to w, when u is

connected to v and v is connected to w (uvw form a

triangle). The Network Clustering coefficient on the

other hand is defined as the average of the local

clustering coefficients of all the nodes.

Assortativity detects the level of homophily in a

network and measures the similarity of connections

in the graph with respect to the node degree

(McPherson, 2001). Assortativity can hint to the

existence of a core-periphery structure where a set of

closely knit nodes constitute the core of a network

and low degree nodes are left on the periphery.

Along these core concepts, many studies have

focused on community detection where algorithms

are applied to uncover locally dense connected

subgraphs (barabasi, 2016). Community detection

allows a deeper understanding of the network’s

structure and hidden connectivity patterns.

2.2 Dynamics Analysis

The study of network dynamics refers to two distinct

phenomena. We borrow the classification given by

(Blonder, 2012) in which they distinguish the

dynamics of the network from the dynamics on the

network. The first examines the growth of the

network, the factors behind the creation or

dissolution of new nodes and edges and the

HEALTHINF 2017 - 10th International Conference on Health Informatics

148

evolution of link strength through time. The second

studies propagation phenomena and the transfer,

throughout the network, of cascades such as

information, trust, opinion, behavior, money, goods

or pathogens etc.

2.2.1 Dynamics of the Network

The study of the dynamics of networks stems from

the need to understand the rules of networks’ growth

in order to predict their evolution. Networks evolve

by adding or removing nodes or links over time.

Research on the evolution of networks focuses on

the various dynamical processes that affect the

change of the network’s structure. The most popular

evolving networks’ models are Barabàsi and

Albert’s Preferential attachment (Barabàsi, 1999)

and (McPherson, 2001)’s homophily model.

▪ In the preferential attachment model, nodes

present a bias to connect to popular nodes that

have a large number of connections. These hubs

gain more connectivity as the network grows,

following a rich-gets-richer model (Bollobás,

2003).

▪ Homophily represents the likeliness of nodes to

connect to nodes that resemble them and which

are generally the neighbors in the network.

Nodes' connections are thus based on a conscious

action with embedded bias (It's more likely for

example to connect to a friend of a friend or an

individual with common interests than it is to a

random person).

While the main goal of these models is to predict the

probability of link formation, enabling thus Link

recommendations, nodes and links dissolution is

another aspect of network evolution that is

increasingly drawing interest. The goal here is to

predict links that are more likely to be dropped from

the network and to understand how it would affect

the structure of the network.

2.2.2 Dynamics on the Network

In an attempt to understand the dynamic effect of

network properties on diffusion, various studies

relied on mathematical models originally used in

fields such as epidemiology, sociology and

economics. Louni et al. (Louni, 2014) classified the

most popular information diffusion models into

three categories:

▪ Contagion Models: these models build on the

idea that a cascade flows in a network in the

same way a contagious disease spreads through a

population. The most widely used models for

studying contagion are usceptible-infected (SI),

susceptible-infected-susceptible (SIS) and

susceptible-infected-recovered (SIR). The

models consider cascades to spread from

adopters (infected) to susceptible nodes and

consider the possibility of retracting the cascade

for recovered nodes.

▪ Social Influence Models: These models assume

that the social influence between nodes affects

the diffusion of cascades (opinions or behaviors

for instance). The most widely studied and used

social influence models are the Linear Threshold

(LT) (Granovetter, 1978) and the Independent

Cascade (IC) (Goldenberg, 2001).

▪ Social Learning Models: In contrast with

previous models which ignore the actions and

decision making of actors, the nodes in social

learning models are considered rational agents

who observe outcomes of prior behaviors and

decide accordingly. The decision of a user to

forward information is modeled using game

theory concepts where the user maximizes some

utility for himself (Jackson, 2008).

Studying the spread of cascades within a network

offers theoretical and empirical tools to not only

quantify the propagation process, but to forecast it as

well.

3 A CATEGORIZATION OF

HEALTHCARE SNA

APPLICATIONS

Healthcare’s purpose is to ensure the well-being of

people by taking both proactive and active actions.

Healthcare organizations take preventive actions like

sharing information about healthy life styles, the

vaccines in the market etc., providing psychological

council, or conducting research for improving health

services and the health life of people. They also take

reactive actions by administrating drugs, doing

surgery helping people with chronic illness etc.

Healthcare research covers a lot of areas such as

clinical, biomedical, health systems and services and

social, cultural, environmental and population health

research. Healthcare research is undertaken to

establish the foundation for developing effective

therapeutic interventions to expose to individuals

and communities, to support enhancing and

understanding illness and health and safeguard and

enhance the health of persons and populations

(Steinwachs, 2008). Due to the complexity of the

healthcare system, a methodological approach is

needed to analyze, monitor and ensure the

A Classiﬁcation of Healthcare Social Network Analysis Applications

149

effectiveness of its endeavors. SNA is thus

introduced as a powerful new way to discover

valuable hidden connections, relationships, trends

and insights.

3.1 Methods and Materials

The purpose of this work is to establish a matching

between SNA methodologies, described in section

II, and healthcare application domains in order to

uncover trends of SNA applications in the healthcare

field. To accomplish this, we classified SNA

applications in healthcare according to their

functional domain and finally assigned SNA

methodologies to each healthcare domain.

To identify the SNA applications in healthcare,

we scanned three databases (Scopus, Science Direct

and IEEExplore Digital Library) for the last 10

years, using various research terms related to: (social

network analysis OR graph analytics) AND

(healthcare, e-health, health organization, behavioral

OR epidemiology) in Title, keywords and abstract.

The search was restricted to English scientific

literatures that are in peer-reviewed venues and

duplicated works were eliminated. A paper is

selected when the algorithm and methodology of

social network analysis in e-health or healthcare

organizations. In this paper, the 16 works listed in

Table 1 will be considered. During the data

extraction process, we included information about

the title of the article, the year of publication, the

authors, the country, the application of healthcare

research, the data sources, the applied methodology

and the type of the modeled graphs.

3.2 Categories of Healthcare SNA

Applications

There are many areas of healthcare that can apply

SNA. In this paper, we focus on the areas that drew

the most attention: healthcare organization and e-

health.

3.2.1 Healthcare Organization

A report of the Institute of Medicine suggested six

aspects for improvement of the healthcare system. It

needs to be: Safe (healthcare services to patients

should be secure and not cause any injuries),

Effective (care services based on scientific evidence

for increasing healthy outcomes), Patient-centered

(present to the patient the care service that respect

their needs, values and preferences), Timely

(provide care assistance early on before any

complications occur), Efficient (make healthcare

services available with minimum costs and without

waste), Equitable (people should have the same

access to healthcare services).

To achieve these purposes, healthcare

organizations need to collaborate to share

information about their operational and research

works, establish policies for more effective and safe

treatments and manage their waste by detecting

fraud of healthcare providers.

i. Health policy

The World Health Organization (WHO) defined

health policy as “the decisions, plans, and actions

that are undertaken to achieve specific health care

goals within a society”. The specification of rules

that healthcare stakeholders should follow in terms

of defining characters for differing groups, making a

reference for treatments and actions undertaken by

healthcare practitioners and sharing this information

with people, are the various things attained by a

health policy institution.

The study proposed by (Millard, 2015) adresses

WHO's Essential Medicine List (EML). EML is a

list of medicines that assists countries on selecting

the treatments of each priority requirement. In this

article, SNA is used to inspect the social, political

and economic areas for adding the encouragement

for Misoprostol’s use for preventing and treating a

postpartum hemorrhage, especially in low income

countries, according to the WHO's EML in 2011. A

study the chronology of WHO misoprostol

applications and evolution of related social networks

are applied to evaluate the relation of health policy

and this social area.

In (Takahashi, 2016), a descriptive analysis of

duplicative prescription practices is performed.

When patients take orders for the same state from

two or more sources, we talk about duplicative

prescription practices. This practice is the origin of

medical waste. Some patients resell drugs for extra

cash and can also cause adverse effects. The

descriptive analysis was conducted by using the

measurement of SNA and describing the prevalence

(the rate of persons with an illness or characteristic)

of duplicative prescription through ages. The study

also calculated the density of the medical facilities

and patients network for each class of drugs defined

by their prevalence.

In (Bramhachari, 2016), the authors conducted a

qualitative ego-network analysis to understand

dominance of Rural Medical Practitioners (RMPs) in

West Bengal, India. They inspected the genesis of

RPMs’ social links with various actors in the health

HEALTHINF 2017 - 10th International Conference on Health Informatics

150

system and showed the operators donating their

subsistence over the years, by using SNA. By

identifying the ties in RMPs’ network that are

formal healthcare providers, the healthcare market

and the community, we can comprehend the

dynamics of the healthcare market.

Guo et al. use healthcare claims data of the

Medical Insurance Association of Anhui Province to

look into details of referrals social network. They

design a referral social network where the nodes are

hospitals and ties are patient-transferred between

hospitals. The authors conduct a structural analysis

to measure the degree and centrality to describe the

relationship between this variable and other patients

and hospital variables. Finally, they explore rules

between the variables of the referral social network

and variables of quality of the healthcare to help

healthcare providers minimize cost and length of

stay in the hospital and increase the efficiency of

medical resources (Guo, 2015).

ii. Healthcare organizational collaboration

Healthcare organizations need to collaborate with

each other in order to improve the quality of care to

the patients, in term of efficient research, cost

decrease, good management of resources etc.

- Intra-organizational Collaboration (actors)

Soulakis et al. made use of patients’ Electronic

Health Records (EHR) with heart failure to explore

the collaboration between healthcare providers and

patients in The Northwestern Memorial Hospital

(NMH). The access to EHR provides a large amount

of data about interactions between providers and

patients (Soulakis, 2015). A structural SNA

methodology is used to describe the collaboration

between patients and providers through a bipartite

network (the source node was a provider, the target

node was a patient and the edge represented the

patient record accessed by this provider), and a

provider collaboration network which is a network

of the common access of patient records by

providers (the node represent providers and the

edges are established when two providers have

access to more than 10 common patient records).

Data is extracted from the Enterprise Data

Warehouse (EDW) of NMH. The network is

afterwards visualized and clique formation is

analyzed. A graph database is used to process

queries and answer questions about care and

provider-patient collaboration.

- Inter-organizational Collaboration (institutions)

Caniato et al. conducted a case study on

management of healthcare waste in a region with

specific characteristics: Gaza Strip (Caniato, 2015).

They employ an SNA and stakeholder analysis to

explore and comprehend the effects of a range of

logistical and socio-economic factors on the

effectiveness of stakeholder networks in the region.

Caniato et al. applied a structural analysis of

interaction frequency and information exchanged

among stakeholders that are public authorities,

health providers, supporting actors and others.

The study performed by (Schoen, 2014) used

SNA to confirm the suggestion that when we take

funding to concentrate on multi-sector collaboration

in Social Innovation for Missouri (SIM) program, a

public health program interventions to prevent

obesity and stop tobacco, develop various

partnership structures than other grantees. The

authors explore different variables as the level of

collaboration and frequency of contact by applying

SNA. They measure the network descriptors such as

average degree, density, betweeness centralization

and degree centralization to evaluate the network of

contacts and the collaboration of different

stakeholders.

Dianas et al. studied an excellence program in

low and middle-income countries provided by the

National Heart, Lung, and Blood Institute-

UnitedHealth to fund 11 centers of excellence

(Dianis, 2016). In order to prove the effect of

collaboration with a federal support, they used SNA.

They created a network of the program’s

stakeholders by considering links as collaborations

on administrative support and research projects.

They later compared the resulting network before

the development of the Centers of Excellence

Program and after.

Kawonga et al. presented a case study of HIV

monitoring and evaluation to examine and

understand the way that Disease Control

Program(DCP) and General Health Services(GHS)

managers communicate when they make a health

reform to make an administrative integration in

South Africa (Kawonga, 2015). For this purpose,

they described the entire network by using density,

degree and betweenness centrality. They also used

density and a measure of homophily to analyze sub-

groups networks. A block-model analysis was used

to identify the connections between management

committees and manager groups.

The paper presented by Khosla et al. introduced

a study of collaboration between HIV agencies in

Baltimore (Khosla, 2016). SNA and relation

coordination were used to analyze the quality of

coordination between HIV agencies when they

accessed resources like information, around seven

A Classiﬁcation of Healthcare Social Network Analysis Applications

151

dimensions such as accuracy of communication,

knowledge of agencies’ work, frequency, problem-

solving communication, timeliness, shared goals and

mutual respect. Density and centrality of the

network of agencies collaboration were calculated as

part of an SNA structural analysis. For the study of

relation coordination, a questionnaire was used

among these seven dimensions about

communication and relationships between HIV

agencies. SNA measures were used to describe the

whole network: density and degree centralization

and to describe a position of an actor in the network:

degree, indegree, centrality, degree centrality,

weighted degree centrality, betweenness centrality

and closeness centrality.

Wang et al. apply SNA to explore the

collaboration between surgeons, assistants and

anesthetist working at different hospitals by using

data from Private Health Insurance (PHI) claims in

Australia. They also studied their impact on quality

and cost of care (Wang, 2014). SNA is used to

analyze the collaboration among the three healthcare

providers, study the topologies of the network to see

how doctors work while treating patients and

examine the effect of these topologies on quality and

cost of care for patients. The effect of network

structure on quality and cost is analyzed around

efficiency metrics that are Length of Stay (LoS),

Medical costs and Complication rate. They

thereafter designed two kind of networks: one for

collaboration between surgeons, assistant surgeons

and anesthetists; and the second centered on a

surgeon collaboration network to study the

connections of each surgeon. The measures of SNA

used in this paper are: the size of node (charged

number of this provider), tie strength (total of

common admission between two providers),

centrality to have an idea on the influence of a

vertex in the network) and density.

- Research collaboration

Collaboration research is important to enhance the

quality of research by determining the leaders in a

subject and affording reasonable proposals and

scientific evidence to make a finance of specific area

of research policy (Wu, 2015).

Bien et al. presented a case study of the use of

SNA in the context of biomedical research grants

collaboration at the University of Arkansas for

Medical Sciences (UAMS) (Bian, 2013). The

objective of this study was to evaluate the research

collaboration networks (RCNs) for both level inter-

and intra-institution in the community of the Clinical

Translational Science Award (CTSA) and examine

the effectiveness of CTSA funded at UAMS and

their influence on environment of research

collaboration in an institution. For categorizing the

network, the authors calculated the network’s path

length and its clustering coefficient. They also

measured the structural characteristics such as

centrality to identify the important (the influencer or

contributor) node in the research community. They

examined the structural characteristics and the

network dynamics of the RCNs.

Wu et al. (Wu, 2015) performed a study on the

scientific research collaboration in the specialty of

psychiatry. This work applied SNA to analyze the

structure of scientific collaboration in psychiatry by

using the notion of co-authorship, which can

determine the authors, institutions and countries

involved in the scientific collaboration network. In

each level of authors, institutions and countries, the

author characterized psychiatry research

collaborative behaviors, K-plex analysis and Core-

periphery are the methods used in this paper to

describe the collaborative connections. The authors

measure centrality to detect the central, the core

position and actor with control and possession of

valuable research resources in each collaboration

network.

3.2.2 E-health

WHO defines e-Health as “the use of information and

communication technologies (ICT) for health”(WHO,

2016). By using ICT in this area, we can assist

patients for treatments; share information about

healthy life styles, follow people with diseases etc.

The study presented by Chomutare et al. adresses

weight loss performance by monitoring online

interaction behaviors for forcast them (Chomutare,

2014). The authors captured data from a sub-forum

of two online communities concerned with obesity.

The first was for people older than 50 years and the

second was for people that needed surgical

interventions as they interacted before and after the

intervention for weight loss performance. Structural

SNA is used to create a classification of people who

lose significant weight (performers) and the others

(non-performers) to predict weight loss. Authors

remarked that the top performers were connected at

different sub-community and were more active

online.

Pachucki et al. measure objectively the social

interaction between 6th-grade students at a private

K-8 School in the State of California by using

accelometers and RFID technology. The purpose of

this paper is to study the relations between social

interaction and mental health behaviors such as self-

HEALTHINF 2017 - 10th International Conference on Health Informatics

152

esteem and depressive symptoms of early

adolescences. Due the focus on health behaviors,

health status and changes in network structure; the

authors measure the characteristics of social

environment, health behavior, social interaction

network and mental health to analyze them using

bivariate associations. They use a stochastic actor-

based modeling (SABM) framework to join the

dynamic co-evolution of social ties and self-esteem

or depressive symptoms (Pachucki, 2015).

Goodall et al. explored the importance of ICT for

searching information behavior by older migrants

with Culturally And Linguistically Diversity

(CALD) (Goodall, 2014). They determined factors

that leverage the use of ICT to locate information.

These factors can be education, migration, socio-

economic status, ethnicity and English proficiency

of older migrants. The study undertook by Goodall

et al. focused on the search of cancer-related

information by the group. The authors used SNA

and a constructivist grounded theory method to

analyze the data captured in the interviews, and then

they studied the preferences and uses of traditional

information sources compared to modern ones (PC,

Internet and mobiles).

Table 1: A methodological classification of each healthcare application represented in this paper.

Reference of

paper and the

country

The SNA application

Healthcare

categorization

Methodological

categorization

Dataset/Size of

the network

Algorithms/Metrics

(Millard,

2015), UK

They establish a

chronology of WHO

misoprostol applications

and they examine the

evolution of related

social networks and the

nested subset network of

the WHO EML

misoprostol applications

Healthcare

organization:

Health policy

Dynamics of the

network

238

organizations

and individuals

Chronological approach

combined with SNA

(evolution of social

network) : density,

geodesic distance,

diameter, centrality,

nesting, clique formation

(Takahashi,

2016), Japan

They conduct a

descriptive analysis of

medical waste by

studying duplicative

prescription practices

Healthcare

organization:

Health policy

Structural

Data are from

health insurance

claims database

1,243,058

insured people

and their

dependents

Statistical analyses:

correlation by using

scatter plots

and the Pearson

correlation coefficient

SNA: bipartite networks,

density

(Bramhachari,

2016), India

They use a qualitative

ego-network method to

understand the RMP

network

Healthcare

organization:

Health policy

Structural 35 participants

Qualitative Ego-network

method

(Guo, 2015),

China

They analyze the

healthcare claims data of

the Medical Insurance

Association of Anhui

Province to design a

referral social network.

Healthcare

organization:

Health policy

Structural

72 hospitals and

8856 patients in

the claim data

from Medical

Insurance

Bureau

Community detection:

spinglass, edge

betweenneess, label

propagation, optimal,

walktrap.

Simple linear regression:

Los, Medical cost,

Degree, closeness

centrality, betweenness

centrality, eigenvector

centrality, rank of

Hospital.

Rules exploration:

Decision tree.

A Classiﬁcation of Healthcare Social Network Analysis Applications

153

Table 1: A methodological classification of each healthcare application represented in this paper (cont.).

(Soulakis,

2015), USA

They make a bipartite

network of providers

accessing patients’

records and a provider

collaboration network to

describe collaboration

between patients and

providers.

Healthcare

collaboration

intra-

organization

Structural

Collaborative

electronic health

record (HER)

1504 nodes and

83 998 edges

Bipartite network

Module and clique

identification: heuristic

community detection

algorithm, kCliques

algorithm

(Caniato,

2015), Italy

They conduct a

Stakeholder analysis and

structural SNA of

interaction frequency

and information

exchanged between

stakeholders

Healthcare

collaboration

inter-

organization

Structural

Dataset

constructed

from 16

structured and

two semi-

structured

interviews

SNA and stakeholder

analysis

(Schoen,

2014), USA

They apply structural

SNA to both contact and

collaboration networks

Healthcare

collaboration

inter-

organization

Structural

23 Missouri

communities

in early 2012

SNA: average degree,

density, degree

centralization, and

betweenness

centralization

(Dianis,

2016),USA

They conduct structural

SNA on the network of

all stakeholders in an

excellence program

Healthcare

collaboration

inter-

organization

Structural

11 contracts in

10 countries

128 nodes

SNA: density, average

distance

(Kawonga,

2015), South

Africa

They apply structural

and dynamic

methodologies on the

communication network

of GHS and DCP

managers

Healthcare

collaboration

inter-

organization

Dynamics of the

network

51 managers in

two provinces

during 2010-

2011

Dataset: HIV

data collation

and HIV data

use

SNA: density, degree,

betweenness centrality

and E-Index (measure of

homophily)

Block modelling

(Khosla,

2016), USA

They combine SNA and

relational coordination

to measure the quality of

coordination among

HIV agencies

Healthcare

collaboration

inter-

organization

Structural 57 agencies

SNA: density, degree

centralization, weighted

degree centralization,

closeness centrality,

betweenness centrality

Relational coordination:

frequency, timeliness and

accuracy of

communication, problem-

solving communication,

knowledge of agencies'

work, mutual respect and

shared goals

(Wang, 2014),

Australia

They use SNA to

explore the collaborative

network and surgeon

centric collaboration

network to analyze the

impact of collaboration

on the quality and cost

of care

Healthcare

collaboration

inter-

organization

Structural

Health insurance

claims: 59256

admissions

performed by

870 surgeons

SNA: degree centrality,

closeness centrality,

betweenness centrality ,

density, clustering

coefficient, number of

triangles

HEALTHINF 2017 - 10th International Conference on Health Informatics

154

Table 1: A methodological classification of each healthcare application represented in this paper (cont.).

(Bian, 2013),

Arkansas,

United States

They apply structural

and dynamic

methodologies on the

network to identify

leaders and influencers

in a research

collaboration network

Healthcare

research

collaboration

Dynamics of the

network

The

Automated

Research

Information

Administrator

(ARIA) and the

Translational

Research

Institute (TRI)

SNA: measures of

centrality, mean path

length, clustering

coefficient, characteristic

path length, diversity

Temporal evolution:

average number of new

edges, centrality leaders

(Wu, 2015),

China

They use the measure of

centrality to conduct an

SNA on authors,

institutions and

countries collaborating

on psychiatric research

Healthcare

research

collaboration

Structural 36557 papers

about psychiatry

from Science

Ciation Index

Expanded (SCI-

Expanded) in

web of

science

SNA: centrality, K-plex

analysis, Core periphery

Hierarchical clustering

(Goodall,

2014),

Australia

They use Grounded

theory and a

qualitative SNA on the

egocentric network of

individuals and their

sources of information,

and compare the

resulting networks

E-health:

Information

technology

access

Structural Interview with

54 participants

aged 63–94

years

Constructivist grounded

theory method (CGTM)

SNA: egocentric network

(Chomutare,

2014),

Norway

They use structural SNA

to classify people

according to their ability

to lose weight

significantly

E-health:

Social

influence and

behavior

analysis

Structural Binomial classification:

Bayes and decision tree

method

SNA: bipartite graph ,

degree centrality,

betweenness centrality

Expansion-reduction

method

Community detection:

hierarchical clustering

(Pachucki,

2015) USA

They measure the

association between

social interactions and

depressive symptoms

and self-esteem of early

adolescences at a private

K-8 School in the State

of California

E-health:

Behavioral

analysis

Dynamics on the

network

40 students of

sixth-graders at

a private K-8

school

Measures of Social

environment, Health

behaviors ( physical

activity and food choice),

Social interaction

networks, dependent

variables ( self-esteem

and depressive symptoms)

to make bivariate

associations and SNA (

size of personal networks,

transitivity and closeness

centrality)

4 SYNTHESIS AND DISCUSSION

The following table represents the matching between

the categorization of healthcare SNA applications

and the SNA methodologies.

Figure 1 shows that the highest number of the

A Classiﬁcation of Healthcare Social Network Analysis Applications

155

included research works studied healthcare

collaboration and focused on collaboration between

institutions. This might be due to the accessibility of

inter-collaboration data compared to intra-

collaboration data (insurance claims vs. EHRs). E-

health is a new area of SNA application and can thus

present new opportunities to researchers, although

the lack of data especially in low-income countries

may be problematic.

Structural SNA methodologies are the most used,

whereas dynamic methodologies are mainly used in

problems related to healthcare organization (cf.

Table 2). The prevalence of structural analyses could

be due to the complexity of dynamic methodologies

compared to structural ones. Associations between

SNA structural metrics and domain-specific metrics

are however rarely examined; which constitutes a

research question that deserves further attention.

There is also a pressing need to move beyond the

static view of the network, visualized in snapshots,

to a visualization that captures more accurately the

dynamic processes that reshape the network (a

movie-like visualization for instance). Another

research opportunity relates to the application of

dynamics on the network methodologies such as

propagation and diffusion models. While these

methodologies have been exclusively used in

behavioral analysis, they have the potential to

examine the propagation of information in social

networks and uncover hidden processes shaping

collaboration or policy making endeavors.

With respect to data collection, a third of the

included studies gathers data from questionnaires.

This raises data completion issues and inaccuracies

arising from informant bias and stresses the

pertinence of alternative data collection methods

relying on RFID technologies, accelerometers or

EHRs etc. Online social networks (OSN) such as

Facebook, Twitter etc. are widely used nowadays

and can help perform social behavioral analyses. We

can comprehend e-health tools and design future IPC

to promote effective interaction behaviors in OSNs

by correlating interaction behaviors and a specific

disease (Chomutare, 2014). However, many

challenges could face such studies. When SNA

methodologies are used on OSNs’ social data, it is

difficult to distinguish between the effects of

Homophily and those of peer influence. Questions

relating to the sufficiency of collected data and its

representativeness of a given behavior to infer

conclusions remain unanswered. There are also the

pressing issues of privacy and ethics regarding data

collection and which are consequences of the

inherent processes of the social graph's construction

and design. Anonymization, consent and privacy are

among the issues that need further attention.

Figure 1: Number of papers of each healthcare application

and methodological classification (Dynamic, Structural).

Table 2: The SNA methodology used in each healthcare

domain.

Healthcare

categorization

Functional

Sub-categorizing

Methodological

categorization

Healthcare

organization

Health policy

Dynamics of the

network, Structural

Healthcare

collaboration

intra-organization

Structural

Healthcare

collaboration

inter-organization

Structural, Dynamics

of the network

Healthcare

research

collaboration

Dynamics of the

network, Structural

E-heath

Information

technology access

Structural

Social influence

and behavior

analysis

Structural, Dynamics

on the network

Furthermore, social media provides a large

amount of data. The resulting networks are thus very

large and new tools are needed to process them. Big

data network analysis is increasingly drawing the

attention of researchers. However, due to the

complexity of healthcare research problems, further

Health policy

Healthcare

collaboration

Behavior

analysis

Information

technology

Number

of papers

Structural

Dynamic

HEALTHINF 2017 - 10th International Conference on Health Informatics

156

research is needed in order to produce domain-

specific tools.

5 CONCLUSIONS

The purpose of this paper was to propose a

classification of healthcare SNA applications based

on a review of papers that used structural and

dynamic SNA methodologies to answer healthcare-

related research problems. We classified these

research works into two categories: One concerning

healthcare organizations and pertaining to policy

making, communication, and collaboration and a

patient-oriented category which concerns patients’

behaviors, social influence and healthcare

information access.

The proposed classification of healthcare SNA

applications is preliminary and requires further

enrichment through the inclusion of other research

works. The level of adequacy of a chosen SNA

methodology to a given Healthcare research problem

is yet to be examined. Experimental studies will

have to be conducted to establish comparative

analyses between variations of a given methodology

for a particular problem. For instance, different

subsets of metrics can be used and compared for

structural SNA methodologies, various propagation

models can be simultaneously tried for dynamic

SNA methodologies.

REFERENCES

Bullmore, E., & Sporns, O., 2009. Complex brain

networks: graph theoretical analysis of structural and

functional systems. Nature Reviews

Neuroscience,10(3), 186-198.

Lewis, K., Kaufman, J., Gonzalez, M., Wimmer, A., &

Christakis, N., 2008. Tastes, ties, and time: A new

social network dataset using Facebook. com.Social

networks, 30(4), 330-342.

Krempel, L., & Plümper, T., 2003. Exploring the

dynamics of international trade by combining the

comparative advantages of multivariate statistics and

network visualizations. Journal of social structure,

4(1), 1-22.

Klofstad, C. A., Sokhey, A. E., & McClurg, S. D., 2013.

Disagreeing about disagreement: How conflict in

social networks affects political behavior.American

Journal of Political Science, 57(1), 120-134.

Paulo, D., Fischl, B., Markow, T., Martin, M., &

Shakarian, P., 2013. Social network intelligence

analysis to combat street gang violence. arXiv preprint

arXiv:1306.6834.

Rubinov, M., & Sporns, O., 2010. Complex network

measures of brain connectivity: uses and

interpretations. Neuroimage, 52(3), 1059-1069.

Chen, Y. D., Tseng, C., King, C. C., Wu, T. S. J., & Chen,

H., 2007. Incorporating geographical contacts into

social network analysis for contact tracing in

epidemiology: A study on Taiwan SARS data. In

Intelligence and Security Informatics: Biosurveillance

(pp. 23-36). Springer Berlin Heidelberg.

Abraham, A., Hassanien, A. E., & Snâaésel, V. (Eds.).,

2010. Computational social network analysis: trends,

tools and research advances. Springer.

Freeman, L. C., 2004. The development of social network

analysis. Vancouver: Empirical Press.

Wasserman, S., 1994. Social network analysis: Methods

and applications (Vol. 8). Cambridge university press.

Albert, Réka, and Albert-László Barabási, 2002.

"Statistical mechanics of complex networks." Reviews

of modern physics 74.1 : 47.

A.-L. Barabási. Linked: The New Science of Networks.

Perseus Books Group, 2002.

Newman, Mark EJ, 2003. "The structure and function of

complex networks." SIAM review 45.2 : 167-256.

D. Watts. Six Degrees: The Science of a Connected Age.

Norton, 2004.

N. Christakis and J. Fowler. Connected: The Surprising

Power of Our Social Networks and How They Shape

Our Lives. Back Bay Books, 2011. (Scott, 2012).

Scott, John. Social network analysis. Sage, 2012.

Blonder B, Wey T, Dornhaus A, James R, Sih A. 2012. Te

mporal dynamics and network analysis. Meth Ecol

Evol. doi: 10.1111/j.2041-210X.2012.00236.x.

Barabási, Albert-László. "Network science" Cambridge

University Press, Cambridge (August 5, 2016).

Kratzer, J., Leenders, R. T. A., & Van Engelen, J. M.,

2005. Informal contacts and performance in

innovation teams. International Journal of Manpower,

26(6), 513-528.

Shrader, C. B., Dellva, W. L., & McElroy, J. C., 1989.

Social networks and group performance – A partial

test of the Pearce and David hypotheses. International

Journal of Small Group Research, 5(2), 145-162.

Watts, Duncan J.; Strogatz, Steven H., June 1998.

"Collective dynamics of 'small-world' networks".

Nature 393 (6684): 440–442.

McPherson, M., Smith-Lovin, L., & Cook, J. M., 2001.

Birds of a feather: Homophily in social networks.

Annual review of sociology, 415-444.

Barabási, A. L., & Albert, R., 1999. Emergence of scaling

in random networks.science, 286(5439), 509-512.

Bollobás, B., & Riordan, O., 2003. Mathematical results

on scale-free random graphs. Handbook of graphs and

networks, 1, 34.

Louni, Alireza, and K. P. Subbalakshmi, 2014. "Diffusion

of information in social networks." Social Networking.

Springer International Publishing. 1-22.

Bollobás, B., & Riordan, O., 2003. Mathematical results

on scale-free random graphs. Handbook of graphs and

networks, 1, 34.

A Classiﬁcation of Healthcare Social Network Analysis Applications

157

Louni, Alireza, and K. P. Subbalakshmi, 2014. "Diffusion

of information in social networks." Social Networking.

Springer International Publishing. 1-22.

Granovetter, Mark, 1978. "Threshold models of collective

behavior." American journal of sociology: 1420-1443.

Goldenberg, J., Libai, B., Muller, E, 2001.: Talk of the

network: a complex systems look at the underlying process

of word-of-mouth. Mark. Lett. 12(3), 211–223.

Jackson,M.O, 2008.: Social and Economic Networks.

Princeton University Press.

Steinwachs DM, Hughes RG, 2008. Patient Safety and

Quality: An Evidence-Based Handbook for Nurses,

Hughes RG, editor.

Millard, Colin, Petra Brhlikova, and Allyson Pollock.

2015. 'Social networks and health policy: The case of

misoprostol and the WHO model essential medicine

list', Social Science & Medicine, 132: 190-96.

Takahashi, Yoshimitsu, Tatsuro Ishizaki, Takeo

Nakayama, and Ichiro Kawachi. 2016. 'Social network

analysis of duplicative prescriptions: One-month

analysis of medical facilities in Japan', Health Policy,

120: 334-41.

Bramhachari, Rittika, and Sabyasachi Mandal. 2016. 'a

social network analysis of rural medical practitioners

in the sundarbans, west bengal', BMJ Global Health,

1: A41-A42.

Guo H., Wei F., Cheng S., Jiang F., 2015. Find Referral

Social Networks, International Symposium on Security

and Privacy in Social Networks and Big Data, 58-63.

Soulakis, Nicholas D., Matthew B. Carson, Young Ji Lee,

Daniel H. Schneider, Connor T. Skeehan, and Denise

M. Scholtens. 2015. 'Visualizing collaborative

electronic health record usage for hospitalized patients

with heart failure', Journal of the American Medical

Informatics Association : JAMIA, 22: 299-311.

Caniato, M., T. Tudor, and M. Vaccari. 2015.

'Understanding the perceptions, roles and interactions

of stakeholder networks managing health-care waste:

A case study of the Gaza Strip', Waste Management,

35: 255-64.

Schoen, Martin W., Sarah Moreland-Russell, Kim Prewitt,

and Bobbi J. Carothers. 2014. 'Social network analysis

of public health programs to measure partnership',

Social Science & Medicine, 123: 90-95.

Dianis, Nancy L., Tracy L. Wolbach, and Maura

Spiegelman. 2016. 'The NHLBI–UnitedHealth Global

Health Centers of Excellence Program: Assessment of

Impact of Federal Funding Through a Social Network

Analysis', Global Heart, 11: 145-48.e1.

Kawonga, Mary, Duane Blaauw, and Sharon Fonn. 2015.

'Exploring the use of social network analysis to

measure communication between disease programme

and district managers at sub-national level in South

Africa', Social Science & Medicine, 135: 1-14.

Khosla, Nidhi, Jill Ann Marsteller, Yea Jen Hsu, and

David L. Elliott. 2016. 'Analysing collaboration

among HIV agencies through combining network

theory and relational coordination', Social Science &

Medicine, 150: 85-94.

Wang, F., U. Srinivasan, S. Uddin, and S. Chawla. 2014.

"Application of network analysis on healthcare." In

Advances in Social Networks Analysis and Mining

(ASONAM), 596-603.

Wu, Y., and Z. Duan, 2015. 'Social network analysis of

international scientific collaboration on psychiatry

research', International Journal of Mental Health

Systems.

Bian, J., M. Xie, U. Topaloglu, T. Hudson, and W. Hogan.

2013. "Understanding biomedicai research

collaborations through social network analysis: A case

study." In Bioinformatics and Biomedicine (BIBM),

2013 IEEE International Conference, 9-16.

World health organization 2016. Health topics, e-health,

url: http://www.who.int/topics/ehealth/en/

Chomutare, T., A. Xu, and M. S. Iyengar. 2014. "Social

Network Analysis to Delineate Interaction Patterns

That Predict Weight Loss Performance." In 2014 IEEE

27th International Symposium on Computer-Based

Medical Systems, 271-76.

Pachucki M. C., Ozer E. J., Barrat A., Cattuto C., 2015.

Mental health and social networks in early

adolescence: A dynamic study of objectively-

measured social interaction behaviors. Social Science

and Medecine 125, 40-50.

Goodall, K. T., L. A. Newman, and P. R. Ward. 2014.

'Improving access to health information for older

migrants by using grounded theory and social network

analysis to understand their information behaviour and

digital technology use', European Journal of Cancer

Care, 23: 728-38.

HEALTHINF 2017 - 10th International Conference on Health Informatics

158