Socialising Health Burden Through Different Network Topologies: A - - PDF document

Socialising Health Burden Through Different Network Topologies: A Simulation Study

Adrian PEACOCKa, Anthony CHEUNGb, Peter KIMb and Simon K. POONa,1

a The University of Sydney, School of Information Technologies, Australia b The University of Sydney, School of Mathematics and Statistics, Australia

• Abstract. An aging population and the expectation of premium quality health

services combined with the increasing economic burden of the healthcare system requires a paradigm shift toward patient oriented healthcare. The guardian angel theory described by Szolovits [1] explores the notion of enlisting patients as primary providers of information and motivation to patients with similar clinical history through social connections. In this study, an agent based model was developed to simulate to explore how individuals are affected through their levels

• f intrinsic positivity. Ring, point-to-point (paired buddy), and random networks

were modelled, with individuals able to send messages to each other given their levels of variables positivity and motivation. Of the 3 modelled networks it is apparent that the ring network provides the most equal, collective improvement in positivity and motivation for all users. Further study into other network topologies should be undertaken in the future.

• Keywords. Guardian Angel, social network, agent-based model, emotional

contagion, NetLogo

Introduction Despite an increase in the number of medical practitioners in Australia over recent years [1], growth in the aging demographic and associated chronic disease, combined with a demand for quality health services from the entire population has driven the need for a change in approach to chronic disease treatment and management. The associated economic strain mounting on the healthcare system is of concern [2], and it is unreasonable to think that physicians can accommodate this pertinent issue without a change in focus to healthcare delivery. Using an agent-based simulation, this study aims to gain insight into the effect that social network topology has on distribution of benefit to individuals, when deploying a Guardian Angel type healthcare system. The concept of a Guardian Angel system was first described as a patient centered system that provides the individual with support via access to education, monitoring progress, transmitting alerts and reminders, and offering support through contact with

• ther patients and medical professionals [3]. The aim was to empower the user to take

an active role in their health, and make better personal decisions in promoting their own

1 Corresponding Author: Associate Professor Simon Poon, Faculty of Engineering and IT, The University

• f Sydney, NSW 2006, Australia; E-mail: simon.poon@sydney.edu.au.

health status. Since its conception, the focus has been magnified further toward a social, patient centered structure to alleviate strain on health practitioners and the health

• system. The use of a telephone based network platform to facilitate support and

improve health behaviors and outcomes between individuals has been investigated with promising results [4], [5]. With significant potential shown by these systems and others, it validates the potential for further application of the guardian angel concept in providing personal support and care for patients with a range of chronic diseases. While these structures in health allow for relations among individuals who share similar diseases, economic, geographic, or cultural backgrounds, and perceptions of health [6], there is minimal research into if these systems equally distribute the user derived health stimulus. Patient driven social networks hold significant potential and value in healthcare due to their unquestionable ability to promote ideas and information to other individuals, provide support to achieve goals, high scalability, and their economic value [7]. Currently however, group dynamics and health outcomes relating to the affect that different network structures have on the ability for all users to receive equally shared improved health is poorly understood and further study is required.

• 1. Methods

A simulation model developed based on the concept of “emotional contagion” introduced by Christakis and Fowler [8], eGuardian Angel is an agent based model of a social innovation for individuals with chronic disease, modelled on assisting them in keeping motivated to reach their individual health goals. The simulation consists of two primary agents: guardians and children. Guardians provide support to their child through motivational messages, support, advice, and personal experience through aspects including achieving goals, diet, exercise, and psychological health [9]. The role

• f the child is to provide feedback to their guardian if they were positively affected by

the message sent to them. Each user can be both a guardian and a child, or either, depending on the network topology being employed. NetLogo is a multi-agent programming language and modelling environment, particularly useful in simulating social phenomena [10]. All agents can interact with each other and perform tasks concurrently, and the model is rendered in a graphical user interface. The simulation can be run in 3 different network topologies: Ring, point- to-point (paired buddies), or random.

Random Paired Ring Figure 1. Factors effecting user motivation & Positivity.

1.1. Agent Attributes It is critical in complex systems modelling to include the necessary variables or rules relevant to each agent. This bottom-up approach emphasises the need for individual agent simplicity, while producing complex social phenomena when they interact. Motivation plays a key role in setting and achieving goals, particularly in changing health behaviors [11]. All actions undertaken are driven by some form of motivation, which can be influenced by multiple factors such as the environment and relationships. This is particularly evident in theories of social influence, where one’s motivation to achieve their set goals comes from another individual [12]. Motivation for the individual to achieve health related tasks is influenced directly by messages received from their guardian. Positivity is influenced by one’s motivation, and closely reflects one’s mood, attitude, and emotions toward a stimulus. The theory of mood contagion elaborates on the ability of one’s mood in affecting another person’s cognitive expressions in a way that influences their motivation [13], as well suggesting that a leader with high positivity saw better performance from those they were prompting [14]. In the model, positivity represents the theoretical measure of how much influence the guardian has on their child.

Figure 2. Factors effecting user motivation & positivity.

1.2. Simulation Schedule User environment can affect motivation, simulating events occurring outside the system. If they have sufficient positivity, a message is composed by the guardian and the degree to which it affects the child depends on their level of positivity. Feedback is then sent from child to guardian. In addition, the increase in motivation seen when the user receives a message is transferred to positivity, where they now have an increased chance of sending their child a message. If the individual does not receive a message from their guardian, the individual’s lack of support causes their motivation to decrease. This is depicted in Figure 2. The basis of the simulation schedule and its functions lay in the foundational principles of emotion and mood contagion [14-16].

• 2. Research Design

At simulation setup, each user is assigned a guardian and an angel. The users are assigned motivation and positivity values with normal distribution to mimic real world variability throughout the population. Decay factor, Messaging factor, Transfer factor, and Feedback factor are held constant between topologies. The first step is a loss of user motivation, occurring if no message is received from their guardian. Where motivation in the previous step is 𝑛(𝑢 − 1), and fd is the decay of motivation factor, current motivation:

𝑛(𝑢) = 𝑛(𝑢 − 1) ∙ (1 − 𝑔

𝑒)

(1)

A guardians own positivity and external factors contribute to whether they send a motivational message to their child or not [4, 15]. If these conditions are satisfied, the guardian will send a message. If a message is sent, the child is affected by a factor of the guardian’s positivity of the previous step [14], with message effectiveness as 𝑔

𝑛.

This makes the child’s motivation 𝐷𝑛(𝑢):

𝐷𝑛(𝑢) = 𝐷𝑛(𝑢 − 1) + (𝐻𝑞(𝑢 − 1) ∙ 𝑔

𝑛)

(2)

Where 𝐷𝑛(𝑢 − 1) is the child’s motivation from the previous step. Once the message is received by the child, feedback is given to the guardian based on the child’s increased change in positivity. The guardian’s updated motivation is given by

𝐻𝑛(𝑢) = (𝐻𝑛(𝑢 − 1) + (𝐷𝑞(𝑢) − 𝐷𝑞(𝑢 − 1)) ∙ 𝑔

𝑔)

(3)

Where 𝐻𝑛(𝑢) is the guardian’s new motivation, 𝐻𝑛(𝑢 − 1) is the guardian’s motivation from the previous step, 𝐷𝑞(𝑢) is the child’s positivity after receiving the message, 𝐷𝑞(𝑢 − 1) is the child’s positivity prior to the message, and 𝑔

𝑔 is the feedback

factor variable [16]. Additionally, there is a transfer of an individual’s motivation into their own positivity. With transfer factor, 𝑢𝑔, all user’s positivity from the previous step, 𝑞(𝑢 − 1), the user’s new positivity, 𝑞(𝑢), is given by

𝑞(𝑢) = 𝑞(𝑢 − 1) ∙ (

(𝑛(𝑢)−𝑞(𝑢−1))∙𝑔

𝑢

𝑞(𝑢−1)

) + 1) (4)

The rate at which the individual’s positivity is influenced is dependent on the difference between their current motivation levels and previous positivity level. The greater the positive difference between one’s new motivation, 𝑛(𝑢), and previous positivity, the greater the influence on increasing their new positivity. Environment affects individuals motivation and represents external factors such as relationships, their work and home environment, or other healthcare they may be concurrently receiving [16]. To represent natural variability, a normal distribution is used, and has chance of effecting their motivation depending on the effect-of- environment slider in the user interface.

• 3. Results and Discussions

Each experiment was run for a total of 2000 ticks (1 tick represents 1 day), with 40

• users. Each network topology was simulated and repeated 500 times. Optimal

performance could be attained by maximising group level motivation and positivity while exhibiting the lowest variation between nodes. From a utilitarian perspective, this multi-goal objective function represents the greatest benefit for majority of users. The random network aimlessly generates at least one link for each user, with the total number of links in the model equaling two times the number of users. This however does not guarantee every user has both a guardian and a child. In the simulation of the random network, users on average experienced an increase in motivation by 944, and an increase in positivity by 409. While this is a significant increase, as a result of the unequal degree of connections between users, the data is significantly skewed to those with a greater number of guardians. These users experience substantial increases in motivation and positivity, while others suffer a considerable decrease. This is reflected by the large standard deviation of 3036.34 of motivation, and 1329.35 of positivity at the end of the trial simulation. The ring topology creates one outgoing link and one incoming link for each agent, each connecting to a different user. As each user in the ring network is connected to each-other in series, there is a flow-on effect of one’s motivation and positivity whereby everyone can influence another individual along the chain. This creates a stable and equitable positive result for majority of users, which can be seen in Figure 3(c). After 2000 ticks, motivation increased to 119.4 (StDev 171.87) and positivity to 79.6 (StDev 95.99). The paired buddy, or point-to-point network represents a single two-way connection between two nodes [17]. The model creates a two-way connection between two users of the system as if they were in a buddy relationship. As seen in Figure 3(b), there was an overall increase in users positivity by 180 and motivation by 329. This however, does not accurately reflect the each individual’s positivity and motivation as noted by the high standard deviation at the end of the trial (motivation = 555.82, positivity = 305.93). This highlights the lack of interconnectivity between nodes compared to the ring network. Users with inherantly low motivation are isolated, and the lack of flow on effect experienced in the ring network makes it difficult for such users to increase their positivity and motivation. On the other hand, users with inherantly high motivation benefit exponentially from a buddy style topology through constant recpricated messages delivered to and from their buddy.

(a) Random Network (b) Paired Buddy

(c) Ring Topology (d) Comparison of StDev between networks Figure 3. Simulation Results

The standard deviations of each network displayed in Figure 3(d) highlights how different network models can significantly affect equal distribution of resources to individuals participating in social type interventions for chronic disease. Although the random network (Figure 3(a)) provided the greatest mean positivity and motivation improvement for the group compared to the other network models, the comparatively massive increase in standard deviation indicates that few experienced great benefit while others did not or were negatively impacted. The ring network on the other hand provided the least mean improvement overall, however achieved the lowest standard deviation indicating a greater number of individuals were able to benefit from the guardian angel type intervention.

• 4. Summary and Limitations

By looking at the 3 network topologies when simulated in a social health context it is clear that all may provide benefit to the user. All simulated trial groups experienced a mean gain in the outcomes of positivity and motivation, however this does not reciprocate to all individuals benefiting from the intervention. Mean values of health

• utcomes do not accurately reflect disparities or distribution of resources throughout

the users in each system. This highlights the need for developing grouping protocols when employing such interventions to ensure those who are inherently less motivated are connected with others in such a way that they can benefit from others motivation and positivity. Of the 3 models simulated, the ring offers the most consistent benefit to everyone as indicated by the lowest standard deviation. The standard deviation is still relatively large however and other common social network topologies should be explored in the future. Additionally, insight from these findings can be used to develop the basis for a clinical trial using a Guardian Angel type messaging system for patients with similar disease, giving the concept further descriptive backing and quantitative evidence for deployment in a healthcare setting. This data can then be used to evaluate the model. A limitation of the model is that although there is sufficient normative backing through theory, there is limited empirical evidence relating to such an innovation. As a result, there are multiple assumptions that must be made including how individuals would use and interact with the eGuardian Angel system, and how users would be influenced by the intervention. In turn this means that the model cannot be empirically

validated, however it can be used as a theoretical model in displaying emerging patterns described in the normative literature of social influence and mood contagion. The model also does not properly take into account people disuse of health related application interventions over time, resulting in inactive and loss of users. References

[1] AIHW, How many medical practitioners are there?, (2015). [2] K. Henry, The Economic Impact of Australia’s Aging Population, SAIS Review 24 (2004), 81–92. [3] P. Szolovits, J. Doyle, W. J. Long, I. Kohane, and S. G. Pauker, Guardian Angel : Health Information Systems, (1994). [4] M. Heisler and J. Piette, “I Help You, and You Help Me”: Facilitated Telephone Peer Support Among Patients With Diabetes, The Diabetes Educator 31 (2005), 869–879. [5] C. Caburnay et al., Evaluating Diabetes Mobile Applications for Health Literate Designs and Functionality, 2014, Prev Chronic Dis 12 (2015), 1–13. [6] A. A. Lazakidou, S. Zimeras, and D. Iliopoulou, mHealth Ecosystems and Social Networks in Healthcare, Springer International, Switzerland, 2016. [7] M. Tsouri, K. Alamantariotou, O. Pediaditaki, V. Harizopoulou, and G. Kontosorou, How Knowledge Flows Through Social Networks and Communities of Practice in a Healthcare Project, in mHealth Ecosystems and Social Networks in Healthcare, Springer, Switzerland, 2016, 11,. [8] N. Christakis and J. Fowler, Connected: The Surprising Power of Our Social Networks and How They Shape Our Lives—How Your Friends’ Friends’ Friends Affect Everything You Feel, Think, and Do, Black Bay Books, New York, 2011. [9] T. Chomutare, E. Rsand, and G. Hartvigsen, Mobile peer support in diabetes, Studies in Health Technology and Informatics 169 (2011), 48–52. [10] M. Swan et al., Sharing Health Data for Better Outcomes on PatientsLikeMe, Journal of Medical Internet Research 12 (2010). [11] S. Tisue and U. Wilensky, Netlogo: A simple environment for modeling complexity, Conference on Complex Systems (2004), 1–10. [12] A. Cohen et al., Framed, Interactive Theory-Driven Texting, Family & Community Health 40 (2017), 43–51. [13] R. Cialdini and N. Goldstein, Social Influence: Compliance and Conformity, Annual Review of Psychology 55 (2004), 591–621. [14] R. Neumann and F. Strack, “Mood contagion”: The automatic transfer of mood between persons., Journal of Personality and Social Psychology 79 (2000), 211–223. [15] S. K. Johnson, Do you feel what I feel? Mood contagion and leadership outcomes, Leadership Quarterly 20 (2009), 814–827. [16] S. K. Johnson, I second that emotion: Effects of emotional contagion and affect at work on leader and follower outcomes, Leadership Quarterly 19 (2008), 1–19. [17] M. Heisler, S. Vijan, F. Makki, and J. Piette, Diabetes Control With Reciprocal Peer Support Versus Nurse Care Management, Annals of Internal Medicine 153 (2010), 507–516.