Understanding Older Adult Use of Data Visualizations as a Resource for Maintaining Health and Wellness

Informatics as a Resource to Support Aging

The older adult demographic group, those at least 60 years old, is growing at an unprecedented rate, and by 2030, this group is expected to represent 25% of the U.S. population (Administration on Aging, 2014). Due to the increased prevalence of chronic diseases and conditions associated with the aging process, the cost of providing health care to an older adult is 3 to 5 times higher than the general population (Administration on Aging, n.d.). Within this context, research in biomedical informatics and gerontology can provide valuable contributions toward developing tools to help older adults maintain well-being and independence.

One approach is through the integration of data from health monitoring technologies. Efforts in this area have the potential to provide near real-time and continuous collection of health data with greater accessibility compared with episodic assessments at clinic visits (Jimison, Pavel, McKanna, & Pavel, 2004; Thompson et al., 2011). These in-home health monitoring tools include wireless physiological assessment devices, motion sensors, embedded infrastructure sensors, and cognitive assessments embedded in computer software. Although there are some potential concerns raised over the obtrusiveness of in-home health monitoring technologies, prior research by Wild et al. (2008) and Demiris et al. (2004) have found that older adults recognize the potential applications of health monitoring tools and accept them when the perceived utility of the monitoring technology exceeds concerns over privacy.

Data Visualizations for Older Adults

Health monitoring technologies make it possible to track health and wellness continuously, unobtrusively, and reliably while providing stakeholders with quantifiable feedback. A significant challenge toward promoting these value propositions lies in demonstrating tangible insights gathered from health monitoring data. It is not enough to have the infrastructure and technology in place to collect data. Appropriately designed visualizations can help bridge the gap between data and information, especially for older adults who may not be familiar with the detailed data sets generated. A well-designed visualization that translates data from health monitoring technologies into a consumable medium for older adults can promote active engagement in health care and further communication among members of the care team.

There is a limited body of literature on data visualization research with a specific older adult consumer focus. Mynatt et al. highlight the importance of maintaining an awareness of long-term health and well-being for older adults (Mynatt, Essa, & Rogers, 2000); this is operationalized through the development of a digital family portrait (Mynatt et al., 2000; Mynatt, Rowan, Craighill, & Jacobs, 2001). The portrait is a visualization of sensor activity collected from a smart home environment with different icons bordering the digital picture frame representing activities of daily life (Mynatt et al., 2001). The authors evaluated the visualization through a field trial with an older adult/family member dyad, finding that the initial designs were too complex and ambitious in conveying 10 levels of information. However, the authors did find that changes in the digital portrait initiated conversation between the older adult and family members (Mynatt et al., 2001).

Reeder et al. (2014) describe the design of sensor visualizations from a 6-month pilot study with older adults. The authors found that visual displays of sensor data were useful for caregivers of older adults experiencing cognitive decline. From the older adult perspective, visualization of sensor data was useful for consultations with their health care providers about activity levels. Reeder et al. integrated the feedback from the semistructured interviews in conjunction with design principles in data visualization to develop three visual displays of sensor data.

Le et al. focus on design of data visualizations for integrated health and wellness, applying Dunn’s wellness model to categorize data collected into cognitive, physiological, social, and spiritual well-being (Dunn, 1959; Le, Wilamowska, Demiris, & Thompson, 2012). This visualization was evaluated with both older adults and health care providers in separate focus groups (Le, Reeder, Thompson, & Demiris, 2013; Le, Reeder, et al., 2015). The authors found that at a high level, older adults identified with the value of having visualizations as a health assessment tool (Le, Reeder, et al., 2015). However, the focus of the authors’ work was on understanding the processes involved in analyzing visualizations as opposed to the utility of the visualizations.

Our work complements the existing literature by examining the perceived utility of an integrated wellness visualization tool. We extended visualizations outside of the sensor focus described by Reeder et al. (2014) and toward integrated wellness collected from multiple data sources. We focused on interactive graphical visualizations as opposed to metaphorical representations as described by Mynatt et al. (2001) to allow for more exploration within the visualization. The exploration process allowed older adult consumers to better conceptualize how they would utilize the visualizations over the course of the evaluation session.

Older Adult Design Guidelines

When designed appropriately, data visualizations can support consumers for both health assessment and shared decision making. However, the older adult population represents a unique demographic group with different design considerations than the general population due to aging associated differences in vision, cognition, and motor control (Charness, Demiris, & Krupinski, 2011). Existing design guidelines for older adults focus primarily on websites, emphasizing how information should be presented to facilitate discoverability and navigation (Demiris, Finkelstein, & Speedie, 2001; National Institute of Aging & National Library of Medicine, 2002; Redish & Chisnell, 2004). Although there exists overlap, design guidelines for data visualizations represent a different set of information needs with an emphasis on supporting effective comparisons and identification of trends as opposed to design guidelines of webpages (focusing more on discovering and navigating through content). In addition, there has been limited work examining the health visualization needs of older adults. We address issues related to utility of health visualizations through a qualitative analysis of interviews conducted with older adults. Having this focus on utility within the design framework is a valuable resource toward encouraging the use of a system.

Prior Work

We iterated on designs of health visualizations with older adults through a user-centered approach (Norman & Draper, 1986). The visualizations were based on Dunn’s (1959) framework of high-level wellness, operationalized further by Hoyman (1975) into dimensions of physiological, social, spiritual, and cognitive health. Dunn’s framework has been tested and applied within gerontology across multiple studies (Demiris, Thompson, Reeder, Wilamowska, & Zaslavsky, 2013; Kleffel, 1991; Miller, 2014). Older adults have also been receptive toward the framework of holistic wellness (Thompson et al., 2011). We presented early visualization mock-ups to older adults through focus groups (Le, Reeder, et al., 2015) and validated older adults’ preferences for abstracted data into an integrated measure of wellness (Le, Reeder, et al., 2015). Based on the findings from the focus groups, we refined the visualizations into higher fidelity interactive prototypes. The visualizations consisted of bar and line charts representing simulated wellness data (Le, Thompson, & Demiris, 2015). Interactions allowed users to view wellness at differing levels of granularity while principles of brushing and linking were applied to connect multiple views of data (Buja, McDonald, Michalak, & Stuetzle, 1991). The visualizations also allowed users to set the weights for each of the varying components of wellness. This was informed by earlier research from focus groups where older adults indicated the utility of breaking wellness into spiritual, social, cognitive, and physiological components, but had varying individual importance placed on the components (Le, Reeder, et al., 2015). For this work, we focused less on the design and evaluation of visualizations and instead we report findings related to the use of health visualizations from an older adult consumer perspective. In particular, we wanted to understand (a) do older adults identify utility in visualizing their wellness data? (b) what are the different use cases for the visualizations? and (c) what are the barriers limiting the use of the visualizations? This article presents these findings and also provides guidelines that support other researchers and designers in the development of health visualizations for older adults.

Semistructured Interview and Evaluation Sessions

We conducted one-on-one semistructured interviews with 21 older adults to obtain their perspectives on use of health visualizations. During this interview, we presented participants with two interactive visualizations developed over the course of iterative user-centered design (see the appendix). We described the design and evaluation of the visualizations as part of prior work in Le, Thompson, and Demiris (2015). We prompted participants during the open exploration phase of the sessions with questions such as the following: What are the different uses for the visualization? Are there barriers to using the visualizations? and What are your thoughts on the information presented through the visualizations? All sessions were audio-recorded for later transcription.

Participants and Setting

We recruited participants for the sessions through contact with independent living facilities and older adult community centers throughout the Seattle, Washington, area. Recruitment methods included distribution of flyers on community boards, postings within newsletters, and snowball sampling. Sessions took place at the community center or a common room within the housing facility, lasting at most 90 min. We presented participants with a US$15 gift card at the completion of the session. All participants had to be at least 60 years old, English-speaking, and willing to participate through informed consent. The University of Washington Institutional Review Board (IRB) approved all procedures involving human subjects (IRB No. 47324).

Qualitative Analysis

We transcribed the sessions verbatim, removing any potential personal identifiers. T.L. conducted an initial high-level review of the transcripts, coding for any reference to use and utility of the visualizations. T.L., N.-C.C., and S.C. then completed an affinity mapping exercise to aggregate coded excerpts into themes. N.-C.C. and S.C. did not previously participate in the study and helped provide an unbiased validation of the initial coding. Affinity mapping is an inductive process that organizes ideas and concepts into themes (Martin, 2012). Each rater independently went through the excerpts to form an initial high-level schema. We then printed excerpts generated from the initial high-level review onto separate notecards. We displayed these notecards on a table and collaboratively reorganized the notecards based on content, grouping together similar excerpts. The process came to a conclusion when all excerpts had been grouped. We then labeled the groups based on their thematic content. Over the course of the affinity mapping, the initial high-level codes were also validated collaboratively, resulting in the exclusion of certain transcript excerpts that did not fit the code of utility. We reviewed the categorizations generated from affinity mapping separately at the conclusion of the exercise and reconciled differences as a group. Affinity mapping allowed us to generate themes associated with use and utility of health visualizations through a bottom-up, inductive approach emergent from the raw data of participant interviews.

Use of Visualization as an Intervention

Participants found that the visualizations could be used primarily as an intervention tool especially for correcting longitudinal decline in health. As an example, participants commented that detecting changes such as cognitive decline was an important use case. These changes would not be as easily noticeable until they manifested themselves in an adverse event; therefore, the visualizations allowed participants to step back and identify earlier patterns or trajectories over time. Detecting differences was only half the story for participants; the visualization provided greater utility if there was some guidance to redirect the course of decline in health:

I think I want to see more information like these are the things you can do to improve your cognitive abilities, these are the things you can do to improve your physiological, each one of these things I would like to see that on a separate page somewhere. That would be more helpful to me rather than just looking at this and not knowing what’s causing the things to go up or down or to change. (P18)

Participants were also positive about sharing the visualization with family members and health care providers primarily as a resource for intervention:

There were other times where it went down significantly. And those would be if the reports were sent to a caregiver or to a family member they would know that it might be appropriate to intervene. (P01)

However, for personal use of the visualization as an intervention, participants were mixed in response. There was a strong perception that, either individually participants already had a strong recognition of their health and wellness, or that if there was a significant decline, especially in cognitive health, there would be challenges in interpreting the visualization. In either case, personal use of the visualization as an intervention tool would not be as helpful for participants.

It [the visualization] wouldn’t be relevant to me because I kind of know. I know how my social thing is. I’m going. I know spiritual, I know my psychological, and I know myself. (P07)

Well, from the looks of it, Jeffrey has deteriorated in all areas so how well he would be able to understand this graphical concept; I just don’t think it would be really there for him. (P03)

Use of Contextual Information to Improve Utility

Participants found that a significant challenge of using the visualizations was the lack of contextual information. They were able to identify differences in patterns within the visualization such as longitudinal changes or certain drop-offs in wellness. This prompted discussions hypothesizing potential causes associated with the changes. However, participants noted that this was all based on inference and they lacked the contextual information to understand what prompted the differences within the graph.

I think once you get used to it, you can pick up a lot of information just glancing at it. And maybe a few clicks. Every now and then I think it would be useful to take to a deeper level to see if you can see patterns. Of course you’re not just mashing what’s there, you’re mashing that with what you know about your life. I think it’s a very large part of how it would be useful. (P09)

As a result, participants found that the visualizations provided a limited understanding of the data. The contextual information, either annotated within the visualization or through conversations with the participant, helped answer questions about why certain patterns appeared.

Perceived Limitations Due to Computer Literacy

Although participants successfully interacted with the visualizations while exploring and completing usability tasks, they expressed uncertainty about making full use of the system due to perceived limitations in computer literacy. For example, participants caveated their exploration process by indicating they used computers for limited tasks such as browsing the Internet or checking their email. Although this did not impede their ability to interact with the visualization, the perception of computer literacy impacted potential use of the visualization. As a participant noted,

Imagine if most people would learn to use this very quickly? Well, most people are more computer literate than I am, I think . . . I would say people with computer skills would learn to use this of course. And the more competent they are, the more quickly they would learn it. (P13)

Participants also commented that a lack of access to computers would limit the potential use of the visualizations. As a result, alternative mediums to present the information would be valuable complements to the visualizations.

I think almost everybody that I know, educated or not, a printout would do just fine. There’s no way . . . most of us aren’t computer savvy anyways, including myself . . . It would be more for a year period of time, maybe a half year period of time or maybe periodically, get a printout like this. (P07)

Time Frame for Representing Data

Participants indicated that visualizations of health data had optimal time frames for representation. Day-to-day visualizations had limited use, primarily because of the high variability of the measures. As one participant noted,

Day-to-day, well, I mean I looked at it day-to-day I might go a couple of days to see, because you go day-to-day you’re going to have days when you’re down and days when you’re up. So I don’t know if I would want to study it that much day-to-day. (P14)

Instead, participants indicated that visualizations were most useful at monthly or yearly intervals. This aligned with their use of the visualizations as a resource to monitor longitudinal changes in health and wellness. It was important for participants to view the visualizations over a long enough period of time that they were able to make an assessment on what constituted a normative pattern of wellness. Shorter time intervals limited the contextual view of the wellness, making it difficult for participants to assess patterns.

. . . from certain date to date why would anybody want to know that anyways, particularly if you’ve got a graph that’s showing you the dates from 2011 to 2012. You’ve got this to this, that’s one year. Those are important and you can just use that . . . I mean a year’s time would be plenty. In other words looking at these graphs would be plenty. You wouldn’t need to do anything. You could note that okay I went down, it went down in any of them except social and that went up . . . (P07)

Sharing of Health Information

Participants were receptive toward sharing the visualizations with other stakeholders, in particular health care providers. Participants envisioned the visualizations as a supplement to their conversation with health care providers:

I think it could help them [health care provider] perhaps come to an understanding of what really is underlying the problems. Because so many medical problems, as you know and I know, have an emotional for want of another word, an emotional background, cause. Many times it’s, why do you really have a headache and why is this really bothering you. And sometimes it’d be good to understand that you’re worried, they could see that their . . . patient [is] suffering with problems with their cognitive skills they could act, whereas they may not have noticed it. I think a good practitioner would have known it already but possibly didn’t and, so I think they could use it. (P10)

However, in contrast to the open communication with health care providers, participants were less willing to share information with family members and expressed concerns with privacy. Although participants recognized the utility of sharing the information with family members—for example, as an alert to monitor changes in health—participants wanted to maintain more control over what components of the visualization their family members were able to see. From our sessions, participants highlighted the tension of two values: sharing information across stakeholder groups and maintaining privacy. Their variability was based on individual preference however one approach toward maintaining this balance is to provide older adults with control, specifying how much detail should be shared to each group.

Limitations

There are limitations to consider when examining the results of this study. We applied a framework for aggregating scores through a weighted linear model. Other methods of inferring relative utility weights may have improved precision; future testing using these methods would be useful. The visualizations contained simulated data which limit personal investment in the outcome. Participants indicated an existing confidence and awareness in their well-being, though that may not be reflected when new trends are noted within their wellness visualization. This is an ideal next step, conducting a pilot study where participants are able to review personal health visualizations generated from a wellness monitoring environment. We recruited participants by convenience, and as a result, the sample was not representative of the older adult population. We were also limited by self-selection bias as those who chose to take part in the study also had a general interest in health visualizations. However, given the breadth of responses and the relatively unique focus on this population, we believe the findings still provide useful insight that can inform follow-up studies with larger and more diverse sample sizes.

Emphasize High-Level Visualization of Health Data

Participants indicated that there was greater utility in visualizing health data over extended periods of time. Visualizations should support this objective by providing overview of health information over a long enough period that participants would be able to assess baseline levels and variations from the norm. This approach toward visualization is not new; Shneiderman (1996) has emphasized this in his mantra of information visualization: “Overview first, zoom and filter, then details-on-demand.”(p. 336) However, we have found that though providing an overview is essential, there was less utility in providing zoomed views of the data. This was because, within the context of wellness assessment, participants found day-to-day variability subject to multiple factors that limited any conclusions drawn from the visualization.

Provide Annotations Within the Visualization

Although participants recognized that there was variability day-to-day, they indicated that contextual information would be a valuable resource in explaining trends in the data. Allowing annotations within the visualizations would help contextualize the data, serving as a personal reminder for participants while also informing family members or health care providers. The visualizations allowed participants to extract information related to trends and patterns in wellness; however, the values were still an abstraction. Participations commented on how this provided little information as to why there was an increase or decrease in the data. Different approaches to providing contextual information could include automatically populating it from external sources such as electronic health records or digital diaries in addition to manual annotations.

Integrate Health Promotion Recommendations

Participants found that though the visualizations showed changes in health and wellness, it left them unclear about next actionable steps, for example, to reverse a declining trend in health. For family members and health care providers, this information was useful to instigate a conversation and intervene if changes were noted (Huh, Le, Reeder, Thompson, & Demiris, 2013; Le, Reeder, et al., 2015). However, participants had trouble translating the information from the visualizations into action. One approach to alleviate this is by providing recommendations to improve wellness based on trends within the visualization. The recommendations allow participants to take a proactive approach, allowing them to ground the abstract visualizations with actionable items. This concept of self-efficacy within health promotion has been shown to be an essential predictor of both behavior and health outcome (Bandura, 1977; Holloway & Watson, 2002; Strecher, DeVellis, Becker, & Rosenstock, 1986). Integrating health promotion recommendations shifts the paradigm of health visualizations toward a data-to-information-to-action approach that older adults find more impactful and of greater value.

Allow Print Media Access to Visualizations

Although older adults are steadily growing in their adoption of technology, as a demographic group they are still far below the national average in the United States (Pew Research Center, 2014). Internet adoption is at a current high of 59% for older adults; however, within this population are two distinct profiles. Younger, higher income, and more educated older adults are comfortable with technology and have positive views toward its benefits. The older and less affluent demographic group is more disconnected with technology and generally has a skeptical attitude toward the benefits of technology (Pew Research Center, 2014). Participants within our sessions expressed concerns about access to the visualizations, especially for themselves or others whom they observed as less comfortable with using computers. Developing visualizations that can be distributed as a printed medium can address some of the challenges associated with technology adoption while also opening up access to a broader audience.

Support Custom Sharing of the Visualizations

We found that participants identified different levels of comfort when sharing the visualizations with others. There was a strong use case for sharing the visualizations with both family members and health care providers. Given the sensitive nature of health data, it was important that participants were allowed to control who was able to view the visualization and what level of detail to present the visualization. This differed depending on the stakeholders. The level of sharing could also be adjusted over time depending on the changing needs of the older adult.