“It’s Not as Simple as Just Looking at One Chart”: A Qualitative Study Exploring Clinician’s Opinions on Various Visualisation Strategies to Represent Longitudinal Actigraphy Data

Wearable technology is increasingly used in healthcare to support the management of chronic conditions [1, 2]. The use of research-grade wearables (e.g., Actigraph) to measure endpoints in clinical trials [2-4], and the rise of consumer wearable companies such as Fitbit [5], demonstrates the acceptance of this technology in the wider community. Despite their popularity, a lack of integration into clinical practice calls into question the potential value offered by these devices to patients and clinicians. It is suggested that unless wearables offer useful insights and outcomes, and not simply data, their promise will not be realised [6]. To be fully integrated into clinical environments, the data from wearables must be simple to access, use and understand [7-9]. Put simply, it must enhance the clinical decision-making capability of the clinician, or empower the patient, to be more engaged in self-management. Thus, the manner in which raw data is processed and presented is critical to the success of these devices within this context [8, 10].

Longitudinal actigraphy data can be presented in two ways: (i) summary metrics of total daily activity, steps or sleep and (ii) a more granular approach whereby within-day physical activity (PA) and sleep data is represented over time. However, a reliance on summary metrics remains, driven by the approach used by consumer wearable companies (e.g., Fitbit, Apple etc.). Clinically, this fails to provide information about the temporal nature of activity. In other words, it provides an understanding of how much people move or sleep, but not how this is dispersed within days, weeks or months. Recent research suggests that more complex methods of data processing may provide more meaningful clinical insights [11], while users may value further detail than that which is currently provided by many devices [12]. More granular visual representations of data may offer the ability to quickly identify patterns and understand behaviour, therefore informing clinical decision-making and driving increased use of wearables in practice [10, 13].

There has been little research exploring the perceived value of commonly employed visual representations of the data from wearable devices [14]. A study in 2012 presented ways in which longitudinal data could be visually represented. However, graphs were limited to summary level information and the intended users’ perceptions were not investigated [7]. To best support the implementation of wearables in practice, it is important to trial advancements on those who will be using them. Clinicians possess the knowledge required to identify important patterns for decision-making [15]. Thus, it is critical to understand whether they perceive any outputs derived from these devices to be clinically useful, whether they can accurately interpret visualisations independently, and what insights they gain from them. Therefore, this study developed a variety of visualisations of longitudinal actigraphy data, both “state of the art” as commonly seen in current commercial devices, and more granular, temporal representations of the same data. The aim of this study was to explore clinicians’ opinions of these data visualisations to determine their perceived clinical utility.

Ethical approval for this cross-sectional, qualitative study was granted by the Human Ethics Board of the local research institution.

The visualisations were created using activity count data (a threshold-based proxy of movement) previously collected from arthritis patients using an Actigraph GT9X Link (Actigraph LLC), as reported elsewhere [11, 16, 17]. Arthritis patients wore the Actigraph during their usual activities for a period of 28 days. Raw data were pre-processed and 60-s epoch activity counts generated. Study days were converted to data analysis days (DADs). A DAD starts at 18:00.000 and ends at 17:59:000 the following day. The absolute values of activity count for each epoch were converted into 4 PA categories: inactive, light, moderate and vigorous [18].

Three participants were selected to provide a comparison for how different activity and pain profiles may present across arthritis populations. Four visualisations were created for each patient; a stacked bar chart (Fig. 1), a cluster bar chart (Fig. 2), a linear heatmap (Fig. 3) and a radial heatmap (Fig. 4). The 4 graphs present identical Actigraph activity count data for each patient across a 28-day period. The visualisations were created in Python using the Matplotlib, HoloView and Seaborn libraries. DADs were labelled with “date day” format where the date was the date in the DAD which covers the morning time. In all but the radial heatmap (Fig. 4), weekdays and weekends were separated to facilitate interpretation of daily activity patterns. In the stacked and cluster bar charts (Fig. 1, 2), total activity for each day was shown, separated into the different PA categories. This represents the current “state of the art” provided by consumer devices such as Fitbit [19] and Apple Watch [20]. A monochromatic ordinal colour scheme from ColorBrewer [21] was used to indicate the different categories. The linear and radial heatmaps portrayed continuous activity count data, facilitating comparison of the temporal patterns of PA within and between days. This approach is not commonly used by consumer devices. Each cell represented the total activity count for a 15-min epoch. A monochromatic sequential colour scheme [21], was used to indicate the activity level in that epoch. To prevent a small number of very high values saturating the scale, colours were mapped to the square root of the activity counts.

Fourteen clinicians were recruited using purposive convenience sampling across 2 hospitals in Ireland (Table 1). The lead clinician of the medical team in each hospital was contacted through email. All interested clinicians were invited to take part in a focus group within their hospital at a suitable time and provided written informed consent to participate.

Focus groups were facilitated by 2 post-doctoral researchers (1 female [A.K.] and 1 male [W.J.]), both of whom were physiotherapists specialising in digital health. Alison Keogh previously published research in qualitative methods and analysis and William Johnston has extensive experience with data derived from wearable devices. Each focus group took 60 min to complete and was audio-recorded and transcribed verbatim by Alison Keogh. No follow-up interviews or clarifications took place.

The focus group explored clinician’s experiences regarding patient-generated activity data (online suppl. File 1, see www.karger.com/doi/10.1159/000512044). Clinicians were provided with the 4 visualisations from each of the arthritis patients, along with a brief clinical history detailing the sex, age, condition, average pain and stiffness and the SF-36 physical and mental domain scores, for each patient profile. Clinicians were asked for their insights into each visualisation independently, including whether they were useful, what clinical understanding could be derived from them, and what further information, if any, they required to support their clinical judgement. As the focus group progressed, comparisons between the visualisations were encouraged.

An inductive, thematic analysis was undertaken using a realist approach whereby the researcher assumed that the opinions of the participants reflected their true perceptions and should be taken as real [22]. Transcripts were analysed by Alison Keogh and William Johnston following the suggested protocol of Braun and Clarke [22]. Both researchers familiarised themselves with the data, generated a list of initial codes relevant to the research question, and refined these codes by grouping them into potential themes. This list was continually reviewed until agreement was reached.

Three major themes were identified: (1) the importance of context, (2) interpreting the visualisations and, (3) applying visualisations to clinical practice. Supporting quotations are provided in Table 2.

None of the clinicians routinely used this type of data with patients. Clinicians extensively reported that, irrespective of the visualisation used, the data alone were not enough to provide clear clinical insights without also understanding the context in which it took place. Context was multi-faceted including wanting to know participants daily levels of pain/stiffness, the type of activity they were engaged in, their employment status, and so on. Without this information, the relevance of the data was questioned, although it was acknowledged that it could be used as an additional tool to drive assessments, evaluate treatment effectiveness and act as a benchmark for patient progress.

• “It’s hard to read into that because we don’t know how they’ve done those steps or what pace they’ve done them or what intensity, whether they’re done them all together or whether it’s just pottering about as the day went on.” [Physiotherapist 1, female]

• “This would be another great thing to have in addition to you know, asking the patient if they have swollen joints, tender joints, eh asking them are they more active or not, early morning stiffness, but this might actually give us a more objective view of if they’re being correct.”[Consultant 2, male]

Linked to this, was a desire to understand how patients compared to other people. There were mixed opinions regarding whether symptoms of pain or stiffness would be expected to correlate with PA patterns; therefore, there was a need to understand how patients were progressing both individually and compared to the general population.

• “This isn’t controlled for anything, this is just a man over a 4-week period and we don’t know much about him. But if we knew what he had been like 1 day and maybe after an intervention that’s where it becomes useful at an individual level but at a population level this doesn’t mean anything to me unless I know what the other people doing his work at his age in life, what they’re doing.” [Consultant 1,male]

There was widespread agreement about the usefulness of separating periods of sleep from activity data (e.g., within Fig. 3, 4). Physiotherapists valued the ability to view a full day temporally, which was somewhat supported by occupational therapists but not by doctors, who felt this level of detail was not important.

• “I suppose the timing of when they’re active isn’t of interest automatically but it would be more in the context of how they were feeling. You know say we put them on a new tablet and we could see that their evening exercise had increased because of the medication you know to be relevant to us it would need to be in the context of their treatment or what they are doing.”[Registrar 1,female]

Figures 1 and 2 were considered easy to interpret as clinicians could quickly identify the categories of activity a patient was spending time in, compared to the gradient approach provided in Figures 3 and 4. Many clinicians required some form of support from the researchers to fully interpret Figures 3 and 4. Indeed, 1 clinician responded viscerally to Figure 3, remarking that “This makes me feel sick.” It was also noted that Figure 4 was more difficult to interpret as it was not possible to separate days, while the nature of the visualisation suggested that the days on the inside of the dial were less significant than those in the outer rings.

• “Unless you’re used to looking at these type of things actually, just in different types of molecular biology you get these different types of graphs and eh, I always have to stand back and think about it before figuring out how to interpret it.” [Consultant 2,male, referring to Figure 3]

There was disagreement between physiotherapists and the other members of the team regarding which visualisation they would use clinically. Specifically, physiotherapists preferred the detail provided to them in Figures 3 or 4. All other clinicians favoured Figure 1, due to its simplicity. Nonetheless, it was acknowledged that Figure 1 still required amendments to enhance its utility, specifically that sleep should be separated from other activity, that summary statistics should be provided alongside the graph and that reference data should be included for comparisons.

• “I get more from this one definitely. I just I feel I have a better understanding of their 24-h pattern.” [Physiotherapist 3, male, referring to Figure 3]

• “Because it’s easier on the eye. It contains the same data, I mean they all contain the same data but that’s readable whereas the others are harder to read. The second one the bars are too narrow and the active parts are almost impossible to see. And then the third and fourth ones are, they contain a lot of qualitative information which you know may be useful from an occupational context but not in a medical context.” [Consultant 1, male, referring to all Figures]

Currently, clinicians rely on patients to provide them with their own data through their own, personally purchased devices. Thus, a range of barriers were noted including the digital literacy of patients, a lack of availability of these devices, the cost of implementing them, an inability to merge the outputs with patient health records, and, overwhelmingly for doctors, a lack of time. In general, allied health professionals desired greater detail than doctors; thus, how visualisations are used in practice will depend on which member of the team they are presented to.

• “Not within the boundaries of a 15-min medical consultation. It’s just not something medically that we would practically be able to engage with. It might be different you know for longer consultations, maybe OT?” [Consultant 1, male]

• “I think in terms of managing fatigue and pacing it would be a really good visual for a patient to say look we talked about pacing, you’re clearly highly active day on day and you’re still saying you’re fatigued. Might this be why?” [Occupational Therapist 2, female]

Nonetheless, clinicians could see the potential value in the information provided by wearable devices. Cohorts that would benefit from this type of data were highlighted, including those with lifestyle-related conditions such as diabetes. Furthermore, it was suggested that an interactive method of selecting which data could be viewed would be beneficial (Table 2). However, for now, it was noted that this would not be suitable for all patients, and thus clinical judgement would be required to decide where and how they were implemented.

• “I think seeing is believing for a lot of people. I think like XX was saying you might psychologically have had a few really crappy months but if it’s here and it’s there to show you well actually look at this, this says otherwise and it’s an objective viewpoint so it does really help reinforce.” [Occupational therapist 1, female]

This study explored clinician’s opinions on a range of visualisation strategies using actigraphy data. Clinicians provided suggestions for how visualisations could be improved. However, what also emerged was a broader picture of the structural barriers and clinical parameters that need to be considered before the use of wearables becomes widespread in clinical practice. Specifically, the current reliance on consumer devices which target the needs of the general population, limits the potential clinical insights that can be derived from their data, due to the lack of context and individualisation provided by them.

Irrespective of the visualisation, clinicians would have required further information to use them in a clinically meaningful manner. The need for patient-specific insights is not new [9, 23, 24]; however, it remains complex as what is required for one patient may not be required for all [25, 26]. In theory, wearables should offer a greater understanding of context than currently used outcome measures [5, 10, 27, 28]. In practice however, clinicians still deemed this approach as too ambiguous, as the multifaceted nature of chronic conditions requires an understanding of numerous aspects of patient’s lives [25]. Central to the need for context was a hypothesis that additional information would help determine whether relationships between health parameters exist. Establishing correlations is an important aspect of evaluating patient progress [15]; however, recent research suggests that reported pain outcomes and PA do not correlate [11]. Though clinicians were aware of this, they nonetheless felt that it would encourage them to use the actigraphy data if they could judge it for themselves.

For visualisations to be implemented, they need to meet clinical needs [10]; however, needs may differ substantially between medical team members. The perceived usefulness of each visualisation relied on who was using it and what information they sought from it. Thus, no single visualisation can be universally better than the other, as the individual data needs of each team member will be different. The role of allied health professionals better supports discussions around lifestyle factors [23, 29] as seen by their desire for more subjective interpretations of data compared to doctors. In contrast, current healthcare structures require doctors to focus on validated endpoints to assess patients swiftly in the limited time available to them. In the absence of a readily interpretable measure linked to PA [30], the barrier of time severely limits doctors’ use of longitudinal actigraphy data [15, 31].

The results of this study support those of previous research which suggest that participants favour data that is presented in a familiar format and yet desire multi-faceted filtering [15]. It has been suggested that “in an ideal world, end-users would be able to select the data visualisations that best suit their needs” [10], a need further endorsed by clinicians in this study. It is clear however that the current “state-of-the-art” method used by activity trackers, particularly consumer devices, of presenting PA data ungrounded in the context of the user’s condition, does not meet clinical needs. As such, if this technology is to be incorporated into practice, solutions which facilitate the amalgamation of actigraphy data alongside contextual data that are relevant to the specific clinical needs of the various healthcare professionals are required. Research-grade devices such as Actigraph already have an application associated with them; therefore, developing these platforms further to incorporate self-report measures should be promoted. However, even with such development there is also a wider barrier that is worth noting, which is beyond the remit of this paper to discuss in depth: the availability of data to clinicians. Furthermore, access to this data remains difficult, thus further limiting clinician’s abilities to utilise it [32]. Currently, Irish hospitals do not have the resources to provide patients with wearable devices; thus, clinicians rely on patients collecting their own data using consumer devices. Recent calls have been made to allow this data to be easily transferrable to clinicians [32-34], however, significant regulatory changes are required before this may happen.

This exploratory study used a sample of convenience; therefore, it is unlikely that data saturation was reached, and so further research in this area is required. Furthermore, recruited clinicians had over 10 years of experience each and so may be less inclined to introduce new tools into practice. Additionally, a pragmatic decision was made to limit the visualisations to 4 to not overburden clinicians within a single focus group. However, a variety of alternatives exist. A user-centred design process is needed to determine the best way to integrate the needs of all clinicians, including the questions that they need answered, into future visualisation strategies to allow them to deliver the individualised, patient-centred care that current structures make difficult to achieve.

Although clinicians agree that the data derived from wearable devices may be useful, current devices fall short. Widely used devices, both consumer and medical grade, have only recently begun to enable users to either actively or passively collect contextual information regarding their PA, and for many devices this feature is still lacking. Significant improvements in both the data captured by these devices and the manner in which it can be displayed are required. Although there was no consensus among clinicians on the depth of context desired, they agreed that without it, the data did not currently support their practice. Irrespective of how data is displayed, until clinicians have a way in which they can apply it in a manner that best supports their role, the potential effectiveness of this data cannot be fully realised.

The study protocol was approved by University College Dublin research institute’s committee on human research (LS-E-20-83-Keogh-Caulfield). The participants of this study gave their written informed consent prior to participation.

The authors have no conflicts of interest to declare.

This study received funding support from Science Foundation Ireland 12/RC/2289_P2.

B.C. devised the study concept which was then refined with the wider team. B.M. and N.S. completed the data processing. B.M., N.S., J.F.D., and M.A. designed the visualisations. S.D., R.M., F.C., and J.F.D. were involved in the collection of the participant actigraphy data. A.K. and W.J. designed, ran and analysed the focus group meetings. All authors contributed to the final manuscript.