Abstract

The usability of a People-Like-Me approach to visualizing health outcomes was assessed, wherein a patient of interest is highlighted and data of patients with similar characteristics are plotted for reference (Kittleson, Hoogeboom, Schenkman, Stevens-Lapsley, & van Meeteren, 2020). Electronic health record data were queried to develop an interactive visualization of Patient-Reported Outcomes over time, with demographic variables included as interactive filters. Nine Physical Therapists and one physical therapist student viewed the chart and read several case studies of hypothetical patients in a low back pain rehabilitation program to facilitate interaction with the interface. Subjects then answered Likert-type questions and open-ended prompts relating to four content areas: Usability, Interpretability, Integration, Helpfulness. The System Usability Scale (Bangor, Kortum, & Miller, 2008) was adapted to assess Usability, and the remaining Likert scales were constructed based on face validity and content validity. After dropping one item from the Interpretability scale, all Likert scales had a Cronbach’s alpha > 0.8, indicating good scale reliability. The interface had average usability (75.0, CI: 60.3 - 89.7), was interpretable (4.28, CI: 3.9 - 4.7), helpful (3.97, CI: 3.5 - 4.5), and therapists were open to integrating the reference chart into practice (3.76, CI: 3.03 - 4.5). However, responses to open-ended prompts indicated integration concerns, such as time constraints and patient interpretability concerns. Therapists find the interface usable, interpretable (by the therapist), and helpful, but it is unclear whether a therapist would integrate a similar reference chart into practice. Future work is needed to collect feedback from a larger sample of physical therapists with diverse backgrounds, and to collect similar feedback from patients in rehabilitation programs.

Introduction

The amount of data available to clinicians is growing at an unprecedented rate (Murdoch & Detsky, 2013). Accessibility to large quantities of meaningful health data (e.g., patient-reported outcomes) provides clinicians with opportunities for exploration and insight, potentially leading to improvements in patient care. Clinicians could query electronic health record data to create patient-facing visualizations, for example, which could be integrated into rehabilitation programs to help patients better understand their prognosis. One reason these types of visualizations are uncommon in practice, however, is because little research has been done to investigate effective ways of integrating patient-facing visualizations (e.g., what kind of graph should be shown, what contextual information is relevant, what reference lines should be shown) (Turchioe et al., 2019).

Research lacks practical ways of using visual analytics for large medical datasets, and there is a need to support research and pilot projects in this area (Caban, & Gotz, 2015). The National Academy of Medicine’s vision is for health information technology to enable potent visual data analyses that can improve care, inform decision making, and accelerate discovery (Shneiderman, Plaisant, & Hesse, 2013); researching novel approaches for patient-facing visualizations is in alignment with their vision. The potential benefits of providing a patient with health data and appropriate context include increased comprehension of health status, engagement in care, and adoption of positive health behaviors (Baker, Rideout, Gertler, & Raube, 2005).

Although there is a critical need to develop such visualizations, there is little research on what visualizations are interpretable and helpful to patients and therapists. A systematic review by Turchioe et al. (2019) concluded there are no consistent approaches in the literature for selecting, developing, or evaluating visualizations according to a particular data type or goal. For example, two articles in their review reported that line graphs were difficult for patients to interpret (Britto et al., 2009; Elder & Barney, 2012), while a panel of experts in a third article recommended line graphs for longitudinal data (Snyder et al., 2019). Although there is little consensus, several articles did show that including contextual information (e.g., reference ranges) improves objective comprehension (Scherer et al., 2018; Elder & Barney, 2012; Snyder et al., 2019).

A novel approach for a patient-facing visualization is a “People-Like-Me” (PLM) reference chart, a patient-facing visualization concept recently introduced by Kittleson, Hoogeboom, Schenkman, Stevens-Lapsley, and van Meeteren (2020). A PLM reference chart is a visualization highlighting a patient of interest, and data of patients with similar characteristics are plotted for reference (see Appendix A for figures from Kittleson et al.). For example, these charts could be used in rehabilitation programs where patients may be interested in understanding how their progress (e.g., outcomes) compares to similar patients. If outcomes are measured at multiple time points, which is often the case for Patient-Reported Outcomes Measurement Information System (PROMIS) scores, patients and clinicians could compare progress over time with similar patients. This type of reference chart is appealing—particularly in conjunction with PROMIS scores—because clinically important difference scores can be added as reference lines, which may be particularly informative to patients and clinicians alike. Turchioe et al. (2019) noted that minimal clinically important differences could be used as meaningful boundaries to enhance patient interpretation and guide behaviors. PLM reference charts could also facilitate a person-centered care approach to health care, often considered the gold standard for long-term care settings (Li & Porock, 2014). The concept of a PLM reference chart seems promising, but whether a PLM chart is practically helpful/useful to a therapist and patient remains untested.

In the current study, we developed an interactive PLM reference chart interface displaying PROMIS Physical Function (PF) scores over time using historical electronic health record data. Users could filter the visualization using various variables, such as age, gender, function level, and acuity level. A trend line was displayed, minimal clinically important difference (MCID) reference lines could be shown or hidden, and the user could highlight a patient of interest by entering a medical record number into a text box, which allowed one to view a patient’s recovery in reference to historical outcomes of “people like them.” Participants were recruited and asked to interact with the interface and answer survey questions about their experience. The purpose of this pilot study was to gain an initial impression from therapists on the usefulness and practicality of such an interface in practice.

Methods

Subjects

Physical Therapists, Physical Therapist Assistants, and Physical Therapist students affiliated with University of Utah Health and practicing at hospitals and community clinics were recruited for this study by reaching out through emails and word of mouth. There were no specific exclusion criteria for this study. All interactions with the subjects occurred remotely, and subjects were able to complete all parts of the study via a web-accessible computer.

People-Like-Me Interface

A PLM reference chart interface was developed with Tableau (Tableau Software Inc., Mountain View, CA), a data analytics and visualization software. The interface was intended for clinicians working with patients from a low back pain rehabilitation program, but the interface could easily be applied toward other rehabilitation programs. Specifically, the PROMIS PF score was selected as the primary outcome to be displayed (given its relevance to this cohort), and relevant demographic variables were included as filters (described in more detail below). However, given different populations of interest, other outcome variables and filters could easily be selected and plotted. PROMIS PF score was the dependent variable presented on the y-axis, and the number of days in rehabilitation on the x-axis. A spaghetti plot showing the change in PROMIS PF score over time for each individual was chosen as the type of graph to display. Data could be filtered by gender, acuity of pain, level of function, age, days in physical therapy, and surgery patients (whether to include or remove them). A trend line that changed according to filters was displayed, and MCID reference lines could be shown or hidden. Users could enter a patient’s medical record number to highlight patients of interest. A still image of the visualization is shown in Figure 2.1.

Data Collection

A website page was developed for displaying and interacting with the PLM interface (link: https://rpubs.com/JasonDude16/PLM_Visualization). The landing page for the site contained information describing the visualization, and instructions for interacting with it were provided in both text and video format (Appendix B). Case studies were also presented to facilitate interaction (Appendix C). These case studies briefly described a patient and their characteristics that were relevant for filtering the PLM chart, such as age, gender, and acuity level. The instructions on the landing page were as follows: “Imagine you are a clinician working with a patient in a low back pain rehabilitation program. After a few weeks/months of being in the program, the patient is interested in seeing how their health outcome (PROMIS Physical Function score, in this case) has changed throughout the program and how their score compares to similar patients. Your task is to manipulate the visualization so that you and the patient can see how their health outcome has changed throughout the program and how this change compares to similar patients. Click on the Case Study tabs shown at the top and try using the case study information to filter the visualization and see how the patient’s PROMIS Physical Function score changes over time relative to patients with similar demographic and clinical information.” Participants were informed they could take as long or as little as they’d like to view the case studies and interact with the interface. When finished, they completed a survey asking about their experience.

A web-accessible survey was given to participants via REDCap (project-redcap.org) immediately following their interaction with the chart (Appendix D). The survey consisted of both open- and closed-ended questions addressing four specific content areas: (1) the usability of the interface, (2) the interpretability of the visualization, (3) how helpful the interface might be at the point of care, and (4) would the provider integrate this interface (or similar) into their practice if given the option? Four scales were developed to answer these questions: Usability, Interpretability, Helpfulness, and Integration. For measuring Usability, a previously validated questionnaire—the 10-item System Usability Scale (SUS; Appendix E) (Bangor, Kortum, & Miller, 2008)—was adapted to this context. The SUS assesses perceived usability of a given product or service. The scale was chosen for its ability to assess a wide range of interfaces, its brevity, easy interpretation by study participants, as well as a single score output that is easy to understand (Bangor, Kortum, & Miller, 2008). Several modifications were made to the SUS items to improve their relevance. For example, the original SUS item 1 asks, “I think that I would like to use this system frequently,” and in this study the item was changed to, “I think that I would like to use this interface frequently.”

Valid and reliable surveys could not be found for measuring scales equivalent to Interpretability, Helpfulness, or Integration. Therefore, we pilot tested our scales to assess these domains. In each domain, survey questions were constructed based on face validity (i.e., “on the surface, does the question tap into the construct of interest?”) and content validity (i.e., “are the questions representative of different aspects of the construct?”), as evaluated by two of the authors (JD and KRL). The Interpretability and Integration scales were each composed of five questions, and the Helpfulness scale was composed of seven. All items for all scales (including Usability) were five-point Likert-type items, ranging from Strongly Disagree (1) to Strongly Agree (5). All items for all scales required a value (i.e., missing values were not allowed). In addition to the scales, eight open-ended prompts were given (exact text is given in the tables below): three prompts related to the Interpretability dimension, two prompts (each) related to the Helpfulness and Integration dimensions, and one prompt asking for additional feedback (Interpretability prompts: What parts of the interface, if any, did not feel intuitive?; Describe how easy or difficult it was to become oriented with the visualization.; Do you think the average patient would be able to interpret the visualization? Please explain.; Helpfulness prompts: Do you think the average patient would find the visualization helpful? Please explain.; Do you think a patient’s experience in a physical therapy program could be improved by using this interface? Please explain.; Integration prompts: If given the option, would you use this interface in practice? Please explain why or why not.; Please describe any reservations you have about using this interface in practice.).

Data Analysis

Before computing the average scores across our different scales, we assessed the reliability of these scales using Cronbach’s alpha coefficient (Cronbach, 1951). Operationally, given this is a pilot study, we defined suitable reliability as ρα = 0.70. If ρα >= 0.70, we calculated the mean response across all items on each scale. If ρα >= 0.70, we removed the item(s) with the lowest correlation and averaged across the remaining items to get a single mean response for each scale.

The Interpretation, Helpfulness, and Integration scale scores were averaged after reversing relevant items. For computing SUS scores, even-numbered items were computed by having each item value subtracted from the numeric value five, and odd-numbered items were computed by subtracting one from each item value (Lewis, 2018). All values are multiplied by 2.5, yielding a final value score in the range of 0-100. Although these values are percentages, their values are based purely on a participant’s individual response and do not reflect the rank of that response in the overall distribution of responses (see also Lewis, 2018). Due to the small sample size, open-ended response data (i.e., qualitative data) were not analyzed using formal methods. Instead, responses to the prompts were simply presented as tables. Observed patterns (e.g., multiple participants making similar comments) were commented on in the discussion section. I want to emphasize that these comments based on the subjective interpretation of the authors and should not be interpreted as a formal qualitative synthesis.

Results

Descriptive Statistics

Ten participants (6 Male, 4 Female, Table 3.1) completed the study, nine of whom were Physical Therapists and one was a Physical Therapist student. The cohort of Physical Therapists had experience in Physical Therapy ranging from 11-30 years (Mean: 17.1). Three participants specialized in Orthopedics and most (7/10) were between the ages of 30-49. Participants spent 4.8 minutes on average exploring the interface, not including reading the instructions, watching the video, and reading the case studies.

Scale Scores

Usability, Helpfulness, and Integration all had ρα > 0.70 when including all items and, subsequently, the mean response was calculated with the inclusion of all items. Interpretability had ρα < 0.70 when including all items; the item with the lowest inter-item correlation was removed and the mean response was calculated from the remaining items. After removing this item, all scales had ρα > 0.80 (Tables 3.1-3.5). Means scale scores and confidence intervals are presented in Table 3.6.

Qualitative Feedback

Qualitative feedback is presented as-is without modifying participants’ responses. Tables 3.7-3.9 relate to the interpretability dimension (e.g., intuitiveness of interface), Tables 3.10 and 3.11 relate to the helpfulness dimension (e.g., would a patient find this interface helpful), and Tables 3.12 and 3.13 relate to the integration dimension (e.g., do you have reservations about using this interface). The prompt asking for additional feedback received only two (brief) responses, so these results were excluded.

Discussion

The mixed methods approach used in the study provides an opportunity for comparing qualitative and quantitative feedback. This section consists of four subsections for each of the scales, each beginning with the main quantitative results followed by a discussion of the observed patterns of the qualitative feedback (except for Usability, since Usability did not have corresponding open-ended prompts).

Usability

The System Usability Scale, which was slightly adapted to create the Usability Scale, is a valid and reliable survey (Lewis, 2018), making the results from the Usability scale of primary interest. Cronbach’s alpha was 0.93 with the inclusion of all ten items, indicating excellent reliability. The sample size was small but prior research has shown that usability problems can be identified with as few as 8-10 participants (Nielson, 1994). One appealing feature of the SUS is the easily interpretable grading scale. Bangor, Kortum, and Miller (2008) have defined adjective ratings corresponding to numeric SUS scores. The average SUS score based on findings from over 200 studies is about 68 (Lewis, 2018), which has a corresponding “C” grade rating and an adjective rating of “Okay.” The average Usability score obtained for this study (Mean 75.0, CI: 60.3 – 89.7) has a “B” grade and an adjective rating of “Good”. Although the mean SUS score for this study was not significantly greater than 68, the 95% confidence interval indicates approximately average usability on the low end, and possibly greater. The above-average usability score suggests the interface is at least moderately usable for therapists.

Interpretability

The Interpretability scale consisted of five items. In general, these items asked if the respondent could interpret the visualization and whether they thought a patient would be able to interpret the visualization with and without help from a therapist. Cronbach’s alpha was < 0.70 when including all items, so the item with the lowest correlation was dropped which asked the therapist if they thought a patient could interpret the visualization independently (i.e., without help from a therapist). Removing this item increased Cronbach’s alpha to 0.80, but, consequently, affected the scale’s meaning. That is, the scale no longer asked the therapist if a patient could interpret the visualization on their own; only if a patient could interpret with help from a therapist. The mean scale sore for Interpretability was 4.28 (CI: 3.9 - 4.7) suggesting good interpretability by therapists. However, the mean score was calculated from three items inquiring about a therapist’s interpretation and one about a patient’s interpretation with help from a therapist. Thus, the score’s meaning is rather vague. It would have been better to include more items inquiring specifically about 1) a therapist’s interpretation, 2) a patient’s interpretation without help from a therapist and 3) a patient’s interpretation with help from a therapist (and to differentiate items by subscales to avoid confounding results).

In general, open-ended responses suggest positive experiences related to the interpretability dimension when using the interface. All respondents commented that it felt “easy” or “fairly easy” to become familiar with the interface. Aspects of the interface that did not feel intuitive (Table 3.7) were related to the presentation of the interface and not the interface itself. For example, one respondent commented that they were unsure of what the cases (case studies) represented on the website. Although responses were generally positive, several respondents raised concerns about a patient’s ability to interpret the interface. One free-response question asked if they thought the average patient would be able to interpret the visualization, and three participants commented that it may be difficult—or at least vary greatly—for neurology patients to interpret the visualization (three of the respondents had a clinical specialty in Neurology). This is perhaps unsurprising given that neurological disorders can adversely affect cognitive function (McDonnell, Smith, & Mackintosh 2011). Implementing this interface for such a population may require additional thought and consideration, as the interface used in this study may be too complex.

Helpfulness

The Helpfulness scale consisted of seven items. These items asked if the interface would be helpful for the therapist and did not ask if the therapist thought this would be helpful to a patient (In hindsight, it likely would have been beneficial to include items inquiring about the latter on a separate subscale). This is a notable distinction because what is helpful to a therapist is not necessarily helpful to a patient, and vice versa. Cronbach’s alpha was 0.90 when including all items, indicating excellent reliability. The mean scale sore for Helpfulness was 3.97 (CI 3.5 - 4.5). Whereas the other scales did not include items asking about specific features of the interface, Helpfulness did: two items asked about the utility of the MCID reference lines. Turchioe et al. (2019) noted that reference lines could be added to increase patient interpretation and guide behavior, but, surprisingly, these two items had lower means than most of the other Helpfulness items. Perhaps it has more to do with how the reference lines were implemented in this interface specifically, or maybe reference lines are less helpful to therapists than patients.

Qualitative feedback seemed to align with quantitative feedback for helpfulness. Six of the eight respondents answered yes when asked if they thought the average patient would find the visualization helpful, and two were unsure (they answered “possibly” or “I think so”). One respondent mentioned that it might be helpful to have a health care provider provide context to the interface, which may be particularly important in situations where a patient compares unfavorably to “people like them.” Kittleson et al. made a similar comment, asking whether patients who scored low relative to their peers would become motivated to improve their outcome or if the experience would be demoralizing. In either case, however, the conversation between a therapist and patient that stems from viewing a patient’s outcome may itself be considered an advancement in health care (Montori, Breslin, Maleska, & Weymiller 2007). Another interesting comment from Table 3.11 mentions that having this interface would allow a therapist to say, “…we haven’t made progress and we’ve been working at this for X length of time, maybe should consider referral vs. other interventions.” Having this interface may provide an opportunity for conversation and reflection between patient and provider, thereby influencing treatment plans as a result.

Integration

The Integration scale consisted of five items and in general these items asked if a therapist would use this interface in practice, or if it’s useful enough to use in practice. Cronbach’s alpha was 0.90, and no items were dropped for computing the average and associated statistics. The mean score for Integration was 3.76 (CI: 3.03 – 4.5), which was the lowest of the four scales. However, the Integration items are, subjectively, the least specific and focused of the four scales. Several items asked directly about integrating the interface into a therapist’s workflow, but some items were less direct. For example, one item stated, “I do not know how I could use this interface in practice.” A high score on this item suggests the participant knows how to use this interface in practice, but this does not necessarily mean they would use the interface in practice. Because of the ambiguity of several scale items the qualitative responses related to Integration are of particular interest.

The qualitative feedback for Integration consisted of mixed responses. When participants were asked what reservations they have about using this interface in practice, two made comments relating to time constraints. Two separate comments were related to concerns about a patient’s ability (or lack thereof) to interpret the interface, but one was referring specifically to cognitively impaired neurology patients. Five of the eight participants who responded to the prompt, “If given the option, would you use this interface in practice?” commented “yes” or “absolutely.” Two participants commented, “no” or “not sure” to the same prompt, and both listed time constraints or low priority as their reasoning. Future research should seek to understand why some therapists feel this is a low priority and what would make them feel differently. Suppose it were emphasized that such an interface can facilitate conversation between therapist and patient, thereby leading to greater patient retention (this is purely speculative). Would a therapist make such an interface a higher priority? Similarly, if therapists were told they could use this to educate patients and facilitate the development of treatment plans simultaneously, would a therapist be more inclined to use it? These are potentially important considerations for a therapist that could influence their perspective on the utility and priority of an interface in practice.

Dos and Dont’s

The interface was usable and interpretably for therapists, but it is potentially less interpretable for patients, specifically those with cognitive impairments. The interface was helpful to therapists and therapists think the average patient would find the interface helpful. Despite this, there is resistance to adopting the interface in practice, namely time constraint and patient interpretation concerns, and interface issues. This section focuses on addressing those concerns as well as describing some of the strengths and weaknesses of the current study.

Asking good questions is essential for any survey. It is suggested to ask more questions and with greater detail than the current study and, if possible, use valid and reliable scales that reflect the dimensions of interest. It would have been useful to separate the Interpretability scale into two subscales: one for assessing interpretability by a therapist and another for assessing interpretability by a patient (but still surveying the therapist). Similarly, adding the same subscales for Helpfulness (helpful for therapist vs. helpful to patient) would have been beneficial. Although the quantitative scales were helpful to the overall goal of assessing usability, the qualitative feedback was much more informative, and it is highly suggested to collect qualitative feedback due to the rich insights it provided even with a small sample size. Asking detailed open-ended questions related to time constraints and priority/importance may improve our understanding of why some therapists are hesitant about adopting the interface. The most informative feedback may come from follow-up questions directed toward those hesitant about adopting (e.g., understanding why they are hesitant and what would make them less so).

When developing these interfaces, it is critical to consider the time required for both a therapist and patient to use and understand it. Seeing that therapists raised time constraint and patient interpretability concerns, it is apparent the interface needs to be simple and quick to use. Although much of the feedback related to interpretability and usability was positive, improvements can be made to mitigate future concerns in these areas. There is little research on best practices for interfaces used specifically as patient-facing visualizations (Turchioe et al., 2019), but more general resources for developing effective data visualizations are ubiquitous (Midway, 2020; Chen, Härdle, & Unwin, 2007; Healy, 2018). Because the literature on this topic is rather scarce, it may be worthwhile to develop and implement several PLM chart versions. An ambitious and informative next step would be testing several PLM chart versions on both therapists and patients, which would help to answer whether patients can interpret the selected interface. Thoughtfully considering the interfaces’ design will likely influence its adoption rate and interpretability.

Several respondents raised interface concerns. One respondent commented on the slow loading time for the visualization, and another mentioned the graph did not always update after selecting filter criteria. Notably, respondents listed these comments in response to the prompt, “Please describe any reservations you have about using this interface in practice.” A user’s experience mustn’t be hampered by a poorly functioning visualization, as this could alter responses and subsequently influence conclusions (this may have contributed to the low Integration score, for example). Regardless of how the interface is implemented, thorough testing is imperative to maximize the user’s experience and reduce the chances of adversely affecting feedback.

One strength of this study is that participants did not have any issues accessing or viewing content on the website. This approach of creating and embedding a visualization into an RMarkdown document (https://rmarkdown.rstudio.com) (or similar) and uploading to a website, such as RPubs (https://rpubs.com/), is also relatively easy to implement and little coding experience is required. A video was embedded on the website with a brief (~5 min) description of how to interact with the interface, which may have been helpful for participants. Because some therapists suggested the interface was not high on the priority list, it may be worthwhile to provide examples of how this interface could be used in practice when presenting the interface to therapists (several examples are provided in the following section). Presenting the interface and accompanying information on a freely accessible website allows future researchers to see what information was shown to participants, providing a useful example framework for them and also helping to generate ideas for future implementations.

Lastly, when asking participants for feedback, it may be useful to distinguish between the example visualization and the concept of the visualization. In this study some respondents mentioned PROMIS scores in their responses, and, in doing so, were answering questions about the example visualization (with specific independent and dependent variables) as opposed to the concept of the visualization (with arbitrary independent and dependent variables). Thus, their opinions about the implementation of such an interface may have been influenced by these variables and the specific interface that was shown to them, which may or may not be desired.

People-Like-Me Charts and Person-Centered Care

Person-centered care is often considered the gold standard for health care (Li, & Porock, 2014). The National Academy of Medicine defines it as “Providing care that is respectful of, and responsive to, individual patient preferences, needs and values, and ensuring that patient values guide all clinical decisions,” (Baker, 2001). A key component of this approach is that the patient plays a role in their own medical treatment decision-making process. A person-centered care framework has been shown to improve health outcomes, increase patient satisfaction, and advance concordance between the care provider and patient on treatment plans (Ekman et al., 2011). Care providers now view person-centered care as an essential component of effective illness management and advocate for its integration into the healthcare system (American Geriatrics Society Expert Panel on Person‐Centered Care, 2016; Dubois, Singh, & Jiwani, 2008; Royen et al., 2010).

A PLM reference chart provides context to a patient’s recovery process (i.e., they can see where they are in relation to similar patients), and this context may provide relevant information to a patient, improving their ability to be an active participant in treatment decisions, putting them more in alignment with a person-centered care framework. Additionally, the reference charts are an opportunity for conversation between the patient and therapist to lead to more personalized care. For example, the recovery trajectory for a particular patient may be unique relative to people like them. In this situation, the therapist may alter some aspect of their program (e.g., dose, intensity, or type of therapy).

In 2016 the American Geriatric Society Expert Panel on Person-Centered Care (2016) provided a definition and essential elements of person-centered care for chronically ill older adults with functional limitations, in which they defined person-centered care as the following: “Person-centered care” means that individuals’ values and preferences are elicited and, once expressed, guide all aspects of their health care, supporting their realistic health and life goals. Person-centered care is achieved through a dynamic relationship among individuals, others who are important to them, and all relevant providers. This collaboration informs decision-making to the extent that the individual desires.

The panelists outlined eight “essential elements to realizing this definition.” Reference charts can incorporate several of the essential elements of person-centered care outlined by the article into a rehabilitation program. The elements of person-centered care that reference charts could potentially incorporate into a rehabilitation program are listed below, followed by a brief explanation of how a PLM reference chart addresses the need; the explanations are speculative, as research has not yet been conducted on the effects of integrating a PLM reference chart into a rehabilitation program.

Essential elements

  1. Ongoing review of the person’s goals and care plan. A PLM reference chart plots an outcome measure over time, often meaningful to both the patient and provider. A PLM reference chart could provide an opportunity for the provider and patient to continuously review and update treatment plans together that is in part based on the information gained from the reference chart. To achieve this aim, the outcome of interest must be carefully considered for relevance to the clinical population and responsiveness to change.

  2. Continual information sharing and integrated communication. A PLM reference chart could facilitate continuous discussion between a therapist and patient. Informative visualizations can spark conversation (Woolen, & Bakken, 2015), which could lead to gaining new information about the patient relevant to their recovery process, an education session by the provider, or more simply a conversation that is not directly relevant to the recovery process, but builds rapport and strengthens the patient-provider relationship.

  3. Education and training for providers and, when appropriate, the person and those important to the person. PLM reference charts display outcome measures, which are measures of a patient’s recovery status. A clinical measurement such as this should have context informative of how the measurement compares to what is expected for a particular person (Kittleson et al., 2020). A reference chart is one way to provide context to outcome measures, which can then be used to educate a patient about an outcome measure and how their recovery compares to historical outcomes data.

  4. Care supported by an interprofessional team in which the person is an integral team member. PLM reference charts could help patients become an integral team member by helping them better understand their prognosis, providing context to a patient’s recovery trajectory relative to similar patients, and providing the patient with information on the sorts of actions that can be taken to improve recovery. This information could potentially help a patient transition to self-management with therapist supervision, as opposed to instruction from the therapist.

Limitations

The biggest limitation is the small sample size. Although this is a critique for many studies, it is especially true for this study with only ten participants. Another limitation is that three of the four scales implemented were developed, and two of those three scales used only five items or less to compute the average and associated statistics. Perhaps another limitation is the highly experienced cohort. Seven of the eight participants were Physical Therapists with an average of nearly 16 years of experience. This specific sample may have thought the interface was easier to interpret than a less experienced cohort would. However, having this cohort may also be advantageous because these therapists may have a better gauge for what patients would or would not be able to interpret in an interface. A related limitation is that it is unknown how well less experienced physical therapist aides/assistants understand the interface, which may be important if they primarily use the interface in practice and engage with a patient. Lastly, the interface was not shown to patients. The interface is ultimately intended to be shown to patients with the broad goal of improving their rehabilitation experience, and a patient’s perspective is, of course, paramount to achieving this goal.

Conclusion

Interactive visualizations are powerful tools that can advance health care and facilitate person-centered care. Before investing a significant amount of time and effort into creating such tools, a critical first step is to gain feedback from clinicians on the usability of such an interface in practice. This pilot study aimed to understand what clinicians think of an interface using a People-Like-Me approach. Results suggest the interface is usable, helpful, interpretable, and therapists would consider using it as a patient-facing visualization; however, some respondents indicated integration concerns, specifically time constraint and patient interpretation concerns, and several issues with the functionality/presentation of the interface. When developing these interfaces it is essential to consider the time required for a therapist and patient to become familiar with the reference chart and how feasible it would be to integrate the interface into a clinician’s workflow. As a next step, we suggest 1) implementing the study with a larger sample size, 2) employing several visualizations to better understand what visual aids are more and less interpretable, and 3) collecting more detailed qualitative feedback from both clinicians and patients.