Empowering Older Patients to Engage in Self Care: Designing an Interactive Robotic Device

Abstract

Objectives:

To develop and test an interactive robot mounted computing device to support medication management as an example of a complex self-care task in older adults.

Method:

A Grounded Theory (GT), Participatory Design (PD) approach was used within three Action Research (AR) cycles to understand design requirements and test the design configuration addressing the unique task requirements.

Results:

At the end of the first cycle a conceptual framework was evolved. The second cycle informed architecture and interface design. By the end of third cycle residents successfully interacted with the dialogue system and were generally satisfied with the robot. The results informed further refinement of the prototype.

Conclusion:

An interactive, touch screen based, robot-mounted information tool can be developed to support healthcare needs of older people. Qualitative methods such as the hybrid GT-PD-AR approach may be particularly helpful for innovating and articulating design requirements in challenging situations.

Introduction

More than 62% of Americans over 65 who suffer from multiple chronic conditions may require managing multiple complex self-care tasks including medications (1). Inability to safely manage medications and perform self-care activities often is a major reason for transfer to an aged care facility (2). Over time, with the tilting of demographic balance and shortage of caregivers, older people may face the tension between increasing cognitive and physical limitations requiring dependence on one hand and desire to maintain independence on the other. Creating technologies to automate assistance for older people is a rapidly growing field (3, 4), however, challenges still remain. Many of those challenges can be attributed to older people’s relationship with technology and its usability. Often, declining visual and auditory capabilities, slurred speech, shaky stiff fingers and other challenges limit the use of existing desktop or mobile phone applications (5–7). Human-computer interaction studies have found touch screen based interfaces that make the interaction simple, directed and goal oriented can offer a potential alternative to address some of these challenges for older people (8).

Other important challenges have been identified as apathy and sustaining older people’s interest in executing self care tasks over longer periods. Increasing physical limitations and social isolation could be an important factor for dampening their motivation. To address this challenge we looked towards the growing field of social robotics. Robots have also been shown to make interaction with computers interesting, engaging and personalized, and to successfully engage older users (9). Robots also offer a unique opportunity to add an interpersonal element to inform, empower and support older users and also leverage formation of an affective or social relationship(10). Not only can they navigate, track, identify and invite users to interact but also serve as an extension of the remote caregiver through tele-presence. As shown in Figure 1, just as touch screens extend usability of standard computers, the information presentation capabilities of a touch screen based user interface can be extended on a robotic platform for successfully engaging older people.

Figure 1.

Incremental enablement of computer usability for older people

With increasing shortages of caregivers to support ageing populations and need for ageing in place (11), a robotic platform could serve as an extension of the human caregiver: as a tool to carry information, invoke patient engagement, access data or to provide intervention in real time to a larger number of patients. Our fundamental basis for proposing this solution is the robot’s influence of physical embodiment (or anthropomorphic presence) that builds shared social context with the user (10) as a key for time-extended relationships with people, including those with special needs. Robots pose an opportunity for enabling tele-presence of healthcare professionals within older and disabled people’s homes, as well as to leverage sophisticated use of information systems in General Practices in New Zealand (12) in supporting self care by older people at home.

There have been attempts to apply social robots in healthcare in recent times (13, 14), but research to date has been deficient in defining social, psychological and clinical aspects of the overall design process as well as in eliciting design requirements from older users in participatory interaction. The current literature also does not satisfactorily address implications around intrusiveness, domination, disempowerment and medicalization (15) that could be imposed by social robots in healthcare and limit their potential use. These are some of the themes explored in this paper.

The device described in this paper is a fully functional standalone IT application. The value from this research is the interpretive approach used to characterize complex healthcare processes, in order to guide a specialized IT application design. We believe this approach and its application to IT design can inform other IT work around healthcare support for older people. The paper describes the conceptual framework of the tool, illustrates the design of a prototype and also provides usability testing results of the prototype tool. The article concludes with a discussion of the research and system design implications of this research and next steps based on the findings. This work is intended to encourage further research on leveraging cognitive robotics technology and robotics programming tools for integrating end users in a highly complex healthcare eco-system, thus encouraging new lines of thinking.

Methods

Taking an interpretive approach, a mixed method grounded theory-participatory interaction design (GT-PD) approach was used for this study, iteratively refining the design using the Action Research (AR) paradigm (16–18). Participatory interaction design is a way to understand knowledge by doing and to make sense of the traditional, tacit, and often invisible ways that people perform their everyday activities (19). Grounded theory (GT) is a “qualitative research method that uses a systematic set of procedures to develop an inductively derived grounded theory about a phenomenon” (20). A GT study does not begin with a preconceived theory that needs to be proven, but rather the theory emerges from the coding of information collected. Action Research (AR) is a research method that is founded upon a qualitative, post-positive philosophy of science that presumes that complex social systems cannot be reduced for meaningful study. Baskerville has identified GT as an important tool that brings rigor to the AR process (21). The most prevalent description of AR (22), details a cyclical process that first requires the establishment of a research environment, followed by five identifiable phases which are iterated, namely: (1) diagnosing, (2) action planning, (3) action taking, (4) evaluating and (5) specifying learning. The 3 iterative cycles followed in this study are shown in Figure 2.

Figure 2.

Action Research Cycles

Data Sources

Multiple information sources were used to provide a comprehensive picture of medication processes as a complex self care task. Along the principles of PD we conducted interviews, field observations and on-site discussions and feedback sessions, in addition to expert panel discussion. The views were complementary as well as divergent at times, mostly cross validating facts while adding finer-grained nuances and subtleties from diverse points of view which were not possible otherwise. Simultaneous review of literature allowed relationships to develop between the GT concepts and categories and also between conceptual models of self management and the technological trends, which provided an understanding of how such conceptual models could be incorporated in the application design.

AR Cycle 1- Framing concepts

After obtaining ethics approval from the University of Auckland’s institutional committee, the researchers followed caregivers while they collected, organized, documented and dispensed medications, following their narrative, asking questions, taking photographs and videos, and making field notes that were later analysed using GT. Thereafter, three residents and their family members, the facility manager, three caregivers, and the resident physician and pharmacist, as actors responsible for performing medication related functions, were interviewed to obtain their perspectives. Eight hours of audio recordings were transcribed and analysed using GT.

AR Cycle 2 – Designing application

After analysing and reflecting upon earlier results in the light of literature on medication compliance and automated healthcare dialogue systems, a paper prototype of the computer interface was prepared and shown to three potential users in a PD manner, eliciting their feedback and views on how the system would look and work. The prototypes were also discussed with a sociologist, two geriatricians, two psychologists, two caregivers, a general practitioner and a senior nurse at the ACF in addition to a computer scientist. Most of the experts were senior faculty members in prominent academic institutions and hospitals. The lead researcher (PT) took notes during these group interactions and personal interviews. Again the data was analysed using GT. The earlier themes were triangulated with this rich information to arrive at system design architecture.

AR Cycle 3 – Testing and refining

We randomly chose ten participants from a list of residents showing interest in being a volunteer to participate in a usability study. With their written consent, prescription data was obtained from the pharmacy in electronic format which informed the server database, which in turn informed the robot to tailor its dialogue accordingly. Participants’ mean age was 80.5 years. The researcher being a trained physician performed a rapid assessment of their cognitive status and three out of ten participants had mild to moderate cognitive impairment. They interacted with the robot as it went into their apartments on a scheduled morning and assisted them in taking their morning medications. The dialogue system was personalised to each user and initiated with a personal greeting after identifying them though face recognition. The think-aloud method was used to capture the testers’ thoughts as they completed the testing cases. During each testing session the computer screen and think-aloud commentary were recorded using a video camera. After the session, a questionnaire was served and a structured interview conducted. Each testing session was approximately two hours in length. The video results were later analysed and coded by the researcher (PT) and verified by the co-authors.

As seen above, GT and PD were combined into a hybrid GT-PD approach to draw on the strengths of both methods. PD provided the means of user engagement to obtain a rich perspective on drawing system requirements and user interface design. GT provides the means of analyzing and coding the data to allow us to understand the attitudes and beliefs surrounding clinical processes and the information used within those processes in order to develop a theoretical framework. Both these methods bring rigor to the phases of AR cycles – for example, PD proved to be a very effective means of obtaining data to facilitate understanding of tacit and explicit user requirements. Those requirements were then analyzed using GT. Design informed by the analyses was iterated to incorporate finer-grained issues. This rigor was important within a sensitive context involving at-risk participants.

Results

The study had three outcomes, one from each cycle. Firstly, we framed concepts in cycle 1; designed the application by the end of cycle 2; and finally, tested and refined the application by the end of cycle 3.

Study outcome 1 - Conceptual framework

The themes arrived at after axial coding confirmed that the general/overall physical and cognitive capabilities of older users are variable. Although people with severe dementia or high needs would be more dependent on human care, those relatively independent and capable people were more likely to engage with their own care. The usability of the system would need to match capability/limitations of the users. Themes also indicated implicit demand to impart basic knowledge about medication (what, how, when, why) even if users were on compliance packaging.

Users demanded privacy and dignity, and expressed the need to be given encouragement and allowed to make independent choices at each step. The system should enable users to engage effectively with the self care task by bringing reminders, knowledge, medications, monitoring equipment and surveillance close to the patient, and maintaining user friendliness. Exploring these themes in the light of the literature, we identified the themes around providing knowledge, providing motivation, and offering users the power of choice at each stage could be further summarised under a single theme of empowerment (23).

The next set of themes indicated the need for collaboration between users, family and healthcare providers. Taking medication was described not just as a routine task but as a socially-contextualised one where patients, family members, caregivers, pharmacists, GPs and specialists all were a part of the process and needed to know what each other was doing, but each from a different perspective (24). This opens the opportunity for healthcare professionals (all using different information systems with different interpretations of the person’s electronic health records) to use the EHR for better communication about their patients. Importantly, the need was felt to create a safe system that is robust, does not introduce new errors and helps its users in timely identification and reporting of adverse events. In summary, selective coding arrived at usability, empowerment, collaboration and safety as four major design themes as shown in Figure 3.

Figure 3.

GT analysis of themes

Study outcome 2 - Architecture

These requirements clearly indicated that we were not looking at designing a simple alarm reminder system but rather at a mechanism to enable better healthcare delivery the includes the possibility of extending the EHR into the home. The richness of information derived earlier, determined our planning of the architecture, design of user interface, logic to determine utterance/display at each state and the background computing required to deliver the same.

The robot’s shape and functionalities were derived earlier in the project with participatory feedback from older residents of an Aged Care Facility(25). Architecture of the robot consists of four layers: robot hardware, a robot software framework (RSF), a robot application programming interface (RAPI), and service applications. Hardware includes two single-board computers, sonar sensors, and a touch-screen mounted on an actuated head integrated with camera, microphone, speakers, and wireless networking ports. The touch-screen enabled interactive dialogue system is based on behaviour change theories (26, 27).

We designed and implemented a dialog and event management system that could display reconfigurable spoken and text/touch screen dialog. A high-level view of design elements informed by the conceptual framework is shown in Table 1, and key dialogue design principles are shown in Table 2.

Table 1.

Key design aspects

Key design aspects for older users
Ability to use computing device	Use of touch screen as an input device. Clear display of one single piece of information with each dialogue and not more than three large soft-buttons offering clear options
Enabling natural interaction	Automated dialogue system with dual output (text and voice). Written display with matched spoken dialogue to reaffirm same information and presenting guidance on what is expected from the user. Mixed initiative, adaptive and adaptable interface that adjusts itself dynamically according to user choices and states to make it as close to natural human interaction as possible.
Empowerment and choice	Respectful communication without any coercion and ability to express choices at each step. Secure data storage and communication. User can control sharing of personal information. Access to tailored drug information and instructions. User given ability to customize preferred name, preferred time, pill organization if used (pill box or blisters). User can refuse, postpone or call for human assistance.
Collaboration	No personal information stored on the robot. Central server holds electronic records that interoperate with larger healthcare system. Provider’s ability to customize information and monitoring parameters. Server informs robot dialogues and also collects user information to inform providers. Real time communication with caregivers/family.

Table 2.

Dialogue design to match clinical requirements

Clinical logic for safe medication use	Dialogue design aspects
Identify right user	Face recognition followed by dialogue confirmation (Are you Mrs...?) with Yes/No options
Estimate right time	Medications scheduled for one or more of breakfast, lunch, tea, dinner time or as needed basis - selected according to clock and user preferences
Prompt user to read labels and identify right medication	Display brand /generic name of medication on screen and prompt user to press “Done” button when successfully located.
Prompt user to take right dose	Tailored dialogues (e.g. take one pill from the bottle or two puffs of inhaler) requiring user confirmation
Prompt user for the right route	Tailored dialogue (e.g. swallow pills with a glass of water, inhale puffs using spacer and so on)
Explore medication efficacy/adverse events as customized by the physician or pharmacist	Tailored dialogue with Yes/No options (Have you been feeling breathless?)
Additional safety mechanism	On screen help, back, exit options. Text message to caregiver/family member at any time experiencing difficulty, weekly log charts displayed to prescriber

The dialogues were based on branching tree logic coded in an XML dialog notation. The system was based on interoperability with the export file from local pharmacy software and a W3C Simple Object Access Protocol (SOAP) service oriented architecture interfacing the robot to a Web-based electronic medical record over 802.11 wireless communications. The prescription information being part of the medical record, stored in a secure server, was invoked on user authentication (28).

The server database holds fields from the pharmacy record (medication name, dose, instructions) and also allows additional information and customization to effectively meet user preferences. The design details of the system are published elsewhere (29).

The device was programmed to display written as well as spoken dialogues with large soft-buttons (i.e. buttons rendered by the software on the touch screen) in contrasting colours as the input mechanism. Tailored information was displayed on screen in large fonts and volume adjustment given for spoken dialogue. Dual output was intended to cover possible visual and/or auditory deficits.

The system offered relevant drug information with doctor/pharmacist’s instructions and explanations if asked for. On screen options also allowed users to refuse a medication and attempted to probe reason for refusal so as to inform the prescriber about changing user preferences and needs. The users could call for help anytime or even were empowered to refuse informing anyone about their personal decisions. These choices were provided to test if the users might choose to assert themselves or tend to passively follow sequential instructions.

Study outcome 3 - Usability Testing

The detailed results of the testing have been presented by the authors elsewhere (30). In summary, results of video analysis showed that all ten users were able to complete the interaction successfully within three to ten minutes irrespective of age or cognitive status. Interestingly, there was no significant relationship observed between occurrences of errors or time taken to complete the session, in relationship to age, computer literacy or previous exposure to the robot. This observation highlights the fact that in this study, mild cognitive impairment and/or unfamiliarity with computers or robots did not hamper usability of our system. As shown in table 3, the main issues identified were technical errors in voice generation, data transmission and screen transitions (mean 4.8 errors per user) and a few errors in choosing the right options (mean 0.9 errors per user).

Table 3.

Usability testing: questionnaire response

Parameters	Observations
Average age of respondents	80.5 years
Gender distribution	50% Males 50% Females
Meds organized as	50% Loose medications 20% Pill Box 20% Sachets 10% Blister packs
Mean (SD) scores on Likert scale (1: Strongly agree – 5: Strongly disagree)
I may like to use this robotic reminder system frequently	2.5 ± 1.58
System would be easy to understand and use by people like me	1.9 ± 1.10
It would be easy to practically use this system in our living quarters	2.4 ± 1.35
I would need technical support to be able to use this system	4.3 ± 1.49
Various functions in this system were appropriately designed	2.0 ± 0.94
Such a system would make us feel confident about our health	2.6 ± 1.17
I would imagine that most people would learn to use it very quickly	2.6 ± 1.26
I found the system very cumbersome to use	4.3 ± 1.16
I felt very confident using the system	1.5 ± 1.08
I needed to learn a lot of things before I could get going with this system	4.2 ± 1.03

The questionnaire asked the participants to rank their opinion on a Likert scale where the questions probed their likes and dislikes about the system. Most (>80%) users found the system easy to use, appropriately designed and felt confident about using it. This is reaffirmed by the observation that most users did not find the system cumbersome, neither did they feel they needed to learn a lot before using the system nor felt that they needed support during its use. However, nearly equivocal responses to questions pertaining to practical usability over long term need to be addressed and explored further in next phase. There is a scope of improvement to make them feel more confident about their health and improve usability for more challenged users.

Triangulating the data from questionnaire and video analysis found appropriateness of information content (82%), good comprehension of graphics and text (87%), few issues in navigation (9%), good system understanding and usability (76%).

The interview focussed on the next steps in the design process and attempted to elicit their participatory inputs in the design. Data from interviews pointed towards the idea of a robot being more intelligently aware of user status and carrying devices, dispensers and water. It was desirable to include a side effect surveillance system and reminder for medication refills and clinic/pharmacy appointments. There was almost unanimous desire by the respondents to keep their family members and/or caregivers in the loop about the medication processes as a fallback safety mechanism.

Informed by these results, the final design iteration (see Figure 4) incorporates context awareness from the home monitoring sensors and physiological monitoring devices worn by the patient. A web based application (called ‘Robogen’) has been designed to read medication information from pharmacy and GP systems (with or without involving a Personal Health Record, or PHR) and also enables caregivers and/or family members to be informed in real time through text, messages and/or video calls. It enables physicians and pharmacists to enter, monitor and edit a medication plan (which could be a part of a larger care plan) and customize dialogues.

Figure 4.

Robotic medication management system components and users conceived at end of AR cycles

The robot not only serves as a reminder, but also as vital signs and side effect monitoring device that flags and shares information through communication devices and PHR. The robot effectively extends the usability of PHR to older people, which otherwise would find it difficult to access, use and populate data into. The robot could eventually be visualized as an agent to assist execution of care plans and help healthcare system learn more about patient behaviour and assist better clinical decision making. The next stages of evaluation and development would test the system over the long term, ultimately aiming towards a larger clinical trial; and also bringing a greater degree of standards compliance to the electronic interchanges between Robogen and other systems. Currently the robot does not share information with GPs or the pharmacist directly. Medication information is demonstrated over Robogen screen to them during development phase. To achieve sharing of data, connecting technology is needed that complies with HL7 CDA, additional technology to connect the robot to a GP’s information system and the use of SNOMED-CT coding to ensure data quality.

Discussion

Although there is the need for IT tools to support complex self-care tasks in older people, there are few empirical studies that show how to design such tools. This article makes a contribution by presenting an interpretive methodological approach to understanding and modelling the medication management tasks in the form of a conceptual model. The article also illustrated how an architectural framework was iteratively refined and implemented as a prototype robot-mounted computing device.

The article has implications for systems design by illustrating the value of a GT-PD-AR method for seeking and understanding multiple perspectives of system use not only from potential users but also from members of the wider healthcare team who are expected to care for the elderly. The article also illustrated the potential for using networked robots that facilitate real time tailoring of interaction and collaboration with care providers. The article also shows how an interactive tool could be developed using an empowering dialogue system to promote participatory models of healthcare delivery, instead of employing commonly-practiced controlling and persuading paradigms. It also lays the groundwork for improved safety and quality use of medications in elderly by verifying medications, and asking for and reporting on side effects in real time. In the future it could match medication data with home sensors and physiological monitoring device data to build a picture for more informed clinical decision making. Poor compliance in older people is common(31), but remains a poorly understood phenomenon (32). We hope that by bringing transparency into day-to-day patient behaviour we can learn more about medication use choices in older people and derive useful insights for the healthcare system to refine its approach.

A key lesson learned was that understanding and modelling difficult users and their needs for IT support tools to support complex self care tasks requires a large time commitment from a multi-disciplinary team and requires multiple iterations to get it right. Also it contributes to the understanding that IT tools for older people could be developed on a robotic device with a simple touch-screen based user interface and using automated dialogue that does not have to depend on artificial intelligence or complex programming tools. Moreover, we discovered patterns about medication usage and reminder systems that could be useful to extrapolate to non-institutionalised or even non-elderly population. For example, independently living elders within the community or people with disabilities who are on multiple medications (e.g. recovering from stroke or disabling injury) who find it difficult to find a helping hand on a day to day basis could be benefitted from such a system. By keeping healthcare professionals and caregivers in the loop, that too in real time, builds an additional layer of safety. The results could generalize for building similar interfaces on other home monitoring devices and extending an already sophisticated use of the electronic health record in primary care in New Zealand. Learning about user behaviour and how an actor makes his/her choices is the best opportunity to inform development of innovative technologies for the future.

Limitations of the study include the fact that despite using a rich and comprehensive set of data to develop the tool, it was still based on the small sample of users and specific to the context of one ACF. This limits the transferability, which is a characteristic of qualitative research (33). However, this paper presents efforts to uncover design elements in the initial phase of a research project that is progressively building towards a larger trial in future iterations.

Conclusion

Delivering effective healthcare to the growing elderly population demands innovative solutions that are appropriate, and that engage and empower them in a safe manner to enable complex self-care tasks. The complexities of healthcare processes make it difficult to model and develop solutions that fit within the context and workflow of its users. Networked computers on a robotic platform, presenting dynamically programmable interactive tools to older users, could promote transparency and efficiency to improve quality of care. Incorporating iterative, user-centred design and evaluation approaches can help in understanding those complexities and enhance our ability to design IT tools to support better care for older people.