Content uploaded by Tobias Kowatsch
Author content
All content in this area was uploaded by Tobias Kowatsch on Feb 23, 2021
Content may be subject to copyright.
Original Paper
Hybrid Ubiquitous Coaching With a Novel Combination of Mobile
and Holographic Conversational Agents Targeting Adherence to
Home Exercises:Four Design and Evaluation Studies
Tobias Kowatsch1,2,3, PhD; Kim-Morgaine Lohse1, MSc; Valérie Erb4,5, BSc; Leo Schittenhelm2, MSc; Helen Galliker1,
BSc; Rea Lehner3, PhD; Elaine M Huang5, PhD
1Centre for Digital Health Interventions, Department of Management, Technology, and Economics, ETH Zurich, Zurich, Switzerland
2Centre for Digital Health Interventions, Institute of Technology Management, University of St Gallen, St Gallen, Switzerland
3Future Health Technologies Programme, Campus for Research Excellence and Technological Enterprise, Singapore-ETH Centre, Singapore, Singapore
4Graduate School of Culture Technology, Korea Advanced Institute of Science and Technology, Daejeon, Republic of Korea
5People and Computing Lab, Department of Informatics, University of Zurich, Zurich, Switzerland
Corresponding Author:
Tobias Kowatsch, PhD
Centre for Digital Health Interventions
Department of Management, Technology, and Economics
ETH Zurich
WEV-G, Weinbergstrasse 56/58
Zurich, 8092
Switzerland
Phone: 41 712247244
Fax: 41 712247301
Email: tkowatsch@ethz.ch
Abstract
Background: Effective treatments for various conditions such as obesity, cardiac heart diseases, or low back pain require not
only personal on-site coaching sessions by health care experts but also a significant amount of home exercises. However,
nonadherence to home exercises is still a serious problem as it leads to increased costs due to prolonged treatments.
Objective: To improve adherence to home exercises, we propose, implement, and assess the novel coaching concept of hybrid
ubiquitous coaching (HUC). In HUC, health care experts are complemented by a conversational agent (CA) that delivers
psychoeducation and personalized motivational messages via a smartphone, as well as real-time exercise support, monitoring,
and feedback in a hands-free augmented reality environment.
Methods: We applied HUC to the field of physiotherapy and conducted 4 design-and-evaluate loops with an interdisciplinary
team to assess how HUC is perceived by patients and physiotherapists and whether HUC leads to treatment adherence. A first
version of HUC was evaluated by 35 physiotherapy patients in a lab setting to identify patients’perceptions of HUC. In addition,
11 physiotherapists were interviewed about HUC and assessed whether the CA could help them build up a working alliance with
their patients. A second version was then tested by 15 patients in a within-subject experiment to identify the ability of HUC to
address adherence and to build a working alliance between the patient and the CA. Finally, a 4-week n-of-1 trial was conducted
with 1 patient to show one experience with HUC in depth and thereby potentially reveal real-world benefits and challenges.
Results: Patients perceived HUC to be useful, easy to use, and enjoyable, preferred it to state-of-the-art approaches, and expressed
their intentions to use it. Moreover, patients built a working alliance with the CA. Physiotherapists saw a relative advantage of
HUC compared to current approaches but initially did not see the potential in terms of a working alliance, which changed after
seeing the results of HUC in the field. Qualitative feedback from patients indicated that they enjoyed doing the exercise with an
augmented reality–based CA and understood better how to do the exercise correctly with HUC. Moreover, physiotherapists
highlighted that HUC would be helpful to use in the therapy process. The longitudinal field study resulted in an adherence rate
of 92% (11/12 sessions; 330/360 repetitions; 33/36 sets) and a substantial increase in exercise accuracy during the 4 weeks.
J Med Internet Res 2021 | vol. 23 | iss. 2 | e23612 | p. 1https://www.jmir.org/2021/2/e23612 (page number not for citation purposes)
Kowatsch et alJOURNAL OF MEDICAL INTERNET RESEARCH
XSL
•
FO
RenderX
Conclusions: The overall positive assessments from both patients and health care experts suggest that HUC is a promising tool
to be applied in various disorders with a relevant set of home exercises. Future research, however, must implement a variety of
exercises and test HUC with patients suffering from different disorders.
(J Med Internet Res 2021;23(2):e23612) doi: 10.2196/23612
KEYWORDS
ubiquitous coaching; augmented reality; health care; treatment adherence; design science research; physiotherapy; chronic back
pain; pain; chronic pain; exercise; adherence; treatment; conversational agent; smartphone; mobile phone
Introduction
Musculoskeletal disorders (MSDs), such as rheumatoid arthritis,
osteoarthritis, low back pain, neck pain, or gout, negatively
impact the locomotor system and are influenced by various risk
factors, such as a sedentary lifestyle, malnutrition, or obesity
[1]. Individuals with MSDs suffer from chronic pain, impaired
mobility and physical function, and reduced quality of life [2].
The average estimated global prevalence of MSDs lies at 18%
[3] and increases with age [4]. MSDs account for 21.3% of all
years lived with disability worldwide [5], with back pain as the
leading cause of disability [3]. Therefore, both the individual’s
psychosocial burden and the financial burden of MSDs are
significant. In the United States, for example, treatment costs
are estimated to account for 5.7% of gross domestic product
[6]. As a consequence of population growth, aging, and
sedentary lifestyles, the burden of MSDs will dramatically
increase in the future and thus poses significant challenges to
global health [4,5]. It is, therefore, crucial to develop effective
interventions for individuals with MSDs.
Physiotherapy is an effective intervention for MSDs [7-10].
However, treatment adherence (ie, the extent to which an
individual’s behavior “corresponds with agreed
recommendations from a health care provider” [11]) is a
common problem in various health care settings [11-13],
including home exercises in physiotherapy [11,14-17]. For
example, treatment adherence ranges from 60% [18] down to
30% [9,19,20], resulting in increased costs because of prolonged
treatment [5,8,21-23]. Common treatment adherence dimensions
(TADs) [24-26] are the session completion rate (the number of
completed vs prescribed exercise sessions, TAD1), the set
completion rate (the number of completed vs prescribed sets
for each exercise, TAD2), the exercise repetition rate (the
number of completed vs prescribed exercise repetitions within
each set, TAD3), the temporal exercise accuracy (the actual vs
prescribed velocity of exercise execution, TAD4), and the spatial
exercise accuracy (the actual vs prescribed body movement
trajectory, TAD5). Reasons for nonadherence can vary. For
example, patients may simply forget to follow the various TADs,
and further patient-related factors, such as disability, attitudes,
motivation, and beliefs about exercise risks and benefits, can
also negatively impact the TADs [14,27-33].
Moreover, adherence support is often limited to on-site
physiotherapy sessions, leaving patients alone at home with
standardized exercise instructions, which results in low
adherence rates [34-36]. Therefore, various technical approaches
have been proposed to help increase home exercise adherence
in physiotherapy [37]. A smartphone app, for example, motivates
patients and sends reminder messages to individuals with an
MSD [38], or a virtual reality training is used to provide
feedback on exercise execution [39]. However, two recent
reviews indicate that adherence remains a challenge [40,41].
Our review of remote patient monitoring tools and current
scientific work (Multimedia Appendix 1 [16,29,39,42-66])
suggests that there is untapped potential in addressing this
problem, too.
To this end, we propose hybrid ubiquitous coaching (HUC) to
improve home exercise adherence for treatments that require
not only personal on-site coaching sessions by health care
experts but also a significant amount of home exercises. HUC
relies primarily on a conversational agent (CA) that delivers
relevant health literacy information about the importance and
benefits of home exercises and personalized motivational
exercise reminders via a smartphone. Moreover, the CA of HUC
delivers real-time exercise support, monitoring, and feedback
in a hands-free augmented reality (AR) environment. In HUC,
health care experts introduce the CA as their trusted personal
assistant that lives not only in the patient’s pocket/smartphone
in the form of a text-based physiotherapy chatbot [48,49,67],
but also in the patient’s AR glasses in the form of an embodied,
holographic instructor [52,55] (see Figure 1 for an overview of
HUC).
HUC combines for the very first time (to the authors’ best
knowledge) research on CAs, human-supported digital
interventions, health psychology, and mobile and wearable
technology–supported physical exercises. First, HUC employs
a CA because they are able to build a working alliance with
patients [49,57,68], which is an important relationship quality
[69] that is robustly linked to treatment success [59].
Accordingly, there is a large body of evidence on the
effectiveness of CAs in delivering clinical and nonclinical
interventions [52]. Second, HUC relies on a hybrid coaching
concept consisting of CA and human physiotherapist teams
because digital interventions supported by humans result not
only in higher treatment adherence [64,66], but also in better
treatment outcomes [70,71]. Third, HUC offers real-time
feedback about intervention progress and correct exercise
execution, which, in turn, is assumed to increase self-efficacy
[29], an important construct that helps shape positive attitudes
toward therapy adherence and health-promoting behavior [63].
Finally, HUC aims to increase adherence by seamlessly
delivering an intervention into the everyday lives of vulnerable
individuals or patients with the help of mobile and wearable
technology (ie, smartphones and AR glasses) [72,73].
J Med Internet Res 2021 | vol. 23 | iss. 2 | e23612 | p. 2https://www.jmir.org/2021/2/e23612 (page number not for citation purposes)
Kowatsch et alJOURNAL OF MEDICAL INTERNET RESEARCH
XSL
•
FO
RenderX
Figure 1. Overview of hybrid ubiquitous coaching (HUC).
Because this research is unprecedented, the following research
questions (RQs) were formulated to guide the design,
implementation, and evaluation of HUC: (RQ1) How is HUC
perceived by (a) patients and (b) health care experts?; and (RQ2)
Does HUC lead to treatment adherence to home exercises?
To answer these questions, we applied HUC exemplarily to the
field of physiotherapy. During the last 3 years, 4
design-and-evaluate loops were conducted: two studies were
carried out in the lab with 50 physiotherapy patients, 11
physiotherapists were interviewed, and finally, empirical data
from 1 patient during a 4-week longitudinal field study were
collected and assessed to reveal real-world benefits and
challenges related to HUC.
Methods
Overview
To address the challenge of nonadherence to home exercises in
physiotherapy, we started the interdisciplinary development of
HUC in collaboration with 2 physiotherapists. In a first step, a
storybook was developed (Multimedia Appendix 2). In a second
step, HUC was developed further as outlined in the next section.
After the conceptual work, we formulated the research questions
that guided the 4 studies in which we iteratively improved,
implemented, and evaluated HUC. Table 1 provides an overview
of the 2-year design and evaluation process.
J Med Internet Res 2021 | vol. 23 | iss. 2 | e23612 | p. 3https://www.jmir.org/2021/2/e23612 (page number not for citation purposes)
Kowatsch et alJOURNAL OF MEDICAL INTERNET RESEARCH
XSL
•
FO
RenderX
Table 1. Overview of the studies.
Study 4, 2019Study 3, 2019Study 2, 2018/2019Study 1, 2018Characteristic
Version 3: Revised AR-
based CA with dynamic be-
havior over time and Wizard
of Oz smartphone-based CA
coaching
Version 2: Revised AR-based CA coaching. One humanlike
exercise instructor with professional physiotherapist
clothing and additional smartphone-based CA coaching
(mockup).
Version 1: ARb-based CAc
coaching. Superhero exer-
cise instructor and human-
like female exercise model.
Maturity of HUCa
Observational longitudinal
field study
Within-subject lab studySemistructured interview
and survey
Lab studyStudy design
1 patient-physiotherapist
dyad, 3 physiotherapists
15 patients, 2 physiothera-
pists
11 physiotherapists35 patientsParticipants
Perceived usefulness, ease
of use & enjoyment. Inten-
tion to continue using HUC,
suggestions for improve-
ments.d
Patient-CA working al-
liance.dPerceived useful-
ness, ease of use, enjoyment,
task load, exercise difficul-
ty.d,e
N/Af
Perceived usefulness, ease
of use & enjoyment.dSug-
gestions for improvements.e
RQ1a: How is HUC per-
ceived by patients?
Perceived usefulness and
relative advantage of HUC,
suggestions for improve-
ments.e
N/APatient-physiotherapist
working alliance, relative
advantage of HUC.d,e
N/ARQ1b: How is HUC per-
ceived by health care ex-
perts?
Session and set completion
rates, exercise repetition
rate, spatial & temporal accu-
racy (TAD1-5)d
Spatial & temporal accuracy
(TAD4-5)d
Session completion rate, sets
& repetition (TAD1-3g)d
Behavioral intention & rec-
ommendation to use HUC.e
RQ2: Does HUC lead to
treatment adherence to home
exercises?
aHUC: hybrid ubiquitous coaching.
bAR: augmented reality.
cCA: conversational agent.
dQuantitative feedback.
eQualitative feedback.
fN/A: not applicable.
gTAD: treatment adherence dimension.
Concept of Hybrid Ubiquitous Coaching
A conceptual overview of HUC is illustrated in Figure 1 and
outlined in the following paragraphs.
Onboarding With a Physiotherapist
First, a physiotherapist performs a physical assessment of the
patient and, depending on the results, defines a tailored home
exercise program. Next, the physiotherapist works together with
the patient on the exercises from the program by focusing on
the TADs (Table 1). Then, the physiotherapist introduces the
CA as his or her scalable personal assistant that will support
the patient regarding the TADs in their everyday life via the
smartphone and the AR glasses. For this purpose, the
physiotherapist shows the patient how to interact with the CA
on the smartphone and via the AR glasses. To link the CA with
the patient (eg, name, age, height, weight, diagnosis) and his or
her specific home exercise program, the physiotherapist prints
out a therapeutic prescription card with a quick response (QR)
code. The QR code not only contains encrypted links to the
patient data and personalized exercise program, but also data
about the physiotherapist so that the CA “knows” the name of
the physiotherapist it was linked to. This allows the CA to link
back to the “real” world during conversations with the patient
and strengthen the relationship and the working alliance with
the specific physiotherapist. A strong working alliance between
health professionals and patients establishes an attachment bond
[74], the development of a shared understanding about goals
and tasks, and, in return, boosts treatment adherence [69,75].
On the way home, the patient scans the QR code with his or her
smartphone to download the mobile HUC app and start the
interaction with the CA. The CA introduces itself as the
physiotherapist’s personal assistant and gives an overview of
the upcoming digital coaching program. During this dialogue
with the CA, the shipment of the AR hardware to the patient’s
home is triggered as the patient’s address is made available via
the link in the QR code.
Everyday Coaching
The smartphone- and AR-based CAs allow for remote everyday
coaching. The combination of the two CAs together results in
a richer communication channel that enables more accurate and
personalized information to be transmitted to the patient [76,77].
The design of the HUC elements was also informed by
behavioral change techniques (BCTs, ie, evidence-based
intervention components aiming to change behavior) [78]. The
smartphone-based CA reminds the patient to do the home
exercises (TAD1), motivates the patient via the smartphone,
and emphasizes the benefits of performing the exercise by
providing psychoeducational material, for example, about sleep
quality and degree of pain (BCTs 2, 13, 15, and 20 [78]). The
J Med Internet Res 2021 | vol. 23 | iss. 2 | e23612 | p. 4https://www.jmir.org/2021/2/e23612 (page number not for citation purposes)
Kowatsch et alJOURNAL OF MEDICAL INTERNET RESEARCH
XSL
•
FO
RenderX
AR-based CA delivers real-time exercise support by
demonstrating the execution (BCT 22), whereby arrows
highlight important angles to be aware of and motivate the user
during the exercise (BCT 13). The AR-based CA guides the
patient through the exercise, monitoring the progress (TAD1-5,
BCT 17) and giving real-time feedback (TAD4-5, BCT 19). It
also informs and emphasizes essential aspects of how to perform
the exercise correctly (TAD4-5, BCT 21). The AR-based CA
counts the number of sets and repetitions out loud and provides
feedback on the exercise execution after a completed set of
exercises (TAD2-5, BCT 19). The feedback is both visual and
auditory and is based on comparing patient data from the AR
system with data about the physiotherapist performing the same
exercise. Lastly, the AR-based CA aims at increasing the
attachment bond, an important relationship quality and
dimension of a working alliance [79], by literally giving high
fives or showing the patient a funny dance move (BCT 13) when
a set or training session is completed.
On-site and Remote Coaching With a Physiotherapist
HUC grants physiotherapists access to patients’ data about the
TADs via a web-based dashboard. Based on that data,
physiotherapists can tailor and optimize planned on-site and
remote coaching sessions more efficiently. For example, if
squats are performed incorrectly and the real-time feedback
from the AR-based CA is not processed correctly by patients,
pressure could be applied on the knees rather than on the
muscles, which could cause knee injuries. If this happens, the
physiotherapist would get an alert message from the CA to
target the weakness directly in the next online or on-site
coaching session.
In the remainder of this paper, we describe 4 build-and-evaluate
loops that we conducted to build and assess HUC according to
our research questions. Table 1 provides an overview of this
process.
Study 1: Perceptions of Individuals With
Physiotherapy Experience
To address RQ1a and RQ2, we implemented a first version of
HUC with a focus on the AR-based CA and tested it with 35
individuals with physiotherapy experience at a public fair in
September 2018.
Development of HUC Version 1
HUC included an AR-based CA that appeared as a male
cartoonlike superhero assistant of a human male physiotherapist
and a female human-looking exercise model, demonstrating the
exercise execution (Figure 2). A personal, empathetic, and
humanlike CA can increase a patient’s intention to change
behavior and to continue to use the CA [80]. Further, the
rationale behind the superhero design was to make the user
experience more fun. The rationale for the design features was
to design a personal and empathetic CA. For ease of
presentation, a squat exercise was implemented as it is a
common physiotherapy exercise. After the CA demonstrated
the exercise (eg, by flying around the female character and
pointing out important aspects of the squat movements) and
commented via voice on how to correctly execute the squat, the
CA asked the patient to perform the exercise. Patients were able
to respond to voice-based questions from the CA with predefined
answer options that appeared as speech bubbles in the AR space
around 20 cm in front of and at eye level for the patient (Figure
3). The patient was able to select an answer by touching it with
a finger. This hands-free approach does not require an additional
controller and thus allows for more intuitive and direct
interactions [81] (Figure 4). A significant advantage of this is
the direct manipulations and interactions in the AR space.
Interactions were primarily used to increase the attachment bond
via small talk (eg, “How are you today?”), to explain the
exercise (eg, “Let’s have a close look at Alexis’ movements”),
or to progress through the several-step process of exercise
execution (“well done, four more to go!”). See Multimedia
Appendix 3 for the video clip and Multimedia Appendix 4 for
the technical details of the implementation and hardware.
Figure 2. Augmented reality–based conversational agent and female humanlike model demonstrating the squat exercise.
J Med Internet Res 2021 | vol. 23 | iss. 2 | e23612 | p. 5https://www.jmir.org/2021/2/e23612 (page number not for citation purposes)
Kowatsch et alJOURNAL OF MEDICAL INTERNET RESEARCH
XSL
•
FO
RenderX
Figure 3. Predefined answer options. The conversational agent communicated visually and auditorily.
Figure 4. Participant performing the squat exercise while wearing the augmented reality hardware.
Evaluation
Participant Acquisition
Participants were recruited from a public fair using convenience
sampling [82]. They were included if they had already
participated in physiotherapy sessions and were interested in
HUC.
Procedure
Participants who expressed an interest in HUC and provided
their consent to participate in the study were invited to
participate. HUC was then introduced, and usage of the AR
glasses was demonstrated. Next, participants were asked to start
the interaction with the AR-based CA and performed one
exercise session. Afterward, participants were asked to fill out
a feasibility questionnaire. Participants received no monetary
compensation.
Measurements
HUC was assessed based on technology acceptance [83-85] and
word-of-mouth [86] research. To reduce participant burden, and
due to the feasibility character of this first study, we used
single-item measures for perceived enjoyment [87], perceived
ease of use, perceived usefulness [85], and intention to use [83].
Consistent with prior work [87], 7-point Likert scales were used,
ranging from strongly disagree (1) to strongly agree (7). Further,
the Net Promoter Score (NPS) [86] was used to assess whether
participants would recommend HUC to other patients. The NPS
is a single-item measurement that indicates satisfaction with a
service. NPS scores are binned into four categories: −100 to 0
J Med Internet Res 2021 | vol. 23 | iss. 2 | e23612 | p. 6https://www.jmir.org/2021/2/e23612 (page number not for citation purposes)
Kowatsch et alJOURNAL OF MEDICAL INTERNET RESEARCH
XSL
•
FO
RenderX
(needs improvement), 1 to 30 (good), 31 to 70 (great), and 71
to 100 (excellent). Finally, qualitative feedback was gathered
on positive aspects of HUC, and suggestions for improvement
were provided by participants.
Study 2: Revision of HUC and Evaluation by
Physiotherapists
To address RQ1b and RQ2, we revised HUC based on the
feedback from study 1 and assessed it with 11 physiotherapists
between November 2018 and March 2019.
Design and Technical Implementation
HUC was revised as follows based on the qualitative results of
study 1. The AR-based CA was scaled up to be human-sized
and adapted in appearance to look like a physiotherapist (ie, the
clothes, including the logo, were copied from a real
physiotherapist, since a professional look and a natural
humanlike style are agent characteristics associated with
increased intention to use [80]). Moreover, due to the limited
field of view of the AR glasses, which sometimes made it hard
to see the interactions between the two characters in HUC
version 1 and thus confused some of the participants in study
1, the female humanlike model was omitted. Therefore, in study
2 the revised AR-based CA demonstrated the squat exercise
with the help of his own virtual body (Figure 5 and the video
in Multimedia Appendix 5). Moreover, an exercise tutorial was
added to help participants better understand how to correctly
execute the exercise (BCT 22). For this purpose, virtual guides
indicated important aspects of exercise execution (Figure 5).
Furthermore, automatic error detection was implemented in
collaboration with 2 physiotherapists. For the error detection,
the position and rotation of the headset were tracked and
compared with the squat movements of the physiotherapists
(Multimedia Appendix 4). Participants were asked to position
themselves on virtual footprints placed on the floor and to
remain in that position during the course of the exercise. Thus,
participants’ initial position and rotation were determined.
Additionally, participants’heights were stored as the difference
between the initial position and the floor along the vertical axis.
Based on these parameters, errors for insufficient depth of the
squat, too fast or too slow execution, and too much deviation
to the left, right, front, or back from an individual’s vertical
center axis were detected. Further technical details of the error
detection are outlined in Multimedia Appendix 4. The errors
detected were saved in the AR app so that the CA would be
able to generate and send an error report to the corresponding
physiotherapist. For example, if a participant did not move low
enough with the upper body during a specific squat exercise,
the AR-based CA would say, “A little bit lower.” In addition
to this error-related real-time feedback, and to target TAD1-3,
the AR-based CA took over the moderation during the exercises
with motivating voice-based instructions and progress reports
(BCTs 12, 13, and 15). For example, the AR-based CA counted
up the number of repetitions (eg, “Only three… two… one.
Great! You already finished the 1st set, take a quick break and
then let’s start with the 2nd set of squats”). Moreover, real-time
feedback was implemented based on these errors and
communicated during and after a specific exercise, thus targeting
temporal (TAD4) and spatial (TAD5) accuracies.
Finally, the HUC concept outlined in Figure 1 was broken down
into a flowchart diagram and specific sketches (eg, of the
web-based dashboard for physiotherapists and the
smartphone-based CA interaction; Multimedia Appendix 6)
that illustrated the various intervention components of HUC in
more detail. In addition to the revision of the AR-based exercise,
the flowchart and sketches were then used for the assessment
of the various aspects of HUC by physiotherapists as outlined
below.
Figure 5. Virtual guides embedded into the revised augmented reality–based conversational agent and used in studies 2, 3, and 4.
J Med Internet Res 2021 | vol. 23 | iss. 2 | e23612 | p. 7https://www.jmir.org/2021/2/e23612 (page number not for citation purposes)
Kowatsch et alJOURNAL OF MEDICAL INTERNET RESEARCH
XSL
•
FO
RenderX
Evaluation
Participant Acquisition
Physiotherapists treating patients with MSDs were recruited
using chain sampling [82] until saturation of the qualitative
feedback was reached [88].
Procedure
First, we explained the HUC concept, the flowchart, and the
sketches to each participating physiotherapist. Then, the
physiotherapist performed the specific squat exercise session
with the AR-based CA. Thereafter, a semistructured interview
was conducted to gather feedback on the utility and feasibility
of HUC. The interview was recorded and transcribed according
to the rules of simple transcription [89]. Thereafter, relevant
themes and concepts were extracted following an iterative
coding process [90]. After the interview, we sent an online
survey to the physiotherapist in which we asked them to assess
the relative advantage of HUC compared to current patient
monitoring and communication applications and asked about
the potential of HUC to strengthen the working alliance between
the physiotherapist and their patient. No monetary compensation
was provided.
Measurements
The guiding questions of the semistructured interview are listed
in Multimedia Appendix 7. The perceived relative advantage
of HUC was adapted from prior work [91]. The answer options
on the 6-item instrument were anchored on 7-point Likert scales,
ranging from strongly disagree (1) to strongly agree (7). The
Session Alliance Inventory [92] was used to assess the working
alliance. Answer options on the 6-item instrument were anchored
on 7-point Likert scales, ranging from never (1) to always (7).
Study 3: Revised HUC (Version 2) Assessed by Patients
To address RQ1a and RQ2, HUC was assessed by 15 patients
seeking physiotherapy treatment in January and February 2019.
To assess the relative advantage of HUC to commonly employed
methods of exercise instruction, HUC was compared to
paper-based and video-based exercise instruction.
Design and Technical Implementation
The revised HUC (version 2) as outlined in study 2 above was
used for the assessment.
Evaluation
Participant Acquisition
Patients were recruited from a physiotherapy center of one of
the collaborating physiotherapists. Inclusion criteria were age
≥18, participation in at least 3 physiotherapy sessions, no
experience with the squat exercise, being in the physical
condition to perform squat exercises, normal vision or wearing
contact lenses, and normal hearing.
Procedure
A two-period crossover study design was used to assess the
relative advantage of HUC compared to paper-based and
video-based exercise instruction, two commonly used methods
in physiotherapy. This study design also allowed us to account
for learning effects [93]. The study was conducted at the
physiotherapist center. After giving consent, each participant
was systematically assigned to a specific order of the exercise
instruction method. Then, a physiotherapist described and
demonstrated the squat exercise in the therapy session and
ensured correct exercise execution. Next, patients received either
handwritten paper-based exercise instructions (Figure 6),
video-based instructions (Figure 7 and Multimedia Appendix
8), or a QR code on a business card to download the HUC app
and the AR hardware (Figure 5 and Multimedia Appendix 5).
The content of the exercise instruction was consistent for all
three instruction methods. Each patient was then instructed to
perform 3 sets of squats with 10 repetitions each. The exercises
were video-recorded so that 2 physiotherapists were able to
assess the accuracy of the exercise execution independently
from each other at a later point in time (see Measurements).
After each of the 3 exercise sessions, patients were asked to
assess the instruction method via an online survey. Finally (ie,
after all exercise sessions), a semistructured interview was
conducted with each participant to gather suggestions to improve
HUC. The overall duration of approximately 1 hour was
compensated with US $50.
J Med Internet Res 2021 | vol. 23 | iss. 2 | e23612 | p. 8https://www.jmir.org/2021/2/e23612 (page number not for citation purposes)
Kowatsch et alJOURNAL OF MEDICAL INTERNET RESEARCH
XSL
•
FO
RenderX
Figure 7. Video-based instructions of study 3.
Measurements
The exercise instructions were assessed with mental effort and
frustration items from the task load index [94]. Moreover,
perceived enjoyment, ease of use, and usefulness of the exercise
instructions were adopted from technology acceptance research
[85,95]. We also assessed the working alliance between the
patient and HUC’s AR-based CA with the Session Alliance
Inventory [92]. Additionally, the intention to continue working
with the corresponding exercise instruction was determined
based on prior work [96]. We also asked patients to rank the
three instruction exercises according to their informativeness.
We finally adopted 5 assessment items from the Exercise
Assessment Scale [97]. With these items, 2 independent
physiotherapists were asked to assess the correctness of the
exercise execution for each instruction method and participant
with a maximum score of 6 points. For this purpose, patients
performing the squat were video-recorded from two perspectives
(front and side). All quantitative survey items are listed in
Multimedia Appendix 9. The questions of the semistructured
interview focused on patients’ perceptions and suggestions for
improvements (Multimedia Appendix 10).
Study 4: HUC Assessed in a 4-Week Field Study
Due to the high burden nature of the study, with experimental
technology that required support from the researchers, we
conducted a 4-week n-of-1 trial to complement our findings
from studies 1-3 and to answer RQ1a, RQ1b, and RQ2. The
goal of this study was to complement the findings from the
previous studies and to add external validity to the lab results
by assessing the long-term adherence and feasibility of HUC
in the everyday life of one patient. By assessing HUC with only
one patient, this study was not designed to illustrate the breadth
of possible use experiences but to show one experience in depth
and thereby potentially reveal real-world benefits and challenges
[98,99].
Design and Technical Implementation
The objective of the third design loop was to adapt HUC for
long-term use. For the field study assessment of HUC, there
was no QR code provided, and no dedicated HUC was
developed. Consistent with recent work [80,100,101], the
following revisions were implemented to increase variation in
HUC coaching.
First, depending on the patient’s familiarity with the AR-based
app (eg, how often it was used), tutorials became skippable,
and explanations were shortened after the first week. Second,
there was randomization of the coach’s speaking texts (“Hey,
good to see you again!”), animations (eg, stretching the upper
limb), and interactions (eg, clapping, high five, thumbs up).
Third, randomization of the coach’s texture (ie, the AR-based
CA had a set of different colored clothes) and the background
music (eg, energetic rock and funk) changed. Lastly, the
AR-based app showed the patient the number of sessions already
completed as an early form of progress visualization, a
well-established behavioral change technique (BCT 19) [78].
Evaluation
Participant Acquisition
A new physiotherapy patient who did not participate in one of
the other studies was recruited from a physiotherapy center.
The eligibility criteria were identical to those outlined in study
3. Additionally, 3 physiotherapists from the same center were
recruited for a follow-up interview. One physiotherapist was
the therapist treating the patient, the second therapist was
involved in the evaluation in the course of the third study, and
the third therapist was not involved in any of the previous
studies.
J Med Internet Res 2021 | vol. 23 | iss. 2 | e23612 | p. 10https://www.jmir.org/2021/2/e23612 (page number not for citation purposes)
Kowatsch et alJOURNAL OF MEDICAL INTERNET RESEARCH
XSL
•
FO
RenderX
Procedure
Before starting the study, the patient signed the informed
consent. Then, at the beginning of the study, the patient received
a study information sheet from the physiotherapist and agreed
to practice a squat exercise 3 times per week. HUC was then
demonstrated, and the patient received an introduction to the
AR headset. Thereafter, the patient was coached by HUC in
their everyday life as outlined in Figure 1. Again, the exercise
in focus was the squat. The coaching included, on average, 4
weekly text messages via WhatsApp, aimed to elicit reminders,
to provide psychoeducational material, and to motivate the
patient to perform the squat as often as the physiotherapists
recommended (BCTs 2, 13, 15, and 20 [78]). According to the
Wizard-of-Oz method [102], the CA was simulated by a
coauthor without the patient being aware of it and reminded the
patient weekly to conduct the exercises over a 4-week period.
Further, real-time feedback was provided during the execution
of the exercises (Multimedia Appendix 5). The TAD data (eg,
session completion rate, sets and repetitions, errors) were
collected and sent to a coauthor via email so that additional
feedback could be provided via a WhatsApp-based text message
(eg, to motivate the patient to perform an exercise). After the
trial, the patient was invited to take part in a debriefing
interview, during which the experience with HUC was explored
and the patient was asked for suggestions for improvement. The
patient received a monetary compensation of US $50 for
participating. In the case of any therapeutic or technical
questions, the patient was able to contact the physiotherapist
and the study team. Finally, the results of the n-of-1 trial were
shown to 3 physiotherapists to gather additional qualitative
feedback about HUC (eg, suggestions for improvement).
Results
The results of the 4 studies are presented in the following
sections. The depersonalized raw data and data analysis script
are made available in Multimedia Appendices 9 and 11,
respectively.
Study 1: Perceptions of Individuals With
Physiotherapy Experience
Overall, 35 (13 females) individuals with a mean age of 35 years
(SD 11) and previous physiotherapy experience participated in
the study. Most of them (32, 91%) had experience with the
exercise. The results listed in Table 2 indicate that participants
enjoyed the exercise sessions and perceived the exercise
moderated by the AR-based CA to be useful and easy to use.
Moreover, participants indicated that they would be willing to
use this form of AR-based exercise at home. All of these
assessments lie significantly above the neutral scale value by
conducting Wilcoxon signed rank tests (Table 2). The NPS was
negative and close to zero, indicating that the HUC needs
improvement before participants would recommend it to other
patients. This result was expected given the prototype character
of this first version of HUC.
Table 2. Augmented reality–based conversational agent coaching assessed by 35 patients.
Pvalueb
Meana(SD)
ItemsConstruct
<.0015.74 (1.06)
I enjoyed the exercise with Alex.c
Perceived enjoyment
<.0016.14 (0.84)It was easy to follow the exercise instructions.Perceived ease of use
<.0016.37 (0.80)I was able to follow the exercise.Perceived usefulness
<.0015.40 (1.49)I would use this type of holographic exercise at home.Intention to use
N/Af
−17.14e
How likely is it that you would recommend this type of exercise to other patients?
Net Promoter Scored
a7-point Likert scales ranging from strongly disagree (1) to strongly agree (7).
bWilcoxon signed rank test with test value 4.
cThe augmented reality–based conversational agent was given the name Alex.
dNPS ranges between −100 and 100.
eThe percentage of detractors subtracted from the percentage of promoters.
fN/A: not applicable.
The qualitative feedback on the positive aspects of HUC and
suggestions for improvement are shown in the thematic maps
(Figure 8). Suggested improvements included “more specific
feedback,” “more helpful instructions,” “coach should look
more like a physiotherapist,” “only one figure,” and “less/no
superman.”
J Med Internet Res 2021 | vol. 23 | iss. 2 | e23612 | p. 11https://www.jmir.org/2021/2/e23612 (page number not for citation purposes)
Kowatsch et alJOURNAL OF MEDICAL INTERNET RESEARCH
XSL
•
FO
RenderX
Figure 8. Thematic map of the patients' qualitative feedback of study 1. Note: the number in brackets indicates the frequency a topic was mentioned.
Study 2: Evaluation by Physiotherapists
Eleven physiotherapists with 3 to 30 years of experience (6
females, age of both sexes between 20 and 49 years) participated
in the study. Five physiotherapists worked in a private practice,
4 in a hospital, and 2 in a rehabilitation clinic.
The physiotherapists indicated in the interviews that patients
could be better supported with the HUC and that
physiotherapists would intend to use it for treatment sessions.
The AR-based CA was perceived as an additional motivator
due to its personal, interactive, and playful approach. The
exercise guidance by the life-sized CA, including the real-time
feedback (eg, automatic counting of exercise repetitions), was
also perceived as an improvement to the status quo. The
smartphone-based CA was perceived as an advantage over
current solutions, mainly due to its ability to remind patients to
execute the exercise, but also by providing psychoeducational
content that targets increasing patients’ health literacy.
Physiotherapists were, however, skeptical of whether HUC
could improve the quality of the treatment and adherence. The
HUC was seen to have some potential to strengthen the working
alliance between physiotherapists and their patients. Establishing
therapeutic goals was seen as an advantage, since tracking
enables patients to be continuously informed about the degree
of goal achievement. The ability to instruct and provide feedback
on the exercise technique was also identified as a factor that
could potentially improve the shared understanding of treatment
tasks. However, the majority of the physiotherapists expressed
the belief that mutual trust, empathy, and, consequently, mutual
goals and tasks were uniquely established during face-to-face
encounters and that technological systems could not successfully
act as assistants to foster a good working alliance. Lastly,
physiotherapists suggested that continuously monitoring a
patient could have a negative influence on mutual trust.
This qualitative feedback is supported by the quantitative data
from the online survey (Table 3). The physiotherapists saw a
clear relative advantage of HUC relative to the status quo of
commercial applications. The working alliance assessments
also resulted in a confirmation of the qualitative feedback.
Overall, the experts assessed the support of the CA as not being
sufficient to build and maintain a robust working alliance
between patients and physiotherapists. This was especially the
case for establishing an attachment bond.
J Med Internet Res 2021 | vol. 23 | iss. 2 | e23612 | p. 12https://www.jmir.org/2021/2/e23612 (page number not for citation purposes)
Kowatsch et alJOURNAL OF MEDICAL INTERNET RESEARCH
XSL
•
FO
RenderX
Table 3. HUC assessed relative to commercial applications by 11 physiotherapists.
Pvalue
Meana(SD)
Items, nExemplary itemConstruct
<.001c
4.97 (0.96)6
With HUCb, patients would perform the exercises more correctly compared
to previous methods.
Relative advantage
Working alliance
.19d
4.23 (1.31)2Alex would help me (my patient and I) work towards mutually agreed
goals.
Goal agreement
.64d
3.82 (1.47)1Alex would help me convince the patient that the way we work on the
problem is correct.
Task agreement
.99d
2.06 (1.32)3Alex would help me make the patient appreciate me.Attachment bond
.99d
3.08 (1.68)6Total
a7-point Likert scales ranging from strongly disagree (1) to strongly agree (7) were used.
bHUC: hybrid ubiquitous coaching.
cttest (2-tailed; t10=8.204) with test value 4.
dWilcoxon signed rank test with test value 4.
Study 3: Revised HUC (Version 2) Assessed by Patients
Fifteen patients (9 female) with a mean age of 37 years (SD
9.93) participated in the study. Participants were generally very
positive toward HUC, the video instruction, and, to some extent,
the paper instructions (Table 4). In total, 8 patients (53%) rated
HUC as their favorite training method, while 6 patients (40%)
rated the video instructions as their favorite method. Further,
in the HUC method, the overall working alliance, task
agreement, and attachment bond values were statistically
significantly higher than the neutral scale value of 4. HUC
resulted in higher average scores on the adapted exercise
assessment scale (TAD4-5) compared to the video- and
paper-based instructions.
The qualitative feedback suggested that the three instruction
methods provided different levels of richness of information.
Eight patients described the paper exercises as boring and not
enjoyable to follow. Even though patients found the paper
instructions helpful to recall the exercise, they often had trouble
interpreting the instructions correctly. Three patients found the
video instructions authentic and found the exercise easy to
understand. However, patients mentioned that they felt insecure
about the correctness of exercise execution. Patients also found
the exercises easier to understand while using HUC and
appreciated the guidance, the reminder app, and the real-time
feedback to understand when mistakes were made. Eight patients
also perceived HUC as fun and enjoyable. Six patients
particularly enjoyed performing the exercise together with a
life-sized CA. One patient, however, found that the CA did not
engage emotionally. A disadvantage of HUC perceived by 9
patients—thereby a frequently perceived disadvantage—was
the bulkiness of the AR hardware used in this study.
The patients suggested several improvements (Figure 9). First,
3 patients wanted to personalize the CA (eg, adjust the speed
and the movement parameters to account for varying abilities,
individually decide the appearance of the coach). Second, 2
patients mentioned a preference for a real person and not an
animated character (eg, “looking a bit more human-like, not
such a computer-figure”). Third, 2 patients suggested adding
more detailed real-time feedback and reminders and elements
of gamification (eg, rewards for regularity, real-time reminders,
and accuracy). Overall, HUC was positively perceived.
J Med Internet Res 2021 | vol. 23 | iss. 2 | e23612 | p. 13https://www.jmir.org/2021/2/e23612 (page number not for citation purposes)
Kowatsch et alJOURNAL OF MEDICAL INTERNET RESEARCH
XSL
•
FO
RenderX
Table 4. HUC-based, video-based, and paper-based exercise instruction methods assessed by 15 physiotherapy patients and exercise assessment scale
assessed by 2 physiotherapists. Note: _ represents the exercise instruction method.
Pvaluec
Paper
M (SD)
Pvaluec
Video
M (SD)
Pvaluec
HUCa
Mb(SD)
n(Exemplary) itemConstruct
Task load linked to exercise instruction methodd
.754.00 (1.36).043.06 (1.43).0052.66 (1.11)1How much mental and percep-
tual activity was required with
_?
Mental capacity
.022.80 (1.69).0042.60 (1.24).0052.66 (1.44)1How frustrated did you feel
during the execution with the
_?
Frustration
Perceived characteristics of the instruction methode
.914.13 (1.06).0055.26 (1.09).065.26 (1.90)1_ was fun to use.Perceived enjoyment
.0055.53 (1.40).0025.80 (1.08).0016.20 (1.01)1I could understand _ very
easily.
Perceived ease of use
.0024.76 (1.42)<.0015.07 (1.52)<.0015.59 (1.63)3_ helped me to do the exercis-
es correctly.
Perceived usefulness
Patient–conversational agent working alliancee
N/AN/AN/A
N/Af
.064.89 (1.96)2Alex and I agree on what is
important for me to work on.
Goal agreement
N/AN/AN/AN/A.025.14 (1.51)1The way Alex and I are
working with my problem is
correct.
Task agreement
N/AN/AN/AN/A<.0015.57 (1.45)3Alex and I respect each other.Attachment bond
N/AN/AN/AN/A<.0015.27 (1.65)6Total
.0025.14 (0.86).015.14 (1.29).105.07 (1.94)1How much would you like to
continuously use the [instruc-
tion method]?
Intention to continuously
usee
<.0013.63 (0.90)<.0014.17 (1.06)<.0014.23 (0.75)4Correct body part moving in
correct plane
Exercise assessment scaleg
aHUC: hybrid ubiquitous coaching.
bM: mean.
cWilcoxon signed rank test with a test value of 4 was used for all constructs but the exercise assessment scale, where a test value of 3 was used.
d7-point Likert scales ranging from very low (1) to very high (7).
e7-point Likert scales ranging from strongly disagree (1) to strongly agree (7).
fN/A: not applicable.
gExercise assessment scale items resulted in a score from completely incorrect (0) to completely correct (6).
J Med Internet Res 2021 | vol. 23 | iss. 2 | e23612 | p. 14https://www.jmir.org/2021/2/e23612 (page number not for citation purposes)
Kowatsch et alJOURNAL OF MEDICAL INTERNET RESEARCH
XSL
•
FO
RenderX
Figure 9. Thematic map of the patients' qualitative feedback of study 3. Note: the number in brackets indicates the frequency a topic was mentioned.
HUC: hybrid ubiquitous coaching.
Study 4: HUC Assessed in a 4-Week Field Study
Frequency and Engagement
Based on the behavioral data recorded by the HUC, a total of
3 exercise sessions per week were performed in 3 out of 4
weeks. This resulted in TAD rates of 92% in frequency, sets,
and repetitions (Table 5). During the last week, the patient was
on vacation and did not take the hardware with them. Therefore,
only 2 sessions were performed. Not going low enough with
the body during the squat exercise was the most common
mistake. The average number of mistakes during the 4 weeks
is depicted in Figure 10. The number of mistakes fell over the
4 weeks, indicating that the HUC’s real-time feedback was
processed by the patient and, at least to some degree, put into
action. In total, 17 standardized text messages were sent to
remind the patient to perform the exercises, to inform the patient
about their progress, and to provide motivational information
about the importance of performing the exercises. The patient
acknowledged these messages with 8 answers (Multimedia
Appendix 12).
Table 5. Behavioral data from the 4-week HUC practice (N=1).
ValueConstruct
11 (92)Session completion rate, n (%)
330 (92)Repetition rate, n (%)
33 (92)Sets, n (%)
8 (89)Errors (1st three sets), n (%)
4 (44)Errors (last three sets), n (%)
4.25 per week (17 in total)
HUCamessages, mean (sum)
2 per week (8 in total)Patient messages, mean (sum)
191 per session (2098 in total)Duration (s), mean (sum)
aHUC: hybrid ubiquitous coaching.
J Med Internet Res 2021 | vol. 23 | iss. 2 | e23612 | p. 15https://www.jmir.org/2021/2/e23612 (page number not for citation purposes)
Kowatsch et alJOURNAL OF MEDICAL INTERNET RESEARCH
XSL
•
FO
RenderX
Figure 10. Box plot of the exercise execution errors during the 4 weeks. The number of errors was aggregated for each week.
Debriefing Interview
Overall, the patient described the experience with the HUC as
very positive. The exercises guided by the AR-based CA were
enjoyable and motivating. The patient liked the feeling of not
being alone while performing the exercises and remarked that
they preferred the HUC to a human personal trainer, as it was
perceived as being “more relaxing.” Additionally, the patient
pointed out usually having difficulties adhering to home
exercises and consequently appreciating the variations of the
WhatsApp-based motivational messages from the
smartphone-based CA and feedback on the execution accuracy
by the AR-based CA, both of which helped the patient with
adherence. The patient expressed an interest in having a more
dynamic intervention program that includes longer sessions and
variations in the exercises. Finally, the patient wanted to
continue working with the smartphone-based and AR-based
CA beyond the period of this n-of-1 trial.
Follow-up Interviews With Physiotherapists
Reflecting on the results of the study, all physiotherapists
confirmed that HUC has the potential to improve adherence to
home-based physiotherapy exercises. Referring to the
progression of the exercise technique, the real-time feedback
about the temporal and spatial accuracy was seen as a clear
advantage. Two physiotherapists also liked the fact that the
system ensured adherence to the prescribed number of sets and
repetitions and described it as having a positive influence on
patient motivation. As a treatment provider, the detailed
monitoring information regarding the exercise technique was
considered to be highly valuable. All physiotherapists
highlighted that the therapy could be optimized and a patient’s
problems could be addressed more individually during on-site
therapy sessions. This could potentially also render the
face-to-face encounters more efficient. In a future version of
HUC, patients should be able to give feedback on their level of
pain or well-being. Particularly, information about a patient’s
pain level could enable more personalized digital coaching.
Lastly, according to 2 therapists, it should be possible to
calibrate the exercises more individually in terms of exercise
technique by adding different variations of one exercise.
Discussion
Principal Findings
The goal of this paper was to address the problem of
nonadherence to home exercises by proposing HUC, a human-
and CA-supported coaching approach that employs smartphone
and AR technology. This paper is, to the best of our knowledge,
the first to investigate the potential of HUC with patients and
health care experts in two lab studies, interview studies, and a
longitudinal n-of-1 trial in the field. HUC required collaborative
and interdisciplinary effort from various stakeholders, including
health care experts, patients, and experts in behavioral medicine,
game design, AR, and human computer interaction. A strength
of this work and HUC is, therefore, that its pragmatic and
diverse investigation helps to broadcast the results to real-world
scenarios and inform practitioners about effective digital designs
for addressing the problem of treatment adherence. Moreover,
the implemented home exercise (ie, the squat) is not only
relevant to physiotherapy patients but also an important
component of other treatments, for example, of high-intensity
interval trainings for patients who are overweight or patients
with cardiovascular disease. In the following section, the results
will be discussed along the research questions and design
challenges. Limitations and future work will also be outlined.
Research Question 1a: How is HUC Perceived by
Patients?
Overall, patients stated that they enjoyed executing the exercise
together with HUC, that they intended to use it, and that they
would recommend the final HUC version to other patients.
J Med Internet Res 2021 | vol. 23 | iss. 2 | e23612 | p. 16https://www.jmir.org/2021/2/e23612 (page number not for citation purposes)
Kowatsch et alJOURNAL OF MEDICAL INTERNET RESEARCH
XSL
•
FO
RenderX
Compared to paper- and video-based instructions, HUC reached
higher overall acceptance rates, required less mental capacity,
was perceived as being easier to follow, and resulted in higher
exercise accuracy ratings. However, significant differences were
only found for the items mental capacity, perceived enjoyment,
and perceived usefulness between HUC and the paper-based
instructions. This is still promising, however, for the following
reasons. First, HUC was rated at least as good as video
instructions, which is encouraging, since video instructions are
associated with reduced symptoms in physiotherapy [103].
Second, perceived enjoyment was significantly higher than with
paper-based instructions, a promising result, since enjoyment
is related to engagement in digital health interventions [104],
something that is often not taken into consideration [105] and
is important for long-term adherence. This was confirmed in
the 4-week field study. In line with previous findings in the
field of CAs [47,49,57,59], patients reported that they were able
to build a working alliance with the CA. This was especially
the case for attachment bonds, which include factors such as
mutual trust, acceptance, and empathy [49,106]. Lastly, in the
field study, the patient indicated that they were more motivated
to do the exercise when using HUC. The positive perception of
HUC and the feeling of collaboration between the patient and
the smartphone- and AR-based CA implies an established
working alliance.
In summary and to answer RQ1a, we conclude that patients
accepted and enjoyed HUC. Patients were also able to build a
working alliance with the CA.
Research Question 1b: How is HUC Perceived by
Health Care Experts?
HUC was also evaluated by physiotherapists to gather their
feedback. Collecting behavioral data about various treatment
adherence dimensions (eg, the session completion rate, the sets
rate, and the exercise repetition rate) was seen as a clear
advantage of HUC. Physiotherapists also considered the
real-time feedback to be an additional motivator due to its
personal, interactive, and gamified approach, which reflects an
improvement relative to the status quo. However,
physiotherapists were initially (study 2) more skeptical about
whether HUC could be conducive to building a strong working
alliance between patients and physiotherapists. The importance
of personal interaction between the physiotherapist and the
patient was one reason for the negative assessment of attachment
bond. Physiotherapists expressed that mutual trust, empathy,
and, consequently, mutual goals and tasks were uniquely
established during face-to-face encounters. Nevertheless, both
the quantitative and qualitative patient feedback (study 3)
underline the potential that a strong working alliance can be
built with a CA. This result is consistent with prior work about
working alliances and CAs [68]. There is a possible explanation
for the deviation between the working alliance ratings of
physiotherapists (study 2) and patients (study 3): While patients
assessed the working alliance with the CA, physiotherapists
assessed it in comparison to the bond that develops during
physical interaction between patient and therapist. However,
HUC foresees the integration of human and CA coaching into
both in-person and remote physiotherapy sessions (Figure 1).
As HUC does not replace but rather complements the bond
between the physiotherapist and patient, it is more equitable to
compare HUC with traditional coaching methods. Thus,
compared to conventional methods, HUC is shown to have the
ability to increase the working alliance.
In the follow-up interviews about the field study, all
physiotherapists agreed that home exercises in physiotherapy
could be optimized with HUC. They were convinced that various
treatment adherence dimensions could be addressed more
efficiently with HUC.
In summary and to answer RQ1b, we conclude that HUC
represents a clear relative advantage to current methods, as
indicated by physiotherapists, in particular, with respect to its
motivational features and real-time feedback.
Research Question 2: Does HUC Lead to Treatment
adherence to Home Exercises?
Results from study 3 and the 4-week n-of-1 trial were promising
with respect to RQ2. Qualitative feedback from study 3 indicated
that HUC addressed adherence challenges better than the paper-
and video-based instruction methods. In particular, patients
found the reminders (TAD1-2) useful, appreciated receiving
guidance (TAD3) and real-time feedback on the spatial and
temporal accuracy of the exercise execution (TAD4-5), and felt
more comfortable about the exercise. Moreover, the results of
the 4-week trial revealed a clear decrease in exercise execution
errors (TAD4-5) and a high adherence rate of 92% in TAD1-3.
Lastly, the patient reported being motivated and committed to
doing the exercise doing the 4 weeks due to the variations in
the messages and the feedback on exercise accuracy. Motivation
and enjoyment are targeted by HUC and, according to prior
work, associated with significantly increased rates of adherence
[14,29,30].
In summary and to answer RQ2, we conclude that HUC did
address the session completion rate (TAD1), set completion
rate (TAD2), exercise repetition rate (TAD3), and temporal and
spatial exercise accuracy (TAD5). Although this study does not
prove beyond a doubt that the use of HUC led to better
adherence, the rate was far higher than has typically been found
in previous studies [18,20] and was also higher than the
participant's previous adherence to home physiotherapy exercise
based on their self-report of their own practice and difficulties.
Design Challenges
A first challenge during the design of HUC was the development
of the CA’s personality in such a way that the patient perceives
the CA coherently as the physiotherapist’s digital assistant via
text messages on the smartphone and voice messages in the AR
space. The first lab study and the NPS revealed that HUC, and
the design in particular, still needed improvement. Accordingly,
the first version of HUC was modified based on this feedback.
In the second lab study, patients reported correspondently higher
levels of satisfaction with HUC and acceptance of the design
of the CA. Further, another challenge was the design of the AR
interactions, which had to take into account the limited exercise
space at patients’ homes, the limited field of view of the AR
glasses, and anthropometric characteristics of patients. Any
additional controller hardware also had to be omitted due to the
requirement that physiotherapy exercises be hands-free. This
J Med Internet Res 2021 | vol. 23 | iss. 2 | e23612 | p. 17https://www.jmir.org/2021/2/e23612 (page number not for citation purposes)
Kowatsch et alJOURNAL OF MEDICAL INTERNET RESEARCH
XSL
•
FO
RenderX
challenge was partly met. Patients ranked HUC as their
first-choice coaching method but still commented on the
bulkiness of the hardware. Another major challenge was the
design of variations in the smartphone-based and AR-based CA
conversations to improve long-term adherence to physiotherapy
in the last study. Variations in the design were implemented to
make the exercises more diversified and to thereby increase
adherence. The positive result for treatment adherence indicated
that the last major design challenge was met.
Limitations and Future Work
The current work has the following limitations. First and
foremost, in the current HUC version, only one home exercise
was implemented, and thus the findings, especially those related
to the AR-based CA, cannot be generalized to a large extent.
Implementations of HUC should be further developed to reach
a level of implementation foreseen in its approach, as outlined
in Figure 1. Future work is therefore required to develop an
intuitive CA that supports health care experts through the
customization of home exercise sessions, for example, via
suggestions given by prior patients [107,108]. Additionally, the
problem with the deviations in the exercises could also be solved
with machine learning. In doing so, exercise suggestions could
be learned, for example, from a basket analysis that considers
past exercise programs, patient-related variables (eg, diagnosis
and preferences) and past treatment outcomes (eg, reduction in
pain) [109,110].
Second, objective adherence rates were based on lab studies
and a 4-week study. Regular physical therapy treatment lasts
approximately 9 weeks, while long-term treatment can last more
than 9 months (36 sessions) [111]. Thus, HUC needs to be
assessed not only by a representative sample of patients but also
for a representative treatment duration that is common not only
in physiotherapy but also in other treatments that require
intensive home exercises (eg, treatments for patients with obesity
or cardiovascular disease).
Third, the current AR-based CA had a fixed coaching approach.
A growing body of research reveals that certain personalities
and therapeutic techniques (eg, interpersonal, person-centered,
behavioral) work better in specific contexts [112,113]. The
suitability of the coaching approach can also depend on
individuals’ personality traits [114]. By tailoring the coaching
technique, the CA could be perceived as more empathetic,
potentially making the coaching experience more enjoyable and
thereby increasing both adherence and the working alliance and
with them, treatment outcomes [112]. Further, we did not
investigate whether the prescribed exercises actually had an
effect on health outcomes. An adaptive algorithm that increases
the difficulty, repetition, and intensity of the exercises could be
implemented to ensure that the patient always performs slightly
above the comfort zone. This might have an effect not only on
health outcomes, but also on adherence. Lastly, the AR-based
part of HUC was not yet perceived to be mobile enough, and
this was a frequently expressed reason why patients were
doubtful about using HUC. Nevertheless, this study was first
and foremost about testing real-time feedback and a hands-free
interaction paradigm with a “human-sized” CA. The AR
hardware is currently changing and shrinking significantly [115],
and future work will therefore rely on significantly smaller AR
and will be able to test HUC on sufficiently more wearable
hardware.
Conclusions
This work provides evidence of the relevance and utility of
HUC that aims to increase adherence to home exercises. It
therefore contributes to the field of digital health by outlining
how CAs hosted by mobile and wearable technologies can
extend the reach and effectiveness of health care experts into
the everyday lives of patients. HUC may be promising not only
in the context of physiotherapy, as exemplarily elaborated in
this work, but also for various other conditions, such as obesity
or cardiovascular disease, as they also require intensive and
longitudinal behavioral support in the everyday lives of patients.
Authors' Contributions
TK, VE, LS, and EMH contributed to the study design. HG provided technical support for the study execution. KML, VE, and
TK contributed to the data analysis of the results. All authors provided critical review and revision of the manuscript. All authors
approved the manuscript before submission.
Conflicts of Interest
TK, KML, LS, and HG are affiliated with the Centre for Digital Health Interventions, a joint initiative of the Department of
Management, Technology and Economics at ETH Zurich and the Institute of Technology Management at the University of St
Gallen, which is funded in part by the Swiss health insurer CSS. TK is also a cofounder of Pathmate Technologies, a university
spin-off company that creates and delivers digital clinical pathways. However, Pathmate Technologies is not involved in any of
the 4 studies described in this paper. All other authors declare no conflict of interest.
Multimedia Appendix 1
Current State Commercial and Research Applications.
[DOCX File , 67 KB-Multimedia Appendix 1]
Multimedia Appendix 2
Storybook of the hybrid ubiquitous coaching approach.
[DOCX File , 179 KB-Multimedia Appendix 2]
J Med Internet Res 2021 | vol. 23 | iss. 2 | e23612 | p. 18https://www.jmir.org/2021/2/e23612 (page number not for citation purposes)
Kowatsch et alJOURNAL OF MEDICAL INTERNET RESEARCH
XSL
•
FO
RenderX
Multimedia Appendix 3
First-person view and interactions with the augmented reality-based conversational agent version 1 of study 1.
[MP4 File (MP4 Video), 165488 KB-Multimedia Appendix 3]
Multimedia Appendix 4
Technical details of the augmented reality-based conversational agents versions 1 and 2.
[DOCX File , 15 KB-Multimedia Appendix 4]
Multimedia Appendix 5
First-person view and interactions with the augmented reality-based conversational agent version 2.
[MP4 File (MP4 Video), 197084 KB-Multimedia Appendix 5]
Multimedia Appendix 6
Sketches of the hybrid ubiquitous coaching approach.
[PDF File (Adobe PDF File), 6067 KB-Multimedia Appendix 6]
Multimedia Appendix 7
Questions of the semistructured interview of study 2.
[DOCX File , 28 KB-Multimedia Appendix 7]
Multimedia Appendix 8
Video-based instructions of study 3.
[MP4 File (MP4 Video), 20604 KB-Multimedia Appendix 8]
Multimedia Appendix 9
Quantitative questionnaire items and depersonalized raw data of all 4 studies.
[XLSX File (Microsoft Excel File), 28 KB-Multimedia Appendix 9]
Multimedia Appendix 10
Questions of the semistructured interview of study 3.
[DOCX File , 15 KB-Multimedia Appendix 10]
Multimedia Appendix 11
Data analysis script written in Python for all 4 studies.
[ZIP File (Zip Archive), 3 KB-Multimedia Appendix 11]
Multimedia Appendix 12
Examples of the smartphone-based conversational turns of study 4.
[DOCX File , 987 KB-Multimedia Appendix 12]
References
1. Tsang A, Von Korff M, Lee S, Alonso J, Karam E, Angermeyer MC, et al. Common chronic pain conditions in developed
and developing countries: gender and age differences and comorbidity with depression-anxiety disorders. J Pain 2008
Oct;9(10):883-891. [doi: 10.1016/j.jpain.2008.05.005] [Medline: 18602869]
2. Briggs AM, Cross MJ, Hoy DG, Sànchez-Riera L, Blyth FM, Woolf AD, et al. Musculoskeletal Health Conditions Represent
a Global Threat to Healthy Aging: A Report for the 2015 World Health Organization World Report on Ageing and Health.
Gerontologist 2016 Apr;56 Suppl 2:S243-S255. [doi: 10.1093/geront/gnw002] [Medline: 26994264]
3. GBD 2017 Disease and Injury Incidence and Prevalence Collaborators. Global, regional, and national incidence, prevalence,
and years lived with disability for 354 diseases and injuries for 195 countries and territories, 1990-2017: a systematic
analysis for the Global Burden of Disease Study 2017. Lancet 2018 Nov 10;392(10159):1789-1858 [FREE Full text] [doi:
10.1016/S0140-6736(18)32279-7] [Medline: 30496104]
J Med Internet Res 2021 | vol. 23 | iss. 2 | e23612 | p. 19https://www.jmir.org/2021/2/e23612 (page number not for citation purposes)
Kowatsch et alJOURNAL OF MEDICAL INTERNET RESEARCH
XSL
•
FO
RenderX
4. Briggs AM, Woolf AD, Dreinhöfer K, Homb N, Hoy DG, Kopansky-Giles D, et al. Reducing the global burden of
musculoskeletal conditions. Bull World Health Organ 2018 May 01;96(5):366-368 [FREE Full text] [doi:
10.2471/BLT.17.204891] [Medline: 29875522]
5. March L, Smith EUR, Hoy DG, Cross MJ, Sanchez-Riera L, Blyth F, et al. Burden of disability due to musculoskeletal
(MSK) disorders. Best Pract Res Clin Rheumatol 2014 Jun;28(3):353-366. [doi: 10.1016/j.berh.2014.08.002] [Medline:
25481420]
6. Malik KM, Beckerly R, Imani F. Musculoskeletal Disorders a Universal Source of Pain and Disability Misunderstood and
Mismanaged: A Critical Analysis Based on the U.S. Model of Care. Anesth Pain Med 2018 Dec;8(6):e85532 [FREE Full
text] [doi: 10.5812/aapm.85532] [Medline: 30775292]
7. Friedrich M, Gittler G, Arendasy M, Friedrich KM. Long-term effect of a combined exercise and motivational program on
the level of disability of patients with chronic low back pain. Spine (Phila Pa 1976) 2005 May 01;30(9):995-1000. [doi:
10.1097/01.brs.0000160844.71551.af] [Medline: 15864148]
8. Hayden JA, van TMW, Tomlinson G. Systematic review: strategies for using exercise therapy to improve outcomes in
chronic low back pain. Ann Intern Med 2005 May 03;142(9):776-785. [Medline: 15867410]
9. Jack K, McLean SM, Moffett JK, Gardiner E. Barriers to treatment adherence in physiotherapy outpatient clinics: a systematic
review. Man Ther 2010 Jun;15(3):220-228 [FREE Full text] [doi: 10.1016/j.math.2009.12.004] [Medline: 20163979]
10. Shori G, Kapoor G, Talukdar P. Effectiveness of home-based physiotherapy on pain and disability in participants with
osteoarthritis of knee: an observational study. J Phys Ther Sci 2018 Oct;30(10):1232-1236 [FREE Full text] [doi:
10.1589/jpts.30.1232] [Medline: 30349155]
11. World Health Organization. Adherence to long-term therapies: evidence for action. 2003. URL: https://apps.who.int/iris/
bitstream/handle/10665/42682/9241545992.pdf [accessed 2021-01-18]
12. Campbell R, Evans M, Tucker M, Quilty B, Dieppe P, Donovan JL. Why don't patients do their exercises? Understanding
non-compliance with physiotherapy in patients with osteoarthritis of the knee. J Epidemiol Community Health 2001
Feb;55(2):132-138 [FREE Full text] [Medline: 11154253]
13. Friedrich M, Gittler G, Halberstadt Y, Cermak T, Heiller I. Combined exercise and motivation program: effect on the
compliance and level of disability of patients with chronic low back pain: a randomized controlled trial. Arch Phys Med
Rehabil 1998 May;79(5):475-487. [Medline: 9596385]
14. Beinart NA, Goodchild CE, Weinman JA, Ayis S, Godfrey EL. Individual and intervention-related factors associated with
adherence to home exercise in chronic low back pain: a systematic review. Spine J 2013 Dec;13(12):1940-1950. [doi:
10.1016/j.spinee.2013.08.027] [Medline: 24169445]
15. Austin S, Qu H, Shewchuk RM. Association between adherence to physical activity guidelines and health-related quality
of life among individuals with physician-diagnosed arthritis. Qual Life Res 2012 Oct;21(8):1347-1357. [doi:
10.1007/s11136-011-0046-x] [Medline: 22038394]
16. Campbell AG, Stafford JW, Holz T, O’Hare GMP. Why, when and how to use augmented reality agents (AuRAs). Virtual
Reality 2013 Dec 1;18(2):139-159. [doi: 10.1007/s10055-013-0239-4]
17. Vasey LM. DNAs and DNCTs — Why Do Patients Fail to Begin or to Complete a Course of Physiotherapy Treatment?
Physiotherapy 1990 Sep;76(9):575-578. [doi: 10.1016/s0031-9406(10)63052-0]
18. Medina-Mirapeix F, Escolar-Reina P, Gascón-Cánovas JJ, Montilla-Herrador J, Jimeno-Serrano FJ, Collins SM. Predictive
factors of adherence to frequency and duration components in home exercise programs for neck and low back pain: an
observational study. BMC Musculoskelet Disord 2009 Dec 09;10:155 [FREE Full text] [doi: 10.1186/1471-2474-10-155]
[Medline: 19995464]
19. Härkäpää K, Järvikoski A, Mellin G, Hurri H, Luoma J. Health locus of control beliefs and psychological distress as
predictors for treatment outcome in low-back pain patients: results of a 3-month follow-up of a controlled intervention
study. Pain 1991 Jul;46(1):35-41. [doi: 10.1016/0304-3959(91)90031-R] [Medline: 1832753]
20. Reilly K, Lovejoy B, Williams R, Roth H. Differences between a supervised and independent strength and conditioning
program with chronic low back syndromes. J Occup Med 1989 Jun;31(6):547-550. [doi:
10.1097/00043764-198906000-00012] [Medline: 2525182]
21. Kolt GS, McEvoy JF. Adherence to rehabilitation in patients with low back pain. Man Ther 2003 May;8(2):110-116.
[Medline: 12890439]
22. McLean SM, Burton M, Bradley L, Littlewood C. Interventions for enhancing adherence with physiotherapy: a systematic
review. Man Ther 2010 Dec;15(6):514-521. [doi: 10.1016/j.math.2010.05.012] [Medline: 20630793]
23. Smith M, Davis MA, Stano M, Whedon JM. Aging baby boomers and the rising cost of chronic back pain: secular trend
analysis of longitudinal Medical Expenditures Panel Survey data for years 2000 to 2007. J Manipulative Physiol Ther 2013
Jan;36(1):2-11 [FREE Full text] [doi: 10.1016/j.jmpt.2012.12.001] [Medline: 23380209]
24. Frost R, Levati S, McClurg D, Brady M, Williams B. What Adherence Measures Should Be Used in Trials of Home-Based
Rehabilitation Interventions? A Systematic Review of the Validity, Reliability, and Acceptability of Measures. Arch Phys
Med Rehabil 2017 Jun;98(6):1241-1256.e45. [doi: 10.1016/j.apmr.2016.08.482] [Medline: 27702555]
25. Friedrich M, Cermak T, Maderbacher P. The effect of brochure use versus therapist teaching on patients performing
therapeutic exercise and on changes in impairment status. Phys Ther 1996 Oct;76(10):1082-1088. [Medline: 8863761]
J Med Internet Res 2021 | vol. 23 | iss. 2 | e23612 | p. 20https://www.jmir.org/2021/2/e23612 (page number not for citation purposes)
Kowatsch et alJOURNAL OF MEDICAL INTERNET RESEARCH
XSL
•
FO
RenderX
26. Bachmann C, Oesch P, Bachmann S. Recommendations for Improving Adherence to Home-Based Exercise: A Systematic
Review. Phys Med Rehab Kuror 2017 Dec 18;28(01):20-31. [doi: 10.1055/s-0043-120527]
27. Alexandre NMC, Nordin M, Hiebert R, Campello M. Predictors of compliance with short-term treatment among patients
with back pain. Rev Panam Salud Publica 2002 Aug;12(2):86-94. [Medline: 12243693]
28. Iversen MD, Fossel AH, Katz JN. Enhancing function in older adults with chronic low back pain: a pilot study of endurance
training. Arch Phys Med Rehabil 2003 Sep;84(9):1324-1331. [doi: 10.1016/s0003-9993(03)00198-9] [Medline: 13680569]
29. Escarti A, Guzman J. Effects of feedback on self-efficacy, performance, and choice in an athletic task. Journal of Applied
Sport Psychology 1999 Mar 11;11(1):83-96. [doi: 10.1080/10413209908402952]
30. Martin LR, Williams SL, Haskard KB, Dimatteo MR. The challenge of patient adherence. Ther Clin Risk Manag 2005
Sep;1(3):189-199 [FREE Full text] [Medline: 18360559]
31. Chan DK, Lonsdale C, Ho PY, Yung PS, Chan KM. Patient motivation and adherence to postsurgery rehabilitation exercise
recommendations: the influence of physiotherapists' autonomy-supportive behaviors. Arch Phys Med Rehabil 2009
Dec;90(12):1977-1982. [doi: 10.1016/j.apmr.2009.05.024] [Medline: 19969157]
32. Basset S. The assessment of patient adherence to physiotherapy rehabilitation. New Zeal J Physiother 2003;31(2):60-66.
33. Picorelli AMA, Pereira LSM, Pereira DS, Felício D, Sherrington C. Adherence to exercise programs for older people is
influenced by program characteristics and personal factors: a systematic review. J Physiother 2014 Sep;60(3):151-156
[FREE Full text] [doi: 10.1016/j.jphys.2014.06.012] [Medline: 25092418]
34. Schoo A, Morris M, Bui Q. The effects of mode of exercise instruction on compliance with a home exercise program in
older adults with osteoarthritis. Physiotherapy 2005 Jun;91(2):79-86. [doi: 10.1016/j.physio.2004.09.019]
35. Schneiders A, Zusman M, Singer K. Exercise therapy compliance in acute low back pain patients. Manual Therapy 1998
Aug;3(3):147-152. [doi: 10.1016/S1356-689X(98)80005-2]
36. Chung BPH, Chiang WKH, Lau H, Lau TFO, Lai CWK, Sit CSY, et al. Pilot study on comparisons between the effectiveness
of mobile video-guided and paper-based home exercise programs on improving exercise adherence, self-efficacy for exercise
and functional outcomes of patients with stroke with 3-month follow-up: A single-blind randomized controlled trial. Hong
Kong Physiother J 2020 Jun;40(1):63-73 [FREE Full text] [doi: 10.1142/S1013702520500079] [Medline: 32489241]
37. Banos O, Nugent C. M-Coaching: Towards the Next Generation of Mobile-Driven Healthcare Support Services. Computer
2018 Aug;51(8):14-17. [doi: 10.1109/mc.2018.3191267]
38. Lambert TE, Harvey LA, Avdalis C, Chen LW, Jeyalingam S, Pratt CA, et al. An app with remote support achieves better
adherence to home exercise programs than paper handouts in people with musculoskeletal conditions: a randomised trial.
J Physiother 2017 Jul;63(3):161-167 [FREE Full text] [doi: 10.1016/j.jphys.2017.05.015] [Medline: 28662834]
39. Verapy Health. Verapy. 2016. URL: https://verapytherapy.com/ [accessed 2021-01-18]
40. Hamine S, Gerth-Guyette E, Faulx D, Green BB, Ginsburg AS. Impact of mHealth chronic disease management on treatment
adherence and patient outcomes: a systematic review. J Med Internet Res 2015;17(2):e52 [FREE Full text] [doi:
10.2196/jmir.3951] [Medline: 25803266]
41. Marcolino MS, Oliveira JAQ, D'Agostino M, Ribeiro AL, Alkmim MBM, Novillo-Ortiz D. The Impact of mHealth
Interventions: Systematic Review of Systematic Reviews. JMIR Mhealth Uhealth 2018 Jan 17;6(1):e23 [FREE Full text]
[doi: 10.2196/mhealth.8873] [Medline: 29343463]
42. Healure Technology. Healure: Physiotherapy Exercise Plans. 2017. URL: http://www.healure.com [accessed 2020-01-18]
43. Cassell J. Embodied conversational interface agents. Commun. ACM 2000 Apr;43(4):70-78. [doi: 10.1145/332051.332075]
44. Laranjo L, Dunn AG, Tong HL, Kocaballi AB, Chen J, Bashir R, et al. Conversational agents in healthcare: a systematic
review. J Am Med Inform Assoc 2018 Sep 01;25(9):1248-1258 [FREE Full text] [doi: 10.1093/jamia/ocy072] [Medline:
30010941]
45. de Cock C, Milne-Ives M, van Velthoven M, Alturkistani A, Lam C, Meinert E. Effectiveness of Conversational Agents
(Virtual Assistants) in Health Care: Protocol for a Systematic Review. JMIR Res Protoc 2020 Mar 9;9(3):e16934. [doi:
10.2196/16934]
46. Fitzpatrick KK, Darcy A, Vierhile M. Delivering Cognitive Behavior Therapy to Young Adults With Symptoms of Depression
and Anxiety Using a Fully Automated Conversational Agent (Woebot): A Randomized Controlled Trial. JMIR Ment Health
2017 Jun 06;4(2):e19 [FREE Full text] [doi: 10.2196/mental.7785] [Medline: 28588005]
47. Kowatsch T, Nißen M, Shih, Chen-Hsuan I, Rüegger D, Volland D, Filler A, et al. Text-based Healthcare Chatbots Supporting
Patient and Health Professional Teams: Preliminary Results of a Randomized Controlled Trial on Childhood Obesity. 2017
Aug Presented at: Embodied Agents for Behavior Change (PEACH2017) Workshop, co-located with the 17th International
Conference on Intelligent Virtual Agents (IVA 2017); August 27, 2017; Stockholm, Sweden p. 1-10.
48. Kramer J, Künzler F, Mishra V, Smith SN, Kotz D, Scholz U, et al. Which Components of a Smartphone Walking App
Help Users to Reach Personalized Step Goals? Results From an Optimization Trial. Ann Behav Med 2020 Jun
12;54(7):518-528 [FREE Full text] [doi: 10.1093/abm/kaaa002] [Medline: 32182353]
49. Hauser-Ulrich S, Künzli H, Meier-Peterhans D, Kowatsch T. A Smartphone-Based Health Care Chatbot to Promote
Self-Management of Chronic Pain (SELMA): Pilot Randomized Controlled Trial. JMIR Mhealth Uhealth 2020 Apr
03;8(4):e15806 [FREE Full text] [doi: 10.2196/15806] [Medline: 32242820]
J Med Internet Res 2021 | vol. 23 | iss. 2 | e23612 | p. 21https://www.jmir.org/2021/2/e23612 (page number not for citation purposes)
Kowatsch et alJOURNAL OF MEDICAL INTERNET RESEARCH
XSL
•
FO
RenderX
50. Prvu Bettger J, Green CL, Holmes DN, Chokshi A, Mather RC, Hoch BT, et al. Effects of Virtual Exercise Rehabilitation
In-Home Therapy Compared with Traditional Care After Total Knee Arthroplasty: VERITAS, a Randomized Controlled
Trial. J Bone Joint Surg Am 2020 Jan 15;102(2):101-109. [doi: 10.2106/JBJS.19.00695] [Medline: 31743238]
51. Vivira Health Lab GmbH. Vivira - Physiotherapy Exercises at Home. 2016. URL: https://www.vivira.com/ [accessed
2021-01-28]
52. Ma T, Chattopadhyay D, Sharifi H. Virtual humans in health-related interventions: A meta-analysis. New York, NY, USA:
ACM; 2019 May 04 Presented at: Extended Abstracts of the CHI Conference on Human Factors in Computing Systems
(CHI EA '19); May 4-9, 2019; Glasgow, UK p. 1-6. [doi: 10.1145/3290607.3312853]
53. Babylon Health. Babylon. 2016. URL: https://www.babylonhealth.com [accessed 2021-01-18]
54. Bickmore TW, Pfeifer LM, Byron D, Forsythe S, Henault LE, Jack BW, et al. Usability of conversational agents by patients
with inadequate health literacy: evidence from two clinical trials. J Health Commun 2010;15 Suppl 2:197-210. [doi:
10.1080/10810730.2010.499991] [Medline: 20845204]
55. GOREHA GmbH. XR Health. 2018. URL: https://www.xr.health/ [accessed 2021-01-18]
56. Nass C, Steuer J, Tauber ER. Computers are social actors. 1994 Presented at: CHI94: ACM Conference on Human Factors
in Computer Systems; April 1994; Boston, MA p. 72-78. [doi: 10.1145/259963.260288]
57. Bickmore T, Gruber A, Picard R. Establishing the computer-patient working alliance in automated health behavior change
interventions. Patient Educ Couns 2005 Oct;59(1):21-30. [doi: 10.1016/j.pec.2004.09.008] [Medline: 16198215]
58. Shamekhi A, Bickmore T, Lestoquoy A, Gardiner P. Augmenting group medical visits with conversational agents for stress
management behavior change. In: Persuasive Technology: Development and Implementation of Personalized Technologies
to Change Attitudes and Behaviors. 2017 Presented at: 12th International Conference, PERSUASIVE 2017; April 4-6,
2017; Amsterdam. [doi: 10.1007/978-3-319-55134-0_5]
59. Flückiger C, Del Re AC, Wampold BE, Horvath AO. The alliance in adult psychotherapy: A meta-analytic synthesis.
Psychotherapy (Chic) 2018 Dec;55(4):316-340. [doi: 10.1037/pst0000172] [Medline: 29792475]
60. Cassell J, Sullivan J, Prevost S, editors. Embodied Conversational Agents. Cambridge, MA: MIT Press; 2000.
61. Milgram P, Takemura H, Utsumi A, Kishino F. Augmented reality: a class of displays on the reality-virtuality continuum.
1995 Presented at: Photonics for Industrial Applications; October 31-November 4, 1994; Boston, MA p. 282-292. [doi:
10.1117/12.197321]
62. Bandura A. Self-efficacy mechanism in human agency. Am Psychol 1982;37(2):122.
63. Schwarzer R. Modeling Health Behavior Change: How to Predict and Modify the Adoption and Maintenance of Health
Behaviors. Applied Psychology 2008 Jan;57(1):1-29. [doi: 10.1111/j.1464-0597.2007.00325.x]
64. Andersson G, Carlbring P, Berger T, Almlöv J, Cuijpers P. What makes Internet therapy work? Cogn Behav Ther 2009;38
Suppl 1:55-60. [doi: 10.1080/16506070902916400] [Medline: 19675956]
65. Spek V, Cuijpers P, Nyklícek I, Riper H, Keyzer J, Pop V. Internet-based cognitive behaviour therapy for symptoms of
depression and anxiety: a meta-analysis. Psychol Med 2007 Mar;37(3):319-328. [doi: 10.1017/S0033291706008944]
[Medline: 17112400]
66. Zachariae R, Lyby MS, Ritterband LM, O'Toole MS. Efficacy of internet-delivered cognitive-behavioral therapy for
insomnia - A systematic review and meta-analysis of randomized controlled trials. Sleep Med Rev 2016 Dec;30:1-10. [doi:
10.1016/j.smrv.2015.10.004] [Medline: 26615572]
67. Kowatsch T, Nißen M, Rüegger D, Stieger M, Flückiger C, Allemand M, et al. The Impact of Interpersonal Closeness Cues
in Text-based Healthcare Chatbots on Attachment Bond and the Desire to Continue Interacting: An Experimental Design.
2018 Jun 23 Presented at: Proc 26th Eur Conf Inf Syst - ECIS; June 23-28 2018; Portsmouth, UK p. A.
68. Bickmore T, Schulman D, Yin L. Maintaining Engagement in Long-term Interventions with Relational Agents. Appl Artif
Intell 2010 Jul 01;24(6):648-666 [FREE Full text] [doi: 10.1080/08839514.2010.492259] [Medline: 21318052]
69. Horvath AO, Greenberg LS. Development and validation of the Working Alliance Inventory. Journal of Counseling
Psychology 1989 Apr;36(2):223-233. [doi: 10.1037/0022-0167.36.2.223]
70. Brintz CE, Miller S, Olmsted KR, Bartoszek M, Cartwright J, Kizakevich PN, et al. Adapting Mindfulness Training for
Military Service Members With Chronic Pain. Mil Med 2020 Mar 02;185(3-4):385-393 [FREE Full text] [doi:
10.1093/milmed/usz312] [Medline: 31621856]
71. Du S, Liu W, Cai S, Hu Y, Dong J. The efficacy of e-health in the self-management of chronic low back pain: A meta
analysis. Int J Nurs Stud 2020 Jun;106:103507. [doi: 10.1016/j.ijnurstu.2019.103507] [Medline: 32320936]
72. Bardram J, Friday A. Ubiquitous Computing Systems. In: Ubiquitous Computing Fundamentals. London, UK: Chapman
and Hall/CRC; 2018:51-108.
73. Weiser M. The Computer for the 21st Century. Sci Am 1991 Sep;265(3):94-104. [doi: 10.1038/scientificamerican0991-94]
74. Castonguay LG, Constantino MJ, Holtforth MG. The working alliance: Where are we and where should we go? Psychotherapy
(Chic) 2006;43(3):271-279. [doi: 10.1037/0033-3204.43.3.271] [Medline: 22122096]
75. Flückiger C, Del Re AC, Wampold BE, Symonds D, Horvath AO. How central is the alliance in psychotherapy? A multilevel
longitudinal meta-analysis. J Couns Psychol 2012 Jan;59(1):10-17. [doi: 10.1037/a0025749] [Medline: 21988681]
76. Daft RL, Lengel RH. Organizational Information Requirements, Media Richness and Structural Design. Management
Science 1986 May;32(5):554-571. [doi: 10.1287/mnsc.32.5.554]
J Med Internet Res 2021 | vol. 23 | iss. 2 | e23612 | p. 22https://www.jmir.org/2021/2/e23612 (page number not for citation purposes)
Kowatsch et alJOURNAL OF MEDICAL INTERNET RESEARCH
XSL
•
FO
RenderX
77. Tseng F, Cheng T, Li K, Teng C. How does media richness contribute to customer loyalty to mobile instant messaging?
Internet Research 2017 Jun 05;27(3):520-537. [doi: 10.1108/intr-06-2016-0181]
78. Michie S, Ashford S, Sniehotta FF, Dombrowski SU, Bishop A, French DP. A refined taxonomy of behaviour change
techniques to help people change their physical activity and healthy eating behaviours: the CALO-RE taxonomy. Psychol
Health 2011 Nov;26(11):1479-1498. [doi: 10.1080/08870446.2010.540664] [Medline: 21678185]
79. Horvath AO, Symonds BD. Relation between working alliance and outcome in psychotherapy: A meta-analysis. Journal
of Counseling Psychology 1991;38(2):139-149. [doi: 10.1037/0022-0167.38.2.139]
80. ter Stal S, Kramer LL, Tabak M, op den Akker H, Hermens H. Design Features of Embodied Conversational Agents in
eHealth: a Literature Review. International Journal of Human-Computer Studies 2020 Jun;138:102409. [doi:
10.1016/j.ijhcs.2020.102409]
81. Seo DW, Lee JY. Direct hand touchable interactions in augmented reality environments for natural and intuitive user
experiences. Expert Systems with Applications 2013 Jul;40(9):3784-3793. [doi: 10.1016/j.eswa.2012.12.091]
82. Schreier M. Fallauswahl. In: Mey G, Mruck K, editors. Handbuch Qualitative Forschung in Der Psychologie. Berlin,
Germany: Springer; 2010:238-251.
83. Venkatesh, Morris, Davis, Davis. User Acceptance of Information Technology: Toward a Unified View. MIS Quarterly
2003;27(3):425. [doi: 10.2307/30036540]
84. Venkatesh, Thong, Xu. Consumer Acceptance and Use of Information Technology: Extending the Unified Theory of
Acceptance and Use of Technology. MIS Quarterly 2012;36(1):157. [doi: 10.2307/41410412]
85. Davis FD, Bagozzi RP, Warshaw PR. User Acceptance of Computer Technology: A Comparison of Two Theoretical
Models. Management Science 1989 Aug;35(8):982-1003. [doi: 10.1287/mnsc.35.8.982]
86. Reichheld FF. The one number you need to grow. Harv Bus Rev 2003 Dec;81(12):46-54, 124. [Medline: 14712543]
87. Kamis, Koufaris, Stern. Using an Attribute-Based Decision Support System for User-Customized Products Online: An
Experimental Investigation. MIS Quarterly 2008;32(1):159. [doi: 10.2307/25148832]
88. Weller SC, Vickers B, Bernard HR, Blackburn AM, Borgatti S, Gravlee CC, et al. Open-ended interview questions and
saturation. PLoS One 2018;13(6):e0198606. [doi: 10.1371/journal.pone.0198606] [Medline: 29924873]
89. Flick U. An Introduction To Qualitative Research, 6th ed. London: SAGE Publications; 2018.
90. Bradley EH, Curry LA, Devers KJ. Qualitative data analysis for health services research: developing taxonomy, themes,
and theory. Health Serv Res 2007 Aug;42(4):1758-1772 [FREE Full text] [doi: 10.1111/j.1475-6773.2006.00684.x] [Medline:
17286625]
91. Moore GC, Benbasat I. Development of an Instrument to Measure the Perceptions of Adopting an Information Technology
Innovation. Information Systems Research 1991 Sep;2(3):192-222. [doi: 10.1287/isre.2.3.192]
92. Falkenström F, Hatcher RL, Skjulsvik T, Larsson MH, Holmqvist R. Development and validation of a 6-item working
alliance questionnaire for repeated administrations during psychotherapy. Psychol Assess 2015 Mar;27(1):169-183. [doi:
10.1037/pas0000038] [Medline: 25346997]
93. Armitage P, Hills M. The Two-Period Crossover Trial. The Statistician 1982 Jun;31(2):119. [doi: 10.2307/2987883]
94. Hart S, Staveland L. Development of NASA-TLX (Task Load Index): Results of empirical and theoretical research. Advances
in Psychology 1988;52:139-183. [doi: 10.1016/s0166-4115(08)62386-9]
95. Davis FD. Perceived Usefulness, Perceived Ease of Use, and User Acceptance of Information Technology. MIS Quarterly
1989 Sep;13(3):319. [doi: 10.2307/249008]
96. Bickmore TW, Mitchell SE, Jack BW, Paasche-Orlow MK, Pfeifer LM, Odonnell J. Response to a Relational Agent by
Hospital Patients with Depressive Symptoms. Interact Comput 2010 Jul 1;22(4):289-298 [FREE Full text] [doi:
10.1016/j.intcom.2009.12.001] [Medline: 20628581]
97. Smith J, Lewis J, Prichard D. Physiotherapy exercise programmes: are instructional exercise sheets effective? Physiother
Theory Pract 2005;21(2):93-102. [Medline: 16392462]
98. Dix A. Human–computer interaction: A stable discipline, a nascent science, and the growth of the long tail. Interacting
with Computers 2010 Jan;22(1):13-27. [doi: 10.1016/j.intcom.2009.11.007]
99. Razak FHA, Dix A. Doing a Single Person Study in Human Computer Interaction (HCI). Malaysian Journal of Computing
2010;1(1):40-61.
100. Parmar D, Olafsson S, Utami D, Bickmore T. Looking the Part: The Effect of Attire and Setting on Perceptions of a Virtual
Health Counselor. New York, NY, USA: ACM; 2018 Nov Presented at: IVA '18: Proceedings of the 18th International
Conference on Intelligent Virtual Agents; November; Sydney, Australia p. 301-306. [doi: 10.1145/3267851.3267915]
101. Creed C, Beale R, Cowan B. The Impact of an Embodied Agent's Emotional Expressions Over Multiple Interactions.
Interacting with Computers 2014 Jan 29;27(2):172-188. [doi: 10.1093/iwc/iwt064]
102. Riek L. Wizard of Oz Studies in HRI: A Systematic Review and New Reporting Guidelines. J Human-Robot Interact 2012
Aug 01;1(1):119-136. [doi: 10.5898/jhri.1.1.riek]
103. Huber S, Priebe JA, Baumann K, Plidschun A, Schiessl C, Tölle TR. Treatment of Low Back Pain with a Digital
Multidisciplinary Pain Treatment App: Short-Term Results. JMIR Rehabil Assist Technol 2017 Dec 04;4(2):e11 [FREE
Full text] [doi: 10.2196/rehab.9032] [Medline: 29203460]
J Med Internet Res 2021 | vol. 23 | iss. 2 | e23612 | p. 23https://www.jmir.org/2021/2/e23612 (page number not for citation purposes)
Kowatsch et alJOURNAL OF MEDICAL INTERNET RESEARCH
XSL
•
FO
RenderX
104. Donkin L, Glozier N. Motivators and motivations to persist with online psychological interventions: a qualitative study of
treatment completers. J Med Internet Res 2012;14(3):e91 [FREE Full text] [doi: 10.2196/jmir.2100] [Medline: 22743581]
105. Micksche M, Sonnbichler G, Isak K. Krebshilfe und Rehabilitation. In: Crevenna R, editor. Onkologische Rehabilitation
Grundlagen, Methoden, Verfahren und Wiedereingliederung. Berlin, Germany: Springer; 2020:295-306.
106. Bordin ES. The generalizability of the psychoanalytic concept of the working alliance. Psychotherapy: Theory, Research
& Practice 1979;16(3):252-260. [doi: 10.1037/h0085885]
107. Alves T, Natálio J, Henriques-Calado J, Gama S. Incorporating personality in user interface design: A review. Personality
and Individual Differences 2020 Mar;155:109709. [doi: 10.1016/j.paid.2019.109709]
108. Sommers J, Engelbert RHH, Dettling-Ihnenfeldt D, Gosselink R, Spronk PE, Nollet F, et al. Physiotherapy in the intensive
care unit: an evidence-based, expert driven, practical statement and rehabilitation recommendations. Clin Rehabil 2015
Nov;29(11):1051-1063 [FREE Full text] [doi: 10.1177/0269215514567156] [Medline: 25681407]
109. Agrawal R, Imieliński T, Swami A. Mining association rules between sets of items in large databases. New York, NY,
USA: ACM; 1993 Presented at: Proceedings of the 1993 ACM SIGMOD international conference on Management of data
(SIGMOD '93); May 26-28, 1993; Washington, DC, USA p. 207-216.
110. Delgadillo J, Rubel J, Barkham M. Towards personalized allocation of patients to therapists. J Consult Clin Psychol 2020
Sep;88(9):799-808. [doi: 10.1037/ccp0000507] [Medline: 32378910]
111. Halfon P, Eggli Y, Morel Y, Taffé P. The effect of patient, provider and financing regulations on the intensity of ambulatory
physical therapy episodes: a multilevel analysis based on routinely available data. BMC Health Serv Res 2015 Feb 07;15:52
[FREE Full text] [doi: 10.1186/s12913-015-0686-6] [Medline: 25889368]
112. Chen R, Rafaeli E, Ziv-Beiman S, Bar-Kalifa E, Solomonov N, Barber JP, et al. Therapeutic technique diversity is linked
to quality of working alliance and client functioning following alliance ruptures. J Consult Clin Psychol 2020
Sep;88(9):844-858. [doi: 10.1037/ccp0000490] [Medline: 32584116]
113. Fadhil A, Schiavo G. Designing for Health Chatbots. ArXiv. Preprint posted online on February 24, 2019 [FREE Full text]
114. Yorita A, Egerton S, Oakman J, Chan C, Kubota N. Self-Adapting Chatbot Personalities for Better Peer Support. : IEEE;
2019 Presented at: 2019 IEEE International Conference on Systems, Man and Cybernetics (SMC); 2019; Bari, Italy p.
4094-4100. [doi: 10.1109/smc.2019.8914583]
115. Wilkins M. Optical headwear for augmented reality. United States BAE SYSTEMS plc (London, GB) D865042. 2019 Oct
29. URL: https://www.freepatentsonline.com/D865042.html [accessed 2021-01-18]
Abbreviations
AR: augmented reality
BCT: behavioral change technique
CA: conversational agent
HUC: hybrid ubiquitous coaching
MSD: musculoskeletal disorder
NPS: Net Promoter Score
QR: quick response
RQ: research question
TAD: treatment adherence dimension
Edited by G Eysenbach; submitted 25.08.20; peer-reviewed by A Lau, J Salisbury; comments to author 16.09.20; revised version
received 28.09.20; accepted 18.01.21; published 22.02.21
Please cite as:
Kowatsch T, Lohse KM, Erb V, Schittenhelm L, Galliker H, Lehner R, Huang EM
Hybrid Ubiquitous Coaching With a Novel Combination of Mobile and Holographic Conversational Agents Targeting Adherence to
Home Exercises: Four Design and Evaluation Studies
J Med Internet Res 2021;23(2):e23612
URL: https://www.jmir.org/2021/2/e23612
doi: 10.2196/23612
PMID:
©Tobias Kowatsch, Kim-Morgaine Lohse, Valérie Erb, Leo Schittenhelm, Helen Galliker, Rea Lehner, Elaine M Huang. Originally
published in the Journal of Medical Internet Research (http://www.jmir.org), 22.02.2021. This is an open-access article distributed
under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits
unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in the Journal of
J Med Internet Res 2021 | vol. 23 | iss. 2 | e23612 | p. 24https://www.jmir.org/2021/2/e23612 (page number not for citation purposes)
Kowatsch et alJOURNAL OF MEDICAL INTERNET RESEARCH
XSL
•
FO
RenderX
Medical Internet Research, is properly cited. The complete bibliographic information, a link to the original publication on
http://www.jmir.org/, as well as this copyright and license information must be included.
J Med Internet Res 2021 | vol. 23 | iss. 2 | e23612 | p. 25https://www.jmir.org/2021/2/e23612 (page number not for citation purposes)
Kowatsch et alJOURNAL OF MEDICAL INTERNET RESEARCH
XSL
•
FO
RenderX
