Heart failure is a common disorder associated with high morbidity and mortality. Early diagnosis and treatment of exacerbations can lower the amount of (re-)hospitalizations. Patients can be supported to self-manage their disease by integrating persuasive coaching in a telemonitoring technology. The Twente TEACH Consortium is a multidisciplinary partnership, under which the iMediSense telemonitoring technology was developed. A mixed-methods approach was used to evaluate and improve the behavioral support of this platform. Methods included log data analysis, stakeholder interviews, usability tests, and a scoping literature review. Results showed that iMediSense is easy to use, achieving high adherence in a sixty days pilot study. A conceptual behavior change module grounded in goal setting theory was developed to provide persuasive coaching. We discuss the potential of our method, the implications of our findings, and present our ideas for further research to advance knowledge of a datadriven persuasive coaching approach to support behavior change.
1
Chronic congestive heart failure (CHF) is a chronic disorder in which the pumping
function of the heart is impaired. CHF has an incidence of 5-10 per 1000 annually [
Telemonitoring technology is a potential solution to provide self-management
support, as it enables both the patient and the caregiver to check up on vital signs of the
patient at home, which could decrease the number of visits to the hospital. However,
providing patients with their own data is not enough to change behavior, strategies to
sustain motivation and engagement are also necessary [
Filling the gap between collecting data and changing behavior can depend more on
the design of engagement strategies than on the specific features or functions of a
technology [
The Twente TEACH consortium is a partnership formed by the University of Twente, Thales, Ziekenhuisgroep Twente (ZGT), Vodafone, and by the healthcare insurance company Menzis. This multidisciplinary cooperation aims to support self-management of patients with CHF with the use of telemonitoring and data-driven persuasive coaching.
Following a holistic and participatory development approach [
There are two main interfaces in the iMediSense platform: the patient interface, which consists of an Android application, and a web application that serves as interface for the caregiver. Both interfaces communicate with a central server. This interface operates on a tablet with an Android operating system. Two sensors communicate with the tablet via Bluetooth: a non-invasive blood pressure monitor and a digital weighing scale. This allows for automatic registration of measurement values. The iMediSense application transmits the encrypted measurement via the 3G/4G wireless network to a hospital network server. As a result, measurements can be performed anywhere, as long as there is a connection with the cellular network. The application menu provides the following main services: registration and transmission of measurements, historical review of measurements, message functionality with the caregiver, profile and settings, and user manual. 3.2
For the caregiver, iMediSense is a web application, which can be accessed from a PC within the secured hospital network. The interface of this application consists of the following main features: patient analysis (e.g., to see measurement values), patient management (e.g., adjusting patient’s alarm thresholds), alarm notification (e.g., if measurement value exceeds a personalized threshold), and message notification (e.g., to send and receive messages to the patients). 3.3
The Android application communicates with the hospital network server through the regular wireless network connection. The caregiver can access the data on the network server from a PC within the hospital network. Figure 1 shows an overview of it.
All transmitted and stored data is encrypted. This includes measurements, messages, settings, and log data. Although not visible in Figure 1, data stored within the system is secured via a Demilitarized Zone environment, which acts as a buffer and communicates with both the internal hospital network and the outside world. Individual environments are separated by firewalls, and transfer between them makes use of security technologies. 4 4.1
A pilot was conducted to determine the use, usability, and usefulness for practice of iMediSense. Specifically, the intention was to know how the patients use, experience, and perceive usefulness of the technology. The leading question on behavioral support was “How should these patients be coached?” Thus, this pilot can be seen as formative evaluation of the current design of iMediSense. The goal was to deliver recommendations to further improve the system in preparation for a large scale implementation.
The pilot included in total twenty-five patients with CHF, who used iMediSense daily during a period of around sixty days each. Patients were selected using the following inclusion criteria: age >18 years; New York Heart Association (NYHA) functional classification II-III; stable CHF, stable symptoms, stable on medication and no admissions within 1 month; and able to provide written informed consent. After obtaining consent, patients received a personal introduction and training from either the researcher or a research nurse in the use of the technology. The training included practice in using the tablet, using the digital blood pressure meter and using the digital weighing scale. During the training patients had to execute at least one measurement by themselves. The training was considered to be completed if the following four conditions were met: 1) execution of at least one (guided) measurement; 2) patient had seen and/or used all menus of the system; 3) patient was confident about conducting measurements themselves; 4) trainer had the confidence that the patient is able to conduct measurements independently.
For the duration of the pilot, patients were instructed to conduct one measurement every morning after the first micturition but before breakfast (including blood pressure, heart rate, weight, and filling up the short questionnaire on health issues). Adherence was therefore defined as the compliance of patients to that instruction (one completed measurement per day). In case of emergencies or health issues, they were instructed to follow the regular guidelines in case of health issues or emergencies.
This study included both qualitative and quantitative research methods:
questionnaires, interviews, usability testing, and log data. Questionnaires were used to
determine the quality of life (via the Dutch version of the EQ5D5L QoL assessment [
Usability testing was performed to determine the use and usability of the technology.
The ‘think aloud’ method was employed [
To create an optimal fit between patients and the technology, the usability tests also
included two “non-users”, these were patients recruited only for the usability tests,
without any experience or training with the iMediSense technology. This was important
to assess because improving intuitivity of the system could increase its efficiency, since
requiring no extensive training would facilitate the use of the technology. The tests
were recorded, transcribed verbatim, and coded for the use of different parts of the
technology. For the analysis, the design observations were sorted according to several
levels [
Log data was used to provide insight in the actual use of the technology. A log data
protocol was applied to collect, transform, and analyze the log files [
For the analysis, sessions were identified from the raw data. A session was defined as a period of activity ended by a period of at least 30 minutes of inactivity. For each session, the used functionalities were determined, enabling analysis, for instance, of the registration and transmission of measurements, the review of the measurements (history), and the messaging function. 4.2
The aim of the second study was to develop a first concept of a behavior change module
for the platform by learning from the evidence of previous research. For this, goal
setting was proposed as the most viable key component because of its added value to
support this target group, as “the act of goal setting motivates the development and use of
self-management skills that increase the likelihood of goal attainment” (p. 431) [
Mechanisms
Moderators
Outcomes Directive func
tion Energizing func
tion Persistence Task-relevant knowledge and strategies
Feedback (on progress towards a goal) Task complexity Commitment - Importance - Self-efficacy
Task performance - Behavior - Outcome
Satisfaction Satisfaction paradox External incentives Feedback (no goal provided) Others (e.g., money)
Goals considering the source:
Personal / Self-set goals Assigned goals
Participatory / Collaborative goals
Guided goals to influence performance should consider:
Goal difficulty
Goal specificity
Proximal or Distal goals
The literature findings were complemented with an additional analysis of the log files
from Study 1. This second log data analysis aimed to provide insight and a deeper
understanding of the usage of the platform at an individual level. The analysis followed
the same protocol from the pilot study [
The participants were predominantly in NYHA functional class 2, 10 out of 25 had a history of hospitalization for decompensated heart failure. Comorbidities were frequent among the sample: There were 8 cases also diagnosed with hypertension, 7 with atrial fibrillation, 7 with diabetes mellitus, 7 with cerebral vascular accident or transient ischemic attack, and 4 with chronic obstructive pulmonary disease.
The results from the EHEALS eHealth literacy showed the median of the score as 3 (out of 5) with an interquartile range of 0.9. This suggests acceptable levels of comfort and perceived skill in using information technology for health among the sample. The mean score on the EQ5D5L was 8.2 on a scale from 5 to 25, which indicates a good quality of life with slight problems or health issues. No differences in quality of life were observed after 2 months of use of the technology.
During the usability tests, all subjects were able to log in, execute a measurement and review their measurements (8 users and 2 non-users). Four users and both nonusers were not able to send a message to the caregiver using iMediSense. Regarding potential coaching support, some positive observations for the “technology” (hardware related) were that the system was easy to use and that it does not take a lot of time, plus the automatic transmission of data was perceived as a useful feature. On the other hand, some critical observations on the “service” (provided by the technology) were that some patients did not understand how to send a message, due to being unclear which box from the interface had to be used to write it.
All patients were interviewed, 8 face to face, and 17 over the phone. After
independent coding of 10% of the quotes by two raters and discussing of disagreements, an
interrater agreement (Cohen’s Kappa [
The log data results showed that all patients were able to use the technology at home
and that it was predominantly used as intended. In total, 1572 sessions were identified
(See Table 1). Twelve out of 25 patients showed perfect or nearly perfect adherence
(one measurement every day). Some patients were unable to perform measurements for
several days due to technical issues, while others were less motivated to continue the
measurements and stopped performing them before the end of the study period. An
analysis of the navigation routes was performed to identify the most common routes
within one session. 1442 sessions were included, eliminating the routes that were
followed less than 10 sessions. It was found that the most common route was: 1)
conducting a measurement after login (n=1341), then 2) sending the measurement (n=1282),
and finally 3) quitting the application (n=418). Finally, as can be seen in Table 1, certain
features of iMediSense were barely used by the patients, such as the message function.
Scoping Review. Thirteen studies were selected from the scoping review that described
or discussed eHealth technology focused on supporting self-management of CHF (or
related conditions) with goal setting and self-monitoring as key components. Among
them, five different technologies were described or evaluated. The SMART2/CHF
PSMS system [
1. Monitoring feedback 2. Collaborative External incentive
Goal setting Core features
Fig. 4. Conceptual goal setting module to support self-care behaviors of CHF The conceptual goal setting module focuses on enhancing self-care of CHF patients by aiming to increase their confidence to perform the recommended behaviors through facilitating education and practice over time. Furthermore, the module requires a general baseline assessment and enquiries on self-care behaviors, requesting the users to input their confidence in their skill to perform each of them (e.g., confidence to increase physical activity or to reduce salt intake). This derives on the system giving feedback and advice based on their results, the users are consequently prompted to set a goal for a specific one, and this is where the coaching begins.
Log Data Analysis. Log data analysis allowed to contrast the individual use and usage of the three representative cases. The case-by-case comparison can be seen in Table 2. Moreover, the log data analysis of these cases also allowed to visualize the distinct patterns of activities across sessions for each individual. Although all users show acceptable adherence rates, the patterns in Figure 5 show how users varied in their use of the iMediSense platform. In Figure 5 the blue dots mark a session and its position in the left Y axis equals the duration of each. The X axis is divided by the log file “cases” and the beginning of every week is marked by a number (1st to 9th). The irregular distance between each week number represents a higher or lower number of interactions with the technology (cases registered in the log file). Matching the recommendations of usage, most sessions across all three cases consisted of one measurement being sent per usage session. It can be observed, from Table 2 and Figure 5, that the activity increases progressively when comparing the lower adherent with the adherent, and then again with the high self-monitoring. The high self-monitoring case showed a remarkably longer interaction with the technology during the 60 days of the pilot (In Table 2, “Total time spent in application”: 2.5 hours, 7 hours, and 12 hours), as well as a greater number of measurements not just during the morning but at several points of the day. In Figure 5, there are data points outside the scope of the “Adherent and high self-monitoring” graph due to sessions at several points in the day by this user, the range of the graphs was kept the same across all users for comparison purposes. However, it can also be perceived that both the adherent and the high selfmonitoring case were actively browsing the history function. The difference in Table 2 in the percentage of sessions with “history opened” is mostly due to the higher amount of “measurement-only” sessions from the high self-monitoring individual. Usage of the message function was so low across all patients that no visible trends can be spotted in Figure 5. 6
Overall, the research on the iMediSense platform has been successful not just for evaluation but most importantly to guide development seeking further improvement in terms of its persuasiveness and coaching support. One of the biggest achievements in terms of facilitating persuasive coaching is that the platform provides self-monitoring support using a simplified guidance of the users in a step-by-step process with a lack of redundant options. In a persuasive data-driven coaching approach, we hold as a premise that keeping self-monitoring as a simple process is important to promote adherence among the patients, which consequently assists on the identification of exacerbation in time through data analysis. From the usage and the usability tests we also know that some features require further improvement, such as the message function. We identified problems with the interface layout on this part of the platform, but even accounting for this the usage was lower than expected during the pilot.
The pilot study contributed to advance our knowledge regarding behavior support through the combination of telemonitoring and persuasive coaching. Findings from Study 1 provided key recommendations on how to coach CHF patients, namely to provide: education and training to use the technology as intended, personalized CHF education, support for the interpretation of measurement values, and support for achieving personal goals. Based on those grounds, the second study generated a first concept of a coaching module that made use of self-monitoring data, although further translation of the concept into actual functionalities, validation, and actual implementation is still due. Derived from our findings through the literature review, we know that a goal setting module can be potentially of added value for behavior support. According to what can be interpreted from the literature, self-management support through goal setting was found to be focused on maintenance goals (e.g., adhering to recommendations), with the most frequent target being increasing the patient’s self-confidence to perform specific and tailored self-management behaviors.
Indeed, an interesting finding derived from our theoretical basis is that the target of
the module is not to directly strive towards health-related goal attainment, but instead
focusing on a key self-care mechanism to promote sustained behavior. In line with the
precepts of goal setting theory, feedback has been integrated in previous technologies
to serve a double function. First, as an external incentive to prompt goal setting. Second,
as a facilitator of behavior change when applied to show progress towards the goals
that are set. The analysis of log data allowed observation of the distinct ways in which
patients used the platform, which raised interesting questions regarding goal setting
support. For instance, users showed diverse motivations to use iMediSense (observed
by levels of adherence, usage of features, and interviews). However, the literature
sometimes did not provide sufficient evidence on how to operationalize goal setting in
terms of these needs. For example, the high self-monitoring case serves as an example
that some patients already have self-set goals based on self-care behaviors (Case 3
aimed at self-monitoring symptoms around three times per day). Despite this
representative case, the literature did not include any case where goal setting was
implemented to influence monitoring, rather than maintenance of the disease. Additionally,
support to other forms of goal setting operationalization was unclear when looking at
both types of evidence, such as the approach of using this technique to “provide a safe
learning environment to practice self-care” [
Therefore, it is also possible to assume that our methods could not allow a full mapping of all of the possibilities by which goal setting can be of use when integrated in a telemonitoring technology. In the end, by leveraging on the strongest part of the evidence, the coaching module seeks to fit with the approach of the system in terms of its simplicity and avoidance of redundant options. For example, rather than allowing users to openly self-set goals, collaborative goal setting assisted by the system, the caregiver, or both, could provide support by making it easier to engage in coaching. 7
Research on the iMediSense platform has delivered a lot of different and extensive data. Thus, we are able to envision the huge potential of a data-driven persuasive coaching approach once greater automatization and personalization is achieved.
First of all, although not addressed in this paper, some of our findings from the caregiver perspectives brought up the added value of eHealth technology like iMediSense in current care organizations. However, full integration of a data-driven approach usually means new work processes and skills for caregivers. How to accomplish this integration is an interesting topic for future research.
Moreover, another possible step is to create and validate user profiles that can be used to tailor support, and mock-ups that can lead design decision-making, especially as the technology scales up. For instance, since participants from the pilot were mostly stable CHF patients and had acceptable eHealth literacy, further validation with unstable patients, with lower literacy, and bigger samples will be required. Furthermore, another topic for future work is the improvement of the system that generates the alarms, which is what guides the use of iMediSense, seeking to allow an early detection of deterioration of the CHF patient. In this matter, we also conducted research on improving the clinical significance of these alarms. This was done by testing and developing personalization algorithms for its alarm system. This analysis found that implementing a moving average algorithm could reduce the amount of false positives and increase the amount of true positives alarms generated by iMediSense. The most successful algorithm was based on systolic blood pressure and heart rate, and then comparing these to preset thresholds. Algorithms based on weight triggered alarms reduced false positives but did not generate alarms of clinical significance. An enhanced alarm system could have the potential to reduce the work load of caregivers by allowing them to focus on the priority cases that are being monitored by iMediSense. However, the trade-off between reducing false positives at the risk of also increasing false negatives demands further research. Finally, additional work is required to validate the mechanics of this goal setting module through iterative evaluation. For example, the underlying mechanisms and outcomes from the original theoretical model (Fig. 3) remain unaddressed by research. Expanding our simplified model by testing these assumptions could be a worthy aim for future studies.
Eventually, if positive results are found, the module could also be extended for new diseases by considering different self-care behaviors along with their key moderators (e.g., enhancing adherence to medication in patients with diabetes), or by integrating new sensors (e.g., adding a physical activity tracker). As the structure of the iMediSense platform allows for modular expansion, in the future we also plan to address other chronic conditions such as other types of cardiovascular diseases, diabetes, or diabetic foot and its complications.
Acknowledgements. We would like to thank and acknowledge Emilie Klaver for carrying out the research focused on improving the alarm system, from which implications for future work were derived and mentioned in this paper.