DiabetesCoach, a Web-based application for supporting self-care among patients with type 2 diabetes, was developed to encourage patients to play a more active role in their own care. The Web application is a low-tech solution for a large group of patients and was provided free of charge as a supplement to regular diabetes care. The application was developed by Medicinfo (Tilburg, Netherlands) in close collaboration with general practitioners, nurses, patients, behavioral scientists, and vendors (ie, health insurance companies). Initial development costs were relatively limited, and the running costs of the application were low. Therefore, a rise in use would not lead to an exponential rise in costs.

The following are the core features of the DiabetesCoach:

The patients’ self-monitored data were made available to the nurses with alerts signaling alarming metabolic values. Each nurse had access to each of her own patients’ DiabetesCoach details via a private account (protected by username and password). The Web application (not integrated with the nurse’s medical record) enabled nurses to set individual goals for their patients, add selected lifestyle programs, and highlights the appropriate chapter of the e-learning program. The patients received no particular instructions with regard to how often they should log on to DiabetesCoach. Patients measured metabolic values both at home and at the primary care practice during office visits. Nurses were allowed to have two extra consultation sessions per patient to compensate for the extra time needed to participate in the study. The information and guidelines provided in DiabetesCoach were in accordance with diabetes care standards and protocols in the Netherlands.

A primary health care foundation in the Netherlands consisting of 10 primary health care practices and a home care organization employing the diabetes nurses (n = 6) agreed to become partners in the pilot. Three primary health care practices volunteered to take part in the DiabetesCoach project.

The selection criteria for patient enrollment were (1) patients with type 2 diabetes mellitus (the primary focus was on fostering lifestyle changes), (2) patients being motivated to perform self-care activities, and (3) patients having access to the Internet and being sufficiently skilled to use the Internet. Through a recruitment letter, 350 patients were invited by the caregivers to use DiabetesCoach. Patients were informed about the purpose and possibilities of the Web application both through the letter and during the office visit. In total, 50 of the 350 invited patients (14%) enrolled in the project and filled out the informed consent forms.

Training sessions (offline) were set up for the enrollees. During the training sessions the participants received instructions on how to use the application, plus a user manual. Also, an email functionality was created for technical support.

We used a mixed-methods research design [24,25] to explore the conditions for long-term use of a Web application among patients with type 2 diabetes. Through usability tests and interviews we were able to explain the actual usage, and the survey provided insights into who uses the technology. All of the results combined provided an insight into the usage pattern and preferences of individual users for specific technology features. Log files enabled us to assess the actual and long-term usage (24 months) of the technology features. Table 1 presents an overview of the research instruments and the accompanying characteristics of the study. Figure 1 presents a chronology of the data collection process.

Directly after collecting the responses to the invitation letter, the nurses interviewed 226 of the 300 nonenrollees (patients who chose not to participate) during the office visit. Using an open-ended question the nurses asked the nonenrollees about their reasons for nonenrollment.

A paper-based survey was administered at baseline during the training sessions of the enrollees (n = 50) to assess patients’ demographics and health-related characteristics (user profiles). The survey consisted of seven closed questions on age, gender, education, health status, diabetes duration, diabetes treatment (medication use), and treatment satisfaction. In total, 42 patients returned the survey completely filled out.

Making use of log files (n = 50), we measured the 2-year usage pattern by patients, the number of log-ins by patients, the mean number of hits by patients of the system’s core features, and the content of the patient–nurse email messages.

We performed u sability tests (n = 20) after 3 months of usage to investigate patients’ experiences with using the Web application. Part 1 of the test consisted of a semistructured interview with open-ended questions aimed at assessing the patients’ eHealth literacy, their reasons for using the Web application, and their positive or negative experiences with using the system based on the critical incidents technique [26,27]. Part 2 contained several tasks related to each feature of the Web application to identify the problems that occurred during real-time use. A trained observer (NN) watched users communicating with the interface of the application while doing simulated tasks and thinking aloud [28]. The participants’ activities were recorded with audiovisual equipment (MORAE version 2.1; TechSmith Corporation, Okemos, MI, USA). The sessions were carried out at the participants’ home or at the health care practice. Each test lasted for about 90 minutes.

One year after the initial use of the Web application, emails were sent to those patients (n = 20) who were not actively using the application by that time. Through an open-ended question patients were asked to report their reason for discontinueing use. We received six responses.

Only 14% (n = 50) of the 350 patients responded positively to the invitation to use the Web application. Nurses interviewed 226 nonenrollees to gain insight into the barriers that inhibited their enrollment. The reasons given (n = 226) were lack of Internet (146/226, 65%), use will not have any added value (25/226, 11%), not in the mood to spend much time on the computer (23/226, 10%), not in the mood to be occupied with the disease (10/226, 4%), lack of skills to use the Internet (10/226, 4%), too busy or no time (4/226, 2%), or other, such as “patient is about to move to another town” (8/226, 4%). Obviously, patients experienced more external barriers to access (not having the equipment and lacking the right skills: 156/226, 69%) than internal motivational barriers (not willing to use it, no added value, too busy: 62/226, 27%).

Multimedia Appendix 2 and Multimedia Appendix 3 present an overview of the monthly use of the Web application during the study period (24 months). It can be seen that use of the Web application fluctuated over time. There was no fixed regimen; each patient used the DiabetesCoach whenever it suited them (free use).

Three groups of users could be distinguished:

(1) Continuous users who are highly active; hardcore users (n = 16):

(b) Continuous users, but with lower levels of activity (n = 18):

(c) Discontinued (inactive) users (n = 16):

Figure 4 presents user activity over a sustained period of time. About 66% of the enrollees continued using the Web application. Of those regular visitors, 30% can be defined as hardcore users; patients who are highly active in using the Web application.

Multimedia Appendix 2 and Multimedia Appendix 3 show that all patients from practice 1 were continuous users, whereas patients from practice 2 were more likely to be discontinued users. One possible reason for this is the closer contact between the patients and their nurse; the nurse of practice 1 was more actively involved in email contact (interactive feedback) with her patients than the nurses of practice 2 and 3 (respectively 4.5, 3.8, and 2.4 messages sent per patient on average).

When taking into account patient characteristics, the discontinued users did not differ substantially from the continuous users, although more of the discontinued users tended not to be taking medication (11/12, 92%).

We believe that more engagement in system use (being highly active) might result in better adherence to self-care activities. This is why we compared highly active users versus low/inactive users with respect to their characteristics and preferences.

We expected that patients with a greater need for care, such as the elderly, people on medication, and patients who had diabetes for a longer time, would benefit most from the technology and would therefore be more inclined to use the Web application. The results displayed in Table 4 show that highly active users were significantly more often medication users than low/inactive users were (2-sided Fisher exact test, P = .005) and had a significantly longer diabetes duration (1-sided analysis of variance, F _{1 , 41} = 5.0, P = .03).

Table 5 presents an overview of the core features and ranks them according to the features that were used most:

Highly active users seemed to have other goals than low/inactive users. Highly active users had a higher need for online monitoring, probably because they were more likely to be frequent medication users who regularly had to pass on their clinical values to their nurse. Particularly for these patients, online monitoring would be convenient (increased access). Low/inactive users, on the other hand, appreciated the ability to document personal details such as treatment plans and medication use (comparable with a personal health record).

The features online monitoring, email, and personal data appealed to both groups, yet the highly active users used all of the features more often, spread over a longer period of time (see Table 6). In particular, they used the interactive features of online monitoring and email more extensively.

The aim of this study was to explore the factors that influenced the use of a Web-based application for supporting the self-care of patients with type 2 diabetes. The major advantages of using the Web application were improved access to care and enhanced patient–nurse communication. The features that appealed to the patients most, and with which they were often engaged, were online monitoring in combination with personal feedback through email and documentation of medication usage. These personalized and interactive features stimulated active participation by both the patient and the nurse. Patients felt better monitored by means of the continuously received feedback and were more motivated to take a more active role in self-managing their diabetes.

Unexpectedly, there was a high preference for the documentation of personal data referring to medication and treatment plans. The documentation feature is not interactive; no communication takes place. However, it is comparable, in a certain way, with a personal health record [35-37], as it includes relevant data over the course of the individual’s lifetime. Patients liked to track how medication use affects their health. The personalized aspect means a lot to the patients.

To foster the widespread use of eHealth technologies like the DiabetesCoach, Internet use should be encouraged among the 65+ age range of the population; it is among the elderly that we have the largest growth potential [41,42]. To do so, the primary health care practices could consider providing training in computer and Internet skills and an opportunity to use a computer with Internet access in the practice itself for those who cannot afford the technology.

Furthermore, we believe that the less motivated or relatively unhealthy patients could benefit the most from the use of eHealth technologies because of their greater need for care and their greater challenge for health improvement. Verheijden et al [39] found that patients with deteriorating health conditions, who are thus more health care dependent, benefitted more from system use and were therefore more inclined to persistently use the Web application. In this study we found evidence, albeit very tentative, for our assumption that use of medication and the duration of the diabetes contributed to technology engagement; highly active users of DiabetesCoach were significantly more often medication users and had significantly longer duration of diabetes. Our results correspond to the findings of Wu et al [43], who found in their study of patients with chronic heart failure that those who used the system had more symptoms. The findings suggest that patients with worse disease conditions are most likely to benefit from eHealth applications. It is therefore expected that the Web application could be most useful for patients who use insulin and have a recent diagnosis of diabetes. Future studies should focus on the encouragement of eHealth among patient populations who can get the most out of it, such as those populations with high rates of behavioral risk factors and multiple chronic conditions [19,44].

In order to understand and overcome technical flaws, users should be able to give feedback during usage so that the system can be fine-tuned to their needs and user profiles. Preferably, users should actively participate in the development of the content (health 2.0) [45]. Patient-centered and -participatory design methods should be used when developing eHealth applications in order to ensure high-quality, user-informed products of demonstrated effectiveness [6,8,46-49]. Through such design approaches we are better able to customize the technology to individual preferences and user profiles. This means that the design of eHealth should start with a careful analysis of individual needs and accompanying system requirements to explore which technology is best suited for whom.

To increase adherence, technology should have push factors for persistence such as feedback mechanisms and triggers [50]. As such, it is relevant to know what kind of technology features or cues would trigger users, such as through words, images, or sounds. Reminders or triggers for use could be sent via text messages [51] and to the patients’ regular email [52]. Mobile phone technology is gaining ground as a simple interface for the health consumer, given the increasing ubiquity of this technology worldwide, and will therefore be especially useful for patients who seldom use their computer.

Personalized feedback appeared to be one of the most promising features for long-term usage. In fact, two types of personalized feedback via email messages can be distinguished: personalized feedback from a caregiver via secure email and personalized feedback via automated messages and prompts. From the results of this study and the findings of Mohr et al [53] and Fry and Neff [54], we can assume that the use of personalized feedback from a real person is more persuasive than automated tailored feedback. Future research should focus on establishing which type of personalized feedback works best for whom (patients with short-term care needs ie, prevention/cure, versus patients with long-term care needs, ie, chronic disease management) and in which situation (purpose of the communication: task focused versus affective).

Moreover, integrating the technology with existing clinical care could serve as a push factor. Stevens et al [55] found that higher levels of engagement can be reached when technology requires users to log in, for example once a month. Therefore, it is expected that the effects of technology use will be stronger on patients who log in every month (fixed regimen) than on patients who log in only once in a while. By integrating eHealth technology into existing traditional offline care (visits), patients will be triggered to log in within the framework of a fixed regimen.

Besides, education should be provided in a more interactive way, for example through Web 2.0 tools that are built around user-generated or user-manipulated content, such as wikis, blogs, podcasts, and social networking sites [45,56-59].

The limitations of this study include the small and select sample of participants. Users were self-selected, as they were motivated to use the Web application. The patients and nurses who chose to participate in the project may possibly differ from other patient groups. Further research should be conducted, preferably with larger sample groups and among nonenrollees, to gain more thorough insights into the technology preferences of the different patient groups. Nevertheless, we believe that our results provide insights beyond the current literature into patients’ engagement in Web-based disease management programs. The use of a mixed-methods design [24,25] has contributed positively to this. By using interviews and usability tests we were able to explain the actual usage, and the survey provided insights into who used the technology. All of the results combined provided an insight into the usage pattern and preferences of individual users for specific technology features. Log files enabled us to assess the actual and long-term usage of the technology features.

In this study, attrition was not measured with the usual measures, such as Kaplan-Meier [22,60,61]. Most attrition measures analyze survival. However, we could not use these measures in our study because they provide insights only into the drop in usage, and not in the pattern of usage. Such survival curves are perhaps more useful for eHealth interventions with a fixed pattern of use—for example, e-therapy interventions. In our study, the pattern of usage was not fixed. Therefore, we searched for activity patterns in measuring continuity of use and we measured the degree of activity to distinguish between the infrequent users and the highly active users.

Our findings confirm the need for further research into usage patterns and user profiles [22]. Strategies that engage users with technology are important for addressing the low take-up of eHealth technologies. This study has set out three key strategies for increasing the initial and long-term use of eHealth technologies: (1) avoiding selective enrollment, (2) making use of participatory design methods, and (3) developing push factors for persistence. Innovations in health care will diffuse more rapidly when technology is employed that both is simple to use and has applicable components for interactivity in order to foresee the patients’ need for continuous and personalized feedback, in particular for patients with a greater need for care. More longitudinal research on the use of eHealth technologies such as this study and recently published studies on attrition and adherence factors [52,53,62-70] is needed to provide insights into the way usage fluctuates over time. Through the present study we gained an insight into the differences between highly active users and nonusage dropouts, which can be seen as a first step toward decreasing attrition. The next step could be found when examining the opportunities technology has to offer. Future research should therefore focus on what kinds of system features can increase the use of eHealth technologies.