Diabetes self-management education (DSME) aims to help people develop the knowledge and skills to manage their physical and emotional health [1]. There is evidence that DSME can improve glucose control [2,3] and prevent complications [4].

The National Diabetes Audit data in the United Kingdom indicates that only 8.3% of people who were referred to structured education attended in 2016 [5]. Established programs involve face-to-face group courses. Qualitative studies suggest that some people find the face-to-face courses difficult to attend because of timings of courses, lack of transport, work and family commitments, or they do not like taking part in groups [6]. Computer-based self-management interventions bypass some of the barriers to face-to-face education [7,8]. A 2013 Cochrane systematic review of computer-based diabetes self-management interventions found a small effect on glycemic control, which was larger in the mobile phone app subgroup [7]. A 2015 systematic review of internet-delivered DSME found significant improvements in glycemic control and clinic attendance compared with usual care [8]. Some studies in the review also found improvements in self-efficacy [9], diabetes knowledge [10-13], exercise behaviors [14,15], and self-care behaviors [14,16]. A 2017 systematic review of the reviews of technology-enabled diabetes self-management interventions found that 18 of 25 reviews reported a significant reduction in glycated hemoglobin; however, a meta-analysis was not conducted because of heterogeneity in interventions and study designs [17]. Reviews of Web-based diabetes self-management interventions are promising, but there are also challenges including low uptake and engagement, that can limit effectiveness and need further research [18].

In light of this, we developed a Web-based structured education program for people with type 2 diabetes mellitus (T2DM), called Healthy Living for People with type 2 Diabetes (HeLP-Diabetes): Starting Out (HDSO). The content and aims of the HDSO program were based on an earlier intervention named HeLP-Diabetes. The key difference between HeLP-Diabetes and HDSO was that HDSO was aimed at people who were newly diagnosed, and so it followed a structured written curriculum with specific aims and objectives, as recommended by the National Institute for Health and Care Excellence [19] and required for general practitioners (GPs) to receive quality and outcomes Framework remuneration for referral [20]. HeLP-Diabetes was developed for people at any stage of their illness, and contained information on 560 webpages, which people could dip in and out of without following a linear pathway. A randomized controlled trial (RCT) showed HeLP-Diabetes to be effective and cost-effective [21,22].

The HDSO program is discussed in this paper. HDSO was a Web-based intervention developed for use on desktop computers and tablets. The program development took an iterative user-driven approach informed by the human-computer interaction (HCI) design lifecycle [23,24] and the Medical Research Council (MRC) guidance on developing and evaluating complex interventions. The development process consisted of 3 phases: (1) initial design; (2) usability testing with volunteers; and (3) in the wild testing in the National Health Service (NHS) with people newly diagnosed with type 2 diabetes. The 3 stages of the development process are described in detail elsewhere [25].

In line with the MRC framework on the development and evaluation of complex interventions, and current advice on development and evaluation of digital health interventions [26,27], following the development process, we undertook a formative evaluation of the program to explore its feasibility, acceptability, and apparent impact on users. As this was a digital health intervention, it was appropriate to draw on methods more familiar to computer science and HCI researchers than biomedical ones, including an emphasis on real-world data, with participants using the intervention as they would in routine practice, rather than as part of an overt research project. Such real-world studies, also known as in the wild studies, provide a contextual backdrop for determining the strengths and weaknesses of the intervention accurately [28]. This allows digital health interventions to be tested by representative users attempting representative tasks in representative environments and makes any recommendations for further research user-led rather than researcher-led [29]. Studies in the wild can reveal complex and unexplained phenomena that can only emerge in the natural setting of the intervention [30]. They benefit from the strong external validity achieved by delivering the intervention as it will be used in routine practice, and not as part of an overt research project. In the wild studies are a necessary precursor to RCTs to determine effectiveness and cost-effectiveness as they allow for further refinement and optimization of the intervention, including the surrounding delivery package provided by health care professionals (HCPs), and a preliminary estimation of any associated changes in outcomes associated with the use of the intervention [27]. Such studies are not intended to, and cannot, determine any causal relationship between observed changes and the intervention tested.

In this study, we took advantage of the naturalistic, real-world setting and data provided by 5 Clinical Commissioning Groups (CCGs) commissioning HDSO in the NHS in the United Kingdom, as part of an overall menu of DSME offered to people with T2DM.

The overall aim of the study was to determine the feasibility, acceptability, and potential impact of delivering a Web-based structured education program (HDSO) in routine primary health care. Specific objectives were as follows:

The study was conducted during implementation of the program in the NHS. People with T2DM in GP practices in 5 London CCGs registered for the program as an NHS service, rather than as research participants. This meant that we were unable to randomize participants to an intervention or control group, or collect clinical outcome data as we did not have access to their clinical records. However, this meant that the participants more accurately reflected the population of interest (people with type 2 diabetes treated in NHS primary health care) [31]. The total population of the 5 CCGs was 1,384,000. The population was diverse, with over 30% of the population from black and minority ethnic (BAME) groups [32]. All 5 CCGs were in the top quartile for deprivation in England [33]. RCTs are more limited in their external validity because of the characteristics of people who volunteer, and the inclusion and exclusion criteria in trial protocols [34,35].

Ethical approval was granted by the Health Research Authority (HRA; reference number 159488). The program was offered to people as part of service delivery, so the use of the data on registrations, activities, and questionnaire scores generated through the Web-based program was permissible under the HRA clause that the secondary use of information collected in the course of normal care is generally excluded from the Research Ethics Committee review [36] and specific informed consent to participate in research is not required. However, everyone who volunteered to be interviewed provided full informed consent before being interviewed and was aware that the interviews were for research purposes.

The target population was adults with T2DM. Referral was not limited to people who were newly diagnosed, as many people only become ready for structured education after having come to terms with the diagnosis [37-39]. Referral therefore included people at any stage of diabetes to enhance uptake. As this was a service, there were no formal exclusion criteria, but we advised HCPs that people who could not use a computer because of severe mental or physical impairment, had insufficient mastery of English, or were currently participating in a trial of an alternative self-management program would not be suitable for referral. A sample size calculation, which would be used to determine a statistically significant treatment effect in RCTs, was not used as this was not appropriate for the study design.

Everyone who registered for the program was invited to take part in interviews. We also invited HCPs working at practices who referred a high or low number of people with T2DM, and staff employed by the CCGs to support implementation of the program.

The program was offered to people with T2DM using referral packs sent by practices who identified eligible people from electronic medical records searches, text messages, flyers, or in consultations with doctors or nurses. In 1 CCG group, there was a dedicated Change Manager who visited practices to talk to staff and people with T2DM about the program. Data on the number of people who were offered the program and declined were not recorded by practices.

People with T2DM were asked to telephone or email the HDSO team, to be registered, and have baseline demographic and clinical data collected (see Outcome Measures). A username and password was emailed to users, along with contact information in case of problems. However, this process of telephone registration proved to be time consuming and caused delay in people being able to access the program. The registration process was therefore modified (see Multimedia Appendix 1) so that users had the option of Web-based self-registration. People with T2DM were given the registration webpage details at referral by HCPs and in referral packs, which they could access to enter their demographic details, and register a username and password to use to log into the program.

The intervention was a Web-based structured self-management program for people newly diagnosed with T2DM. It is described in detail in Multimedia Appendix 1, using the Template for Intervention Description and Replication (TIDier) checklist [40], and summarized here.

The content was based on HeLP-Diabetes [41], an online self-management intervention for everyone with T2DM. The theoretical basis was the Corbin and Strauss theory that self-management involves 3 tasks: managing the disease process, managing the emotional consequences, and managing the changes that occur in daily life [42]. HDSO was a 4-session program, comprising 4 or 5 modules per session, and questionnaires measuring diabetes-related distress and diabetes self-management self-efficacy score (DSMSES) in weeks 1 and 4. Each module took about 15 to 20 min to complete (see Table 1). Information was presented using text, images, and videos (see Multimedia Appendix 2). Email reminders were sent to users if they had not logged on for 7 days or more.

Users were provided with their scores from the diabetes-related distress and DSMSES questionnaires, and individualized feedback developed by GPs and Diabetes Specialist Nurses (DSNs). The feedback in week 1 helped users identify gaps in knowledge or skills, and signposted to sessions of the program that could help them improve. The feedback in week 4 focused on improvements made and directed users to the HeLP-Diabetes website for ongoing support. Users were asked to set specific, measurable, achievable, realistic, time-bounded goals [43], and they were given opportunities to review their goals.

The data collected reflect the objectives of the study (see Table 2).

The server side of the website automatically collected anonymized data on user ID, date and time of login, and pages visited. These were used to determine how many people registered for and completed the program.

Data on personal characteristics included the following: (1) age; (2) gender; (3) ethnicity; (4) highest educational attainment; (5) internet access (home or public); (6) information technology skill level (basic, intermediate, or advanced; (7) duration of diabetes; (8) previously offered face-to-face education (yes or no); (9) previously attended face-to-face education (yes or no); and (10) diabetes management (lifestyle alone, or tablets and/or insulin). Education level was categorized using UK and US qualifications, and the International Standard Classification of Education [44].

Data on change in diabetes-related distress and self-efficacy in self-management were collected using the Problem Areas in Diabetes (PAID) [45] and DSMSES [46] questionnaires.

A total of 791 people registered for the program, 188 started it (completed at least the first module of the first session), and 74 completed all 4 sessions (see Multimedia Appendix 3).

Data on the characteristics of people with T2DM is given in Table 3. A total of 791 people with T2DM registered to use the program. Demographic data were self-reported at registration (by either telephone or Web-based questionnaire), and there is a large amount of missing data because of nondisclosure, particularly on previous offer and attendance at structured education and diabetes management. The average age of people with T2DM registering to use the program was 57.6 years, over half (316/586, 53.9%) were male, over half (310/605, 51.2%) were from BAME backgrounds, and nearly one-third (181/602, 30.1%) had no qualifications beyond high school leaving age (Table 3). Just over one-quarter (170/589, 28.9%) had had their diabetes <1 year, and while half (193/394, 49.0%) recalled being offered face-to-face education, only 9.4% (37/394) had attended it.

The only factors associated with completion were duration of diabetes (P=.04), and having been offered (P=.001) and having previously attended (P=.002) face-to-face education. Having advanced information technology skills was not associated with completing the program (see Table 3).

Median DSMSES (self-efficacy) scores were significantly higher (better) in week 4 compared with week 1. Median PAID (distress) scores were significantly lower (better) in week 4 than week 1 (see Table 4).

Interviews were conducted with 17 participants (10 people with T2DM and 7 HCPs). Of the 10 people with T2DM, 7 had completed the program and 3 had registered for the program, but not completed it. Other characteristics are listed in Multimedia Appendix 4. Of the HCPs, 3 were DSNs, 2 were GPs, 1 was a HeLP-Diabetes Change Manager (employed by the local CCG to liaise with the GP Practices and promote the HDSO program) and 1 was a CCG Project Officer (providing support to senior CCG project managers). The data from people with T2DM and HCPs are combined in the results, as many of the subthemes are shared. Where a subtheme is unique to a particular group, this is stated in its description and illustrated with a quote.

Four major themes emerged from the analysis of the interview data, each with a number of subthemes (see Table 5). We mapped 2 of the major themes to NPT constructs. NPT explains whether and how complex interventions become embedded in health care practice. Hence, the themes that mapped to the constructs were those that related to health care system factors (the value of discussion between HCPs and people with T2DM about DSME at the time of referral; and improving uptake of the HDSO program) rather than factors that related to people with T2DM or the program. The NPT constructs and the HDSO-specific meanings we defined for each construct are listed in Table 6. The themes and subthemes we mapped to the constructs are listed in Table 7.

This study found that it is feasible to deliver Web-based structured education in the NHS; a wide demographic can use it; and it may improve self-efficacy and diabetes-related distress. The quantitative data showed that there were problems with uptake and completion, with completion positively associated with duration of diabetes <1 year, and self-report of having been offered and/or attended structured education previously. The qualitative data helped us to explore the low uptake more fully. Findings from interviews with people with T2DM and health professionals suggested that professional factors, personal factors for people with T2DM, and program factors affected program acceptability and attrition. More research is needed on improving uptake and determining the relative effectiveness and cost-effectiveness of Web-based and group-based structured education.

The strengths of this study include the strong external validity, with our data drawn from real-world experience of implementation and delivery of the service within routine primary health care. Other strengths include the use of a mixed method, with quantitative data on uptake, usage, and outcomes, and qualitative data to explore the underlying reasons for these observed data, and the use of theory (NPT) in the data analysis. Using NPT allowed us to make a theoretically informed interpretation of the qualitative data in regard to implementation. The design was appropriate for the study objectives, namely, to determine acceptability, feasibility, and apparent impact of the program. Acceptability and feasibility related not only to patients but also to health professional factors, which NPT allowed to explore more deeply.

It is important to state clearly that this design cannot be used to determine effectiveness of the intervention, nor can any causal links be inferred. Determining effectiveness and ascribing a causal impact of an intervention requires an RCT design, with an appropriate comparator, and a sample size calculation. In light of the multiple tests undertaken, it is possible that the observed associations between likelihood of completion and duration of diabetes, and self-reported offer or attendance at structured education are because of chance. A specific weakness of our study was the lack of clinical outcome data and our reliance on proximal outcomes collected through self-reported outcome measures. This was a direct result of our emphasis on external validity and real-world data, so that people with T2DM used the program as part of their routine NHS care, and not as part of a research study. Hence, we could not obtain formal informed consent, except from those who participated in interviews, and as such, it would not have been ethical to have access to clinical data. A further limitation was our inability to invite people who were offered the program but did not register to use it to take part in interviews because of the ethical limitations. Interviewing these people would have helped us to understand why some people do not want to use Web-based structured education.

The percentage of completers was low (9.4%) but compares favorably with attendance at face-to-face education (8.3%) [5] and adherence to other digital self-care interventions [8,59-63]. The interviews helped explore attrition from the program, and NPT improved interpretation and transferability of the themes. Reasons for low engagement were similar to findings from a 2016 systematic review of diabetes education programs [6] and suggest that personal factors for people with T2DM and HCP factors, including nonprioritization, lack of enthusiasm from HCPs for education, and people feeling that they would benefit or that knew enough already, contributed to poor uptake and completion [6].

The study included people with any duration of diabetes, but one of the factors associated with completion was duration of diabetes of less than a year (P=.04). The rationale for including people with any duration of diabetes was that the literature suggests that many people only become ready for structured education after having come to terms with the diagnosis [37-39], and we did not want to limit uptake by only including people who were newly diagnosed. Our previous research had shown that uptake was lower when people with diabetes of duration greater than a year were excluded [25]. Overall, our findings therefore suggest that while offering structured education to people with diabetes of any duration does help improve uptake, overall uptake is still low, and it is the newly diagnosed group who are more likely to complete the course. This is consistent with a 2014 study by Roelofson et al, which found low overall participation in a Web-based patient platform for T2DM containing health data and educational information (110 people used the intervention out of 974 who were registered, 11.3%), but interest was higher in people with shorter duration of illness [64]. This suggests that DSME should be offered to everyone with T2DM, but people who are newly diagnosed are a group who should continue to be targeted and offered referral in the first year following diagnosis.

Having been offered face-to-face education (P=.001) and having attended face-to-face education (P=.002) also seemed to be associated with completion. The association between likelihood of completion and having been previously offered, or attended, structured education has not been reported previously in the literature. This association could reflect intrinsic characteristics, whereby people with more interest in and commitment to structured education are more likely to remember the offer and attend whatever education they are offered, or could reflect an effect of structured education, in that it makes people aware of how much there is to learn, and hence promotes engagement with subsequent offers. Further research is needed on whether people are more likely to take up structured education if they are offered it more than once, or whether more incentives are needed to increase intrinsic motivation and interest and commitment to complete a course.

If Web-based structured education can be found to be effective and cost-effective and have acceptable reach, this could give people with T2DM more options for learning about self-management and help improve structured education uptake.