Web-based self-management support for people with type 2 diabetes (HeLP-Diabetes): randomised controlled trial in English primary care

Objective To determine the effectiveness of a web-based self-management programme for people with type 2 diabetes in improving glycaemic control and reducing diabetes-related distress. Methods and design Individually randomised two-arm controlled trial. Setting 21 general practices in England. Participants Adults aged 18 or over with a diagnosis of type 2 diabetes registered with participating general practices. Intervention and comparator Usual care plus either Healthy Living for People with Diabetes (HeLP-Diabetes), an interactive, theoretically informed, web-based self-management programme or a simple, text-based website containing basic information only. Outcomes and data collection Joint primary outcomes were glycated haemoglobin (HbA1c) and diabetes-related distress, measured by the Problem Areas in Diabetes (PAID) scale, collected at 3 and 12 months after randomisation, with 12 months the primary outcome point. Research nurses, blind to allocation collected clinical data; participants completed self-report questionnaires online. Analysis The analysis compared groups as randomised (intention to treat) using a linear mixed effects model, adjusted for baseline data with multiple imputation of missing values. Results Of the 374 participants randomised between September 2013 and December 2014, 185 were allocated to the intervention and 189 to the control. Final (12 month) follow-up data for HbA1c were available for 318 (85%) and for PAID 337 (90%) of participants. Of these, 291 (78%) and 321 (86%) responses were recorded within the predefined window of 10–14 months. Participants in the intervention group had lower HbA1c than those in the control (mean difference −0.24%; 95% CI −0.44 to −0.049; p=0.014). There was no significant overall difference between groups in the mean PAID score (p=0.21), but prespecified subgroup analysis of participants who had been more recently diagnosed with diabetes showed a beneficial impact of the intervention in this group (p = 0.004). There were no reported harms. Conclusions Access to HeLP-Diabetes improved glycaemic control over 12 months. Trial registration number ISRCTN02123133.


Strengths and Limitations of this study
• Self-management improves outcomes and health status in people with type 2 diabetes; but uptake is poor, partly because some people find the current model of group-based structured education inconvenient or unappealing. • Web-based self-management support may help improve uptake.
• The trial recruited to target and achieved reasonable follow-up; however trial participants may not be representative of the total population of people with type 2 diabetes in primary care. There is a global epidemic of type 2 diabetes mellitus (T2DM). An estimated 422 million adults, or 10% of the global population, were living with diabetes in 2014 of whom around 90% had type 2 diabetes. 1 Poorly controlled diabetes is associated with premature mortality, and a high risk of complications, including cardiovascular disease, nephropathy and retinopathy. The risk of complications can be reduced by good control of glycaemia and cardiovascular risk factors. 2 3 Interventions which improve self-management skills for patients with diabetes can improve health outcomes and reduce health care costs 4 and international guidelines support training patients in self-management. 3 5 However, it is not clear how best to support patients in developing such skills, and uptake of diabetes selfmanagement education remains low. In England, despite over 90% of eligible patients being referred, 6 only 5.3% attended self-management training in 2014 - 15. 7 Poor uptake may be related to the dominant model of structured education, which is groupbased sessions, lasting a half or whole day, or spread over regular sessions over several weeks. 8 Many patients, such as those who work, those with caring commitments, or those who are uncomfortable in groups, may find it difficult to attend. 9 10 Web-based support for self-management could address some of these barriers, particularly in high income countries, where levels of web-access are high. In the UK over 80% of households had Internet access in 2015, and internet access amongst older people continues to grow steadily. 11 12 Potential advantages include convenience, anonymity, regular updates, and the potential to use video and graphics to present complex information in a format accessible to those with low literacy. 13 Although systematic reviews have confirmed that computer-based interventions can improve health outcomes in diabetes 14 , not all such interventions have a beneficial impact, with meta-analyses showing substantial heterogeneity related to widely differing interventions, including in the use of theory to develop the intervention, 15 outcomes, 14 16 and the duration of follow-up, with most trials having relatively short follow-up (less than 12 months). 14 This is the first UK-based trial of a comprehensive, web-based self-management support programme for people with type 2 diabetes.
This trial assessed the effects of a web-based self-management programme, called Healthy Living for People with Diabetes (HeLP-Diabetes), on glycated haemoglobin (HbA1c) and diabetes-related distress over 12 months.

Methods.
Trial Design and participants: multi-centre, two-arm individually randomised controlled trial in 20 General Practices in England with a mix of urban, suburban and rural practices. Practices were required to have two nurses -one to facilitate access to the intervention, and one to collect data. Participants were adults, aged 18 or over, with T2DM, registered with participating general practices. Patients were excluded if they were unable to provide informed consent; unable to use a computer due to severe mental or physical impairment; had insufficient spoken or written English to use the intervention (operationalised as unable to consult without an interpreter); were terminally ill with less than 12 months life expectancy; or were currently participating in a trial of an alternative self-management programme. Participants were not required to have home Internet access or prior experience of using the Internet to participate. Participants with previous or current experience of self-management education were eligible to participate. Recruitment took place between September 2013 and December 2014. The trial protocol was submitted for publication in June 2014. 17 There were no changes to the methods after the protocol was agreed and the start of the trial. Ethical From an NHS perspective, the important research question was whether the proposed intervention could improve health outcomes when compared to current practice. However, to improve acceptability to participants and to maintain blinding, all participants had access to a website. Participants in the control arm were given access to a simple information website, based on the information available on the website of the main UK diabetes charity (Diabetes UK) or National Health Service patient information website (NHS Choices). They received the same initial facilitation meeting as participants in the intervention group, in which they were shown how to log on, set a user name and password, and how to use the website.

Primary outcomes
The outcomes reflected the dual goals of improving health outcomes and reducing diabetesrelated distress. The two joint primary outcomes were glycated haemoglobin (HbA1c) and diabetes-related distress, measured by the Problem Areas in Diabetes (PAID) scale, both at 12 months post-randomisation. PAID has 20 items focusing on areas that cause difficulty for people living with diabetes, including social situations, food, friends and family, diabetes treatment, relationships with health care professionals and social support. 20 PAID scores range from 0 -100, with higher scores indicating more distress. A score of 40 or more indicates significant distress, and around 40% of patients with diabetes experience significant distress. 21 outside the 10 -14 month window were not used directly in the primary analysis, but were entered into the imputation model.

Baseline characteristics
Baseline demographic and clinical characteristics are shown in Table 1. The mean age was nearly 65 years, over two-thirds (n = 258, 69%) were male, and most were White British (n = 300, 80%). Nearly all (n=370, 99%) had a computer with access to the internet at home and just over half (n = 210, 56%) rated themselves as experienced computer users. Around onethird (n = 134; 36%) had been diagnosed for less than 5 years, with a further third (n = 115, 31%) having been diagnosed between 5 and 9 years ago. Overall, this was a population with well-controlled diabetes at baseline (mean HbA1c was 7.3% (56 mmol / mol)) and low levels of distress (mean PAID = 19).

Primary outcomes
At twelve months the primary analysis showed a significant difference in change in HbA1c between the randomised groups with participants in the HeLP-Diabetes group having a lower HbA1c than those in the control group (mean difference = -0.24%; 95% confidence intervals -0.44 to -0.05, p=0.014) (Table 2, Figure 2). There was no difference in change in PAID scores between the groups at 12-months (mean difference -1.5; 95% CI -3.9, 0.9, p=0.209), though both groups showed a decrease in PAID over the follow-up of the trial (Table 2, Figure 3).

Secondary outcomes
There was no difference in secondary outcomes at 12 months, with the exception of systolic blood pressure, which decreased more in the intervention group than in the control group (p = 0.010) ( Table 2). There were no significant differences between groups on any of the outcome measures amongst individuals who completed three month outcomes (Supplementary Table 2). No adverse effects or events were recorded during follow-up.

Usage data
The mean number of log-ins was significantly higher in the intervention group than the control group (18.7 vs. 4.8, p= 0.0001), as was the mean number of pages visited per log-in (10.5 vs. 7.7, p <0.0001) and the mean number of days in which the website was accessed (10.1 vs. 3.3, p<0.0001) ( Table 3). The causal analyses estimated that for a "high-usage" population (those with usage ≥ the median of 4 days) the HeLP-Diabetes intervention could on average reduce HbA1c by -0.44% (95% CI -0.81, -0.06) and PAID by -2.8 (95% CI -7.2, 1.7) over 12 months ( Supplementary Figures 2 and 3). The mean usage in the "high-usage" group was 18 days. It should benoted that the usage data presented do not include the initial facilaition visit. There was a technical error in the software which led to usage data not being collected before 1 January 2014. At this point 16 participants had been randomised (7 to intervention, 9 to control). For these 16 participants, the usage data is not based on a full year, but for all other participants, data are summarised for the 12 months postrandomisation.

Sensitivity analyses
The findings from the sensitivity analyses, including a complete-case analysis, were similar to the main analysis (Supplementary Table 3). Participants who were missing 12-month HbA1c had significantly higher mean baseline HbA1c measures (7.9% vs. 7.1%, p<0.001) leading to higher imputed HbA1c at 12-months in the non-completers and a greater mean difference between the randomised groups than from complete case analyses (Supplementary Figure 4, Supplementary Table 3).

Subgroup analyses
Pre-specified subgroup analyses showed that there was no evidence of baseline measures of HbA1c or PAID being effect modifiers for the mean difference between the groups. There was strong statistical evidence (interaction p=0.004) to suggest that the duration of diabetes acted as an effect modifier, with those who had been diagnosed more recently (<7 years) showing more of a reduction in PAID than those who had been diagnosed for longer periods of time. Duration of diabetes had no effect on change in HbA1c (Supplementary Table 4).

Harms
There were no reported harms in either group.

Discussion
In this first UK-based trial of a web-based self-management programme for people with T2DM, participants randomised to HeLP-Diabetes demonstrated improved glycaemic control at 12 months compared to those randomised to a simple information website. This improvement is both clinically and statistically meaningful, appears robust across all prespecified sensitivity analyses, and was not dependent on duration of diabetes, baseline glycaemic levels or level of diabetes-related distress. Each 1% reduction in HbA1c is associated with a risk reduction of 21% for deaths related to diabetes and a 37% risk reduction for microvascular complications. 26 Given that this web-based intervention could be delivered at low-cost and at scale across the UK, the potential for population benefit is considerable. Moreover, in contrast to group-based education, where the effects appear to wane with time, 29 the effects of HeLP-Diabetes were greater at 12 months than at 3 months.
There was no overall impact on diabetes-related distress, but some evidence that HeLPdiabetes appeared to reduce distress in recently diagnosed individuals. However, it is worth noting that baseline PAID scores were exceptionally low in this trial population. In a small pilot study, participants offered supported access to HeLP-Diabetes reduced their PAID scores by 6 points (p=0.04) over 6 weeks. 30 The trial has many strengths. It was a pragmatic trial, open to nearly all patients with T2DM in participating practices. Concealment of allocation was complete, as randomisation occurred after baseline data collection. Baseline prognostic factors were well balanced between groups. Every effort was made to achieve blinding, including requiring practices to have two nurses, so that data collection was undertaken by a nurse blind to participant allocation. Data for the co-primary outcomes at the primary outcome point were available for 78% and 86% of participants for HbA1c and PAID respectively. All analyses were on an intention-to-treat basis, supplemented by a CACE analysis. Although response rates for the co-primary outcomes were good, some potential for bias existed. Our primary analysis used multiple imputation methods because evidence shows that the assumptions underpinning this method are more defensible than those assumed using other approaches to missing data. 31 We also undertook sensitivity analyses including complete cases, non-contaminated cases, and a linear model excluding centre; all yielded similar results.
The two co-primary outcomes reflected the twin aims of the intervention: to improve diabetes control and to reduce diabetes-related distress. Around 40% of patients with diabetes have significant levels of distress, which severely impacts on quality of life, 32 and diabetes-related distress is an important outcome for patients. 33 Our patient and public involvement (PPI) panel were clear that this should be a primary outcome, and a recent meta-ethnography emphasised the importance of empowerment and quality of life in promoting long term engagement with self-management. 34 In contrast, many health care professionals are more interested in glycaemic control. In line with previous trials in this area, 35 we decided to adopt both as co-primary outcomes and to test both at a 5% level of significance. 36 There are some limitations. Despite maximising the inclusivity of the trial by minimising the exclusion criteria, participants were not representative of the overall population of patients with type 2 diabetes in England. Compared to the overall population, participants had better control of their diabetes and cardiovascular risk factors, 6 7 and were much less distressed. 21  10 This finding mirrors that of a recent systematic review of demographic factors associated with web portal usage amongst people with diabetes which found that those with well controlled diabetes were more likely to use such portals than those with poor control. 37 However, fewer of our participants self-rated their computer skills as excellent (57% of our sample compared to a national average of 73%). 12 This good control at baseline has two implications -first, that there was little room for improvement in this population, and secondly, that this population may have been unusually motivated to self-manage their diabetes. Although every effort was made to maintain blinding, it is possible that some participants may have discussed their use of the intervention with research nurses, making it possible to infer which arm they had been allocated to. This could have affected secondary clinical outcomes, such as blood pressure or weight, but could not have affected assessment of glycated haemoglobin. There appeared to be high potential for contamination between two participants who shared the same surname and address, and a further two participants did not receive their allocated intervention due to an error at practice level; excluding these four made no difference to the results. A further limitation of the trial is that it provides little insight into the mechanism of action of HeLP-Diabetes. This was the result of a deliberate decision to focus on clinically important outcomes and minimise both the response burden and the potential impact of measurement on participants.
This is the first UK-based trial of a web-based self-management programme for people with type 2 diabetes, and internationally, the first trial of such a comprehensive intervention that aims to address the three main tasks of self-management: emotional, medical and role management. 19 In the Cochrane review of computer-based self-management interventions for people with T2DM, only 4 of the included studies had follow-up of 12 months or more. 14 Of these, three interventions were clinic-based, with participants completing self-assessment tools on a touch screen and receiving tailored advice during their baseline visit to their diabetes clinician [38][39][40] and one was a mobile phone-based intervention which provided tailored messages in response to participant's results of blood glucose self-monitoring data. 41 A more recent systematic review of internet delivered diabetes self-management identified 2 trials with 12 or more months follow-up. 42 One trial was on a structured intervention based on a peer-led, group-based, diabetes self-management course 43 . There were six sessions, with each session available for one week. Each session required participants to make a specific action plan to address a problem they were experiencing. Peer facilitators encouraged use of the programme. Follow up was planned at 6 and 12 months; however HbA1c data were only available at 6 months. The other trial compared two versions of a web-based intervention (with and without additional social support) to enhanced usual care. The web-based intervention was designed using social-cognitive theory and a social-ecological model, with a focus on three main behaviours: dietary intake, physical activity and medication adherence. Users of either web-based intervention received motivational phone calls to encourage adherence and development of action plans. Those randomised to the enhanced intervention (with additional social support) received two additional phone calls and an invitation to attend a group session. There was no statistically significant difference between groups in HbA1c or other biological outcomes at 12 months. 44 Thus the results of this trial add significantly to the available literature.
On the basis of these results, HeLP-Diabetes should be considered as an addition to the current menu of self-management support for people with type 2 diabetes, and may help increase overall access and uptake. Most commissioned services currently focus on newly diagnosed patients, leaving clear unmet need for people who have had their diabetes for longer, but are looking for ways to improve their health. Many patients are not ready to engage in self-management early in their illness journey, 9 but become motivated to do so later, often as a result of a change in medication or development of a complication. 45 The intervention is low cost, and as most costs are fixed, irrespective of number of users, is likely to be cost-effective, particularly if widely used. A cost-effectiveness analysis of HeLP-Diabetes will be reported separately.  12 delivery of the intervention and / or the management and oversight of the trial, staff at the participating practices, PCRN staff, and all our participants.

Data Sharing.
Patient level data, the full dataset, statistical code are available from the corresponding author. Consent for data sharing was not obtained from participants, but the potential benefits of sharing these data outweigh the potential harms as the data are anonymised.

Rationale 4
The aim of HeLP-Diabetes was to enable people with type 2 diabetes to lead healthier lives 5 and improve overall quality of life. In order to do this, we adopted a strong theoretical 6 framework, which included: 7 narrative, acceptance of the role of "patient", and negotiating inevitable changes in 28 relationships with family, friends and colleagues. This characterisation of the work of self-29 management provided the overall framework for the content, tone and style of the 30 intervention. As these tasks are intertwined and inter-dependent, we designed the content 31 to address all three tasks in an integrated fashion. 32 Normalisation Process Theory is a mid-range sociological theory which predicts and explains 33 whether and why certain innovations will be effectively implemented into routine health care 34 and become "normalised", that is, become so much part of routine practice that they 35 disappear from view. 3 It posits that interventions which are easily understood and 36 distinguished from other interventions (coherence), which relevant professionals can see the 37 advantage of compared to current or alternative practice (cognitive participation), which fit 38 well with existing professional relationships and workflows (collective action), and where 39 users are provided with formal or informal feedback and evidence about the impact of the 40 intervention (reflexive monitoring) are more likely to be implemented than those that do not. 41 These constructs guided the development of the intervention by, for example, ensuring 42 HeLP-Diabetes integrated well with existing workflows in general practice. 43 Effective self-management in diabetes often requires patients to change deep-rooted 44 behaviours, particularly around diet and physical activity. Rather than adopt a single 45 behavioural theory, we opted to design our behaviour change modules around specific 46  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59   HeLP-Diabetes was designed to be used as part of an overall package of care for people 93 with diabetes. Low usage, or non-adherence to internet interventions is well-recognised 94 problem, and our preparatory work with patients and health care professionals indicated that 95 integrating the intervention into routine care was likely to improve uptake and adherence. 96 Hence we made the programme available to registered users only, and encouraged health 97 care professionals to register patients. Once registered, patients could use the programme 98 as much (or as little) as they wanted. There was no prescribed level of use, as our proposed 99 users included patients at all stages of their illness journey, from those newly diagnosed to 100 those who had lived with diabetes for many years. As such, we anticipated that each user 101 would have different needs and priorities, and the programme was designed to allow users 102 to pick and choose sections that were most relevant and beneficial for them personally.

103
There was a limited amount of tailoring. Additional resources and sources of help were 104 tailored by the patient's CCG, but otherwise tailoring was limited to the behaviour change 105 and health record sections where users entered their own goals or data. 106 107 Registration and facilitation. 108 During the trial, registration was performed by a practice nurse. During the registration 109 procedures users were asked to select a password and username. After registration, the 110 nurse was asked to demonstrate the intervention to the patient, and integrate use of the 111 intervention into the patient's personal diabetes care plan by, for example, discussing what 112 goals the patient would like to set for the coming period, and showing the patient how to use 113 the intervention to set goals, monitor progress, and, if desired, arrange for automated 114 reminders by SMS or email to be sent by the programme. 115 Registration was undertaken in the patient's practice. All subsequent use of the intervention 116 was at any location convenient for the patient with internet access. For most people, we 117 expected this to be at home, or at the home of a relative. However, all users were given 118 information about local services (usually libraries) offering free access to an internet-119 connected computer. 120 121

Encouraging engagement 122
In addition to the registration and facilitation procedures described above, engagement to 123 the programme was encouraged through regular prompts, provided by email or SMS. 124 Emailed prompts were of two types: a) short (2 -3 lines) emails with one message and a 125 link to the relevant section of the programme. An example of this type of prompt was a 126 seasonal reminder of the importance of flu vaccination for people with diabetes, with a link to 127 the page of the programme explaining how and why flu vaccination is beneficial, and a 128 reminder that flu vaccination is available free from their general practice.

Item
Line numbers Provide the name or a phrase that describes the intervention 2,3 Describe any rationale, theory or goal of the elements essential to the intervention

-48
Materials: describe any physical or informational materials used in the intervention, including those provided to participants or used in intervention delivery or in training of the intervention providers 51-90 Procedures: Describe each of the procedures, activities and /or processes used in the intervention, including any enabling or support activities

-137
For each category of intervention provider, describe their expertise, background, and any specific training given.

109
Describe the modes of delivery (such as face-to-face or by some other mechanism, such as internet or telephone) of the intervention, and whether it was provided individually or in a group

-106
If the intervention was planned to be personalised, titrated or adapted, then describe what, why, when and how.

-106
If the intervention was modified during the course of the study, describe the changes (what, why, when and how)

-144
Planned: if intervention adherence or fidelity was assessed, describe how and by whom, and if any strategies were used to maintain or improve fidelity, describe them.

-149
Actual: if intervention adherence or fidelity was assessed, describe the extent to which the intervention was delivered as planned.

Statistical Methods for Multiple Imputation
Multiple Imputation using chained equation was used as the primary method to account for missing data (in both baseline and follow-up data). 1 A set of imputation models were specified, one for each variable with missing data. Each variable was then regressed on all other variables, including completely recorded baseline and follow-up variables and stratified by randomised group. Imputations were performed using predictive mean matching using the five nearest neighbours to the prediction as a set to draw from. The full list of variables considered in the MICE approach is shown in Supplementary Table 1, together with the number of missing values for each variable and time period.
Since only measurements within a 10-14 month window period were used within the main analyses of HbA1c and PAID, the following imputation procedure was implemented for these two co-primary outcomes. Twelve month measurements were subdivided into those that were measured within 10-14 months (the primary outcome variable) and those that were measured outside 10-14 months (a variable used for imputing only). For HbA1c, two additional variables were created for use within the imputation model; 1) the time in days from randomisation that the "12-month" HbA1c measurement was actually taken for values inside of the window period (and set to 365 for measurements taken outside the window period), 2) the time in days from randomisation that the "12-month" HbA1c measurement was actually taken for values outside the window period (and set to missing for measurements taken within the window period). The first of these variables gives the desired time for imputing HbA1c measurements when they are missing, whilst the second gives information on how far outside of the window the actual measurements were taken. Corresponding variables were created for the "12-month" PAID measurement. Finally, two additional variables were created defining the time in days at which HbA1c and PAID were measured at "3-months". All variables were included within the chained equations and imputed where necessary. 40 imputed datasets were created, the analysis models were fitted to each imputed dataset separately, and the estimates were pooled using Rubin's rules.
Statistical Methods for causal analyses The causal analysis proposed attempts to address how the effectiveness of the intervention is mediated through the frequency of website usage. In particular, it is important to understand whether prolonged usage of the website modifies the efficacy of the intervention. Since website usage is measured post-randomisation a naïve analysis of correlating usage with outcomes in the intervention group may give biased and misleading results, since there may be unmeasured confounders also correlated with the outcomes that distinguish the motivated users who regularly log-in from the less motivated ones. Causal analyses using instrumental variables (IV) were therefore used to determine the effect of website usage on outcomes. This approach preserves randomisation (i.e. provides a comparison independent of observed and unobserved confounders).
"Usage" is defined as the proportion of follow-up (rescaled as no. days in a year) that the HeLP-Diabetes website is accessed. It was assumed that the efficacy of the intervention is zero for individuals who never log-in (the exclusion restriction assumption). Website usage in the control group was ignored in the model as it was assumed that the control website was unlikely to be effective. (NB. usage statistics were collected for the comparator website but were not used in these analyses).  This main underlying assumption of the causal analysis is that the effect of randomisation to the HeLP-Diabetes intervention on 12-month outcomes occurs only through use of the website (Supplementary Figure 1; Z is randomised intervention, WU is website-usage and Y is 12-month outcome, e.g. HbA1c or PAID). This relies of the "exclusion restriction" assumption that the HeLP Diabetes intervention has no effect when usage is zero (i.e. for individuals who never log-in). Hence randomisation is assumed to be an instrumental variable. Figures 4 and 5 show the estimated causal effects of HeLP Diabetes on HbA1c and PAID at 12-months, respectively, by level of website usage (the "exclusion restriction" assumption leads to zero efficacy at zero usage). To determine the casual estimate of efficacy for a "high-usage" group, the individual predicted efficacy was calculated for each individual greater than or equal to the median usage of 4 days (shown as the red boxes in the histograms in Figures 4 and 5) and the average efficacy for this group was obtained. The mean usage in the "high-usage" group was 18 days.

Reported on page No
Title and abstract 1a Identification as a randomised trial in the title 1b Structured summary of trial design, methods, results, and conclusions (for specific guidance see CONSORT for abstracts)

Introduction
Background and objectives 2a Scientific background and explanation of rationale 2b Specific objectives or hypotheses

Trial design 3a
Description of trial design (such as parallel, factorial) including allocation ratio 3b Important changes to methods after trial commencement (such as eligibility criteria), with reasons Participants 4a Eligibility criteria for participants 4b Settings and locations where the data were collected Interventions 5 The interventions for each group with sufficient details to allow replication, including how and when they were actually administered Outcomes 6a Completely defined pre-specified primary and secondary outcome measures, including how and when they were assessed 6b Any changes to trial outcomes after the trial commenced, with reasons Sample size 7a How sample size was determined 7b When applicable, explanation of any interim analyses and stopping guidelines Randomisation: Sequence generation 8a Method used to generate the random allocation sequence 8b Type of randomisation; details of any restriction (such as blocking and block size) Allocation concealment mechanism 9 Mechanism used to implement the random allocation sequence (such as sequentially numbered containers), describing any steps taken to conceal the sequence until interventions were assigned Implementation 10 Who generated the random allocation sequence, who enrolled participants, and who assigned participants to interventions Blinding 11a If done, who was blinded after assignment to interventions (for example, participants, care providers, those  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48 assessing outcomes) and how 11b If relevant, description of the similarity of interventions Statistical methods 12a Statistical methods used to compare groups for primary and secondary outcomes 12b Methods for additional analyses, such as subgroup analyses and adjusted analyses

Results
Participant flow (a diagram is strongly recommended) 13a For each group, the numbers of participants who were randomly assigned, received intended treatment, and were analysed for the primary outcome 13b For each group, losses and exclusions after randomisation, together with reasons Recruitment 14a  Sources of funding and other support (such as supply of drugs), role of funders *We strongly recommend reading this statement in conjunction with the CONSORT 2010 Explanation and Elaboration for important clarifications on all the items. If relevant, we also recommend reading CONSORT extensions for cluster randomised trials, non-inferiority and equivalence trials, non-pharmacological treatments, herbal interventions, and pragmatic trials.
Additional extensions are forthcoming: for those and for up to date references relevant to this checklist, see www.consort-statement.org.    Objective: To determine the effectiveness of a web-based self-management programme for people with type 2 diabetes in improving glycaemic control and reducing diabetes-related distress.

Methods
Design: Individually randomised two-arm controlled trial

Setting: 21 General Practices in England
Participants: Adults aged 18 or over with a diagnosis of type 2 diabetes registered with participating general practices Intervention and comparator: Usual care plus either HeLP-Diabetes, an interactive, theoretically informed, web-based self-management programme or a simple, text-based website containing basic information only.
Outcomes and data collection: Joint primary outcomes were glycated haemoglobin (HbA1c) and diabetes-related distress, measured by the Problem Areas in Diabetes (PAID) scale, collected at 3 and 12 months after randomisation, with 12 months the primary outcome point. Research nurses, blind to allocation collected clinical data; participants completed self-report questionnaires online.

Analysis:
The analysis compared groups as randomised (intention to treat) using a linear mixed effects model, adjusted for baseline data with multiple imputation of missing values.

Results
Of the 374 participants randomised between September 2013 and December 2014, 185 were allocated to the intervention and 189 to the control. Final (12 month) follow up data for HbA1c were available for 318 (85%) and for PAID 337 (90%) of participants. Of these, 291 (78%) and 321 (86%) responses were recorded within the pre-defined "window" of 10-14 months. Participants in the intervention group had lower HbA1c than those in the control (mean difference -0.24%; 95% Confidence Interval -0.44 to -0.049; p=0.014). There was no significant overall difference between groups in the mean PAID score (p=0.21), but prespecified subgroup analysis of participants who had had diabetes for less than 7 years showed a beneficial impact of the intervention in this group (p = 0.004). There were no reported harms.

Conclusions
Access to HeLP-Diabetes improved glycaemic control over 12 months.

Registration
Trial registration ISRCTN02123133.
Word count of abstract = 299.

Strengths and Limitations of this study
• The trial recruited to target and achieved reasonable follow-up; hence the results for the population of participants are robust; • However, despite wide inclusion criteria and a deliberately pragmatic design, trial participants were well-controlled at baseline, and therefore the extent to which the trial results generalise to the wider population of people with type 2 diabetes is open to discussion.

Introduction.
There is a global epidemic of type 2 diabetes mellitus (T2DM). An estimated 422 million adults, or 10% of the global population, were living with diabetes in 2014 of whom around 90% had type 2 diabetes. 1 Poorly controlled diabetes is associated with premature mortality, and a high risk of complications, including cardiovascular disease, nephropathy and retinopathy. The risk of complications can be reduced by good control of glycaemia and cardiovascular risk factors. 2 3 Interventions which improve self-management skills for patients with diabetes can improve health outcomes and reduce health care costs 4 and international guidelines support training patients in self-management. 3 5 However, it is not clear how best to support patients in developing such skills, and uptake of diabetes selfmanagement education remains low. In England, despite over 90% of eligible patients being referred, 6 only 5.3% attended self-management training in 2014 - 15. 7 Poor uptake may be related to the dominant model of structured education, which is groupbased sessions, lasting a half or whole day, or spread over regular sessions over several weeks. 8 Many patients, such as those who work, those with caring commitments, or those who are uncomfortable in groups, may find it difficult to attend. 9 10 Web-based support for self-management could address some of these barriers, particularly in high income countries, where levels of web-access are high. In the UK over 80% of households had Internet access in 2015, and internet access amongst older people continues to grow steadily. 11 12 Potential advantages include convenience, anonymity, regular updates, and the potential to use video and graphics to present complex information in a format accessible to those with low literacy. 13 Although systematic reviews have confirmed that computer-based interventions can improve health outcomes in diabetes 14 , not all such interventions have a beneficial impact, with meta-analyses showing substantial heterogeneity related to widely differing interventions, including in the use of theory to develop the intervention, 15 outcomes, 14 16 and the duration of follow-up, with most trials having relatively short follow-up (less than 12 months). 14 This is the first UK-based trial of a comprehensive, web-based self-management support programme for people with type 2 diabetes.
This trial assessed the effects of a web-based self-management programme, called Healthy Living for People with Diabetes (HeLP-Diabetes), on glycated haemoglobin (HbA1c) and diabetes-related distress over 12 months.

Methods.
Trial Design and participants: multi-centre, two-arm individually randomised controlled trial in 21 General Practices in England with a mix of urban, suburban and rural practices. Practices were required to have two nurses -one to facilitate access to the intervention, and one to collect data.
Recruitment: standard opt-in recruitment procedures were followed. Each practice had a register of patients with T2DM. The electronic medical record of every patient on this register was reviewed to screen out ineligible patients, and the remainder were sent a letter from their GP, inviting them to participate in the study. Eligible participants were adults, aged 18 or over, with T2DM, registered with participating general practices. Patients were excluded if they were unable to provide informed consent; unable to use a computer due to severe mental or physical impairment; had insufficient spoken or written English to use the intervention (operationalised as unable to consult without an interpreter); were terminally ill with less than 12 months life expectancy; or were currently participating in a trial of an alternative self-management programme. Participants were not required to have home  17 There were no changes to the methods after the protocol was agreed and the start of the trial. Ethical approval was obtained from Camden and Islington National Research Ethics Service (NRES) committee, reference 12/LO/1571.

Patient involvement
Patients were involved in all stages of the study, including contributing to the original application for funding as co-investigators; substantive and ongoing contribution to intervention development; contributing to the trial design, including the decision to have two co-primary outcomes; active membership of the Trial Steering Committee and Trial Management Group; and contributing to the writing of this paper. This last role is recognised through co-authorship (MK).

Randomisation and blinding
Randomisation marked the point of study entry. It was performed centrally (independently of the trial team), after written informed consent was obtained and all baseline data were completed, using a web-based randomisation system, at the level of the individual participant. Randomisation was conducted in a 1:1 ratio using random permuted blocks of sizes 2, 4 and 6, stratified by recruitment centre. Participants were informed the trial compared two forms of web-based support, and were blinded as to which was the intervention and which the comparator. Nurses who offered facilitation for the intervention could not be blinded, but were asked not to discuss details of allocation with the nurses who gathered follow-up data. The research team obtaining and analysing data from participants were blind to allocation.

Intervention
The intervention consisted of facilitated access to HeLP-Diabetes. Facilitation consisted of an introductory training session with the practice nurse. In this appointment patients were were shown how to log on, set a user name and password, and introduced to the content of the website. HeLP-Diabetes was a theoretically informed web-based programme whose overall goals were to improve health outcomes and reduce diabetes-related distress. 18 Overall content was guided by the Corbin and Strauss model of managing a long term condition which posits that patients must undertake medical, emotional and role management. 19 It was developed using participatory design principles, with substantial input from users, defined as patients with T2DM and health professionals caring for such patients. All content was evidence-based, drawing on evidence on management of diabetes, promoting behaviour change and emotional wellbeing, and maximising usability and engagement. Content was designed to be accessible to people with a wide range of literacy and health literacy skills, with all essential content provided in both video and text. There were information sections on diabetes, how diabetes is treated, possible complications of diabetes, possible impacts of diabetes on relationships at home and at work, dealing with unusual situations like parties, holidays, travelling or shift work, and what lifestyle modifications will improve health. There were sections addressing skills and behaviour change, including behaviour change modules on eating healthily, losing weight, being more physically active, smoking cessation, moderating alcohol consumption, managing medicines, glycaemic control and blood pressure control. Users could set the programme to send themselves reminder text messages or emails, and could specify the content and frequency of such reminders. The third strand of components focused on emotional well-being with self-help tools based on cognitive behavioural therapy and mindfulness. There were multiple personal stories (used with license from health talk online), and a moderated forum. Participants were free to use the programme as much or as little as they chose. Engagement with the programme was promoted through regular newsletters, emails and SMS containing updates on latest diabetes-related research or practice, seasonally-relevant advice (e.g. fasting during Ramadan, benefits of 'flu vaccinations), and links to specific relevant parts of the programme. Two or three prompts were sent each month, although users could opt out of receiving them. Further details are provided in Appendix 1.

Comparator
From an NHS perspective, the important research question was whether the proposed intervention could improve health outcomes when compared to current practice. However, to improve acceptability to participants and to maintain blinding, all participants had access to a website. Participants in the control arm were given access to a simple information website, based on the information available on the website of the main UK diabetes charity (Diabetes UK) or National Health Service patient information website (NHS Choices). They received the same initial facilitation meeting as participants in the intervention group, in which they were shown how to log on, set a user name and password, and how to use the website.

Primary outcomes
The outcomes reflected the dual goals of improving health outcomes and reducing diabetesrelated distress. The two joint primary outcomes were glycated haemoglobin (HbA1c) and diabetes-related distress, measured by the Problem Areas in Diabetes (PAID) scale, both at 12 months post-randomisation. PAID has 20 items focusing on areas that cause difficulty for people living with diabetes, including social situations, food, friends and family, diabetes treatment, relationships with health care professionals and social support. 20 PAID scores range from 0 -100, with higher scores indicating more distress. A score of 40 or more indicates significant distress, and around 40% of patients with diabetes experience significant distress. 21

Secondary outcomes
Clinical secondary outcomes included systolic and diastolic blood pressure; body mass index; total cholesterol and HDL (not fasting); and completion of the "9 essential processes" for effective management of diabetes, mandated by NHS England (= weight, BP, smoking status, measurement of serum creatinine, cholesterol and HbA1c, urinary albumin and assessment of eyes and feet) within the previous 12 months. 3 Patient-reported outcomes included depression and anxiety, measured using the Hospital Anxiety and Depression Scale (HADS); 22 diabetes-related self-efficacy measured using the Diabetes Management Self-Efficacy Scale (DMSES); 23 and satisfaction with treatment, measured using the Diabetes Satisfaction with Treatment Questionnaire status and change version (DTSQs & DTSQc). 24

Data collection
Data were collected at baseline, 3 and 12 months, with 12 months the primary endpoint. Patient-reported data were collected using online questionnaires emailed to participants. Clinical outcomes were collected by nurses in participating practices. Participants were asked to complete their online questionnaires before visiting the nurse for clinical measurements and blood tests. Blood samples were analysed at the local NHS laboratory used by participating practices for routine clinical analyses. Data on completion of the "9 essential processes" were collected from the GP record for the 12 months prior to randomisation and the 12 months after randomisation at the 12 month follow-up point to avoid triggering behaviour change amongst the study nurses. Use of the intervention was recorded automatically using bespoke software that recorded the date, and time of each

Sample size Calculation
Our original sample size calculation was that randomising 350 participants with 85% followup would provide 90% power at the 5% level of significance to detect a 0.25% difference in HbA1c and a 4.0 point difference in PAID score at 12 months post-randomisation between the randomised groups. 25 26 Since HbA1c and PAID were joint primary outcomes measuring different aspects of T2DM, both were tested at a 5% significance level.

Analysis
The analysis followed a pre-specified analysis plan, based on comparing the groups as randomised (intention-to-treat). The analysis plan was approved by the Trial Steering Committee before unblinding and uploaded to the trial website, https://www.ucl.ac.uk/pcph/research-groupsthemes/ehealth/projects/projects/helpdiabetesrct. Only HbA1c and PAID measured within a 10-14 month window period following randomisation was used in the primary analysis with missing 12-month outcomes multiply imputed using baseline and other outcome data (e.g. 3m data and final follow-up data collected outside the 10-14 month window). Further information on the imputation method is given in Appendix 2.
A linear mixed effects model with random centre effects was used to analyse each of the primary outcomes separately, adjusting for the baseline level of the outcome, age, gender, previous participation in other self-management programmes, pre-existing cardiovascular disease and time since diagnosis of diabetes. Secondary outcome measures were analysed similarly using generalised linear mixed models, with a normal residual error structure for continuous outcomes and a logit link for the binary outcome `Completion of 9 essential processes'. Pre-specified sub-group analysis for the co-primary outcomes was undertaken by baseline glycaemic control (HbA1c outcome only), baseline PAID (PAID outcome only), and duration of diabetes, treating all potential effect modifiers as continuous. The interaction between randomised group and each effect modifier was included in the model separately and assessed using a Wald test.
Use of the intervention was investigated as a mediator for efficacy, using instrumental variable methods, with randomisation as the instrument (Supplementary Figure 1). 27 28 Potential contamination was monitored by recording participants with similar family names and identifying those with the same addresses. Where this occurred, it was dealt with in the analysis by reporting the extent and undertaking a sensitivity analysis excluding these individuals.
A number of other sensitivity analyses were performed to assess the robustness of the primary analyses: 1) performing two complete case analyses disregarding outcomes measured outside 10-14 months and 11-13 months post-randomisation; 2) repeating the analysis using multiple imputation of baseline covariates only; 3) fitting linear models excluding centre random-effects; and 4) fitting an unadjusted model using only outcome measured in 10-14 months postrandomisation.
The TSC took on the role of the data monitoring committee. Trial registration ISRCTN02123133. Recruitment took place between September 2013 and December 2014. An initial 421 patients consented to participate, but of these 47 did not fully complete their baseline questionnaires and were therefore not randomised and did not enter the study. A total of 374 participants were randomised, of whom 86% (n = 321) provided data on PAID, and 78% (n = 291) had HbA1c measured within 10 to 14 months of randomisation. Additional final outcome data, obtained outside the 10 -14 month pre-defined window, were available for a further 27 participants for HbA1c and 16 participants for PAID ( Figure 1). Data obtained outside the 10 -14 month window were not used directly in the primary analysis, but were entered into the imputation model (Supplementary Table 1).

Baseline characteristics
Baseline demographic and clinical characteristics are shown in Table 1. The mean age was nearly 65 years, over two-thirds (n = 258, 69%) were male, and most were White British (n = 300, 80%). Nearly all (n=370, 99%) had a computer with access to the internet at home and just over half (n = 210, 56%) rated themselves as experienced computer users. Around onethird (n = 134; 36%) had been diagnosed for less than 5 years, with a further third (n = 115, 31%) having been diagnosed between 5 and 9 years ago. Overall, this was a population with well-controlled diabetes at baseline (mean HbA1c was 7.3% (56 mmol / mol)) and low levels of distress (mean PAID = 19).

Primary outcomes
At twelve months the primary analysis showed a significant difference in change in HbA1c between the randomised groups with participants in the HeLP-Diabetes group having a lower HbA1c than those in the control group (mean difference = -0.24%; 95% confidence intervals -0.44 to -0.049, p=0.014) (Table 2, Figure 2). There was no difference in change in PAID scores between the groups at 12-months (mean difference -1.5; 95% CI -3.9, 0.9, p=0.209), though both groups showed a decrease in PAID over the follow-up of the trial (Table 2, Figure 3).

Secondary outcomes
There was no difference in secondary outcomes at 12 months, with the possible exception of systolic blood pressure, which decreased more in the intervention group than in the control group (p=0.010) ( Table 2); though the result was not statistically significant after correction for multiple testing of secondary outcomes. There were no significant differences between groups on any of the outcome measures amongst individuals who completed three month outcomes (Supplementary Table 2). No adverse effects or events were recorded during follow-up.

Usage data
The mean number of log-ins was significantly higher in the intervention group than the control group (18.7 vs. 4.8, p= 0.0001), as was the mean number of pages visited per log-in (10.5 vs. 7.7, p <0.0001) and the mean number of days in which the website was accessed (10.1 vs. 3.3, p<0.0001) ( Table 3). The causal analyses estimated that for a "high-usage" population (those with usage ≥ the median of 4 days) the HeLP-Diabetes intervention could on average reduce HbA1c by -0.44% (95% CI -0.81 to -0.06) and PAID by -2.8 (95% CI -7.2 to 1.7) over 12 months ( Supplementary Figures 2 and 3). The mean usage in the "highusage" group was 18 days. It should be noted that the usage data presented do not include the initial facilitation visit. There was a technical error in the software which led to usage data not being collected before 1 January 2014. At this point 16 participants had been randomised (7 to intervention, 9 to control). For these 16 participants, the usage data is not based on a full year, but for all other participants, data are summarised for the 12 months post-randomisation. The findings from the sensitivity analyses, including a complete-case analysis, were similar to the main analysis (Supplementary Table 3). Participants who were missing 12-month HbA1c had significantly higher mean baseline HbA1c measures (7.9% vs. 7.1%, p<0.001) leading to higher imputed HbA1c at 12-months in the non-completers and a greater mean difference between the randomised groups than from complete case analyses (Supplementary Figure 4, Supplementary Table 3).

Subgroup analyses
Pre-specified subgroup analyses showed that there was no evidence of baseline measures of HbA1c or PAID being effect modifiers for the mean difference between the groups. There was strong statistical evidence (interaction p=0.004) to suggest that the duration of diabetes acted as an effect modifier, with those who had been diagnosed more recently (<7 years) showing more of a reduction in PAID than those who had been diagnosed for longer periods of time. Duration of diabetes had no effect on change in HbA1c (Supplementary Table 4).

Harms
There were no reported harms in either group.

Discussion
In this first UK-based trial of a web-based self-management programme for people with T2DM, participants randomised to HeLP-Diabetes demonstrated improved glycaemic control at 12 months compared to those randomised to a simple information website. This improvement is both clinically and statistically meaningful, appears robust across all prespecified sensitivity analyses, and was not dependent on duration of diabetes, baseline glycaemic levels or level of diabetes-related distress. Each 1% reduction in HbA1c is associated with a risk reduction of 21% for deaths related to diabetes and a 37% risk reduction for microvascular complications. 26 Given that this web-based intervention could be delivered at low-cost and at scale across the UK, the potential for population benefit is considerable. Moreover, in contrast to group-based education, where the effects appear to wane with time, 29 the effects of HeLP-Diabetes were greater at 12 months than at 3 months.
There was no overall impact on diabetes-related distress, but some evidence that HeLPdiabetes appeared to reduce distress in recently diagnosed individuals. However, it is worth noting that baseline PAID scores were exceptionally low in this trial population. In a small pilot study, participants offered supported access to HeLP-Diabetes reduced their PAID scores by 6 points (p=0.04) over 6 weeks. 30 The trial has many strengths. It was a pragmatic trial, open to nearly all patients with T2DM in participating practices. Concealment of allocation was complete, as randomisation occurred after baseline data collection. Baseline prognostic factors were well balanced between groups. Every effort was made to achieve blinding, including requiring practices to have two nurses, so that data collection was undertaken by a nurse blind to participant allocation. Data for the co-primary outcomes at the primary outcome point were available for 78% and 86% of participants for HbA1c and PAID respectively. All analyses were on an intention-to-treat basis, supplemented by a CACE analysis. Although response rates for the co-primary outcomes were good, some potential for bias existed. Our primary analysis used multiple imputation methods because evidence shows that the assumptions underpinning this method are more defensible than those assumed using other approaches to missing data. 31 We also undertook sensitivity analyses including complete cases, non-contaminated cases, and a linear model excluding centre; all yielded similar results.
The two co-primary outcomes reflected the twin aims of the intervention: to improve diabetes control and to reduce diabetes-related distress. Around 40% of patients with diabetes have significant levels of distress, which severely impacts on quality of life, 32 and diabetes-related distress is an important outcome for patients. 33 Our patient and public involvement (PPI)  10 panel were clear that this should be a primary outcome, and a recent meta-ethnography emphasised the importance of empowerment and quality of life in promoting long term engagement with self-management. 34 In contrast, many health care professionals are more interested in glycaemic control. In line with previous trials in this area, 35 we decided to adopt both as co-primary outcomes and to test both at a 5% level of significance. 36 There are some limitations. Despite maximising the inclusivity of the trial by minimising the exclusion criteria, participants were not representative of the overall population of patients with type 2 diabetes in England. Compared to the overall population, participants had better control of their diabetes and cardiovascular risk factors, 6 7 and were much less distressed. 21 This finding mirrors that of a recent systematic review of demographic factors associated with web portal usage amongst people with diabetes which found that those with well controlled diabetes were more likely to use such portals than those with poor control. 37 However, fewer of our participants self-rated their computer skills as excellent (57% of our sample compared to a national average of 73%). 12 This good control at baseline has two implications -first, that there was little room for improvement in this population, and secondly, that this population may have been unusually motivated to self-manage their diabetes. Although every effort was made to maintain blinding, it is possible that some participants may have discussed their use of the intervention with research nurses, making it possible to infer which arm they had been allocated to. This could have affected research nurses' measurements of secondary clinical outcomes, such as blood pressure or weight, but could not have affected assessment of glycated haemoglobin as this was measured by laboratory staff who were blinded. There appeared to be high potential for contamination between two participants who shared the same surname and address, and a further two participants did not receive their allocated intervention due to an error at practice level; excluding these four made no difference to the results. A further limitation of the trial is that it provides little insight into the mechanism of action of HeLP-Diabetes. This was the result of a deliberate decision to focus on clinically important outcomes and minimise both the response burden and the potential impact of measurement on participants.
This is the first UK-based trial of a web-based self-management programme for people with type 2 diabetes, and internationally, the first trial of such a comprehensive intervention that aims to address the three main tasks of self-management: emotional, medical and role management. 19 In the Cochrane review of computer-based self-management interventions for people with T2DM, only 4 of the included studies had follow-up of 12 months or more. 14 Of these, three interventions were clinic-based, with participants completing self-assessment tools on a touch screen and receiving tailored advice during their baseline visit to their diabetes clinician [38][39][40] and one was a mobile phone-based intervention which provided tailored messages in response to participant's results of blood glucose self-monitoring data. 41 A more recent systematic review of internet delivered diabetes self-management identified 2 trials with 12 or more months follow-up. 42 One trial was on a structured intervention based on a peer-led, group-based, diabetes self-management course 43 . There were six sessions, with each session available for one week. Each session required participants to make a specific action plan to address a problem they were experiencing. Peer facilitators encouraged use of the programme. Follow up was planned at 6 and 12 months; however HbA1c data were only available at 6 months. The other trial compared two versions of a web-based intervention (with and without additional social support) to enhanced usual care. The web-based intervention was designed using social-cognitive theory and a social-ecological model, with a focus on three main behaviours: dietary intake, physical activity and medication adherence. Users of either web-based intervention received motivational phone calls to encourage adherence and development of action plans. Those randomised to the enhanced intervention (with additional social support) received two additional phone calls and an invitation to attend a group session. There was no difference between groups in HbA1c or other biological outcomes at 12 months. 44 Thus the results of this trial add significantly to the available literature.

Page 11 of 40
For peer review only -http://bmjopen.bmj.com/site/about/guidelines.xhtml 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  On the basis of these results, HeLP-Diabetes may be considered as an addition to the current menu of self-management support for people with type 2 diabetes, and may help increase overall access and uptake. Most commissioned services currently focus on newly diagnosed patients, leaving clear unmet need for people who have had their diabetes for longer, but are looking for ways to improve their health. Many patients are not ready to engage in self-management early in their illness journey, 9 but become motivated to do so later, often as a result of a change in medication or development of a complication. 45 The intervention is low cost, and as most costs are fixed, irrespective of number of users, is likely to be cost-effective, particularly if widely used. A cost-effectiveness analysis of HeLP-Diabetes will be reported separately.

Funding.
This paper presents independent research funded by the National Institute for Health Research (NIHR) under its Programme Grants for Applied Research Programme (Grant Reference Number RP-PG-0609-10135). The views expressed are those of the author(s) and not necessarily those of the NHS, the NIHR or the Department of Health. The funder had no role in the study design, data collection, data analysis, data interpretation, or writing of the report. All authors had full access to all the data in the study and can take responsibility for the integrity of the data and data analysis. The lead author had final responsibility for the decision to submit for publication, affirms that the manuscript is an honest, accurate, and transparent account of the study being reported; that no important aspects of the study have been omitted; and that there were no significant discrepancies from the published protocol.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  We gratefully acknowledge the permission to use under license the validated behaviour change modules for weight loss (POWeR), alcohol reduction (DownYour Drink) and smoking cessation (StopAdvisor and the diabetes module from Healthtalk online (http://www.healthtalk.org/) in HeLP-Diabetes.
We are grateful to Orla O'Donnell for outstanding project management, Fiona Giles for administrative support, all our PPI who contributed to the development, maintenance and delivery of the intervention and / or the management and oversight of the trial, staff at the participating practices, PCRN staff, and all our participants.

Data Sharing.
Patient level data, the full dataset, statistical code are available from the corresponding author. Consent for data sharing was not obtained from participants, but the potential benefits of sharing these data outweigh the potential harms as the data are anonymised.

Rationale 4
The aim of HeLP-Diabetes was to enable people with type 2 diabetes to lead healthier lives 5 and improve overall quality of life. In order to do this, we adopted a strong theoretical 6 framework, which included: 7 1. The Corbin and Strauss model of living with a long term condition, 2 which determined 8 the overall scope of the intervention; 9 2. Normalisation Process Theory, 3 to maximise likelihood of successful implementation 10 of the intervention into routine health care 11 3. A taxonomy of behaviour change techniques, 4 to guide the content of the behaviour 12 change modules; 13 4. Finally, we worked within the paradigm of evidence-based medicine, ensuring all 14 content was congruent with current best evidence. This applied both to the medical 15 content (e.g. information about treatments) and to the delivery of the intervention 16 (e.g. evidence about maximising acceptability, uptake and adherence). 17 Self-management of a long term condition is a complex process, which Corbin and Strauss 18 characterised as requiring medical, emotional and role management. 2 Medical management 19 includes behaviour change, such as eating healthily, being physically active, stopping 20 smoking and taking medication. Working with health professionals and the health care 21 system (e.g. keeping appointments, remembering to schedule check-ups or regular 22 monitoring tests) is also included in medical management. Equally important, and often just 23 as challenging for individuals concerned, are emotional management and role management.

24
Emotional management refers to managing the complex negative emotions that result from 25 being diagnosed with a long term condition, such as depression, anger, guilt, shame and a 26 sense of stigma. Role management requires adjusting to the change in biographical 27 narrative, acceptance of the role of "patient", and negotiating inevitable changes in 28 relationships with family, friends and colleagues. This characterisation of the work of self-29 management provided the overall framework for the content, tone and style of the 30 intervention. As these tasks are intertwined and inter-dependent, we designed the content 31 to address all three tasks in an integrated fashion. 32 Normalisation Process Theory is a mid-range sociological theory which predicts and explains 33 whether and why certain innovations will be effectively implemented into routine health care 34 and become "normalised", that is, become so much part of routine practice that they 35 disappear from view. 3 It posits that interventions which are easily understood and 36 distinguished from other interventions (coherence), which relevant professionals can see the 37 advantage of compared to current or alternative practice (cognitive participation), which fit 38 well with existing professional relationships and workflows (collective action), and where 39 users are provided with formal or informal feedback and evidence about the impact of the 40 intervention (reflexive monitoring) are more likely to be implemented than those that do not. 41 These constructs guided the development of the intervention by, for example, ensuring 42 HeLP-Diabetes integrated well with existing workflows in general practice. 43 Effective self-management in diabetes often requires patients to change deep-rooted 44 behaviours, particularly around diet and physical activity. Rather than adopt a single 45 behavioural theory, we opted to design our behaviour change modules around specific 46 HeLP-Diabetes was designed to be used as part of an overall package of care for people 93 with diabetes. Low usage, or non-adherence to internet interventions is well-recognised 94 problem, and our preparatory work with patients and health care professionals indicated that 95 integrating the intervention into routine care was likely to improve uptake and adherence. 96 Hence we made the programme available to registered users only, and encouraged health 97 care professionals to register patients. Once registered, patients could use the programme 98 as much (or as little) as they wanted. There was no prescribed level of use, as our proposed 99 users included patients at all stages of their illness journey, from those newly diagnosed to 100 those who had lived with diabetes for many years. As such, we anticipated that each user 101 would have different needs and priorities, and the programme was designed to allow users 102 to pick and choose sections that were most relevant and beneficial for them personally. 103 There was a limited amount of tailoring. Additional resources and sources of help were 104 tailored by the patient's CCG, but otherwise tailoring was limited to the behaviour change 105 and health record sections where users entered their own goals or data. 106 107

Registration and facilitation. 108
During the trial, registration was performed by a practice nurse. During the registration 109 procedures users were asked to select a password and username. After registration, the 110 nurse was asked to demonstrate the intervention to the patient, and integrate use of the 111 intervention into the patient's personal diabetes care plan by, for example, discussing what 112 goals the patient would like to set for the coming period, and showing the patient how to use 113 the intervention to set goals, monitor progress, and, if desired, arrange for automated 114 reminders by SMS or email to be sent by the programme. 115 Registration was undertaken in the patient's practice. All subsequent use of the intervention 116 was at any location convenient for the patient with internet access. For most people, we 117 expected this to be at home, or at the home of a relative. However, all users were given 118 information about local services (usually libraries) offering free access to an internet-119 connected computer. 120 121

Modifications during the trial 139
One of the key functions of the programme was to provide up-to-date, evidence-based 140 information. Hence the site was regularly reviewed to ensure all content was up-to-date, 141 evidence-base, and congruent with current NICE guidelines. In practice, this meant small 142 updates each month, with a complete review when the NICE guidelines on management of 143 diabetes were updated. 9 144 145

Assessment of fidelity 146
Intervention use was assessed through bespoke software which recorded the date, time and 147 pages viewed for each log in by each user. Practice nurses were trained in registration and 148 facilitation procedures, but we were unable to monitor how well they adhered to them. 149 150

Availability of the intervention 151
The intervention is currently available for commissioning from a not-for-profit Community 152 Interest

Item
Line numbers Provide the name or a phrase that describes the intervention 2,3 Describe any rationale, theory or goal of the elements essential to the intervention

-48
Materials: describe any physical or informational materials used in the intervention, including those provided to participants or used in intervention delivery or in training of the intervention providers 51-90 Procedures: Describe each of the procedures, activities and /or processes used in the intervention, including any enabling or support activities

-137
For each category of intervention provider, describe their expertise, background, and any specific training given.

109
Describe the modes of delivery (such as face-to-face or by some other mechanism, such as internet or telephone) of the intervention, and whether it was provided individually or in a group

-106
If the intervention was planned to be personalised, titrated or adapted, then describe what, why, when and how.

-106
If the intervention was modified during the course of the study, describe the changes (what, why, when and how)

-144
Planned: if intervention adherence or fidelity was assessed, describe how and by whom, and if any strategies were used to maintain or improve fidelity, describe them.

-149
Actual: if intervention adherence or fidelity was assessed, describe the extent to which the intervention was delivered as planned.

Statistical Methods for Multiple Imputation
Multiple Imputation using chained equation was used as the primary method to account for missing data (in both baseline and follow-up data). 1 A set of imputation models were specified, one for each variable with missing data. Each variable was then regressed on all other variables, including completely recorded baseline and follow-up variables and stratified by randomised group. Imputations were performed using predictive mean matching using the five nearest neighbours to the prediction as a set to draw from. The full list of variables considered in the MICE approach is shown in Supplementary Table 1, together with the number of missing values for each variable and time period.
Since only measurements within a 10-14 month window period were used within the main analyses of HbA1c and PAID, the following imputation procedure was implemented for these two co-primary outcomes. Twelve month measurements were subdivided into those that were measured within 10-14 months (the primary outcome variable) and those that were measured outside 10-14 months (a variable used for imputing only). For HbA1c, two additional variables were created for use within the imputation model; 1) the time in days from randomisation that the "12-month" HbA1c measurement was actually taken for values inside of the window period (and set to 365 for measurements taken outside the window period), 2) the time in days from randomisation that the "12-month" HbA1c measurement was actually taken for values outside the window period (and set to missing for measurements taken within the window period). The first of these variables gives the desired time for imputing HbA1c measurements when they are missing, whilst the second gives information on how far outside of the window the actual measurements were taken. Corresponding variables were created for the "12-month" PAID measurement. Finally, two additional variables were created defining the time in days at which HbA1c and PAID were measured at "3-months". All variables were included within the chained equations and imputed where necessary. 40 imputed datasets were created, the analysis models were fitted to each imputed dataset separately, and the estimates were pooled using Rubin's rules.
Statistical Methods for causal analyses The causal analysis proposed attempts to address how the effectiveness of the intervention is mediated through the frequency of website usage. In particular, it is important to understand whether prolonged usage of the website modifies the efficacy of the intervention. Since website usage is measured post-randomisation a naïve analysis of correlating usage with outcomes in the intervention group may give biased and misleading results, since there may be unmeasured confounders also correlated with the outcomes that distinguish the motivated users who regularly log-in from the less motivated ones. Causal analyses using instrumental variables (IV) were therefore used to determine the effect of website usage on outcomes. This approach preserves randomisation (i.e. provides a comparison independent of observed and unobserved confounders).
"Usage" is defined as the proportion of follow-up (rescaled as no. days in a year) that the HeLP-Diabetes website is accessed. It was assumed that the efficacy of the intervention is zero for individuals who never log-in (the exclusion restriction assumption). Website usage in the control group was ignored in the model as it was assumed that the control website was unlikely to be effective. (NB. usage statistics were collected for the comparator website but were not used in these analyses). This main underlying assumption of the causal analysis is that the effect of randomisation to the HeLP-Diabetes intervention on 12-month outcomes occurs only through use of the website (Supplementary Figure 1; Z is randomised intervention, WU is website-usage and Y is 12-month outcome, e.g. HbA1c or PAID). This relies of the "exclusion restriction" assumption that the HeLP Diabetes intervention has no effect when usage is zero (i.e. for individuals who never log-in). Hence randomisation is assumed to be an instrumental variable. Figures 4 and 5 show the estimated causal effects of HeLP Diabetes on HbA1c and PAID at 12-months, respectively, by level of website usage (the "exclusion restriction" assumption leads to zero efficacy at zero usage). To determine the casual estimate of efficacy for a "high-usage" group, the individual predicted efficacy was calculated for each individual greater than or equal to the median usage of 4 days (shown as the red boxes in the histograms in Figures 4 and 5) and the average efficacy for this group was obtained. The mean usage in the "high-usage" group was 18 days.     * Partial mean matching performed using 5 nearest neighbours † 12-month measurements were subdivided into those that were measured within 10-14 months (primary outcome variable) and those that were measured outside 10-14 months (variable used for imputing only).  ƚ excluding outcomes outside 10-14 months post-randomisation and those missing baseline covariates ǂ excluding outcomes outside 11-13 months post-randomisation and those missing baseline covariates ¥ Excluding patients who were suspected to have been exposed to the alternative intervention * likelihood ratio test for including centre in the model as a fixed effect; p=0.617 (HbA1c) ** likelihood ratio test for including centre in the model as a fixed effect; p=0.357 (PAID)   1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60 Introduction Background and objectives 2a

Z WU Y
Scientific background and explanation of rationale 2b Specific objectives or hypotheses

Trial design 3a
Description of trial design (such as parallel, factorial) including allocation ratio 3b Important changes to methods after trial commencement (such as eligibility criteria), with reasons Participants 4a Eligibility criteria for participants 4b Settings and locations where the data were collected Interventions 5 The interventions for each group with sufficient details to allow replication, including how and when they were actually administered Outcomes 6a Completely defined pre-specified primary and secondary outcome measures, including how and when they were assessed 6b Any changes to trial outcomes after the trial commenced, with reasons Sample size 7a How sample size was determined 7b When applicable, explanation of any interim analyses and stopping guidelines Randomisation: Sequence generation 8a Method used to generate the random allocation sequence 8b Type of randomisation; details of any restriction (such as blocking and block size) Allocation concealment mechanism 9 Mechanism used to implement the random allocation sequence (such as sequentially numbered containers), describing any steps taken to conceal the sequence until interventions were assigned Implementation 10 Who generated the random allocation sequence, who enrolled participants, and who assigned participants to interventions Blinding 11a If done, who was blinded after assignment to interventions (for example, participants, care providers, those  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48 assessing outcomes) and how 11b If relevant, description of the similarity of interventions Statistical methods 12a Statistical methods used to compare groups for primary and secondary outcomes 12b Methods for additional analyses, such as subgroup analyses and adjusted analyses

Results
Participant flow (a diagram is strongly recommended) 13a For each group, the numbers of participants who were randomly assigned, received intended treatment, and were analysed for the primary outcome 13b For each group, losses and exclusions after randomisation, together with reasons Recruitment 14a Dates defining the periods of recruitment and follow-up 14b Why the trial ended or was stopped Baseline data 15 A table showing baseline demographic and clinical characteristics for each group  Numbers analysed  16 For each group, number of participants (denominator) included in each analysis and whether the analysis was by original assigned groups Outcomes and estimation 17a For each primary and secondary outcome, results for each group, and the estimated effect size and its precision (such as 95% confidence interval) 17b For binary outcomes, presentation of both absolute and relative effect sizes is recommended Ancillary analyses 18 Results of any other analyses performed, including subgroup analyses and adjusted analyses, distinguishing pre-specified from exploratory Harms 19 All important harms or unintended effects in each group (for specific guidance see CONSORT for harms)

Limitations 20
Trial limitations, addressing sources of potential bias, imprecision, and, if relevant, multiplicity of analyses Generalisability 21 Generalisability (external validity, applicability) of the trial findings Interpretation 22 Interpretation consistent with results, balancing benefits and harms, and considering other relevant evidence

Registration 23
Registration number and name of trial registry Protocol 24 Where the full trial protocol can be accessed, if available Funding 25 Sources of funding and other support (such as supply of drugs), role of funders *We strongly recommend reading this statement in conjunction with the CONSORT 2010 Explanation and Elaboration for important clarifications on all the items. If relevant, we also recommend reading CONSORT extensions for cluster randomised trials, non-inferiority and equivalence trials, non-pharmacological treatments, herbal interventions, and pragmatic trials.

Strengths and Limitations of this study
• The trial recruited to target and achieved reasonable follow-up; hence the results for the population of participants are robust (internal validity); • The two co-primary outcomes reflected the goals of the intervention, namely improving diabetes control and reducing diabetes-related distress; • However, despite wide inclusion criteria and a deliberately pragmatic design, trial participants were well-controlled at baseline, and therefore the extent to which the trial results generalise to the wider population of people with type 2 diabetes is open to discussion (external validity).

Page 4 of 41
For peer review only -http://bmjopen.bmj.com/site/about/guidelines.xhtml 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  There is a global epidemic of type 2 diabetes mellitus (T2DM). An estimated 422 million adults, or 10% of the global population, were living with diabetes in 2014 of whom around 90% had type 2 diabetes. 1 Poorly controlled diabetes is associated with premature mortality, and a high risk of complications, including cardiovascular disease, nephropathy and retinopathy. The risk of complications can be reduced by good control of glycaemia and cardiovascular risk factors. 2 3 Interventions which improve self-management skills for patients with diabetes can improve health outcomes and reduce health care costs 4 and international guidelines support training patients in self-management. 3 5 However, it is not clear how best to support patients in developing such skills, and uptake of diabetes selfmanagement education remains low. In England, despite over 90% of eligible patients being referred, 6 only 5.3% attended self-management training in 2014 - 15. 7 Poor uptake may be related to the dominant model of structured education, which is groupbased sessions, lasting a half or whole day, or spread over regular sessions over several weeks. 8 Many patients, such as those who work, those with caring commitments, or those who are uncomfortable in groups, may find it difficult to attend. 9 10 Web-based support for self-management could address some of these barriers, particularly in high income countries, where levels of web-access are high. In the UK over 80% of households had Internet access in 2015, and internet access amongst older people continues to grow steadily. 11 12 Potential advantages include convenience, anonymity, regular updates, and the potential to use video and graphics to present complex information in a format accessible to those with low literacy. 13 Although systematic reviews have confirmed that computer-based interventions can improve health outcomes in diabetes 14 , not all such interventions have a beneficial impact, with meta-analyses showing substantial heterogeneity related to widely differing interventions, including in the use of theory to develop the intervention, 15 outcomes, 14 16 and the duration of follow-up, with most trials having relatively short follow-up (less than 12 months). 14 This is the first UK-based trial of a comprehensive, web-based self-management support programme for people with type 2 diabetes.

BMJ Open
This trial assessed the effects of a web-based self-management programme, called Healthy Living for People with Diabetes (HeLP-Diabetes), on glycated haemoglobin (HbA1c) and diabetes-related distress over 12 months.

Methods.
Trial Design and participants: multi-centre, two-arm individually randomised controlled trial in 21 General Practices in England with a mix of urban, suburban and rural practices. Practices were required to have two nurses -one to facilitate access to the intervention, and one to collect data.
Recruitment: standard opt-in recruitment procedures were followed. Each practice had a register of patients with T2DM. The electronic medical record of every patient on this register was reviewed to screen out ineligible patients, and the remainder were sent a letter from their GP, inviting them to participate in the study. Eligible participants were adults, aged 18 or over, with T2DM, registered with participating general practices. Patients were excluded if they were unable to provide informed consent; unable to use a computer due to severe mental or physical impairment; had insufficient spoken or written English to use the intervention (operationalised as unable to consult without an interpreter); were terminally ill with less than 12 months life expectancy; or were currently participating in a trial of an alternative self-management programme. Participants were not required to have home Internet access or prior experience of using the Internet to participate. Participants with previous or current experience of self-management education were eligible to participate. Recruitment took place between September 2013 and December 2014. The trial protocol was submitted for publication in June 2014. 17 There were no changes to the methods after the protocol was agreed and the start of the trial. Ethical approval was obtained from Camden and Islington National Research Ethics Service (NRES) committee, reference 12/LO/1571.

Patient involvement
Patients were involved in all stages of the study, including contributing to the original application for funding as co-investigators; substantive and ongoing contribution to intervention development; contributing to the trial design, including the decision to have two co-primary outcomes; active membership of the Trial Steering Committee and Trial Management Group; and contributing to the writing of this paper. This last role is recognised through co-authorship (MK).

Randomisation and blinding
Randomisation marked the point of study entry. It was performed centrally (independently of the trial team), after written informed consent was obtained and all baseline data were completed, using a web-based randomisation system, at the level of the individual participant. Randomisation was conducted in a 1:1 ratio using random permuted blocks of sizes 2, 4 and 6, stratified by recruitment centre. Participants were informed the trial compared two forms of web-based support, and were blinded as to which was the intervention and which the comparator. Nurses who offered facilitation for the intervention could not be blinded, but were asked not to discuss details of allocation with the nurses who gathered follow-up data. The research team obtaining and analysing data from participants were blind to allocation.

Intervention
The intervention consisted of facilitated access to HeLP-Diabetes. Facilitation consisted of an introductory training session with the practice nurse. In this appointment patients were were shown how to log on, set a user name and password, and introduced to the content of the website. HeLP-Diabetes was a theoretically informed web-based programme whose overall goals were to improve health outcomes and reduce diabetes-related distress. 18 Overall content was guided by the Corbin and Strauss model of managing a long term condition which posits that patients must undertake medical, emotional and role management. 19 It was developed using participatory design principles, with substantial input from users, defined as patients with T2DM and health professionals caring for such patients. All content was evidence-based, drawing on evidence on management of diabetes, promoting behaviour change and emotional wellbeing, and maximising usability and engagement. Content was designed to be accessible to people with a wide range of literacy and health literacy skills, with all essential content provided in both video and text. There were information sections on diabetes, how diabetes is treated, possible complications of diabetes, possible impacts of diabetes on relationships at home and at work, dealing with unusual situations like parties, holidays, travelling or shift work, and what lifestyle modifications will improve health. There were sections addressing skills and behaviour change, including behaviour change modules on eating healthily, losing weight, being more physically active, smoking cessation, moderating alcohol consumption, managing medicines, glycaemic control and blood pressure control. Users could set the programme to send themselves reminder text messages or emails, and could specify the content and frequency of such reminders. The third strand of components focused on emotional well-being with self-help tools based on cognitive behavioural therapy and mindfulness. There were multiple personal stories (used with license from health talk online), and a moderated forum. Participants were free to use the programme as much or as little as they chose. Engagement with the programme was promoted through regular newsletters, emails and SMS containing updates on latest diabetes-related research or practice, seasonally-relevant advice (e.g. fasting during Ramadan, benefits of 'flu vaccinations), and links to specific relevant parts of the programme. Two or three prompts were sent each month, although users could opt out of receiving them. Further details are provided in Appendix 1.

Comparator
From an NHS perspective, the important research question was whether the proposed intervention could improve health outcomes when compared to current practice. However, to improve acceptability to participants and to maintain blinding, all participants had access to a website. Participants in the control arm were given access to a simple information website, based on the information available on the website of the main UK diabetes charity (Diabetes UK) or National Health Service patient information website (NHS Choices). They received the same initial facilitation meeting as participants in the intervention group, in which they were shown how to log on, set a user name and password, and how to use the website.

Primary outcomes
The outcomes reflected the dual goals of improving health outcomes and reducing diabetesrelated distress. The two joint primary outcomes were glycated haemoglobin (HbA1c) and diabetes-related distress, measured by the Problem Areas in Diabetes (PAID) scale, both at 12 months post-randomisation. PAID has 20 items focusing on areas that cause difficulty for people living with diabetes, including social situations, food, friends and family, diabetes treatment, relationships with health care professionals and social support. 20 PAID scores range from 0 -100, with higher scores indicating more distress. A score of 40 or more indicates significant distress, and around 40% of patients with diabetes experience significant distress. 21

Secondary outcomes
Clinical secondary outcomes included systolic and diastolic blood pressure; body mass index; total cholesterol and HDL (not fasting); and completion of the "9 essential processes" for effective management of diabetes, mandated by NHS England (= weight, BP, smoking status, measurement of serum creatinine, cholesterol and HbA1c, urinary albumin and assessment of eyes and feet) within the previous 12 months. 3 Patient-reported outcomes included depression and anxiety, measured using the Hospital Anxiety and Depression Scale (HADS); 22 diabetes-related self-efficacy measured using the Diabetes Management Self-Efficacy Scale (DMSES); 23 and satisfaction with treatment, measured using the Diabetes Satisfaction with Treatment Questionnaire status and change version (DTSQs & DTSQc). 24

Data collection
Data were collected at baseline, 3 and 12 months, with 12 months the primary endpoint. Patient-reported data were collected using online questionnaires emailed to participants. Clinical outcomes were collected by nurses in participating practices. Participants were asked to complete their online questionnaires before visiting the nurse for clinical measurements and blood tests. Blood samples were analysed at the local NHS laboratory used by participating practices for routine clinical analyses. Data on completion of the "9 essential processes" were collected from the GP record for the 12 months prior to randomisation and the 12 months after randomisation at the 12 month follow-up point to avoid triggering behaviour change amongst the study nurses. Use of the intervention was recorded automatically using bespoke software that recorded the date, and time of each page visited. A new log-in to the intervention was defined as any page that was accessed 30 minutes or more after the last accessed page.

Sample size Calculation
Our original sample size calculation was that randomising 350 participants with 85% followup would provide 90% power at the 5% level of significance to detect a 0.25% difference in HbA1c and a 4.0 point difference in PAID score at 12 months post-randomisation between the randomised groups. 25 26 Since HbA1c and PAID were joint primary outcomes measuring different aspects of T2DM, both were tested at a 5% significance level.

Analysis
The analysis followed a pre-specified analysis plan, based on comparing the groups as randomised (intention-to-treat). The analysis plan was approved by the Trial Steering Committee before unblinding and uploaded to the trial website, https://www.ucl.ac.uk/pcph/research-groupsthemes/ehealth/projects/projects/helpdiabetesrct. Only HbA1c and PAID measured within a 10-14 month window period following randomisation was used in the primary analysis with missing 12-month outcomes multiply imputed using baseline and other outcome data (e.g. 3m data and final follow-up data collected outside the 10-14 month window). Further information on the imputation method is given in Appendix 2.
A linear mixed effects model with random centre effects was used to analyse each of the primary outcomes separately, adjusting for the baseline level of the outcome, age, gender, previous participation in other self-management programmes, pre-existing cardiovascular disease and time since diagnosis of diabetes. Secondary outcome measures were analysed similarly using generalised linear mixed models, with a normal residual error structure for continuous outcomes and a logit link for the binary outcome `Completion of 9 essential processes'. Pre-specified sub-group analysis for the co-primary outcomes was undertaken by baseline glycaemic control (HbA1c outcome only), baseline PAID (PAID outcome only), and duration of diabetes, treating all potential effect modifiers as continuous. The interaction between randomised group and each effect modifier was included in the model separately and assessed using a Wald test.
Use of the intervention was investigated as a mediator for efficacy, using instrumental variable methods, with randomisation as the instrument (Supplementary Figure 1). 27 28 Potential contamination was monitored by recording participants with similar family names and identifying those with the same addresses. Where this occurred, it was dealt with in the analysis by reporting the extent and undertaking a sensitivity analysis excluding these individuals.
A number of other sensitivity analyses were performed to assess the robustness of the primary analyses: 1) performing two complete case analyses disregarding outcomes measured outside 10-14 months and 11-13 months post-randomisation; 2) repeating the analysis using multiple imputation of baseline covariates only; 3) fitting linear models excluding centre random-effects; and 4) fitting an unadjusted model using only outcome measured in 10-14 months postrandomisation.
The TSC took on the role of the data monitoring committee. Trial registration ISRCTN02123133. Recruitment took place between September 2013 and December 2014. An initial 421 patients consented to participate, but of these 47 did not fully complete their baseline questionnaires and were therefore not randomised and did not enter the study. A total of 374 participants were randomised, of whom 86% (n = 321) provided data on PAID, and 78% (n = 291) had HbA1c measured within 10 to 14 months of randomisation. Additional final outcome data, obtained outside the 10 -14 month pre-defined window, were available for a further 27 participants for HbA1c and 16 participants for PAID ( Figure 1). Data obtained outside the 10 -14 month window were not used directly in the primary analysis, but were entered into the imputation model (Supplementary Table 1).

Baseline characteristics
Baseline demographic and clinical characteristics are shown in Table 1. The mean age was nearly 65 years, over two-thirds (n = 258, 69%) were male, and most were White British (n = 300, 80%). Nearly all (n=370, 99%) had a computer with access to the internet at home and just over half (n = 210, 56%) rated themselves as experienced computer users. Around onethird (n = 134; 36%) had been diagnosed for less than 5 years, with a further third (n = 115, 31%) having been diagnosed between 5 and 9 years ago. Overall, this was a population with well-controlled diabetes at baseline (mean HbA1c was 7.3% (56 mmol / mol)) and low levels of distress (mean PAID = 19).

Primary outcomes
At twelve months the primary analysis showed a significant difference in change in HbA1c between the randomised groups with participants in the HeLP-Diabetes group having a lower HbA1c than those in the control group (mean difference = -0.24%; 95% confidence intervals -0.44 to -0.049, p=0.014) (Table 2, Figure 2). There was no difference in change in PAID scores between the groups at 12-months (mean difference -1.5; 95% CI -3.9, 0.9, p=0.209), though both groups showed a decrease in PAID over the follow-up of the trial (Table 2, Figure 3).

Secondary outcomes
There was no difference in secondary outcomes at 12 months, with the possible exception of systolic blood pressure, which decreased more in the intervention group than in the control group (p=0.010) ( Table 2); though the result was not statistically significant after correction for multiple testing of secondary outcomes. There were no significant differences between groups on any of the outcome measures amongst individuals who completed three month outcomes (Supplementary Table 2). No adverse effects or events were recorded during follow-up.

Usage data
The mean number of log-ins was significantly higher in the intervention group than the control group (18.7 vs. 4.8, p= 0.0001), as was the mean number of pages visited per log-in (10.5 vs. 7.7, p <0.0001) and the mean number of days in which the website was accessed (10.1 vs. 3.3, p<0.0001) ( Table 3). The causal analyses estimated that for a "high-usage" population (those with usage ≥ the median of 4 days) the HeLP-Diabetes intervention could on average reduce HbA1c by -0.44% (95% CI -0.81 to -0.06) and PAID by -2.8 (95% CI -7.2 to 1.7) over 12 months (Supplementary Figures 2 and 3). The mean usage in the "highusage" group was 18 days. It should be noted that the usage data presented do not include the initial facilitation visit. There was a technical error in the software which led to usage data not being collected before 1 January 2014. At this point 16 participants had been randomised (7 to intervention, 9 to control). For these 16 participants, the usage data is not based on a full year, but for all other participants, data are summarised for the 12 months post-randomisation. The findings from the sensitivity analyses, including a complete-case analysis, were similar to the main analysis (Supplementary Table 3). Participants who were missing 12-month HbA1c had significantly higher mean baseline HbA1c measures (7.9% vs. 7.1%, p<0.001) leading to higher imputed HbA1c at 12-months in the non-completers and a greater mean difference between the randomised groups than from complete case analyses (Supplementary Figure 4, Supplementary Table 3).

Subgroup analyses
Pre-specified subgroup analyses showed that there was no evidence of baseline measures of HbA1c or PAID being effect modifiers for the mean difference between the groups. There was strong statistical evidence (interaction p=0.004) to suggest that the duration of diabetes acted as an effect modifier, with those who had been diagnosed more recently (<7 years) showing more of a reduction in PAID than those who had been diagnosed for longer periods of time. Duration of diabetes had no effect on change in HbA1c (Supplementary Table 4).

Harms
There were no reported harms in either group.

Discussion
In this first UK-based trial of a web-based self-management programme for people with T2DM, participants randomised to HeLP-Diabetes demonstrated improved glycaemic control at 12 months compared to those randomised to a simple information website. This improvement appears robust across all pre-specified sensitivity analyses, and was not dependent on duration of diabetes, baseline glycaemic levels or level of diabetes-related distress. Each 1% reduction in HbA1c is associated with a risk reduction of 21% for deaths related to diabetes and a 37% risk reduction for microvascular complications. 26 A reduction in HbA1c of 0.24% across a population level could translate into considerable population benefit, particularly as this web-based intervention could be delivered at low-cost and at scale across the UK. Moreover, in contrast to group-based education, where the effects appear to wane with time, 29 the effects of HeLP-Diabetes were greater at 12 months than at 3 months. There was no overall impact on diabetes-related distress, but some evidence that HeLP-diabetes appeared to reduce distress in recently diagnosed individuals. However, it is worth noting that baseline PAID scores were exceptionally low in this trial population. In a small pilot study, participants offered supported access to HeLP-Diabetes reduced their PAID scores by 6 points (p=0.04) over 6 weeks. 30 The trial has many strengths. It was a pragmatic trial, open to nearly all patients with T2DM in participating practices. Concealment of allocation was complete, as randomisation occurred after baseline data collection. Baseline prognostic factors were well balanced between groups. Every effort was made to achieve blinding, including requiring practices to have two nurses, so that data collection was undertaken by a nurse blind to participant allocation. Data for the co-primary outcomes at the primary outcome point were available for 78% and 86% of participants for HbA1c and PAID respectively. All analyses were on an intention-to-treat basis, supplemented by a CACE analysis. Although response rates for the co-primary outcomes were good, some potential for bias existed. Our primary analysis used multiple imputation methods because evidence shows that the assumptions underpinning this method are more defensible than those assumed using other approaches to missing data. 31 We also undertook sensitivity analyses including complete cases, non-contaminated cases, and a linear model excluding centre; all yielded similar results.
The two co-primary outcomes reflected the twin aims of the intervention: to improve diabetes control and to reduce diabetes-related distress. Around 40% of patients with diabetes have significant levels of distress, which severely impacts on quality of life, 32 and diabetes-related distress is an important outcome for patients. 33 Our patient and public involvement (PPI)  10 panel were clear that this should be a primary outcome, and a recent meta-ethnography emphasised the importance of empowerment and quality of life in promoting long term engagement with self-management. 34 In contrast, many health care professionals are more interested in glycaemic control. In line with previous trials in this area, 35 we decided to adopt both as co-primary outcomes and to test both at a 5% level of significance. 36 There are some limitations. Despite maximising the inclusivity of the trial by minimising the exclusion criteria, participants were not representative of the overall population of patients with type 2 diabetes in England. Compared to the overall population, participants had better control of their diabetes and cardiovascular risk factors, 6 7 and were much less distressed. 21 This finding mirrors that of a recent systematic review of demographic factors associated with web portal usage amongst people with diabetes which found that those with well controlled diabetes were more likely to use such portals than those with poor control. 37 However, fewer of our participants self-rated their computer skills as excellent (57% of our sample compared to a national average of 73%). 12 This good control at baseline has two implications -first, that there was little room for improvement in this population, and secondly, that this population may have been unusually motivated to self-manage their diabetes. Although every effort was made to maintain blinding, it is possible that some participants may have discussed their use of the intervention with research nurses, making it possible to infer which arm they had been allocated to. This could have affected research nurses' measurements of secondary clinical outcomes, such as blood pressure or weight, but could not have affected assessment of glycated haemoglobin as this was measured by laboratory staff who were blinded. There appeared to be high potential for contamination between two participants who shared the same surname and address, and a further two participants did not receive their allocated intervention due to an error at practice level; excluding these four made no difference to the results. A further limitation of the trial is that it provides little insight into the mechanism of action of HeLP-Diabetes. This was the result of a deliberate decision to focus on clinically important outcomes and minimise both the response burden and the potential impact of measurement on participants.
This is the first UK-based trial of a web-based self-management programme for people with type 2 diabetes, and internationally, the first trial of such a comprehensive intervention that aims to address the three main tasks of self-management: emotional, medical and role management. 19 In the Cochrane review of computer-based self-management interventions for people with T2DM, only 4 of the included studies had follow-up of 12 months or more. 14 Of these, three interventions were clinic-based, with participants completing self-assessment tools on a touch screen and receiving tailored advice during their baseline visit to their diabetes clinician [38][39][40] and one was a mobile phone-based intervention which provided tailored messages in response to participant's results of blood glucose self-monitoring data. 41 A more recent systematic review of internet delivered diabetes self-management identified 2 trials with 12 or more months follow-up. 42 One trial was on a structured intervention based on a peer-led, group-based, diabetes self-management course 43 . There were six sessions, with each session available for one week. Each session required participants to make a specific action plan to address a problem they were experiencing. Peer facilitators encouraged use of the programme. Follow up was planned at 6 and 12 months; however HbA1c data were only available at 6 months. The other trial compared two versions of a web-based intervention (with and without additional social support) to enhanced usual care. The web-based intervention was designed using social-cognitive theory and a social-ecological model, with a focus on three main behaviours: dietary intake, physical activity and medication adherence. Users of either web-based intervention received motivational phone calls to encourage adherence and development of action plans. Those randomised to the enhanced intervention (with additional social support) received two additional phone calls and an invitation to attend a group session. There was no difference between groups in HbA1c or other biological outcomes at 12 months. 44 Thus the results of this trial add significantly to the available literature.  On the basis of these results, HeLP-Diabetes may be considered as an addition to the current menu of self-management support for people with type 2 diabetes, and may help increase overall access and uptake. Most commissioned services currently focus on newly diagnosed patients, leaving clear unmet need for people who have had their diabetes for longer, but are looking for ways to improve their health. Many patients are not ready to engage in self-management early in their illness journey, 9 but become motivated to do so later, often as a result of a change in medication or development of a complication. 45 The intervention is low cost, and as most costs are fixed, irrespective of number of users, is likely to be cost-effective, particularly if widely used. A cost-effectiveness analysis of HeLP-Diabetes will be reported separately.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  We gratefully acknowledge the permission to use under license the validated behaviour change modules for weight loss (POWeR), alcohol reduction (DownYour Drink) and smoking cessation (StopAdvisor and the diabetes module from Healthtalk online (http://www.healthtalk.org/) in HeLP-Diabetes.
We are grateful to Orla O'Donnell for outstanding project management, Fiona Giles for administrative support, all our PPI who contributed to the development, maintenance and delivery of the intervention and / or the management and oversight of the trial, staff at the participating practices, PCRN staff, and all our participants.

Item
Line numbers Provide the name or a phrase that describes the intervention 2,3 Describe any rationale, theory or goal of the elements essential to the intervention

-48
Materials: describe any physical or informational materials used in the intervention, including those provided to participants or used in intervention delivery or in training of the intervention providers 51-90 Procedures: Describe each of the procedures, activities and /or processes used in the intervention, including any enabling or support activities

-137
For each category of intervention provider, describe their expertise, background, and any specific training given.

109
Describe the modes of delivery (such as face-to-face or by some other mechanism, such as internet or telephone) of the intervention, and whether it was provided individually or in a group 2, 123 -137 Describe the type(s) of location(s) where the intervention occurred, including any necessary infrastructure or relevant features 116 -120 Describe the number of times the intervention was delivered and over what period of time including the number of sessions, their schedule and their duration, intensity or dose.

-106
If the intervention was planned to be personalised, titrated or adapted, then describe what, why, when and how.

-106
If the intervention was modified during the course of the study, describe the changes (what, why, when and how)

-144
Planned: if intervention adherence or fidelity was assessed, describe how and by whom, and if any strategies were used to maintain or improve fidelity, describe them.

-149
Actual: if intervention adherence or fidelity was assessed, describe the extent to which the intervention was delivered as planned.

Statistical Methods for Multiple Imputation
Multiple Imputation using chained equation was used as the primary method to account for missing data (in both baseline and follow-up data). 1 A set of imputation models were specified, one for each variable with missing data. Each variable was then regressed on all other variables, including completely recorded baseline and follow-up variables and stratified by randomised group. Imputations were performed using predictive mean matching using the five nearest neighbours to the prediction as a set to draw from. The full list of variables considered in the MICE approach is shown in Supplementary Table 1, together with the number of missing values for each variable and time period.
Since only measurements within a 10-14 month window period were used within the main analyses of HbA1c and PAID, the following imputation procedure was implemented for these two co-primary outcomes. Twelve month measurements were subdivided into those that were measured within 10-14 months (the primary outcome variable) and those that were measured outside 10-14 months (a variable used for imputing only). For HbA1c, two additional variables were created for use within the imputation model; 1) the time in days from randomisation that the "12-month" HbA1c measurement was actually taken for values inside of the window period (and set to 365 for measurements taken outside the window period), 2) the time in days from randomisation that the "12-month" HbA1c measurement was actually taken for values outside the window period (and set to missing for measurements taken within the window period). The first of these variables gives the desired time for imputing HbA1c measurements when they are missing, whilst the second gives information on how far outside of the window the actual measurements were taken. Corresponding variables were created for the "12-month" PAID measurement. Finally, two additional variables were created defining the time in days at which HbA1c and PAID were measured at "3-months". All variables were included within the chained equations and imputed where necessary. 40 imputed datasets were created, the analysis models were fitted to each imputed dataset separately, and the estimates were pooled using Rubin's rules.
Statistical Methods for causal analyses The causal analysis proposed attempts to address how the effectiveness of the intervention is mediated through the frequency of website usage. In particular, it is important to understand whether prolonged usage of the website modifies the efficacy of the intervention. Since website usage is measured post-randomisation a naïve analysis of correlating usage with outcomes in the intervention group may give biased and misleading results, since there may be unmeasured confounders also correlated with the outcomes that distinguish the motivated users who regularly log-in from the less motivated ones. Causal analyses using instrumental variables (IV) were therefore used to determine the effect of website usage on outcomes. This approach preserves randomisation (i.e. provides a comparison independent of observed and unobserved confounders).
"Usage" is defined as the proportion of follow-up (rescaled as no. days in a year) that the HeLP-Diabetes website is accessed. It was assumed that the efficacy of the intervention is zero for individuals who never log-in (the exclusion restriction assumption). Website usage in the control group was ignored in the model as it was assumed that the control website was unlikely to be effective. (NB. usage statistics were collected for the comparator website but were not used in these analyses).  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  This main underlying assumption of the causal analysis is that the effect of randomisation to the HeLP-Diabetes intervention on 12-month outcomes occurs only through use of the website (Supplementary Figure 1; Z is randomised intervention, WU is website-usage and Y is 12-month outcome, e.g. HbA1c or PAID). This relies of the "exclusion restriction" assumption that the HeLP Diabetes intervention has no effect when usage is zero (i.e. for individuals who never log-in). Hence randomisation is assumed to be an instrumental variable. Figures 4 and 5 show the estimated causal effects of HeLP Diabetes on HbA1c and PAID at 12-months, respectively, by level of website usage (the "exclusion restriction" assumption leads to zero efficacy at zero usage). To determine the casual estimate of efficacy for a "high-usage" group, the individual predicted efficacy was calculated for each individual greater than or equal to the median usage of 4 days (shown as the red boxes in the histograms in Figures 4 and 5) and the average efficacy for this group was obtained. The mean usage in the "high-usage" group was 18 days.

Z WU Y
Scientific background and explanation of rationale 2b Specific objectives or hypotheses

Methods
Trial design 3a Description of trial design (such as parallel, factorial) including allocation ratio 3b Important changes to methods after trial commencement (such as eligibility criteria), with reasons Participants 4a Eligibility criteria for participants 4b Settings and locations where the data were collected Interventions 5 The interventions for each group with sufficient details to allow replication, including how and when they were actually administered Outcomes 6a Completely defined pre-specified primary and secondary outcome measures, including how and when they were assessed 6b Any changes to trial outcomes after the trial commenced, with reasons Sample size 7a How sample size was determined 7b When applicable, explanation of any interim analyses and stopping guidelines Randomisation: Sequence generation 8a Method used to generate the random allocation sequence 8b Type of randomisation; details of any restriction (such as blocking and block size) Allocation concealment mechanism 9 Mechanism used to implement the random allocation sequence (such as sequentially numbered containers), describing any steps taken to conceal the sequence until interventions were assigned Implementation 10 Who generated the random allocation sequence, who enrolled participants, and who assigned participants to interventions Blinding 11a If done, who was blinded after assignment to interventions (for example, participants, care providers, those  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48 assessing outcomes) and how 11b If relevant, description of the similarity of interventions Statistical methods 12a Statistical methods used to compare groups for primary and secondary outcomes 12b Methods for additional analyses, such as subgroup analyses and adjusted analyses

Results
Participant flow (a diagram is strongly recommended) 13a For each group, the numbers of participants who were randomly assigned, received intended treatment, and were analysed for the primary outcome 13b For each group, losses and exclusions after randomisation, together with reasons Recruitment 14a Dates defining the periods of recruitment and follow-up 14b Why the trial ended or was stopped Baseline data 15 A table showing baseline demographic and clinical characteristics for each group  Numbers analysed  16 For each group, number of participants (denominator) included in each analysis and whether the analysis was by original assigned groups Outcomes and estimation 17a For each primary and secondary outcome, results for each group, and the estimated effect size and its precision (such as 95% confidence interval) 17b For binary outcomes, presentation of both absolute and relative effect sizes is recommended Ancillary analyses 18 Results of any other analyses performed, including subgroup analyses and adjusted analyses, distinguishing pre-specified from exploratory Harms 19 All important harms or unintended effects in each group (for specific guidance see CONSORT for harms) Sources of funding and other support (such as supply of drugs), role of funders *We strongly recommend reading this statement in conjunction with the CONSORT 2010 Explanation and Elaboration for important clarifications on all the items. If relevant, we also recommend reading CONSORT extensions for cluster randomised trials, non-inferiority and equivalence trials, non-pharmacological treatments, herbal interventions, and pragmatic trials.