Study Design and Randomisation
Trial Design and Registration.
A pragmatic, two-arm RCT design was conducted from October 2014 to April 2015. The study was retrospectively registered with Australian New Zealand Clinical Trials Registry, six weeks after recruitment commenced and before randomisation (ACTRN12614001110673, UTN No.: U1111-1161-9803).
Sample Size Calculation and Change to Trial Design after Trial Commencement.
Informed by previous studies (9, 10, 21-23), we anticipated 10% recruitment, 50% baseline screening rate and 40% study attrition. We initially selected a Solomon 4-group design (24) to account for anticipated Question-Behaviour-Effect (QBE), where answering questions about a specific behaviour can influence an individual’s related cognitions, emotions and behaviour (25). Thus, our initial sample size calculation required 200 unengaged participants (50 per condition in the 4-group design). Calculation used input parameters were: effect size of 0.3 (26), 80% power, significance level of 0.05, two-tailed.
However, lower than expected recruitment of unengaged participants in the first few weeks foreshadowed risk of lack of power to detect change in the primary outcome. Consequently, changes were made to trial design (see Supplementary file 2 for more detail) where: prior to randomisation, the design was modified to a conventional two-arm (intervention/control) RCT, and alternate methods were used to minimise the potential impact of QBE (see Concealment of Study Purpose below). Using the revised design and existing input parameters, 25 unengaged participants in each arm were required to sufficiently power the study to determine effect on the primary outcome.
Young adults with T2D (18-39 years) registered with Australia’s National Diabetes Services Scheme (NDSS) were eligible to participate, with registration date used as a proxy for diabetes diagnosis. With approximately 90% of Australians with T2D registered, the NDSS is considered the “best available source to monitor type 2 diabetes in children and young people in Australia” (27, p.36). Exclusion criteria were non-proficiency in English and other diabetes types.
Of the approximately 32,000 young adults with T2D registered on the NDSS, 5,354 had consented to be contacted for research purposes; all were invited to participate. To protect confidentiality, NDSS staff coordinated study recruitment, including an introductory letter (on NDSS/Diabetes Australia letterhead) and study invitation. Two incentives were offered: a chance to win one of three iPad minis at registration/Stage 1 data collection, AUD$20 upon study completion. A reminder invitation was posted four weeks later and recruitment continued until online study enrolment waned.
Concealment of Study Purpose
The purpose of the study was initially concealed to mitigate risk of young adults with T2D who had not engaged in screening declining to participate in a study focused on the behaviour. Consequently, the study invitation advertised the opportunity to participate in a study about ‘diabetes self-management’ with the question about screening status embedded within a suite of items exploring diabetes self-management activities.
Data collection was managed via Qualtrics secure online survey platform (Qualtrics, Provo, UT). Baseline data was collected in two stages to allow identification of participants who had already engaged in screening. Participants were stratified based on engagement with screening and then randomly allocated to ‘leaflet’ intervention or ‘no leaflet’ control. Randomisation sequence was generated by the project manager (AJL) via an online random number generator using a 1:1 ratio (28).
Following a two-month development and piloting process (January to February 2015), the leaflet was posted to all intervention participants in March 2015. Four weeks later, all participants were emailed an invitation to complete a follow-up survey, with survey logic programmed to ensure that previously unengaged participants were asked whether they had engaged in DR screening “since completing the last survey”. The survey also contained all Stage 2 social cognitive items and a fidelity question which asked intervention group participants whether they received the leaflet, and if so, whether they had read it.
Upon completion of the post-intervention survey, all study participants were provided with a transparency statement which explained: the reason for concealment of the true study purpose, why screening is important for all people with diabetes and a link to more information. Control group participants who had been exposed to real-world ‘usual care’ were invited to receive the Who is looking after your eyes? leaflet upon provision of their postal address.
Development of the 8-panel, Who is looking after your eyes? leaflet (Figures 1a and b) is described elsewhere (18).
Our survey was reviewed and pilot-tested by stakeholders including young adults with T2D. Baseline data were collected in two stages: i) demographics and clinical characteristics, and ii) social cognitive determinants. At 4-weeks post-intervention, items assessing engagement with diabetes-related health checks (including screening), emotional well-being and all social cognitive determinant items were repeated.
Stage 1 Demographic data.
Participants provided written consent and demographic data, including gender, age, country of birth, language spoken at home, marital status, level of education, employment status, primary diabetes management, family history of T2D and number of health comorbidities.
Engagement with diabetes-related health checks: was assessed via six separate questions, worded as follows: ‘Since you were diagnosed with diabetes, have you had your [cholesterol, blood pressure, average blood glucose (HbA1c), kidney function, eye health, feet] checked?’. A standard definition was provided for each with the aim of minimising reported confusion between standard vision check and screening for DR (29). Responses to the eye health check component of this question were used to identify: unengaged participants to determine uptake of screening 4-weeks post-intervention (primary outcome).
Depressive symptoms: the Patient Health Questionnaire-2 (PHQ-2, 30), a brief, validated depression screening tool, was included, to identify potential harms arising from the intervention. Responses to PHQ-2 items were summed to produce a total score (range: 0-6), with higher scores indicating more depressive symptoms.
Stage 2 Social cognitive determinants.
We have previously reported the theory-based development of the 54-item survey used in this study (18); all items are listed in Supplementary file 1. In brief, 16 items assessed three knowledge constructs (link between diabetes and vision loss, knowledge of DR, and knowledge of screening). Responses were scored dichotomously (correct / incorrect). Knowledge items were aggregated to form a composite score with higher scores indicating greater knowledge.
Twenty-one items assessed three attitude constructs: i) attitudes to screening, ii) perception of personal risk and iii) anticipated regret at not screening. Three items assessed normative beliefs (such as approval of others and beliefs about the behaviour of similar others) and a further three items assessed intention to screen for DR. For attitudes, normative belief and intention items, responses were scored on either a 5- or 7-point Likert scale, with higher scores representing stronger agreement (items reverse scored where necessary).
Finally, eleven items assessed two behavioural skills constructs: i) perceived control (e.g. ability to seek and attend screening) and ii) overcoming barriers (e.g. ability to identify and address common environmental and psychosocial barriers). Responses were scored on a 5-point Likert scale with higher scores representing greater confidence.
Data were analysed using the Statistical Package for Social Sciences (SPSS, IBM Corp, Armonk, NY; Ver.23, 2015). To assess factors associated with loss to follow-up, chi-square and independent t-tests (two-tailed) were used to compare baseline demographic characteristics and scores on modifiable behavioural determinants between those who completed and did not complete the study.
Primary outcome: we planned to perform inferential statistical analyses to determine the effect of the intervention on uptake of DR screening. However, insufficient ‘unengaged’ participants provided post-intervention data. As such, the study was underpowered to determine effect of the leaflet on the primary outcome.
Secondary outcome: to assess intervention effects on secondary outcomes, we: i) created change scores by subtracting the baseline composite scores from those at follow-up, ii) conducted independent samples t-tests (two-tailed) on the change scores to assess between-group differences and also conducted paired-samples t-tests to assess within-group changes over time, and iii) calculated effect sizes to determine the relevance of the finding.
Although intention-to-treat and per-protocol analyses were planned, high attrition precluded reliable analysis. Consequently, we elected to exclude cases with missing secondary outcome data pairwise, restricting results to complete cases only for each individual behavioural determinant composite score. Data are presented as means±standard deviation (SD), median (interquartile range, IQR) or n(%). Statistical significance was defined as p<0.05. Effect sizes are described with partial eta squared (h2, range: 0-1); guidelines for interpretation are: h2=0.01 (small), h2=0.06 (moderate), and h2=0.14 (large effect), (31).
 The NDSS is an initiative of the Australian Government administered with the assistance of Diabetes Australia. It supports people with diabetes to self-manage their condition by providing subsidised access to diabetes-related products and services.