NEW UPDATE: Completed trials may now upload and register supplementary documents (e.g. null results reports, populated pre-analysis plans, or post-trial results reports) in the Post Trial section under Reports, Papers, & Other Materials.
The effect of technology-assisted behavioral interventions in type 2 diabetes
Last registered on July 01, 2020


Trial Information
General Information
The effect of technology-assisted behavioral interventions in type 2 diabetes
Initial registration date
May 27, 2020
Last updated
July 01, 2020 10:35 AM EDT

This section is unavailable to the public. Use the button below to request access to this information.

Request Information
Primary Investigator
University of Cape Town
Other Primary Investigator(s)
PI Affiliation
University of Cape Town
PI Affiliation
University of Cologne
PI Affiliation
Max Planck Institute for Research on Collective Goods
PI Affiliation
Max Planck Institute for Research on Collective Goods
Additional Trial Information
In development
Start date
End date
Secondary IDs
The burden of diabetes is a growing global problem, not only for patients and families, but also for health insurance providers and the wider economy. Much of this is driven by lifestyle, such as what we eat and drink, smoking and how little we exercise. Health-related behaviour is difficult to shift. Measuring and tracking behaviour in the field is often a challenge. Wearable health-monitoring technology may offer innovative solutions for lifestyle modification, as well as the study of it. We want to understand how we can support diabetes patients form sustainable healthy-eating habits and increase glucose control. Our study is incorporating the behavioral changes that the COVID-19 pandemic may influence in this high risk population. It will be critical to understand the impact of the pandemic on this at risk population and to learn how we could better support the diabetic community to adapt to these changes. Better disease management will reflect a lower risk of co-morbidities such as cardiovascular disease and severe COVID-19 illness. Overall, this research is highly relevant in the context of the global chronic disease burden, the COVID-19 pandemic, and preventing diabetes-related complications in particular.

External Link(s)
Registration Citation
Larmuth, Kate et al. 2020. "The effect of technology-assisted behavioral interventions in type 2 diabetes." AEA RCT Registry. July 01. https://doi.org/10.1257/rct.5871-1.1.
Sponsors & Partners
Experimental Details
This RCT will measure behavioral and health outcomes. The study employs a staircase design that will include three different treatment conditions. The three groups will vary in the amount of personalised feedback and healthcare practitioner contact they get. We aim to find out which treatment is the most helpful to diabetes patients to reach their diet and glucose control goals. All patients will receive an intervention which includes nutrition education, medical advice, and regular check ups with a study physician to monitor progress. We will also use health assessments, questionnaires, and mobile apps.
Intervention Start Date
Intervention End Date
Primary Outcomes
Primary Outcomes (end points)
1. Percentage Time in Target Glucose Range (3.9 - 8 mmol/L)
2. Haemoglobin A1c
Primary Outcomes (explanation)
1. To measure Percentage Time in Target Range we will use Abbott's Continuous Glucose Monitoring technology. We will use two different sensors: The Libre Pro and the FreeStyle Libre (with LibreLink app or scanner). The former provides no feedback, while latter provides real-time feedback to the wearer. To insert, the patient (or in our case, the nurse) lines up a disposable plastic applicator device with a spring-loaded needle on the back of the upper arm and pushes down. The spring-loaded needle introduces the sensor under the skin and retracts. The sensor is connected to a small round external disc and held in place with a water resistant adhesive, where it remains for 14 days. The technology allows for readings of real-time glucose values with the scanner or the LibreLink smartphone app, as well as visual trends and variability (e.g. Time in Target). The study facilitator will assist the patient to log in to the LibreLink app. The FreeStyle Libre sensor is factory calibrated. In order for glucose data to be stored, the wearer must flash over the device with their scanner or smartphone at least once every 8 hours. The LibreLink app is compatible with NFC-enabled smartphones running Android OS 5.0 or higher and with iPhone 7 and higher running OS 11 and higher. There is also a scanner available as an alternative to using the mobile app.

2. Haemoglobin A1c (HbA1c) is recognised as the key surrogate marker for the development of long-term diabetes complications and has been used as the primary endpoint for many CGM studies. While HbA1c reflects average glucose over the past 2-3 months, its limitation is the lack of information about acute glycaemic excursions and the acute complications of hypo- and hyperglycaemia (colloquially, “glucose spikes”). HbA1c fails to show the magnitude and frequency of intra- and interday glucose variation. The utility of HbA1c measures are enhanced when used as a complement to CGM glycaemic data. A blood sample will be taken to test HbA1c at baseline and endline. This test is standard in diabetes diagnosis and management. It would be ordered regularly, even in the absence of our study.
Secondary Outcomes
Secondary Outcomes (end points)
1. Diet adherence
2. Lipid panel
3. Body composition
4. Blood pressure
5. Subjective wellbeing
6. Diabetes distress score
Secondary Outcomes (explanation)
1. Two dietary assessment tools will be used to characterize subjects’ diets. A Food Frequency Questionnaire (FFQ) and a Medicross paper food diary for diabetes patients.
First, at Visit 1, participants will complete a self-administered FFQ (Appendix C) asking about foods eaten in the past month. The instrument has been used in previous studies in South Africa and can also be used online (Webster, Murphy, Larmuth, Noakes and Smith, 2018). The FFQ was developed according to guidelines proposed by Cade, Thompson, Burley, & Warm (2002) by adapting the South African Medical Research Council’s (MRC) FFQ (Steyn & Senekal, 2004) to include food items frequently eaten by people following a low carbohydrate diet. It also includes standard portion sizes (to increase accuracy of macronutrient ratio estimation), and frequency options (necessary for an online questionnaire). The FFQ data will be used to assess the types of foods eaten and macronutrient ratios rather than absolute nutrient quantities. Macronutrient ratios will be determined from FFQ data by multiplying the frequency of food items eaten per month, by their respective assigned macronutrient contents. Macronutrient content will be derived from the FatSecret database and MRC’s SAFOODS database, because MRC’s SAFOODS database omits many low carbohydrate food choices. The FFQ will be re-administered (Visit 5) to assess the sustainability of healthy-eating habits formed during the intervention.
Second, participants will be asked to keep a food diary for four days during the beginning of the baseline and four days towards the end of the CGM intervention. The four days will include weekends and weekdays. We are interested in quality over quantity of data so we do not insist on daily entries. However, patients may enter the diary daily if they wish. Besides the common limitations of self-report measures (e.g. deception, guessing, non-compliance), there is a risk of missing data when asking individuals to enter their food intake since they could forget or become fatigued. Omissions in the food diary will be followed up on. We will also learn from compliance rates in entering data about the usefulness of such diaries as a measure of self-reported dietary intake. For the treatment groups, it will be interesting to correlate compliance behaviours such as number of food diary days completed and number of scans per day using Freestyle Libre CGM (engagement in self-monitoring). Qualitative comparisons will be discussed between temporal glucose patterns and reported dietary intake. Our analysis of participants’ food choices will be explorative.

2. Lipid panel will be measured at baseline and endline (blood test).

3-4. Body composition (including waist and hip circumference, weight, height) and blood pressure will be taken at baseline and regular checkups with study physician.

5-6. The participant will complete a self-administered baseline questionnaire to assess demographics (including income bracket and education), tobacco and alcohol consumption, physical activity, diabetes, cardiovascular disease and blood pressure history, risk and time preferences, math ability, subjective wellbeing, and diabetes distress. Our baseline questionnaire includes sections from validated questionnaires. Prevalence and severity of diabetes is strongly associated with socioeconomic and cultural factors, such as income, education, gender, ethnicity and culture. Subjective wellbeing and diabetes distress will be measured at baseline and endline.
Subjective wellbeing will be measured using a single standard item and scored on a scale of 1 to 10 where 1 means “very dissatisfied” and 10 means “very satisfied”: “How do you feel about your life as a whole right now?”. The wording of this standard measure of subjective wellbeing is sourced from the South African National Income Dynamics Study (NIDS). This measure may be affected by social distancing and economic hardship due to SARS-CoV2 restrictions, which will be recorded.
We will use the Diabetes Distress Scale (DDS) for adults with type 2 diabetes. The DDS is a 17-item self-report instrument. Each item is rated on a 6-point scale from (1) “not a problem” to (6) “a very significant problem”. The scale yields an overall distress score based on the average responses on the 1-6 scale for all 17 items. It can also be used to get a measure of emotional burden (average of 5 items), physician distress (average of 4 items), regimen distress (average of 5 items) and interpersonal distress (average of 3 items). Average score < 2.0 reflects little or no distress. Average score 2.0 – 2.9 reflects moderate distress and > 3.0 reflects high distress. A total or subscale score > 2.0 is considered clinically significant.
Experimental Design
Experimental Design
This RCT will measure behavior change and health outcomes. We will explore which type of feedback and support is most useful to patients with type 2 diabetes to fine tune their diet and help them reach their metabolic health goals. We will use a rolling recruitment strategy and sequential assignment to treatments. Eligible adult type 2 diabetes patients will be referred to the study physicians by the network of physicians at a medical centre in South Africa. The investigators will make use of remote monitoring technology as far as possible to limit the need for person-to-person contact. We will explore the influence of the pandemic on patients' ability to manage their diabetes.
Experimental Design Details
Not available
Randomization Method
We will randomize sequentially, as patients are enrolled on a rolling basis at the medical practice, using a computer and Stata statistical software.
Randomization Unit
Was the treatment clustered?
Experiment Characteristics
Sample size: planned number of clusters
90 individuals and three treatment arms.
Sample size: planned number of observations
90 individuals.
Sample size (or number of clusters) by treatment arms
30 individuals CONTROL (physician standard care), 30 individuals INFORMATION (physician standard care plus CGM real-time feedback), 30 individuals COACHING (physician standard care plus CGM real-time feedback and health coaching online).
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
Individuals will be sequentially randomly assigned and matched for sex and carbohydrate intake to one of the three groups, during phase 1, until a sample size of 30 participants per group is achieved. We have repeated observations for all participants during baseline, intervention and sustainability phases. The number of participants was chosen based on power calculations (default significance level alpha 0.05, power 0.80) using the free software G*Power for statistical power analysis (Faul, F., Erdfelder, E., Lang, A. et al., 2007). We also took budget and logistical factors into consideration. We aim to be able to detect a clinically significant change in our primary behaviour of interest (5% Time in Target (3.9 - 8 mmol/L)) and in our primary health outcome (HbA1c of 0.5%). We aim to do both a within-subject (before and after) and between subject experimental analysis. However, it is likely that the sample size will limit us to a within-subject analysis and difference-in-difference analysis. Difference-in-difference analysis will allow for the comparison of the treatment against the control group (Wing, Simon and Bello-Gomez, 2018). This is a quasi-experimental research design that will allow us to study causal relationships if a classical randomized control trial design is not feasible. A within-subject analysis considers a t-test using measurements before and after treatment. We set G*Power to calculate the difference between two sample means (matched pairs). An a priori type of power analysis computes the required sample size, given significance level alpha, power, and effect size. We use the default significance level alpha 0.05 and power 0.80. We consider a one-tailed t-test because the hypothesis we have is that compared to baseline the outcome improves with the behavioural intervention. The null hypothesis is that there will be no difference from baseline. We use the Wilcoxon signed rank test which relaxes the assumption of t-tests that the data are normally distributed (parent distribution min ARE). For HbA1c, we consider the sample size necessary to detect a difference in mean HbA1c from 7.3% to 6.3% (i.e. a clinically significant difference of 1.0 as observed in a previous health coaching study McKenzie et al., 2017) with standard deviation (SD) of 3.5 before treatment and standard deviation of 2.0 after (standard deviations based on clinical experience of the study physicians). Since both measures are from the same individual the correlation between them is high (0.95). Using these values, the standardized effect size (dz) is 0.58. G*Power output states that a total sample size of 23 participants is necessary to detect this effect size. For Time in Target range, we consider the required sample size necessary to detect a small improvement from 50% to 55% with SD of 20 before treatment and SD of 15 after. This produces an effect size (dz) of 0.67. G*Power states that we require a sample size of 18 participants to detect a small clinically significant improvement in Time in Target range. Since we will recruit 30 participants within each group, the study is adequately and conservatively powered for within-subjects analysis of treatment effect for both primary outcome measures HbA1c and Time in Target range. The between-subjects a priori power analysis in G*Power considers a Wilcoxon-Mann-Whitney test (two groups) to test for a treatment effect i.e. treatment group versus control group. For HbA1c, if we consider the case of the control group with mean 7.0 and SD 2.0 and the treatment group with mean 6.0 and SD 2.0, this is an effect size (d) of 0.5 and requires a sample size of 101 subjects in each of the two groups. We will have only 30 subjects in each group and thus will not be able to detect a small to moderate effect size of treatment on HbA1c with these assumptions. Turning to a post hoc power analysis we consider what effect size we could actually achieve with 30 subjects in each group. G*Power states that we can detect an effect of 0.71 with our proposed design. For HbA1c we would be able to detect a difference of about 1.5 pp (e.g. control 7.5% (SD 2) versus treatment group 6.0% (SD 2)), which would be somewhat unlikely to observe in the data. However, SD greatly affects these estimates of power. Balancing the groups on sex and baseline carb intake will help to reduce the SD. The possibility of between-subjects analysis of HbA1c will only be known after we have collected the data. We continue the hypothetical post hoc power analysis for Time in Target range in G*Power. If we consider the case of the control group with mean 60% (SD 15) and treatment group with mean 70% (SD 15), this is an effect size of 0.67 and power of 0.76 (just below default threshold of adequate statistical power). However, if control SD is 14 and treatment group SD is 14, this implies an effect of 0.71 and adequate power of 0.81. Thus, with smaller SDs the study would in this case be powered to detect a clinical difference of 10% between treatment and control group in our primary behavioural outcome. Overall, we expect the study will have adequate statistical power for a within-subjects analysis, and potentially between-subjects analysis for Time in Target.
Supporting Documents and Materials

There are documents in this trial unavailable to the public. Use the button below to request access to this information.

Request Information
IRB Name
IRB Approval Date
IRB Approval Number