This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR Formative Research, is properly cited. The complete bibliographic information, a link to the original publication on https://formative.jmir.org, as well as this copyright and license information must be included.
Type 2 diabetes (T2D) is increasingly prevalent in society, in part because of behavioral issues, with sedentary behavior, reduced exercise, and the consumption of foods with a high glycemic index being major contributors. There is evidence for the efficacy of mobile apps in promoting behavior change and lifestyle improvements in people with T2D. Many mobile phone apps help to monitor the condition of people with T2D and inform them about their health. Some of these digital interventions involve patients using apps on their own or in conjunction with health care professionals.
This study aimed to test the acceptability of receiving app-based, daily physician feedback for patients with T2D that is informed by the continuous monitoring of their activity, food choices, and glucose profiles, with the aim of encouraging healthier behavior. The
A total of 15 patients diagnosed with T2D wore a glucose monitor and an Apple Watch for 12 days. The uploaded data were integrated into the
Over the 12 days of the study, there was a significant reduction of 0.22% (
A mobile app system that provides people with T2D daily, physician-generated, personalized feedback can produce favorable changes in glycemic and cardiovascular risk parameters—even in the short term—and encourage better self-management of their condition. Study participants found the experience of using the mobile app system acceptable and were motivated to establish longer-term lifestyle improvements through behavior changes.
Type 2 diabetes (T2D) is a widespread chronic health condition that is increasingly prevalent in society, in part because of people’s behavior. A lack of physical exercise and the consumption of foods with a high glycemic index are at the core of the current epidemic of obesity and increased risk of diabetes and consequent cardiovascular disease [
The core activities in chronic disease self-management are medical management (medication and dietary advice adherence), management of necessary behavior changes, and managing emotions and feelings around coping with chronic diseases. This aspect of T2D treatment is relatively underdeveloped worldwide [
Several mobile health apps have been developed to help people with T2D self-monitor their condition and provide them with diabetes education and information. The increased use of health-related apps is partly because of their convenience, portability, and reach [
Mobile phone interventions for diabetes self-management have been found to be a useful support in promoting health-related behavior changes. People tend to keep their phones with them constantly—even at night—thus providing an inexpensive, real-time delivery mechanism for health and behavioral support messaging [
More than 10 systematic literature reviews of studies on the use of mobile apps for self-management of diabetes have been published in the past 5 years [
Core features in diabetes self-management apps vary widely among different apps [
Previous studies on diabetes mobile app interventions have explored different types of feedback messaging based on several behavior change theories. These include targeting messages based on the patient’s disease stage within the transtheoretical model of behavior change [
Examples of feedback provided by physicians after reviewing the previous day’s data.
Participant number | Date | Physician feedback |
2001 | February 22, 2019 | “Main issue is the high sugars after lunch and dinner, White flour-based bread and pizza base are causing problems, consider multigrain bread and a healthier choice for dinner. Good steps but no recorded activity. Try for 20 minutes extra exercise of moderate level per day.” |
2001 | February 23, 2019 | “Good morning activity and low carbohydrate breakfast kept things nicely controlled thorough the morning. Low activity after lunch and multiple carbohydrate snacks in early afternoon kept blood sugar high in the afternoon. Try some low-GI snacks e.g. cheese on Vita Wheats or fruit (banana, berries etc.).” |
2001 | February 24, 2019 | “Excellent morning after low carb breakfast. The coatings of schnitzels are a trap as they contain rapidly absorbed carbohydrate. Steamed chicken breast may have been a better choice. Well done for the extra exercise on the bike. Although exercise can acutely put the blood sugar up, the overall effect will be positive.” |
The reviewed studies suggested that the design and development of diabetes self-management apps must be informed by an understanding of the needs and desires of the people who will use them and must incorporate the features and support mechanisms that patients value [
This clinical study at the Eastern Health Clinical School (Box Hill Hospital, Melbourne, Australia) aims to evaluate the satisfaction of patients with T2D with wearable technology alongside using a diabetes management app and examine the potential effectiveness of physicians’ real-time feedback in promoting behavior change around participants’ diet, activity, and health choices. The 12-day
Over a 5-month period in 2019, we aimed to test the hypothesis that wearable devices with real-time feedback from a physician might motivate behavior change in participants with T2D. The study used wearable sensor technology to track glucose profiles, medication, insulin dose, food and drink intake (through self-reported photographs of every meal), and activity levels of participants with T2D. Participants were provided with daily personalized SMS text message advice from a physician who had access to all the study participants’ collected data. A physician rather than a dietician reviewed the data and offered recommendations via the app as we monitored medication use and exercise, as well as diet.
We analyzed and identified the preferred features of a mobile app diabetes management system, as discussed in prior study. The analysis informed the design of the mobile app
The system combines user input and sensor data to track patients’ behavior and food intake data and medication and insulin use and record patients’ daily activities. Other personal data were tracked using the mobile device’s built-in sensors; these included insulin use, number of steps, heart rate, and glucose levels. The uploaded data were integrated into the smartphone app, which also enabled participants to record their activities and food intake. The study physician received a report on these integrated data, which allowed the creation of personalized feedback for each study participant.
Schematic of the
The
Communication-driven design process applied in the
A technology package was prototyped for the study, assembling 3 applications to gather data and provide 1-way communication between the physician and participants (
The iOS app (
Technology used in the
Examples of
All data in the app were mirrored on a webpage that was accessed by the physician, who reviewed the previous day’s data each morning and provided text-based feedback, which would appear both in the app and as a notification. There were no opportunities for 2-way communication with the physician. Feedback was limited to 2 to 3 sentences and concentrated on a few aspects of health-related behavior, with the aim of providing positive recommendations for improvements. Examples of daily feedback responses are listed in
A total of 15 patients with T2D were recruited for the pilot study from a larger cohort of registered outpatients attending a large public hospital in Melbourne, Australia.
Participants were preselected from the public hospital database and associated clinics based on their T2D diagnosis (none were recently diagnosed), basic computer skills, and access to digital media. In addition, the recruited participants had no disabilities or health conditions that could interfere with their activity levels. Of the 15 participants, there were 4 (27%) women and 11 (73%) men. Patient ages ranged from 42 to 65 years. Participants aged >65 years were excluded because of presumed unfamiliarity with the use of smartwatches, smartphones, and SMS text message communications. It was important that participants had a reasonable familiarity with using smart technology (smartphones) and the ability to adapt to any smart gadgets and devices provided in this study (
Study participant being taken through the
Two Monash University design researchers conducted and audio recorded 30 in-depth, semistructured, and descriptive interviews with the 15 study participants. The interviews were conducted twice over the period of the 12-day study: once at the beginning—while participants were being fitted with the sensor and trained on
Interview questions were based on the study’s primary outcome measures and covered the participants’ background, how they managed their diabetes before and after the study, their digital literacy, their attitudes toward managing their health and diet, and their levels of satisfaction with using digital eHealth technology represented by the
Although sourced from public clinics, only 1 patient was known to the lead physician of the study before the study. The feedback was anonymous, and the physician was not identified.
Patient characteristics were compared from baseline to the final visit using the Student
All audio recordings from participant interviews were reviewed by 2 design researchers using a deductive framework approach for data analysis [
Ethical approval for this pilot project was applied for and was granted (project ID LR63/2017) by the Eastern Health Research Ethics committee on October 9, 2017.
Our study’s participants were on a wide range of antidiabetes medications (
Glucose lowering therapies at baseline (n=15).
Therapy | Participants, n (%) |
Diet alone | 0 (0) |
Insulin | 2 (13) |
Metformin | 14 (93) |
SUa | 2 (13) |
TZDb | 0 (0) |
DPP-4c | 1 (7) |
GLP-1d | 6 (40) |
SGLT-2e | 3 (20) |
aSU: sulfonylurea.
bTZD: thiazolidinedione.
cDPP-4: dipeptidyl peptidase-4 inhibitor.
dGLP-1: glucagon-like peptide-1 agonist.
eSGLT-2: sodium-glucose cotransporter–2 inhibitor.
Baseline characteristics of the patients in the study (n=15)a.
Characteristic | Values, mean (SD; range) | ||
Age (years) | 54.07 (7.16; 41-65) | ||
Height (cm) | 135.67 (12.44; 110-158) | ||
Weight (kg) | 98.09 (10.50; 80.8-115.6) | ||
Systolic BPb (mm Hg) | 135.67 (12.44; 110-158) | ||
Diastolic BP (mm Hg) | 85.07 (9.11; 71-103) | ||
BMI (kg/m2) | 31.95 (3.64; 26.0-40.1) | ||
Waist to hip ratio (n=13) | 0.98 (0.04; 0.91-1.06) | ||
|
HbA1cc (%) | 7.94 (2.14; 5.8-13.6) | |
|
HbA1c (mmol/mol) | 63.3 (23.4; 40-125) | |
Fructosamine (mmol/L) | 295.80 (80.78; 221-545) | ||
Total cholesterol (mmol/L) | 4.78 (0.89; 3.7-6.70) | ||
HDLd cholesterol (mmol/L) | 2.39 (1.06; 0.90-4.70) | ||
Triglycerides (mmol/L) | 1.19 (0.22; 0.90-1.87) | ||
Creatinine (µmol/L) | 82.27 (34.30; 39-163) |
aFemale to male ratio was 4:11.
bBP: blood pressure.
cHbA1c: hemoglobin A1c.
dHDL: high-density lipoprotein.
Electronic data collected by wearable arrays were incomplete for a variety of technical reasons. The original data set contained 30,000 data points for each patient. It was estimated that approximately 79.86% (2300/2880) of the glucose trace and 76.19% (4320/5670) of the activity and pulse rate data were available for analysis. There were sufficient data for a feedback response on 83% (10/12) of the days of the study. As a pilot or feasibility study, this study was underpowered to detect changes in HbA1c, nor did it have a control group that did not receive feedback.
The effects of the intervention on anthropometry and parameters derived from bioelectrical impedance are shown in
Changes in measured parameters from day 1 to day 12 of the study.
Parameter | Mean change from day 1 to day 12 | |
Systolic BPa (mm Hg) | −4.47 | .21 |
Diastolic BP (mm Hg) | −2.93 | .09 |
Heart rate (beats/min) | −1.67 | .41 |
Weight (kg) | −0.64 | .06 |
BMI (kg/m2) | −0.91 | .12 |
Waist hip ratio | 0.01 | .27 |
HbA1cb (%) | −0.22 | .004 |
Fructosamine (mmol/L) | −10.36 | .16 |
Creatinine (μmol/L) | 3.27 | .10 |
Total cholesterol (mmol/L) | −0.25 | .15 |
LDLc cholesterol (mmol/L) | −0.15 | .30 |
Triglycerides (mmol/L) | −0.24 | .43 |
HDLd cholesterol (mmol/L) | 0.01 | .66 |
aBP: blood pressure.
bHbA1c: hemoglobin A1c.
cLDL: low-density lipoprotein.
dHDL: high-density lipoprotein.
The changes in blood parameters and vital signs between the baseline and final visits are shown in
Activity was assessed by the number of steps per hour, as measured by the Apple Watch. The average number of steps per hour declined from 442 in the first 4 days to 399 in the last 4 days of the study, and this did not reach significance. Heart rate, as measured by the Apple Watch, was analyzed for changes in maximum heart rate, SD of heart rate, and resting heart rate (defined as heart rate at 5 AM), comparing values for the first 4 days of the intervention with those for the last 4 days of the intervention; no significant changes were found.
Changes in the continuous blood glucose trace were examined from the first 4 days of the study (days 1-4) and compared with those for the last 4 days of the study (days 9-12;
Mean interstitial glucose recorded over the first 4 days of the study compared with last 4 days of the study.
Analysis of the interviews showed that participants were curious about their personal health information and were keen to learn how to use real-time tracked information to manage their health. Although they tended to have a long-standing relationship with their family physician, they felt that the depth of information about their diabetes from scheduled general practitioner (GP) checkups was limited. Most participants planned their meals as per family and convenience rather than nutrition, influencing meal choices and quantities. Activity levels varied; fewer than half engaged in planned exercise, and only 13% (2/15) were
Of the 15 participants, 6 (40%) had previously used health-tracking devices. Of these 15 participants, 10 (66%) had used smartphone apps before for monitoring their health, and 11 (73%) had sought additional information from the internet during the study period.
All patients reported having a relationship with their existing family GP for a long duration (between 2 and 30 years), with an average of 12 years. Adherence to regular health practitioner visits varied; almost half (7/15, 47%) made 3-monthly GP visits for prescription renewal or checkups. Many had been referred to dieticians and other allied health professionals but did not attend regularly after the initial education following a diabetes diagnosis. Only 20% (3/15) of patients visited their diabetes nurse educator at either the 3- or 6-month intervals. Patients who saw their GP more regularly visited their diabetes nurse educator more frequently. Patients felt that the usefulness and amount of advice and follow-up on diabetes from health professionals varied. One of the study’s patients who initially saw a diabetes nurse educator felt that “there was nothing that she could tell me that I didn’t already know” (P2014). Another reported that their physician did not give advice on aspects of diabetes management: “He’s a doctor, not a physical educator” (P2002).
A patient who visited their GP strictly for prescription renewal suggested a strong desire to be given relevant information about T2D self-management: “...don’t tell me that I’m a naughty boy and that I’m sick but tell me how to manage it” (P2009).
Patients we talked with showed curiosity and willingness to learn even in the perceived absence of professional health advice: “I’m trying to figure this stuff out on my own” (P2010).
One of the patients expressed exasperation at having to deal with a chronic disease with so many individualized variables:
I’ve been diabetic for a bloody long time. Twenty-odd years. And I still don’t get it...I just found the whole process damn confusing.
Of the 15 patients, 11 (73%) cooked for themselves at least part of the time. For many, family requirements restricted their free choice of meals. Some noted a lack of time, not wanting to plan, or simply finding it
Only 27% (4/15) of the participants actively engaged in moderate planned exercise and only 13% (2/15) in high-level exercise. One of the participants felt that
So he’s like, “can you see if you can just increase a little bit?” and suggested a five-minute walk. And I’m thinking, well, he understood that I do have issues. And he’s like, “We’re not asking you to go for a 10-kilometer run”
Of the 15 participants, 13 (87%) said they would have been happy to continue using the
Some participants felt that photographing food was difficult to do in public, especially when dining out. From feeling self-conscious to feeling as if it was a nuisance, participants noted that this aspect of the study was the biggest burden. However, at the same time, participants commented that photographing their food made them feel more accountable (to themselves) and aware of what they were eating and drinking. Participants would have liked 2-way communication with the physician, especially to clarify aspects of the meal they had photographed or to ask a question about the SMS text message feedback:
It would have been nice if there was an option to be able to respond to the feedback and ask questions, because...he [the physician responding] gives you a direct, “if you do this, this and this....” But that was based on assumptions. Like, for example, yesterday my lunch was a quiche where it’s not a real egg and bacon pie. He’s like, “Your sugar spiked because of the crust on that.” But it doesn’t have a crust on it because I made it myself. It’s just basically egg and bacon in a pie dish. So just to give back and say, “Well, you actually haven’t quite got the advice right”...at the moment it’s sort of a one-way street.
All participants were satisfied with the tone and helpfulness of the feedback, even when the physician’s comments on food choices were not always positive: “It’s nice having someone in your corner” (P2015).
They looked forward to receiving the feedback and appreciated the personalized aspects. Some perceived that the information was already known to them; it validated their own knowledge about their diet and habits as a patient suggested:
...it didn’t provide me anything I didn’t know...I might not have known it at the front of my mind, but I DO know it
Others gained useful new insights and felt more in control of their choices.
Patient 2015 felt more in control and could see how continued app use might improve knowledge of their specific health behavior: “If I’d known what I know now, then things would be so much better” (P2015).
Of the 15 participants, 3 (20%) noted an increase in their feelings of positivity and well-being following the study, and some felt that participation increased discussions and changed family routines around healthy food choices. Others determined that they would measure glucose levels daily rather than every few days as a result of being in the study. An awareness of the need to change behavior by acting upon the advice was suggested: “It’s like anything, if you’re getting the information, it’s worth nothing if you don’t work off it” (P2001).
Of the 15 participants, 2 (13%) participants mentioned feeling supported and motivated to change their behavior as a result of having the physician’s feedback on a daily basis, as patient 2007 commented the following:
All this, I consider learning; it’s a learning thing. And it’s understanding. For diabetes is massive. And I mean, even this—being held accountable. And I think, as I said, my diabetes people are fantastic. But they can’t phone me daily.
This pilot study used wearable technology to gather data on activity, exercise, pulse rate, interstitial fluid glucose, and food intake and give patients with T2D daily text-based feedback that would provide short advisory comments (nudges) on food intake and activity based on the previous day’s data.
Advice on behavior change is often based on the average responses of groups to particular foods rather than on individual responses. It has recently become clear that there are large differences between individual glycemic responses to food and that approaches based on average responses, such as a glycemic index, may be inherently flawed [
It is crucial that new technologies are brought to bear to facilitate behavior change both through providing real-time visibility of blood glucose profiles and by providing nudging messages to reinforce positive messages on a frequent (daily) basis. We believe that regular, frequent, positive, and suggestive feedback using the strategy of nudging will, over time, significantly modify behavior and prevent the development of diabetes and reduce other cardiovascular risk factors such as hypertension and hypercholesterolemia, thus leading to weight loss or a positive change in body fatness.
A survey of Australian patients with T2D about a
The main difference between our research and others is in testing the usefulness of the
Although the optimum frequency of messages is not yet known, our study and others [
The diabetes self-management app studies we reviewed deployed both generic (not based on individual data inputs) and personalized tailored messages. The degree of personalization varied. On one end, Dobson et al [
In our
I like having someone to report to. I find it keeps me on track. It keeps me honest; it keeps me motivated
Although it is possible that participants might become
The study had some limitations because of limited resources, including time and specific technical components (eg, Medtronic continuous glucose monitor, Apple Watch, and Android smartphone) used for the
The study was intended to address the consumption of foods with a high glycemic index rather than a poor diet in general (eg, one high in saturated fat and sodium); this is a limitation of the study. The tool had a basic functionality to support glucose levels, activity data monitoring, and physician feedback. Feedback from the participants regarding their experience using the
Despite the small sample sizes in our pilot study, our findings support the potential of mobile app–based, daily personalized physician feedback as an intervention for positive changes in behavior and health outcomes in people with T2D. A follow-up study is needed to ascertain both the long-term engagement with the app and the extent to which long-term behavior change is feasible. Specific areas to be explored in follow-up studies include the following:
How feedback that is analyzed and delivered via AI rather than by a physician might reproduce the personal and motivational effect of coaching-style feedback
Whether 2-way communication enhances motivation for sustained behavior change
How behavior change models could be deployed to personalize feedback messaging [
Interviews with patients with T2D in this study provided important insights into how the experience could be made more engaging and presumably more effective. Future wide-scale applications of daily personalized feedback delivered through an app would be limited by the availability of physicians or even specially trained dieticians to provide this feedback. This factor, and the learning points from our pilot study, could be incorporated in the design of new integrated wearable technology that will enable scaling up of this kind of intervention through the use of newer and less invasive sensor technology, possibly by deploying an AI approach in the generation of feedback. This approach aligns with predicted advances in customized diabetes treatment by incorporating machine-based algorithms [
Our study suggests that providing daily physician-generated personalized feedback based on wearable sensor information and recorded food intake and activity data can produce favorable changes in glycemic and cardiovascular risk parameters even in the short term. The participants found the experience acceptable, and it provided them opportunities for positive long-term behavior changes. We plan to address the feedback collected through the interviews to redevelop the system and conduct a longer study with more participants.
artificial intelligence
general practitioner
hemoglobin A1c
type 1 diabetes
type 2 diabetes
This work was supported by a seeding grant from the Monash Institute of Medical Engineering granted to CG. Contributions were also provided by the Monash Faculty of Information Technology and the Special Purpose Fund of the Endocrine Department of Eastern Health. The authors would like to thank Flavia Xie-Ku of Monash University Art, Design, and Architecture and Yuxin Zhang and Farid Raisi of Monash University Faculty of Information Technology for their contributions to the
CG led and conducted the clinical study design and implementation and contributed to manuscript writing and reviewing. MME assisted CG and helped conduct the clinical study. FB led the system development, contributed to the study design and technical evaluation, and drafted the first manuscript. DF led the
None declared.