We conducted a randomized controlled trial with 2 parallel groups between February 2017 and June 2019. The DiaCert study design [21] has been described in detail previously. Study participants were randomized 1:1 to the intervention or control group at baseline. The primary study outcome was MVPA (minutes/day) at 3 months of follow-up. Secondary outcomes included MVPA at 6 months of follow-up and the clinical variables BMI, waist circumference, HbA_1c, cholesterol (total, LDL, and HDL), triglycerides, and systolic and diastolic blood pressure at 3 and 6 months of follow-up. No changes to methods were done after trial commencement. The study is reported according to the CONSORT (Consolidated Standards of Reporting Trials) statement [22] and the CONSORT-EHEALTH (Consolidated Standards of Reporting Trials of Electronic and Mobile HEalth Applications and onLine TeleHealth) checklist, which is developed for eHealth or mHealth interventions [23].

The trial was approved by the ethics committee of the regional ethical review board in Stockholm, Sweden (Dnr: 2016/2041-31/2, 2016/99-32, 2017/1406-32) and registered at ClinicalTrials.gov (NCT03053336). All participants received both oral and written information about the study and provided their written informed consent to participate. Participants received no compensation for participation in the study. After data collection, data were anonymized.

The inclusion criteria were (1) having a diagnosis of type 2 diabetes, (2) being 18 years of age or older, (3) being able to communicate in Swedish, and (4) having access to and being able to use a smartphone. The exclusion criterion was not being able to walk.

Patients were recruited continuously from 5 primary health care centers and 1 specialized medical center in the Stockholm area, Sweden. Patients at the participating centers received initial information about the study from their physician or diabetes nurse, and those interested in participating were contacted by study personnel and given more detailed information. Thereafter, patients either agreed to participate and were scheduled for a baseline introductory meeting, or declined participation. We did not record the number of patients who did not agree to participate. All study participants met with study personnel at baseline and after 3 and 6 months. On each occasion, study outcomes including physical activity and other lifestyle factors, as well as clinical variables, were assessed.

Study participants randomized to the intervention group continued to receive standard care as usual but also downloaded the DiaCert app during the baseline meeting. To access the individual user account in the app, a personal 6-digit code was entered. The code was given to participants by study personnel 7 days after the baseline meeting to avoid overlap with baseline accelerometer measurements. The intervention group was encouraged to use the app daily for 3 months. At the 3-month follow-up meeting, participants deleted the app from their phones together with study personnel. Participants then received no intervention during 3 months and were offered access to the app again at the 6-month follow-up.

The DiaCert app displayed daily steps that through connection to a digital platform were shared with study personnel. An individual step goal between 1000 and 10,000 steps [24] was set at baseline based on the participant’s usual activity level. Participants in the intervention group were contacted by study personnel every second week by phone. During these follow-ups, the participant could revise his or her step goal with an even 500 steps increase or decrease. The maximum goal set at any time point was 10,000 steps. Users received automatic positive feedback messages including the user’s name in the app on days when the goal was met. In addition to daily steps, information on HbA_1c taken during the study period was also displayed in the app.

The app design has previously been described in detail, including a presentation of the app screen by screen [9]. In brief, features of the app included a home screen displaying the daily steps in relation to an individual step goal during the past week. A circle was gradually filled as the user walked toward the step goal. It was completely filled and marked with a checkmark when the goal was reached. Through the home screen, the user was also able to access information on previous daily steps, questionnaires, and results of HbA_1c. The app was continuously updated to run with the current iOS and Android versions, but no changes were made to the content during the study. Users were asked to contact study personnel if they experienced malfunction of the app. The app was developed within the research project and is no longer available.

Study participants randomized to the control group received standard care, that is, continued their usual care as prescribed by their regular primary care physician and diabetes nurse, also after inclusion to the study. For ethical reasons, they were offered access to the DiaCert app at the 6-month follow-up.

Power calculations were performed a priori to determine the sample size need to detect a clinically significant difference of 8 minutes/day of MVPA [21]. A total of 250 participants (125 in each group) were estimated to provide 80% power at a 5% significance level. This included an expected dropout rate of 20%. Baseline data collection ended in June 2018 before reaching 250 participants because the DiaCert app was no longer compatible with the upgrades of iOS and Android.

A random allocation sequence list was generated by the first author (SEB) in Stata (version 14.0; StataCorp). Women and men were randomized separately in blocks of 10 within each participating primary health care center and the specialized medical center. Patients who agreed to participate in the study were continuously allocated to the next available spot on the list by study personnel (authors SEB and CA). Participants were informed about their group allocation at the end of the baseline meeting. Because of the nature of the intervention, neither participants nor study personnel were blinded to participants’ group allocation.

Baseline characteristics of study participants are presented as mean (SD) or number (%) for continuous and categorical variables, respectively. Variables were checked for normality and outliers. The Student t test or the chi-square test was used to assess potential differences in baseline characteristics between the intervention and control group.

We used linear mixed models with fixed and random intercept and slope for the time variables to assess if there were longitudinal differences in MVPA at 3 months of follow-up (primary outcome) and secondary outcomes including MVPA at 6 months of follow-up and BMI, waist circumference, HbA_1c, total cholesterol, LDL cholesterol, HDL cholesterol, triglycerides, and systolic and diastolic blood pressure at 3 and 6 months of follow-up between the intervention and control group. We have therefore included, in addition to time and group terms, a group×time interaction term to assess if any differences in the study outcomes were constant at 3 and 6 months. For outcomes that differed significantly between the study groups at baseline, that is, MVPA, additional sensitivity analysis adjusting the models for baseline levels of the outcome was performed according to the methods described by Twisk et al [32]. The analysis of intervention effect was made following the intention-to-treat approach [33]. Missing data were associated with a primary care center, with 17/25 (68%) participants with missing accelerometer data at baseline belonging to primary care center 1. Missing data were assumed to be missing at random and not depending on the study group as the degree of missing data was similar in both the intervention and control group during the intervention. Participants with complete data at baseline for each specific outcome variable were included in the analysis of intervention effect at 3 and 6 months.

Post hoc sensitivity analyses using self-reported data on physical activity from the baseline questionnaire were carried out to deal with the unblinded nature of our study and the potential bias that may have been present during baseline accelerometer measurements. Although participants were not connected to the DiaCert app until after the completion of baseline accelerometer measurements, they were aware of which group they were randomized to during measurements. This could potentially have affected baseline levels of MVPA, which might not have represented the “true” levels of MVPA before the start of the study. Multiple imputation based on MICE (multiple imputation by chained equations) was used to address this issue [34]. We first set all baseline values of MVPA for the intervention group to missing. To predict MVPA at baseline for participants in the intervention group in the hypothetical scenario in which participants were blind to the intervention, we implemented MICE using all relevant baseline variables. The variables included in the model were age, sex, height, weight, education, household income, marital status, smoking and snuff habits, the year of diabetes diagnosis, the center of recruitment, self-reported levels of physical activity, and the number of valid accelerometer days at baseline. We compared the imputed and the observed values of MVPA at baseline using a 2-tailed t test. Linear mixed models, as described above, were thereafter fitted to contrast differences in MVPA between the intervention and control group using the imputed data for MVPA at baseline.

Statistical tests were 2-sided, and the significance level was set to P<.05. Statistical analyses were performed using Stata (version 17; StataCorp).

Results from between-group analysis and the intervention effect on minutes/day of MVPA are shown in Table 2. The mean change in minutes/day of MVPA from baseline to follow-ups is graphically shown in Figure 2. The statistically significant difference in minutes/day of MVPA seen between the groups at baseline, with participants in the intervention group being more active than participants in the control group, remained at the 3-month follow-up. The predicted mean difference between the groups after 3 months was 10.05 minutes (95% CI 1.66-18.44). At the 6-month follow-up, there was no statistically significant difference in MVPA between the groups (β=5.02, 95% CI –3.72 to 13.75).

We found no statistically significant effect of the intervention (group by time interaction) on MVPA at either the 3- or the 6-month follow-up (Table 2). When adjusting for baseline levels of MVPA, results remained nonsignificant at both 3 (β=4.38, 95% CI –2.11 to 10.88) and 6 (β=–0.65, 95% CI –7.58 to 6.29) months. Additionally, results from sensitivity analyses using imputed baseline data also remained nonsignificant at follow-up after both 3 (β=6.86, 95% CI –4.05 to 17.78) and 6 (β=1.44, 95% CI –9.87 to 11.76) months.

Detailed results from between-group analyses and the intervention effect on clinical variables included as secondary outcomes are shown in Table 2. The mean change in outcomes from baseline to follow-ups is graphically shown in Multimedia Appendix 2. We found no statistically significant differences in any of the secondary outcomes at the 3- or 6-month follow-ups, except for in the analysis of BMI where a statistically significant effect of the intervention was seen at 6 months (group by time interaction: 0.52, 95% CI 0.20-0.84). However, there was no difference in mean BMI between the groups at the 6-month follow-up (predicted difference in mean: 0.11, 95% CI –1.45 to 1.67). Participants in the control group had a slightly higher BMI at baseline than participants in the intervention group (30.6 vs 30.2 kg/m²), although this difference was not statistically significant (P=.61).

In this 2-armed randomized controlled trial, we found no clear effect of the DiaCert app promoting daily walking in persons with type 2 diabetes. We found no increase in objectively measured MVPA after neither 3 (primary outcome) nor 6 months of follow-up compared with standard care when accounting for baseline levels of MVPA. This is in line with results from Thorsen et al [20]. They were not able to show an effect of app-based interval walking on MVPA over 52 weeks compared with standard care among persons with type 2 diabetes. On the contrary, Poppe et al [19] found that an eHealth and mHealth intervention that primarily targeted physical activity in persons with type 2 diabetes led to increased physical activity and decreased sedentary behavior. Participants in our study and the study by Thorsen et al [20] reported more time in MVPA at baseline compared with those in the study by Poppe et al [19]. This might have contributed to the difference in effect between the studies, as a higher baseline level of MVPA may imply less room for improvement.

Other previous studies have also shown an effect of mHealth solutions to increase physical activity. Hochsmann et al [35] showed that an interactive smartphone game aimed at increasing daily steps in persons with type 2 diabetes had a significant effect on activity compared with a control group receiving standard lifestyle counseling. Glynn et al [36] evaluated the effect of an app that aimed to increase physical activity in primary care patients and found that the intervention group increased their number of daily steps compared with the control group. In the randomized Sophia step study, comprising persons with prediabetes or type 2 diabetes in Sweden [37], self-monitoring of daily physical activity using pedometers with registration of steps on the web in combination with counseling did not increase levels of physical activity but seemed to prevent the decrease in physical activity seen in the control group. Although our primary hypothesis was rejected and we could not show that the use of the DiaCert app led to increased MVPA among persons with type 2 diabetes, results from other studies still indicate that mHealth solutions can have positive effects on physical activity.

Although we found a statistically significant effect of the DiaCert intervention on BMI at the 6-month follow-up, there was no difference in mean BMI between the groups at any time point. Therefore, we cannot draw any conclusion regarding the effect on BMI. Not all studies evaluating interventions targeting physical activity include anthropometric outcomes, and the effect on BMI in previous studies are mixed, showing no effect or indicating a small reduction favoring the intervention group [10,12]. Thorsen et al [20] did not report BMI as an outcome, but found a nonsignificant reduction of waist circumference. Similar to the Sophia step study, our results provided no evidence for an effect on waist circumference [37].

Reviews and meta-analyses [11,13,15] provide strong evidence for a positive effect of mobile apps for lifestyle modification in persons with type 2 diabetes on HbA_1c levels. However, all of the evaluated apps included monitoring of blood glucose. The effects on other clinical markers including body weight, blood lipids, and blood pressure are less clear [38]. Our results are in line with the Sophia step study, where no effects on cardiometabolic variables, including HbA_1c, triglycerides, HDL cholesterol, and LDL cholesterol, were seen [37]. Neither our study nor the Sophia step study included a glucose-monitoring component, which may explain the lack of an effect on HbA_1c. Nevertheless, results from a meta-analysis by Lee et al [12] indicated that mHealth interventions in persons with type 2 diabetes >65 years of age may improve blood lipid profiles, which we found no evidence of. One explanation for the lack of an effect on cardiometabolic markers is that a changed lifestyle behavior for the better could have led to lowered medication, leaving HbA_1c, serum lipid levels, or blood pressure unchanged. We did not assess changes in medications during the intervention, which is a limitation of our study.

The use of different behavior change techniques may also explain differences in efficacy between apps. Fanning et al [39] investigated the effect of a basic self-monitoring app (tracking, feedback, information) and 2 theory-based tools (goal setting and point-based feedback) on MVPA among healthy but inactive adults. Four different groups received either a basic self-monitoring app only or the basic app together with (1) goal setting, or (2) feedback, or (3) goal setting and feedback. All groups increased their MVPA, but the feedback group showed the highest increase. While the DiaCert app comprised several behavioral change techniques that previously have been included in successful interventions, such as monitoring, goal setting, and positive feedback, it did not include other features associated with effective results, for example, frequent reminders or the option to share data with peers [40]. How the included components are designed may also affect results; for example, goal setting can be personalized or generic, and a goal can be more or less challenging.

The design of this randomized controlled study is one of the strengths of our study. The fact that study participants were recruited from 6 different care centers located in different areas with diverse populations and levels of socioeconomic status is another strength. This likely increased generalizability of our results. However, a limitation of our study is that we did not record the number of patients who turned down participation after being contacted by study personnel. Nevertheless, the mean age in our study was slightly lower than that of the average person with type 2 diabetes in Sweden, but levels of BMI and HbA_1c were similar [41]. Younger persons may be more inclined to participate in an app-based intervention, although internet access and smartphone usage are high also among older age groups in Sweden [42]. Another strength of our study is that both men and women were included. Earlier studies have shown that women participate in physical activity programs more than men [43]. The larger proportion of men in our study could partly be explained by the higher prevalence of type 2 diabetes among men; almost 60% of persons with type 2 diabetes within primary care in Sweden are men [41]. It could also be speculated that older men are more interested in using technology than older women and therefore more likely to participate in an mHealth intervention.

The fact that participants were not blinded to the intervention is a limitation. While baseline information regarding the clinical variables is unlikely to have been affected by participants being aware of their group allocation, this knowledge may have had an impact on accelerometer measurements. Participants can, intentionally or unintentionally, change their physical activity behaviors during the measuring period. The statistically significant differences between the intervention and control group in accelerometer-measured MVPA at baseline were not seen for physical activity assessed in the baseline questionnaire. This could be interpreted as an immediate effect of being randomized to the intervention group and thereby feeling encouraged to become more active. Future studies should be careful not to disclose allocation information to participants until all baseline measures have been completed. It is also a limitation that personnel working in the study were not blinded during measurements. Nevertheless, objective accelerometer measurement of physical activity and clinical biomarkers including HbA_1c and lipid levels represent a strength in our study design as they are less prone to biased estimates resulting from unblinded study personnel. The continuous recruitment of study participants during the whole year also reduced the risk of results being biased due to season.

We did not reach our goal of including 250 participants in the study, which is a limitation. Because a digital solution must be continuously updated to run with the current iOS and Android versions, we had to end recruitment after 2.5 years for practical reasons, before reaching the goal. However, dropout rates were lower than the estimated 20% in our power calculation [21]. Lack of data on user engagement and adherence to the app is a limitation as higher user engagement has also been associated with more favorable outcomes [44]. Decreased app engagement over time during an intervention period has been suggested as a potential explanation for the lack of more long-term effects of physical activity promoting apps [45].

We found no evidence that persons with type 2 diabetes being randomized to use an app promoting daily walking increased their levels of MVPA at 3 or 6 months of follow-up compared with controls receiving standard care. Further, the effect on BMI was unclear, and we found nothing to support an effect of being randomized to use the DiaCert app on waist circumference, HbA_1c, total cholesterol, LDL cholesterol, HDL cholesterol, triglycerides, and systolic or diastolic blood pressure compared with standard care. Further research is needed to determine what type of mHealth physical activity intervention could be effective to increase physical activity and improve cardiometabolic markers among persons with type 2 diabetes.