The One Drop | Mobile app was launched in April 2015. It is available for free on iOS, WatchOS, and Android operating systems.

One Drop | Mobile has a variety of features to support diabetes management (see Figure 1). Users can manually and passively (via HealthKit, Google Fit, and the Bluetooth-enabled One Drop | Chrome blood glucose meter) store and track blood glucose readings, medication doses, physical activity, and foods consumed. In addition, users can view daily, weekly, and monthly summary statistics regarding these data. A built-in food library facilitates tracking food. An optional medication scheduler reminds users when a dose is due and facilitates tracking medications. Users can also view the percentage of in-range blood glucose readings over time and store and track HbA_1c values and body weight. Importantly, they can set daily goals (for time in range, medication adherence, carbohydrate intake, and physical activity) and monitor their progress toward these goals. Users can also access a wide array of diabetes-relevant information by using an in-app newsfeed that delivers health tips, articles, infographics, user polls, expert interviews, and scientific study results. A community section lets the user view and learn from other users’ data. A map displays dots representing other One Drop | Mobile users in a local area, anywhere, and provides an option to view another user’s data and give badges to offer support and encouragement. A notifications inbox delivers data-driven insights, achievements, reminders, and lists badges accumulated from other users.

Solutions Institutional Review Board approved analyses and reporting of One Drop | Mobile’s data for research purposes.

Summary statistics characterized the sample overall and stratified by diabetes type. Distributions of continuous variables were asymmetrical, so Mann Whitney U tests compared mean ranks of continuous user characteristics, app-tracked data, and HbA_1c percent. Chi-square tests assessed differences in dichotomous variables by diabetes type. To examine and exclude invalid self-reported blood glucose and HbA_1c data, we converted each user’s 90-day average blood glucose to an estimated HbA_1c using the formula HbA_1c=(90-day mean blood glucose+77.3)/35.6 [18]. We calculated the difference between the converted HbA_1c and self-reported HbA_1c and excluded users with more than a 2.0% difference. For the remaining users, Spearman’s rho correlations tested the relationship between self-reported HbA_1c values and the prior 90-day average blood glucose to ensure consistency with the literature [19]. Because most users enter their first HbA_1c when they initiate using the app, we were unable to assess the relationship between 90-day average blood glucose prior to the first self-reported HbA_1c.

Missing data were handled using multiple imputation [20]. We used predictive mean matching (PMM) [21,22] to impute 100 datasets. PMM is a multiple-imputation method robust to violations of distributional assumptions (eg, normality) [23,24]. Multiple imputation was carried out in SPSS 23.0 (IBM Corp).

Next, 3 mixed effects repeated measures models tested mean HbA_1c change by time (pre- to posttest), diabetes type (T1D vs T2D), and their interaction. Only these effects were in the first model (ie, the unadjusted model). The second model adjusted for a priori covariates: gender, location (US vs non-US), years since a diagnosis of diabetes, and the number of months between the first and second HbA_1c entries. We restricted the third model to users with T2D, excluded the time by diabetes type interaction term, and adjusted for gender, location, years since diagnosis, number of months between HbA_1c entries, and insulin use.

Finally, 4 multiple regression models assessed the relationships between change in HbA_1c and using the app to track blood glucose, activity, medications, and food. The first, unadjusted model assessed the relationships between HbA_1c change and the 4 types of self-care tracking. The second model included diabetes type (T1D vs T2D), and the third model included gender, location, years since diagnosis, and number of months between the first and second HbA_1c entries. We restricted the fourth model to users with T2D and included insulin use as well as the a priori covariates. Given the skewness of self-care data and assumption violations for statistical testing, we dichotomized each variable to indicated tracking or nontracking of blood glucose, medications, activity, and food.

We assessed changes in HbA_1c for people with T1D or T2D who used the One Drop | Mobile app over a period of 1 year. We also evaluated relationships between tracking self-care with the app and HbA_1c change during that time. App users reported up to a 1.27% absolute decrease in HbA_1c depending on their diabetes type. Using the app to track food intake was associated with greater HbA_1c reductions.

Landmark studies, including the Diabetes Control and Complications Trial [25] and United Kingdom Prospective Diabetes Study [26], found lowering HbA_1c closer to normal levels reduced the risk of diabetes complications. According to recent reviews, diabetes apps are associated with reduction of HbA_1c [1-3], but their effectiveness varies widely across studies and by diabetes type.

One review reported people with T1D who used diabetes apps had a 0.36% HbA_1c reduction in 3 to 9 months [1]. For people with T1D using the dBees self-care and blood glucose tracking app, there was no HbA_1c reduction over time or relative to controls using a paper logbook [5]. In another trial, 34 children and teenagers with T1D using the mDiab/Mobil Diab tracking and self-care support app reduced their HbA_1c by 0.72%, but HbA_1c also fell by 0.98% in the control group [8]. In a nonrandomized controlled trial, 90 adults with T1D and HbA_1c≥8% using the Diabeo digital diary and insulin calculator lowered their HbA_1c by 0.91% relative to controls [6]. Among 36 people with T1D in Australia using the Glucose Buddy tracking app, HbA_1c was reduced by 1.10% [7].

Based on 367 people with T1D using the One Drop | Mobile app, we found HbA_1c declined by 0.86%—an amount consistent with other studies evaluating publicly available apps but more than two-fold larger than the overall effect of diabetes apps tested among people with T1D [1]. Moreover, unlike the previous trials described above, we related HbA_1c change to tracking self-care with an app, and found, regardless of diabetes type, using the app to track food consumption was associated with a greater HbA_1c reduction.

For people with T2D, an evaluation of 10 studies testing diabetes apps found an average HbA_1c reduction of 0.49% [3]. One of those studies was an RCT evaluating the publicly available WellDoc app (now available as Bluestar) that reported a 2.03% drop in HbA_1c among 15 people with T2D in one urban area. Our observational study with no control group or randomization included a sample of 921 people with T2D, and found HbA_1c decreased by 1.27%. This HbA_1c improvement is comparable to the difference in HbA_1c improvement between the WellDoc intervention and control groups and more than double the effect of diabetes apps used by people with T2D in a recent meta-analysis by Hou et al [3]. In that meta-analysis, one other trial evaluated a publicly available app [9]. The trial evaluated the mDiab/Mobil Diab app as used by 40 people with T2D in Butembo, Democratic Republic of Congo [9]. HbA_1c improved by 1.78% [9]. The baseline HbA_1c was 0.54% higher than in our study.

This study has limitations. There was no control group or randomization. Multiple potential confounding factors may have contributed to the observed results, making it impossible to ascribe causal relationships between using the One Drop | Mobile app and HbA_1c change. The significant relationship between using the app to track self-care and HbA_1c benefit enhances confidence of a direct link. Users were also self-selected in terms of their using the app and self-reporting 2 or more HbA_1c values, introducing external validity and generalizability concerns. This possibility, however, is also a concern with any RCT in which participants self-select to participate. Our sample also reflects people willing to use a diabetes app. It is plausible to assume these people are younger, have a higher socioeconomic status (ie, a higher income, education) and are more comfortable using technology. To protect privacy, One Drop | Mobile does not collect user age, precluding the ability to describe this and other characteristics (eg, education, income, insurance status) of the sample or adjust for them in analyses. One Drop | Mobile also has other features we did not relate to HbA_1c change or adjust for in our analyses. HbA_1c was self-reported rather than assessed with a laboratory assay. Because the app is a tool for the user and not subject to review by others, it is unlikely users altered their HbA_1c values in response to social desirability bias. Consistent with prior studies that used laboratory HbA_1c values, we found a greater HbA_1c improvement among people with T2D than people with T1D [1]. Also, self-reported HbA_1c was highly correlated with average blood glucose 90 days before the HbA_1c, increasing confidence in its utility as an indicator of glycemic control in this study. Finally, our sample included over 1200 PWD from both within and outside the United States, differentiating it from other studies that included people from only one country or region.

There are currently no best practices for evaluating mobile health apps [27], and clearly more research is needed. This study adds to that body of work. Diabetes app developers collect data that can both improve product offerings and user experience and evaluate how users may be benefiting.

We believe people want and deserve mobile health apps that address their self-care needs and enhance their ability to improve the management of their chronic health condition [17,28]. Selecting an app is challenging. There are over 1500 diabetes apps to choose from with more being developed. A review of 65 publicly available diabetes apps concluded 86% were unfit for promoting self-management [29]. Ratings by consumers can be a poor indication of an app’s clinical efficacy [30]. The results of carefully developed clinical evaluations will help consumers select better apps and assist providers in recommending efficacious apps to patients.