Country-Specific Differences in Glycemic Outcomes Using Real-World Data from the mySugr® App for Diabetes Self-Management

Abstract

Introduction:

Digital diabetes technologies have the potential to enhance disease self-management and reduce the socioeconomic burden on health care systems. This retrospective real-world data study aimed to investigate the impact of the mySugr® app on glycemic outcomes on a global scale and within individual countries to identify socioeconomic factors that influence country-specific outcomes.

Methods:

From January 2018 to October 2024, data were collected from 13,826 mySugr® app users, who met the inclusion criteria (at least two blood glucose logs per day on ≥14 days/month for 3 months with at least one insulin log entry). Change in mean estimated HbA1c (eHbA1c) was assessed over 3 months compared with baseline, analyzed globally, and stratified by diabetes type, country of origin, and World Bank region. A linear regression model was fit to study the relationship between country-specific eHbA1c change and socioeconomic factors (gross domestic product [GDP] per capita, health care expenditure [HE], ratio and World Bank region).

Results:

App use was associated with a significant mean eHbA1c reduction from 7.55% at baseline to 7.19% over 3 months (P < 0.001). Significant improvements were observed for both type 1 and type 2 diabetes (P < 0.001) and across geographic regions. A linear model incorporating GDP per capita, HE, World Bank region, and a GDP*HE interaction was able to explain differences between countries to a large degree (adjusted R² value = 0.535). Higher GDP and HE were associated with a better eHbA1c improvement, with the HE effect being stronger in high-GDP countries. These results show an improved glycemic control with the app’s use across diverse health care systems and highlight socioeconomic factors that can predict a majority of the outcome differences between countries, offering valuable insights for health economic modeling and informing policy discussions on digital health reimbursement.

Introduction

Diabetes mellitus, a chronic metabolic disease, is one of the biggest health challenges worldwide which can lead to serious secondary diseases and complications.^1–10 According to the 11th edition of the International Diabetes Federation report, the total number of adults living with diabetes is expected to rise from 588.7 million in 2024 to 852.5 million in 2050.¹ Continuous monitoring and tight control of blood glucose (BG) is a cornerstone of diabetes management. Moreover, BG monitoring is the prerequisite for continuously optimized and personalized diabetes management with regard to nutrition, medication, medical devices, and associated technologies.^11–13

Over the past few years, traditional methods of diabetes monitoring have been increasingly supplemented with digital diabetes technologies such as mobile applications (mobile health, m-health), which aim to help individuals monitor their BG levels and make informed decisions for their daily diabetes management. Various randomized controlled trials (RCTs) showed the benefits of digital diabetes technologies in terms of improved glycemic control and reduction of diabetes distress in people with both type 1 and type 2 diabetes (T1D, T2D).^14–16 While RCTs assess the efficacy of digital diabetes technologies under controlled conditions, real-world complexities and variability inherent in the real-world settings such as diverse populations with varying comorbidities and socioeconomic backgrounds can lead to different outcomes. Therefore, real-world data (RWD) studies are essential to validate the effectiveness of digital diabetes technologies and their clinical utility.^17,18

In this study, we analyzed RWD from the mySugr® mobile application (mySugr GmbH) with the aim of quantifying the changes in estimated glycated hemoglobin (eHbA1c) both globally and at a country-specific level during 90 days of application use.¹⁹ To understand differences in outcomes for the respective countries, we built a predictive model that relates the country’s socioeconomic attributes with changes in eHbA1c.

Methods

In this retrospective RWD study, we assessed the impact of the mySugr® mobile application (mySugr GmbH) on glycemic outcomes measured by eHbA1c from a global perspective as well as stratified by different countries and regions according to the definition of the World Bank. Based on these results, we developed a linear model to explain country-specific differences in glycemic outcomes by using socioeconomic parameters on the country level (see below for details).

All data analyzed in this publication were collected between January 1, 2018 and October 10, 2024, through the use of the mySugr® mobile application. This CE-marked digital diabetes technology hosts multiple medical devices to support diabetes self-management such as a digital logbook and an insulin bolus calculator. All mySugr® users provided their consent to their data being used for research purposes, which covers their use within the context of this publication.

Finally, we comply with the Reporting of studies Conducted using Observational Routinely-collected health Data (RECORD) guidelines to ensure adequate reporting and reproducibility.²⁰

In addition to the mobile application’s data, we used the following external datasets to classify countries into World Bank regions and to relate socio-economic metrics to outcomes: •

Global health care expenditure database from the World Health Organization.²¹

•

Gross domestic product (GDP) per capita from the World Bank open data platform.²²

•

World Bank region classification.²³

These external datasets were mapped based on 3166 standard of the International Organization for Standardization 3-digit code for each country.²⁴ We used economic information from 2021, for reasons of completeness.

Cohort definitions

The study cohort consisted of people with diabetes that used the mySugr® mobile application between 2018 and 2024, who met the following criteria: •

At least two BG logs per day on at least 14 individual days within a 30-day period (G2D14 logging class) in all months from baseline (the month before their first mySugr® use) to month 3 after their first mySugr® use. Here, the baseline data were imported into the app from Bluetooth-connected meters or third-party services such as Apple Health.

•

At least one insulin entry in the logbook between the baseline month and month 3, which could be manually entered or synced from connected devices.

This cohort was further stratified based on different attributes for subset analyses such as self-reported diabetes type (for example T1D and T2D), the reported country of origin, and the World Bank region according to the country of origin. For all stratifications, at least 25 app users were required to ensure anonymity.

Objectives of the data analysis

The analysis’ primary objective was to assess the mobile application’s impact on glycemic outcomes within 90 days of use, as measured by the absolute and relative mean change in eHbA1c values compared with the baseline month in the global cohort. eHbA1c values were calculated as described by Nathan et al.¹⁹ The 30-day period before the first use (first logbook entry) of the application was defined as the baseline month, while months 1, 2, and 3 were the first 30, 31–60, and 61–90 days of the mobile application’s use. One month was defined as 30 days. This analysis was additionally performed with the following user stratifications: •

Mean change of eHbA1c values in month 3 stratified by diabetes type (T1D, T2D).

•

Mean change of eHbA1c values in month 3 stratified by diabetes type and by World Bank region.

•

Mean change of eHbA1c values in month 3 stratified by diabetes type and by country.

As a secondary objective, we aimed to understand the relationship between the respective countries’ socioeconomic attributes and their specific eHbA1c change through a modeling approach described in detail below. We hypothesized that patients in countries able to afford higher investments in health had better access to diabetes treatments.

Statistical methods

Baseline characteristics such as eHbA1c, diabetes type, age, gender, and years with diabetes were described for the full analysis set (FAS) of all included app users fulfilling the inclusion criteria as well as stratified per country of origin and per World Bank region. We also categorized the cohort by three classes of baseline eHbA1c values: less than 7%, 7%–9%, and greater than 9%. These descriptive analyses of glycemic outcomes were based on the total population as well as stratified per diabetes type (T1D, T2D), country of origin, and per World Bank region. For all outcome stratifications, the eHbA1c was calculated from the average mean BG per month and reported both in absolute terms and in relation to the baseline month. To quantify the uncertainty of estimation, the 95% confidence interval (CI) around the mean was calculated using bootstrapping statistics.²⁵

Two-sided, one-sample t-tests were performed to quantify whether the change in eHbA1c at the end of the observation period (month 3) as compared with baseline is significantly different from 0 for the overall cohort and the overall cohort stratified by diabetes type. The statistical evaluations were carried out with python 3.11.4, statsmodels 0.14.0, scipy 1.11.1, pandas 2.1.1, and numpy 1.26.4.

Modeling of outcome based on baseline information

In order to relate the dependent variable of glycemic change at month 3 versus the baseline month with country-level information, the following information was included in our model, as they would be readily available for countries without the need for specific medical testing: •

Average age per country.

•

GDP per capita of the respective country.

•

Health care expenditure reported as a fraction of GDP.

•

Average years with diabetes per country.

•

World Bank region membership of the country.

•

Income group of the country from the World Bank region classification.

As a preprocessing step, only countries with all available data were selected and the continuous variables were standardized to mean 0 and unit standard deviation. The categorical variables (Income and Region) were one-hot encoded before fitting the model. The number of countries with sufficient users (25) was comparatively low (n = 40). Therefore, we used the following stepwise procedure to ensure that only relevant information was included in the final model: 1.

All possible subsets of single features were computed.

Among these features, all possible pairwise interactions were computed.

Ordinary Least Squares models were fitted using the subset of single features and including up to a maximum of three interactions.

Among these models, the adjusted Akaike information criterion for small sample sizes²⁶ was calculated and the feature combination with the lowest value was taken as the final model.

To ensure that the model also performs well on held out data, 10-fold cross validation was formed and the mean absolute error (MAE) for each fold was calculated and compared with the overall MAE.

The model fit was evaluated using an F-test as well as the adjusted R² value. We used a two-sided one-sample t-test to verify whether the relevant coefficients were significantly different from 0. The statistical modeling was done by using python 3.11.4, statsmodels 0.14.0, scipy 1.11.1, pandas 2.1.1, and numpy 1.26.4.

Results

This retrospective RWD analysis included a total of 13,826 individuals who used the mySugr® mobile application between January 1, 2018 and October 10, 2024, and met the inclusion criteria. The majority came from the Europe and Central Asia region (48%), followed by Latin America and Caribbean (27%), and North America (20%). The proportions of people with T1D and T2D in the FAS were 47.7% and 43.5%, respectively. The remaining proportion consisted of people with other diabetes types such as latent autoimmune diabetes in adults or patients that chose not to disclose their diabetes type. Table 1 shows the detailed baseline demographics for the total population including age groups, gender, and years with diabetes. As a minimum number of 25 included individuals were required for all stratifications to ensure anonymity, the analyses for diabetes types per World Bank region and per country only differentiated between T1D and T2D (Table 2 and Supplementary Table S1). Looking at baseline eHbA1c values, we distinguished between the T1D and T2D subgroups since it is known that there are differences in baseline values. Figure 1 indicates better glycemic control at baseline in the subgroup with T2D. In total, 49.2% of the individuals with T2D reported an eHbA1c <7%, while this proportion was only 36.6% in the T1D subgroup.

FIG. 1.

Number of individuals per baseline eHbA1c group (<7%, 7%–9%, >9%) in the subgroups with T1D and T2D. eHbA1c, estimated HbA1c; T1D, type 1 diabetes; T2D, type 2 diabetes.

Table 1.

Baseline Demographics Total Cohort, N = 13,826

Parameter	Number of included individuals	%
Gender
Male	5639	40.8%
Female	3332	24.1%
Unknown	4855	35.1%
Age groups (years)
<20	196	1.4%
20–30	795	5.8%
30–40	1007	7.3%
40–50	1191	8.6%
50–60	1538	11.1%
60–70	1729	12.5%
70–80	1243	9.0%
80–90	341	2.5%
>90	24	0.2%
Unknown	5762	41.7%
Diabetes type
T1D	6590	47.7%
T2D	6018	43.5%
Latent autoimmune diabetes in adults (LADA)	464	3.4%
Gestational diabetes mellitus	247	1.8%
Other	223	1.6%
Maturity-onset diabetes of the young (MODY)	30	0.2%
Unknown	254	1.8%
Years with diabetes
<5	2155	15.6%
5–10	1544	11.2%
10–15	899	6.5%
>15	3289	23.8%
Unknown	5937	42.9%
World Bank region
South Asia	98	0.7%
Middle East and North Africa	189	1.4%
Sub-Saharan Africa	228	1.6%
East Asia and Pacific	934	6.8%
North America	2753	19.9%
Latin America and Caribbean	2993	21.6%
Europe and Central Asia	6631	48.0%

T1D, type 1 diabetes; T2D, type 2 diabetes.

Table 2.

Baseline Characteristics by World Bank Regions; N = 13,826

World Bank region	Gender (N = 13,826), [n]			Diabetes type (N = 12,608), [n]		eHbA1c [% and 95 % CI]
World Bank region	Male	Female	Unknown	T1D	T2D	T1D	T2D
East Asia and Pacific	321 (34%)	179 (19%)	434 (46%)	427 (53%)	385 (47%)	7.75 [7.63,7.86]	7.28 [7.16,7.39]
Europe and Central Asia	3,184 (48%)	1,426 (22%)	2,021 (30%)	2,990 (49%)	3,070 (51%)	7.61 [7.57,7.65]	7.34 [7.3,7.38]
Latin America and Caribbean	809 (27%)	909 (30%)	1,275 (43%)	2,186 (81%)	515 (19%)	7.98 [7.93,8.03]	7.56 [7.45,7.68]
Middle East and North Africa	77 (41%)	52 (28%)	60 (32%)	113 (65%)	62 (35%)	7.51 [7.33,7.69]	7.92 [7.62,8.22]
North America	1,150 (42%)	678 (25%)	925 (34%)	679 (27%)	1,878 (73%)	7.88 [7.77,7.99]	7.37 [7.31,7.42]
South Asia	33 (34%)	16 (16%)	49 (50%)	49 (55%)	40 (45%)	7.83 [7.5,8.16]	8.08 [7.71,8.46]
Sub-Saharan Africa	65 (29%)	72 (32%)	91 (40%)	146 (68%)	68 (32%)	8.16 [7.9,8.42]	7.62 [7.34,7.9]
Overall	5,639 (41%)	3,332 (24%)	4,855 (35%)	6,590 (52%)	6,018 (48%)	7.78 [7.75,7.81]	7.38 [7.35,7.41]

Changes in eHbA1c values associated with application use

We analyzed the changes in mean eHbA1c values from the baseline period within a 90-day period of app use. In the FAS, the mean eHbA1c value decreased from 7.55% at baseline to 7.19% in month 3, which is a relative eHbA1c change of −0.36% (two-sided one-sample t-test P < 0.001). The eHbA1c value decreased the most within the first month of app use and then increased again slightly (Fig. 2). When we stratified the eHbA1c changes either by country and World Bank region, we saw that while the specific changes varied geographically, qualitatively every geographic stratification showed an improvement in eHbA1c (Table 3 and Supplementary Tables S2 and S3). When we stratified the overall cohort by diabetes type, we observed a significant decrease from baseline eHbA1c to month 3 for both patients with T1D and T2D (two-sided one sample t-test P < 0.001 for both comparisons; Fig. 3). The mean eHbA1c value in the T2D group remained relatively stable after the initial decrease, while it increased again in months 2 and 3 among people with T1D (Fig. 3). This rebound trend in the T1D group was observed in most World Bank regions except the Middle East and North Africa and South Asia (Table 3).

FIG. 2.

Changes in eHbA1c values from the baseline period within 90 days of app use in the total cohort; N = 13,826; (A) absolute eHbA1c change; (B) relative eHbA1c change.

FIG. 3.

Changes in eHbA1c values from the baseline period within 90 days of app use in the total cohort, stratified by diabetes type; (A) absolute eHbA1c change; (B) relative eHbA1c change.

Table 3.

eHbA1c Changes from Baseline to Month 3 by World Bank Regions and by Diabetes Type; All Continuous Values Are Reported as Mean with the 95% Confidence Interval

World Bank region	Overall					T1D					T2D
	Users (n)	eHbA1c at baseline [%]	Relative (Δ) eHbA1c [%]			Users (n)	eHbA1c at baseline [%]	Relative (Δ) eHbA1c [%]			Users (n)	eHbA1c at baseline [%]	Relative (Δ) eHbA1c [%]
			Month 1	Month 2	Month 3			Month 1	Month 2	Month 3			Month 1	Month 2	Month 3
East Asia and Pacific			934	7.42 [7.36,7.47]	−0.33 [−0.36,−0.3]			−0.32 [−0.36,−0.29]	−0.33 [−0.37,−0.29]	427			7.75 [7.63,7.86]	−0.34 [−0.39,−0.29]	−0.24 [−0.3,−0.18]	−0.22 [−0.27,−0.16]	385	7.28 [7.16,7.39]	−0.34 [−0.39,−0.29]	−0.41 [−0.47,−0.35]	−0.43 [−0.49,−0.37]
Europe and Central Asia	6631	7.43 [7.41,7.45]	−0.39 [−0.4,−0.38]	−0.38 [−0.4,−0.37]	−0.37 [−0.39,−0.36]	2990	7.61 [7.57,7.65]	−0.43 [−0.45,−0.41]	−0.35 [−0.37,−0.33]	−0.30 [−0.32,−0.28]	3070	7.34 [7.3,7.38]	−0.37 [−0.39,−0.35]	−0.43 [−0.45,−0.41]	−0.45 [−0.47,−0.43]
Latin America and Caribbean	2993	7.87 [7.45,7.51]	−0.37 [−0.39,−0.36]	−0.32 [−0.34,−0.29]	−0.29 [−0.31,−0.27]	2186	7.98 [7.93,8.03]	−0.38 [−0.4,−0.36]	−0.28 [−0.31,−0.26]	−0.25 [−0.28,−0.23]	515	7.56 [7.45,7.68]	−0.39 [−0.43,−0.34]	−0.45 [−0.5,−0.39]	−0.45 [−0.51,−0.39]
Middle East and North Africa	189	7.66 [7.55,7.77]	−0.44 [−0.5,−0.37]	−0.44 [−0.52,−0.36]	−0.50 [−0.59,−0.42]	113	7.51 [7.33,7.69]	−0.42 [−0.49,−0.34]	−0.40 [−0.49,−0.32]	−0.44 [−0.53,−0.35]	62	7.92 [7.62,8.22]	−0.42 [−0.57,−0.27]	−0.51 [−0.69,−0.33]	−0.63 [−0.81,−0.45]
North America	2753	7.48 [7.83,7.9]	−0.39 [−0.41,−0.37]	−0.40 [−0.43,−0.38]	−0.42 [−0.45,−0.39]	679	7.88 [7.77,7.99]	−0.43 [−0.47,−0.38]	−0.33 [−0.37,−0.28]	−0.28 [−0.33,−0.23]	1878	7.37 [7.31,7.42]	−0.38 [−0.4,−0.35]	−0.43 [−0.46,−0.4]	−0.47 [−0.5,−0.44]
South Asia	98	7.93 [7.76,8.1]	−0.45 [−0.56,−0.35]	−0.49 [−0.62,−0.36]	−0.55 [−0.69,−0.41]	49	7.83 [7.5,8.16]	−0.16 [−0.28,−0.03	−0.23 [−0.38,−0.09]	−0.21 [−0.38,−0.04]	40	8.08 [7.71,8.46]	−0.81 [−1.0,−0.63]	−0.80 [−1.05,−0.56]	−0.96 [−1.2,−0.73]
Sub-Saharan Africa	228	7.96 [7.82,8.1]	−0.45 [−0.53,−0.36]	−0.45 [−0.54,−0.36]	−0.36 [−0.45,−0.26]	146	8.16 [7.9,8.42]	−0.46 [−0.58,−0.35]	−0.46 [−0.59,−0.34]	−0.33 [−0.46,−0.2]	68	7.62 [7.34,7.9]	−0.42 [−0.54,−0.29]	−0.38 [−0.5,−0.27]	−0.40 [−0.52,−0.27]

Modeling approach to predict outcomes based on baseline information

Adopting the variable selection procedure described in detail in the methods section, we arrived at a final model that only relies on the World Bank region membership, the GDP per capita, and the health care expenditure ratio of the respective country: $eHba 1 c hange \sim H E + GDP + Region + GDP * H E$

Remarkably, this model was able to explain the differences between countries to a large degree (adjusted R² value: 0.535; F(3.834) = 0.00255).

Table 4 and Supplementary Figure S1 show the model’s relevant coefficients. Here, a higher GDP per capita as well as a higher health care expenditure ratio are associated with better glycemic outcomes. There is a significant interaction effect between the GDP per capita and the health care expenditure ratio, suggesting that a higher health care expenditure ratio is associated with better glycemic outcomes, especially in countries with a high GDP per capita. There are significant differences between World Bank regions: the Middle East & North Africa, and South Asia showed better outcomes, while North America shows worse outcomes when other factors are considered. In order to exclude that the model overfits to the relatively low number of included countries, we performed 10-fold cross validation and found a comparable mean absolute error on the train and test folds (0.13 vs. 0.09; see Supplementary Fig. S2 for an example).

Table 4.

Coefficients of the Linear Model to Explain the Differences in the Glycemic Outcomes per Country

	Coef	Std err	t	P > \|t\|	[0.025	0.975]
Intercept	−0.3038	0.060	−5.043	0.000	−0.427	−0.181
wb_region[T.Europe& Central Asia]	0.0117	0.069	0.170	0.866	−0.129	0.152
wb_region[T.Latin America & Caribbean]	−0.0688	0.081	−0.854	0.400	−0.233	0.096
wb_region[T.Middle East & North Africa]	−0.3456	0.109	−3.164	0.004	−0.569	−0.123
wb_region[T.North America]	0.3097	0.138	2.248	0.032	0.028	0.591
wb_region[T.South Asia]	−0.4009	0.156	−2.566	0.016	−0.720	−0.082
wb_region[T.Sub-Saharan Africa]	−0.1382	0.144	−0.957	0.346	−0.433	0.157
che_gdp	−0.0940	0.033	−2.806	0.009	−0.162	−0.026
gdp_per_capita	−0.0516	0.029	−1.775	0.086	−0.111	0.008
che_gdp:gdp_per_capita	−0.0558	0.027	−2.074	0.047	−0.111	−0.001

Discussion

This global RWD analysis highlights that the use of digital diabetes technology is associated with improved glycemic control when analyzed globally, on a geographic region, and country level. We further show that the changes in glycemic control are highly variable between countries and find that a simple set of socioeconomic attributes readily available from public datasets are sufficient to explain the majority of these differences.

The glycemic improvement associated with usage of digital diabetes technologies are in line with previous studies both for other applications as well as for the mySugr® application.^27,28 This work extends these findings with a global dimension and demonstrates that usage of a digital diabetes technologies is associated with a decrease in eHbA1c across different health care systems and other geographic differences.

Interestingly, while we observe a qualitative decrease in all World Bank regions and countries, there are substantial geographic differences between countries. To our knowledge, this is an effect that has not been described before systematically with the same digital diabetes technologies and holds important implications for health economic modeling of the potential impact of such an application.

Since this finding presents a unique opportunity to understand how differences between countries affect the outcomes associated with using a digital diabetes technology, we compiled a list of candidate predictors that we reasoned would be available for a large fraction of the world’s countries and performed feature selection based on these. Remarkably, the best model that resulted from this procedure was extremely simple and only relied on the GDP per capita, the health care expenditure as a fraction of the GDP and the World Bank region membership of the respective country. Despite its simplicity, the model was able to explain a large degree of the variance present in the data, suggesting that these factors are among the most important variables determining inter-country variability.

The positive impact on glycemic outcomes of the GDP per capita is an expected association and also has been described before for specific countries and health care systems.^29,30 Here our findings suggest that this relationship is broadly applicable for different countries and health care systems. The underlying mechanisms behind these associations are expected to be complex, but we can hypothesize that not only the better overall quality of life associated with higher income exerts an effect but also better access to medication and treatment and the digital readiness of the specific country,³¹ the socioeconomic status,³² and the adoption of digital diabetes technologies by the particular health care system may play a role.³³

The association of the health care expenditure ratio with positive glycemic outcomes has not been described directly before in the literature as systematic studies between countries are missing. However, also this association is plausible since lower health care expenditure will constrain infrastructure development, reduce available resources such as access to medication and treatment, and impact the ability to invest in and scale up health innovations like digital diabetes technologies.³⁴ In addition, there are studies that look at the impact of diabetes incidence on health care expenditure,³⁵ suggesting that appropriate treatment for diabetes poses a significant burden on the health care system.

A particularly interesting finding of the model is the interaction effect between GDP per capita and the health care expenditure ratio. Here, our data suggest that when a country has a high GDP per capita, increases in the health care expenditure ratio are associated with bigger glycemic improvements. Similarly, this finding suggests that if a country has a low GDP per capita, additional health care expenditure is associated with smaller improvements.

The significant association of glycemic improvements with different World Bank regions is particularly noteworthy as it persists after controlling for the GDP per capita and the respective health care expenditure ratio. Here, it might reflect the overall baseline glycemic control before starting with a digital diabetes technology since worse initial control is associated with better glycemic outcomes³⁶ and glycemic control is generally poorer in low- and middle-income countries (LMICs).³⁷ This might explain why regions with a high proportion of LMICs (Middle East and North Africa, South Asia) might be associated with bigger improvements as compared with North America.

While the associations found through modeling are interesting by themselves, the model also enables prediction of expected improvements in countries where we were not able to measure the impact of the mySugr® application on glycemic outcomes directly. This is relevant for health economic modeling of the impact of digital diabetes technologies use, enabling the prediction of economic impact only through knowledge of publicly available data and the use of a model relating HbA1c changes to economic impact, while applying the “best available data” principle.³⁸

One of the primary strengths of this study is its broad database, which includes a large number of users with both T1D and T2D from multiple countries. This diverse dataset allows for a comprehensive analysis of the impact of a digital diabetes technology on glycemic control across various demographics and geographic regions over time. In addition, the large number of users investigated allows for several stratifications, including per country and World Bank region. Finally, the use of the same digital diabetes technology across different countries and health care systems during the same time period allows for the comparison of per-country differences.

Despite its strengths, the study has several limitations. First, the diabetes type, age, and gender of the users are self-reported voluntarily, which may introduce inaccuracies and potential biases in the data, especially in terms of misclassification, which could impact the interpretation of the outcomes. Future studies should aim to validate user-reported data through independent datasets where possible. Furthermore, this is an observational study that does not include confounding variables that are potentially associated with receiving the treatment or modulating or mediating its effects such as medication use, physical exercise, inherent motivation to improve diabetes management or general quality of treatment. While these factors were not controlled for in the reported results to maximize the number of countries we can include in the analysis, a detailed sensitivity analysis could strengthen causal inference in future work. Additionally, the reported geographical distribution of users is unequal, with a low number of users in some World Bank regions and countries, which may skew the findings. Another limitation is the selection bias related to the study population; users must be able to afford a smartphone and have a certain level of digital affinity to use digital diabetes technologies, which may not represent the broader diabetic population.³² This is particularly relevant for countries where digital literacy might be lower due to limited access to infrastructure and affordability of devices, therefore, it should be emphasized that the study findings might overestimate the overall population-level effects of digital diabetes technologies. The short follow-up duration of only 3 months is another constraint, as it may not capture long-term glycemic control improvements. This limitation highlights an opportunity for future longitudinal studies with longer follow-ups to investigate the sustainability of glycemic improvements. It is also challenging to determine whether the observed effects are due to the application itself or the treatment associated with its use as detailed data on antidiabetic drugs have not been available, other than the inclusion criterion of at least one recorded insulin entry in the logbook between the baseline month and month 3. This lack of detailed data on antidiabetic treatments may limit the ability to differentiate the direct impact of the app from the effects of the accompanying treatment regimens. A clearer distinction in future studies could provide deeper insights into the mechanism of glycemic improvement. The availability of treatment options may explain the effect of economic variables on glycemic control, i.e., the more affluent a country the more antidiabetics drugs availability. Finally, while the predictive model developed in this study is informative, care should be taken when applying it to countries substantially different from the ones investigated here since the associations found may not extend to health care systems that are radically different. A more cautious interpretation of model generalizability is warranted, particularly when comparing outcomes across countries with varying health care dynamics, regulatory frameworks, or digital adoption rates.

Overall, these results show an improved glycemic control with the app’s use across diverse health care systems and highlight socioeconomic factors that can predict a majority of the outcome differences between countries, offering valuable insights for health economic modeling and informing policy discussions on digital health reimbursement. Importantly, the findings suggest that beyond access to digital tools, investments in health care infrastructure and affordability of treatment options are essential to achieve equitable outcomes across regions. Investments in digital literacy campaigns, affordable smartphone access, and regulatory support can promote adoption in resource-constrained settings and empower individuals to take better control of their health. In higher-income settings, the scalability and personalization of these tools can further enhance their impact, reaching underserved populations and optimizing diabetes care for diverse patient needs. This study reflects the growing promise of leveraging digital health technologies to bridge gaps in health care access and quality globally. The integration of digital diabetes technologies into national health programs and reimbursement schemes has the potential to support not only better glycemic outcomes but also significant reductions in the economic burden of diabetes, benefiting patients, health care systems, and societies.

Footnotes

Acknowledgments

The authors thank Susanne Moser for her medical writing support which was funded by Roche Diagnostics International, in accordance with the Good Publication Practice 2022 guidelines. The authors thank Brigitta Monz, MD, PhD, Paco Cerletti, PhD, and Johanna Kober, PhD, Roche employees, for their scientific review and input to the article.

Authors’ Contributions

M.M., D.T., and O.U.G. developed the ideation of the article and the RWD study design. M.M. performed the statistical analyses. All authors contributed to the data interpretation, read, critically reviewed, and edited the article.

Author Disclosure Statement

All authors declare financial competing interests and are employees of Roche, and may hold stock in Roche, at the time of writing the article. All authors declare no nonfinancial competing interest. MYSUGR is a trademark of Roche.

Funding Information

This publication has been sponsored by Roche Diagnostics International AG. This study received no external funding.

Supplementary Material

Abbreviations Used

References

International Diabetes Federation. IDF Diabetes Atlas. 2025. Available from: https://diabetesatlas.org/resources/idf-diabetes-atlas-2025/ [Last accessed: September 30, 2025].

GBD 2021 Diabetes Collaborators. Global regional and national burden of diabetes from 1990 to 2021, with projections of prevalence to 2050: A systematic analysis for the global burden of disease study 2021. Lancet, 2023; 402(10397):203–234; doi: 10.1016/S0140-6736(23)01301-6

Mobasseri

, Shirmohammadi

, Amiri

, et al. Prevalence and incidence of type 1 diabetes in the world: A systematic review and meta-analysis. Health Promot Perspect, 2024; 14(2):202–205; doi: 10.34172/hpp.2020.18

Ogrotis

, Koufakis

, Kotsa

. Changes in the global epidemiology of type 1 diabetes in an evolving landscape of environmental factors: Causes, challenges, and opportunities. Medicina (Kaunas), 2023; 59(4):668; doi: 10.3390/medicina59040668

Gregory

, Robinson

TIG

, Linklater

, et al.; International Diabetes Federation Diabetes Atlas Type 1 Diabetes in Adults Special Interest Group. Global incidence, prevalence, and mortality of type 1 diabetes in 2021 with projection to 2040: A modelling study. Lancet Diabetes Endocrinol, 2022; 10(10):741–760; doi: 10.1016/S2213-8587(22)00218-2

Avogaro

, Fadini

. Microvascular complications in diabetes: A growing concern for cardiologists. Int J Cardiol, 2019; 291:29–35; doi: 10.1016/j.ijcard.2019.02.030

Crasto

, Patel

, Davies

, et al. Prevention of microvascular complications of diabetes. Endocrinol Metab Clin North Am, 2021; 50(3):431–455; doi: 10.1016/j.ecl.2021.05.005

Mishriky

, Cummings

, Powell

. Diabetes-related microvascular complications—a practical approach. Prim Care, 2022; 49(2):239–254; doi: 10.1016/j.pop.2021.11.008

Tomic

, Shaw

, Magliano

. The burden and risks of emerging complications of diabetes mellitus. Nat Rev Endocrinol, 2022; 18(9):525–539; doi: 10.1038/s41574-022-00690-7

10.

Fowler

. Microvascular and macrovascular complications of diabetes. Clin Diabetes, 2008; 26(2):77–82; doi: 10.2337/diaclin.26.2.77

11.

Davis

, Fischl

, Beck

, et al. 2022 National standards for diabetes self-management education and support. Sci Diabetes Self Manag Care, 2022; 48(1):44–59; doi: 10.1177/26350106211072203

12.

Weinstock

, Aleppo

, Bailey

, et al. The Role of Blood Glucose Monitoring in Diabetes Management. American Diabetes Association; Arlington, VA; 2020.

13.

ElSayed

, Aleppo

, Aroda

, et al. 6. Glycemic targets: Standards of care in diabetes-2023. Diabetes Care, 2023; 46(Suppl 1):S97–S110; doi: 10.2337/dc23-S006

14.

Stevens

, Gallagher

, Andrews

, et al. The effectiveness of digital health technologies for patients with diabetes mellitus: A systematic review. Front Clin Diabetes Healthc, 2022; 3:936752; doi: 10.3389/fcdhc.2022.936752

15.

American Diabetes Association Professional Practice Committee. Diabetes technology: Standards of care in diabetes— 2024. Diabetes Care, 2024; 47(Suppl 1):S126–S144; doi: 10.2337/dc24-S007

16.

Ehrmann

, Hermanns

, Finke-Gröne

, et al. Efficacy of a digital diabetes logbook for people with type 1, type 2, and gestational diabetes: Results from a multicenter, open-label, parallel-group, randomized controlled trial. J Diabetes Sci Technol, 2025; 19(5):1362–1373; doi: 10.1177/19322968241239870

17.

US Food & Drug Administration (FDA). Framework for FDA’s real-world evidence program. 2018. Available from: https://www.fda.gov/media/120060/download [Last accessed: September 30, 2025].

18.

Dang

. Real-world evidence: A primer. Pharmaceut Med, 2023; 37(1):25–36; doi: 10.1007/s40290-022-00456-6

19.

Nathan

, Kuenen

, Borg

, et al.; A1c-Derived Average Glucose Study Group. Translating the A1C assay into estimated average glucose values. Diabetes Care, 2008; 31(8):1473–1478; doi: 10.2337/dc08-0545

20.

Benchimol

, Smeeth

, Guttmann

, et al.; RECORD Working Committee. The reporting of studies conducted using observational routinely-collected health data (RECORD) statement. PLoS Med, 2015; 12(10):e1001885; doi: 10.1371/journal.pmed.1001885

21.

World Health Organization. Global healthcare expenditure database. Available from: https://apps.who.int/nha/database [Last accessed: September 30, 2025].

22.

World Bank Group. Open Data Platform. GDP per capita. Available from: https://data.worldbank.org/indicator/NY.GDP.PCAP.CD [Last accessed: September 30, 2025].

23.

World Bank Group. Countries and Economies. Available from: https://data.worldbank.org/country [Last accessed: September 30, 2025].

24.

International Organization for Standardization. 2020. Codes for the representation of names of countries and their subdivisions. Available from: https://www.iso.org/obp/ui/#iso:std:iso:3166:-1:ed-4:v1:en [Last accessed: September 30, 2025].

25.

Hesterberg

. What teachers should know about the bootstrap: Resampling in the undergraduate statistics curriculum. Am Stat, 2015; 69(4):371–386; doi: 10.1080/00031305.2015.1089789

26.

Hurvich

, Tsai

. Regression and time series model selection in small samples. Biometrika, 1989; 76(2):297–307; doi: 10.2307/2336663

27.

Debong

, Mayer

, Kober

. Real-world assessments of mysugr mobile health app. Diabetes Technol Ther, 2019; 21(S2):S235–S240; doi: 10.1089/dia.2019.0019

28.

Zivkovic

, Mitter

, Theodorou

, et al. Transitioning from self-monitoring of blood glucose to continuous glucose monitoring in combination with a Mhealth app improves glycemic control in people with type 1 and type 2 diabetes. Diabetes Technol Ther, 2025; 27(1):10–18; doi: 10.1089/dia.2024.0169

29.

, Ma

, Zheng

, et al. The effects of economic status on metabolic control in type 2 diabetes mellitus at 10 metabolic management centers in China. J Diabetes, 2024; 16(1):e13466; doi: 10.1111/1753-0407.13466

30.

Louie

, Shiffman

, Rowland

, et al. Predictors of lack of glycemic control in persons with type 2 diabetes. Clin Diabetes Endocrinol, 2024; 10(1):2; doi: 10.1186/s40842-023-00160-7

31.

Bober

, Rollman

, Handler

, et al. Digital health readiness: Making digital health care more inclusive. JMIR Mhealth Uhealth, 2024; 12:e58035; doi: 10.2196/58035

32.

Gomes

, Tang

, Chen

, et al. Socioeconomic factors associated with glycemic measurement and poor HbA1c control in people with type 2 diabetes: The global discover study. Front Endocrinol (Lausanne), 2022; 13:831676; doi: 10.3389/fendo.2022.831676

33.

Essén

, Stern

, Haase

, et al. Health app policy: International comparison of nine countries’ approaches. NPJ Digit Med, 2022; 5(1):31; doi: 10.1038/s41746-022-00573-1

34.

Gardner

DSL

, Saboo

, Kesavadev

, et al. Digital health technology in diabetes management in the asia-pacific region: A narrative review of the current scenario and future outlook. Diabetes Ther, 2025; 16(3):349–369; doi: 10.1007/s13300-025-01692-0

35.

Boye

, Bae

, Thieu

, et al. An economic evaluation of the relationship between glycemic control and total healthcare costs for adults with type 2 diabetes: Retrospective cohort study. Diabetes Ther, 2024; 15(2):395–407; doi: 10.1007/s13300-023-01507-0

36.

Garvey

, Cohen

, Butera

, et al.; GRADE Research Group. Association of baseline factors with glycemic outcomes in grade: A comparative effectiveness randomized clinical trial. Diabetes Care, 2024; 47(4):562–570; doi: 10.2337/dc23-1782

37.

Aschner

, Galstyan

, Yavuz

, et al. Glycemic control and prevention of diabetic complications in low- and middle-income countries: An expert opinion. Diabetes Ther, 2021; 12(5):1491–1501; doi: 10.1007/s13300-021-00997-0

38.

Fortwaengler

, Parkin

, Neeser

, et al. Description of a new predictive modeling approach that correlates the risk and associated cost of well-defined diabetes-related complications with changes in glycated hemoglobin (HbA1c). J Diabetes Sci Technol, 2017; 11(2):315–323; doi: 10.1177/1932296816662048