Many studies have shown that diabetes mellitus not only results in specific complications but also leads to increased risks of cardiovascular disease and cancer [1-3]. Although it might have adverse outcomes, diabetes is now considered a chronic disorder that can usually be controlled with appropriate treatment, lifestyle management, and self-care measures to keep blood glucose in the target range [4,5].

Among self-care measures, the self-monitoring of blood glucose (SMBG) is a critical component for checking blood glucose levels [5-7]. Several studies have provided evidence that SMBG has notable benefits on glycemic control, and a recent meta-analysis showed that SMBG has beneficial effects on glucose control in both the short- and long-term [8-10]. Specifically, previous research articles have shown that SMBG is helpful for patients with diabetes mellitus type 1 in controlling blood sugar level. Furthermore, one randomized controlled trial (RCT) recruited patients with diabetes mellitus type 2 (DMT2) and observed them for at least 12 months, and the results suggested that SMBG improves diabetes control [11]. However, for patients with non–insulin-treated DMT2, the value of SMBG was inconsistent among the studies [11-15]. The difference may be due to the different research designs or targets of diabetes type in the studies.

To address the limitations of previous studies, we focused on patients with non–insulin-treated DMT2 who were receiving care from a single clinic and investigated the relationship between the frequency of SMBG and the patients’ glucose levels. In addition, there is growing evidence suggesting that digital technologies are being adopted as an additional method for health care systems to increase patient contact and enhance the effect of conventional care practices for diabetes patients [16,17]. We also focused on the patients who used a diabetes management app with self-care measures during the observation period. Therefore, the objective of this study was to analyze whether the app and SMBG affected glycemic control.

Table 1 presents the patients’ demographic characteristics stratified by the 2 groups. There were no significant differences in gender (Χ²₁=0; P>.99), diabetes duration (t₄₁₈=–0.69; P=.49), and the baseline blood glucose level (t₄₁₈=–0.58; P=.56) between the 2 groups. However, significant differences were found in age (t₄₁₈=–6.66; P<.001) and education level (Χ²₁=45.44; P<.001).

Linear mixed modeling was used to estimate the effects from factors that could affect the patients’ blood glucose percentage change. In addition, due to the above analysis of patient characteristics, we know that there are significant differences in the patients’ age and education level between the patient groups. Therefore, we have to confirm whether the interactions of these factors and the groups should be put into the original model.

First, we built a basic model to check whether the patients’ group, age, and education level have an effect on the blood glucose percentage change. This basic model only included the patients’ group, age, and education level; these factors exhibited significant effects on the blood glucose percentage change (P<.001 for the patients’ age; P<.001 for the patients’ education level; P<.001 for the patients’ group). Second, we wanted to examine whether an interaction effect of the patients’ age and group has an effect on the blood glucose percentage change. We built a second model that included the same factors as the basic model, but also added an interaction effect of the patients’ age and group. The second model showed that the interaction effect of the patients’ age and group did not have a significant effect on the blood glucose percentage change (P=.53). We then built a third model that included the same factors as the basic model and the interaction effect of the patients’ education level and group. The third model showed that the interaction effect of the patients’ education level and group did not have a significant effect on the blood glucose percentage change (P=.48). Finally, we built a fourth model with the same factors as the basic model and added the 2 interaction effects—one for the patient’s age and group and another for the patient’s education level and group. In the fourth model, we found that the interaction effects did not have a statistically significant effect on the blood glucose percentage change (P=.109 for the interaction effect of the patients’ age and group; P=.94 for the interaction effect of the patients’ education level and group). Therefore, we decided not to incorporate the age-group and education level-group interaction factors into the original model.

After conducting the above analyses, our original model was kept, and the final, included variables consisted of the patients’ group, gender, diabetes duration, age, education level, the interaction effect of the nth month after registering an account, and the count of blood glucose measured in the nth month. Table 2 presents a summary of the new model.

We found that the app-engaged-user group had significantly greater decreases in blood glucose percentage change than the only-data-uploader group (β estimate=–2.430 × 10^-2; t=–3.873, df=1.81 × 10⁴; P<.001 for the patients of the app-engaged-user group). In addition, for patients with shorter diabetes duration and those who are younger, the magnitudes of the drops in blood glucose percentage change were more profound (β estimate=1.010 × 10^-3; t=2.823, df=1.71 × 10⁴; P=.005 for diabetes duration; β estimate=7.474 × 10^-4; t=2.652, df=1.71 × 10⁴; P=.008 for the age of the patients; Figures 2-3). We also found that the frequency of SMBG enlarged the decreases in blood glucose along the interaction months (β estimate=–3.352 × 10^-4; t=–8.266, df=1.97 × 10⁴; P<.001 for the nth month × the count of blood glucose in the nth month; Figure 4). Additionally, when the group of patients with an education level of junior high school or lower was set as the baseline, these patients had significantly greater decreases in blood glucose percentage change than those with an education level of senior high school or higher (β estimate=2.927 × 10^-2; t=4.996, df=1.72 × 10⁴; P<.001 for patients with an education level of senior high school or higher; Figure 5). Lastly, the gender of the patients did not significantly affect the percentage change (β estimate=2.753 × 10^-3; t=0.534, df=1.74 × 10⁴; P=.59 for female patients, with male patients as the baseline).

Our study was based at a single clinic to minimize differences between the frequency of patient visits, level of health education, and quality of care received for the app-engaged-user and only-data-uploader groups. Our results showed that there were 6 significant factors—the patients’ group (app-engaged-user or only-data-uploader), age, diabetes duration, education level, gender, and the count of blood glucose measured in the nth month—that were more strongly associated with changes in the patients’ blood glucose. We found that patients who are app-engaged, younger, and less-educated and have shorter diabetes duration saw a steeper decrease in their blood glucose percentage change. We also found that the interaction between the nth month of recording SMBG and the SMBG count of that month affected blood glucose level significantly. Therefore, this interaction deserves more attention than total SMBG count. However, we found that the gender of the patients did not significantly affect the percentage change.

As previously mentioned, in many studies for patients with non–insulin-treated DMT2, the value of SMBG is inconsistent [8,11,22-24]. Some studies have demonstrated that SMBG was effective in controlling blood glucose [8,22,23,25], whereas other studies claimed that SMBG was not effective [12,26]. These inconsistencies are mainly due to differences in the trial designs, populations studied, and outcome indicators. However, in our findings, we used the count of blood glucose measured in the same month instead of the use of SMBG as a measurement. This is different from some previous studies. Diabetes patients test SMBG differently according to their current blood glucose status. Generally, when a patient’s blood glucose becomes more stable, the count of SMBG will decrease. Therefore, it is more accurate to look at how the month of SMBG testing and count of SMBG each month affects blood glucose levels than total SMBG count.

Patients with longer diabetes duration may be affected by more diabetes symptoms [27-29], so their control of diabetes is usually worse than patients with a shorter diabetes duration. In addition, aging is associated with physiological changes that may lead to systemic alterations [30]. These systemic alterations may affect mental and physical functioning, increasing the chances of morbidity, multimorbidity, and mortality [30]. Older patients with diabetes may have macrovascular and microvascular complications and geriatric syndromes [28,31], so their control of diabetes is usually worse than younger patients with diabetes.

A common assumption is that patients with higher educational levels would have more knowledge about diseases and therapies, and thus, they would be able to better comply with therapies. However, previous studies have found that even highly educated patients may not sufficiently understand their conditions or truly believe in the benefits of therapy compliance, whereas patients with lower education levels may trust the doctor’s advice more and exhibit better compliance [32,33]. This could explain our finding that the blood glucose percentage drop of the patients with an education level of junior high school or lower was greater than those with an education level of senior high school or higher. In addition, our study showed that the blood glucose percentage decrease of the patients who used the Health2Sync mobile app was more than those who did not use the app. For patients with the same education level, those using the Health2Sync mobile app had a greater decrease in blood glucose percentage than the patients who did not use it. Furthermore, regardless of the level of education, the patients who used the Health2Sync app experienced a larger drop in blood glucose levels than those who do not use it. The Health2Sync app benefits users because it allows them to record their daily behaviors together with blood glucose readings, and the app has a bot that provides automated analyses, alerts, encouragements, and personalized educational content [16]. For the app-engaged-user group, 66% (69/104) of the patients recorded behavioral factors in addition to self-reported outcomes, most commonly entering diet, and 85% (88/104) viewed at least one educational content or interactive educational guide the app provided. Previous studies have shown that patient education and diet management are crucial for improving blood glucose [8,23,25,27]. There is growing evidence suggesting that gender affects the pathophysiology of many diseases, but in our study, gender did not significantly affect blood glucose percentage change [34-37]. Our study focused on a single clinic with limited samples; future studies should consider including a few more clinics to obtain more data samples for analyses. The other limitation is that we did not consider the differences in daily behaviors such as exercise and diet across the patients, and that these factors could have vital impacts on glycemic control. Future studies should also include these behaviors for analyses.

Diabetes is approaching epidemic proportions globally, and it places an enormous burden upon both the patients and countries’ health systems. It is especially difficult for low- and middle-income countries, due to insufficient equipment and clinics, to cope with the rise in diabetes and other chronic diseases [25,38]. The Health2Sync app can enhance the care for patients with diabetes and solve resource-limited problems.

Additionally, our study showed positive results at a single diabetes management clinic using real-world data without prior RCT settings. RCTs are generally considered by regulators to be the gold standard for establishing the causal relationship between medication and patient outcomes, but it is incapable of reflecting real clinical practice in which heterogeneous scenarios exist [39-41]. As digital interventions are to be applied to all patients, we believe that our study with real-world data is more convincing in demonstrating efficacy.

In conclusion, through the retrospective analyses, we showed that the Health2Sync app and SMBG contribute to the improvement and control of blood glucose. Further studies are needed to reveal whether different clinical care methods have an impact on diabetes treatment.