A total of 160 patients (80 in each group) were included in the matched cohort. At baseline, the mean (SD) age of patients was 47.13 (10.56) years, with 66 (41.3%) patients being female. The mean (SD) values for BMI, fasting blood glucose, and HbA_1c levels were 25.88 (4.06) kg/m², 9.17 (2.96) mmol/L, and 8.54% (2.36%), respectively. Regarding diabetes treatment, 71.9% (115/160) of the patients were treated with OADs alone, whereas 28.1% (45/160) of the patients were on insulin regimens, either alone or in combination with OADs. Diabetic microvascular complications were present in 53.8% (86/160) of the patients, whereas diabetic macrovascular complications were present in 4.4% (7/160) of the patients. The baseline characteristics of the 2 groups are shown in Table 1 (for characteristics before matching, see Table S1 in Multimedia Appendix 1). The 2 groups were similar in most baseline characteristics, with absolute standardized differences <10%, except for the duration of T2DM, which had a difference of 15.57%.

In the mobile app–assisted SMBG group, the median (IQR) frequency of blood glucose monitoring was 0 (0-2) times per week, with 28% (22/80) of patients monitoring their blood glucose at least twice per week. The median (IQR) app usage frequency was 1 (0-3) time per week, with 40% (32/80) of patients logging in at least twice per week.

Figure 2 shows unadjusted outcomes at 12 months. The mean (SD) fasting blood glucose in the mobile app–assisted SMBG and control groups was 7.09 (2.09) mmol/L and 7.31 (2.38) mmol/L, respectively (MD −0.21 mmol/L, 95% CI −0.91 to 0.49 mmol/L; P=.55). The proportion of patients achieving or maintaining a fasting blood glucose level of <7 mmol/L in the mobile app–assisted SMBG and control groups was 60% (48/80) and 61% (49/80), respectively (OR 0.95, 95% CI 0.50-1.79; P=.87). The mean (SD) of HbA_1c in the mobile app–assisted SMBG and control groups was 6.78% (1.51%) and 6.91% (1.61%), respectively (MD −0.13%, 95% CI −0.62% to 0.36%; P=.59). The proportion of patients achieving or maintaining an HbA_1c level of <7% in the mobile app–assisted SMBG and control groups was 68% (54/80) and 69% (55/80), respectively (OR 0.94, 95% CI 0.49-1.84; P=.87).

Table 2 reports adjusted outcomes at 12 months. There were no statistically significant differences observed between the mobile app–assisted SMBG group and the control group in glycemic control outcomes at 12 months. Specifically, the results showed (1) no significant difference in fasting blood glucose and HbA_1c levels (MD −0.17 mmol/L, 95% CI −0.85 to 0.51 mmol/L; P=.62 and −0.12%, 95% CI −0.58% to 0.33%; P=.59, respectively) and (2) no significant difference in the proportion of patients achieving or maintaining a fasting blood glucose level of <7 mmol/L and an HbA_1c level of <7% (OR 0.89, 95% CI 0.46-1.73; P=.74 and OR 0.91, 95% CI 0.44-1.88; P=.79, respectively). An additional data analysis at 6 months is provided in Table S2 in Multimedia Appendix 1. At 6 months, the mobile app–assisted SMBG group showed a reduction in HbA_1c compared with the control group (adjusted MD −0.43%, 95% CI −0.86% to 0%; P=.05), reaching borderline statistical significance. No significant differences were observed for fasting blood glucose levels or target achievement rates.

Our study assessed the long-term effectiveness of mobile app–assisted SMBG in improving glycemic control in patients with T2DM at 12 months, alongside standard care, in Ningbo, China. Although the mobile app–assisted SMBG group showed lower fasting blood glucose and HbA_1c levels compared with the control group, these differences were not statistically significant. Similarly, the proportion of patients achieving or maintaining a fasting blood glucose level of <7 mmol/L and an HbA_1c level of <7% was lower in the intervention group; however, these differences did not reach statistical significance.

Despite the lack of statistical significance, our findings offer valuable insights into the real-world application of mobile app–assisted SMBG. While standard care in this well-resourced setting has been shown to be effective [28-30], opportunities for further improvement remain. In our study, the supplementary mobile app–assisted SMBG did not yield clear benefits, likely due to process-related factors such as limited engagement, rather than a ceiling effect. Addressing these implementation challenges should be a key focus of future research. Specifically, previous studies have demonstrated that standard care at MMC effectively manages glycemic control in patients with T2DM [28-30]. Similarly, in our study, at baseline (ie, the point of registration at MMC for standardized diabetes care), the mean fasting blood glucose in both groups exceeded 9 mmol/L. By 12 months, this had decreased to approximately 7 mmol/L. Likewise, the mean HbA_1c at baseline was ≥8.5%, which declined to <7% at 12 months. These findings suggest that MMC’s standardized care may be effective in improving glycemic control in individuals with T2DM. Patients in the intervention group monitored their blood glucose less frequently than the approximately 2 times per day typically associated with meaningful glycemic improvements [31], which may have limited the intervention’s impact. The supplementary 6-month analysis showed a borderline significant reduction in HbA_1c in the intervention group, but this was not sustained at 12 months. This trend likely reflects a decline in engagement over time—a common challenge in digital health interventions—and aligns with previous studies reporting reduced effectiveness of mobile app–assisted SMBG as usage wanes [32-35]. Furthermore, given that our study participants had a median diabetes duration of 3 to 4 years, they were likely not newly diagnosed and may have already been familiar with SMBG through standard care, potentially reducing the added benefit of mobile app–assisted SMBG. Research indicates that mobile app–assisted SMBG is most beneficial for patients newly diagnosed with T2DM, particularly within the first 3 to 6 months, when treatment adjustments are most critical [34,36,37].

Systematic reviews examining factors associated with the uptake, engagement, and retention of mobile apps, including those used for self-management of chronic conditions such as diabetes, have identified a range of influences aligned with the Capability, Opportunity, Motivation for Behavior model. These include challenges such as usability challenges, technical difficulties, lack of customizable features, absence of supportive functionalities (eg, reminders or feedback), and limited perceived usefulness of the apps [38-40]. In addition, a previous study evaluating Chinese mobile apps for diabetes self-management identified concerns related to app quality, functionality, and features and recommended improvements through co-design involving researchers, health care professionals, and target users [41]. While it is possible that the developers of the mobile app used in our study received some informal feedback from key stakeholders, we are not aware of any systematic co-development approach undertaken.

To the best of our knowledge, this is the first study to assess the long-term (12 mo) effectiveness of mobile app–assisted SMBG in a real-world cohort of patients with T2DM in Ningbo, China. Unlike randomized controlled trials, which are rigorously controlled but may not fully reflect the complexities of real-world clinical settings [42,43], our study leveraged real-world data from routine clinical practice within a high-quality treatment environment. Propensity score matching was used to create a balanced control group and address potential confounders, with known confounders further adjusted for in the analysis. Despite this, the absolute standardized difference in diabetes duration between groups remained above the accepted 10% threshold, with the intervention group having a shorter average duration. This imbalance may have conferred an advantage, as shorter diabetes duration is typically associated with better glycemic control [24], and mobile app interventions tend to be more effective in individuals with more recent diagnoses [44]. Nevertheless, we did not observe statistically significant improvements in glycemic control in the intervention group, even with this potential bias in its favor.

Several other limitations should also be acknowledged. First, the small sample size reduced statistical power, which may have limited our ability to detect subtle differences. As this was a retrospective analysis of existing data, no formal sample size calculation was performed. Instead, we included all eligible patients (n=1570) and applied propensity score matching to design this retrospective cohort study (n=160). Because patients voluntarily purchased the mobile app–assisted glucometers, the intervention group may have differed from the control group, introducing potential confounding factors. Apart from the app development and engagement issues discussed earlier, the study encountered additional process-related limitations due to restricted dataset access granted by the app developers for research purposes. First, although we reported the overall frequency of app engagement by patients, including SMBG, the dataset available to us did not include longitudinal data needed to evaluate usage patterns over time. This constrained our ability to assess whether declining engagement may have contributed to the waning effectiveness of the intervention. In addition, we did not have access to data on how clinicians and patients interacted with specific app features—such as doctor-patient communication, delivery of clinical advice, or patient self-management—limiting insights into the mechanisms of engagement. Safety data were also unavailable in this retrospective study, as they were not routinely collected, limiting the comprehensive evaluation of the intervention. Because the study was conducted in a highly specialized, resource-intensive setting, the generalizability of our findings to lower-resource environments may be limited. There are inherent trade-offs in real-world observational studies, which prioritize practical applicability and relevance over the controlled conditions typically found in randomized controlled trials. Future large-scale, prospective studies are needed to evaluate high-quality mobile app–assisted SMBG interventions, incorporating strategies to optimize engagement and focusing on both long-term effectiveness and safety. Causal inference frameworks such as the Bradford Hill criteria or the 9 principles for determining causality in epidemiology should be applied to strengthen the evidence base [45].

In conclusion, in a real-world cohort of patients with T2DM in Ningbo, China, mobile app–assisted SMBG did not lead to statistically significant improvements in glycemic control at 12 months. This suggests that in a well-resourced setting, standard care alone may be relatively effective. However, opportunities for further improvement remain. The lack of observed benefit may be due to process-related issues, such as suboptimal engagement with the intervention. Addressing these challenges should be a focus of future research.