All procedures were in accordance with the ethical standards of the responsible committee on human experimentation (institutional and national) and with the Helsinki Declaration of 1975, as revised in 2000. This study was approved by the Institutional Review Board of St. Mary’s Hospital, Seoul (KIRB-018524-025). Written informed consent was not required because this retrospective observational study was performed without intervention.

This study was conducted on patients with diabetes who visited the outpatient clinic. The medical staff introduced the diabetes management service to the patients using the EMR-linked mobile application iCareD. For patients who consented to use the service after receiving overall instructions on their health data and iCareD, the medical staff installed the application on their mobile phones and provided instructions on how to use it. When a patient registered as a member, consent was obtained for their data to be linked to the service platform and for medical staff to utilize their data for treatment.

During regular medical visits thereafter, doctors utilized information from iCareD to provide customized treatment and education to patients. This study targeted patients with type 1 and type 2 diabetes who were over 19 years old and agreed to use the application.

The iCareD integrates an EMR-linked physician’s website with the patients’ mobile application; it was introduced in the endocrinology department of our hospital. Of the five physicians who volunteered to use iCareD, two actively used it in their clinics. We retrospectively extracted 1-year real-world data of patients seen by the two physicians (i.e., data recorded since the patients’ first visit [April 2017 to November 2019]). The iCareD integrates two databases (the CMCnU [EMR of Catholic Medical Center] and the patient-faced databases) connected to an interface server. The system is powered by a mobile phone where the iCareD application is installed.

First, the CMCnU database (physician side) contains joining or withdrawal information, institutional codes, and hospital registration numbers. The physicians can see the patient list virtually, select the patient, and then access the EMR-linked website in a real-world setting. Second, the self-management server, i.e., the customer database, consists of biological and hospital data (laboratory results, body weight, height, body mass index [BMI], and messages from the physicians or medical staff). On the other hand, patients were required to download the application, request service usage, complete real-name authentication, and actively utilize the application. On the day the patient registered, they were first asked to read and agree to the terms of use. Thereafter, a trained clinical research coordinator (CRC) checked the patient’s identity by comparing it with the EMR data for the first time only and provided a brief introduction of the iCareD application for approximately 15 minutes; the patient’s identification (ID) was then linked to their assigned physician. After this process, the designated physician could see the registered patient’s status. Subsequently, during an outpatient visit, physicians reviewed the patient’s self-monitored dataset, comprising data such as blood sugar levels, blood pressure readings, physical activity metrics, and dietary intake. This information was then utilized to facilitate both patient care and counselling objectives. Specifically, tailored counselling sessions were provided to patients who demonstrated proficient engagement with the application and effectively amassed relevant data, while those who had not utilized the application were proactively encouraged to adopt its usage.

The iCareD has several features that reduce the workload for physicians. The weekly and monthly reports are generated automatically, and each patient’s current health status is analyzed and summarized. Without requiring their physicians’ intervention, patients can see their analysis results and maintain their blood glucose at an optimal level. Motivational messages are also automatically sent to the patients on a weekly and monthly basis considering various metabolic factors (such as the type of diabetes, BMI, fasting blood sugar [FBS] level, sex, age, and comorbidities) and other risk factors (prehypertension, abdominal obesity, elevated triglycerides, decreased high-density lipoprotein cholesterol [HDL-C] levels, and prediabetes status). The functionalities integrated into the application are implemented as algorithms that conform to the guidelines for diabetes treatment [23].

After initiating the application service, clinical data were collected on the HbA1c, FBS, blood pressure (systolic [SBP], diastolic [DBP]), and lipid profiles (triglycerides, HDL-C, and LDL-C levels) to examine the glucose-lowering effect and chronic disease-prevention impact of this application in real-world settings.

We obtained the clinical results over time. Because this was a non-interventional study, we gathered the clinical data retrospectively. In terms of the baseline results, we allowed –30 days to 1 day (application registration day) from baseline; based on the initiation of the application service, we collected clinical laboratory data at 3, 6, 9, and 12 months after initiation. After each visit, we also allowed an interval of ±1 month for each time point due to the retrospective study design. In addition, we collected data on medication use before and after 1 year of iCareD usage. When it comes to the change in medication (insulin or oral medication) before and after, we considered a timeframe of –30 days to 1 day (starting from the day of application registration).

The administrator and authorized persons could access a patient’s registration file only with the patient’s registration number. Thus, the CRC matched the patient’s number using the application registration number, and the data were locked, i.e., any access or editing was not permitted. Using an electronic mail service, the administrator then transferred the patient’s ID with an automatically generated password to the assigned CRC to retrospectively collect the laboratory data. The CRC collected the laboratory data at five time points: baseline and 3, 6, 9, and 12 months after registration. Using these data, we analyzed the blood glucose-lowering effect of the iCareD system.

To define the degree of compliance, we calculated the number of self-monitored blood glucose (SMBG) inputs and message subscription rates using a specific formula: (total number of messages read/total number of messages sent)×100. This can be used to detect the level of compliance with the diabetes management system. After uploading the patients’ SMBG data in the iCareD system application, additional tasks included linking a Bluetooth glucometer or manual recording and saving the patient’s assigned data. For the message subscription, when the “message read” status was on, the patient clicked the message and entered the message section to see the contents. Therefore, when the patient inputted the SMBG data or had a high message subscription rate, it indicated that the patient was complying well with the mobile application services.

We classified the patients into three groups: “non-users,” “poor-compliance users,” and “good-compliance users.” During enrolment, the patients’ baseline SMBG data were input involuntarily. Users (both poor- and good-compliance users) comprised those who used the iCareD system for more than 3 days (including the first day), i.e., their SMBG data were input voluntarily for more than 2 days. Conversely, non-users comprised those who used the iCareD system for less than 3 days or achieved non-compliance with SMBG input or message subscription. Good-compliance users were those who complied with the recommended frequency of SMBG inputs and showed good-compliance with the message subscription rate. At enrolment, we instructed the patients to check their SMBG data at least once a day for when the target range of HbA1c was achieved. Good compliance was defined as achieving more than 50% of the recommended SMBG frequency [21] and showing 100% compliance with message subscriptions. Poor-compliance users were those who did not comply with the recommended frequency of SMBG inputs, achieved less than 50% of the recommended SMBG frequency, or showed poor-compliance with message subscriptions.

Therefore, to analyze the factors that could affect glucose levels or chronic disease care, we merged each patient’s data on the application (including the number of SMBG inputs and messages read), with the clinical laboratory data from the EMR (both collected over a period of 1 year). Because not all patients were registered on the same day, we separately followed each patient for 1 year. The first and last patients were enrolled in April 2017 and November 2018, respectively.

For supplemental analysis, we navigated ABC drug usages across groups (non-user, poor-compliance users, good-compliance users) from baseline to 1-year and if there were no change in type and dosage of medication, we defined its status as ‘no change,’ and if addition or increased dosage of medication, and if reduction or decrease in the number of medication, we classified those to ‘increased’ and ‘decreased,’ respectively.

At the time of the study, privacy protection and confidentiality were ensured by encrypting data that could identify the patients based on their personal information in accordance with the Korean Personal Information Protection Act. The data were managed by assigning a random ID to classify the patients during data collection, and only a designated researcher was able to access the data linking the patients’ identifying information with their IDs in the hospital.

To test the differences among variables at baseline and follow-up between the groups (non-users, poor-compliance users, and good-compliance users), we used the chi-square test for categorical variables and either analysis of variance or the Kruskal-Wallis test depending on the normality of the continuous variables. If there were significant differences between groups in the continuous variables, multiple comparison tests based on the Bonferroni correction were applied.

In each group, we tested the differences in the changes of the 1-year levels of HbA1c, weight, BMI, SBP, DBP, and LDL-C using a paired t test or a non-parametric test, Wilcoxon signed rank test. Moreover, within the P values in each group, differences in changes in 1-year ABC target achievement rate were estimated by performing the McNemar’s test. The participants who met all three criteria (HbA1c <7.0%, SBP/DBP <130/80 mm Hg, and LDL-C <100 mg/dL) were categorized as the ABC target achievement group. To investigate the effect of a 1-year use of the EMR-linked diabetes self-management system on HbA1c levels and ABC target achievement rates compared to baseline, generalized estimating equation (GEE) models were used. GEE models for HbA1c were built with an exchangeable correlation structure to allow for analyses of repeated measurements at baseline, 3, 6, 9, and 12 months. If significant interaction effects between time and group were present, to understand the interaction, the significance of the differences in each variable over time was tested according to the group. To study 1-year pre-post change of ABC medication data among groups, ABC medication usage after 1-year usage of iCareD was compared to baseline and the number of patients of each group was analyzed using chi-square test and Fisher’s exact test was applied in cases where the expected frequencies were less than 5 and constituted more than 20% of the total. Statistical analyses were performed using the SPSS software version 24.0 for Windows (IBM Corp., Armonk, NY, USA).

A total of 308 patients were included in this study (female, 146 [47.4%]) (Table 1). Among these, 29 (9.4%) and 279 (90.6%) patients had type 1 and type 2 diabetes, respectively. Their average age was 58.3±12.9 years. Furthermore, 92 (30%), 170 (55%), and 46 (15%) patients were classified as “non-users,” “poor-compliance users,” and “good-compliance users,” respectively. The BMIs did not differ significantly among these groups (non-users 24.0±3.1 kg/m², poor-compliance users 24.2±3.3 kg/m², and good-compliance users 24.5±3.5 kg/m²; P=0.763). No significant differences in clinical outcomes were observed among the three groups. The baseline ABC medication was analyzed among groups and categorized into three categories: glucose-modifying agents (metformin, dipeptidyl peptidase-4 inhibitors, sodium-glucose co-transport 2 [SGLT2] inhibitors, sulfonylureas, alpha-glucosidase inhibitors, thiazolidinediones, glucagon-like peptide-1 receptor agonists, rapid/short-acting insulins, and basal insulins [intermediate/long-acting]), blood pressure-modifying agents (renin-angiotensin system blockers [angiotensin-converting inhibitors and angiotensin II receptor blockers], calcium channel blockers, and diuretics), and lipid-modifying agents (statins, fibric acid derivatives, and ezetimibe). Among glucose-modifying drugs, the SGLT2 inhibitors and basal insulins (intermediate/long-acting) were statistically different (P<0.05), but overall usage of glucose-modifying agents, blood pressure-modifying agents, and lipid-modifying agents were not significantly different (P>0.05) across groups.

The changes in the HbA1c values over time were not significantly different among the groups, and there were no interactions among the groups (P for interaction=0.171) (Fig. 1). The HbA1c value from baseline in non-users did not change significantly over time (baseline, 3, 6, 9, and 12 months: 7.7, 7.5, 7.5, 7.5, and 7.5, respectively; P=0.407). However, the HbA1c level from baseline was significantly improved over time in good-compliance users (baseline, 3, 6, 9, and 12 months after registration: 8.3, 7.3, 7.1, 7.2, and 7.2, respectively; P<0.001) and in poor-compliance users (baseline, 3, 6, 9, and 12 months after registration: 7.8, 7.4, 7.3, 7.4, and 7.4, respectively; P<0.001).

The changes in 1-year HbA1c levels according to the time of the visit significantly differed among the groups (P=0.01) (Table 2). The mean values of HbA1c in non-users did not change significantly over time (baseline 7.7±1.8; and 12 months after registration 7.5±1.3; P=0.642). However, they changed significantly over time in good-compliance users (baseline 8.3±1.7; and 12 months after registration 7.2±0.9; P<0.001) and in poor-compliance users (baseline 7.8±1.4; and 12 months after registration 7.4±1.1; P<0.001). Post hoc analysis of the degree of HbA1c level change at 12 months in the three groups showed a significant difference (P<0.05).

The mean values of BMI in non-users also did not change significantly over time (baseline 66.7±12.2; and 12 months after registration 66.4±12.2; P=0.251). However, they changed significantly over time in good-compliance users (baseline 24.5±3.5; and 12 months after registration 24.1±3.2; P<0.01). Post hoc analysis of the degree of 1-year BMI change was not significantly different among the groups (P=0.06 and P=0.07, respectively).

The changes in the LDL-C according to the time of visit were significantly different among the groups (P<0.05). The mean values of LDL-C in non-users did not change significantly over time (baseline 81.6±29.4; and 12 months after registration 81.5±25.5; P=0.442). However, they changed significantly over time in good-compliance users (baseline 89.2±26.4; and 12 months after registration 79.4±23.4; P<0.05) and in poor-compliance users (baseline 87.4±32.8; and 12 months after registration 79.3±24.3; P<0.05). Post hoc analysis of the degree of 1-year LDL-C change in the three groups showed significant differences from the baseline levels (P<0.05).

The ABC target achievement rates were defined as HbA1c <7% [24], blood pressure (SBP/DBP) <130/80 mm Hg [25], and LDL-C <100 mg/dL [26]. The changing trends in ABC target achievement rate over 1 year were significantly different among the groups (P for interaction=0.034) (Table 3). The ABC target achievement rate was not significantly different among the groups at each point. However, when we analyzed groups separately, we observed a significant improving trend of ABC target achievement rate in a good-compliance group before and after 1 year. The ABC target achievement rate in non-users and the poor-compliance users did not change significantly after 1 year of usage (20.7 and 14.1 [baseline and 1-year], P=0.201; and 13.5 and 18.8 [baseline and 1-year], P=0.106; respectively). However, they changed significantly over time in good-compliance users (baseline 10.9; and 12 months after registration 23.9; P=0.034).

The number of changes in the ABC medications at before and after using of iCareD was not significantly different among the groups (P>0.05). We investigated the number of patients whose prescription was changed (including changes in the type and dosage of medication, as well as addition or increased dosage of medication, and reduction in dose or decrease in medication number in the ABC medication) and the number of patients whose prescription was not changed of each group. We compared changes from baseline to 1-year medication, including dosage in every route. The numbers for all three agents showed no significant differences from the baseline in total users: glucose-modifying agents (P=0.632), blood pressure-modifying agents (P=0.092), and lipid-modifying agents (P=0.195) (Supplemental Table S1).