The Kangbuk Samsung Health Study is a large-scale retrospective cohort of adult male and female who have undergone comprehensive annual or biennial health checkups at the Kangbuk Samsung Hospital Total Healthcare Center in Seoul and Suwon, Korea [23]. Within this cohort, a 12-week mobile-based chronic disease management program was implemented in 2021, targeting employees identified as having HTN, T2DM, or obesity based on their 2019 health checkup, with participation being voluntary. This study focused on employees enrolled in a mobile chronic disease management program delivered as part of an EAP by Kangbuk Samsung Hospital and used their longitudinal health checkup data to assess long-term outcomes.

The initial population of 408,526 individuals who underwent at least 3 health checkups between 2002 and 2023 was further screened for eligibility. Eligible participants were those who completed at least 3 health checkups between 2002 and 2018 and who underwent follow-up health checkups in 2019 (baseline) and 2023 (follow-up). To ensure data consistency in measurement, health checkups from 2020, 2021, and 2022 were excluded from the analysis. This decision was based on changes in health checkup protocols and participation patterns during the COVID-19 pandemic that may have introduced systematic biases. During this period, social distancing policies and disruptions to routine health care services led to irregular participation in health checkups and an increased proportion of missing or incomplete data.

Based on these criteria, condition-specific cohorts were constructed for HTN, T2DM, and obesity, with participants classified into intervention and control groups according to enrollment in the mHealth program. Individuals with missing values in variables required for propensity score matching (PSM)—systolic blood pressure (SBP) and diastolic blood pressure (DBP), use of antihypertensive medication (for the HTN cohort), glycated hemoglobin (HbA_1c), fasting blood glucose (FBG), use of antidiabetic medication (for the diabetes mellitus [DM] cohort), and BMI (for the obesity cohort)—were excluded. A 1:2 PSM procedure was then performed to construct comparable intervention and control groups.

This study was approved by the Institutional Review Board (IRB) of Kangbuk Samsung Hospital (IRB: KBSMC 2024-10-032). All intervention participants provided written informed consent for participation in the program, as well as for the use of their health checkup data for research purposes. For the control group, deidentified health checkup data were used, and the requirement for informed consent was waived by the IRB. All data were anonymized prior to analysis to ensure participant privacy and confidentiality, and access to the data was restricted to authorized researchers. Participants in the intervention program received small nonmonetary incentives (eg, thank-you gifts) upon completion of the program.

The intervention consisted of a structured 12-week mobile-based chronic disease management program delivered as part of an EAP to support workplace health promotion and chronic disease risk reduction. Employees identified through annual health checkups as having HTN, T2DM, or obesity-related metabolic risk were invited to enroll in 1 of 2 separate cycles between May 24 and December 26, 2021. At enrollment, participants underwent a preintervention physical and laboratory examination at the affiliated medical institution, provided informed consent, and received a dedicated health management mobile application. Participants were provided with a personalized care plan and Bluetooth-enabled self-monitoring devices (BP monitor, glucometer, or smart scale), depending on the specific disease management program in which they were enrolled.

During the intervention period, participants were encouraged to regularly record biometric and lifestyle data using a mobile application. Biometric data, including BP, blood glucose, and body weight, were automatically synchronized through connected devices, while lifestyle information such as diet, physical activity, and medication adherence was manually entered. These data were continuously transmitted to a web-based platform, allowing health care professionals to monitor participant status in real time. As the program was delivered through a mobile app, participants were able to actively self-monitor their health data, and real-time feedback from the health care coaching team further supported engagement and motivation for behavior change.

A multidisciplinary health coaching team consisting of nurses, clinical dietitians, and clinical exercise physiologists reviewed participant data weekly and provided tailored feedback based on the primary condition and real-time records. Feedback included health information messages, individualized commentary on submitted measurements, and guidance on behavior modification to improve metabolic risk factors. The intensity and frequency of this one-on-one feedback were adapted either automatically or manually according to individual risk levels and engagement, with more frequent monitoring and interaction provided for participants with higher clinical risk or lower adherence.

To enhance participant engagement and adherence, multiple supportive strategies were implemented. Participants received regular prompts and reminders through the mobile application to encourage consistent data entry and engagement. For individuals with low adherence, the health coaching team conducted follow-up contacts, such as phone calls, to promote continued participation.

At program completion, participants underwent follow-up assessments equivalent to the preintervention examinations and received individualized reports summarizing their progress in clinical indicators. Aggregate results were shared with health coordinators and participants to inform ongoing health planning. On completing the entire process, participants received thank-you gifts as an additional incentive to encourage participation and retention (Figure 1).

Although pre- and postprogram assessments were conducted as part of the 12-week intervention, this study focused on its long-term longitudinal associations observed after program participation. Outcomes were assessed by comparing health examination data from 2019 (baseline) and 2023 (follow-up), enabling the analysis of sustained changes beyond the immediate intervention period.

The primary outcomes were 9 clinical indicators commonly associated with chronic disease risk and management: SBP, DBP, BMI, FBG, HbA_1c, triglycerides, total cholesterol, high-density lipoprotein (HDL) cholesterol, and low-density lipoprotein cholesterol. These biomarkers were selected based on both clinical judgment by specialists in family medicine and endocrinology and established clinical evidence, as they are essential indicators for HTN, DM, and obesity, as well as for overall cardiometabolic risk assessment and management [24-26].

In addition, short-term outcomes, including intervention-related survey responses and self-reported lifestyle behaviors assessed immediately before and after the 12-week intervention, were descriptively analyzed to complement the long-term findings. Satisfaction and perceived usefulness were assessed using 5 questionnaire items on a 5-point Likert scale, which were converted into a 100-point scale for analysis. Lifestyle behaviors were evaluated using self-reported questionnaires on smoking (average number of cigarettes per day), alcohol consumption (average number of drinks per day), and physical activity (total weekly exercise time).

To reduce confounding and improve comparability between the intervention and control groups, 1:2 PSM was performed using baseline data from 2019. Separate PSMs were estimated for each condition group based on their relevant covariates. The selection of matching variables was aligned with the real-world enrollment criteria of the intervention program, in which participants were identified based on key disease-specific clinical indicators. Age and sex were included as common covariates across all conditions. In addition, SBP, DBP, and use of antihypertensive medication were included for the HTN cohort; HbA_1c, FBG, and use of antidiabetic medication were included for the DM cohort; and BMI was included for the obesity cohort.

A caliper width of 0.2 SD of the logit of the propensity score was applied for matching. Covariate balance before and after matching was assessed using absolute standardized mean differences (ASDs), with values less than 0.1 indicating adequate balance.

To compare longitudinal changes in outcomes associated with program participation, linear mixed effects models were applied separately within each HTN, DM, and obesity cohort. Fixed effects included group (intervention vs control), time (baseline vs follow-up), and their interaction (group × time), while a random intercept was included for each participant. The models were adjusted for baseline values and covariates with residual imbalance after PSM (ASDs≥0.1) within each cohort. Estimated marginal means were calculated to obtain adjusted group-specific outcomes at each time point, and the primary inference was based on the group × time interaction term, representing differences in longitudinal changes between groups. As a supplementary analysis, conventional statistical tests, such as paired and independent comparisons based on group comparisons, were also performed. Statistical significance was determined by a 2-sided P<.05, and all analyses were performed using R software version 4.3.1 (R Core Team).

The prematching cohorts included 30 intervention and 111,482 control individuals for HTN, 28 intervention and 111,445 control individuals for DM, and 44 intervention and 111,381 control individuals for obesity. After 1:2 PSM, the final analytical cohorts comprised 30 intervention and 60 control individuals for HTN, 26 intervention and 52 control individuals for DM, and 44 intervention and 88 control individuals for obesity (Figure 2).

Table S1 in Multimedia Appendix 1 presents the covariate balance between the intervention and control groups before and after PSM across the HTN, DM, and obesity cohorts, focusing on the variables used for PSM. Prior to matching, there were substantial baseline imbalances in several key covariates between the groups, as evidenced by high ASDs, which indicate the degree of dissimilarity between groups. In this study, we considered an ASD of 0.1 or greater to indicate significant imbalance between groups.

In the HTN cohort, significant differences were observed in SBP and DBP, with ASDs of 0.755 and 1.013, respectively. These values reflect considerable baseline differences in BP control between the intervention and control groups. Similarly, the DM cohort showed ASDs with FBG and HbA_1c exceeding 1, highlighting a large difference in glycemic status prior to matching. The obesity cohort showed a slightly more modest imbalance, but BMI still had an ASD of 0.388, suggesting unequal body composition.

After applying 1:2 PSM, intervention participants were successfully matched to comparable controls from a large eligible pool, resulting in improved balance in key covariates. Notably, all ASDs were reduced below the threshold of 0.1 across all measured covariates. In many cases, the postmatching ASDs approached zero, reflecting near-perfect alignment of baseline characteristics between the groups.

Table 1 summarizes the baseline distribution of key clinical outcomes for the matched intervention and control groups across the HTN, DM, and obesity cohorts. Notably, SBP and DBP were well-aligned in both the HTN and obesity cohorts, with ASDs close to 0. In addition, FBG and HbA_1c were consistently balanced across all 3 cohorts, reflecting comparable glycemic profiles at baseline.

Despite improvements in key disease-specific indicators, several variables showed persistent imbalances across multiple cohorts. Total cholesterol exhibited significant differences between groups in all 3 cohorts (ASD=0.148 in HTN, 0.188 in DM, and 0.492 in obesity). BMI also demonstrated notable imbalances in both HTN (ASD=0.295) and DM (ASD=0.690) cohorts.

While PSM successfully achieved balance in key disease-specific indicators within each cohort, residual differences in certain baseline characteristics, particularly lipid-related and anthropometric variables, persisted across cohorts. To account for these residual imbalances, variables with an ASD≥0.1 were included as covariates in the subsequent linear mixed effects models.

Table 2 presents longitudinal changes in key clinical indicators based on linear mixed effects models using estimated marginal means across the matched HTN, DM, and obesity cohorts. Baseline values were well balanced between the intervention and control groups across all outcomes.

In the HTN cohort, SBP levels were comparable between groups at both baseline (2019) and follow-up (2023), and the group × time interaction was not statistically significant (P=.41), indicating no differential change over time. In contrast, DBP was significantly lower in the intervention group compared to the control group at follow-up (76.61 vs 79.31 mm Hg; P=.045); however, the interaction term was not significant (P=.13), suggesting that although a between-group difference was observed in 2023, the overall longitudinal change pattern did not differ significantly between groups.

In the DM cohort, SBP was significantly lower in the intervention group compared to the control group at follow-up (113.42 vs 117.68 mm Hg; P=.03), although the interaction term was not statistically significant (P=.10), indicating a trend without a significant difference in change patterns. Notably, DBP showed both a significant between-group difference at follow-up (71.71 vs 76.38 mm Hg; P<.001) and a significant group × time interaction (P=.02), with a decrease over time observed in the intervention group. Other glycemic indicators, including FBG and HbA_1c, did not show significant differences in either between-group comparisons or interaction terms.

In the obesity cohort, triglycerides showed a significant between-group difference at follow-up (117.81 vs 157.74 mg/dL; P<.001), along with a significant interaction (P<.001), suggesting a favorable change pattern associated with the intervention. Similarly, HDL cholesterol was significantly higher in the intervention group at follow-up (59.96 vs 54.88 mg/dL; P<.001), with a significant interaction term (P<.001), suggesting a favorable change pattern associated with the intervention.

Analyses based on observed (unadjusted) values, including within- and between-group comparisons, yielded generally consistent patterns and are presented in Table S2 in Multimedia Appendix 1.

Survey responses collected immediately after the intervention indicated high levels of satisfaction and perceived usefulness of the program, with mean scores of 83.8 and 83.7, respectively.

Consistent with these findings, self-reported lifestyle behaviors assessed immediately before and after the intervention showed changes across multiple domains. In the HTN, DM, and obesity cohorts, some participants reported reductions in cigarette consumption after the intervention. Alcohol intake demonstrated mixed patterns, with a greater proportion of participants reporting decreases rather than increases. Physical activity distributions shifted toward higher activity categories, with an increased number of participants achieving ≥180 minutes per week compared to preintervention levels (Table S3 in Multimedia Appendix 1).

This study provides preliminary real-world evidence suggesting long-term associations between participation in a relatively short-term digital health intervention delivered as part of an EAP and several cardiometabolic indicators across multiple chronic disease risk groups over 4 years of follow-up. The intervention group was associated with favorable changes in several cardiometabolic indicators, particularly in DBP, triglycerides, and HDL cholesterol, based on adjusted longitudinal analyses. Although some statistically significant changes were modest in magnitude, even small changes in these clinically relevant cardiometabolic markers may be meaningful at the population level [27-29].

In the HTN cohort, while cross-sectional differences in DBP were observed at follow-up, there was insufficient evidence of significantly different DBP trajectories over time between the groups. In contrast, in the DM cohort, DBP showed both a significant between-group difference at follow-up and a significant group × time interaction, indicating different longitudinal DBP trajectories between the groups. These findings are broadly consistent with previous studies reporting associations between mHealth-based interventions and BP control over time [30,31].

In the obesity cohort, triglycerides and HDL cholesterol showed the most pronounced favorable changes in the intervention group, with both significant between-group differences at follow-up and longitudinal change patterns. These observed lipid changes are clinically relevant, as triglycerides and HDL cholesterol are established markers associated with cardiovascular risk [32,33].

Although this study did not directly evaluate behavioral or engagement mechanisms, several hypothesized pathways may help explain the observed associations. Participants reported high levels of satisfaction and perceived usefulness during the intervention period, along with favorable short-term changes in lifestyle behaviors. Previous studies have suggested that mHealth interventions combining real-time self-monitoring, personalized feedback, and behavioral coaching may support long-term adherence and lifestyle modification [34-38]. However, because detailed app usage metrics and longitudinal behavioral adherence data were not systematically collected in this study, these proposed mechanisms should be considered exploratory and were not directly tested in this analysis.

From a health care system perspective, the findings suggest that workplace-based mHealth programs may serve as a scalable complement to existing chronic disease management infrastructure. In countries with established national health checkup systems, integrating digital support into employee health programs may help strengthen long-term follow-up and secondary prevention after health screenings [13,18,21]. In addition, mHealth platforms may help reduce logistical and psychological barriers to ongoing disease management by providing continuous support integrated into daily routines, rather than relying solely on episodic clinic-based care [39]. These findings are also consistent with growing international interest in digital therapeutics and reimbursement models for mHealth-based prevention strategies targeting HTN, DM, and obesity [1,40,41]. In occupational settings, digital interventions embedded within EAPs may improve accessibility and continuity of care for working populations with limited opportunities for regular in-person follow-ups [2,10,12]. Collectively, these findings support the potential role of workplace-integrated digital health programs in chronic disease prevention and management.

Several limitations should be acknowledged. First, the retrospective nature of the analysis and the limited program cycle resulted in a relatively small sample size with complete follow-up data in 2023. This may reduce the statistical power of the study and increase the likelihood of type II errors, particularly for outcomes with relatively small effect sizes or nonsignificant longitudinal trends. Therefore, the findings should be interpreted cautiously, and the observed associations require further validation in larger prospective studies. Second, selection bias cannot be completely ruled out, as participants may have been more motivated or health-literate than the general population. Although PSM and linear mixed effects models were used to balance clinical variables, unmeasured confounders such as health behaviors, income, or education level may still have influenced the results. In addition, several baseline variables, including BMI and lipid-related markers, remained partially imbalanced after matching. Although these variables were additionally adjusted for in the longitudinal models, residual imbalance may still limit the strength of causal interpretation. Third, the absence of intermediate assessments between baseline (2019) and follow-up (2023) limits our ability to assess the trajectory of behavioral maintenance or identify when improvements plateaued or declined. As a result, the observed long-term outcomes should be interpreted as associations rather than direct causal effects of the intervention. Additionally, although short-term survey-based assessments provided some indication of behavior changes following the intervention, the absence of detailed and systematically collected app usage metrics (eg, log-ins, device adherence, coaching frequency) precluded detailed dose-response analyses of the underlying mechanisms. Finally, the generalizability of the findings may be limited, as the study population was derived from a single corporate system in Korea and consisted predominantly of male office workers.

Based on the findings and limitations of this study, future research should investigate the active elements of digital interventions—such as feedback frequency, personalization of advice, accountability of self-tracking, or credibility of professional coaching—that most effectively drive long-term behavior change. Furthermore, prospective studies with larger, more diverse cohorts are needed to generalize these results and investigate the compounding effects of cumulative interventions. Linking digital intervention data with claims databases or electronic health records would also enable assessments of downstream health care utilization and cost-effectiveness. Additionally, incorporating adaptive AI-based personalization, nudge algorithms, or gamification could enhance engagement, particularly among younger or more tech-savvy populations. Ultimately, future mHealth programs should aim to create hybrid models that combine medical, behavioral, and digital precision.