Association of a Health Belief Model-Based Digital Health Intervention with Glycolipid Metabolism and Health Behaviors in Older Adults with Type 2 Diabetes: A Quasi-Experimental Before–After Study

Abstract

Purpose

Integrated management of blood glucose and lipids reduces complications in type 2 diabetes mellitus (T2DM). However, few community-based programs combine mobile health (mHealth) technology, Health Belief Model (HBM)-based education, behavioral strategies, and family support within one framework, and evidence in Chinese community settings is limited. We examined whether a 12-month HBM-based intervention was associated with changes in glycolipid metabolism and health behaviors among community-dwelling older adults with T2DM in Tianjin, China.

Patients and Methods

This quasi-experimental, single-arm, uncontrolled before–after (pre–post) study included 114 T2DM patients (mean age 72.25 ± 5.63 years) managed at a community health service center; no parallel control group was enrolled. Paired samples t-tests and McNemar tests assessed changes in metabolic parameters and categorical health behaviors, respectively.

Results

All glucose and lipid parameters improved (all P < 0.0001). The primary outcomes changed as follows: HbA1c −0.63% points (95% CI −0.80 to −0.46), 2-hour postprandial glucose −5.16 mmol/L (95% CI −5.54 to −4.78), and triglycerides −2.61 mmol/L (95% CI −2.82 to −2.40). Fasting plasma glucose, total cholesterol, and LDL-cholesterol also decreased (Tables 3 and 4). Medication adherence rose from 73.68% to 86.84% (P = 0.013), disease knowledge from 50.88% to 76.32%, and regular exercise from 28.95% to 61.40% (both P < 0.0001); smoking and alcohol use were unchanged.

Conclusion

The HBM-based digital health intervention was associated with improved glycolipid metabolism and health behaviors in this cohort. Because the single-arm design lacked a control group and concurrent changes in glucose- and lipid-lowering medications were not recorded, all metabolic improvements — and the large triglyceride reduction in particular — may partly reflect pharmacological intensification; randomized controlled trials are needed to confirm these findings.

Introduction

Diabetes mellitus is one of the most pressing global public health challenges of the twenty-first century. An estimated 537 million adults aged 20–79 years were living with diabetes worldwide in 2021, and this figure is projected to reach 783 million by 2045, with the steepest relative increases in low- and middle-income countries.¹ China bears a disproportionate share of this burden: the absolute number of diabetes cases has continued to rise over the past three decades and is expected to grow further through 2050, driven largely by population aging and lifestyle transitions.² Type 2 diabetes mellitus (T2DM) accounts for the majority of these cases and places a substantial burden on healthcare systems through its chronic complications and associated economic costs.^3,4 Because the micro- and macrovascular complications of T2DM arise from joint dysregulation of glycemic, lipid, and blood pressure parameters, professional societies now recommend integrated management of these cardiometabolic risk factors rather than a narrow focus on glycemic control alone.^5–7

Glycemic control and lipid metabolism are closely intertwined in T2DM. Alzahrani et al⁸ reported that HbA1c levels correlated positively with triglycerides and total cholesterol, indicating that poor glycemic control exacerbates dyslipidemia. Glycemic variability itself is also an independent risk factor for several chronic complications in T2DM,³ reinforcing the case for integrated glycolipid management.⁴ Community-based lifestyle interventions have been central to this integrated approach. A systematic review of randomized trials by Haw et al⁹ found sustained reductions in diabetes incidence after lifestyle programs, although effect sizes attenuated once active support ended. Objectively measured physical activity is independently associated with glycemic control, adiposity, and overall cardiometabolic risk in T2DM.¹⁰ Yet community programs targeting physical inactivity, poor nutrition, and medication non-adherence^11,12 face real-world barriers: funding and infrastructure demands,¹² difficulty tailoring content across cultural contexts,¹³ and limited sustainability once external support is withdrawn.^9,12

Behavioral frameworks offer a way to embed durable self-management into such programs. Cognitive behavioral therapy (CBT) helps patients identify and modify the maladaptive thoughts and behaviors that hinder self-management. In T2DM, CBT improves glycemic indices and reduces psychological distress.^11,14 The Health Belief Model (HBM) complements CBT by mapping behavior change onto perceived susceptibility, severity, benefits, barriers, cues to action, and self-efficacy.¹⁵ Patient perceptions framed by the HBM explain a substantial share of the variance in diabetes self-care.¹⁶ In Chinese T2DM populations, HBM-related beliefs mediate the link between diabetes knowledge and self-management behaviors.^17,18 Interventions that strengthen self-efficacy, a core HBM construct, improve self-management behavior.¹⁹ Trials in Malaysia, Iran, and elsewhere show that HBM- and CBT-informed interventions enhance metabolic control and promote healthier lifestyle choices.^20–22 Family-based components that mobilize social support further improve treatment adherence and quality of life, including in Chinese families caring for elderly relatives with T2DM.^22–24 Perceived social support independently predicts diabetes self-management behavior.²⁵ Family function shapes self-management in part through diabetes distress and self-efficacy.²⁶

Mobile health (mHealth) technologies have opened new channels for delivering these behavioral frameworks at scale, supporting continuous monitoring, personalized feedback, and remote follow-up.^27–29 Digital health interventions have been associated with improvements in glycated hemoglobin (HbA1c), lipid profiles, and health behaviors in T2DM.^30–32 Application- and wearable-based programs report comparable gains in glycemic control and patient-centered self-management.^33–35 Systematic reviews document an expanding role for these technologies in both primary prevention and ongoing management.³⁶ Low-cost messaging platforms extend structured diabetes education to resource-limited settings.^37,38 Pragmatic trials in Chinese primary care likewise show that multicomponent, mHealth-enabled management systems can improve T2DM outcomes at the population level.^39,40

Three gaps remain. First, behavioral theory, mHealth tools, and family support have rarely been integrated within a single intervention, and most existing studies examine these components in isolation.^41,42 Second, evidence from community-dwelling older adults in China is limited, even though this is the population the community health service system is chiefly designed to serve. Third, few studies examine glycemic outcomes, lipid outcomes, and self-management behaviors together, so it remains unclear whether metabolic and behavioral gains move in parallel. Previous digital diabetes-management programs in Chinese primary care have been predominantly single-component, comprising glucose telemonitoring or education alone; the present program integrates four components within one theory-driven framework and additionally engages family caregivers.

Rapid population aging in China has accelerated T2DM prevalence and placed growing demands on the community health service system.^2,6 Urban community health centers are therefore a practical delivery setting for structured chronic-disease intervention programs.⁴³ We embedded a theory-driven digital program that also engaged family caregivers within routine community care, using the HBM to structure education, CBT-informed strategies to translate health beliefs into behavior, a mobile application and a wearable device for continuous engagement, and family participation to sustain it. Using a quasi-experimental before–after design, we evaluated this 12-month program in community-dwelling older adults with T2DM at a community health service center in Tianjin, China, with two aims: first, to assess whether the program was associated with coordinated improvement in glucose and lipid metabolism; and second, to assess whether it was associated with parallel gains in self-management behaviors, including medication adherence, disease knowledge, exercise participation, and family supportive behaviors. Because the program was delivered under real-world community conditions, we report metabolic and behavioral outcomes together to provide feasibility evidence and effect-size estimates for a future controlled trial rather than a test of efficacy.

Materials and Methods

Participants

We conducted this quasi-experimental, single-arm before-after (pre-post) intervention study at Xingnan Street Community Health Service Center, Nankai District, Tianjin, China. Between January 2022 and December 2023, we enrolled 114 patients with T2DM (58 men and 56 women) through random sampling from the community health management registry. Each patient served as their own control: baseline data collected in 2022 (pre-intervention) were compared with data collected at the end of the 12-month HBM-based intervention program in 2023 (post-intervention).

All patients met the diagnostic criteria for T2DM established by the World Health Organization (1999) and confirmed according to the Chinese Guidelines for the Prevention and Control of Type 2 Diabetes Mellitus (2022 edition).⁴⁴ Patients were eligible if they had a confirmed diagnosis of T2DM, were at least 18 years of age, had resided in the Xingnan Street community for at least two years, were literate at or above the junior high school level, used a smartphone with internet access regularly, and provided written informed consent. Patients were excluded if they had severe mental illness, cognitive impairment, or a language disorder, or were unable to participate in health education activities. No participants with cognitive impairment were enrolled, and all participants provided written informed consent personally. Eligible registrants in the community chronic-disease (diabetes) health-management registry were ordered by registry identification number, and systematic sampling with a random starting point was applied; the sampling interval was determined by the ratio of eligible registrants to the target sample size. No a priori power calculation was performed; the sample size was determined by the number of eligible registrants accessible within the recruitment window. Although the approved protocol set a minimum age of 18 years, the sampling frame was the community’s elderly chronic-disease health-management registry, which enrolls older residents, so the realized cohort consisted of older adults (mean age 72.25 ± 5.63 years) as a direct consequence of the sampling frame. All 114 sampled participants completed both assessment waves, with no loss to follow-up and no missing outcome data.

Ethics Statement

The study was conducted in accordance with the Declaration of Helsinki and adhered to China’s “Measures for Ethical Review of Life Science and Medical Research Involving Humans” as well as international ethical standards. The study protocol was reviewed and approved prior to initiation by the Ethics Committee of Xingnan Street Community Health Service Center, Nankai District, Tianjin (approval number: XNWS01005; approved on December 10, 2021). Written informed consent was obtained from all participants after the study purpose, procedures, risks, and privacy measures were explained; for elderly participants with limited literacy, simplified language and verbal explanations were provided. Because cognitive impairment was an exclusion criterion, no participant was enrolled through proxy or guardian consent. All personal identifiers were anonymized during data collection, and research data were stored in encrypted databases with restricted access in compliance with China’s Personal Information Protection Law. Participants were informed of their right to withdraw at any time without penalty.

Data Collection

Demographic and clinical information was collected using a standardized self-administered questionnaire that patients completed independently or with assistance from a trained investigator. The questionnaire covered demographic details, medical history, lifestyle habits, comorbid conditions, medication use, and participation in health education activities.

For laboratory assessments, patients fasted from 22:00 on the evening before blood collection. Venous blood samples were drawn the following morning and analyzed for fasting plasma glucose (FPG), total cholesterol (TC), triglycerides (TG), high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), alanine aminotransferase (ALT), aspartate aminotransferase (AST), creatinine (Cr), blood urea nitrogen (BUN), and uric acid (UA). HbA1c was measured separately. 2-hour postprandial glucose (2hPG) was measured after a standard breakfast.

All data collection procedures followed a written standard operating protocol to minimize observer bias. Patient records were anonymized with coded identifiers to protect confidentiality. Data were accessed for research analysis purposes after completion of the 12-month intervention period. Authors did not have access to any information that could identify individual participants during or after data collection.

Interventions

The intervention used the Health Belief Model (HBM) as its theoretical framework and delivered its content through mHealth technology and family-based social support, with the goal of improving self-management capacity and treatment adherence. The content was organized along the six HBM constructs: perceived susceptibility, perceived severity, perceived benefits, perceived barriers, cues to action, and self-efficacy. It moved from threat appraisal and outcome evaluation through self-efficacy building to behavioral activation. The program comprised one group lecture per week (45 minutes, delivered by community health center physicians) and two individualized sessions per week (15–20 minutes each, delivered by the responsible nurses), corresponding to a planned total of approximately 52 group lectures and 104 individualized sessions per participant over the 12-month period. Family doctor teams recorded each patient’s progress and ran a mutual-support group in which patients exchanged experiences and strategies. Extended operational details of the digital components are provided in Supplementary Material S1.

Perceived Susceptibility

A custom WeChat mini-program (“Diabetes Risk Calculator”) generated an individualized 10-year complication probability curve based on the UKPDS risk model, incorporating each patient’s body mass index, family history, and lifestyle parameters. Patients whose records indicated inadequate progress were flagged for a home visit by the family doctor and were assigned by the responsible nurse to a higher-intensity tier of the intervention.

Perceived Severity

Short virtual-reality scenarios illustrated the natural history of diabetic foot ulcers, showing progression from peripheral neuropathy to amputation alongside an animation of the underlying molecular mechanism and the associated tissue changes. A smart wristband visualized the related data for each patient. Patients also viewed video testimonials from people living with diabetic retinopathy, including scenes of cooking with impaired vision, paired with timeline comparisons of 10-year outcomes between patients with and without glycemic control.

Perceived Benefits

The recommended 30 minutes of daily exercise was split into three 10-minute brisk walks, and the WeChat Exercise function displayed the corresponding calorie expenditure in real time. Patients who reached predefined activity targets unlocked a “Metabolic Pioneer” achievement badge as reinforcement. A meal-photo recognition feature delivered glycemic-load (GL) feedback and linked food substitutions to an HbA1c prediction model, so that patients could see the expected long-term effect of daily food choices.

Perceived Barriers

For patients with a General Self-Efficacy Scale (GSES) score below 20, we launched a “5% Improvement Program” that began with keeping a simple food diary in week 1 and adding one vegetable serving in week 2, lowering the threshold for behavioral activation through small, incremental goals. A “Sugar Control Alliance” function was integrated into the WeChat step-count ranking: patients reaching the required step total could form teams to redeem prizes. In addition, following the National Basic Public Health Service Specification, we measured FPG, 2hPG, HbA1c, and body mass index (BMI) quarterly. The weekly group lectures described above covered diabetes self-management, dietary principles, and the role of exercise (Table 1).

Table 1.

Evidence-Based Health Education Curriculum Revised Based on ADA Standards for Diabetes Education

Module	Topics Covered in the Educational Curriculum	Evidence Base
Medication	Insulin injection techniques, oral medication time-window management	IDF Global Guideline for Managing Older People with Type 2 Diabetes
Dietary management	Food Glycemic Index (GI) calculation, plate allocation methods	ESPEN guidance on nutrition in older adults
Exercise prescription	Target heart-rate monitoring, resistance-training periodization program	ACSM physical activity guidance
Chinese medicine intervention	Foot Sanli–Sanyinjiao acupoint moxibustion	China Diabetes Prevention Guidelines
Psychological support	Mindfulness-Based Stress Reduction (MBSR), cognitive behavioral therapy	ADA Standards of Care in Diabetes

Notes: Table 1 lists the topics covered by the educational curriculum; not all topics were delivered as separately scheduled activities.

Abbreviations: ACSM, American College of Sports Medicine; ADA, American Diabetes Association; ESPEN, European Society for Clinical Nutrition and Metabolism; GI, glycemic index; IDF, International Diabetes Federation; MBSR, Mindfulness-Based Stress Reduction.

Evaluation Indicators and Determination Criteria

We assessed three groups of outcomes: metabolic parameters, cardiovascular parameters, and self-reported health behaviors. Basic demographic and anthropometric variables included age, sex, height, weight, and BMI. The glucose metabolism indicators were FPG, 2hPG, and HbA1c, reflecting short-term and long-term glycemic control. The lipid metabolism indicators were TG, TC, and LDL-C, given their established association with cardiovascular risk in T2DM.

Because hypertension commonly coexists with T2DM and modulates cardiometabolic risk, we also recorded systolic blood pressure (SBP) and diastolic blood pressure (DBP). Health behaviors were captured as categorical indicators and included medication adherence, disease knowledge mastery, regular exercise, smoking, alcohol consumption, periodic follow-up visits, and family caring behaviors. Two behavioral indicators were classified against predefined criteria: regular medication use was classified as present if the participant reported two or fewer missed doses in the preceding month, and regular exercise was classified as present if the participant reported at least three days per week of moderate-intensity activity lasting at least 30 minutes per session. Disease knowledge mastery and family caring behaviors were assessed by trained interviewers using a program-developed structured interview and were classified on the basis of the interviewer’s overall judgment of the participant’s responses rather than a predefined numerical threshold; the instrument was not formally validated. Taken together, these metabolic, cardiovascular, and behavioral indicators offer a profile that links the biochemical changes observed during the intervention to the behavioral changes that may underlie them.

Statistical Analysis

We performed all statistical analyses using SPSS 27.0 (IBM Corp., Armonk, NY, USA). The study used a within-subjects, pre-post design in which each of the 114 participants contributed paired measurements from 2022 (pre-intervention) and 2023 (post-intervention). Paired samples t-tests were applied to the six metabolic outcomes (FPG, 2hPG, HbA1c, TG, TC, and LDL-C), and McNemar’s test was applied to each binary behavioral indicator. Categorical variables were expressed as frequencies and percentages (n (%)) with Wilson 95% confidence intervals. For the multi-category variable of daily exercise duration, a summary contrast (any regular exercise vs none) is reported as the principal comparison, with each duration category additionally analyzed dichotomously (category present vs absent) using McNemar’s test; these duration-specific comparisons are exploratory. Continuous variables were expressed as mean ± standard deviation (M ± SD). We assessed the normality of continuous variables using the Shapiro–Wilk test and inspection of histograms; normally distributed variables were compared using paired samples t-tests (two-sided), and non-normally distributed variables using the Wilcoxon signed-rank test. No outcome data were missing: all 114 participants contributed complete paired measurements for every reported variable, so no imputation or available-case exclusion was required. Confidence intervals for mean changes were computed from the summary statistics under the conservative assumption of a non-negative pre–post correlation, an approach that if anything overstates interval width; this is stated here for transparency. No formal correction for multiple comparisons was applied, given the exploratory, hypothesis-generating nature of the secondary behavioral outcomes; P-values for secondary outcomes should therefore be interpreted as descriptive. All tests were two-sided, and P < 0.05 was considered statistically significant.

Results

General Demographic Description

At baseline, the 114 patients had a mean age of 72.25 ± 5.63 years, a mean diabetes duration of 15.66 ± 7.21 years, a mean BMI of 24.14 ± 2.36 kg/m², and a mean weight of 67.09 ± 9.12 kg; coronary heart disease was present in 42.11% of patients, hypertension in 36.84%, and cerebrovascular disease in 9.65% (Table 2).

Table 2.

Baseline Characteristics of the Study Participants (n = 114)

Variable	Value
Age (years)	72.25 ± 5.63
Duration of diabetes (years)	15.66 ± 7.21
Height (cm)	166.54 ± 7.50
Weight (kg)	67.09 ± 9.12
BMI (kg/m2)	24.14 ± 2.36
Hemoglobin (g/L)	136.91 ± 10.00
SBP (mmHg)	128.87 ± 8.36
DBP (mmHg)	81.85 ± 6.93
Coronary heart disease, n (%)	48 (42.11)
Hypertension, n (%)	42 (36.84)
Cerebrovascular disease, n (%)	11 (9.65)

Notes: Continuous variables are presented as mean ± SD; categorical variables as n (%).

Abbreviations: BMI, body mass index; SBP, systolic blood pressure; DBP, diastolic blood pressure; SD, standard deviation.

Changes in Glucose Metabolism Indicators

All three glucose metabolism indicators decreased over the 12-month intervention (Table 3). Fasting plasma glucose fell from 8.23 ± 0.82 to 7.43 ± 0.94 mmol/L, 2-hour postprandial glucose from 16.06 ± 1.50 to 10.90 ± 1.42 mmol/L, and HbA1c from 7.99 ± 0.60% to 7.36 ± 0.69% (all P < 0.0001). The mean HbA1c reduction was 0.63% points.

Table 3.

Changes in Patients’ Glucose Metabolism Indices Before and After Intervention

Variable	Pre-Intervention	Post-Intervention	Mean Change (95% CI)	Cohen’s d	P-value
Fasting plasma glucose (mmol/L)	8.23 ± 0.82	7.43 ± 0.94	−0.80 (−1.03, −0.57)	0.91	<0.0001
2-hour postprandial glucose (mmol/L)	16.06 ± 1.50	10.90 ± 1.42	−5.16 (−5.54, −4.78)	3.53	<0.0001
HbA1c (%)	7.99 ± 0.60	7.36 ± 0.69	−0.63 (−0.80, −0.46)	0.97	<0.0001

Notes: Data are mean ± SD; P-values are from paired samples t-tests. Confidence intervals for mean changes were computed from summary statistics under a conservative assumption of non-negative pre–post correlation.

Abbreviations: HbA1c, glycated hemoglobin.

Figure 1 shows the individual-level changes in HbA1c (Figure 1A) and FPG (Figure 1B). The downward slope of the connecting lines is visible in the large majority of the 114 patients rather than a subset of high responders.

Figure 1.

Individual-level changes in glycated hemoglobin (HbA1c) and fasting plasma glucose (FPG) before and after the 12-month intervention. (A) HbA1c (%). (B) FPG (mmol/L). Each thin gray line connects the pre- and post-intervention values of one patient (n = 114); blue dots represent pre-intervention values and orange dots post-intervention values. Black horizontal bars indicate group means. Paired samples t-test; ****P < 0.0001 refers to the overall paired pre–post comparison within each panel.

Changes in Lipid Metabolism Indicators

Lipid parameters also changed markedly over the intervention period (Table 4). Triglycerides fell from 4.12 ± 0.98 to 1.51 ± 0.58 mmol/L (a 63.3% reduction), total cholesterol from 6.20 ± 0.73 to 4.68 ± 0.76 mmol/L (24.6%), and LDL-cholesterol from 3.60 ± 0.77 to 3.03 ± 0.77 mmol/L (15.8%); all P < 0.0001. The triglyceride reduction was the most pronounced; the total and LDL-cholesterol changes were more modest. Given the magnitude of the triglyceride reduction and the absence of systematic medication tracking, this finding should be interpreted with caution (see Discussion).

Table 4.

Control of Lipid Metabolism in Patients Before and After Intervention

Variable	Pre-Intervention	Post-Intervention	Mean Change (95% CI)	Cohen’s d	P-value
Triglycerides (mmol/L)	4.12 ± 0.98	1.51 ± 0.58	−2.61 (−2.82, −2.40)	3.24	<0.0001
Total cholesterol (mmol/L)	6.20 ± 0.73	4.68 ± 0.76	−1.52 (−1.72, −1.32)	2.04	<0.0001
LDL-cholesterol (mmol/L)	3.60 ± 0.77	3.03 ± 0.77	−0.57 (−0.77, −0.37)	0.74	<0.0001

Notes: Data are mean ± SD; P-values are from paired samples t-tests. Confidence intervals for mean changes were computed from summary statistics under a conservative assumption of non-negative pre–post correlation.

Abbreviations: LDL, low-density lipoprotein.

Figure 2 presents the individual-level trajectories for each lipid parameter (Figure 2A, triglycerides; Figure 2B, total cholesterol; Figure 2C, LDL-cholesterol). The downward slope is present in the majority of the 114 patients for all three lipids, as it was for the glycemic indicators. The reduction was largest for triglycerides (panel A), corresponding to the group-level decrease of 63.3%, whereas total cholesterol (panel B) and LDL-cholesterol (panel C) showed smaller reductions.

Figure 2.

Individual-level changes in lipid metabolism parameters before and after the 12-month intervention. (A) Triglycerides (TG, mmol/L). (B) Total cholesterol (TC, mmol/L). (C) LDL-cholesterol (LDL-C, mmol/L). Each thin gray line connects one patient’s pre- and post-intervention values (n = 114); blue dots, pre-intervention; orange dots, post-intervention. Black horizontal bars indicate group means. Paired samples t-test; ****P < 0.0001 refers to the overall paired pre–post comparison within each panel.

Changes in Health Behaviors

After the 12-month standardized health education program (one group lecture and two individualized sessions per week), patients’ disease knowledge mastery rose from 50.88% to 76.32% (P < 0.0001), and family caring behaviors rose from 57.02% to 78.07% (P = 0.0007) (Table 5). Both indicators therefore increased over the intervention period.

Table 5.

Health Behaviors and Self-Management Before and After the Intervention

Variable	Pre-Intervention, n (%) [95% CI]	Post-Intervention, n (%) [95% CI]	P-value
Disease knowledge mastery	58 (50.88) [41.8–59.9]	87 (76.32) [67.7–83.2]	<0.0001
Family caring behaviors	65 (57.02) [47.8–65.7]	89 (78.07) [69.6–84.7]	0.0007
Regular medication use	84 (73.68) [64.9–80.9]	99 (86.84) [79.4–91.9]	0.013
Periodic follow-up visits	83 (72.81) [64.0–80.1]	94 (82.46) [74.4–88.3]	0.080
Smoking	17 (14.91) [9.5–22.6]	15 (13.16) [8.1–20.6]	0.226
Alcohol use	23 (20.18) [13.8–28.5]	20 (17.54) [11.7–25.6]	0.662
Any regular exercise	33 (28.95) [21.4–37.9]	70 (61.40) [52.2–69.8]	<0.0001
Exercise: none	81 (71.05) [62.1–78.6]	44 (38.60) [30.2–47.8]	<0.0001
Exercise: 0.5 h/day	1 (0.88) [0.2–4.8]	4 (3.51) [1.4–8.7]	0.175
Exercise: 1.0 h/day	11 (9.65) [5.5–16.5]	41 (35.96) [27.7–45.1]	<0.0001
Exercise: 1.5 h/day	14 (12.28) [7.5–19.6]	15 (13.16) [8.1–20.6]	0.842
Exercise: 2.0 h/day	7 (6.14) [3.0–12.1]	10 (8.77) [4.8–15.4]	0.449

Notes: Data are n (%); P-values are from McNemar tests for paired binary outcomes. Wilson 95% confidence intervals are given in brackets. Duration-specific comparisons are exploratory.

Medication-taking and follow-up behaviors showed differing responses to the intervention (Table 5). The rate of regular medication use rose from 73.68% to 86.84% (P = 0.013). The rate of periodic follow-up visits increased from 72.81% to 82.46% but did not reach statistical significance (P = 0.080).

Changes in smoking, alcohol use, and exercise habits are summarized in Table 5. Smoking and alcohol consumption showed small decreases that did not reach statistical significance: smoking fell from 14.91% to 13.16% (P = 0.226), and alcohol consumption from 20.18% to 17.54% (P = 0.662). Exercise habits changed substantially. The proportion of patients reporting no regular exercise fell from 71.05% to 38.60% (P < 0.0001), and the proportion exercising for one hour per day rose from 9.65% to 35.96% (P < 0.0001). A summary contrast of any regular exercise vs none is reported in Table 5.

Across the seven health behavior indicators (Table 5), the largest absolute gain was in regular exercise participation (+32.5% points), followed by disease knowledge mastery (+25.4), family caring behaviors (+21.1), and medication adherence (+13.2). Smoking (−1.8) and alcohol consumption (−2.6) showed small, non-significant changes. Periodic follow-up visits showed an intermediate pattern (+9.6% points, P = 0.080). No adverse events or unintended effects related to the intervention were observed or reported during the 12-month program.

Discussion

In this 12-month quasi-experimental, single-arm before-after (pre-post) study, a Health Belief Model (HBM)-based digital health intervention delivered through a Chinese urban community health service center was associated with improvements in both glycolipid metabolism and several self-reported health behaviors in 114 older adults with T2DM. To our knowledge, few previous studies have evaluated a single intervention framework that simultaneously integrates mHealth technology, HBM-based health education, cognitive behavioral strategies, and family support for community-dwelling older adults in the Chinese primary care setting. The main finding is consistent across glycemic, lipid, and behavioral domains: patients showed measurable improvements in FPG, 2hPG, HbA1c, TG, TC, and LDL-C, alongside gains in medication adherence, disease knowledge, regular exercise, and family caring behaviors.

Changes in Glucose Metabolism Indicators

The mean HbA1c reduction of 0.63% points falls within the range reported for comparable community-based digital interventions. A meta-analysis by Zhang et al⁴⁵ reported reductions of 0.40% to 0.80% across telemedicine-based primary care interventions. Martin et al³² further showed that digital health coaching yields larger HbA1c gains in patients with higher baseline values, which matches our finding in a cohort with a pre-intervention mean HbA1c of 7.99%. The recent lifestyle-intervention trial by Kitazawa et al,⁴⁶ which combined a smartphone application with intermittently scanned continuous glucose monitoring in adults at high risk for T2DM, reported improvements in time in range and body weight without meaningful HbA1c change, underscoring that glycemic benefits vary with baseline disease status and intervention design. Taken together, our observed HbA1c reduction is plausible in magnitude and consistent with the broader digital health literature.

The 32% reduction in 2hPG was more pronounced than the FPG reduction and deserves separate interpretation. The meal photo recognition function in the intervention gave patients immediate visual feedback on the glycemic load of their meals, and the three 10-minute brisk walks distributed through the day targeted postprandial excursions specifically. We hypothesize that both components contributed to improved postprandial glucose handling, and the pattern we observed (a larger 2hPG change than FPG change) is consistent with the mechanistic expectations of dietary and activity-level shifts acting on meal-related glucose spikes. Such attributions cannot be established in a single-arm design and remain hypotheses. Participants with the highest baseline HbA1c (above 9%) showed the steepest individual slopes in Figure 1, which accords with the greater responsiveness of poorly controlled patients reported in the digital health literature.^28,30

Changes in Lipid Metabolism Indicators

The most pronounced lipid change was the 63.3% fall in triglycerides; total cholesterol and LDL-cholesterol decreased more modestly, by 24.6% and 15.8%, respectively. The triglyceride reduction exceeds what is typically reported for lifestyle-only interventions, whereas the cholesterol changes are within a behaviorally achievable range. Chen et al³¹ reported more modest lipid improvements in a meta-analysis of health literacy interventions in diabetic patients, and Devaraj et al⁴⁷ found smaller lipid effects in a yearlong cardiovascular lifestyle program. The literature on the HbA1c–lipid relationship is itself mixed. Alzahrani et al⁸ reported positive correlations between HbA1c and TG and TC but no significant correlation with LDL-C or HDL-C. A meta-analysis of mHealth interventions in older adults with type 2 diabetes by Lee et al⁴⁸ reported pooled reductions of 0.24 percentage points in HbA1c and 0.09 mmol/L in triglycerides, with no significant change in LDL-cholesterol. These discrepancies likely reflect differences in baseline population characteristics, medication regimens, and duration of diabetes, and we interpret our lipid findings against this heterogeneous background rather than against a single benchmark.

Several factors may explain the lipid changes in our cohort. First, the baseline triglyceride level was markedly elevated (mean 4.12 mmol/L), which mechanically allows a greater absolute reduction than cohorts starting closer to target, whereas baseline LDL-cholesterol (3.60 mmol/L) was only modestly above target and changed correspondingly little. Second, the dietary module emphasized low-glycemic-index food selection and plate-allocation methods, giving patients a concrete framework for sustained dietary change, to which triglycerides are particularly responsive. Third, we cannot fully rule out concurrent pharmacological contributions, as lipid-lowering medication adjustments during the intervention period were not systematically recorded. We therefore interpret the triglyceride reduction in particular as reflecting a combination of behavioral change and possible medication adjustment, rather than behavioral change alone. Responses were also heterogeneous: a minority of participants showed increased LDL-cholesterol after the intervention, visible as upward-sloping trajectories in Figure 2C, underscoring variable individual responses and the need for individualized follow-up.

Even with this caveat, the concurrent improvement in both glucose and lipid metabolism carries clinical value. Chinese consensus statements on cardiovascular prevention in T2DM emphasize integrated glycolipid management to reduce cardiovascular risk,⁷ and programs that move multiple cardiometabolic parameters together, irrespective of whether the mechanism is lifestyle, medication, or both, are aligned with that goal.

Changes in Health Behaviors

Four health behaviors improved significantly: medication adherence, disease knowledge mastery, regular exercise participation, and family caring behaviors (Table 5). These behavioral gains plausibly underlie part of the observed metabolic improvement, given the established links between self-management behaviors and glycolipid control in T2DM.^31,49

Several design features may explain why these specific behaviors responded. The HBM-based content addressed perceived threat (via the Diabetes Risk Calculator and VR scenarios), perceived benefits (via the exercise medal system and GL feedback), and self-efficacy (via the graduated “5% Improvement Program”), in line with evidence that targeting multiple HBM constructs together outperforms single-construct interventions.¹⁵ The short service radius of the community health center (most patients within a 15-minute walk) likely reduced practical barriers to medication refill, a known predictor of adherence. The gamified WeChat features, such as step-count rankings and team-based rewards, may have sustained exercise motivation beyond the initial novelty period.

The gain in family caring behaviors deserves specific attention. Panagiotidis et al²³ showed that family-based health literacy interventions improve both patient outcomes and family engagement in T2DM, and Duarte-Diaz et al²² reported that patient empowerment was associated with reduced anxiety and depression symptoms. We speculate that our parallel gains in family support, disease knowledge, and adherence reflect a feedback loop between household-level behavior and individual-level motivation — a hypothesis that was not directly tested in this study and remains to be examined.²⁴

Not every behavior responded, however. Smoking fell from 14.91% to 13.16% (P = 0.226) and alcohol use from 20.18% to 17.54% (P = 0.662); neither change reached statistical significance. This pattern was expected and is consistent with the broader literature showing that addictive behaviors require targeted, high-intensity interventions, such as pharmacotherapy, structured counseling, or behavioral cessation programs, that were not part of the present intervention design.

Limitations

This study has several limitations. First, it was a quasi-experimental, single-arm before-after (pre-post) study without a parallel control group receiving usual care over the same calendar period. Consequently, the observed changes cannot be attributed to the intervention alone: secular trends in diabetes management, seasonal variation, regression to the mean, and concurrent medication adjustments all may have contributed. Randomized controlled trials with parallel control arms are needed to establish causality.

Second, the study was conducted at a single community health service center in Tianjin with 114 participants. The findings may not generalize to other regions of China, to rural or lower-resource primary care settings, or to community-dwelling older adults with different socioeconomic and cultural backgrounds. Multi-center replication with larger and more diverse samples is warranted.

Third, the intervention combined multiple active ingredients, including mHealth technology, HBM-based education, CBT-informed strategies, exercise prescription, dietary guidance, gamification, and family engagement. The present design cannot separate the individual contributions of these components, and dismantling trials or factorial designs would be needed to identify the elements that drive the observed effects.

Fourth, the large triglyceride reduction (a 63.3% decrease) is of a magnitude that can be difficult to attribute to lifestyle change alone. We did not systematically record concurrent glucose-lowering or lipid-lowering medication initiation, dose changes, or adherence during the intervention period, so a substantial portion of the metabolic improvement, most clearly the lipid effect but potentially also part of the glycemic effect, may reflect pharmacological intensification rather than behavioral change. Future studies should capture all concurrent medication changes to separate the two contributions.

Fifth, behavioral outcomes were captured with self-reported, categorical indicators (for example, whether a patient exercised regularly). Self-report is subject to recall bias and social-desirability bias, especially in a program where patients interact repeatedly with healthcare providers, and the binary or coarse exercise categories likely miss quantitatively important variation. Objective measures, such as step counters, continuous glucose monitoring, and pharmacy refill records, would strengthen future evaluations. In addition, two of the behavioral indicators — disease knowledge mastery and family caring behaviors — were classified without predefined operational criteria and using an instrument that was not formally validated, which introduces potential rater variability; the corresponding results should be regarded as indicative rather than precise estimates.

Sixth, follow-up was limited to the 12-month active intervention period. Whether the metabolic and behavioral gains persist once the structured program ends is unknown, and the long-term durability of such community-based mHealth interventions is a recognized gap in the field. Extended follow-up studies are needed to assess maintenance beyond the active phase. Seventh, the dichotomized, category-by-category testing of exercise duration is statistically suboptimal for ordered categories; together with the absence of a correction for multiple comparisons, the behavioral results should be regarded as exploratory. Eighth, post-intervention anthropometric and blood-pressure values were not systematically compiled, so change in body weight, BMI, and blood pressure over the study period could not be evaluated. Finally, because the study used a single-group design with no comparison condition, blinding of participants, intervention providers, or outcome assessors to study condition was not applicable.

Conclusions

A 12-month HBM-based digital health intervention delivered through a Chinese urban community health service center was associated with improvements in glycolipid metabolism and several self-reported health behaviors in 114 older adults with T2DM. Among the behavioral outcomes, regular exercise participation showed the largest absolute gain (+32.5% points), followed by disease knowledge mastery and family caring behaviors, whereas smoking and alcohol use did not change significantly. Because changes in glucose- and lipid-lowering medications were not systematically recorded, all metabolic improvements — and the triglyceride reduction in particular — should be interpreted as associations that may partly reflect concurrent pharmacological adjustment. The study’s distinct contribution lies in integrating mHealth tools, HBM-based education, cognitive behavioral strategies, and family support within a single community-delivered framework. These findings suggest that a theory-informed digital intervention can be delivered through routine community health center workflows with existing staff, supporting its potential as a scalable template for community-based diabetes management, pending confirmation in controlled studies. Randomized controlled trials with parallel control arms, systematic medication capture, and extended follow-up are needed to confirm the magnitude of effect, disentangle the contribution of individual intervention components, and establish long-term durability.

Funding Statement

This research received no external funding.

Abbreviations

T2DM, type 2 diabetes mellitus; HBM, Health Belief Model; CBT, cognitive behavioral therapy; mHealth, mobile health; FPG, fasting plasma glucose; 2hPG, 2-hour postprandial glucose; HbA1c, glycated hemoglobin; BMI, body mass index; TG, triglycerides; TC, total cholesterol; LDL-C, low-density lipoprotein cholesterol; HDL-C, high-density lipoprotein cholesterol; SBP, systolic blood pressure; DBP, diastolic blood pressure.

Data Sharing Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Ethics Approval and Informed Consent

Ethics approval and the informed-consent process are described in the Materials and methods (Ethics statement). The study was approved by the Ethics Committee of Xingnan Street Community Health Service Center, Nankai District, Tianjin (approval number XNWS01005), and all participants provided written informed consent.

Author Contributions

Mengqi Wang and Daiqing Li contributed to the conceptualization and methodology of the study. Mengqi Wang, Jiaqi Wang, and Jinqiu Liu were responsible for the investigation and data curation; Jiaqi Wang and Jinqiu Liu additionally contributed to writing through review and editing. Mengqi Wang performed the formal analysis, prepared the visualizations, and wrote the original draft. Daiqing Li provided supervision and project administration and contributed to writing through review and editing. All authors made a significant contribution to the work reported; took part in drafting, revising, or critically reviewing the article; gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Disclosure

The authors report no conflicts of interest in this work.