Evaluating eRAMP telecare for diabetes in outpatient clinics: a hybrid effectiveness-implementation study

Abstract

This study evaluated the effectiveness and implementation of a telecare-enhanced hybrid consultation model for diabetes management in public general outpatient clinics. In a single-blind, non-inferiority randomized controlled trial involving 786 adults with type 2 diabetes across seven clinics, participants were assigned to either a hybrid care group, receiving alternating telecare and in-person consultations, or a control group, receiving in-person care only, over 84 weeks. The intervention group (IG) showed comparable HbA1c levels to the control group (CG). Blood pressure readings were consistently lower in the IG, and medication adherence remained stable, unlike the CG, which experienced a midline decline. All clinics sustained the telecare model post-trial. Implementation was facilitated by strong leadership and a reliable digital infrastructure, though challenges such as digital literacy and staff workload remained. The model proved effective, comparable cost, and feasible within a public healthcare setting.

Introduction

Diabetes mellitus (DM) is a pressing global health and economic challenge, contributing significantly to morbidity, premature mortality, and escalating healthcare expenditures^1,2. In Hong Kong, the burden is substantial, affecting approximately 10% of the population and resulting in over HK$2 billion in annual healthcare costs. DM remains a leading cause of cardiovascular disease and stroke, placing considerable strain on the healthcare system^1,2.

To improve chronic disease management, the Hong Kong Hospital Authority (HA) implemented the Risk Assessment and Management Program (RAMP), a structured, multidisciplinary model of care embedded in general out-patient clinics (GOPCs)^3–5. RAMP has demonstrated improvements in glycaemic control and care coordination^5–7. However, its traditional face-to-face delivery format limits accessibility, especially for patients with mobility limitations, time constraints, or those living in remote areas⁶. In recent years, technological advances have spurred interest in telecare consultations as a strategy to address these barriers.

Telecare consultations—remote delivery of healthcare via telecommunications—expanded rapidly worldwide during COVID-19^8,9. In Hong Kong, the pandemic catalysed telecare uptake and broadened use of HA Go, the Hospital Authority’s mobile health app, across outpatient services¹⁰. Lessons from specialist outpatient deployments during this period informed our adaptation of a hybrid diabetes care model for GOPCs. In practice, this approach lessens travel burden, speeds access to care, and enables service delivery irrespective of patients’ locations¹⁰. Evidence from randomized trials suggests that telecare consultations can improve glycaemic control, blood pressure, and lipid profiles in patients with DM, with outcomes comparable or superior to face-to-face care¹¹. Moreover, research indicates that telecare does not compromise therapeutic relationships⁹. Systematic reviews further underscore the clinical effectiveness, cost-effectiveness, and potential for reduced healthcare service utilization associated with telecare for diabetes management^12–15.

Despite these advantages, the implementation of telecare in routine outpatient settings remains limited. Challenges include inadequate infrastructure, resistance to workflow changes, staffing shortages, lack of regulatory clarity, and concerns about missing physical examinations¹⁶. These barriers highlight a critical gap between evidence and practice. Furthermore, while telecare’s clinical efficacy is well-documented, fewer studies have examined its real-world implementation, sustainability, and scalability using robust frameworks. As noted in prior reviews, hybrid effectiveness-implementation trials remain scarce, and few studies apply comprehensive models such as RE-AIM to evaluate both patient- and system-level outcomes.

To address this gap, we conducted a hybrid Type II effectiveness–implementation study to evaluate an enhanced version of the RAMP model that incorporates telecare consultations. Guided by the RE-AIM framework, this study simultaneously assessed the clinical impact and implementation outcomes of the program in seven public GOPCs. The RE-AIM framework enabled a systematic evaluation across five dimensions: Reach (recruitment and representativeness), Effectiveness (clinical outcomes such as HbA1c, BMI, and lipid profiles), Adoption (willingness of providers and institutions to initiate the program), Implementation (delivery fidelity), and Maintenance (sustainability post-trial).

Accordingly, this study aims to generate actionable, context-specific insights on the integration and scalability of telecare consultations within Hong Kong’s public healthcare system. By examining both health outcomes and operational feasibility, our findings provide a comprehensive evidence base to guide future dissemination and policy adoption of hybrid telecare models in primary care settings.

Results

Reach

From December 2021 to December 2024, a total of 1094 participants were screened for eligibility (Fig. 1). Out of these, 786 participants were recruited and randomly assigned to either the intervention group (n = 408) or the control group (n = 378).

Fig. 1. Flowchart of the study population throughout the study.

This flowchart summarizes participant progress throughout the trial, from initial screening to analysis. A total of 1,094 individuals were assessed for eligibility, of whom 308 (28%) were excluded—104 did not meet inclusion criteria and 204 declined to participate for various reasons such as inconvenience, preference for face-to-face consultation, or difficulties using telecare. A total of 786 participants were randomized: 408 assigned to the intervention group and 378 to the control group. Follow-up occurred at baseline (T1), mid-intervention (T2, 42 weeks), and post-intervention (T3, 84 weeks). Reasons for attrition and numbers included in intention-to-treat (ITT) and per-protocol (PP) analyses are shown for each group.

Table 1 shows that the enrolled participants tended to be younger (mean age: 59.65 ± 10.62 years for enrolled group compared to 62.53 ± 10.69 years for the not enrolled one, p < 0.001) and had a lower proportion of females (39.9% compared to 54.1%, p < 0.001). Additionally, these participants had higher levels of education and were more likely to be employed full-time (45.3% vs. 29.9%, p < 0.001). They also reported using digital devices more frequently for health-related activities, such as booking appointments or searching for health information, and showed a greater willingness to share medical information electronically (p < 0.001). Furthermore, the enrolled patients reported fewer functional limitations (95.7% vs. 89.6%, p = 0.008) and expressed greater confidence in using virtual meeting platforms, with 34.6% indicating they were “very confident” compared to 21.2% in the other group (p < 0.001).

Table 1.

Comparison of demographic characteristics between patients enrolled and not enrolled

Characteristic		Patients not enrolled (n = 308, 28%)	Patients enrolled (n = 786, 72%)	P-value
Age (years)		62.53 ± 10.69a	59.65 ± 10.62a	<0.001
Gender	F/M (% of Female)	166/141 (54.1%)a	314/472 (39.9%)a	<0.001
Heights (m)		1.64 ± 0.09a	1.64 ± 0.09a	0.671
Weights (kg)		72.32 ± 16.70a	72.56 ± 15.57a	0.846
BMI (kg/m2)		26.76 ± 5.29a	26.80 ± 4.84a	0.943
DM type	Type1/2 (% of Type 1)	14/266 (5%)	9/755 (1.2%)	0.005
DM duration (years)	<1 y 1–2 yb 3–4 yb 5–6 yb 7–8 yb >9 yb	71 (23.4%) 29(9.5%) 36(11.8%) 25(8.2%) 143(47.0%) 0	179 (22.9%) 135 (17.2%) 100 (12.8%) 68 (8.7%) 298 (38.1%) 3 (0.4%)	0.038
Marital status	Not married Married Divorced Widowed	29 (9.7%) 216 (72.0%) 26 (8.7%) 29 (9.7%)	79 (10.1%) 576 (73.3%) 77 (9.8%) 54 (6.9%)	0.302
Education Level	No formal educational attainment Primary or lower Secondary Tertiary or higher	12 (4.0%) 103 (34.2%) 144 (47.8%) 42 (14.0%)	14 (1.8%) 162 (20.6%) 470 (59.9%) 139 (17.6%)	<0.001
Employment	Full-time employed Part-time employed Unemployed Retired Housewife	90 (29.9%) 28 (9.3%) 11 (3.7%) 160 (53.2%) 12 (4.0%)	355 (45.3%) 83 (10.6%) 24 (3.1%) 299 (38.1%) 23 (2.9%)	<0.001
Living with	Alone Spouse Child	44 (14.6%) 52 (17.3%) 205 (68.1%)	95 (12.1%) 165 (21.0%) 525 (66.8%)	0.780
Financial status	More than sufficient Barely sufficient Not sufficient Far from sufficient	118 (39.2%) 163 (54.2%) 15 (5%) 5 (1.7%)	317 (40.3%) 127 (54.3%) 35 (4.5%) 7 (0.9%)	0.444
Financial support from	Salary (Y/N, (% of Yes))	110/198 (36.7%)	401/385 (51.1%)	0.004
Accommodation	Apartment A room A bed Others	289 (96.7%) 7 (2.3%) 0 (0%) 3 (1.0%)	757 (96.6%) 14 (1.8%) 2 (0.3%) 11 (1.4%)	0.650
Functional limitations	No limitation Walks without assistive devices indoors but limited outdoor walking Walks with assistive devices both indoors and outdoors on flat ground Limited self-mobility, walks with assistive devices and support from others	275 (89.6%) 19 (6.2%) 11 (3.6%) 2 (0.7%)	752 (95.7%) 16 (2.0%) 14 (1.8%) 4 (0.5%)	0.008
IT Literacy
Experience in using a smartphone (years)	1–2 yb 3–4 yb 5–6 yb 7–8 yb >9 yb	11 (3.6%) 8(2.6%) 8 (2.6%) 16 (5.2%) 263 (85.9%)	12 (1.5%) 15(1.9%) 19 (2.4%) 17 (2.2%) 721 (92%)	0.019
Confidence in using a virtual meeting platform	Not confident at all Slightly confident Somewhat confident Moderately confident Very confident	35 (11.4%) 45(14.7%) 62 (20.2%) 100 (32.6%) 65 (21.2%)	16 (2.0%) 35 (4.5%) 147 (18.7%) 316 (40.2%) 272 (34.6%)	<0.001
C-eHEALS47		23.65 ± 9.02a	27.41 ± 7.13a	<0.001
In the past 12 months, have you used a computer, smartphone, or tablet to (Y/N, (% of Y))	Book an appointment Contact with healthcare Purchase medicine Search for health information Connect with people who have similar health or medical issues	71/235 (23.2%) 63/243 (20.6%) 12/294 (3.9%) 102/204 (33.3%) 55/250 (17.9%)	226/560 (28.8%) 219/567 (27.9%) 40/746 (5.1%) 414/372 (52.7%) 187/599 (23.8%)	0.057 0.010 0.416 <0.001 0.032
How willing are you to exchange the following types of medical information electronically with healthcare professionals via a smartphone or tablet?
Health Tips	Not Willing Slightly Willing Somewhat Willing Very Willing	102 (33.4%) 49 (16.1%) 92 (30.2%) 62 (20.3%)	93 (11.8%) 134 (17.0%) 299 (38.0%) 260 (33.1%)	<0.001
Medication Reminder	Not Willing Slightly Willing Somewhat Willing Very Willing	158 (52.0%) 40 (13.2%) 55 (18.1%) 51 (16.8%)	326 (41.5%) 77 (6.2%) 163 (20.7%) 220 (28.0%)	<0.001
Lifestyle Behaviours (e.g., physical activity, food intake, sleep patterns)	Not Willing Slightly Willing Somewhat Willing Very Willing	137 (45.1%) 50 (16.4%) 56 (18.4%) 61 (20.1%)	219 (27.9%) 96 (6.2%) 244 (31.0%) 277 (28.9%)	<0.001
Symptoms	Not Willing Slightly Willing Somewhat Willing Very Willing	118 (38.7%) 42 (13.8%) 73 (23.9%) 72 (23.6%)	156 (19.8%) 98 (12.6%) 249 (31.7%) 283 (36.0%)	<0.001
Test Results	Not Willing Slightly Willing Somewhat Willing Very Willing	98 (32.2%) 33 (10.9%) 73 (24.0%) 100 (32.9%)	94 (12.0%) 69 (8.8%) 208 (26.5%) 415 (52.8%)	<0.001

^aData are reported as mean ± SD or n (%).

^by = year(s).

Effectiveness

Table 2 summarizes baseline characteristics. The intervention and control groups were broadly similar across demographic and clinical variables, with modest differences in age, blood pressure, and medication adherence. These imbalances were accounted for in the adjusted analyses. Intention-to-treat (ITT) results are presented first, in accordance with the study protocol. ITT analyses, based on multiply imputed datasets, showed patterns consistent with the per-protocol (PP) findings and confirmed non-inferiority for HbA1c. Detailed ITT estimates are provided in Supplementary Table 1.

Table 2.

Baseline characteristics of participants

Characteristic		CG (n = 378, 48%)	IG (n = 408, 52%)	P-value
Age (years)		61.38 ± 10.92	58.04 ± 10.10	<0.001
Gender	F/M (% of Female)	155/223 (41%)	159/249 (39%)	0.561
Heights (m)		1.62 ± 0.08	1.65 ± 0.09	<0.001
Weights (kg)		71.62 ± 14.22	73.17 ± 16.01	0.161
BMI (kg/m2)		27.14 ± 4.71	26.79 ± 4.85	0.317
DM type	Type1/2 (% of Type 1)	4/371 (1.1%)	5/384 (1.3%)	0.780
DM duration (years)	<1 y 1–2 y 3–4 y 5–6 y 7–8 y >9 y	103 (27.5%) 58 (15.5%) 45 (12.0%) 29 (7.7%) 139 (37.1%) 1 (0.3%)	76 (18.6%) 77 (18.9%) 55 (13.5%) 39 (9.6%) 159 (39.0%) 2 (0.5%)	0.079
Client-related outcomes
HbA1c (%)		6.96 ± 0.05	6.93 ± 0.05	0.206
Lipid profile (mmol/L)	LDL HDL TG	2.23 ± 0.03 1.30 ± 0.02 1.69 ± 0.31	2.13 ± 0.03 1.28 ± 0.02 1.35 ± 0.04	0.072 0.821 0.692
Blood pressure (mm/Hg)	SBP DBP	136.04 ± 0.86 76.93 ± 0.58	130.05 ± 0.63 75.00 ± 0.54	<0.001 0.023
SF-12	PCS MCS	47.63 ± 8.25 51.63 ± 9.51	47.17 ± 7.26 51.73 ± 7.82	0.574 0.380
ARSM	Medication Adherence Medication Refill Overall Score	9.83 ± 0.12 6.39 ± 0.09 16.22 ± 0.16	9.95 ± 0.11 6.54 ± 0.08 16.48 ± 0.14	0.456 0.238 0.003
SDSCA	Medication Diet Exercise Blood Sugar Test Foot Care	5.92 ± 0.09 4.93 ± 0.10 4.16 ± 0.11 1.38 ± 0.10 3.75 ± 0.13	6.70 ± 0.05 4.94 ± 0.09 3.89 ± 0.09 1.53 ± 0.10 4.06 ± 0.13	<0.001 0.866 0.073 0.398 0.083
Health service utilisation outcomes
Non arrange Private GP	Frequency/past 3 m	0.34 ± 0.05	0.32 ± 0.05	0.804
Arrange public GP	Frequency/past 3 m	2.47 ± 0.05	2.19 ± 0.03	<0.001
Non arrange public GP	Frequency/past 3 m	0.60 ± 0.07	0.39 ± 0.04	0.002
AE visit	Frequency/past 3 m	0.18 ± 0.03	0.16 ± 0.03	0.749
Inpatient	Frequency/past 3 m Total day/past 3 m	0.16 ± 0.03 0.64 ± 0.24	0.14 ± 0.02 0.41 ± 0.08	0.378 0.516

Data are reported as mean ± SD or n (%).

y = year(s), m = month(s)

In terms of healthcare utilization, the intervention group reported fewer arranged public GP visits than the control group (2.19 ± 0.03 vs. 2.47 ± 0.05 visits in the past three months, p < 0.001). Non-arranged public GP visits were also lower in the intervention group (0.39 ± 0.04 vs. 0.60 ± 0.07, p = 0.002). However, there were no significant differences between groups in private GP visits, emergency visits, or hospital stays (p > 0.05 for all comparisons). Although some differences in baseline characteristics and clinical outcomes were observed, these were statistically adjusted for in the subsequent analyses to minimize potential confounding and ensure a valid comparison of outcomes between groups.

We conducted both intention-to-treat (ITT) and per-protocol (PP) analyses. As the findings were consistent between the two approaches, this report will primarily focus on the results from the PP analysis. The follow-up data and corresponding effects on clinical and healthcare utilization outcomes at T2 and T3 are summarized in Figs. 2–4 and Table 3, are further elaborated in the sections below.

Fig. 3. Changes in self-reported outcomes over time between intervention and control groups.

Panels show mean ± SD scores at T1, T2, and T3 for a Physical Component Summary (PCS, SF-12), b Mental Component Summary (MCS, SF-12), c Medication adherence (ARMS), d Medication refill (ARMS), e Overall self-care score, f Medication behaviour (SDSCA), g Diet (SDSCA), h Exercise (SDSCA), i Blood sugar testing (SDSCA), and j Foot care (SDSCA). Red circles = intervention group; blue triangles = control group.

Fig. 2. Changes in clinical outcomes over time between intervention and control groups.

a HbA1c (%), b Systolic blood pressure (SBP, mmHg), c Diastolic blood pressure (DBP, mmHg), d Low-density lipoprotein cholesterol (LDL, mmol/L), e High-density lipoprotein cholesterol (HDL, mmol/L), and f Triglycerides (TG, mmol/L) measured at baseline (T1), mid-intervention (T2), and post-intervention (T3). Data are presented as mean ± SD. Red lines with circles represent the intervention group, and blue lines with triangles represent the control group.

Fig. 4. Changes in health Service Utilization outcomes over time between intervention and control groups.

a Non-arranged private GP visits, b Arranged public GP visits, c Non-arranged public GP visits, d Accident & Emergency (AE) visits, e Inpatient visits, and f Inpatient days from T1 to T3. Red lines with circles denote the intervention group, and blue lines with triangles denote the control group.

Table 3.

Per-protocol intervention effects at T1, T2, and T3 on all effectiveness outcomes in GEE model

Variables			β	95% CI	Wald χ²	p
HbA1c (%)	Between group effect		−0.098	(−0.242, 0.047)	1.754	0.185
	Within group effect	T2 vs T1	−0.098	(−0.194, -0.001)	3.948	0.047
		T3 vs T1	−0.034	(−0.149, 0.082)	0.329	0.566
	Group*Time interaction effect	Group*T2 vs T1	0.170	(0.032, 0.307)	5.849	0.016
		Group*T3 vs T1	0.188	(0.026, 0.351)	5.161	0.023
Lipid profile (mmol/L)—LDL	Between group effect		−0.086	(−0.177, 0.005)	3.439	0.064
	Within group effect	T2 vs T1	−0.108	(−0.173, −0.044)	10.844	<0.001
		T3 vs T1	−0.114	(−0.191, −0.036)	8.245	0.004
	Group*Time interaction effect	Group*T2 vs T1	0.074	(−0.015, 0.163)	2.643	0.104
		Group*T3 vs T1	0.066	(−0.040, 0.173)	1.496	0.221
Lipid profile (mmol/L)—HDL	Between group effect		−0.008	(−0.057, 0.041)	0.106	0.745
	Within group effect	T2 vs T1	0.002	(−0.015, 0.019)	0.046	0.830
		T3 vs T1	0.031	(0.008, 0.053)	7.242	0.007
	Group*Time interaction effect	Group*T2 vs T1	0.009	(−0.017, 0.034)	0.455	0.500
		Group*T3 vs T1	−0.005	(−0.034, 0.024)	0.108	0.743
Lipid profile (mmol/L)—TG	Between group effect		−0.035	(0.146, 0.076)	0.376	0.540
	Within group effect	T2 vs T1	−0.031	(−0.106, 0.044)	0.642	0.423
		T3 vs T1	−0.015	(−0.095, 0.065)	0.137	0.711
	Group*Time interaction effect	Group*T2 vs T1	0.110	(−0.011, 0.231)	3.181	0.075
		Group*T3 vs T1	0.026	(−0.105, 0.156)	0.147	0.701
Blood pressure (mm/Hg)—SBP	Between group effect		−6.190	(−8.245, −4.135)	34.856	<0.001
	Within group effect	T2 vs T1	−5.870	(−12.311, 0.870)	2.914	0.088
		T3 vs T1	−8.561	(−11.274, −5.847)	38.232	<0.001
	Group*Time interaction effect	Group*T2 vs T1	2.974	(−4.096, 10.043)	0.680	0.410
		Group*T3 vs T1	3.865	(0.379, 7.351)	4.722	0.030
Blood pressure (mm/Hg)—DBP	Between group effect		−1.894	(−3.481, −0.307)	5.470	0.019
	Within group effect	T2 vs T1	−3.922	(−6.789, −1.056)	7.191	0.007
		T3 vs T1	−4.211	(−5.887, −2.535)	24.257	<0.001
	Group*Time interaction effect	Group*T2 vs T1	1.240	(−2.183, 4.662)	0.504	0.478
		Group*T3 vs T1	1.018	(−1.395, 3.431)	0.684	0.408
SF-12-PCS	Between group effect		0.542	(−0.548, 1.632)	0.950	0.330
	Within group effect	T2 vs T1	−0.400	(−1.670, 0.871)	0.381	0.537
		T3 vs T1	0.480	(−0.508, 1.469)	0.907	0.341
	Group*Time interaction effect	Group*T2 vs T1	1.470	(−0.221, 3.160)	2.904	0.088
		Group*T3 vs T1	0.628	(−0.692, 1.949)	0.870	0.351
SF-12-MCS	Between group effect		0.097	(−1.125, 1.391)	0.024	0.876
	Within group effect	T2 vs T1	3.557	(2.318, 4.796)	31.656	<0.001
		T3 vs T1	2.754	(1.708, 3.801)	26.607	<0.001
	Group*Time interaction effect	Group*T2 vs T1	−1.168	(−2.774, 0.438)	2.033	0.154
		Group*T3 vs T1	0.064	(−1.411, 1.540)	0.007	0.932
ARMS—Medication Adherence	Between group effect		0.113	(0.197, 0.424)	0.511	0.475
	Within group effect	T2 vs T1	−0.338	(−0.625, −0.050)	5.304	0.021
		T3 vs T1	−0.386	(−0.710, −0.062)	5.453	0.020
	Group*Time interaction effect	Group*T2 vs T1	−0.143	(−0.538, 0.253)	0.498	0.480
		Group*T3 vs T1	0.118	(−0.410, 0.646)	0.192	0.661
ARMS—Medication Rill	Between group effect		0.143	(−0.094, 0.379)	1.399	0.237
	Within group effect	T2 vs T1	−1.596	(−1.836, −1.357)	171.354	<0.001
		T3 vs T1	−1.803	(−2.064, −1.543)	184.382	<0.001
	Group*Time interaction effect	Group*T2 vs T1	−0.039	(−0.385, 0.308)	0.048	0.826
		Group*T3 vs T1	0.336	(−0.051, 0.723)	2.902	0.088
ARMS—Overall Score	Between group effect		0.254	(−0.162, 0.671)	1.435	0.231
	Within group effect	T2 vs T1	−1.944	(−2.335, −1.553)	94.792	<0.001
		T3 vs T1	−2.192	(−2.663, −1.720)	83.093	<0.001
	Group*Time interaction effect	Group*T2 vs T1	−0.165	(−0.722, 0.392)	0.337	0.562
		Group*T3 vs T1	0.476	(−0.271, 1.223)	1.562	0.211
SDSCA—Medication	Between group effect		0.108	(0.014, 0.201)	5.130	0.024
	Within group effect	T2 vs T1	−0.158	−0.325, 0.009)	3.447	0.063
		T3 vs T1	0.127	(0.038, 0.217)	7.771	0.005
	Group*Time interaction effect	Group*T2 vs T1	0.154	(−0.028, 0.335)	2.758	0.097
		Group*T3 vs T1	−0.127	(−0.243, −0.011)	4.589	0.032
SDSCA—Diet	Between group effect		0.025	(−0.245, 0.295)	0.032	0.857
	Within group effect	T2 vs T1	0.359	(0.108, 0.611)	7.843	0.005
		T3 vs T1	0.949	(0.669, 1.229)	44.276	<0.001
	Group*Time interaction effect	Group*T2 vs T1	0.161	(−0.193, 0.516)	0.794	0.373
		Group*T3 vs T1	−0.429	(−0.830, −0.029)	4.418	0.036
SDSCA—Exercise	Between group effect		−0.280	(−0.560, 0.000)	3.832	0.050
	Within group effect	T2 vs T1	0.089	(−0.186, 0.364)	0.402	0.526
		T3 vs T1	−0.192	(−0.499, 0.116)	1.495	0.221
	Group*Time interaction effect	Group*T2 vs T1	−0.283	(−0.654, 0.088)	2.237	0.135
		Group*T3 vs T1	0.123	(−0.293, 0.538)	0.334	0.563
SDSCA—Blood Sugar Test	Between group effect		0.182	(−0.094, 0.457)	1.665	0.197
	Within group effect	T2 vs T1	−0.014	(−0.240, 0.213)	0.014	0.906
		T3 vs T1	0.288	(−0.062, 0.638)	2.603	0.107
	Group*Time interaction effect	Group*T2 vs T1	−0.0241	(−0.599, 0.117)	1.747	0.186
		Group*T3 vs T1	−0.636	(−1.080, 0.191)	7.862	0.005
SDSCA—Foot Care	Between group effect		0.315	(−0.048, 0.678)	2.891	0.089
	Within group effect	T2 vs T1	0.754	(0.404, 1.103)	17.880	<0.001
		T3 vs T1	1.579	(1.195, 1.964)	64.867	<0.001
	Group*Time interaction effect	Group*T2 vs T1	0.110	(−0.383, 0.602)	0.190	0.663
		Group*T3 vs T1	−0.897	(−1.446, −0.348)	10.267	0.001
Non arrange Private GP	Between group effect		−0.015	(−0.149, 0.119)	0.048	0.826
	Within group effect	T2 vs T1	0.127	(−0.040, 0.294)	2.221	0.135
		T3 vs T1	0.242	(0.050, 0.434)	6.104	0.016
	Group*Time interaction effect	Group*T2 vs T1	−0.191	(−0.395, 0.014)	3.332	0.068
		Group*T3 vs T1	−0.195	(−0.483, 0.094)	1.754	0.186
Arrange public GP	Between group effect		−0.319	(−0.436, −0.202)	28.540	<0.001
	Within group effect	T2 vs T1	−2.450	(−2.548, −2.352)	2384.388	<0.001
		T3 vs T1	−2.483	(−2.580, −2.386)	2504.102	<0.001
	Group*Time interaction effect	Group*T2 vs T1	0.295	(0.174, 0.416)	22.758	<0.001
		Group*T3 vs T1	0.319	(0.198, 0.439)	26.923	<0.001
Non arrange public GP	Between group effect		−0.252	(−0.408, −0.095)	9.891	0.002
	Within group effect	T2 vs T1	−0.485	(−0.612, −0.359)	56.473	<0.001
		T3 vs T1	−0.470	(−0.597, −0.342)	52.206	<0.001
	Group*Time interaction effect	Group*T2 vs T1	0.228	(0.072, 0.383)	8.211	0.004
		Group*T3 vs T1	0.228	(0.074, 0.381)	8.447	0.004
AE visit	Between group effect		−0.016	(−0.091, 0.059)	0.177	0.674
	Within group effect	T2 vs T1	−0.104	(−0.164, −0.045)	11.723	<0.001
		T3 vs T1	−0.114	(−0.171, −0.057)	15.485	<0.001
	Group*Time interaction effect	Group*T2 vs T1	0.001	(−0.080, 0.083)	0.001	0.973
		Group*T3 vs T1	0.015	(−0.064, 0.095)	0.146	0.702
Inpatient	Between group effect		−0.032	(−0.107, 0.043)	0.701	0.402
	Within group effect	T2 vs T1	−0.081	(−0.147, −0.014)	5.664	0.017
		T3 vs T1	−0.069	(−0.140, 0.002)	3.680	0.055
	Group*Time interaction effect	Group*T2 vs T1	0.033	(−0.087, 0.086)	0.544	0.461
		Group*T3 vs T1	0.000	(−0.054, 0.119)	0.000	0.995
Inpatient Day	Between group effect		−0.155	(−0.642, 0.332)	0.390	0.532
	Within group effect	T2 vs T1	−0.266	(−0.743, 0.210)	1.200	0.273
		T3 vs T1	−0.336	(−0.786, 0.113)	2.153	0.142
	Group*Time interaction effect	Group*T2 vs T1	0.173	(−0.361, 0.708)	0.403	0.525
		Group*T3 vs T1	0.170	(−0.322, 0.663)	0.458	0.499

^*β coefficients are interpreted relative to the control group at baseline (T1); positive values indicate higher outcomes in the intervention group.

At baseline (T1), the mean HbA1c levels were comparable between the control group (6.99 ± 1.03%) and the intervention group (6.89 ± 0.97%). Over time, the control group exhibited a slight decline at T2 (6.90 ± 0.88%) but returned close to baseline at T3 (6.96 ± 0.91%). In contrast, the intervention group showed a gradual increase in HbA1c at T2 (6.97 ± 0.96%) and further at T3 (7.08 ± 0.98%). Compared to baseline, a modest but statistically significant reduction in HbA1c was observed at T2 across both groups (β = −0.098, 95% CI: −0.194 to −0.001, p = 0.047), but this effect was not sustained at T3 (p = 0.566). Significant group*time interactions at T2 (β = 0.170; 95% CI, 0.032 to 0.307; p = 0.016) and T3 (β = 0.188; 95% CI, 0.026 to 0.351; p = 0.023) indicated divergent trajectories, with a transient improvement in the control group and a slight increase in the intervention group. To evaluate non-inferiority, as specified in the published study protocol with respect to the primary outcome and non-inferiority margin, an independent-samples t-test was performed to compare HbA1c levels at T3. The mean difference between the intervention and control groups was –0.12% (intervention minus control), with a one-sided 95% CI upper bound of 0.03%, which was lower than the predefined non-inferiority margin of 0.40%. This confirmed that the hybrid telecare model was non-inferior to standard consultation for HbA1c control.

With regard to lipid profiles, LDL levels showed a significant reduction at both T2 (β = −0.108, 95% CI: −0.173 to −0.044, p < 0.001) and T3 (β = −0.114, 95% CI: −0.191 to −0.036, p = 0.004) across groups, though no significant differences were observed between the intervention and control groups. HDL levels increased modestly but significantly at T3 compared to T1 (β = 0.031, 95% CI: 0.008 to 0.053, p = 0.007), while TG levels remained stable over time, with no significant differences between or within groups.

In terms of blood pressure, decreased over time in both groups, with a substantial reduction observed at T3 (β = −8.561; 95% CI, −11.274 to −5.847; p < 0.001). DBP also declined significantly over time among all participants (T2 vs. T1: β = −3.922, 95% CI: −6.789 to −1.056, p = 0.007; T3 vs. T1: β = −4.211, 95% CI: −5.887 to -2.535, p < 0.001) and was consistently lower in the intervention group than in the control group (β = −1.894, 95% CI: −3.481 to −0.307, p = 0.019). However, no significant time-by-group interaction effects were observed for any of the outcomes.

The PCS scores remained consistent over time, and no significant changes were observed from T1 to T3. There were no significant differences of group or group-by-time interaction effects were observed. In contrast, the MCS scores showed significant improvements over time in both groups, with increases at T2 (β = 3.557, 95% CI: 2.318 to 4.796, p < 0.001) and T3 (β = 2.754, 95% CI: 1.708 to 3.801, p < 0.001). However, no significant main effects of group or group-by-time interaction effects were observed.

The overall ARMS score improved significantly over time, with reductions observed at both T2 (β = −1.944, 95% CI: −2.335 to −1.553, p < 0.001) and T3 (β = −2.192, 95% CI: −2.663 to −1.720, p < 0.001) compared to T1. Similarly, medication refill adherence significantly improved at T2 (β = −1.596, 95% CI: −1.836 to −1.357, p < 0.001) and T3 (β = −1.803, 95% CI: −2.064 to −1.543, p < 0.001). No significant group or interaction effects were observed for these outcomes. However, medication-taking adherence did not change significantly over time or differ between the intervention and control groups.

The medication adherence improved overtime, with a significant increase at T3 (β = 0.127, 95% CI: 0.038 to 0.217, p = 0.005), as shown in the SDSCA scores. Diet adherence significantly improved over time, with increases at both T2 (β = 0.359, 95% CI: 0.108 to 0.611, p = 0.005) and T3 (β = 0.949, 95% CI: 0.669 to 1.229, p < 0.001). Foot care adherence improved significantly from T1 to T2 (β = 0.754, 95% CI: 0.404 to 1.103, p < 0.001) and showed further improvement at T3 (β = 1.579, 95% CI: 1.195 to 1.964, p < 0.001). In contrast, no significant changes were found for exercise or blood sugar testing adherence over time or between groups.

In terms of health-service utilization, the frequency of non-arranged private GP visits significantly increased from T1 to T3 across both groups (β = 0.242, 95% CI: 0.050 to 0.434, p = 0.016), with no significant interaction effects. In contrast, arranged public GP visits significantly decreased at T2 (β = −2.450, 95% CI: −2.548 to −2.352, p < 0.001) and further at T3 (β = −2.483, 95% CI: −2.580 to −2.386, p < 0.001). Non-arranged public GP visits also declined from T1 to T2 (β = −0.485, 95% CI: −0.612 to −0.359, p < 0.001) and T3 (β = −0.470, 95% CI: −0.597 to −0.342, p < 0.001).

Emergency visits declined significantly at T2 (β = −0.104, 95% CI: −0.164 to −0.045, p < 0.001) and T3 (β = −0.114, 95% CI: −0.171 to −0.057, p < 0.001) in both groups, with no significant differences between groups or interaction effects. Inpatient admissions decreased significantly at T2 compared to T1 (β = −0.081, 95% CI: −0.147 to −0.014, p = 0.017), but no significant group differences or interactions were found. The number of inpatient days remained stable over time, with no statistically significant changes.

Adoption

Figure 5 compares organizational, technological, and personnel-related readiness factors over time. The analysis revealed the significant effect of time on both Organizational Environment (p = 0.005) and Implementation Process (p = 0.019). The Organizational Environment score declined from a mean of 9.00 ± 0.44 at T1 to 7.47 ± 0.31 at T2, with a partial rebound to 8.66 ± 0.33 at T3. Similarly, the Implementation Process score decreased from 9.40 ± 0.38 at T1 to 7.29 ± 0.55 at T2, followed by a slight increase to 7.44 ± 0.45 at T3.

Fig. 5. Comparison of seven Readiness for Implementation sub-scores at T1 (n = 26), T2 (n = 50), and T3 (n = 38).

Mean ± SD scores of organizational readiness factors at baseline (T1, yellow bars), mid-implementation (T2, orange bars), and post-implementation (T3, red bars). Factors include organizational environment, motivational readiness, technology usefulness, promotion, implementation process, department–technology fit, and key personnel awareness/support. Red asterisks (*) indicate statistically significant differences between timepoints (p < 0.05).

Key Personnel Awareness and Support showed a significant decline over time (p = 0.002), from 16.07 ± 3.41 at T1 to 13.19 ± 4.02 at T2, and further to 12.26 ± 3.97 at T3. In contrast, no statistically significant changes were observed across the timepoints for Organizational Motivation, Technology Usefulness, Promotion, and Department–Technology Fit. Nonetheless, Key Personnel Awareness and Support demonstrated a consistent downward trend, despite the lack of statistical significance in interaction effects.

Figure 6 shows a decline in the Global Readiness for Implementation Score from 71.08 ± 1.63 at T1 to 62.20 ± 3.28 at T2, with little additional change at T3 (61.50 ± 2.67). The overall decrease was not statistically significant; however, significant declines emerged in selected subdomains, including Organizational Environment, Implementation Process, and Key Personnel Awareness and Support.

Fig. 6. Comparison of Readiness for Implementation total scores at T1 (n = 26), T2 (n = 50), and T3 (n = 38).

Mean ± SD of the overall readiness score at T1 (orange), T2 (yellow), and T3 (red). Error bars represent standard deviations.

Implementation

To evaluate provider adherence to the program protocol, each step was assessed using a standardized checklist, covering a total of 1014 patient interactions (Table 4). Patient recruitment was successfully completed in 99.90% of cases, with only one omission. Written informed consent was obtained in 100% of cases. Baseline data collection demonstrated a 98.62% completion rate. Both SMS reminders and device readiness checks were successfully implemented in all cases (100%). Pre-consultation phone confirmations achieved a 98.32% success rate. Other key procedures—including teleconference sign-in, patient identification, adherence to data privacy protocols, and documentation of consultation notes—achieved full compliance (100%).

Table 4.

Results of performance list

Person in Charge	Items	Satisfactory (%)
Research Assistant	Recruit patient according to the eligibility criteria	99.9
	Obtain written consent	100.0
	Measure baseline data	98.6
Clerk	1 day before telecare consultation: Use SMS send meeting ID to patient and phone to confirm the readiness of device for telecare consultation	100.0
	1 hour before telecare consultation: call patients to confirm availability and give Zoom password	98.3
	Prepare the name list, standard dialogue card and zoom device to associate consultant for starting telecare consultation	100.0
Doctor	Sign in Zoom with the Clinic account	100.0
	Patient identification: verify patient’s name and ID number against with CMS and patient list	100.0
	Standard dialogue for data privacy, security & patient consent with documentation in CMS	100.0
	Consultation notes with keyword “TELE”	100.0

Facilitators and barriers of program implementation were thematically summarized from qualitative interview data. Six themes emerged as facilitators for the successful implementation of telecare, including effective management and multifaced support, comprehensive training and guidance, perceived convenience and continuity of care, openness to technology and digital skills, positive impacts during the covid-19, reliable infrastructure and technical support. Conversely, four themes emerged as barriers, including limited digital literacy and reluctance to change, inadequate services, environmental and technological challenges, operational challenges and staff capacity constraints. The detailed results and illustrated quotes are presented in the Supplementary table 3 and 4.

Guided by the Theoretical Domains Framework (TDF), we examined implementation determinants by behavioural construct. Knowledge, skills, and environmental context/resources functioned as both facilitators and barriers: comprehensive understanding and strong technical competence among patients and providers enhanced engagement, whereas gaps in these areas led to misunderstanding and underuse. Effective delivery required both soft resources (information, communication, problem-solving) and hard resources (staffing, telecare applications, facilities) aligned across individual, team, and organizational levels. Operational frictions—most notably difficulties with the Hospital Authority’s routine e-payment for standard fees and limitations in medication delivery—impeded some telecare consultations; these were routine service charges, not study costs. Social influence (support from family, community, and government) was a key facilitator. Experiences during COVID-19 increased public familiarity with online consultations, and the Hospital Authority’s expansion of HA Go across outpatient services further smoothed adoption in GOPCs and supported the hybrid model’s feasibility. Conversely, concerns about scams and inconvenience, scepticism about clinical effectiveness, and emotional factors—perceiving telecare as impersonal, resistance to change, and reliance on the reassurance of physical examination—dampened uptake. Detailed belief statements are provided in Supplementary Table 5.

Maintenance

At the individual level, withdrawal rates were higher in the intervention group, with 44% at T2 and 64% at T3, compared to 35% at T2 and 57% at T3 in the control group, evidenced by participants voluntarily withdrawing or being unresponsive to follow-up telephone calls. There was no documented evidence of participants re-engaging with the intervention or independently continuing to use its components after withdrawal. Qualitative data provided possible explanations for the additional dropout compared to in-person consultations. The main reasons centred on patients’ personal attitudes and capabilities. Patients’ insufficient proficiency with digital technologies necessary for telecare sessions led to negative perceptions, such as viewing telecare as bothersome and difficult to use, resulting in its abandonment. Additionally, service deficiencies, such as the lack of medication delivery and the inability to provide sick notes, along with telecare’s inherent limitation in not offering physical examinations, contributed to the dropout rate.

All participating GOPCs continued the telecare program beyond the study period, indicating strong sustainment. Clinics integrated the model into routine workflows, with staff operating independently and without external support. Core intervention elements were embedded into HA Go, the Hospital Authority’s teleconsultation app, supporting institutionalization and scalability.

In addition to operational continuation, organizational sustainability was further assessed through a CMA from both provider and societal perspectives over the 84-week study period. A brief summary is provided below, with detailed methodology, assumptions, and summary tables available in Supplementary Table 6-15.

We conducted a cost-minimization analysis over 84 weeks from provider and societal perspectives (methods, assumptions, and tables in Supplementary Tables 6–15). From the provider perspective, the total cost per subject was HK$8,158.1 in the intervention group versus HK$8,230.1 in the control group. Intervention setup costs were HK$84,301 across seven clinics—HK$1.3 per RAMP patient (n = 67,181)—covering 21 iPads (HK$2,394 each), stands (HK$240), speakerphones (HK$771), and one-hour staff training; salary costs followed the Hong Kong Civil Service Pay Scale, and equipment was amortized over total RAMP enrolment (Supplementary Tables 6). Based on time-and-task logs for 50 subjects, the average staff-time cost per telecare consultation was HK$152.9 (clerical confirmation HK$18.7; nursing HK$90.6; administrative wrap-up; doctor consultation 8.3 min = HK$96.4), using standard hourly wages and estimated durations (Supplementary Tables 7–8). In the control arm, face-to-face consultations averaged HK$99.4 per subject, or HK$100.9 excluding tasks not feasible via telecare (physical exams, eye exams, medication dispensing) (Supplementary Tables 9–11). Health-service utilization outside RAMP (from CMS data) showed 3-month mean costs per subject of HK$1,199 (intervention) vs HK$1,246.5 (control) at 42 weeks (T2), and HK$886.9 vs HK$906.2 at 84 weeks (T3) (Supplementary Tables 12), annualized across the study with a weighted approach. Given the intended schedules—three telecare plus four in-person visits (intervention) versus seven in-person visits (control)—the total provider cost per subject was HK$8,158.1 and HK$8,230.1, respectively (Supplementary Tables 13).

From the societal perspective (structured interviews: n = 33 intervention; n = 56 control), telecare participants reported 28.5 min in virtual consultations plus 90.8 min for travel/waiting to collect medications (HK$185.6 per subject). Control participants spent 114.7 min per clinic visit (some accompanied), yielding HK$227.1 per subject (Supplementary Tables 14). Combining patient costs with provider and utilization costs, the total societal cost per subject over 84 weeks was HK$9,623.3 in the intervention group versus HK$9,819.8 in the control group (Supplementary Tables 15).

Discussion

This study provides robust evidence that a telecare-enhanced hybrid consultation model for diabetes management can be both clinically effective and organizationally feasible within Hong Kong’s public healthcare system. The intervention led to improvements in blood pressure and medication adherence, while maintaining comparable cost over 84 weeks⁷. The transient midline reduction in HbA1c—no longer evident at endpoint—likely reflects normalization of post-pandemic routines, stabilization after early treatment adjustments, and inherent limitations of telecare (e.g., fewer opportunities for physical examination). Higher attrition in the telecare arm by T3 may also have biased late follow-up estimates.

Although within-group improvements were observed, between-group differences were generally non-significant. Crucially, the upper bound of the one-sided 95% CI for HbA1c remained below the non-inferiority margin, confirming that the hybrid model was non-inferior to conventional face-to-face care. Rather than demonstrating superiority, the model sustained comparable clinical outcomes while achieving organizational feasibility and cost neutrality. These findings underscore the value of a hybrid approach as a scalable alternative for routine outpatient services where accessibility and resource optimization are priorities; notably, the model has been embedded into routine care across seven GOPCs, supported by a multidisciplinary team and the HA Go digital platform.

Despite strong initial adoption, disparities in digital readiness and retention emerged over time. Although 72% of screened individuals were enrolled, those who ultimately accessed and engaged with telecare were disproportionately younger, more educated, employed, and digitally literate, with fewer functional limitations. They achieved greater improvements in clinical and behavioral outcomes, including enhanced medication adherence and reduced outpatient service use. Conversely, those with lower digital literacy, older age, or more complex health and social needs were less likely to enroll or remain engaged. This interaction between reach and effectiveness underscores a key implementation challenge: individuals who may benefit most from improved access were among the least represented^6,17.

To address this gap, digital inclusion strategies must be embedded within future telecare models. These could include community-based onboarding, simplified user interfaces, culturally tailored education, digital literacy training, caregiver-assisted sessions, and hybrid consultation formats that combine virtual and in-person visits⁶.

At the organizational level, early integration of telecare into clinical workflows was facilitated by structured training, leadership support, and reliable infrastructure⁷. Readiness assessments, however, showed declining engagement over time, most notably in environmental support, process readiness, and frontline staff morale from baseline to endpoint⁷. Despite these subdomain declines, the global readiness index did not change significantly across the three time points, and all clinics sustained the telecare model post-trial. Together, these findings suggest that initial enthusiasm alone is insufficient for long-term sustainment; ongoing reinforcement and support is essential to maintain momentum¹⁸.

Qualitative findings provided important context. While strong leadership and clear protocols supported early implementation, operational challenges quickly emerged. Staff faced increased workloads, often without compensation, particularly when assisting patients with technical issues. These disruptions, combined with insufficient IT support and the absence of structured feedback mechanisms, contributed to declining motivation and organizational fatigue. To support long-term viability, clinics must implement robust infrastructure—including dedicated technical support, change management strategies, and staff incentives—to sustain implementation quality^6,19–22.

Sustained integration also requires adaptive leadership, continuous capacity building, and streamlined workflows. Ongoing, targeted training programs are essential for maintaining staff competence and preventing burnout¹⁹. Reliable technical support and reduced workflow friction can ease the burden on frontline providers²⁰. Additionally, embedding feedback loops and aligning incentives with implementation goals may foster a culture of shared ownership and continuous improvement¹⁸.

In contrast to organizational-level maintenance, individual engagement with telecare declined considerably. By the end of the study period, 64% of participants in the intervention group had disengaged, with no evidence of continued independent use of telecare services⁷. While some participants cited convenience, flexibility, and increased self-efficacy as key motivators for engagement, others discontinued use due to digital anxiety, unfamiliarity with technology, or a preference for in-person care once pandemic restrictions eased^6,23,24.

To bridge this gap, future models should incorporate long-term engagement strategies such as digital coaching, caregiver support, periodic check-ins, and meaningful incentives to promote sustained patient involvement⁶.

Importantly, over 84 weeks, the hybrid model was cost-neutral. Slightly higher staffing costs in the intervention arm were offset by lower healthcare utilization, particularly in public outpatient services⁷. We used an ingredient-based costing approach to identify, measure, and value resources. Direct provider costs included setup (hardware, software, staff training) and recurrent expenditures (clinician time, clerical support, IT assistance), while the societal perspective captured patient time and transportation. All staff costs—including clerical and IT—were valued using prevailing salary scales. Together, these findings indicate that telecare did not add financial burden and may enable cost minimization in resource-constrained settings. The integration of core telecare functions into HA Go further strengthens the model’s scalability and supports alignment with Hong Kong’s digital health infrastructure²⁵.

Several limitations warrant consideration. First, the study was conducted within a single, integrated public healthcare system, which may constrain generalizability to settings with different digital capacities or care models. Second, selection bias may have operated at two levels: (i) at enrolment, participants were more likely to be younger, functionally independent, and digitally prepared than decliners (Table 1); and (ii) during follow-up, differential attrition—disproportionately affecting older or less digitally literate participants—may have biased estimates. Higher attrition in the telecare arm during the post-pandemic phase may also reflect shifting patient preferences. Finally, the hybrid Type II trial design, while pragmatic for real-world implementation, limits causal attribution of effects to the telecare component alone⁷.

In conclusion, this study demonstrates that a telecare-enhanced RAMP model can be feasibly implemented and sustained within a public primary care setting. It provides evidence that hybrid consultations can improve clinical outcomes and maintain cost efficiency while identifying important equity and implementation challenges. Addressing digital access gaps and supporting long-term patient and provider engagement will be key to ensuring that telecare becomes a sustainable, patient-centred feature of chronic disease management^6,7.

Methods

Study design

The study utilized a large single-blind, randomized trial with a hybrid effectiveness-implementation design (Type II) to concurrently evaluate the effectiveness and implementation outcomes of the program²⁶. This design was selected based on prior evidence suggesting that telecare consultation is a safe and acceptable intervention for improving the health of DM patients, despite preliminary findings and limited understanding of its implementation in healthcare settings. Eligible participants who provided consent were randomized into two parallel groups: the intervention group (n = 408; alternating between telecare and face-to-face consultations) and the control group (n = 378; conventional face-to-face consultations) in a 1:1 allocation ratio. A block randomization method with a block size of four ensured balanced group representation.

The random assignment schedule, generated using the Research Randomizer software, was created by a team member not involved in participant recruitment. Group assignments were sealed in envelopes and revealed sequentially during randomization. Following successful participant recruitment, the research assistant contacted the team member responsible for group assignment, who assigned participants to intervention or control groups based on a computer-generated number (‘1’ for intervention and ‘2’ for control). While data collectors remained blinded to group assignments, participants and healthcare providers involved in the intervention were not. Non-consenting or ineligible patients were reassured that declining participation would not affect their access to usual services.

This study received IRB approval from both the HA (REC Ref. No.: 20119, The Joint NTWC REC) and the Hong Kong Polytechnic University (No. HSEARS20200619003), and was registered on ClinicalTrials.gov (Identifier: NCT05183685) and followed the Consolidated Standards of Reporting Trials (CONSORT) guidelines (Supplementary Note 1). The study protocol was previously published and provides further methodological details²⁷.

Participants

This study was conducted in seven government out-patient clinics (GOPCs) under the Hospital Authority, which serve as primary healthcare centres for over 51,000 diabetic patients annually within the New Territories West Cluster (NTWC) ²⁸Diabetic patients who had scheduled appointments with a doctor, either as new cases or for reviews aimed at improving diabetes management, were approached and invited to participate. Eligible participants were adults aged 18 years or older with a confirmed diabetes diagnosis who were clinically stable—that is, without acute diabetic complications (e.g., ketoacidosis, hyperosmolar hyperglycaemic state, or severe hypoglycaemia requiring hospitalization) and without urgent comorbidity-related instability—and who attended routine clinic follow-ups. HbA1c levels were not used as inclusion or exclusion criterion. All participants were required to be able to provide informed consent. Exclusion criteria included the presence of dementia, hearing or vision impairments without the support of a companion or family member, or a lack of internet access at home or through nearby community centres.

Sample size

Power analysis was conducted using non-inferiority criteria, focusing on the primary outcome of HbA1c. A non-inferiority margin of 0.4%, derived from a previous similar study²⁹, was applied. With a standard deviation of 2%³⁰, a type 1 error of 5%, a statistical power of 80%, and an estimated discontinuity rate of 20%³¹, it was calculated that 350 subjects per group would be necessary. Consequently, the total sample size required for the program was determined to be 700 participants. Because attrition over the 84-week follow-up was expected to exceed this 20% allowance, we continued consecutive recruitment across all seven clinics, enrolling 786 participants to preserve statistical power despite withdrawals.

Data collection

Data collection occurred at three intervals: at baseline pre-intervention (T1); 42 weeks after the start of the intervention (T2); and at 84 weeks, immediately following the conclusion of the intervention (T3). A trained research assistant, blinded to group allocation and uninvolved in both the intervention and control groups, collected the data.

Semi-structured individual interviews (n = 44) with healthcare providers, including GOPC managers, doctors, nurses, and clerical staff, were conducted at T2 and T3 to gather descriptive information on program implementation and perceptions of the overall program. This provided supplementary evidence to the quantitative data. An interview guide was designed following the qualitative inquiry approach of the RE-AIM framework (Supplementary Note 2).

Intervention group

Durlak and DuPre’s implementation framework (Supplementary Table 16) guided the program³², highlighting the influence of innovation characteristics (compatibility and adaptability) and provider characteristics (skill and knowledge proficiency) on successful implementation. To enhance compatibility and adaptability, the research and GOPC service teams conducted multiple meetings to review and refine the RAMP program protocol, referral criteria, and team member responsibilities. Healthcare providers and managers expressed positive attitudes toward the program, agreeing on its potential as a long-term service development within the clinic. Implementation also unfolded during COVID-19, when telecare pilots were expanding across the Hospital Authority—context that further increased staff readiness and patient receptivity to telecare.

The research team conducted program orientation sessions, providing healthcare providers with essential knowledge and skills for effective implementation. These sessions covered the study’s rationale, timeline, roles, responsibilities, recruitment process, and referral criteria. Ongoing support, such as telephone consultations and one-on-one meetings, was provided to assist staff in overcoming challenges encountered during the study.

Following implementation setup and staff preparation, the intervention was launched for patient participation. Participants in this study engaged in an enhanced RAMP program, which consisted of three primary components:

1. Alternating consultations: Beginning with face-to-face consultations, followed by telecare sessions, and alternating for a total of seven consultations (four face-to-face and three telecare).

2. Support from a multidisciplinary healthcare team: Delivering integrated care for diabetic management.

3. Involvement of family members or informal caregivers: Encouraging their active participation in the program.

The study spanned 84 weeks, with consultations occurring approximately every 14 weeks, utilizing both face-to-face and telecare methods.

During the first on-site consultation, healthcare providers introduced the telecommunication app, HA Go, to the intervention group. This app facilitates virtual consultations. Patients were educated on accessing telecare sessions via the app, and before the first telecare consultation, they received detailed instructions for effective use. A clinic clerk supported the process by providing passwords for online meetings, contacting patients an hour prior to consultations for availability checks, and conducting rehearsals. Any technical issues were promptly resolved by the clerk or IT team to ensure a seamless telecare experience.

Before the initial face-to-face consultation with the GOPC doctor, patients underwent a thorough assessment of diabetic complications and micro- and macro-vascular symptoms by a nurse. This evaluation included HbA1c, fasting lipid profile, body mass index, eye and foot examinations. Based on the assessment results and associated diabetic and cardiovascular risks, the doctor reviewed reports, adjusted diabetic medications, provided education, offered self-management support, and referred patients to the multidisciplinary team as necessary.

The multidisciplinary team—composed of doctors, nurses, optometrists, dietitians, podiatrists, physiotherapists, and patient care assistants—provided care through both face-to-face and telecare sessions as appropriate. Team member responsibilities were clearly defined in the program protocol. Multidisciplinary case conferences were held every eight months to review patient progress, address concerns, discuss treatment options, and adjust management plans accordingly. As part of the standard RAMP program, the team developed educational materials and structured empowerment courses to enhance patients’ knowledge and self-management skills. These resources were standardized across all seven clinics and made available to both intervention and control groups, reinforcing usual self-management support rather than adding a distinct intervention.

Patients’ families and informal caregivers were actively encouraged to participate in consultations. Their input on patients’ lifestyles and treatment adherence proved invaluable. Informal caregivers, particularly those assisting older adults, also supported patients in resolving technical issues encountered during telecare consultations.

Control group

Control group participants received the same baseline clinical assessment, multidisciplinary management, education, and empowerment courses as the intervention arm under the standard RAMP model. The sole distinction was modality: all consultations were conducted exclusively in person, with no telecare component.

Outcomes

The measures corresponding to the five dimensions of the RE-AIM framework—Reach, Effectiveness, Adoption, Implementation, and Maintenance—are outlined below. The Effectiveness dimension involved evaluating outcome measures for both the intervention and control groups, while the other dimensions specifically pertained to the telecare consultation group. A set of items reflecting best practices in applying the RE-AIM framework was adopted, based on the RE-AIM checklist (Fig. 7)³³.

Fig. 7. RE-AIM study model, measures, and data source.

This figure illustrates how this hybrid effectiveness-implementation study was designed and evaluated under the RE-AIM framework. Key dimensions of RE-AIM were emphasized, followed by the measures and analyses used to address and evaluate each dimension. The data sources are listed, detailing the approaches by which supporting data were collected and utilized. *CMS=Clinical Management System.

The reach (per RE-AIM) dimension captured the number, proportion, and representativeness of participants. The proportion reached was calculated as the number enrolled divided by the total pool of potentially eligible patients. Because recruitment targets were set by the power analysis, Reach emphasized representativeness rather than absolute numbers. Data on non-participants came from two sources: (1) de-identified records in the Clinical Management System (CMS) and (2) a brief, voluntary questionnaire on basic sociodemographic and digital literacy. Patients were verbally informed that the questionnaire was optional and declining would have no consequences; responses were anonymized at source, and written consent was not required given the minimal, non-identifiable data collected.

The effectiveness of the study was assessed through client-related outcomes and health service utilization outcomes. For client-related outcomes, the primary outcome of this study was the mean change in HbA1c levels, which was assessed by comparing values between the intervention group and the control group at various time points. HbA1c, which reflects the average blood glucose levels over the preceding 2-3 months, is the standard measure for assessing glycaemia³⁴. Blood samples were collected by a nurse during the initial health check, following the participants’ consent. Along with HbA1c, the serum lipid profile was analysed to evaluate cardiovascular risk, focusing on three key parameters: high-density lipoprotein (HDL), low-density lipoprotein (LDL), and triglycerides (TG). Blood pressure (BP), including systolic blood pressure (SBP) and diastolic blood pressure (DBP), was measured using a digital monitor. Three readings were taken at 5-minute intervals from the participant’s right arm (unless contraindicated) while seated, and the average of these readings was used to determine the final BP at each time point. Clinical outcome targets were based on the 2018 European Society of Cardiology prevention guidelines, with BP targets set at <130 mmHg for patients under 65 years of age and <140 mmHg for patients aged 65 and older. Lipid targets included HDL cholesterol ≥ 1.0 mmol/L for both males and females, LDL cholesterol <2.6 mmol/L, and triglycerides <1.7 mmol/L. Body mass index (BMI)³⁵ was calculated using the formula: weight in kilograms divided by height in m². Height and weight measured to the nearest 0.10 cm and 0.10 kg, respectively, during the health check, with participants removing shoes for accuracy.

Diabetes self-care management was assessed using the Summary of Diabetes Self-care Activities (SDSCA)³⁶, a 10-item scale that measures patients’ adherence to five key activities: general diet, consumption of fruits and low-fat foods, daily physical activity, medication adherence, glucose monitoring, and foot care. This scale has demonstrated high validity and reliability³⁷. Medication adherence was measured using the Adherence to Refills and Medications Scale (ARMS)³⁸, a 12-item questionnaire designed to evaluate participants’ ability to take and refill prescribed medications under various circumstances. This scale has shown high internal consistency, with a Cronbach’s alpha of 0.81³⁹. Quality of life was assessed using the Chinese version of the 12-item Short Form Health Survey version 2¹⁹. Participants rated the items on Likert-type scales, and the results were summed to provide distinct physical and mental health component scores. The validity, reliability, and standard scoring algorithm of this instrument have been confirmed in numerous studies^20,40.

Health service utilisation outcomes were assessed by tracking the number of visits individuals made to various healthcare facilities, including general practitioners (GPs), emergency departments, hospitals, and outpatient clinics. Data for all service encounters, with the exception of GP visits, were extracted from the Clinical Management System (CMS) database. GP visit frequencies, however, were self-reported by the participants.

The adoption dimension was measured using the Readiness for Implementation Model (RIM) Survey was employed to assess the success of telecare consultation adoption at both clinic and staff levels⁴¹. Comprising 42 items, the RIM identifies barriers and facilitators to implementing new technologies in healthcare settings. Responses were aggregated into a total score ranging from 0 to 100, with a score of 70 or above signifying successful adoption. The survey’s validity and reliability, including inter-rater reliability, have been well established⁴². Ward managers and doctors completed the questionnaire at three key time points: baseline pre-intervention (T1), 42 weeks (T2), and 84 weeks (T3).

The implementation dimension evaluated to determine the extent to which it was carried out as intended within the clinic setting. A performance checklist, developed by the Principal Investigator (PI) based on the program’s workflow, served as the primary tool for assessing adherence to the planned implementation. The checklist underwent validation by a panel of healthcare professionals—including doctors, optometrists, and nurses—as well as international researchers specializing in implementation science. Data were collected on the percentage of perfectly delivered consultations for each participant across the study. A consultation was deemed perfectly delivered when all checklist items were completed per protocol—patient identification, device readiness verification, data-privacy confirmation, and accurate documentation in the CMS. To complement quantitative fidelity data, semi-structured group interviews with the General Outpatient Clinic manager, clinical providers, patients, and informal caregivers explored facilitators, barriers, and the feasibility of implementation.

Finally, maintenance was evaluated at both the individual and organizational levels^23,43. At the individual level, we examined effectiveness outcomes at 84 weeks, withdrawal rates, and patient-incurred costs (e.g., travel and time). At the organizational level, we conducted a cost-minimization analysis (CMA) comparing total and per-capita costs between intervention and control arms. Given the study’s non-inferiority design, a full cost-effectiveness analysis was unnecessary unless telecare demonstrated clinical superiority (e.g., gains in quality-adjusted life years). Accordingly, we used an ingredient-based costing approach to identify, measure, and value resource use from both the provider (details and questionnaire in Supplementary Note 3) and societal (details and questionnaire in Supplementary Note 4) perspectives.

Statistical analysis

Statistical analyses were performed in SPSS version 29 under both intention-to-treat (ITT) and per-protocol (PP) frameworks. Missing outcome data were handled using multiple imputation (m = 20) under a missing-at-random assumption. The ITT analysis was designated as the primary analysis, and all primary and secondary outcomes were first analysed using the imputed datasets. The PP analyses were conducted as secondary analyses and included participants who completed ≥5 of the 7 scheduled consultations and contributed outcome data at both T2 and T3. All outcome models adjusted for baseline values of the respective measure. Baseline group comparisons used independent-samples t tests for continuous variables, Mann–Whitney U tests for ordinal variables, and Pearson’s χ² tests for categorical variables (with exact p values where appropriate).

To determine whether the hybrid telecare model was non-inferior to standard consultation, we compared HbA1c levels between the intervention and control groups at T3 using an independent-samples t-test, as prespecified. The mean difference (intervention minus control) and its one-sided 95% confidence interval (CI) were estimated. Non-inferiority was concluded if the upper bound of the one-sided 95% CI did not exceed the established margin of 0.4%¹⁹. Longitudinal changes in clinical outcomes were additionally examined using generalized estimating equations (GEE) with group, time, and group-by-time interaction terms to provide adjusted estimates and to describe trajectories over time. In this model, the control group and baseline (T1) were set as the reference categories. Thus, the between-group terms in GEE model represent the difference at baseline; the time terms represent within-control-group change from baseline; and the group*time interaction terms represent the additional change from baseline in the intervention group relative to the control group. Accordingly, a positive β coefficient indicates a higher outcome value in the intervention group compared with the control group, whereas a negative β coefficient indicates a lower outcome value in the intervention group. Additionally, one-way ANOVA was used to evaluate between-group differences in all Readiness for Implementation Model (RIM) score outcomes.

To explore the facilitators and barriers of program implementation from qualitative data, inductive thematic analysis was first performed to generate themes and codes from the interview transcripts with healthcare providers⁴⁴. To ensure consistency in coding and interpretation, an audit trail was maintained, and any discrepancies were resolved through consensus. Second, a deductive framework analysis was employed using the Theoretical Domains Framework (TDF) based on the results of the thematic analysis (Supplemental Table 2)⁴⁵. This mapping allowed for a structured interpretation of how each theme and code related to the theoretical constructs within the TDF, providing insights into the underlying determinants influencing behavioural decisions^45,46. A detailed analysis within each TDF domain was validated through cross-referencing with existing literature and expert discussions.

While the primary outcome, non-inferiority margin, and overall analytical framework were specified in the published protocol, some secondary analyses and implementation-related evaluations were refined during the conduct of the study.

Author contributions

A.W. and F.W. developed the conception and design of the initial study. A.W. was responsible for obtaining funding. A.W., S.C., and L.L. performed the statistical analyses, interpreted the data, and drafted the manuscript. J.L., D.T., M.C., M.W., B.W., V.H., C.T., W.H., and S.C. provided a critical review of the manuscript and intellectual input on the study design, methodology, and evaluation. All authors contributed to, reviewed, and approved the manuscript. All authors were responsible for the decision to submit for publication.

Data availability

All proposals for data use will need the approval of the study team before any data are released. Usage rights for the raw data are retained by the study team. De-identified individual participant data will be available upon reasonable request to researchers. Requests should be directed to the corresponding author via email, accompanied by a detailed research proposal. Access will be granted following the approval of the proposal and the signing of a data access agreement. Only data from participants who have consented to share their data will be provided. Data sharing will be authorized following proposal approval and the execution of a signed data access agreement, strictly for scientific use.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41746-026-02424-9.