Preferences of frail elderly patients with cardiovascular disease for web-based exercise telerehabilitation interventions in China: protocol for a discrete choice experiment study

Abstract

Introduction

Individuals with cardiovascular disease are prone to frailty, while frailty accelerates disease progression and worsens outcomes. Exercise interventions are essential non-pharmacological treatments for frail elderly patients with cardiovascular disease; however adherence remains low and strategies are inconsistent. Telerehabilitation improves accessibility, allows continuous monitoring with timely clinical responses and helps overcome mobility and geographical barriers for frail older adults. Guided by the patient-centred approach, exercise telerehabilitation should consider both clinical effectiveness and patient preferences to enhance acceptance and adherence. However, patient preferences for web-based exercise telerehabilitation remain poorly understood.

Methods and analysis

This study is designed as a discrete choice experiment to elicit and quantify preferences for key features of web-based exercise telerehabilitation among frail older adults with cardiovascular disease. Candidate attributes and levels were developed through a systematic literature review, patient focus groups and expert consultation. An orthogonal fractional-factorial design will generate the choice sets. A questionnaire survey will recruit older adults with cardiovascular disease who are identified as frail based on validated frailty assessment scales. Participants will be recruited from a cardiovascular specialty hospital in Xiamen, China, with a planned sample size of 157. Final data will be analysed using mixed logit models to estimate attribute importance, quantify preference weights and identify preference heterogeneity.

Ethics and dissemination

This study has received ethical approval from the relevant institutional ethics committee (2025–46). All participants will provide informed consent. Findings will be disseminated to patient groups, clinicians and policymakers and published in peer-reviewed journals and presented at national and international conferences.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• Adopts the discrete choice experiment (DCE) method to explore the research subjects’ preferences for interventions.

• Determines DCE attributes and levels via systematic literature review, qualitative research and expert consultation.

• Constructs DCE choice sets using an orthogonal experimental design to reduce the cognitive burden on frail elderly respondents.

• The single-centre study in China may limit the generalisability of the methods to other populations and regions.

• Relies on hypothetical DCE choice tasks, which have an inherent methodological deviation from real-world decision-making behaviour.

Introduction

China’s ageing population is projected to reach 480 million by 2050.¹ Age is the most significant risk factor for cardiovascular disease (CVD).² With demographic shifts, both the incidence and prevalence of CVD in older adults are expected to rise, imposing a substantial burden of morbidity, mortality and healthcare costs.² Frailty is a complex geriatric syndrome characterised by a significant decline in strength, endurance and physiological function.^{3 4} Studies have shown that the prevalence of frailty in the elderly with established CVD is higher than that in the general population, affecting up to 80% of patients with heart failure, 74% of patients with aortic stenosis and 30% of patients with coronary artery disease.²

The relationship between frailty and CVD is bidirectional, driven by shared risk factors (advanced age, lifestyle, chronic disease, adverse mental state) and overlapping pathophysiological mechanisms (inflammation, metabolic imbalance, coagulation disorders). Individuals with CVD are prone to frailty, while frailty accelerates disease progression and worsens outcomes.^{5 6} Therefore, integrating frailty into CVD management for older adults is essential to improve patient outcomes.⁶

Numerous studies have established exercise-based cardiac rehabilitation (CR) as a Class IA recommended intervention.^{2 7 8} Exercise rehabilitation is also the preferred strategy for preventing and treating frailty. Early exercise rehabilitation may delay CVD progression and potentially reverse frailty in older patients. However, despite robust evidence supporting the benefits of exercise, participation and adherence rates in traditional hospital-centred CR among older Chinese patients are generally low.^{2 7 8} Key barriers include transportation difficulties, time constraints, limited physical function and inadequate family support.⁹ Furthermore, growing evidence indicates that home-based exercise interventions for patients with CVD are effective. Existing studies have found that structured home exercise programmes can improve cardiopulmonary function, exercise endurance and heart rate recovery in patients with multiple cardiometabolic conditions during the intervention period.^{10 11} Despite these benefits, adherence to home-based interventions remains variable, particularly among older adults. Factors such as diminished motivation, lack of supervision and concerns about adverse events may impact sustained participation. These findings underscore the need for patient-centred strategies to optimise engagement.¹²

Web-based exercise telerehabilitation leverages the Internet, wearable devices and digital platforms to overcome geographical and time limitations, enabling home-based or community-based rehabilitation.¹³ The cost-effectiveness of such programmes has also been systematically evaluated, supporting sustainable healthcare implementation.¹⁴ Randomised trials have confirmed that home-based telerehabilitation is non-inferior to centre-based programmes in improving exercise capacity and quality of life.¹⁵ In recent years, web-based cardiac telerehabilitation has demonstrated efficacy comparable to traditional centre-based rehabilitation in improving exercise capacity, cardiac function and quality of life, with no increased risk of cardiovascular adverse events observed during long-term follow-up.⁹ Wearable sensor-based programmes, such as the TELEWEAR-CR (Telehealth Wearable‑based Cardiac Rehabilitation) study using real-time heart rate monitoring devices, have further demonstrated the feasibility of remote supervision for home-based exercise.¹⁶ Furthermore, remote monitoring technologies enable timely clinical feedback and risk management, further enhancing their safety and feasibility.^{16 17} Systematic reviews and meta-analyses have consolidated this evidence, supporting telerehabilitation as an effective alternative to traditional centre-based CR.¹⁸ This evidence provides a safety basis for its promotion among elderly patients with CVD. However, its applicability to frail elderly individuals may be limited by multiple implementation barriers, including insufficient digital literacy, limited access to technology (eg, smartphones, wearable devices, stable internet), inadequate training and technical support, as well as safety concerns and delayed emergency responses.¹⁹ Currently, discrete choice experiments (DCEs) investigating exercise preferences among frail elderly patients with CVD remain insufficient. Given this group’s heightened vulnerability and complex care needs, their preferences for web-based exercise telerehabilitation may differ from those of general patients. Notably, a feasibility study has demonstrated the applicability of information and communications technology (ICT)-based home CR for patients with heart failure and frailty,²⁰ yet patient preferences for specific programme features in this population remain unexplored.

In 2001, the Institute of Medicine’s report Crossing the Quality Chasm identified patient-centred care as one of six aims of high-quality healthcare.²¹ Patient-centred care is defined as care that is ‘respectful of and responsive to individual patient preferences, needs, and values’, where preference reflects the individual’s subjective inclination toward an outcome and underlies behaviour.²² Treatment adherence can be improved by identifying behavioural motivations from the patient’s perspective and incorporating preferences into therapeutic programmes.²³ Guided by the patient-centred approach, exercise telerehabilitation should consider both clinical effectiveness and patient preferences to enhance acceptance and adherence.²⁴ DCEs, a quantitative method based on choice behaviour, decompose rehabilitation programmes into key attributes and levels, simulate patient trade-offs and quantify the importance and substitutability of each element.^{23 24} Applying DCEs to frail elderly patients with CVD helps identify priority rehabilitation elements, reveal preference heterogeneity and provide an evidence base for stratified and personalised web-based telerehabilitation services. Therefore, this study aimed to: (1) identify essential components of web-based exercise telerehabilitation for frail elderly patients with CVD; (2) quantify patient preferences and examine related characteristics influencing choice; and (3) determine whether preferences vary by participant characteristics and classify patient subgroups by sociodemographic factors.

Methods and analysis

Design

A DCE was designed to investigate the preferences of frail elderly patients with CVD for web-based exercise telerehabilitation. The manuscript was prepared in accordance with the DIRECT (Discrete Choice Experiment Reporting Checklist) guidelines.²⁵ The DCE followed four main steps: (1) identifying attributes and levels; (2) designing the experiment and developing the questionnaire; (3) conducting the survey and collecting data; and (4) analysing data.

Attributes and identification levels

The attributes and levels included in the DCE survey were determined based on a targeted systematic review, patient focus groups and expert consultations conducted between July and August 2025.

A systematic review of six major English and Chinese databases was conducted to identify attributes relevant to web-based exercise telerehabilitation. This systematic review included 12 highly relevant articles in the analysis,2324 26,35 resulting in the identification of eight potential attributes: (1) safety and emergency response, (2) guidance mode, (3) programme structure, (4) incentive and reward mechanism, (5) interactive function, (6) charging mode, (7) training and technical support and (8) social participation mode. The detailed findings of the systematic review are presented in online supplemental material table S1.

Patients participated in the attribute development process. In alignment with previously published recommendations, qualitative work was conducted during attribute development.³³ Following the systematic review, patient focus group discussions were conducted to further explore the attributes and levels obtained from the literature review. Focus groups were chosen because they encouraged participants to reflect and express their subtle thought processes. We conducted a targeted sampling based on factors such as gender, age and educational level to ensure adequate representation. To assess the comprehensibility and relevance of the selected attributes and levels, we first organised a focus group comprising five frail elderly patients. During this discussion, no new attributes or levels emerged beyond those already identified through the literature review. Given concerns about sample adequacy, we conducted a supplementary focus group discussion with five additional participants using the same discussion guide. Similarly, this discussion did not yield any new themes, attributes or levels. Based on the principle of thematic saturation in qualitative research, subsequent discussions failed to generate new conceptual information, leading us to conclude that the data had reached saturation. It is worth noting that some elderly participants were relatively unfamiliar with the concept of online remote rehabilitation. Notably, some elderly participants were relatively unfamiliar with the concept of online remote rehabilitation. Therefore, we additionally integrated insights from multidisciplinary healthcare professionals specialising in geriatric rehabilitation and cardiovascular care.

This expert panel comprises seven rehabilitation nurses and three rehabilitation physicians, all with over 5 years of experience in cardiovascular exercise rehabilitation. All members are affiliated with specialised cardiovascular hospitals. Although formally classified as rehabilitation physicians or rehabilitation nurses, all members possess dual or interdisciplinary roles, with cardiology and cardiovascular nursing as their primary specialties. Consequently, this panel is deemed suitable for optimising the implementation of exercises in relation to patient preferences. Initial definitions and corresponding levels for each attribute were submitted to both patients and experts for review. Experts were invited to assess the importance, clinical feasibility and practicality of each attribute and level, and to provide supplementary suggestions for relevant attributes. Final DCE attributes and levels were determined based on consensus between patient feedback and expert priority rankings, ensuring both patient-centred relevance and clinical feasibility.

Consequently, we removed the two attributes of the interaction function and the incentive and reward mechanism due to low correlation or high redundancy. Other attributes and associated levels are not provided. Three experts pointed out that the communication and feedback mechanism referred to by the ‘interactive function’ attribute had already been covered by the four attributes of ‘safety and emergency response’, ‘exercise guidance mode’, ‘exercise participation mode’ and ‘training and technical support’. Keeping the ‘interactive function’ would cause content redundancy, so it was deleted. All patients and three experts noted that rewards offer only short-term incentives, whereas long-term adherence depends more on intrinsic motivation, making this attribute less important. All patients and experts believed that safety was the priority in rehabilitation medical treatment in remote settings. Due to their diseases, patients were doubtful about the safety of exercise rehabilitation, and offline rehabilitation had risks. Remote rehabilitation required a reliable safety guarantee and emergency response mechanism, making safety and emergency response more relevant considerations in remote rehabilitation. In addition, language adjustments were made to the remaining attributes and level representations to ensure they were understood by older subjects, based on patient and expert recommendations. In addition to methodological recommendations suggesting that the number of attributes in a DCE should generally not exceed six to maintain statistical efficiency and respondent manageability, particular consideration was given to the target population’s frailty and potential cognitive vulnerability. To reduce cognitive burden and decision fatigue among frail older adults, six attributes were retained as a balance between comprehensiveness and feasibility (table 1).³⁶

Table 1. Characteristics of attributes and levels included in the main survey.

Attributes	Levels	Levels description
Safety and emergency response	Self-protection mode	Only rely on self-perception/family members’ report, no equipment alarm.
	Basic wearable monitoring+family/system alerts	For example, equipped with a smart bracelet to synchronise heart rate, blood oxygen and blood pressure in real time. Relevant data are synchronised to the user terminal, professional personnel terminal and family terminal. When abnormalities occur, family members and the platform are automatically reminded.
	Multifaceted intelligent guarantee+rapid hospital response	Integrate multidimensional monitoring, wear smart bracelets and behaviour perception devices (such as smart shoes, radar and other devices to detect falls and movement deviations), build a virtual patient model and synchronise data to the user terminal, professional personnel terminal and family terminal in real time. When data is abnormal, an alarm is issued to assist family members/hospitals/emergency rescue teams to respond according to procedures.
Exercise guidance mode	Guidance via videos and graphic materials	Complete training independently, only according to standardised videos and graphic materials, upload records and professional personnel provide regular feedback and adjust exercise programmes.
	Regular artificial remote guidance	Professional personnel regularly provide guidance, feedback and adjustments through real-time video and voice according to the exercise programme.
	Virtual coach guidance	AI-driven virtual images provide standardised, dynamic guidance: real-time feedback, error prompts and personalised training suggestions based on users’ action data. Manual review is conducted regularly or as needed. This level does not require advanced digital literacy. Users should be able to master basic device operation and seek assistance from family members or caregivers when necessary.
Exercise programme structure	Fixed exercise mode	Standardised exercise programme based on clinical guidelines: including clear aerobic training, resistance training and flexibility training, formulated and adjusted by professionals according to participants’ conditions, with fixed schedules and clear duration/intensity.
	Fragmented exercise mode	Seamlessly integrate functional movements (such as tiptoeing, squatting and resistance training) into daily life scenarios (such as brushing teeth, cooking), and achieve regular physical activity without occupying extra time and space.
Charging method	Charged by service item	Independently priced for specific services such as remote monitoring, consultation and equipment, with a separate fee for each item.
	Package and membership-based charging	Package multiple services into packages or membership services, with one-time or regular payment covering comprehensive content.
	Pricing by risk stratification	Priced according to patients’ exercise risk levels. High-risk patients require more services and thus incur higher costs, whereas low-risk and medium-risk patients incur lower costs.
Training and technical support	Easy to use, no training required	The interface is extremely simple and can be used with assistance from family members.
	Short-term training	Once for 15–30 min, plus telephone support.
	Continuous technical support	Normal hotline+regular appointment for on-site/community support.
Exercise participation mode	Group exercise in community/park/gym	Offline group exercise rehabilitation in the community, where voice communication and mutual encouragement with others are possible during exercise.
	Individual exercise in community/park/gym	Offline individual exercise rehabilitation in the community, with no social interaction.
	Home individual exercise	Complete exercise alone at home, with no social interaction or only interaction with family members.
	Home virtual group exercise	Real-time online synchronisation (such as live follow-along, video conference interaction), where voice communication and mutual encouragement with others are possible during exercise.

AI, artificial intelligence.

The overall development of attributes and grades followed the methodological recommendations for DCEs outlined in the DIRECT guidelines. According to this framework, attribute identification should be accomplished through systematic literature reviews, qualitative research and expert consultation.²⁵ No single theoretical model or specific guideline was adopted as the sole framework for attribute selection. Instead, attributes were inductively derived from the evidence base and refined through consultations with patient focus groups and expert panels.

Experimental design and questionnaire development

After determining the experimental attributes and corresponding levels, it is necessary to construct choice sets with different combinations of attributes and levels through experimental design. Across the six attributes and their levels, a complete factorial design generated 1152 (32×43×2) hypothetical scenarios and 662976 [= (1152×1151)/2] pairwise choice tasks, which would be impractical for respondents. Efficient experimental design followed four principles: orthogonality, balance, minimal overlap and utility balance.²³ Therefore, to enhance the efficiency and precision of the study design, an orthogonal design using SPSS software (V.26.0) generated 25 manageable scenarios (online supplemental material table S2). One of the medium-level scenarios (Scenario 10) was designated as Option A because it represents a moderate combination of attribute levels without extreme values, thereby reducing the likelihood of dominant effects and maintaining balanced trade-offs across all alternative sets. It was then paired with the remaining 24 scenarios to form 24 alternative sets. Respondents were then asked to choose their preferred options. In addition, this study used a combination of pictures and text to present the choice tasks in order to help patients with different education levels better understand the options. An example of a DCE choice set is shown in figure 1. The following two additional considerations were made during the choice set design process: (1) To avoid exaggerating the relative weight of each attribute and improve questionnaire efficiency, this study did not include an opt-out option. However, this approach may not fully reflect real-world decision-making scenarios, particularly for frail elderly individuals.³⁷ (2) To reduce the respondents’ cognitive burden, the 24 choice sets were split into three versions of the DCE questionnaire (online supplemental material questionnaire).³⁸ Participants were randomly assigned to complete one version with eight DCE choice sets.

Figure 1. Example of a choice set.

Four sections comprised the questionnaire: an introduction, selection tasks, a general information sheet and a section on disease status. The introduction explained the study’s purpose, questionnaire requirements and the importance of exercise intervention. The section on selection tasks provides an overview of the attributes, levels and descriptions to help patients better understand the meaning of each attribute and level before selecting a protocol. The general information sheet includes several demographic variables (age, gender, level of education, occupation and marital status) and physical activity status, which may influence patient preferences. Professional clinicians will complete the disease status section. The purpose of the pre-experiment is to determine whether the questionnaire’s content is clearly expressed and easily understood. The questionnaire was pretested on 12 patients to assess both the clarity of attribute descriptions and the overall feasibility of the choice tasks. The average time required to complete the task for the 12 patients was 15 minutes. The majority of patients felt the survey was ‘appropriate in length’ and ‘easy to understand in content’, but some felt the attributes and levels were abstract and required an additional explanation from family members or investigators. As a result, we included examples of partial levels during the investigators’ training. Furthermore, the data from the pilot test will not be included in the final analysis of the main study.

Sample and recruitment

In this study, the inclusion criteria will be: (1) age: ≥60 years old, (2) diagnosed CVD: coronary heart disease, heart failure, hypertension, arrhythmia, etc, (3) frailty status: the Essential Frailty Toolset is a brief and objective instrument used to assess frailty among older patients with CVD.⁶ The toolset consists of four components, with total scores ranging from 0 to 5, classifying patients as non-frail (0), prefrail (1–2) or frail (3–5). The components include performance on five chair stands without arm assistance, cognitive function assessed by the Mini-Mental State Examination (MMSE), haemoglobin level and serum albumin level. The exclusion criteria will be included: (1) serious physical disorders or physical disabilities such as heart, liver and kidney, (2) Cognitive function was formally assessed using the MMSE, which was incorporated into this study as part of the frailty assessment toolkit. Participants with cognitive function below the predetermined MMSE threshold were excluded. Those unable to fully understand the study content were excluded. Sample size was calculated using Johnson and Orme’s³⁹ equation N>500 c/(t×a). In this formula, N represents the required sample size, and c is the largest number of levels across all attributes. The study included six attributes, each with three to four levels. The maximum number of levels was four for the attribute ‘exercise participation mode’ (table 1); therefore, c=4. The parameter t refers to the number of choice tasks per respondent, and a refers to the number of alternatives per task. In this study, each participant completed eight choice tasks. Each task presented two alternatives. Accordingly, t=8 and a=2. Based on the established guidelines, the calculation is as follows: N>500×4/ (8×2)=2000/16=125. Therefore, the minimum sample size required for the DCE is 125. Given that Johnson and Orme’s formula provides a conservative minimum for main effects estimation, we additionally adhered to Pearmain’s⁴⁰ recommendation of 100 respondents for robust preference estimation and subgroup analysis. Accounting for an ineffective response rate of 20%, the target was set at 157 participants.

Participants will be recruited at the study site across various cardiovascular departments, rehabilitation centres and rehabilitation clinics of a specialised cardiovascular hospital, and interested volunteers will be informed in detail about the purpose and process of the cross-sectional survey. The survey will be conducted in a separate, quiet room within the clinic. Two researchers will be present to assist patients and answer all their questions. The survey will be available in both paper and electronic formats, depending on the patient’s wishes and abilities. Electronic questionnaires will be collected via the anonymous online platform WenjuanXing (www.wjx.cn), and paper questionnaires will be entered independently by two researchers using Excel 2021 and systematically checked. Participants will be asked to carefully weigh each set of choices and then tick the box below the preferred option (A or B). The survey included a dominant choice set at the beginning to assess internal consistency, and those who failed this test were excluded from analysis. Questionnaires with straight-line responses to all items will also be excluded.

The inclusion of heterogeneous CVD populations (such as coronary artery disease, heart failure and arrhythmias) stems from the fact that exercise-based CR shares common core elements, safety principles and behavioural support strategies across these conditions. Given that this study aims to reveal patient preferences rather than evaluate clinical efficacy, a broader cardiovascular population perspective will enhance the generalisability of the findings. Furthermore, a mixed logic model and subgroup analyses will be employed to explore preference heterogeneity and identify potential subgroup differences. In future practice, personalised exercise recommendation systems could provide targeted guidance based on patient characteristics.

Statistics and data analysis

Data will be analysed using SPSS V.26.0 and R V.4.4. Choice data will be coded using dummy variables. Subsequently, a discrete choice modelling framework will be constructed to address the following objectives: (1) to assess preferences of frail older adults with CVD for web-based exercise telerehabilitation interventions; (2) to examine whether preference heterogeneity exists due to individual differences; (3) to explore heterogeneous preferences across subgroups with different characteristics; and (4) to evaluate the contribution of each attribute to overall preferences.

Model fit will be evaluated using the log-likelihood, Akaike information criterion and Bayesian information criterion. These indicators will be used to assess model performance and compare alternative model specifications. All statistical tests will be two-sided, with a significance level set at α=0.05.

A mixed logit model based on random utility theory will be used as the primary analytical approach to analyse the DCE data.⁴¹ This model will estimate overall preferences for web-based exercise telerehabilitation interventions (Objective 1). Model coefficients (β) will represent the relative utility associated with each attribute level. A positive (or negative) β will indicate that an attribute level is preferred (or not preferred) relative to the reference level. Larger absolute values will indicate stronger preferences. Statistical significance will be determined using 95% CIs. Unobserved preference heterogeneity (Objective 2) will be assessed through the random parameters in the mixed logit model. The SD of each coefficient will indicate the extent of variability in preferences across individuals. Subgroup analysis will be conducted to address Objective 3. Interaction terms between DCE attributes and participants’ baseline characteristics will be introduced into the mixed logit model. Sociodemographic variables and disease-related characteristics will be treated as independent variables, and participants’ choices of remote exercise telerehabilitation programmes will be treated as the dependent variable. This approach will be used to examine preference heterogeneity in referral choices across different demographic and clinical subgroups. Post-estimation analyses will be conducted based on the estimated model coefficients. The relative importance (RI) of each attribute will be calculated to quantify its contribution to overall preferences (Objective 4). Effect coding will be applied to all attributes and levels. The RI of each attribute will be derived by calculating the difference between the highest and lowest utility levels for that attribute and dividing it by the sum of differences across all attributes.⁴² All RI values will sum to 100%, with higher values indicating greater importance. Predicted uptake probabilities will also be estimated to support interpretation of preference results (Objective 1). Scenario-based simulations will be performed using the estimated coefficients to predict the probability of selecting specific intervention profiles under different attribute combinations. These probabilities will be calculated using post-estimation procedures (eg, the nlcom command in Stata V.18.0). This analysis will illustrate how changes in intervention design influence the likelihood of patient uptake.

For data quality control, incomplete questionnaires will be excluded from the final analysis. Missing responses will not be imputed due to the structured nature of DCE data. All responses will be screened for internal consistency and completion time to ensure data reliability.

Patient and public involvement

Patients were involved in specific stages of the DCE development process, including qualitative interviews to identify relevant attributes and pilot testing to assess questionnaire clarity and feasibility. Their input informed the refinement of attributes and levels to ensure relevance and comprehensibility for frail older adults with CVD. However, patients were not involved in the overall study design, data analysis, interpretation of results or manuscript preparation.

Ethics and dissemination

This study has been approved by the Ethics Committee of Xiamen Cardiovascular Hospital, Xiamen University (registration number 2025–46, registration date 4 September 2025). Patient recruitment for this study began in September 2025 and is expected to end around June 2026. In accordance with the principles of voluntariness and confidentiality, the investigator will explain the background, purpose and potential risks of the study to the patients or specialists participating in the interview and survey, and participants will be required to sign a written informed consent form before participating in the study. All interview materials and questionnaires will be used only for this study and are provided to researchers in an anonymous manner to ensure confidentiality. Patients can withdraw from this research at any time. Data analysis will be performed in accordance with the principles of good scientific research on DCEs, as developed by the International Society for Pharmacoeconomics and Outcomes Research. Our findings will be disseminated to interested patient groups and the general public through online blogs, policy briefs, national and international conferences and peer-reviewed journals. Data are available in a public, open-access repository.

Discussion

Although telemedicine is expected to improve resource allocation, its success ultimately depends on patient acceptance and adherence.²⁴ For frail elderly populations, successful implementation requires particular attention to digital literacy barriers, equitable access to necessary technology and internet connectivity and comprehensive training and technical support systems. These factors, which are explicitly incorporated as attributes in our DCE, represent critical determinants of real-world engagement and should inform service design and policy planning.

Therefore, future telemedicine services should be designed from the patient’s perspective. This study employed a DCE to examine the preferences of frail elderly patients with CVD for remote rehabilitation, providing an empirical basis for designing and implementing patient-centred interventions. It will also identify which future telerehabilitation designs are most preferred and which treatment attributes are most important.

The main strengths of this study are as follows: first, targeting frail elderly patients with CVD, a high-risk subgroup that has been neglected in telerehabilitation research, could fill the evidence gap and improve the accessibility and targeting of interventions. Second, attributes and levels were generated through literature review, focus groups and expert consultation. A standardised DCE design and mixed Logit/ latent class modelling were used to characterise heterogeneity.²³ In addition, the resulting preference weights can be directly used to optimise service elements (such as safety monitoring, remote guidance models and payment methods), guide the selection of elements for pilot interventions and provide actionable evidence for implementation research and policy development.⁴³

The preference weights generated by this study provide a reference for quantitatively assessing frail elderly individuals’ relative emphasis on different functional elements of online cardiac remote rehabilitation. By delineating the relative strength of preferences for each attribute and the potential trade-off ranges that individuals may accept, the findings offer decision-making guidance for service model design to some extent. During project development, attributes with relatively stronger utility effects may warrant priority consideration, while components with lesser impact could potentially allow for some flexibility in implementation. Based on this, the findings offer insights for optimising the architecture of remote rehabilitation services, such as exploring phased or differentiated implementation strategies under varying clinical risk levels or support needs. Concurrently, as real-world safety evidence accumulates and remote monitoring technology for wearable devices advances, the feasibility of moderately adjusting programme intensity based on patient-valued elements may increase. However, further practical validation remains necessary.

Despite these strengths, several limitations should be acknowledged. First, this study used a single-centre cross-sectional design, which may limit extrapolation given the sample size and cultural background. Second, DCEs reflect hypothetical choices, which may not fully align with actual behaviour. Third, potential selection bias should be considered. Participants who had access to digital devices and were willing to engage in a web-based survey may differ systematically from those with limited digital access or lower motivation to participate, potentially affecting the representativeness of the results. In addition, limited digital literacy and the cognitive burden of frail elderly patients with CVD may affect the representativeness of the results. Finally, patient preferences for telemedicine may evolve over time as individuals gain experience with telerehabilitation or as their health status changes. Because this study captured preferences at a single time point prior to exposure to the intervention, we were unable to assess temporal changes in preferences. Future multicentre longitudinal studies with more inclusive sampling strategies are warranted to validate and extend these findings.