Coronary artery bypass grafting patients’ preferences and willingness to pay for remote cardiac rehabilitation: discrete choice experiment study

Abstract

Background

Cardiac rehabilitation (CR) uptake among coronary artery bypass grafting (CABG) patients in China is extremely low. Remote cardiac rehabilitation (RCR) offers a promising alternative, but its design must be informed by patient’s preferences to ensure adoption. This study aimed to quantify CABG patients’ preferences and willingness-to-pay (WTP) for key RCR attributes to inform patient-centered service design.

Methods

A discrete choice experiment was administered to CABG patients, assessing preferences across six RCR attributes: service content, service provider, degree of personalization, monitoring method, content model, and cost. Data were analyzed using a mixed logit model to estimate preferences, WTP, and relative importance (RI). Subgroup analyses by gender and age were performed.

Results

A total of 343 CABG patients participated in the study, of whom 69.39% were male, with a mean age of 61.69 (SD = 10.33) years. Within the present DCE model, service provider emerged as the most critical attribute (RI = 45.54). Patients strongly preferred services led by multidisciplinary teams (β = 1.70^***, WTP = CNY 92.65/month), followed by doctor-led services (β = 1.09^***, WTP = CNY 59.21/month). Continuous automatic monitoring via wearable devices (β = 0.90^***, WTP = CNY 49.32/month) was also highly valued. Personalized plans based on discharge assessment were preferred over standardized fixed protocols (β = 0.36^*, WTP = CNY 19.50/month). Cost had a significant negative impact on choice (RI = 8.21; β=-0.02^**). Video content was preferred over text (β = 0.52^***, WTP = CNY 28.33/month), while voice format showed no significant effect. In this subgroup analysis, male patients showed significant preferences for a wider range of attributes, whereas female patients showed significant preferences primarily for service provider. Younger patients (≤ 60 years) also favored personalized based on discharge assessment, while older patients (> 60 years) preferred AI assistant + regular review by team members and comprehensive programs.

Conclusions

CABG patients prioritize professional, integrated support and convenient technology in RCR. Program development should emphasize multidisciplinary teams, wearable devices, and video content. These patient-informed strategies are essential to enhance engagement and address the low CR uptake.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12913-026-14817-3.

Background

Coronary artery bypass grafting (CABG) remains the primary surgical treatment for complex coronary artery disease [1]. In China, the number of CABG procedures has increased steadily, reaching 80,016 cases in 2024, reflecting an 11.2% growth [2]. Despite improvements in intraoperative techniques and in-hospital care, long-term outcomes of CABG patients remain threatened by a critical gap in secondary prevention. Cardiac rehabilitation (CR), a comprehensive intervention strategy encompassing exercise, nutrition, psychological support, and health education, is strongly recommended as a Class I, Level A intervention by international guidelines for its proven efficacy in reducing mortality and hospital readmissions, and improving functional capacity and quality of life [3–9].

However, CR uptake among CABG patients in China is extremely low [10–12]. This low participation reflects the uneven distribution of healthcare resources in China, with only 23% of healthcare institutions offering CR programs and fewer than 54% of these providing both phase I and phase II rehabilitation [13]. Furthermore, postoperative patients are often hesitant to engage in CR due to inadequate health literacy, lack of motivation, and prevalent fear of movement [10, 14]. In addition, financial barriers further limit CR participation, as CR services are not yet fully integrated into the basic medical insurance system, leaving patients to bear most of the costs out-of-pocket [15, 16]. This low participation represents a major gap in healthcare delivery, as it is closely associated with poor post-discharge continuity of care, resulting in preventable negative outcomes including significant decline in quality of life, recurrent angina, increased readmission risk, and higher long-term healthcare costs for CABG patients [14, 17–22].

The rapid advancement of mobile health and telemedicine offers a promising solution to these barriers. Remote cardiac rehabilitation (RCR) utilizes digital technologies to provide remote guidance and monitoring, potentially overcoming traditional access limitations [23, 24]. For a health system, RCR represents a scalable and potentially cost-effective strategy to expand CR coverage. Nevertheless, current RCR services remain nascent, with one survey reporting that only 5.7% of patients participating in CR engaged in remote formats [25, 26]. The successful implementation and uptake of such innovative health service models depend fundamentally on understanding and incorporating patient preferences [27]. Without robust data on the attributes patients value in an RCR program, developers and clinicians lack the evidence needed to design services that patients will adopt and sustain long-term.

The Discrete Choice Experiment (DCE) is a robust quantitative method drawn from health economics that quantifies how individuals trade off different service attributes, thereby identifying the features that drive acceptance and value [28, 29]. Although previous research has begun exploring patient’s preferences for CR, a critical gap remains. Existing qualitative studies and DCEs have largely focused on mixed cardiovascular populations or patients undergoing percutaneous coronary intervention [12, 30], and their findings may not be directly generalizable to CABG patients. CABG patients represent a distinct subgroup with unique challenges. They typically undergo more invasive and complex surgical procedure, experience longer and more strenuous recovery periods, and often have a higher comorbidity burden [17, 19]. These factors may translate into distinct preferences, including a greater need for professional oversight, personalized rehabilitation plans, and enhanced safety monitoring. However, the specific preferences and willingness to pay (WTP) of CABG patients for RCR attributes remain unknown, posing a significant barrier to designing effective and equitable services for this high-risk group.

Therefore, this study employed a DCE to systematically investigate the preferences, WTP, and relative importance (RI) that CABG patients assign to key RCR program attributes. By addressing this specific evidence gap, this research aims to generate crucial evidence to directly inform the design of patient-centered, mHealth-facilitated rehabilitation services tailored to the unique needs of the CABG population.

Methods

Participants

This study was conducted at the Department of Cardiac and Vascular Surgery of Shandong Provincial Hospital in China. Using convenience sampling, we recruited CABG patients between January and September 2025. The department performs over 3,500 cardiac surgeries annually, including approximately 2000 CABG procedures. The hospital serves a population of approximately 100 million people across Shandong Province and its surrounding areas. Its patient base includes individuals from both urban centers and rural areas; approximately 40% of the province’s population resides in rural areas, reflecting the broader demographic distribution across China. At the time of this study, the department offered traditional center-based CR services to a limited number of patients, but there was no formal RCR program in place, which provided an opportunity to assess patient preferences for future RCR service development.

Inclusion criteria were: (a) age ≥ 18 years; (b) first-time CABG surgery, including isolated CABG or CABG combined with other procedures such as valve surgery; (c) clear consciousness, able to communicate effectively and understand the questionnaire; (d) voluntary participation and provision of written informed consent. The exclusion criteria were: (a) mental disorders or cognitive impairments that prevented cooperation with the survey; (b) uncontrolled severe arrhythmias, hypertension, fever, acute severe infections, or multiple organ failure that rendered patients unsuitable for CR exercise intervention upon evaluation.

Measures

Clinical and socio-demographic information questionnaire

A self-designed questionnaire was used to collect participants’ clinical and sociodemographic characteristics. The questionnaire included the following variables: gender, age, BMI, education level, marital status, type of health insurance, residence, occupational status, monthly income, surgical status (isolated CABG vs. CABG with concomitant surgery), number of bypass grafts, postoperative New York Heart Association (NYHA) functional class, and presence of other chronic diseases.

Determining attributes and levels in the discrete choice experiment

Identifying attributes and their levels is a critical step in conducting a DCE [31–33]. This study employed a multi-stage approach that combined literature analysis, patient interviews, and group discussions to develop the final attributes.

(a) Literature review. The literature search was conducted in PubMed and Web of Science for publications up to November 30, 2024, using search terms including “Cardiac Rehabilitation,” “Heart Rehabilitation,” “Coronary Artery Bypass Grafting,” “CABG,” “remote,” “telemedicine,” “eHealth,” and “mobile health.” Cross-sectional surveys, qualitative studies, and reviews that addressed CABG patients’ preferences for CR were screened to extract potential RCR service attributes. This process initially identified multiple candidate attributes, such as service content, service provider, monitoring method, and cost.

(b) Patient interviews. To supplement any potential gaps in the literature and directly obtain patient perspectives, semi-structured interviews were conducted with a purposive sample of 8 CABG patients who met the same inclusion criteria as the main study. These interviewees were independent of the 343 patients enrolled in the subsequent DCE survey to ensure that the attribute development process did not influence the final study sample. The interviews aimed to understand patients’ awareness, preferences, and concerns about RCR. Interviews were conducted by a trained cardiac surgery nurse (DY S), each lasted approximately 10–15 min. All interviews were audio-recorded with participant consent and transcribed verbatim within 24 h. Two researchers (DY S and SS L) independently coded the transcripts using a deductive approach to thematic analysis [34], which is suitable for identifying manifest content based on pre-existing frameworks. Discrepancies were resolved through discussion. The interviews revealed four key themes: preferences for personalized rehabilitation programs, concerns about service costs and insurance coverage, a need for real-time monitoring and emergency response capability, varying perspectives on data privacy and security.

(c) Group discussions. Based on the literature review and patient interview findings, the research team initially identified 10 potential attributes: service content, service provider, degree of personalization, monitoring method, frequency of sessions, content model, out-of-pocket cost, reimbursement/insurance coverage, data privacy/security, emergency response capability. To ensure clinical relevance and practicality, we convened a group discussion. The panel consisted of a chief cardiac surgery physician, a cardiac rehabilitation nurse specialist, and two cardiac surgery charge nurses. The experts thoroughly evaluated the 10 initial attributes based on clinical experience, patient interview themes, and the methodological requirements of DCE design.

The panel unanimously agreed to exclude the following four attributes for the reasons provided below:

• Frequency of sessions: This attribute was considered implicitly captured within “Service content” and “Degree of personalization,” which would make it redundant if listed separately.

• Data privacy/security and Emergency response capability: The panel regarded these as essential prerequisites for any acceptable RCR service rather than negotiable attributes for patients to trade off when choosing among service options. To avoid confusion in the choice tasks, we addressed and assured these aspects in the introductory section of the DCE questionnaire.

• Reimbursement/insurance coverage: This attribute was deemed closely related to “Out-of-pocket cost”; including both would create attribute overlap and increase respondents’ cognitive burden.

To avoid choice overload, we limited the number of attributes to between 5 and 8, as recommended in the literature [35–37]. The panel unanimously agreed on six core attributes: (1) Service content; (2) Service provider; (3) Degree of personalization; (4) Monitoring method; (5) Content model; (6) Cost. The complete list of attributes and levels is presented in Table S1 (Appendix A).

(d) Experimental design. In designing the DCE choice sets, the principles of balance, orthogonality, and minimal overlap were followed [38, 39]. A fractional factorial design was employed, and 24 choice sets were generated using a D-efficient design. These choice sets were randomly divided into three versions, each containing 8 choice sets [40]. To prevent participants from being unable to express no preference when choosing between A and B, the study also included a “neither” option. To assess internal consistency, we repeated one choice set (i.e., we asked participants to complete the same choice task twice). Therefore, each version contained a total of 9 choice sets. An example of a choice set is shown in Fig. 1. See Appendix B for the full questionnaire.

Fig. 1.

An example of discrete choice experiment selection sets. Notes: CNY: Chinese yuan, ¥, 1¥ ≈ 0.14 USD (based on the average exchange rate during the study period, January–September 2025)

Sample size

This study used the Orme equation to estimate the sample size for the DCE questionnaire. The formula for calculating the minimum acceptable sample size is as follows: n ≥ 500 C / ta [41, 42]. Based on the number of choice sets per version (t = 8), the number of options per set (a = 3), and the maximum number of levels for any attribute (C = 4), the minimum required sample size for each version was calculated as 84 participants. Because the study employed three versions of the questionnaire, the total minimum required sample size was 252 participants. To improve the accuracy of parameter estimates and ensure that the pre-specified subgroup analyses had sufficient statistical power, we included 343 valid responses in the final analysis.

Data collection

The survey team comprised the researchers and nursing staff from the relevant departments. Prior to distributing the questionnaires, training and guidance were provided to the team members. Investigators were required to familiarize themselves with the questionnaire content, understand the meaning of each question and its options, and be aware of the precautions for completing the questionnaire. Patients were approached and completed the survey after they were transferred from the cardiac surgery intensive care unit to the general ward, once their condition had stabilized. Face-to-face surveys were conducted with patients who met the inclusion criteria and did not fall under the exclusion criteria. The completed questionnaires were checked on-site and collected immediately.

Statistical Analysis

Descriptive statistics for participant characteristics were analyzed using SPSS 26.0, with continuous variables presented as means (SD) and categorical variables as frequencies (%). The primary analysis of the DCE data was performed using Stata 16.0. Both conditional logit and mixed logit models were estimated. Attribute levels were coded using dummy coding, with the first level of each attribute specified as the reference category. Model fit was assessed using the Akaike Information Criterion (AIC), the Bayesian Information Criterion (BIC), the Log-Likelihood (LL), and adjusted R^². Lower AIC, BIC, and absolute LL values, and a higher adjusted R^², indicate better model fit [43]. The model with superior fit statistics was selected for primary reporting of results. To account for differences in respondents’ preferences, all non-cost attribute levels were set as random parameters following a normal distribution and were allowed to be correlated, while the cost attribute was treated as a fixed continuous variable. Based on the model coefficients, the RI of each attribute was calculated as the percentage contribution of its utility range (the difference between the highest and lowest part-worth) to the sum of the utility ranges across all attributes. For calculating RI, cost was treated as a categorical variable to ensure consistency with the other attributes. WTP for specific attribute levels was derived using the formula: WTP = -β attribute/β cost. The 95% confidence intervals (CI) for WTP were estimated using a bootstrap procedure with 5,000 replications. CIs were derived using the percentile method (2.5th and 97.5th percentiles). To further explore preference heterogeneity, subgroup analyses were conducted based on gender (male vs. female) and age (≤ 60 vs. >60 years). Statistical significance was defined as a two-sided P value < 0.05. All data presented in the tables and text have been rounded to two decimal places for consistency and clarity, while all statistical analyses were performed using the original, unrounded data.

Results

Sociodemographic and clinical characteristics

A total of 343 CABG patients were included in this study, including 238 males and 105 females. The mean age was 61.69 (SD = 10.33) years. Education levels were primarily at primary school or below (33.53%), and 79.30% were married. The majority had urban and rural resident medical insurance (62.46%) and most resided in village (42.57%). Additionally, 76.68% were unemployed, and 52.37% had a monthly income of <CNY3000. A total of 179 patients underwent isolated CABG surgery, while 164 patients underwent CABG combined with other procedures. Grafting with fewer than three vessels was performed in 58.02% of cases. Postoperative heart function (NYHA classification) was primarily class II (54.90%), and 66.47% had comorbid chronic diseases. Detailed information is provided in Table 1.

Table 1.

Descriptive statistics for sociodemographic and disease information (n = 343)

Variables	n (%)/ M(SD)
Gender
Male	238(69.39)
Female	105(30.61)
Age, years	61.69(10.33)
BMI
Underweight	7(2.04)
Normal weight	160(46.65)
Overweight	111(32.36)
Obese	65(18.95)
Educational level
Primary school or below	115(33.53)
Junior high school	116(33.82)
High school/high vocational school	87(25.36)
College education or higher	25(7.29)
Marital status
Married	272(79.30)
Single/widowed/divorced	71(20.70)
Types of health insurance a
Urban employee medical insurance	120(35.19)
Urban and rural resident medical insurance	213(62.46)
Commercial Insurance	8(2.35)
Residence
City	80(23.32)
Town	117(34.11)
Village	146(42.57)
Occupational status
Unemployed	263(76.68)
Employed	80(23.32)
Monthly income (CNY) b
<3000	177(52.37)
3000–5000	102(30.18)
5000–8000	47(13.91)
>8000	12(3.55)
Surgical Status
CABG	179(52.19)
CABG + Other Surgeries	164(47.81)
Number of bypass surgeries
<3	199(58.02)
≥ 3	144(41.98)
New York Heart Association (After surgery) c
Ⅰ	76(22.55)
Ⅱ	185(54.90)
Ⅲ	69(20.47)
Ⅳ	7(2.08)
Presence of other chronic diseases d
Yes	222(66.47)
No	112(33.53)

Notes: ^a Two missing values; ^b Five missing values; ^c Six missing values; ^d Nine missing values; CABG: Coronary Artery Bypass Grafting; CNY: Chinese yuan, ¥, 1¥ ≈ 0.14 USD (based on the average exchange rate during the study period, January–September 2025)

Results from the mixed logit model

The mixed logit model demonstrated superior fit when compared to the conditional logit model. Therefore, results from the mixed logit model are presented below. (The results of the Conditional Logit model were referred to Table S2 in the appendix A, and the results of the Mixed Logit model were presented in Fig. 2).

Fig. 2.

Results of mixed logit model analysis on patients’ preferences for remote cardiac rehabilitation services (n = 343). Notes: AIC = 4228.80; BIC = 4370.80; LL = -2090.39; Adjusted R²=0.30; RI: relative importance; ^*P<0.05; ^**P<0.01; ^***P<0.001; CNY: Chinese yuan, ¥, 1¥ ≈ 0.14 USD (based on the average exchange rate during the study period, January–September 2025)

The results showed that service provider, degree of personalization, monitoring method, content model, and cost had a significant impact on patients’ choices of RCR services (P < 0.05), whereas service content did not have a significant impact (P > 0.05). Specifically, compared to services provided by nurses, patients preferred RCR services provided by multidisciplinary teams (β = 1.70^***), doctors (β = 1.09^***), or AI assistants + regular review by team members (β = 0.63^*). In degree of personalization, the personalized based on discharge assessment (β = 0.36^*) was preferred over the standardized fixed protocol. Regarding the monitoring method, compared to manual data upload, patients preferred continuous automatic monitoring via wearable devices (β = 0.90^***). Compared to text format, video format was preferred (β = 0.52^***), and voice format had no significant effect. The cost had a significant negative impact on patients’ choice (β= -0.02^**), indicating that the higher the cost, the less likely patients were to choose the service.

In the present DCE model, the RI analysis showed that the service provider had the greatest impact on patients’ choices (RI = 45.54), followed by the monitoring method (RI = 16.43), service content (RI = 15.89), degree of personalization (RI = 8.75), and cost (RI = 8.21). Content model had the lowest RI (RI = 5.18).

Willingness to pay for remote cardiac rehabilitation services

For service provider, patients had the highest WTP for the multidisciplinary team service model, at CNY 92.65/month, followed by services provided by doctors (CNY 59.21/month) and by AI assistant combined with regular team reviews (CNY 34.34 /month). In terms of degree of personalization, patients were willing to pay CNY 19.50/month more for personalization based on discharge assessment. For monitoring method, patients were WTP CNY 49.32/month more for continuous automatic monitoring via wearable devices. Regarding the model, the WTP for video format is CNY 28.33/month higher than for text format. Please refer to Table 2 for details.

Table 2.

Analysis of patients’ willingness to pay for remote cardiac rehabilitation services

Attributes and levels	WTP (CNY)	95%CI
Service content
Physical exercise training	Reference
Physical exercise training; Medication management	-1.60	(-19.50, 16.35)
Physical exercise training; Medication management; Nutritional guidance	4.04	(-17.82, 26.30)
Comprehensive program (exercise; medication; nutrition; psychological and behavioral interventions)	18.68	(-5.27, 42.67)
Service provider
Nurse	Reference
Doctor	59.21	(40.18, 78.71)
Multidisciplinary team (doctors, nurses, rehabilitation therapists, psychological counselors, nutritionists, etc.)	92.65	(68.74, 117.27)
AI assistant + regular review by team members	34.34	(7.58, 60.18)
Degree of personalization
Standardized fixed protocol	Reference
Personalized based on discharge assessment	19.50	(2.85, 36.88)
Real-time AI-driven protocol adjustment	7.29	(-22.17, 36.88)
Monitoring method
Manual data upload	Reference
Continuous automatic monitoring via wearable devices	49.32	(22.28, 76.36)
Content model
Text	Reference
Voice	2.76	(-18.70, 23.93)
Video	28.33	(11.06, 45.09)

Notes: WTP: Willingness to pay; CNY: Chinese yuan, ¥, 1¥ ≈ 0.14 USD (based on the average exchange rate during the study period, January–September 2025)

Subgroup analysis

Gender

The subgroup analysis of gender differences in CABG patients’ preferences for RCR revealed notable differences in service attribute preferences, WTP, and RI. Male patients showed significant preferences for several service attributes. Male patients showed significant preferences for a comprehensive rehabilitation program (β = 0.79^**), doctor-led services (β = 1.00^***), multidisciplinary team services (β = 1.84^***), AI assistant + regular review by team members (β = 1.36^***), personalized plans based on discharge assessment (β = 0.39^*), continuous automatic monitoring via wearable devices (β = 1.58^***), and video formats (β = 0.77^***). Cost had a significant negative effect (β = -0.04^***). Female patients showed significant preferences for multidisciplinary team services (β = 1.63^***) and AI assistant + regular review by team members (β = 0.98^*), but did not demonstrate significant preferences for other attributes.

In terms of RI, both groups ranked service provider as the most important attribute (male: RI = 43.14; female: RI = 52.44). Male patients placed greater importance on monitoring method (RI = 16.97) and service content (RI = 13.58), whereas female patients prioritized cost (RI = 28.21). For more details, please refer to Table S3 in Appendix A.

Age

Both younger (≤ 60 years) and older (> 60 years) patients showed significant preferences for doctor-led services, multidisciplinary team services, continuous automatic monitoring via wearable devices, and video formats (all P < 0.01). Cost had a significant negative effect in both groups (P < 0.01). In addition, younger patients preferred personalization based on discharge assessment (β = 0.79^**). Older patients also preferred AI assistant + regular review by team members (β = 0.76^**) and a comprehensive rehabilitation program (β = 0.85^**).

In terms of RI, service provider ranked highest in both groups (≤ 60 years: RI = 30.45; >60 years: RI = 40.74). Younger patients placed greater importance on monitoring method (RI = 17.07) and degree of personalization (RI = 15.57), whereas older patients prioritized service content (RI = 40.14). For more details, please refer to Table S4 in Appendix A.

Discussion

This study provides comprehensive quantitative evidence regarding the preferences and WTP of CABG patients for RCR. Within the attribute set and levels defined in this study, service provider had the largest impact on patient choices. Significant heterogeneity across patient subgroups underscores the need for tailored service design. Although this study was conducted in a single region in China, the findings may provide transferable insights for other health systems facing similar challenges in CR uptake, particularly those with resource constraints or nascent digital health infrastructure. Understanding patient preferences in such contexts can inform the design of scalable RCR interventions that align with patient priorities.

Among the attributes examined in this study, service provider was the most influential. Patients preferred multidisciplinary team, followed by doctor-led services and AI assistant + regular review by team members. This pattern may reflect the need for professional oversight and perceived safety during the postoperative period. Patients may view multidisciplinary teams as a reliable structure that integrates care and reduces the risk of complications. These findings align with prior research on the effectiveness of multidisciplinary teams in chronic disease management [44–47], but the DCE results quantify the substantial economic value patients attach to this model. The relatively lower coefficient for the AI assistant + regular review by team members model suggests that, while patients are open to automation, they retain greater trust in human-led care, likely due to concerns about reliability during recovery. This supports existing evidence on hybrid human-AI models in digital health [48–50], offering a pragmatic approach for health systems aiming to expand service reach while maintaining perceived quality of care.

The strong preference for continuous automatic monitoring via wearable devices reflects patients’ valuation of low-burden data collection. This finding aligns with the broader goal of digital health of integrating care seamlessly into daily life [51]. Previous studies have shown that wearables can enhance engagement and outcomes in home-based CR [52]. The results extend this evidence by quantifying patients’ WTP for such monitoring, providing an economic rationale for integrating wearables into RCR programs. From an implementation perspective, this preference underscores the need to ensure data reliability and support clinicians in interpreting device outputs, particularly given known limitations related to activity type and intensity [53].

Patients preferred personalized plans based on discharge assessment over standardized fixed protocols. In contrast, the preference for real-time AI-driven protocol adjustment did not reach statistical significance. This difference may reflect the fact that patients value initial personalization but are less convinced about the added value of AI-based dynamic adjustments. Personalized discharge planning has been shown to improve patient engagement and outcomes, while AI-driven personalization may offer benefits in CR [54]. However, the findings suggest that continuous AI-driven adjustments may not be a priority for patients, and such features may require clearer communication of their value to gain acceptance.

Given the hypothetical nature of the DCE, these WTP estimates should be interpreted as stated rather than revealed preferences [55]. Cost had a relatively low RI, ranking fifth among the six attributes, suggesting that patients were less sensitive to price compared to other attributes. However, the significant negative coefficient for cost indicates that patients still preferred lower-priced options, consistent with typical economic expectations. Social desirability bias may have led patients to overstate their WTP relative to real-world behavior. It is also possible that patients viewed the listed cost levels as reasonable given the expected benefits, or that they assumed that insurance would cover part of the expenses. Future studies using actual financial commitments or revealed preference methods are needed to further explore these factors.

Although content model had the lowest RI among the attributes, patients showed a clear preference for video formats over text formats. This preference may be explained by differences in cognitive load and health literacy, given that a considerable proportion of participants had an educational level at junior high school or below. Video formats deliver information in a visually accessible manner, potentially improving comprehension and retention compared to text-based materials [56]. Similar preferences for video have been reported in other digital health contexts [57], and video-based rehabilitation has been associated with satisfactory patient engagement [58]. These findings suggest that while content model may not be a primary driver of patient choice, incorporating video formats could still support information delivery in RCR programs.

Regarding service content, its impact on patients’ preferences did not reach statistical significance, despite a RI of 15.89%. One possible explanation is that hospitalized patients recovering from surgery may prioritize immediate concerns, such as who delivers the care and how it is delivered, over the specific components of the program. More nuanced elements, such as psychosocial support and nutritional guidance, may become more relevant later in the recovery process. Additionally, the attribute levels may not have provided sufficient contrast to capture meaningful differences, and collinearity with other attributes cannot be ruled out. This suggests that when designing RCR programs for hospitalized CABG patients, initial efforts should focus on establishing trusted care teams and clear delivery mechanisms, with comprehensive content introduced progressively as patients advance through recovery. It is worth noting that RI reflects the utility range within an attribute, whereas statistical significance tests each level against the reference. For service content, the coefficients showed numerical differences but had large standard errors; therefore, no level was significantly different from ‘physical exercise training’, and we cannot reliably identify which specific level is preferred.

Subgroup analyses revealed notable heterogeneity in preferences across gender and age groups. Male patients’ engagement with both team-based and technology-driven features may indicate a greater inclination toward efficiency and autonomy during recovery. The higher priority placed on cost by female patients suggests that financial considerations may play a more prominent role in their decision-making. These findings align with previous research suggesting that women may have different priorities and barriers in CR participation [59, 60]. Age-related differences were also observed. Younger patients, generally more familiar with digital tools, may value features that offer convenience, personalization, and real-time feedback. Older patients, on the other hand, may place greater importance on the comprehensiveness and clarity of service content, reflecting a greater need for detailed information and reassurance during recovery [61, 62]. The additional preference among older patients for AI assistant with regular review, combined with their strong preference for multidisciplinary team oversight, suggests that they may accept automation when paired with human supervision rather than as a standalone feature.

Limitations and future research

This study has several limitations that warrant consideration. First, convenience sampling from a single center in Shandong Province may introduce selection bias and limit the generalizability of the findings. The study sample may not fully represent the diverse needs of patients across different regions or healthcare tiers in China. Future multicenter studies across varied geographic and healthcare settings are needed to validate these findings and explore regional differences in patient preferences. Second, the inclusion criteria of this study required participants to have clear consciousness and the ability to communicate effectively and understand the questionnaire, which may have led to the underrepresentation of patients with lower health literacy, cognitive impairments, or limited digital capability. Therefore, the findings may not fully capture the preferences of these subgroups, who may face greater barriers to engaging in RCR. Future research should specifically target these underserved populations to ensure equitable service design. Third, the D-efficient design used in this study was generated using uninformative priors due to the absence of prior parameter estimates. Future studies may benefit from conducting pilot work to obtain informative priors for more efficient experimental designs. Fourth, subgroup analyses were limited to gender and age, leaving other sociodemographic factors unexplored. Future studies could apply latent profile analysis or clustering methods to classify patients into distinct subgroups based on multidimensional characteristics, and further examine the differences in preferences across these subgroups. Fifth, the subgroup analyses by gender and age should be considered exploratory, as the sample sizes for female subgroup (n = 105) may limit statistical power to detect smaller effects. Moreover, no formal interaction tests were conducted; therefore, statements about differences in the number of significant attributes across groups are descriptive only. Nevertheless, these findings provide valuable preliminary insights into preference heterogeneity across patient subgroups and can inform hypothesis generation for future confirmatory studies with larger, more diverse samples. Sixth, the WTP estimates were derived as a ratio involving the cost coefficient. Given the relatively small cost coefficient, these estimates may be sensitive and should be interpreted as indicative of relative preferences rather than precise monetary values. Future studies using WTP-space models could further validate the robustness of these estimates.

Implications for practice

The findings of this study provide concrete, evidence-based guidance for clinicians, healthcare administrators, and digital health developers. To address the documented gap in CR uptake, the design of RCR programs should be fundamentally informed by the patient preferences identified here. Key priorities include establishing multidisciplinary collaborative networks with clearly defined roles for cardiac surgeons, rehabilitation nurses, physiotherapists, and nutritionists, supported by a designated care coordinator to ensure continuity of remote care. For wearable technologies, integrating reliable devices that enable continuous automatic monitoring should be accompanied by patient education during discharge and protocols for data integration into electronic health records. For video content, developing culturally appropriate educational videos covering exercise, medication, nutrition, and psychological support, and making them accessible via widely used platforms, can enhance information delivery. For healthcare policymakers, the relatively low importance of cost suggests that a patient-contributory financing model for digital rehabilitation may be a feasible strategy. Additionally, the marked heterogeneity by gender and age underscores the need for tailored approaches: technology-driven features may enhance engagement for male and younger patients, while a human-centered approach with trusted multidisciplinary teams may be more effective for female and older patients. By ensuring that the development of RCR is closely aligned with the expressed priorities of CABG patients, we can enhance patient engagement, improve long-term health outcomes, and contribute to the broader advancement of patient-centered care models.

Conclusion

This study employed a DCE to quantify the preferences and WTP of 343 CABG patients for RCR services. Within the present DCE design, service provider had the greatest influence on patient preferences, with patients expressing a strong preference for multidisciplinary teams, followed by doctor-led services. Continuous automatic monitoring via wearable devices and video-based content were also valued. Personalized plans based on discharge assessment were preferred over standardized protocols, while service content did not reach statistical significance. Cost had a relatively low relative importance, ranking fifth among the six attributes. Subgroup analyses revealed that male patients showed significant preferences for a wider range of attributes, whereas female patients showed significant preferences primarily for service provider. Younger patients additionally favored personalized discharge assessment, while older patients additionally favored AI-assisted services and comprehensive programs. These findings suggest that patient-centered RCR programs should prioritize multidisciplinary team structures, integrate wearable monitoring technologies, and incorporate video-based educational materials. Tailored approaches may be considered to address the distinct preferences observed across gender and age groups.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

DS and JZ conceived and designed the study. DS designed the questionnaire and wrote the initial draft of the manuscript. HW, GL, SJ and XL were responsible for the implementation of the study and data collection. DS, SL, HZ and JZ reviewed and revised the manuscript. DS was responsible for the overall study management. All authors read and approved the final manuscript.

Funding

This work was supported by Nursing research Project of Shandong Provincial Hospital (HL2025-16) and Shandong Province Traditional Chinese Medicine Science & Technology Project (M20243601).

Data availability

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

This study was conducted in accordance with the Declaration of Helsinki and was approved by the Ethics Committee of Shandong Provincial Hospital, affiliated with Shandong First Medical University (SWYX: NO.2025-011). All participants voluntarily consented to take part in the study and signed informed consent forms, confirming their understanding of the study’s objectives. To ensure confidentiality, all data collected is anonymized and deidentified, protecting participants’ identities. The data will be securely stored and accessible only to authorized research personnel. No financial compensation is provided to participants. Additionally, no identifiable images or personal data of participants will be included in the publication or supplementary materials to safeguard their privacy.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.