Factors Influencing Healthcare-Seeking Preferences Among Patients in County Medical Consortium in China: A Discrete Choice Experiment

Abstract

Purpose

Under China’s county medical consortium system, this study explores how level of care integration and other key attributes of medical institutions influence patients’ healthcare choices and examines differences in preferences among population subgroups.

Patients and Methods

This study constructed a virtual medical institution choice set using a discrete choice experiment (DCE) with eight institutional attributes. A total of 676 respondents were recruited via an online convenience sample on the Credamo platform using a self-developed questionnaire. The conditional logit model and fixed effects model were adopted to analyze 10,816 valid observations.

Results

Medical quality emerges as the primary factor influencing patients’ healthcare choices, followed by the level of care integration. Heterogeneity analysis reveals that rural residents and individuals with lower educational attainment prioritize practical factors such as convenience and medical costs. Individuals with higher education levels and female patients demonstrate greater concern for medical care quality.

Conclusion

All findings reflect stated preferences in hypothetical scenarios rather than observed healthcare-seeking behavior. Within this stated-preference framework, higher levels of integrated care, streamlined referral procedures and higher levels of information sharing were consistently associated with stronger patient choice. Importantly, we make no causal claim that these preferences directly predict real-world care-seeking or system-level outcomes. To move from stated preference to observed behavior, future research combining offline representative sampling with real utilization data is needed to validate and extend these patterns.

Introduction

The fragmentation of healthcare systems represents a persistent and pervasive challenge for health systems worldwide.¹ Both high-income and low- and middle-income countries have struggled with excessive specialization, disconnect between vertical planning and primary care, and insufficient health information exchange. These structural deficiencies not only generate inefficient resource allocation and fragmented patient journeys but also increase referral burdens and out-of-pocket costs. Consequently, they constrain overall health system performance. As economies develop and health literacy improves, patient expectations are shifting from simpl“ “accessing medical services” to “pursuing high-quality, continuous, and humanized care experience”. This shift mirrors a global transition from disease-centered to patient‑centered healthcare models,² providing a crucial direction for health reforms aimed at reducing fragmentation and building integrated systems.

Countries worldwide have therefore actively pursued integrated care models suited to their contexts.³ Kaiser Permanente in the United States is a classic vertically integrated delivery system. It integrates medical services with insurance mechanisms and adopts a system-wide unified electronic health record system to enable continuous and coordinated care. This model delivers effective cost containment and efficiency improvement;⁴ the United Kingdom’s National Health Service has enabled continuous care for common conditions at the primary level through primary care networks; and China has promoted downward referral of high-quality resources through county-level integrated healthcare consortia and urban medical groups. Despite different forms, these models share core goals: enhancing continuity and accessibility via resource integration, process optimization, and information connectivity.

Yet, existing research has predominantly focused on institutional design or organizational performance,^5,6 with surprisingly little attention to how integration shapes patients’ perceptions, experiences, and choice behaviors. Patients’ healthcare-seeking decisions are no longer static evaluations of a single facility’s “product attributes”. Instead, they represent dynamic journey experiences spanning multiple institutions, touchpoints, and time periods.⁷ Prior evaluations of county-level consortia have largely adopted policy or managerial perspectives, focusing on resource allocation efficiency while neglecting patient-level behavioral responses. As the ultimate service users, patients’ experiences and choices not only reflect the true effects of integration but also determine whether the system achieves its “patient-centered” goal. Therefore, examining how attributes such as level of care integration, information sharing, and referral convenience influence patients’ choices is essential for understanding the behavioral mechanisms of integration reform and for improving policy implementation.

Discrete choice experiments (DCEs) have been widely used to elicit patients’ healthcare-seeking preferences.⁸ Existing DCEs consistently identify medical quality, out-of-pocket costs, waiting time, and travel distance as core attributes, with medical quality being the most influential factor and costs and distance exerting negative utility.^9–12 However, most existing studies have focused on technical or node-specific attributes within a single healthcare facility and have seldom incorporated system-level integration attributes such as levels of care integration, information sharing, or referral convenience. Few Chinese county medical consortium DCE studies have adopted a patient-journey theoretical lens to quantify the relative weight of integrated care against traditional service factors.

To address these research shortcomings, this study asks two questions: (1) How do levels of care integration and other key institutional attributes influence patients’ healthcare-seeking preferences? (2) What differences in preferences exist across population subgroups? To answer these questions, we designed a discrete choice experiment with eight institutional attributes: levels of care integration, medical quality, distance to medical facilities, medical expenses, referral convenience, levels of information sharing, waiting time, and appointment services. A total of 676 respondents were recruited via the Credamo online platform, yielding 10,816 valid observations analyzed using conditional logit and fixed-effects models.

Drawing on the integrated framework of Service-Dominant Logic and the patient journey perspective, we operationalize the multi-layer theoretical logic into eight measurable experimental attributes. The upstream resource integration layer is captured by two core dimensions: level of care integration and cross-institutional information sharing. The pathway orchestration layer is reflected in referral convenience and appointment service. Touchpoint accessibility factors include distance to medical facilities and waiting time, while medical expenses and medical quality constitute the core value-sacrifice trade-off dimensions. Each attribute directly maps to a specific stage of the patient healthcare-seeking journey, ensuring a consistent mapping from theoretical constructs to empirical measures.

Notably, this study measures respondents’ perceived level of care integration within hypothetical scenarios, rather than objective administrative integration indicators of real medical alliances. Prior research suggests that patients’ subjective perceptions of coordinated services may diverge from formal institutional integration standards;^13,14 our analysis therefore captures preference responses to perceived integration, rather than providing an assessment of objective system performance.

Research Framework and Research Hypotheses

Research Framework

In contemporary healthcare delivery systems, patient healthcare-seeking behavior is no longer a static evaluation of the “product attributes” of a single healthcare facility but rather a dynamic journey experience across institutions, touchpoints, and time.⁷ Service-Dominant Logic (S-D Logic) emphasizes that value is not unilaterally created and delivered by service providers but rather “co-created” through resource integration and interaction among multiple actors within the service ecosystem.¹⁵ In the healthcare field, healthcare institutions, information platforms, payers, regulators, and patients jointly participate in value co-creation. Operant resources such as interoperability, pathway orchestration capability, and the level of care integration become important drivers determining patient experience and choice.¹⁶ Compared with the traditional supply-side “product logic”,¹⁷ S-D Logic provides a theoretical perspective that can explain how system coordination and patient experience are linked through resource integration. This enables us to understand the “level of care integration” as collaborative capacity at the ecosystem level, to interpret the “level of information sharing” as availability and fluidity at the data level, and to further trace their impacts on access, processes, and perceived value.

The “patient journey” perspective operationalizes this value co-creation process as an orderly connection of a series of touchpoints and pathways.¹⁸ In a typical healthcare-seeking journey, patients need to complete multi-stage transitions from appointment and system entry, waiting and treatment, to referral and follow-up.¹⁹ Each touchpoint is both a source of experience and a source of costs. Travel distance to healthcare facility and waiting time constitute spatiotemporal accessibility at the access level; the level of perfection of appointment services reflects the efficiency of front-end scheduling and capacity matching; referral convenience reflects cross-institutional pathway orchestration and routing capability; level of information sharing determines whether previous examinations and medical records can flow smoothly within the network, thereby avoiding duplicate examinations and information asymmetry. On this basis, patients’ trade-offs between medical quality and medical expenses occur both at individual touchpoints and at the overall journey level. Patients’ preference for “high quality” may be weakened by “long waiting”,²⁰ and preferences for “low costs” may change due to “duplicate examinations and process blockages”.²¹ The journey perspective therefore provides a process-oriented framework for explaining the dynamic trade-off of “value-sacrifice”.

Within this integrated framework (as shown in Figure 1), the level of care integration is positioned at the upstream “resource integration tier” of the ecosystem. Higher integration signifies stronger network governance mechanisms and coordination mechanisms, enabling stable collaborative relationships and standardized interfaces to form among healthcare institutions of different levels and types. The level of information sharing, as the technical and data dimension manifestation of integration, characterizes the interoperability of electronic health records, imaging, and laboratory results among different institutions. Existing research has demonstrated that enhanced interoperability can reduce information silos, decrease the probability of duplicate examinations, and improve the timeliness of cross-institutional collaboration.²²

Figure 1.

Analytical Framework. This figure illustrates the conceptual framework of the study, which integrates Service-Dominant Logic and the patient journey perspective. The framework is structured across four layers: resource integration, pathway orchestration, touchpoint accessibility, and value trade-off. Care integration and information sharing are positioned as upstream operant resources that influence referral convenience and appointment services, which in turn shape patients’ perceptions of waiting time, distance to medical facilities, medical expenses and quality, ultimately affecting healthcare facility choice and rating.

At the operational level, integration and sharing “drive” improvements at the pathway orchestration tier. Smoother referral processes and more efficient appointment systems are direct manifestations of integration spillover effects on processes. This can achieve better load balancing and capacity management on both the supply and demand sides,²³ this will indirectly alleviate congestion and waiting in the downstream areas. In addition, a well-developed appointment service and high-level integration can, to a certain extent, alleviate the adverse effects caused by the distance to medical treatment by optimizing the supply layout and route selection, and reduce the burden of “running around and uncertainty” at the patient’s perceived level.

The waiting time at the “touchpoint accessible tier” and the distance to medical facilities constitute the first threshold for patients to enter the system. Waiting is a time cost and also a source of risk and anxiety. It not only affects the experience but may also have an indirect impact on clinical outcomes.²⁴ Distance directly reflects the constraints of travel cost and accessibility. Against the backdrop of the urban-rural gap and the imbalance of regional resources, its impact is even more pronounced.²⁵ Integration - process improvement - access mitigation constitutes a mechanism chain that is passed from the upstream to the downstream. Integrated networks and interoperability have improved referrals and scheduling, reduced congestion and repetitive steps, thereby shortening waiting times and enhancing subjective perception of the adverse effects of distance. This chain provides a clear path for understanding how system reform affects individual choices.

The “value trade-off tier” is the direct manifestation of patients forming preferences and making choices during their journey. According to the perceived value theory, a patient’s perceived value = perceived benefit/perceived sacrifice.²⁶ Medical quality is closely related to safety, professionalism and prognosis, and it is the core source of perceived value for patients. Medical expenses are a sacrifice of money, waiting is a sacrifice of time, and distance is a sacrifice of travel. The Service-Dominant Logic emphasizes that the generation of perceived value depends on “quality” and also on whether the value co-creation process is effective. When information can flow within the ecosystem and path arrangement can reduce unnecessary transitions and delays, patients will obtain similar or higher expected treatment benefits at a lower sacrifice, thereby increasing their preference for institutions. On the contrary, if insufficient integration leads to information congestion, a cumbersome referral process and inefficient appointment services, patients may have to spend more time and money to obtain the same quality, thereby reducing their willingness to choose. In this sense, integration and sharing have changed the “average level” as well as the “marginal trade-off”. In a high-quality environment, patients may show greater tolerance to long waiting times. When the degree of information sharing is high and the convenience of referral is high, the negative utilities of cost and distance may decrease.

The integration of Service-Dominant Logic and patient journey provides an explanatory framework for healthcare integration, which encompasses governance at the macro level and experience at the micro level. The concept of “actionable resources” in the Service-Dominant Logic defines integration and interoperability as investable, governable, and accumulable system capabilities, not merely institutional labels. The patient journey provides a path for these upstream capabilities to choose behaviors. Furthermore, viewing patients with chronic conditions and elderly populations as groups highly sensitive to “journey smoothness” helps policy find a measurable balance between equity and efficiency. For example, interoperability interfaces and green referral channels can be prioritized for chronic disease pathways, or appointment quotas can be preferentially secured for these groups during high-congestion periods.

Research Hypotheses

Based on the above theories and research findings, this study proposes the following research hypotheses.

First, according to Service-Dominant Logic theory, healthcare value is co-created by multiple actors including patients and healthcare institutions through resource integration and interaction. Among these, operant resources serve as important driving factors. In healthcare scenarios, they can be decomposed into the level of care integration and the level of information sharing. A high level of care integration means more coherent healthcare-seeking processes. A high level of information sharing can reduce duplicate examinations and information asymmetry. Both can reduce patient costs. Therefore, the following research hypothesis is proposed:

H1: The resource integration tier (level of care integration, level of information sharing) has a positive impact on patient choice of healthcare facilities.

Second, from the patient journey perspective, a complete patient journey includes multi-stage services with touchpoints and pathways such as appointment, waiting, and referral. These touchpoint experiences, along with the healthcare costs incurred during the journey and patient assessments of medical quality, all influence patient choice of healthcare facilities. Therefore, the following research hypotheses are proposed:

H2: The touchpoint accessibility tier (waiting time, distance to medical facility) has a negative impact on patient choice of healthcare facilities.

H3: The pathway orchestration layer (referral convenience, appointment service) has a positive impact on patient choice of healthcare facilities.

According to perceived value theory, patient perceived value during the healthcare-seeking process is jointly determined by their assessed perceived benefits (with medical quality at the core) and perceived sacrifice (time, cost, and distance costs). Patients engage in dynamic trade-offs between the value gains brought by medical quality and the cost sacrifices incurred, ultimately forming the overall perceived value in a healthcare-seeking journey. Therefore, the following research hypothesis is proposed:

H4: Within the value-sacrifice tier, medical quality (core perceived benefit) positively predicts patients’ facility preference, whereas medical expenses (monetary perceived sacrifice) exert negative effects on healthcare facility selection.

Finally, demographic characteristics (gender, age, education level, chronic condition status) may create differences in patient choice preferences for healthcare facilities by influencing patient needs, information acquisition capabilities, and sensitivity to healthcare-seeking costs. Previous discrete choice experiments have shown that preferences vary significantly across these subgroups.^27–29 Patients with chronic conditions may be more concerned about the convenience of travel distance to healthcare facility. Patients with higher education levels place greater emphasis on medical quality. Patients of different genders may also differ in their demands for healthcare process efficiency. Therefore, the following research hypothesis is proposed:

H5: Patient preferences for healthcare facility choice exhibit certain differences based on demographic characteristics (gender, age, education level, chronic condition status).

Methods

Survey Experiment

The survey experiment is an innovative method that combines experimental research with social surveys.³⁰ It possesses the dual characteristics of both surveys and experiments and has been widely welcomed by scholars in the social sciences in recent years. According to different research questions, it can be divided into two categories. Early survey experiments were primarily conducted in the form of randomized questioning to improve survey precision and reduce social desirability bias. The mainstream design adopted in recent years is the vignette experiment. This method conducts causal inference by creating randomized virtual scenarios.³¹ It can effectively eliminate interference from other irrelevant factors and has advantages in reducing endogeneity problems and estimating causal effects.³²

The vignette experiment can be further divided into single-factor and multi-factor designs according to the number of experimental variables. Given that this study involves multiple experimental variables, a multi-factor vignette experiment method was adopted to construct virtual healthcare facility scenarios containing eight dimensions including the level of care integration and medical quality for hypothesis testing.

Participants

We utilized the professional online data collection platform, Credamo, to design the questionnaire and recruit participants from Dongyang People’s Hospital. The inclusion criteria required that participants: 1) had previously received hospital treatment services; 2) were aged 18 years or older; and 3) explicitly understood and consented to the study ethics and informed consent before completing the questionnaire. After data collection, 24 participants were excluded due to incomplete responses, incorrect answers, or failing an attention check (eg, showing strong consistency in their questionnaire responses). A final sample of 676 valid participants was obtained. Among the final sample, 381 participants were male (56.36%) and 295 were female (43.64%). In terms of age distribution, there were 504 participants aged 18–45 (74.56%), 168 participants aged 46–65 (24.85%). Additional demographic details (eg, education level, residence, income, chronic disease status) are provided in Table 1.

Table 1.

Demographic Information of the Subjects (N=676)

Variable	Statistical Value
Gender (%)	—
Male	56.36
Female	43.64
Age (%)	—
18–45 years old	74.56
46–65 years old	25.44
65 and older	0.59
Education Level (%)	—
Junior high school and below	1.92
High School/Vocation School	8.73
College degree and above	89.35
Place of Residence (%)	—
Urban	74.41
Rural	25.59
Annual Household Income (%)	—
Under 50,000 yuan	2.96
50,000–100,000 yuan	23.52
100,000–200,000 yuan	46.89
Over 200,000 yuan	26.63
Chronic disease status (%)	—
Yes	41.12
No	58.88
Frequency of medical visits (%)	—
1–2 times	53.25
3–5 times	37.72
6 times and above	9.02
Medical Insurance Coverage (%)	—
Urban and Rural Residents’ Medical Insurance	38.76
Urban Employee Medical Insurance	59.91
Other	1.33
Distance from home (%)	—
1–2km	12.87
2–5km	43.05
5–10km	27.96
Over 10km	16.12

According to the sample analysis of 676 respondents (Table 1), the main demographic characteristics distribution is as follows: males (56.36%) slightly outnumbered females (43.64%). Age distribution was dominated by the young and middle-aged group of 18–45 years (74.56%), with 46–65 years accounting for 24.85% and over 65 years accounting for 0.59%. The low proportion of elderly samples is not only constrained by the digital divide for elderly people in online surveys, but more importantly, in the Chinese context, healthcare-seeking choices of elderly people often follow the arrangements of their children at home. The healthcare-seeking choice preferences of middle-aged people often also represent the actual healthcare-seeking choices of elderly family members.^33,34 In terms of education level, those with college degree or above accounted for the vast majority (89.35%), reflecting the overall high cognitive ability of the sample, which facilitates understanding of the experimental scenarios. Residence distribution showed that urban residents accounted for 74.41% and rural residents for 25.59%. Those with annual household income above 100,000 yuan accounted for 73.52%, indicating that most families have a certain economic foundation. Chronic disease status distribution was relatively balanced, with 41.12% having chronic conditions. Annual healthcare utilization frequency was mainly 1–2 times (53.25%) and 3–5 times (37.72%), consistent with the general characteristics of healthcare-seeking behavior in the general population. Medical insurance types were mainly Urban Employees’ Basic Medical Insurance (59.91%) and Urban and Rural Residents’ Basic Medical Insurance (38.76%). The differences in the coverage mechanisms of these two types may influence healthcare-seeking choices. In terms of travel distance to healthcare facility, 83.88% were within 10 kilometers, indicating that most respondents preferred seeking healthcare nearby.

Ethical Considerations

This study was approved by the Medical Ethics Committee of Dongyang People’s Hospital prior to its conduct (Ethical Approval No.: 2025-YX-297) and was conducted in accordance with the Declaration of Helsinki. All participants provided informed consent by reviewing and agreeing to a consent statement on the first page of the online questionnaire before participation.

Setting Healthcare Facility Characteristic Indicators

The eight attributes were finalized via a multi-stage validation process. First, we systematically reviewed healthcare-seeking preferences and integrated care studies to extract frequently measured service dimensions. Second, semi-structured expert discussions were held with two health policy researchers and four clinical administrators from county medical consortia to identify system-level integration features absent from existing literature. Third, a paper-based pretest among 32 outpatient respondents was conducted to refine attribute wording and eliminate ambiguous expressions. All dimensions were retained after pretesting, confirming both face and content validity for patient population.

This study selected eight characteristic indicators of healthcare facilities: (see Table 2 for details): level of care integration, medical quality, distance to medical facilities, medical expenses, referral convenience, level of information sharing, waiting time, and appointment service.

Table 2.

Healthcare Facility Attributes

Attributes	Description	References
Level of care Integration	The degree to which healthcare services are coordinated across different providers and levels of care within the county medical consortium	[35]
Medical Quality	The perceived level of medical skill, treatment outcomes, and patient safety	[9]
Distance to medical facility	The one-way travel distance from the patient’s home to the healthcare facility
Medical expenses	The direct medical expenses paid by the patient out of pocket for a single visit	[10]
Referral convenience	The ease and speed of being referred from one facility to another when needed	[11,12]
Level of information sharing	The extent to which patients’ medical records and examination results are shared electronically among different facilities	[36]
Waiting time	The time from making an appointment or arriving at the facility until seeing a doctor	[9]
Appointment service	The availability and usability of systems for scheduling a medical visit	[12]

Specifically, level of care integration includes three values: highly integrated, partially integrated, and minimally integrated. Medical quality includes three values: higher, average, and lower. Travel distance to healthcare facility includes two values: closer and farther. Healthcare costs include three values: lower, moderate, and higher. Referral convenience includes three values: convenient, average, and inconvenient. Level of information sharing includes three values: fully shared, partially shared, and minimally shared. Waiting time includes three values: short, medium, and long. Appointment service includes two values: relatively well-developed and relatively lacking.

On this basis, this experiment adopted the D-optimal design from fractional factorial design to generate choice sets. The D-optimal design was conducted through the %ChoicEff macro command in SAS software. After 100 iterations, a choice design containing 16 choice sets was finally generated, with each choice set containing two alternatives. The relative D-efficiency value was 76.02. Finally, these 16 choice sets were divided into two sets of questionnaires, with each questionnaire containing eight groups. Each group presented two healthcare facilities with different characteristics. Respondents needed to select one healthcare facility they would be more willing to visit from the two healthcare facilities displayed in each choice set. Additionally, respondents needed to rate the two healthcare facilities in each group according to their healthcare-seeking preferences. Figure 2 presents an example of using the vignette experiment method to study healthcare facility choice preferences.

Figure 2.

Example of Experimental Design Choice Set. This figure shows a hypothetical choice set presented to respondents. Each choice set consists of two healthcare facilities (A and B) with different levels of eight attributes: level of care integration, medical quality, distance to medical facilities, medical expenses, referral convenience, level of information sharing, waiting time, and appointment service. Respondents were asked to select their preferred facility and rate each option on a 1–7 scale, where 1 indicates “very unwilling to choose” and 7 indicates “very willing to choose”.

Questionnaire Pretesting

Before launching the main online survey, we conducted a pretest of the questionnaire with outpatients recruited from the outpatient clinic of Dongyang People’s Hospital. A total of 32 participants completed the paper-based questionnaire. The pretest was also designed as a form of cognitive interviewing to examine content and face validity. Participants were asked to elaborate their understanding of each attribute level; confusing terminology was supplemented with plain-language definitions embedded in the formal questionnaire. Attention screening items were also embedded to identify careless responders, further strengthening data validity.

Specifically, the pretest revealed that the term “Level of Medical Integration” was not immediately understood by some respondents. To address this, we added plain-language explanations for each level directly in the questionnaire (eg, “highly integrated” was explained as “different healthcare services work closely together and share resources”). No other major comprehension issues were reported. The attention check items successfully identified a small number of inattentive respondents, confirming their utility for data quality control. No attribute levels were removed or added based on the pretest.

Survey Sample Selection

Regarding the calculation of the optimal sample size for discrete choice experiments, no completely unified standard has yet been formed in academia. However, there are some widely recognized rules of thumb and methods.

The rule of thumb by Pearmain et al,³⁷ indicates that for DCE design, a sample size exceeding 100 can provide a good data foundation for preference modeling Orme’s rule of thumb is a commonly used method to determine the minimum sample size for discrete choice experiments.³⁸ The formula for calculating the minimum sample size is: N > 500c/(t×a), where 500 is a fixed variable, c is the maximum number of levels an attribute can achieve, t represents the number of choice sets contained in each questionnaire, and a refers to the number of alternatives in each choice set. According to this formula, the minimum sample size required for each version in this study should be: N > 500×3/(8×2) = 93.75, meaning at least 94 respondents are needed. As the discrete choice questionnaire in this study is divided into two sets, A and B, the minimum sample size is 2×94 = 188 respondents.

In conclusion, since this study employs both the conditional logit model and the fixed effects model for data analysis, a certain sample size is required to ensure the reliability of the selection set analysis. Taking into account the precision requirements of the research, the number of variables, and possible invalid samples (such as incomplete questionnaire responses and failure to pass the attention test), the sample size of this study is 700. Respondents were recruited on the Credamo online platform and data was collected by filling out questionnaires online. The respondents were patients who had received hospital treatment services, covering different genders, ages, educational levels, places of residence and chronic diseases. A total of 700 questionnaires were distributed. After eliminating 24 questionnaires that did not meet the response requirements or contained incorrect or random answers, 676 valid questionnaires were retrieved.

Dependent Variables and Model Specification

Two dependent variables were included in this study. The first variable refers to respondents’ choices across medical institution selection sets, analyzed using a discrete choice model (DCM). Rooted in random utility theory, DCM is a well-established econometric approach for studying discrete decision-making in finite choice scenarios.³⁹ We applied the conditional logit model specified in Equation (1):

(1)

In the Equation 1, represents the probability that individual i chooses the jth healthcare facility out of a total of J facilities. This probability is influenced by both facility characteristics and individual characteristics , where β and α denote their respective regression coefficients. Conditioning on each choice set implicitly controls for time-invariant individual heterogeneity, thereby eliminating potential bias in the estimates of facility attributes.

We acknowledge that mixed logit and latent class models are widely used to capture unobserved random preference heterogeneity in DCE research. However, three key considerations justify our conditional logit (CL) specification. First, all choice sets adopt binary two-alternative designs, under which the IIA assumption holds theoretically, eliminating the primary motivation to adopt more complex mixed logit models. Second, the core research objective targets population-averaged attribute weights rather than unobserved latent preference subgroups; observable demographic stratification fully addresses heterogeneity research questions without latent class estimation. Third, CL maintains high interpretability and avoids overparameterization, while cluster-robust standard errors at the individual level fully correct repeated within-person choice correlations, resolving the key statistical concern addressed by mixed logit. For completeness, we additionally adopted individual fixed-effects linear regression for facility rating outcomes to cross-validate preference patterns.

The second dependent variable is the respondent’s rating for each healthcare facility, analyzed using a fixed-effects model. The model is specified in Equation 2:

(2)

Here, represents the rating of respondent i for healthcare facility j, which is determined by a set of facility attributes , where β denotes the corresponding regression coefficient. To control for individual factors, we included an individual-level fixed effect α_i in the model. Finally, the model includes an error term ε_ij representing other unobserved confounding factors. The fixed-effects model assumes that unobserved time-invariant heterogeneity across respondents is captured by the individual intercepts, and that the error term is uncorrelated with the included attributes after conditioning on α_i. To account for within-respondent correlations caused by repeated rating observations from the same participant, we adopted cluster-robust standard errors with the respondent ID as the cluster variable.

Independent Variables and Moderating Variables

The core independent variables in this study are the eight virtual healthcare facility characteristics set in the vignette experiment. These specifically include: level of care integration, medical quality, distance to medical facilities, medical expenses, referral convenience, level of information sharing, waiting time, and appointment service. The specific definitions and value levels of each variable are as described previously.

To deeply explore group heterogeneity in patient healthcare facility choice preferences, nine variables were used as moderating variables for subgroup regression in subsequent analysis: gender, age, education level, urban-rural status, annual household income, chronic disease status, healthcare utilization frequency, medical insurance type, and travel distance to healthcare facility.

Results

According to Table 3, for the two dependent variables, respondents’ average rating for all hypothetical healthcare facilities was 4.4 points (standard deviation = 1.65). In the binary choice between healthcare facilities A and B, the selection proportions for both A and B were 50% each. Among the independent variables, the eight variables describing healthcare facility characteristics (level of care integration, medical quality, distance to medical facility, out-of-pocket costs, ease of referral, level of information sharing, waiting time, and appointment service) demonstrated a relatively balanced distribution across all levels. This distribution usually conforms to the principle of random distribution necessary for experimental design. This reduces the estimation error due to distribution error in subsequent analyses. It provides an appropriate and balanced sample for empirical analysis.

Table 3.

Descriptive Statistics of Virtual Healthcare Institution Characteristics in Scenario Experiments (N=10,816)

Variable	Statistical Value
Healthcare Institution Rating (Points)	4.4 (1.65)
Selection Situation (%)
A	50.0
B	50.0
Level of Medical Integration (%)
Highly Integrated	34.26
Partially Integrated	34.49
Low Integration	31.25
Healthcare Quality (%)
Moderately High	34.49
Average	34.26
Lower	31.25
Distance to Medical Care (%)
Closer	50.0
Far	50.0
Medical Expenses (%)
Lower	34.04
Moderate	28.24
Higher	37.72
Referral Convenience (%)
Convenient	31.25
Average	34.49
Inconvenient	34.26
Information Sharing Level (%)
Full sharing	31.03
Partial sharing	37.72
Less sharing	31.25
Waiting Time (%)
Short	31.47
Medium	37.5
Long	31.03
Appointment Services (%)
Fairly well-developed	50.0
Relatively Lacking	50.0

Factors Influencing Residents’ Healthcare Facility Choice

To measure residents’ choice preferences for healthcare facilities, this study employed a conditional logit model and a fixed effects model to analyze the two dependent variables: “binary choice between healthcare facility” and “healthcare facility rating”, respectively. Meanwhile, to enhance the explanatory power of the models, the analysis incorporated multiple levels of eight healthcare facility characteristics as independent variables. Since all independent variables were categorical variables, dummy coding was applied to all levels of the independent variables to avoid multicollinearity interference. Furthermore, for the multiple levels of each characteristic, the level with the objectively poorest performance was selected as the reference group and assigned a value of 0. All other levels were compared against this reference group to more intuitively illustrate the direction and magnitude of the impact of superior levels on residents’ healthcare-seeking choices.

As shown in Table 4, all eight healthcare facility characteristics in both models exerted significant positive effects on patients’ healthcare-seeking choices, and the results were highly consistent. Specifically, healthcare facilities with a higher level of care integration, higher medical quality, shorter distance to medical facility, lower out-of-pocket costs, greater ease of referral, complete level of information sharing, shorter waiting time, and comprehensive appointment service demonstrated stronger attractiveness to patients seeking care.

Table 4.

Factors Influencing Patients’ Choice of Healthcare Provider

	Healthcare Facility Rating		P	95% CI	Healthcare Facility Choice (Binary)		P	95% CI
	Coefficient	Standard Error			Coefficient	Standard Error
Level of Healthcare Integration (Low Integration=0)	—	—			—	—
Highly Integrated	0.829	0.045	<0.001	(0.74, 0.92)	1.280	0.066	<0.001	(1.15, 1.41)
Partially integrated	0.333	0.033	<0.001	(0.27, 0.40)	0.430	0.055	<0.001	(0.32, 0.54)
Healthcare Quality (Lower=0)	—	—			—	—
Higher	1.618	0.052	<0.001	(1.52, 1.72)	2.395	0.089	<0.001	(2.22, 2.57)
Average	0.721	0.041	<0.001	(0.64, 0.80)	1.265	0.073	<0.001	(1.12, 1.41)
Distance to Medical Care (Far = 0)	—	—			—	—
Closer	0.320	0.030	<0.001	(0.26, 0.38)	0.107	0.041	<0.05	(0.03, 0.19)
Medical expenses (higher = 0)	—	—			—	—
Lower	0.396	0.038	<0.001	(0.32, 0.47)	0.623	0.046	<0.001	(0.53, 0.71)
Moderate	0.260	0.038	<0.001	(0.19, 0.33)	0.365	0.065	<0.001	(0.24, 0.49)
Referral Convenience (Inconvenient=0)	—	—			—	—
Convenient	0.376	0.033	<0.001	(0.31, 0.44)	0.335	0.054	<0.001	(0.23, 0.44)
Average	0.224	0.036	<0.001	(0.15, 0.29)	0.111	0.053	<0.05	(0.01, 0.21)
Level of Information Sharing (Less Shared=0)	—	—			—	—
Full sharing	0.164	0.046	<0.001	(0.07, 0.25)	0.513	0.079	<0.001	(0.36, 0.67)
Partial sharing	0.0178	0.036	0.624	(−0.05, 0.09)	0.424	0.066	<0.001	(0.29, 0.55)
Waiting time (Long=0)	—	—			—	—
Short	0.105	0.034	<0.05	(0.04, 0.17)	0.555	0.069	<0.001	(0.42, 0.69)
Medium	0.184	0.029	<0.001	(0.13, 0.24)	0.603	0.056	<0.001	(0.49, 0.71)
Appointment Services (Relatively Lacking=0)	—	—			—	—
Relatively well-developed	0.296	0.033	<0.001	(0.23, 0.36)	0.141	0.043	<0.05	(0.06, 0.22)
Intercept	2.337	0.058	<0.001	(2.22, 2.45)	—	—
R2or χ2	0.2782	—			0.3323	—
N	10816	—			10,816	—
ll	−18469.1				2503.03
aic	36,966.2	—			5034.1	—
bic	37068.2	—			5136.1	—

Notes: For the fixed-effects regression model, the modified Wald test was applied to detect heteroskedasticity, which was mitigated via cluster-robust standard errors at the respondent level. All Cook’s distance statistics were substantially less than 1, confirming the absence of influential outlier observations.

In terms of effect size, medical quality exhibited the most significant impact in the fixed effects model. Compared with healthcare facilities with lower medical quality, facilities with higher medical quality showed a rating increase of approximately 1.62 points. This indicates that medical quality has a substantial impact on patient choice. This conclusion is consistent with findings from existing literature,^40,41 further enhancing the robustness of this study’s conclusions. Additionally, level of care integration also demonstrated substantial influence, which is highly consistent with research findings from other scholars.⁴² High integration compared with low integration resulted in a rating increase of 0.83 points. Moreover, the effect sizes of other characteristics were relatively smaller but all reached statistical significance levels.

To further assess the relative impact of the eight healthcare facility characteristics on patients’ healthcare facility choice, this study adopted marginal contribution analysis. By progressively incorporating each characteristic variable, the marginal contribution of each variable to the goodness of fit of the models (R² for the fixed effects model and pseudo-R² for the conditional logit model) was calculated. The calculation results are presented in Figures 3 and 4.

Figure 3.

Marginal contribution of each feature to the fixed-effects model R². This figure displays the increase in R² contributed by each healthcare facility attribute when added sequentially to the fixed‑effects model for healthcare facility ratings. The bars represent the marginal explanatory power of each attribute. Medical quality shows the largest marginal contribution, followed by level of care integration.

Figure 4.

Marginal contribution of each characteristic to the discrete choice model χ². This figure displays the increase in chi‑square (χ²) contributed by each healthcare facility attribute when added sequentially to the conditional logit model for binary choice. The bars represent the marginal explanatory power of each attribute. Consistent with Figure 3, medical quality shows the largest marginal contribution, followed by level of care integration.

Figures 3 and 4 present the marginal contributions of each healthcare facility characteristic to the explanatory power of the fixed effects model and the conditional logit model, respectively. The results show that medical quality occupied an absolute dominant position in enhancing the explanatory power of both models (marginal R² reached 0.1972 in the fixed effects model, and marginal χ² was approximately 2252 in the conditional logit model). This finding aligns with reality and the essential attributes of healthcare services. The primary purpose of patients seeking healthcare is to obtain effective treatment. Medical quality is directly related to patients’ life safety and treatment outcomes.⁴³ Therefore, it is the most critical factor influencing healthcare-seeking choices. Notably, the marginal contribution of level of care integration was also relatively significant (0.1041 in the fixed effects model and 491.85 in the conditional logit model), substantially higher than other characteristics such as distance to medical facility, level of information sharing, and appointment service. This indicates that medical quality and level of care integration together constitute the two core factors influencing residents’ healthcare-seeking choices, while the marginal enhancement of other characteristics to the models’ explanatory power is relatively limited.

In this study, although medical quality consistently remains the primary consideration in patients’ healthcare-seeking choices, level of care integration, as the second most influential factor, also suggests that integrated care models represented by integrated healthcare consortiums have played a positive role in guiding reasonable patient distribution and optimizing healthcare-seeking experiences. This also provides empirical evidence for further advancing the development of integrated healthcare consortiums and deepening the tiered healthcare delivery system.

Heterogeneity Analysis

To explore the differential impacts of population characteristics on healthcare facility choices, this study analyzed group differences across nine dimensions (gender, age, educational attainment, urban-rural residence, annual household income, chronic disease status, healthcare utilization frequency, medical insurance type, and travel distance to healthcare facilities) using the fixed effects model (consistent with the core analytical logic of the study).

From the perspective of gender differences (Figure 5A), the overall influence of healthcare facility characteristics on patients of different genders was similar, but significant variations emerged in medical quality, level of care integration, and level of information sharing. Female patients showed higher sensitivity to medical quality, with high-quality services exerting stronger attractiveness to them than to males; in contrast, male patients paid more attention to the level of care integration and information sharing, and were more inclined to choose facilities with higher integration and information connectivity. This difference may be attributed to variations in gender roles (eg, women often undertaking more family health management responsibilities) and healthcare-seeking cognition (eg, men prioritizing service systematization and process efficiency).

Figure 5.

Heterogeneity analysis by gender, age, education, and urban–rural residence. Heterogeneity analysis by (A) gender, (B) age group, (C) education level, and (D) urban-rural residence. Points represent coefficient estimates; horizontal lines indicate 95% confidence intervals. The vertical dashed line at zero indicates no effect. Coefficients greater than zero indicate a stronger preference for the attribute level relative to the reference category.

The eight attributes had relatively balanced effects on the 18–45 and 46–65 age groups (Figure 5B). However, participants aged 65 years and older accounted for only 0.59% of the sample, and the estimates for this subgroup lacked statistical power. Therefore, no reliable conclusions can be drawn about preference patterns among older adults in this study. Future research with larger samples of older adults is needed to examine whether their preferences differ from those of younger age groups.

There are also differences between urban and rural residents (Figure 5C): urban residents are more concerned about the overall quality and medical service level (tending to choose larger and broader economic capabilities), while rural residents are relatively less sensitive (geographical location restrictions, access to medical resources). Distinct preference gaps also exist across educational subgroups (Figure 5D). Participants with higher educational attainment attach greater importance to integrated service mechanisms and care coordination; in contrast, respondents with high school education or below prioritize out-of-pocket costs and geographically accessible services, reflecting greater price sensitivity and financial constraints.

In terms of economic and health characteristics (annual household income, chronic diseases, frequency of use of health services), annual household income shows a gradient effect (Figure 6A): As incomes increase, the attention of patients to the medical quality and the level of integration of care gradually increases, as high-income groups have higher solvency and more choices, shifting from basic accessibility to medical quality. Patients with chronic diseases (Figure 6B) and high-frequency healthcare users (1–5 visits per year, consistent with the utilization pattern of the general population) are more sensitive to the level of care integration (Figure 6C). They have long-term and continuous healthcare needs and frequent referrals. Integrated healthcare may help ensure the continuity of care, reduce repetitive examinations, and improve the healthcare experience. In contrast, patients without chronic diseases and those seeking medical treatment infrequently prioritize immediacy and convenience (for instance, short medical distances and comprehensive appointment services), as they have short-term and less frequent healthcare needs.

Figure 6.

Heterogeneity analysis by annual income, chronic disease status, frequency of medical visits, and medical insurance type. Heterogeneity analysis by (A) annual income, (B) chronic disease status, (C) frequency of medical visits, and (D) medical insurance type. Points represent coefficient estimates; horizontal lines indicate 95% confidence intervals. The vertical dashed line at zero indicates no effect. Coefficients greater than zero indicate a stronger preference for the attribute level relative to the reference category.

From the perspective of differences in medical insurance types (Figure 6D) and travel distances (Figure 7), patients under the basic medical insurance for urban and rural residents with a lower reimbursement ratio are more sensitive to medical integration, convenience in seeking medical treatment, and economic factors (reducing the burden of out-of-pocket expenses), while patients under the basic medical insurance for urban employees with a more comprehensive coverage are less sensitive to these factors. It is worth noting that the travel distance to medical institutions is not a decisive obstacle: patients are still willing to bear the additional time and transportation costs to choose institutions with higher medical quality and comprehensive levels, which reflects their strong demand for high-quality medical resources.

Figure 7.

Heterogeneity analysis by distance to medical facilities. Heterogeneity analysis by distance to medical facilities. Points represent coefficient estimates; horizontal lines indicate 95% confidence intervals. The vertical dashed line at zero indicates no effect. Coefficients greater than zero indicate a stronger preference for the attribute level relative to the reference category.

Discussion

This study explores the association between healthcare integration levels and patients’ stated healthcare-seeking preferences by constructing a multi-level analytical framework of “resource integration-pathway orchestration-touchpoint accessibility-value trade-off”. The findings indicate that, in this discrete choice experiment, healthcare integration was the second most important attribute influencing patients’ stated choices, after medical quality. This result extends the application of Service-Dominant Logic to the healthcare field and offers a theoretical perspective for understanding how integration might be perceived by patients. Unlike traditional supply-side evaluations of integration, this study adopts a patient-centered, stated-preference approach. It provides insights into how attributes such as care integration, information sharing, and referral convenience may shape patients’ perceived value. Nevertheless, because the study measures stated preferences in hypothetical scenarios rather than actual healthcare-seeking behaviour, the findings reflect patients’ declared valuations and trade-offs, not necessarily observed real-world choices.

The results suggest a hierarchical pattern in how integrated care attributes relate to patient‑perceived value. At the resource integration level, a higher stated level of care integration was associated with stronger patient preference, possibly reflecting a desire for continuous and coordinated care. At the pathway orchestration level, better referral convenience and information sharing were also preferred, which may reduce perceived process burdens. At the touchpoint level, shorter waiting time and closer distance were associated with higher preference. Together, these findings provide empirical support for applying service ecosystem theory in healthcare, but they remain at the level of stated preference and should not be directly equated with actual utilization or clinical outcomes.

In terms of theoretical contributions, this study introduces Service-Dominant Logic and the perspective of the patient journey, expanding the theoretical boundaries of medical service research. The Service-Dominant Logic emphasizes that value co-creation occurs in a complex service ecosystem and requires the integration and interaction of resources among multiple participants.⁴⁴ This study applies this theoretical framework to the medical environment, suggesting how integrated healthcare may promote value co-creation through mechanisms such as building collaborative networks and enhancing interoperability. The patient journey perspective provides a new analytical tool for understanding the micro-mechanisms of integrated healthcare, which helps to determine the ways in which comprehensive services realize their value at different touchpoints and pathways. This theoretical integration enriches the application of Service-Dominant Logic in specific fields and also provides a new perspective for studying the value creation of complex service systems.

The research also found that patients’ evaluation of medical services appears to have shifted from a single focus on medical quality to an overall service experience. This transformation reflects that modern medical services are undergoing a paradigm shift from “treatment-oriented” to “experience-oriented”. In this process, integrated healthcare is an organizational form but also a service innovation that systematically enhances the patient experience. This study provides potential evidence consistent with this theoretical viewpoint.

The results of the heterogeneity analysis indicate the potential inclusive value of integrated care and suggest differences between groups. Groups of patients with chronic conditions show increased demand for and sensitivity to integrated services, according to the theoretical discussion of the special value of integrated healthcare for groups with complex health needs.^45,46 Previous research suggests that socio-economic factors, such as urban-rural residence and education level, may influence patients’ attitudes towards integrated services. This finding suggests the possibility that when promoting integrated healthcare, more consideration may need to be given to the heterogeneous characteristics of the service recipients. The results of this study may offer relevant insights for the development in China, but also to the suggestions for global integrated healthcare practice. From the primary care networks of the UK’s NHS⁴⁷ to Accountable Care Organizations (ACOs) in the United States,⁴⁸ Countries around the world are exploring comprehensive nursing models that suit their national conditions. Previous research suggest that, regardless of the form of integration adopted, measures such as strengthening the exchange of health information and optimizing the referral process may help enhance patients’ recognition of integrated services.

Limitations of the Study

This research had several limitations. First, the sample composition limits generalizability. The study population was predominantly young, highly educated, and urban, with insufficient representation of older adults. This underrepresentation partly reflects the digital divide among older adults, who are less likely to participate in web-based surveys. Consequently, the findings may not be generalizable to older, rural, or less-educated populations. Future research could address this limitation by conducting offline surveys targeting older adults to obtain a more representative sample.

Second, the study measured stated preferences in hypothetical scenarios, not actual healthcare‑seeking behaviour. While discrete choice experiments are a validated method for preference elicitation, responses reflect what patients say they would choose under controlled conditions, not necessarily what they do in real-life settings. Consequently, our results should be interpreted as preferences and perceived value rather than direct behavioural predictions. Future research using real-world data is needed to further validate these findings.

Future research can further explore the long-term effects and mechanisms of integrated healthcare by expanding the research background, refining the categories of patient populations, and providing more theoretical and practical basis for the continuous optimization of the medical service system.

Conclusion

This study explores patients’ stated preferences for county-level integrated medical consortia using a discrete choice experiment grounded in Service-Dominant Logic and the patient journey framework. Our empirical results consistently indicate that medical quality constitutes the primary driver of patients’ hypothetical healthcare facility choices, followed by the level of care integration as the second most influential attribute, indicating that system-level coordination and continuity of care are valued by patients alongside clinical excellence. Subgroup heterogeneity analysis reveals distinct preference patterns across population segments: rural residents and participants with lower educational attainment prioritize low-cost and geographically accessible medical services, whereas highly educated and female respondents place greater weight on high-quality clinical care.

While our findings confirm that respondents hold stronger preferences for medical consortia featuring higher care integration, seamless cross-institutional information sharing, and streamlined referral pathways, these preference results cannot be interpreted as direct evidence that integrated reform will objectively reshape real healthcare-seeking behaviour, optimize system resource allocation, or improve overall healthcare efficiency. All service recommendations below are derived from hypothetical patient preferences rather than from empirically demonstrated real-world intervention effects.

For county medical consortia managers, advancing integrated care networks, cross-hospital information exchange, and simplified referral workflows can align with respondents’ expressed preferences. Targeted service design strategies may be developed for vulnerable groups with distinct demand—including rural residents, low-education populations, and patients with chronic diseases—with customized accessibility-focused arrangements to accommodate their stronger sensitivity to travel distance and out-of-pocket costs.

Interpretations of subgroup preference disparities must account for clear limitations in our unbalanced sample composition. More broadly, three critical constraints contextualize conclusions. First, all data capture hypothetical stated preferences within simulated choice scenarios, which may diverge from patients’ actual medical-seeking choices in real-life contexts. Second, online recruitment via a commercial questionnaire platform introduces inherent selection bias, as groups with limited digital literacy (particularly older rural adults) are largely excluded from our sample. Third, the skewed demographic distribution restricts the external validity of our heterogeneity results across the full population of county-level medical consortium users. Future research combining real-world medical utilization records with offline stratified sampling targeting underrepresented groups is required to validate and extend the preference patterns reported in this work.

Funding Statement

Qilu Research Key Project on the Construction and Development of Medical Consortia in Zhejiang Province (2025ZHA-QL104).

Ethics Approval and Consent to Participate

This study was approved by the Medical Ethics Committee of Dongyang People’s Hospital prior to its conduct (Ethical Approval No.: 2025-YX-297), and conformed to the ethical guidelines of the Declaration of Helsinki.

Disclosure

The authors declare no competing interests in this work.