Research on the Network Structure Characteristics of Doctors and the Influencing Mechanism on Recommendation Rates in Online Health Communities: A Multi-Dimensional Perspective Based on the “Good Doctor Online” Platform

Abstract

(1) Background: Online health communities (OHCs) serve as ecosystems connecting doctors, patients, and medical resources. Studying their deep network structure and impact mechanisms on medical service quality provides a comprehensive understanding of digital healthcare ecosystems and has guiding significance for platform service optimization. (2) Methods: Using the “Good Doctor Online” platform as the data source, we employed social network analysis methods to construct network models from the professional title and disease-type dimensions, and used multiple linear regression statistical analysis to identify the influencing factors of doctor recommendation rates. (3) Results: Our analysis found that depression doctors exhibit the highest network connectivity (average degree = 17.378), and chief physicians demonstrate significantly higher internal connectivity (average degree = 9.353) compared to resident physicians (average degree = 0.804). The doctor recommendation rate is significantly correlated with post-consultation evaluation (r = 0.602, p < 0.001) and shows a 45% variance explanation (R² = 0.450) in our regression model. (4) Conclusions: Different disease types in OHCs demonstrate distinct organizational patterns, with depression networks showing significantly denser connections than diabetes networks. Professional titles strongly influence network position, with chief physicians forming highly connected hubs while resident physicians remain peripheral. Recommendation rates emerge through multi-dimensional trust processes primarily driven by post-consultation evaluation quality.

1. Introduction

With the development of internet technology and the emergence of digital healthcare, healthcare service models are undergoing a profound transformation. Online Health Community (OHC) has become a critical platform for patients to acquire medical information and seek medical support [1]. In recent years, OHC has evolved beyond mere information dissemination channels, developing into a complex ecosystem that interconnects patients, physicians, and medical resources. However, despite OHCs playing an increasingly vital role in medical services, systematic research on their inner operational mechanisms and service quality evaluation remains insufficient. This research gap urgently necessitates an in-depth exploration of the structural characteristics and intrinsic logic of physicians’ social networks within online health communities from multiple dimensions.

Existing research predominantly focuses on surface-level characteristic analyses of OHCs, such as user interaction patterns [2,3] and information transmission pathways [4,5], but investigations into deep-level network structural characteristics and their influence on mechanisms on medical service quality remain relatively limited. Particularly, there is a lack of empirical research regarding the formation mechanism of the critical indicator of physician recommendation rates. This research gap not only impedes our comprehensive understanding of digital medical ecosystems but also constrains practical guidance for OHC service optimization. Recent research by Fineschi et al. (2022) highlights the importance of understanding differences between patient and physician perceptions in healthcare settings, which further underscores the need to examine recommendation mechanisms in online health communities [6].

In light of these considerations, this study aims to explore the network structural characteristics of OHCs and their impacts on medical services through the following research questions:

• What are the differences in the network’s structural characteristics across different disease types of physicians? How do these differences reflect the functional division of labor and organizational logic within medical specialties?
• What heterogeneous structural characteristics do physician groups of different professional titles (chief physicians/resident physicians) demonstrate within medical collaboration networks? What influences do these structural differences exert on medical service quality transmission?
• What are the key factors that influence physician recommendation rates in online health communities? Specifically, how does post-consultation evaluation function as a core element of social trust negotiation within recommendation mechanisms?
• How do network structural characteristics in digital medical ecosystems influence physician recommendation rates through social trust negotiation processes? What implications does this operational mechanism hold for platform management?

This research will employ Social Network Analysis (SNA) methods combined with statistical analysis, using physician data from the “Good Doctor Online” platform to systematically analyze the network structural characteristics of OHCs and their influence mechanisms on medical service recommendation rates. Through this investigation, we hope to reveal the intrinsic structural logic of physician networks within OHCs and explain the multi-dimensional influence mechanisms of recommendation rates. By quantitatively analyzing the complex interactive relationships in online medical services, we aim to provide practical references for medical service quality enhancement and patient-physician trust construction, and offer novel theoretical support and practical guidance for constructing digital medical ecosystems.

2. Related Studies

2.1. Definition and Classification of OHC

An Online Health Community (OHC) is a virtual community based on an internet platform for users to share health information and experiences. Johnston et al. (2013) defined OHC as a “platform allowing patients, caregivers, and medical professionals to communicate health information and experiences online” [7]. With the development of mobile internet, a study by MNSD Okour (2023) further expanded this definition, emphasizing that OHC is a “comprehensive health management platform integrating health information sharing, emotional support, and professional consultation” [8].

Currently, representative OHCs both domestically and internationally can be broadly categorized into two types: first, Patient-to-Doctor (P2D) communities and professional medical health websites, such as Medhelp, Patientslikeme, Good Doctor Online, Spring Rain Doctor, Sweet Home, Search Medical Consultation Network, and 39 Health Network; second, Patient-to-Patient (P2P) communities or health sections on social platforms, such as Sweet Home, Baidu Quit Smoking Bar, and Sina Health [9].

2.2. Quality Evaluation of OHC Information Services

Currently, the construction of OHC information service quality evaluation systems primarily focuses on several key directions: multi-dimensional evaluation, user experience, reliability and authority, cultural and design factors, and consequence and antecedent indicators [10]. First, in terms of multi-dimensional evaluation, Wang and Strong (2020) proposed a four-dimensional framework including information accuracy, timeliness, comprehensibility, and usability [11], while Jiang et al. (2020) combined technical levels, user perception, information quality, and service models to construct a more comprehensive evaluation system using fuzzy comprehensive evaluation method [12]. Sun et al. (2019) conducted a systematic literature review, summarizing 25 evaluation criteria and 165 specific indicators covering credibility, professionalism, and objectivity, providing comprehensive evidence for OHC information quality assessment [13].

In terms of user experience, Xu and Shang (2021) constructed a user experience evaluation index system based on five elements (strategic layer, scope layer, structural layer, framework layer, and presentation layer) using the fuzzy hierarchical analysis method, and proposed suggestions to enhance user satisfaction and continuous usage intention [14]. In measuring reliability and authority, Boyer (2007) proposed eight core indicators through the HON certification system, including content authority and privacy protection [15], and Venkatasubramanian (2021) constructed 15 reliability evaluation indicators (technical reliability, access flexibility, time management flexibility, and communication capability across time and distance, etc.) to ensure information credibility and authority [16]. Some studies focused on cultural and design factors. For example, Sun et al. (2021) used the LIDA tool to evaluate website layout, color scheme, and search tools, emphasizing the impact of website design on user experience and information acquisition efficiency [17]. Finally, Zhang and Kim (2022) constructed a consequence and antecedent indicator system through meta-analysis, deeply exploring the influence of individual differences and information sources on information quality, as well as cognitive evaluation and behavioral intention indicators, which systematically elucidating the influence mechanism of OHC information service quality [18].

A critical comparison of these evaluation approaches reveals several limitations. Multi-dimensional frameworks often lack contextual sensitivity to different medical specialties, while user experience models frequently overlook professional dimensions of healthcare quality. Reliability measures tend to focus on technical aspects rather than knowledge exchange quality, and cultural factors are often underrepresented in standardized evaluation systems. Our research addresses these gaps by integrating network structural analysis with recommendation metrics, providing a more comprehensive understanding of how professional networks influence service quality perception.

2.3. Application of Social Network Analysis (SNA) in Online Health Communities

Social network research in online health communities has become a focal point in academic fields. Through in-depth analyses of user interactions, support networks, and knowledge flow across multiple dimensions, researchers have revealed the complex social network structures and dynamic characteristics of digital social platforms.

In user interaction research, scholars have constructed social network models, discovering the fragility and high dispersion of user interactions. For instance, Zhai et al.’s study on the Baidu Quit Smoking Bar demonstrated that users lack sustained interactions, with the network presenting a highly fragmented state. Xu and Zhang (2016) analyzed the Douban Depression Group, and revealed the depth and breadth of user interactions by using clustering coefficients and reciprocity indicators to precisely characterize the structural features of social networks [19].

Support networks and knowledge flow represent the core functions of online health communities. Lu et al.’s (2021) research on online depression communities profoundly explored the transmission mechanisms of information support and emotional support, analyzing how user attributes influence support network formation [20]. Through centrality indicators and community detection, researchers not only identified key roles within networks but also systematically compared network characteristics across different platforms. Curran and Abidi’s (2006) research on emergency practitioners’ forums was particularly outstanding; they constructed information-seeking and sharing networks, evaluated knowledge dissemination efficiency, and accurately located influential key nodes [21]. Cross-platform comparative studies also provided important perspectives for understanding the diversity of digital social networks. Chuang and Yang’s (2013) comparative analysis of forums, blogs, and notes [22], and Zhang and Yang’s (2012) comparison of QuitNet forums and Facebook [23], both revealed differences in user interactions and information dissemination across platforms.

In role identification, researchers adopted multi-dimensional analytical methods. Through centrality indicators, core-periphery analysis, and other techniques, scholars successfully identified key user roles in online communities and their influences on network dynamics. Durant et al. (2010) divided users into five roles in melanoma discussion groups, conducting an in-depth analysis of interaction patterns across different roles within support networks [24]. Dias et al. (2012) focused on key roles in knowledge sharing within diabetes forums, systematically revealing these users’ critical contributions to network structure and knowledge dissemination [25].

This research not only enriched our understanding of online health communities but also provided important theoretical and methodological references for subsequent social network research.

2.4. Influencing Factors of OHC Information Services

In the age of digital healthcare, Online Health Communities are key platforms for health information exchange. Their service effectiveness and user experience are comprehensively influenced by multi-dimensional factors. Existing research predominantly systematically analyzed key factors affecting OHC service quality from technological, content, user, and social perspectives [26], presenting a complex multi-factor ecosystem.

Firstly, technological infrastructure and usability are fundamental supports for OHC services. Topaloglu et al. (2013) revealed through ANP methodology that functional factors (such as information personalized classification and search options) and usability factors (such as memorability and interactivity) jointly shape user experiences, with functional factors slightly predominating [27]. This finding resonates with Calvano et al.’s (2021) conclusions regarding digital health center website usability, both emphasizing the critical role of technological dimensions in user interaction [28]. Venkatasubramanian (2021) further focused on health information website reliability, proposing primary influencing factors like search engine rankings, certification marks, and editorial policies, providing specific pathways to enhance platform credibility [16].

Secondly, information quality and content depth constitute the core competitiveness of OHCs. Shim and Jo (2020) emphasized four dimensions of information quality: timeliness, comprehensibility, completeness, and relevance, indicating that high-quality information services can significantly improve user satisfaction, thereby promoting continuous usage and health management benefits [29]. Maloney-Krichmar and Preece’s (2005) ethnographic research went further, not only focusing on information itself but also highlighting the intrinsic connections between high-quality content, community norms, and member interactions, revealing the dialectical interaction between social support and technological factors [30].

User individual differences represent a critical perspective for understanding OHC service effects. Hong (2006) analyzed personal factors from the user access willingness perspective, such as network dependency and health information knowledge levels [31]; Czaja et al. (2013) focused on elderly populations, discovering that network usage experience and cognitive capabilities significantly influence usage effects. Interestingly, after controlling other variables, age itself does not constitute a decisive factor [32]. These studies collectively point to a crucial conclusion: OHC services should possess differentiated and inclusive designs capable of satisfying personalized needs across different user groups.

Social support and interaction mechanisms represent the unique advantages distinguishing OHCs from traditional health information channels. Maloney-Krichmar and Preece (2005) emphasized active interactions between members, experience sharing, emotional support, and stable community member composition. These social network elements not only enhance information dissemination efficiency but also provide users with emotional identity and psychological support [30].

Although existing OHC research has achieved significant progress in user interactions, network characteristics, and service quality evaluation, current studies primarily focus on descriptive analyses. They lack systematic exploration of deep-level correlation mechanisms between network structural characteristics and information service quality, with insufficient cross-dimensional and cross-hierarchical comprehensive analyses. Most research remains at surface-level descriptions of user interactions and support networks, failing to fully reveal the complex composite influences of multiple dimensions like disease and professional title types on network structures.

Simultaneously, research on physician recommendation rate influence mechanisms remains limited, often concentrating on platform-level user experience and information service quality influencing factors. Therefore, this study intends to employ Social Network Analysis (SNA) methods to systematically analyze OHC network characteristics from a network topology structural perspective [33]. Through a multi-dimensional indicator system, it will deeply parse the complex influence mechanisms of professional capabilities, social psychology, and institutional environments on physician recommendation rates.

This research not only introduces a network structural perspective methodologically but also attempts to reveal the multi-dimensional and complex process of medical service recommendation rate generation. By providing a more comprehensive and in-depth theoretical insight into understanding digital medical ecosystems, it holds significant academic and practical significance for optimizing online health community service quality.

3. Research Design

3.1. Data Sources

Research data were sourced from physicians’ information crawled from the “Good Doctor Online” platform (https://www.haodf.com/) using Python 3.8.8 with BeautifulSoup and Selenium [34]. During data cleaning, invalid samples with incomplete information were first eliminated (such as doctors with only IDs due to military unit delisting). Subsequently, physicians’ personal information tables and individual website data tables were merged, using physician ID as the unique identifier to integrate recommendation rates, online consultation volumes, professional titles (text fields such as “Chief Physician” and “Attending Physician”), affiliated hospitals and departments, ultimately obtaining physician information for 7409 individuals, including 2673 depression-related physicians, 1102 leukemia-related physicians, and 3634 diabetes-related physicians.

Depression, leukemia, and diabetes were specifically selected for analysis as they represent distinct categories of medical conditions (mental health, acute complex disease, and chronic metabolic disease) that typically involve different treatment approaches and specialist networks, allowing us to investigate diverse network structures across the medical specialty spectrum.

A missing data analysis was conducted using pandas.read_csv() to load the data, and df.info() to quickly detect field types and the proportion of non-null values. To ensure analytical reliability, we eliminated incomplete records rather than imputing missing values, as the complete dataset remained sufficiently large (n = 7409) to support robust statistical analysis without potential bias from imputation.

3.2. Research Methods

This study was conducted from two perspectives: overall network characteristic analysis and investigation of factors influencing physician recommendation rates. For network characteristic analysis, Social Network Analysis (SNA) was employed to examine the structural characteristics of hospital affiliation networks in online health communities and to analyze physician distribution patterns across different classification levels. For investigating physician recommendation rate influencing factors, correlation analysis and multiple linear regression statistical methods were utilized to assess the impact of various factors on recommendation rates and visualize the analysis results.

3.2.1. Network Analysis

This research constructed physician network models using social network analysis software Gephi 0.10.1 [35], where network nodes represented individual physicians, and network edges indicated physicians belonging to the same hospital. Utilizing Gephi’s “filter” function, physicians were categorized into four groups based on professional titles: “Chief Physician”, “Associate Chief Physician”, “Attending Physician”, and “Resident Physician”, constructing corresponding subnetworks. Key network indicators for each subnetwork were analyzed, including average degree, graph density, number of connected components, and modularity index, with comparative analyses conducted across different professional titles. Additionally, physicians were filtered by disease type (depression, leukemia, diabetes) to construct corresponding disease subnetworks, comparing their network indicators to reveal distribution characteristics and structural differences across different classifications. Through these methods, the aim was to comprehensively understand the network distribution of physicians across the professional title and disease-type dimensions and their inherent patterns.

3.2.2. Influencing Factors Investigation

Correlation analysis was performed using SPSS 26.0.0.0 software [36], screening out variables with non-significant correlations with recommendation rates based on Pearson correlation coefficients (r). Recommendation rates on the “Good Doctor Online” platform are numerical indicators of a physician’s popularity and reputation, calculated based on patient reviews, thank-you notes, and professional endorsements. For the screened independent variables, multiple linear regression analysis was further conducted to assess the independent impact of each factor on recommendation rates using regression coefficients (β). Simultaneously, Python was used to generate correlation heat maps and box plots for result visualization, intuitively displaying the relationships and distribution characteristics between variables. Through these methods, the objective was to systematically analyze the primary factors influencing physician recommendation rates and provide clear graphical support to enhance the credibility of research conclusions.

Correlation analysis was performed using SPSS software [36], screening out variables with non-significant correlations with recommendation rates based on Pearson correlation coefficients (r) at a significance level of p < 0.05.

4. Results

4.1. Social Network Analysis

The visualization of the doctor affiliation network in online health communities is presented in Figure 1. Each node in the network represents a doctor, and the edges between nodes indicate that these doctors belong to the same hospital. Doctor nodes for depression, leukemia, and diabetes are rendered in orange, green, and purple, respectively. Relevant network indicators are shown in

Table 1. Network Indicators of Doctors by Different Disease Types.

Network Indicators	Depression	Leukemia	Diabetes
Average Degree	17.378	7.929	7.234
Graph Density	0.007	0.007	0.002
Connected Components	720	371	1240
Modularity	0.94	0.936	0.982

and

Table 2. Network Indicators of Doctors by Different Professional Titles.

Network Indicators	Chief Doctor	Deputy Chief Doctor	Attending Doctor	Resident Doctor
Average Degree	9.353	6.647	3.444	0.804
Graph Density	0.003	0.003	0.002	0.003
Connected Components	961	946	682	209
Modularity	0.982	0.98	0.971	0.942

.

The average degree indicates the average number of edges connected to each node (doctor); the number of connected components represents the number of independent subnetworks in the network (typically corresponding to different hospitals); the modularity index measures the quality of network division into different modules (hospitals), with higher values indicating clearer divisions; graph density measures the ratio of actual existing edges to potentially existing edges in the network.

The results in Figure 2 and Table 1 demonstrate that under different disease-type classifications, the depression doctor network exhibits the highest average degree (17.378) and a relatively high graph density (0.007), indicating extremely dense connections among doctors within the same hospital in the depression network. This may reflect that medical teams related to depression collaborate more closely within hospitals, or that the hospitals involved have a higher number of depression specialist doctors with a more concentrated distribution. The graph density of all disease types remains low (0.002–0.007), showing that although internal connections are dense, the proportion of edges in the overall network is still minimal. The diabetes doctor network possesses the highest number of connected components (1240) and modularity index (0.982), indicating that diabetes doctors are distributed across most hospitals, with extremely tight internal connections and almost no cross-hospital connections. In comparison, the connected component numbers for depression (720) and leukemia (371) doctor networks are relatively fewer, implying that doctors in these disease types are concentrated in fewer hospitals, with equally dense internal connections.

In the visualization analysis, using Gephi to plot the network topology of high-recommendation-rate doctors (top 20%), we discovered that high-recommendation-rate depression doctors are primarily concentrated in a few core medical institutions (such as Beijing Anding Hospital and Shanghai Mental Health Center), forming dense cross-departmental subnetworks. In contrast, high-recommendation-rate diabetes doctors are scattered across multiple tertiary hospitals, relying on the integration of internal resources within each hospital.

The results in Figure 3 and Table 2 reveal distinctive network characteristics across professional title categories. Chief doctors exhibited the highest average degree (9.353), indicating approximately 9 connections per chief doctor within each hospital, suggesting dense internal professional networks. Conversely, resident doctors demonstrated the lowest average degree (0.804), implying minimal interconnectivity within hospitals. Connected component analysis revealed hospital distribution: chief and deputy chief doctors were distributed across approximately 950 hospitals, attending doctors across 682 hospitals, and resident doctors across 209 hospitals. All professional title categories demonstrated high modularity indices (>0.94), indicating clear network partitioning with highly dense internal hospital connections and minimal inter-hospital linkages. The overall graph density remained low (0.002–0.003), reflecting the minimal proportion of potential connections realized within the network. This predominantly resulted from the numerous independent hospitals, leading to inherently low network connectivity.

4.2. Exploration of Recommendation Rate Influencing Factors

Correlation analysis revealed significant correlations between doctor recommendation rates and multiple variables: post-consultation evaluation (r = 0.602, p < 0.001), final patient completion count (r = 0.562, p < 0.001), and number of thank-you letters (r = 0.554, p < 0.001) demonstrated moderate positive correlations. Article count showed no significant correlation with recommendation rate (r = 0.122), and thus was excluded from subsequent multilinear regression analysis to enhance analytical precision. Pearson correlation coefficients between variables are presented in

Table 3. Pearson Correlation Coefficients r Between Variables.

	Doctor Recommendation Rate	Doctor Online Work Quantity	Personal Webpage Visits	Article Count	Total Post-Consultation Patient Count	Post-Consultation Evaluation	Thank-You Letter Count	Gratitude Gift Count
Doctor Recommendation Rate	1	0.459	0.369	0.122	0.526	0.602	0.554	0.42
Doctor Online Work Quantity	0.459	1	0.263	0.274	0.717	0.743	0.706	0.724
Personal Webpage Visits	0.369	0.263	1	0.151	0.273	0.278	0.259	0.176
Article Count	0.122	0.274	0.151	1	0.163	0.148	0.137	0.148
Total Post-Consultation Patient Count	0.562	0.717	0.273	0.163	1	0.807	0.743	0.696
Post-Consultation Evaluation	0.602	0.743	0.278	0.148	0.807	1	0.981	0.787
Thank-You Letter Count	0.554	0.706	0.259	0.137	0.743	0.981	1	0.755
Gratitude Gift Count	0.420	0.724	0.176	0.148	0.696	0.787	0.755	1

. Standard errors between correlations are presented in

Table 4. Standard Errors Between Correlations.

	Standard Error
Doctor Recommendation Rate	0.4424
Doctor Online Work Quantity	1924.798
Personal Webpage Visits	232.508
Article Count	84.311
Total Post-Consultation Patient Count	741.467
Post-Consultation Evaluation	115.195
Thank-You Letter Count	52.065
Gratitude Gift Count	169.859

.

We used Python 3.8.8 to generate a correlation heatmap (Figure 4). The coordinate axes encompassed indicators including doctor recommendation rate, online work quantity, personal webpage visits, article count, total post-consultation patient count, post-consultation evaluation, thank-you letter count, and gratitude gift count. Numerical values within grid cells represent Pearson correlation coefficients (r) between the corresponding variables. Additionally, grid cell color intensity reflects correlation coefficient magnitude, with darker colors indicating more significant inter-variable correlations.

Multiple regression model goodness-of-fit test results demonstrated that the model could explain 45% of the dependent variable variance (R² = 0.450). F-test results (F = 864.142, p < 0.001) indicated statistically significant overall regression model performance. Standardized regression coefficient analysis revealed that post-consultation evaluation (diagnosis_evaluation) possessed the highest standardized regression coefficient (β = 1.419), signifying its most substantial influence on the dependent variable. Thank-you letter count (thanks_letter) also exhibited significant influence (β = −0.833), with a negative directional effect. Doctor online work quantity (doctor_online_work_num) (β = −0.012) and final patient count (patient_final_finish) (β = 0.109) demonstrated relatively minor impacts on recommendation rate.

Python was employed to generate a box plot illustrating the regression relationship between the primary recommendation rate influencing factor (post-consultation evaluation) and the recommendation rate (Figure 5). The horizontal axis represents the doctor recommendation rate, progressively increasing from 2.3 to 5, arranged sequentially; the vertical axis represents post-consultation evaluation. The upper and lower box edges correspond to the third (Q3) and first (Q1) quartiles, respectively. The horizontal line within the box represents the median, reflecting the central tendency of sample data. Whiskers above and below the box indicate maximum and minimum values, with dispersed points marked as outliers. Box positioning demonstrates the distribution of 50% of the data, collectively reflecting the centralized trend across different groups. The results explicitly indicate a significant positive correlation between doctor recommendation rate and post-consultation evaluation levels.

These findings provide comprehensive insights into the complex mechanisms underlying doctor recommendation rates in online health communities. By systematically analyzing the intricate relationships between various professional and interactive indicators, the study illuminates the multi-dimensional nature of recommendation rate generation, emphasizing the critical role of post-consultation evaluation in shaping patient trust and service perception.

5. Discussion

The network analysis of the professional title dimension reveals the hierarchical characteristics of the medical system. The chief physician network shows a high degree of internal connectivity, reflecting the core position of senior physicians in the dissemination of professional knowledge and the integration of resources. In contrast, the residency network appears relatively closed and fragmented, highlighting the gradient character of medical professional development. The conclusion of this study reveals the heterogeneity of physician groups with different professional titles in the medical collaboration network. From the perspective of network topological structure attributes, chief physicians, with their ability to integrate information resources and advantages in structural holes endowed by their senior titles, have significant advantages in core indicators such as Degree Centrality and Betweenness Centrality. The formation mechanism of this difference can be attributed to three dimensions: First, the occupational role division dimension, senior physicians through leading multidisciplinary diagnosis and treatment, difficult case consultation and other collaborative scenarios, to build a cross-department strong connection network, the subgroup of which shows a higher density and clustering coefficient; second, the dimension of knowledge dissemination. As the pivotal node of clinical experience and scientific research information, chief physicians effectively promote the transfer and diffusion of tacit knowledge through vertical guidance and horizontal coordination, thus strengthening the cohesion of the network. Zhu and Wang (2020) pointed out in their study that key opinion leaders (Kols) play a core role in the knowledge diffusion process in the medical network, and high school psychological physicians significantly improve the overall knowledge level and innovation ability of the network through frequent interaction and information sharing [37], which is consistent with the conclusion of this study. Third, in the dimension of career development trajectory, residents are limited by the hierarchical characteristics of the training system and the rules of the career life cycle, and their social networks show significant local aggregation, which is significantly correlated with the growth path and resource acquisition mechanism of junior physicians. Lee and Chen (2020) found through the analysis of social networks that, Junior physicians rely on tight local networks in the process of resource acquisition, which not only limits the breadth of their networks, but also affects the diversity and flexibility of their professional development [38], which is also confirmed by the conclusion of this study.

The analysis of the disease-type network deeply reveals the functional differences and internal organizational logic of the medical profession and presents the uniqueness and complexity of the disease field in the construction of a professional network. The depression network exhibits significantly different network characteristics from other disease types, and its unusually dense internal connections reflect multi-level professional collaboration mechanisms. From the point of view of the complexity of diagnosis and treatment, depression is a disease involving physiological, psychological, and social dimensions, and its medical network shows a high degree of interdisciplinary integration. This network structure may stem from several deep-seated reasons: First, the highly professional nature of depression diagnosis and treatment requires the medical team to have the comprehensive ability to cross fields. Unlike diabetes, which is a relatively single physical disease, depression requires the coordinated treatment of psychiatry, psychology, neurology, endocrinology, and other disciplines. The high density of the network means that there is frequent knowledge exchange and experience sharing among professionals. Second, the individualization and complexity of depression treatment have led to tighter professional networks. The causes, symptoms, and treatment options of each depression patient are highly personalized, which requires close internal communication and collaboration among the healthcare team. The high connectivity in the network actually reflects the high degree of collaboration and knowledge integration in the diagnosis and treatment process. Moreover, medical resources in the field of depression are relatively concentrated, and this concentration further strengthens the tight connection of the network. The study found that doctors with high referral rates for depression were mainly concentrated in a few core medical institutions, such as the Beijing Anding Hospital and Shanghai Mental Health Center, which formed a close inter-departmental specialty subnetwork.

In contrast, the diabetes network showed very different network characteristics. The distribution of specialties is more dispersed, reflecting the degree of standardization of diabetes care and the breadth of resources. There are multiple organizational logics behind this network structure: (1) Diabetes, as a common chronic disease, has a relatively standardized diagnosis and treatment and requires less professional collaboration than depression. Professional dependence among individual physicians is relatively weak, resulting in relatively loose network connectivity (2) Diabetes medical resources are more widely distributed and not as highly centralized as depression. Doctors with high referral rates are scattered among multiple top three hospitals and rely on the integration of internal resources within each hospital rather than close collaboration across hospitals. (3) The diagnosis and treatment of diabetes involves endocrine, nutrition, cardiovascular, and other fields, but the professional boundaries between these fields are relatively clear, reducing the urgency of interdisciplinary collaboration.

On a larger scale, this network difference reflects the profound impact of the complexity of the disease itself on the organization of the medical profession. Different disease types present a unique professional network structure, which not only reflects the needs of diagnosis and treatment but also reflects the internal mechanism of medical resource allocation, professional knowledge dissemination, and innovation. This analysis not only reveals the impact of disease types on the construction of professional networks, but also suggests that our future medical service system should pay more attention to such differences, and it is necessary to establish differentiated medical resource allocation and professional collaboration models for different disease fields.

The study on the mechanism of recommendation rate revealed the multi-dimensional heterogeneity of the mechanism of medical service recommendation rate. From the perspective of the network structure of post-diagnosis evaluation, the recommendation rate presents a complex generative logic. First of all, the professional competence dimension, the actual effect of medical service is the core driving factor of the recommendation rate, which not only focuses on the cure rate, but also provides insight into the systematicness and comprehensiveness of medical treatment. Secondly, from the social-psychological dimension, the process of patient trust generation involves the emotional quality of doctor-patient interaction, including the communication style, empathy, and humanistic care of doctors. Such micro-interaction has a profound impact on the formation mechanism of recommendation rate. Furthermore, in the institutional context dimension, the medical service referral rate reflects the broader medical ecosystem, which involves the complex interaction of professional reputation, social capital, and information dissemination networks. Abbasi-Moghaddam, Mohammad Ali et al. (2019) pointed out that medical service evaluation has gone beyond the traditional technology-oriented and shifted to a more holistic and comprehensive evaluation model [39], which is consistent with the conclusion of this study. Wang and Chen (2024) further emphasize that the generation of recommendation rate is a dynamic process of social trust negotiation, which integrates multiple elements of professional competence, affective dimension, and institutional environment [40], and these results echo our research findings. This multi-dimensional analysis reveals the deep logic of modern medical service evaluation: it is not only a technical repair, but also a complex social and cultural generation process.

Collectively, these findings reveal the complex interplay between professional hierarchies, disease characteristics, and trust mechanisms in shaping online health communities. Our network analysis provides a more nuanced understanding of digital medical ecosystems than previous studies by simultaneously examining structural positions, collaborative patterns, and reputation mechanisms across multiple dimensions. This integrated approach offers both theoretical insights into healthcare network dynamics and practical guidance for platform optimization that addresses the specific needs of different professional groups and disease communities.

6. Conclusions

This research, through an in-depth analysis of doctor networks in Online Health Communities (OHCs), reveals the complexity and diversity of medical service ecosystems. By systematically examining the structural characteristics of doctor networks from the professional title and disease-type dimensions, and exploring the multi-dimensional influence mechanisms of recommendation rates, the study provides a novel theoretical perspective for understanding the inherent operational logic of digital medical platforms. The study highlights the heterogeneity of medical professional networks and the distinct roles of different professional titles and disease domains in collaboration, knowledge dissemination, and resource integration.

However, the study acknowledges several limitations: First, the data sourced from a single platform may have certain sample representativeness limitations; second, network analysis and recommendation rate factor measurements might be influenced by some unobservable variables [41]. Future research can further expand the sample size, introduce more dimensional variables, and deepen the understanding of OHC network dynamics.

Despite these limitations, this network-based analytical approach offers significant theoretical and practical value for understanding digital medical ecosystems. Our findings provide concrete recommendations for optimizing online health platforms through targeted enhancement of cross-specialty collaboration networks for complex conditions like depression, while leveraging the concentrated expertise in chief physician networks. The strong correlation between post-consultation evaluation and recommendation rates (r = 0.602) indicates that platforms should prioritize quality patient feedback mechanisms. Future research should explore how technological innovations can further strengthen these professional networks to improve healthcare outcomes while maintaining the trust-based recommendation mechanisms identified in this study.