2.1 Data Sources

The data set analyzed in the present study was derived from the Chinese Livelihood Status Survey administered by the Development Research Center of the State Council in 2021. The survey encompasses a diverse cross-section of China's provinces and regions, spanning the eastern, central, western, and northeastern regions, including Anhui, Fujian, Gansu, Guangxi Zhuang Autonomous Region, Jilin, Jiangsu, Xinjiang Uygur Autonomous Region, Yunnan, and Zhejiang. This extensive geographical reach ensures a representative sample of the national demographic. The survey collected data from 5010 rural residents across 138 villages and 64 districts and counties, capturing a comprehensive array of individual-level variables such as demographics, healthcare behavior and attitudes, and social security information, complemented by district- and village-level data encompassing healthcare resources, infrastructure, and socioeconomic indicators.

2.2 Variable Explanations

The dependent variable was the perceived healthcare accessibility, a construct with multifaceted dimensions. While traditional metrics often rely on objective factors such as distance to healthcare facilities and associated costs, the subjective perceptions of individuals are equally pivotal. With the advent of digital technologies, the perception of accessibility can diverge from physical proximity. Thus, it is appropriate to measure the evolution of healthcare accessibility through individual assessments. The survey question was: “In the past year, have you experienced any improvement in your access to healthcare services?” Responses were dichotomized as “significant improvement” (coded as “1”) and “no significant improvement” (coded as “0”).

The independent variable, telemedicine, was analyzed through the survey question: “Has the village clinic established a telemedicine collaboration with a higher-level hospital?” Affirmative responses were coded as “1,” while negative responses were coded as “0.”

Control variables were categorized into individual and district or county levels. At the individual level, these variables comprised age, gender, education level, household income, self-assessed health status, medical insurance, and family doctor affiliation. Education was quantified based on years of schooling, with a standardized coding system in which 0 years was considered “illiterate/semi-illiterate,” 6 years was considered “elementary school,” 9 years was considered “middle school,” 12 years was considered “high school/technical school,” 16 years was considered “undergraduate,” and 19 years was considered “master's.” The self-assessed health status was divided into five categories, with “healthy” assigned a value of “1,” and the other categories designated a value of “0.” District- or county-level controls encompassed the per capita GDP, number of healthcare professionals per 1000 people, paved road access, distance to the county center, and presence of a compact medical community.

2.3 Regression Model

In this model, $Y_{i,v,p}$ was the binary dependent variable that indicated whether the perceived healthcare accessibility had improved for residents $i$ of village $v$ in province $p$. $P(Y_{i,v,p}=1|{\mathit{X}}_i)=\Phi \left({\mathit{X}}_i\right)$ represented the conditional probability of improvement in perceived healthcare accessibility. $Teleme{d}_i$ denoted whether the village clinic of the resident $i$ had established a telemedicine collaboration with a higher-level hospital. C_i was the vector of individual-level control variables. C_v referred to the county-level control variables. ${\eta }_p$ represented the province-level dummy variable accounting for unobserved regional heterogeneity, and ${\epsilon }_{i,v,p}$ denoted robust standard errors. A significantly positive coefficient for ${\beta }_1$ suggested that telemedicine was associated with an increased likelihood of perceived healthcare accessibility improvement.

To address potential endogeneity stemming from reciprocal causality, an instrumental variable (IV) approach was applied to bolster the internal validity of the estimates. To check the robustness of the findings, we used an alternative dependent variable: villagers' assessments of healthcare accessibility convenience. The study further delved into heterogeneity and mechanism analyses to elucidate the nuanced effects of telemedicine on perceived healthcare accessibility.

3.1 Descriptive Statistics

After the exclusion of cases with missing data, the study sample comprised 3311 rural residents, among which 1266 had access to telemedicine services and 2045 did not. An analysis of perceived healthcare accessibility and the assessment of healthcare convenience revealed that individuals with access to telemedicine had nearly two times greater average values of relevant variables and had a proportion of primary care consultations that was approximately 5% higher than those without access to telemedicine. These preliminary findings suggest a positive correlation between telemedicine access and perceived healthcare accessibility. Table 1 displays the descriptive statistics for the total cohort and the two subgroups.

3.2 Baseline Regression

Table 2 presents the impact of telemedicine on perceived healthcare accessibility. Model 1 included the telemedicine variable alongside provincial fixed effects, while Models 2 and 3 incrementally incorporated individual- and county-level control variables. The regression analyses yielded marginal effects of telemedicine on perceived healthcare accessibility at 0.086, 0.087, and 0.091, respectively, all of which were statistically significant at the 1% level. These results indicate that telemedicine significantly improved the perceived healthcare accessibility of rural residents.

3.3 Robustness Test

This study acknowledges the potential for endogeneity issues stemming from mutual causality, which could compromise the internal validity of the results. Specifically, the establishment of telemedicine relationships between village clinics and higher-level hospitals may influence villagers' perceived healthcare accessibility. Conversely, when villagers' perceived healthcare accessibility is poor, local governments are more likely to implement telemedicine projects to improve those perceptions. To address this endogeneity concern, we applied two IVs: county-level terrain relief [24] and the number of broadband users [25] in 2011. These IVs were selected based on their relevance and exogeneity, which are essential for ensuring robust research findings.

To rigorously assess the validity of our IVs, we applied a comprehensive statistical approach. Specifically, we conducted a weak instrument test and an over-identification test to evaluate both the relevance and exogeneity of our IVs. (1) Weak Instrument Test: Table 3 presents the results of the weak instrument test, which was crucial for ensuring that our IVs were sufficiently correlated with the endogenous explanatory variables. The regression coefficients of our IVs were statistically significant at the 1% level, indicating a robust relationship. Additionally, the Kleibergen-Paap F statistic was significantly greater than the 10% critical value (19.93). This result confirmed that our IVs were not weak and were strongly correlated with the endogenous explanatory variables. (2) Over-Identification Test: Given that we used two IVs, we further used the Hansen J test to assess the exogeneity assumption to verify that our IVs were not correlated with the error term in the regression model. After controlling for the distribution of medical resources—specifically, the number of health technicians per 1000 people in the county—the results of the over-identification test indicated that we could not reject the null hypothesis that all IVs were exogenous. This finding provided strong evidence that our IVs were valid and suitable for use in the analysis. (3) Empirical Results and Interpretation: Building on the qualitative analysis, the statistical test results further reinforced our confidence in the validity of the IVs. Table 3 presents the IV-probit model estimates, with columns (2) and (4) confirming that telemedicine significantly improved both perceived healthcare accessibility and the perception of healthcare convenience. These results held even after accounting for potential endogeneity, thereby reinforcing the robustness of our baseline regression findings. It is important to note that the IV-probit regression results were interpreted as the local average treatment effect [26]. This approach focuses on the causal impact of telemedicine on a specific subgroup of individuals who are most influenced by the instrument. As a result, the regression coefficients from the IV-probit model were larger in magnitude compared with those from the baseline model. This difference was expected and aligned with the theoretical underpinnings of the local average treatment effect, which captures the treatment effect for compliers rather than the entire population.

We demonstrated the validity of our IVs through a combination of qualitative reasoning and rigorous statistical testing. The results of the weak instrument test and the over-identification test provided compelling evidence that our IVs were both relevant and exogenous. The IV-probit model estimates confirmed that telemedicine significantly enhanced perceived healthcare accessibility and convenience, even after accounting for endogeneity. These findings underscore the robustness of our analysis and highlight the potential of telemedicine to improve healthcare outcomes in rural areas.

3.4 Heterogeneity Analysis

To explore the conditions under which telemedicine was more effective in enhancing perceived medical accessibility, the present study examined two dimensions: external objective factors and the influence of medical systems. (1) External Objective Factors: External objective factors comprised the city area, population density (defined as the permanent population of the city divided by the city area), and availability of high-speed networks in rural areas. These factors are critical in understanding how telemedicine can overcome geographical and infrastructural barriers to improve healthcare access. (2) Influence of Medical Systems: The influence of medical systems was assessed through the implementation of the family doctor system and compact medical communities, both of which are integral components of China's healthcare reform. The family doctor system aims to provide personalized healthcare services, while compact medical communities focus on integrating healthcare resources to improve service efficiency and accessibility. (3) Methodology: The heterogeneity in the effectiveness of telemedicine was assessed using subgroup regression analysis. This approach allowed us to assess how different conditions and systems impacted the perceived medical accessibility among various subgroups. The significance of differences in regression coefficients across subgroups was evaluated using the seemingly unrelated estimation-based test (SUEST). Specifically, the empirical p-values derived from the SUEST indicate whether the differences in coefficients are statistically significant. A significant empirical p-value suggests that telemedicine has a differential impact on perceived medical accessibility across groups. (4) Results (Table 4): The results revealed that telemedicine was particularly effective in larger administrative areas with lower population densities, especially in sparsely populated rural regions (columns 1–4). Additionally, the availability of high-speed networks significantly enhanced the effectiveness of telemedicine in improving perceived medical accessibility (columns 5 and 6). These findings highlight the importance of infrastructure support in maximizing the benefits of telemedicine. The analysis of the family doctor system and compact medical communities further underscored the role of healthcare reforms in enhancing the impact of telemedicine. For rural residents enrolled in family doctor programs, telemedicine significantly improved perceived medical accessibility (columns 7 and 8). Moreover, telemedicine was particularly effective within compact medical communities, where it significantly enhanced perceived medical accessibility compared with loose medical communities (columns 9 and 10). The SUEST results confirmed that telemedicine had a more pronounced impact in compact medical communities, suggesting that integrated healthcare systems amplified the benefits of telemedicine.

In summary, the present study demonstrated that the effectiveness of telemedicine in enhancing perceived medical accessibility was contingent on both external objective factors and the structure of medical systems. The findings highlight the importance of high-speed network infrastructure and integrated healthcare systems in maximizing the potential of telemedicine. These insights provide valuable guidance for policymakers aiming to optimize telemedicine implementation in rural areas.

3.5 Mechanism Analysis

Primary care stands as a cornerstone of the hierarchical medical system, with its reach at the grassroots level being indicative of the system's overall efficacy. The present study captured the essence of grassroots primary care through the inquiry: “When you feel unwell, where do you usually seek treatment first?” The analysis was designed to ascertain whether telemedicine not only bolstered the use of primary care but also enhanced the perception of medical accessibility. For the purposes of the present study, primary care was delineated as the initial recourse to township-level or lower medical facilities, which was coded as “1,” while all other options were coded as “0.” A probit model was used to scrutinize the nexus between telemedicine (independent variable) and the use of primary care (dependent variable).

To address potential endogeneity issues stemming from omitted variables, such as the distance from village to county hospitals, the present study used county terrain relief and the number of broadband users in 2011 as the IVs for telemedicine. These IVs were selected based on their relevance and exogeneity, ensuring that they were correlated with the telemedicine variable while being uncorrelated with the error term in the primary regression model.

The regression outcomes delineated in Tables 5 and 6 reveal that telemedicine had a significant positive impact on the propensity of rural residents to use primary care services. Specifically, the findings suggest that village health clinics could enhance their diagnostic and treatment capabilities by leveraging telemedicine to receive medical support from higher-tier hospitals. This improvement in service quality and accessibility, in turn, encouraged a greater uptake of primary care among rural patients.

The use of IVs in the present analysis helped to mitigate the bias that may arise from omitted variables, thereby providing more robust and reliable estimates of the causal effect of telemedicine on primary care utilization. The significant positive relationship identified in the regression outcomes underscored the potential of telemedicine to strengthen the capacity of rural health clinics and promote better healthcare access in underserved areas.