Introduction

The concentration of high-quality medical resources in large cities (Ruger and Horton, 2020; Wang and Nie, 2024) and the growing demand for high-quality healthcare (Kruk et al., 2018; Coube et al., 2023) has led to a rapid increase in intercity patient mobility globally (Wang and Nie, 2024), when patients travel across city boundaries in search of better healthcare services (Ding et al., 2023; Zhang et al., 2023; Zhong, Wu and Tao, 2024). However, research on intercity patient mobility remains limited, especially in low- and middle-income countries (LMICs).

Recent advancements in transportation, particularly high-speed rail (HSR), have transformed intercity travel (Lin, Yang and MacLachlan, 2018). As a modern rail system operating at significantly higher speeds than conventional rail, HSR offers a cost-effective and more accessible alternative to air travel, while also providing faster and more reliable service compared to conventional rail and road transportation (Duan, Sun and Zheng, 2020; Lin, 2017; Ren et al., 2019; Wang et al., 2016). Beyond its role as a transportation mode, HSR offers transformative potential to address broader societal challenges: it improves accessibility for less developed areas (Cavallaro, Buzzone and Nocera, 2023; Zhou, Yang and Li, 2018; Kim and Sultana, 2015), spurs economic growth (Jin et al., 2020; Liang et al., 2020), and facilitates regional connectivity (Lin, Yang and MacLachlan, 2018; Ren et al., 2019; Luo and Zhao, 2021).

Specifically, HSR has the potential to influence intercity patient mobility by enabling individuals—particularly those in remote or underserved areas—to overcome geographical barriers and access higher-quality healthcare services in other cities within reasonable travel times (Zhang et al., 2023; Chen et al., 2021). This is especially important for patients requiring frequent or specialized care, such as those with chronic illnesses or complex medical conditions, for whom local healthcare options may be insufficient (Liu et al., 2021). Moreover, HSR offers a more affordable alternative to air travel, helping to reduce transportation costs—an essential factor for many patients who face financial limitations (Bu, Tang, and Liu, 2022). However, despite its transformative potential, the impact of HSR on intercity patient mobility remains largely unexplored. Understanding how HSR influences intercity patient mobility is crucial to ensuring that medical infrastructure adapts to shifting patient flows and evolving regional healthcare demands in the HSR era.

To bridge this gap, this study takes an interdisciplinary perspective to examine the impacts of HSR on intercity patient mobility. Based on difference-in-differences (DID) models, linear regression models and yearly regression models, this study addresses three research questions: (1) What are the effects of HSR on intercity patient mobility? (2) Does the effect exhibit any difference among cities with different sizes? (3) Are there discernible lagging effects associated with the years following HSR implementation? The answers to these questions aim to enhance our understanding of the relationship between the emerging transportation technology like HSR and intercity patient mobility, thereby informing more effective policies and decisions regarding the advancement of intercity transportation network and the allocation of healthcare resources.

It is worth mentioning that this study leverages a groundbreaking dataset that enables such analysis. We utilized a large-scale, high-resolution, official dataset of intercity chronic kidney disease (CKD) hospitalization records extracted from the Hospital Quality Monitoring System (HQMS) of China—a representative developing country suffering from the displacement between patients and healthcare resources and a highly uneven distribution of medical facilities (Liu, 2013; Yang et al., 2020; Zhang et al., 2020b). Our dataset includes over 4 million daily records of CKD inpatients from tertiary hospitals across China, spanning 2014–2018. Critically, it records both the residence and hospitalization cities for each patient, enabling precise mapping of intercity patient flows. This unparalleled level of detail provides an ideal foundation for analyzing the impacts of HSR on patient mobility.

Literature review

Intercity patient mobility

The number of patients seeking healthcare resources across cities is increasing globally, with this trend being particularly pronounced in China. Government statistics indicate that a significant proportion of the Chinese population travels intercity for medical care. From 2014 to 2017, approximately 7.9% of patients sought hospital services outside their city of residence, with this data soaring to 24.2% in surgical departments (National Health Commission of the People’s Republic of China, 2019). This data underscores the extensive intercity patient mobility and highlights the critical demand for accessible healthcare across cities.

Accompanying this trend, numerous recent studies have begun to examine the intercity patient mobility behaviors. Research confirms the escalating trend of intercity patient mobility (Aggarwal et al., 2017) and reveals that such behaviors display distinct spatial patterns influenced by factors such as travel time and distance (Ding et al., 2023; Wang and Nie, 2024), advancements in intercity transportation (Zhang et al., 2023), socio-demographic status of the patients (Vega Naranjo et al., 2023), and pertinent policies like intercity medical care insurance and medical referral systems (Jia et al., 2020; Ao et al., 2022).

Some studies have leveraged big data to analyze intercity patient mobility. For instance, Wang, Wang, and Onega (2021) utilized Medicare enrollment and claims data to examine cancer patients’ spatial behavior in the U.S. Hekmat et al. (2022) gathered data from 602 public hospitals in Iran to explore factors influencing patient mobility rates. Wang and Nie (2024) retrieved more than 200 thousand patients’ information from Xianghubao, a Chinese mutual aid platform for severe diseases, to explore the variations in travel willingness, time, and distance of intercity patient mobility. However, due to the difficulty of obtaining official, large-scale, individual patient mobility data, most of these studies either used the unofficial data crawled from the Internet, or had to limit their research to state/provincial/regional levels. Notable exceptions include Ding et al. (2023) and Ao et al. (2022), who used official nationwide hospitalization data to explore intercity mobility patterns and assess the cost-effectiveness of such mobility among CKD patients.

This study utilized the same nationwide patient mobility dataset as Ao et al. (2022) and Ding et al. (2023) to examine the impact of HSR on intercity patient mobility. Extracted from the Hospital Quality Monitoring System (HQMS), the dataset includes over 4 million daily records of CKD inpatients from AAA-level tertiary hospitals across China, covering the period from 2014 to 2018. This large-scale, high-resolution dataset provides an ideal basis for our study.

Impacts of high-speed rail on mobility

High-speed rail (HSR) has profoundly transformed the landscapes of transportation and mobility. Research indicates that HSR enhances regional accessibility by improving connectivity between cities and the accessibility for less developed areas (Zhou, Yang and Li, 2018; Cavallaro, Buzzone and Nocera, 2023). Consequently, HSR has the potential to mitigate spatial inequalities (Kim and Sultana, 2015), spur economic growth (Jin, 2020; Liang et al., 2020), and facilitate mobility across regions and cities (Lin, Yang and MacLachlan, 2018; Ren et al., 2019; Luo and Zhao, 2021).

HSR reduces travel time dramatically, which in turn has reshaped travel behavior and preferences across regions (Wang et al., 2016; Ren et al., 2019). HSR offers a cost-effective alternative to air travel and provides faster, more reliable service compared to traditional rail and road transport (Lin, 2017; Duan, Sun and Zheng, 2020), thus positioning itself as an ideal solution for intercity mobility (Lin, Yang and MacLachlan, 2018). Moreover, the enhanced accessibility facilitated by HSR has promoted greater labor mobility and more extensive commuting patterns (Guirao, Campa and Casado-Sanz, 2018; Heuermann and Schmieder, 2019; Feng et al., 2023; Gao and Pan, 2023). HSR has also been shown to foster intercity knowledge spillovers and innovation diffusion by enabling faster and more frequent face-to-face interactions between firms and institutions (Xiao, Wang, and Du, 2022). Collectively, these findings demonstrate that HSR has significantly boosted intercity mobility and altered the dynamics of regional development and accessibility (Wang et al., 2019).

HSR’s influence also extends into the healthcare sector by enhancing accessibility to medical services and thus potentially improving health outcomes. By reducing travel times, HSR enables patients, particularly those in remote or underserved regions, to access specialized healthcare services located in major urban centers (Chen et al., 2021; Zhang et al., 2023). This accessibility is crucial for patients requiring regular and specialized treatments that might not be available locally, such as those with chronic diseases or complex medical conditions (Liu et al., 2021; Bu, Tang and Liu, 2022). Studies also suggest that HSR may help in balancing the healthcare resource distribution across regions, which can alleviate the pressure on overburdened urban healthcare facilities and improve overall healthcare service quality and utilization (Choi, Kim and Park, 2019; Song, Qiao and Luo, 2021; Wu, Rong and Zhao, 2022).

However, existing research has not thoroughly explored how HSR influence intercity patient mobility. By reducing travel time and improving travel comfort, HSR could drastically change the healthcare-seeking behaviors of patients, facilitating access to specialized medical care that is often concentrated in major urban centers. Therefore, addressing this research gap could provide insights into how transportation infrastructure improvements can lead to more equitable healthcare distribution and improve overall public health outcomes. To bridge this gap, this study uses statistical models to examine the impacts of the HSR implementation on the dynamics of intercity patient mobility across all city pairs in China, aiming to comprehensively quantify and understand the relationship between transportation technology development and intercity patient mobility behaviors.

CKD hospitalization in China

With over 697 million cases worldwide in 2019, CKD has been recognized as a major public health concern globally (Abbafati et al., 2020; Francis et al., 2024). Studies have shown that individuals with CKD have a higher hospitalization rate compared to the general population (Scahrauben et al., 2020), and their average length of stay is longer than that of patients with other disease categories (Schneider, O’Donnell and Dean, 2009). Moreover, the treatment of CKD consumes a disproportionately large share of healthcare resources (Daratha et al., 2012). These characteristics underscore the intensive medical demands associated with CKD.

China has the largest population with CKD in the world (Zhang et al., 2012), with a prevalence of 8.2% in China (Wang et al., 2023). As a developing country, China’s distribution of healthcare resources for such severe diseases is markedly uneven, with many cities experiencing shortages in clinical hospital capacity, specialist physicians, available kidney resources for transplantation, and advanced therapeutic equipment (Zhang et al., 2020b). These resources are predominantly concentrated in major urban centers such as Beijing, Shanghai, and Shenzhen, leaving smaller cities with a significant scarcity of such services (Liu, 2013; Yang et al., 2020). This disparity necessitates that CKD patients often seek healthcare in other cities, leading to intercity hospitalizations.

In addition, China’s healthcare system is structured hierarchically to mirror its administrative divisions—province, city, and township. Medical resources are disproportionately concentrated in higher-tier cities, particularly in provincial capitals. This structure leads to a high volume of cross-city healthcare-seeking, especially among patients in lower-tier cities with limited access to specialized care (Yan et al., 2022). Prominent tertiary hospitals in cities such as Beijing and Shanghai attract patients nationwide, often for complex or chronic conditions like CKD. On the other hand, there is a lack of well-established referral systems (Jia et al., 2022), resulting in the phenomena that most patients have to seek intercity hospitalization by themselves. In this context, high-speed rail (HSR) may facilitate access to specialized care by reducing travel time and increasing mobility. This study therefore employs the mobility of CKD patients as a case study to explore the impacts of HSR on intercity patient mobility.

Materials and methods

Data

Intercity patient mobility data

The panel data for analysis is sourced from multiple sources. The primary data on intercity mobility of CKD patients is extracted from China’s HQMS dataset. HQMS is a comprehensive, patient-level national database essential for hospital accreditation, managed by the National Health Commission (NHC). As of December 2018, the HQMS covered more than 75% of tertiary hospitals—defined as hospitals having at least 500 beds and holding the highest ranking within the Chinese medical hierarchy—across 31 provinces of the Mainland China. HQMS collected patient-level data from the nationally uniform front page of the hospitalization medical records, also known as the standardized electronic inpatient discharge summary records. As a mandatory and legally binding document, the electronic inpatient discharge summary record must be completed by the attending physician, who possesses the most accurate and comprehensive understanding of the patient’s condition. It captures detailed patient-level information, including demographics, hospitalization details, primary diagnoses, and basic treatment and cost data. Further information about HQMS can be found in previous studies (Zhang et al., 2016; Zhang et al., 2020b).

With the support of the World Health Organization (i.e., China-World Health Organization Biennial Collaborative Projects 2018-2019), we accessed the HQMS database with the initial aim of providing evidence for policy making regarding kidney disease. As such, only data for CKD patients were available and analyzed in this study, either as the primary diagnosis or as one of the first two secondary diagnoses, following the ICD-10 coding system (Wan et al., 2021). In China, where the CKD prevalence reaches 8.2%, an estimated 82 million adults are affected (Wang et al., 2023). However, the number of nephrologists is insufficient to meet healthcare demands, and there exists significant spatial disparity in nephrology service availability (Wu et al., 2023). Consequently, CKD patients constitute a substantial portion of intercity medical care flows. A preliminary analysis of this dataset reveals the geographic mismatch between healthcare resources and the needs of CKD patients, as well as the network structure of CKD intercity patient mobility (Ding et al., 2023).

The study period spans from 2014 to 2018, coinciding with the onset of the second wave of China’s HSR expansion and ending before the outbreak of the COVID-19 pandemic in 2019, to avoid the impacts of travel restrictions during the pandemic. Based on the residential city and hospitalization city of each patient in the dataset, we extract the yearly intercity patient mobility counts between city pairs during the study period. After deleting cities that do not have data, we obtain the mobility counts across 18,809 city pairs for year. Therefore, we have in total 18,809 (city pairs) × 5 (years) = 94,045 observations. Figure 1 plots the number of patient outflows and inflows of CKD intercity patient mobility in the year of 2014 and 2018.

Fig. 1: Spatial distribution of intercity CKD patient mobility in 2014 and 2018.
Fig. 1: Spatial distribution of intercity CKD patient mobility in 2014 and 2018.
Full size image

a Inflow counts in 2014; b inflow counts in 2018; c outflow counts in 2014; d outflow counts in 2018.

We also plot the distribution of hospitals and CKD nephrologists across cities, to visualize the spatial disparities of the healthcare supplies. The number of hospitals is obtained from China City Statistical Yearbook, and the number of CKD nephrologists is collected from two prominent online healthcare platforms based on our previous work (Wu et al., 2023). As shown in Fig. 2, both the density of hospitals (Fig. 2a) and the availability of CKD nephrologists (Fig. 2b) per unit area are heavily clustered in the eastern and southeastern regions, particularly in large metropolitan areas, leaving smaller cities with a significant scarcity of such services (Liu, 2013; Yang et al., 2020). The darkest shades in both maps represent the highest concentration levels, and these are largely confined to provincial capitals and coastal megacities. When adjusted for population, similar patterns exist (Fig. 2c, d), with large cities showing higher numbers of hospitals and CKD specialists per capita compared to vast areas in western and central China. The uneven distribution of hospitals and CKD nephrologists necessitates that patients often seek healthcare in other cities, leading to intercity hospitalizations.

Fig. 2: Spatial distribution of medical resources relevant to CKD in 2018.
Fig. 2: Spatial distribution of medical resources relevant to CKD in 2018.
Full size image

a Number of hospitals per 10,000 km² in each city in 2018; b number of nephrologists per 10,000 km² in each city in 2018; c number of hospitals per 10,000 population in each city in 2018; d number of nephrologists per 10,000 population in each city in 2018.

HSR data

The HSR data is sourced from the railway information dataset provided by CnOpenData (https://www.cnopendata.com/en/), including the opening dates of HSR stations and the HSR timetable. The data contains all train services with train codes starting with G, D, and C, which are operated at significantly higher speeds than conventional trains and offer more comfortable onboard environments. This definition of HSR follows the standard classification used by Chinese railway authorities and is consistent with previous research. We extract two metrics from the data: (1) the year when HSR become available between city pairs, to measure the effect of HSR implementation, and (2) the shortest travel time by HSR between city pairs, calculated from the HSR timetable. Notably, HSR travel time is only available for city pairs that have active HSR services, encompassing 4412 (city pairs) × 5 (years) = 22,060 observations.

China’s HSR network represents a landmark in modern transportation. The inauguration of the Beijing-Tianjin Intercity Railway in 2008 marked the beginning of a transformative era for HSR development in the country. Following this milestone, the network entered a rapid expansion phase, with a second wave of construction commencing in 2014. This period witnessed substantial growth in connectivity. Figure 3 illustrates the total number of cities connected to each city by HSR in 2014 and 2018, showcasing the remarkable increase in HSR accessibility over these years.

Fig. 3: Accessibility of cities via HSR in 2014 and 2018.
Fig. 3: Accessibility of cities via HSR in 2014 and 2018.
Full size image

a Number of cities accessible via HSR from each city in 2014; b Number of cities accessible via HSR from each city in 2018.

Other control variables

To control for other factors that might influence the intercity patient mobility, we collect multisource data that measures the socio-economic factors, hospitalization factors, and transportation factors of each city. The socio-economic factors include yearly average population of the origin (o_pop) and the destination cities (d_pop) and GDP of the origin (o_GDP) and the destination cities (d_GDP), both obtained from China City Statistical Yearbook.

Measures of the hospitalization factors are also from China City Statistical Yearbook, including the number of hospitals in the origin (o_hospital) and destination (d_hospital) cities, hospital bed capacity in the origin (o_bed) and destination (d_bed) cities, number of doctors in the origin (o_doctor) and destination (d_doctor) cities, enrollments in the medical insurance in the origin (o_UBMI) and the destination (d_UBMI) cities, enrollments in the basic pension insurance in the origin (o_UEBPI) and the destination (d_UEBPI) cities, and enrollments in unemployment insurance in the origin (o_UI) and the destination (d_UI) cities.

Transportation factors are measured by the yearly road transportation passenger volume in the origin (o_passenger) and the destination (d_passenger) cities, obtained from China City Statistical Yearbook. To quantify intercity connectivity, we constructed a national transportation network by integrating both air routes (sourced from OpenFlights) and railway lines (extracted from OpenStreetMap, OSM). For the railway component, we filtered the OSM data to retain only major intercity lines (e.g., railway=rail), excluding local transit systems such as subways (railway=subway) and light rail (railway=light_rail). Urban road networks were not considered, as our analysis focused on long-distance intercity mobility rather than intra-city movement. We then applied a shortest travel time algorithm to estimate the annual minimum travel time between all city pairs via air and rail. Using this travel-time-based intercity network, we computed betweenness centrality to evaluate each city’s accessibility and its role as a transportation hub within the national network. Betweenness centrality quantifies how often a city appears on the shortest paths between other city pairs and is defined as \({C}_{B}\left(v\right)={\sum }_{s\ne v\ne t}\frac{{{\rm{\sigma }}}_{{st}}\left(v\right)}{{{\rm{\sigma }}}_{{st}}}\), where \({{\rm{\sigma }}}_{{st}}\) is the total number of shortest paths between nodes s and t, \({{\rm{\sigma }}}_{{st}}\left(v\right)\) is the number of those paths that pass through node v. A higher betweenness centrality score indicates greater strategic importance within the national transportation system.

To avoid multicollinearity, we calculate the correlation matrix between each pair of the control variables (Supplementary Table 1), and remove variables that have significant and high correlation (correlation index > 0.80 and P value < 0.05) with other variables. The descriptive statistics of the variables being included in the regressions are summarized in Supplementary Table 2.

Difference-in-differences model

Model specification

We adopt DID model to examine the effects of HSR implementation on the counts of CKD patient mobility between city pairs. DID model has been widely used in both transportation and public health research to examine the impacts of policies or technology advancement on the mobility or health outcomes (Wing, Simon and Bello-Gomez, 2018; Xiao, Wang and Du, 2022).

In the DID model, two groups and two time periods are defined. The initial period (t = 0) represents the pre-treatment phase, prior to the first implement of HSR between a given city pair. The second period (t = 1) denotes the post-treatment phase, during which HSR services are active between the city pair. The treatment group (s = 1) consists of city pairs with active HSR services, while the control group (s = 0) includes city pairs without HSR connectivity. The treatment effect of HSR implementation is estimated by comparing the average change in outcomes between the treatment and control groups over time.

Specifically, the DID estimate (\(\gamma\)) is estimated as follows:

$$\gamma =\left[E\left({y|s}=1,t=1\right)-E\left({y|s}=1,t=0\right)\right]-\left[E\left(s=0,t=1\right)-E({y|s}=0,t=0)\right]$$
(1)

\(\gamma\) can be derived through an ordinary least squares (OLS) regression:

$$y=\alpha +{\beta }_{1}t+{\beta }_{2}s+\gamma \left(t* s\right)+{\beta }_{3}x+u$$
(2)

where y denotes the outcome variable of interest; t represents a binary time variable, which is assigned as 0 for the pre-treatment period and 1 for the post-treatment period; s is a binary group variable, where 1 represents the treatment group and 0 represents the control group; \(\gamma\) represents the effect of interest; t \(*\) s is the interaction between the time and group variables; and x encompasses a set of controlled variables.

Considering that HSR commenced operations in various cities in different years, we construct a time-varying DID model to ascertain the impact of HSR on intercity patient mobility, as illustrated below:

$${y}_{{ij},t}=\gamma {{HSR}}_{{ij},t}+{\beta }_{i}{x}_{{it}}+{\beta }_{j}{x}_{{jt}}+{c}_{{ij}}+{m}_{t}+{u}_{{ij},t},t=1,\ldots ,5$$
(3)

where \({y}_{{ij},t}\) is the count of CKD patient mobility from city i to city j in year t; \({{HSR}}_{{ij},t}\) is a binary variable, which equals to 1 if there is at least one HSR train operating from city i to city j in year t, and equals to 0 if there isn’t; \({x}_{{it}}\) and \({x}_{{jt}}\) represents time-varying control variables of the origin cities (city i) and destination cities (city j), respectively; \({c}_{{ij}}\) is the city-pair fixed-effects to control for the unobserved time-invariant city heterogeneity; \({m}_{t}\) is the time fixed-effect to control for temporal trends; \({u}_{{ij},t}\) is the error term.

Statistical tests are conducted for the DID models: the Hausman test determines the suitability of fixed effects (FE) versus random effects (RE) models; the Dickey-Fuller test assesses the presence of stochastic trends in the dependent variables; the Breusch-Godfrey/Wooldridge (BG) test checks for serial correlation, and the Breusch-Pagan (BP) test examines for heteroskedasticity.

Inverse probability of treatment weighting (IPTW)

One precondition of the DID model is parallel trend assumption, which requires similar temporal trends between the treatment and the control group if the treatment is absent, otherwise biases might exist in the model estimation. To mitigate this bias, we implement inverse probability of treatment weighting (IPTW) based on propensity scores (Austin and Stuart, 2015):

$${w}_{{ij},t}=\left\{\frac{1}{{{ps}}_{{ij},t}},{{Treatment}}_{{ij},t}=1;\frac{1}{1-{{ps}}_{{ij},t}},{{Treatment}}_{{ij},t}=0\right\}$$
(4)

where \({w}_{{ij},t}\) is the weight for the city pair i to j in year t; \({{Treatment}}_{{ij},t}\) is the binary variable that measures if active HSR service is available between city i and city j in year t; \({{ps}}_{{ij},t}\) is the propensity score of the city pair i to j in year t, which is obtained by fitting a logit model of \({{Treatment}}_{{ij},t}\) on a set of control variables:

$${{Treatment}}_{{ij},t}=f\left({z}_{{ij},t}\right)+{\epsilon }_{{ij},t}$$
(5)

where \(f\) is a logit model; \({z}_{{ij},t}\) contains all the eight control variables used in this study; and \({\epsilon }_{{ij},t}\) is the unobserved error.

Event study analysis

To further validate the parallel trends assumption and explore the dynamic effects of HSR implementation over time, we conduct an event study analysis. This approach allows us to estimate how the effect of HSR on intercity patient mobility evolves in the years before and after the introduction of high-speed rail services between city pairs. Specifically, we estimate the following event-study specification:

$${y}_{{ij},t}=\mathop{\sum }\limits_{t=-4,t\ne -1}^{4}{\gamma }_{t}{D}_{{ij},t}+{\beta }_{i}{x}_{{it}}+{\beta }_{j}{x}_{{jt}}+{c}_{{ij}}+{m}_{t}+{u}_{{ij},t}$$
(6)

where \({y}_{{ij},t}\) is the count of CKD patient mobility from city i to city j in year t; \({D}_{{ij},t}\) is a set of binary variables indicating the number of years t before or after HSR service starts between city i and city j; \({\gamma }_{t}\) captures the relative effect of HSR t years from implementation.

This specification enables us to assess whether patient mobility trends were parallel prior to HSR implementation (i.e., whether \({\gamma }_{t}\approx\)0 for t < 0), and to observe how the effect of HSR evolves post-implementation (t\(\ge\)0). We present the estimated coefficients \({\gamma }_{t}\) and their 95% confidence intervals in a dynamic event-time plot. A non-significant and flat trend in the pre-treatment years would support the parallel trends assumption.

OLS model

We then use OLS model to examine the effects of the travel speed improvement by HSR on the count of intercity CKD patient mobility:

$${y}_{{ij},t}=\alpha {V}_{{ij},t}+{\beta }_{i}{x}_{{it}}+{\beta }_{j}{x}_{{jt}}+{\varepsilon }_{{ij},t},t=1,\ldots ,5$$
(7)

where \({V}_{{ij},t}\) measures the travel speed improvement by HSR from city i to city j in year t, and \({\varepsilon }_{{ij},t}\) represents the error term. \({V}_{{ij},t}\) is measured as the ratio between HSR speed and the traditional rail speed:

$${V}_{{ij},t}=\frac{{V}_{{ij},t}^{{HSR}}}{{V}_{{ij},t}^{{rail}}}$$
(8)

where \({V}_{{ij},t}^{{HSR}}\) is the travel speed by HSR from city i to city j in year t, and \({V}_{{ij},t}^{{rail}}\) is the travel speed by traditional rail from city i to city j in year t.

Due to data limitation, we do not have access to travel speed data, so we use the ratio between the distance between two cities and the travel time to calculate the travel speed:

$${{\rm{V}}}_{{ij},t}^{{HSR}}=\frac{{d}_{{ij}}}{{t}_{{ij},t}^{{HSR}}}$$
(9)
$${{\rm{V}}}_{{ij},t}^{{rail}}=\frac{{d}_{{ij}}}{{t}_{{ij},t}^{{rail}}}$$
(10)

where \({d}_{{ij}}\) is the distance between city i and city j, \({t}_{{ij},t}^{{HSR}}\) and \({t}_{{ij},t}^{{rail}}\) are the travel time by HSR and traditional rail from city i to city j in year t, respectively.

Based on formula (8)–(10), we got:

$${V}_{{ij},t}=\frac{{V}_{{ij},t}^{{HSR}}}{{V}_{{ij},t}^{{rail}}}=\frac{{d}_{{ij}}/{t}_{{ij},t}^{{HSR}}}{{d}_{{ij}}/{t}_{{ij},t}^{{rail}}}=\frac{{t}_{{ij},t}^{{rail}}}{{t}_{{ij},t}^{{HSR}}}$$
(11)

In addition, we run another OLS model to examine the effects of the HSR service frequency on the count of intercity CKD patient mobility:

$${y}_{{ij},t}=\alpha {F}_{{ij},t}+{\beta }_{i}{x}_{{it}}+{\beta }_{j}{x}_{{jt}}+{\varepsilon }_{{ij},t},t=1,\ldots ,5$$
(12)

where \({F}_{{ij},t}\) denotes the daily average number of train services available between from city i to city j in year t.

Yearly regression model

HSR might have different effects on intercity patient mobility over time, depending on how long the HSR service has become available between city pairs. To measure this effect, we further apply a yearly panel data regression. The temporal evolution of HSR is approximately measured by the number of years after HSR is implemented between each city pair, as below:

$$\begin{array}{l}{y}_{{ij},t}={\gamma }_{1}{{HSR}1}_{{ij},t}+{\gamma }_{2}{{HSR}2}_{{ij},t}+{\gamma }_{3}{{HSR}3}_{{ij},t}\\\qquad\quad+\,{\gamma }_{4}{{HSR}4}_{{ij},t}+{\gamma }_{5}{{HSR}5}_{{ij},t}+{\beta }_{i}{x}_{{it}}+{\beta }_{j}{x}_{{jt}}\\\qquad\quad+\,{c}_{{ij}}+{m}_{t}+{u}_{{ij},t},t=1,\ldots ,5\end{array}$$
(13)

where \({{HSR}1}_{{ij},t}\) - \({{HSR}5}_{{ij},t}\) are binary indicators that represent the number of years that HSR services are available. Specifically, \({{HSR}1}_{{ij},t}\) equals to 1 if year t is less than a year after HSR is available from city i to city j, \({{HSR}(N)}_{{ij},t}\) (\(N=\mathrm{2,3,4}\)) equals to 1 if year t is between \(N-1\) to \(N\) years after HSR is available from city i to city j, and \({{HSR}5}_{{ij},t}\) equals to 1 if year t is more than 4 years after HSR is available from city i to city j.

Robustness check

Placebo test

We examine the robustness of the DID model by conducting a placebo test. We randomly select 6000 city pairs from the control group to create a ‘fake’ treatment group, and randomly assign the treatment label to each sample. We then run the DID model to see if the fake samples show “zero” effect. Besides running the 100% “fake” treatment test, we gradually increase the number of ‘real’ treatment samples in the 6000 city pairs, and run the DID using a 75%-fake treatment group, a 50%-fake treatment group, and a 25%-fake treatment group. For example, in the 75%-fake treatment group, we replace 1500 city pairs (25% of the ‘fake’ sample) by the “real” treatments from the treatment group, to see if the effect becomes significant as we reduce the ‘fake’ treatment in our testing samples.

2SLS regression

Considering the potential endogeneity in HSR implementation – HSR may be implemented between cities with more intercity mobility, we utilize an IV regression model with 2SLS estimation to address this endogeneity. We use the average distance of the origin and destination cities to the nearest HSR hub city as the IV. A city is classified as an HSR hub if it hosts one of the 18 Chinese National Railway Bureau headquarters (Supplementary Fig. 1).

A valid IV should satisfy both the relevance and the exclusion conditions. Regarding relevance condition, the distance to HSR hubs is strongly correlated with the probability of HSR implementation, because cities closer to HSR hubs (i.e., locations hosting National Railway Bureau headquarters) are significantly more likely to be prioritized in HSR development plans, due to their administrative, logistical, and network coordination functions. As for the exclusion restriction, the geographic distance to HSR hub cities is largely determined by the spatial configuration of the national railway system and does not directly influence patient mobility between non-hub cities, especially after controlling for observable factors such as GDP, healthcare infrastructure, and population. To further validate the IV used, we conducted the balance test (see Supplementary Table 3). We divide cities near and far from HSR hubs using the distance from each city to the nearest hub city (the threshold is the median value of all distances). Then we compare the pre-treatment (year 2013) characteristics between city pairs near and far from HSR hubs. According to the results, in 2013, the cities near and far from HSR hubs do not have significant difference in GDP, population, and healthcare availability (i.e. number of hospitals, beds, and doctors) at 0.001 significance level, indicating the plausibility of the instrument.

In the first stage, we use the IV to estimate the HSR implementation and HSR travel speed increase. Because the HSR implementation is measured using a binary variable, while the HSR travel time increase is measured by a continuous variable, we estimate a logistic regression of the HSR implementation from city i to city j in year t, and estimate an OLS regression of the HSR speed increase from city i to city j in year t.

$${{HSR}}_{{ij},t}=\gamma {{IV}}_{{ij},t}+{\beta }_{i}{x}_{{it}}+{\beta }_{j}{x}_{{jt}}+{\varepsilon }_{{ij},t},t=1,\ldots ,5$$
(14)
$${V}_{{ij},t}=\alpha {{IV}}_{{ij},t}+{\beta }_{i}{x}_{{it}}+{\beta }_{j}{x}_{{jt}}+{\varepsilon }_{{ij},t},t=1,\ldots ,5$$
(15)

where \({{IV}}_{{ij},t}\) is the average distance from city i and city j to the nearest HSR hub city, serving as the IV in the analysis. In the second stage, we conduct regression of the count of CKD patient mobility from city i to city j in year t on the predicted \({{HSR}}_{{ij},t}\) and \({V}_{{ij},t}\) using the DID and OLS models (formula (1)–(5) and formula (7)).

Subsample test

To further address concerns about non-random HSR deployment, we conduct a series of subsample analyses. Specifically, we stratify the sample based on pre-treatment GDP of the origin cities, namely, the first 50% of cities with higher GDP are sampled into the high-GDP group, while the rest 50% of cities are stratified into the low-GDP group. Then we re-estimate the DID model within each group. If the HSR effect is consistent across both, it reduces concern that effects are driven by baseline economic characteristics of cities.

Results

Effects of HSR on intercity patient mobility

To explore the effects of HSR on intercity patient mobility, we firstly run a time-varying DID model that investigates the effects of HSR implementation on intercity patient mobility, and two OLS models that examine the effects of HSR travel speed improvement measured by the ratio between the HSR travel speed and conventional rail travel speed, and HSR service frequency measured by the daily average number of trains operating from city i to city j. Table 1 presents the results from the DID model (model 1) and the two OLS model (model 2 and 3). Notably, the DID model is applied to 18,809 city pairs with valid intercity patient mobility data, resulting in a total of 94,045 observations (18,809 city pairs × 5 years). In contrast, the OLS models are restricted to 4412 city pairs with active HSR services, comprising 22,060 observations (4,412 city pairs × 5 years).

Table 1 DID regression and OLS regression results: effects of HSR implementation and travel speed improvement on intercity patient mobility.
Full size table

From Model (1), the coefficient of HSR is positive (\(\beta\) = 0.905, S.E. = 0.362) and significant at the 0.05 level, indicating that the availability of HSR services between cities is associated with a significant increase of 0.905 ± 0.362 CKD patient flows. This corresponds to an 8.24 ± 3.28% increase in intercity patient mobility, based on the average patient flow (10.983) across all city pairs. This finding demonstrates the potential of HSR implementation in facilitating intercity patient mobility. Because of the fact that intercity patient mobility enables patients to access higher-quality healthcare resources in other cities (Zhang et al., 2023; Zhong, Wu and Tao, 2024), HSR has the effect to enhance the quality and accessibility of healthcare services (Choi, Kim and Park, 2019). Therefore, investment in the advancement of intercity transportation could be a potential strategy to mitigate healthcare disparities and improve the overall well-being.

The event study analysis (Fig. 4) shows insignificant trends before HSR implementation, which supports the parallel trends assumption and thus implies the validity of DID results. In addition, the immediate post-treatment effects (0–2 years after HSR implementation) is positive and significant, suggesting that HSR increases patient mobility right after its implementation, with the largest effect observed two years after implementation. However, the effect drops sharply after 3 years of HSR implementation, indicating that the initial boost in patient mobility is temporary, or that other dynamics begin to reverse the effect over time.

Fig. 4: Event study analysis results showing the effect of HSR implementation on intercity patient mobility in all cities.
Fig. 4: Event study analysis results showing the effect of HSR implementation on intercity patient mobility in all cities.
Full size image

All regressions use the year prior to HSR implementation as the baseline period; point estimates represent changes in patient mobility relative to the baseline, with error bars indicating 95% confidence intervals.

As presented in Model (2), the coefficient for improvements in HSR travel speed is positive (\(\beta\) = 8.285, S.E. = 1.818) and highly significant at the 0.001 level. This suggest that doubling HSR travel speed compared to conventional rails results in an additional 8.285 ± 1.818 patient flows between the cities. This change represents a 28.51 ± 6.26% increase in intercity patient mobility, calculated based on the average patient flow (29.060) across city pairs with HSR in operation. The substantial increase in patient flow underscores how advancements in transportation speed directly influence healthcare-seeking behaviors, emphasizing the broader benefits of investing in faster and more efficient travel modes like HSR.

As shown in Model (3), the coefficient for HSR service frequency is also positive (\(\beta\) = 3.422, S.E. = 0.095) and significant at the 0.001 level, indicating that every one more train operating from city i to city j is associated with a 11.8% ± 0.33% increase in intercity patient mobility. Therefore, the number of HSR services available could also influence health-care seeking behaviors.

Distinctive effects of HSR in megacities, large cities, and small cities

City size is a critical factor that might result in the distinctive patterns of the intercity patient mobility (Coube et al., 2023; Hekmat et al., 2022). To examine whether the effects of HSR differ in cities with different sizes, we categorize our samples into megacities, large cities, and small cities based on the population in their urbanized areas, then run the DID models on each group (Table 2). The grouping is applied to the destination cities only since healthcare resources, service capacity, and hospital attractiveness are more determined by the characteristics of the destination cities, and such grouping enables the analysis on what corresponding policies should be considered to accommodate changes in patient volumes due to intercity patient mobility. Megacities are the most developed urban centers with populations exceeding 5 million (e.g., Beijing, Shanghai, and Guangzhou); large cities, with populations over 1 million, are typically less developed than megacities but serve as key regional hubs (e.g., Xiamen, Taiyuan, and Changzhou); small cities have populations of 1 million or less (e.g., Sanya and Tongling). The event study analysis (Fig. 5) supports the parallel trends assumption for megacities and large cities. However, the parallel trends assumption is only partially satisfied in the small cities subgroup, indicating that the results for small cities should be interpreted with caution.

Table 2 Results of DID regression in cities with different sizes: effects of HSR implementation on intercity patient mobility in megacities, large cities, and small cities.
Full size table
Fig. 5: Event study analysis results of the effect of HSR implementation on intercity patient mobility across city sizes.
Fig. 5: Event study analysis results of the effect of HSR implementation on intercity patient mobility across city sizes.
Full size image

a Mega cities; b Large cities; c Small cities. All regressions use the year prior to HSR implementation as the baseline period; point estimates represent changes in patient mobility relative to the baseline, with error bars indicating 95% confidence intervals.

Model (4)–(6) are the DID modeling results, showing the effects of HSR availability on intercity patient mobility. The coefficients are positive for mega cities (\(\beta\) = 2.518, S.E. = 1.277, P value < 0.05) and large cities (\(\beta\) = 0.793, S.E. = 0.472, P value < 0.1), but insignificant for small cities (\(\beta\) = 0.199, S.E. = 0.237, P value > 0.1). The results indicate that the introduction of HSR leads to an increase of CKD patient flows to megacities and large cities by 23.02 ± 11.57% and 7.26 ± 4.26%, respectively. This pattern is likely attributed to the enhanced connectivity facilitated by HSR, which enables more patients to access high-quality healthcare resources and services typically concentrated in larger urban centers such as Beijing, Shanghai, and Shenzhen (Li et al., 2020). Simultaneously, the previously common trend in the road transportation era—where patients primarily sought care at hospitals in nearby small cities (e.g., from a small town to the nearest county hospital)—diminishes with the advent of HSR (Syed, Gerber and Sharp, 2013).

Lagging effects of HSR implementation

To examine the temporal evolution of HSR’s impact, we conducted yearly regression models separately for megacities, large cities, and small cities. As shown in Fig. 6, during the initial three years following HSR implementation, its effect on intercity patient mobility is significantly positive in both megacities and large cities. However, after this period, the effect in megacities turns negative, while it remains significantly positive in large cities, with an even greater magnitude. These findings suggest that in the long run, HSR implementation appears to attract more patients to large cities, which can function as regional hubs, rather than to megacities like Beijing and Shanghai. This patient mobility pattern has the potential to release healthcare service pressures on megacities and help balance medical demands across regions. Therefore, strategic investments should be considered in large cities to improve their healthcare infrastructure, to strengthen their roles as regional healthcare hubs. At the same time, it is important to consider the capacity of healthcare systems and economic costs, to ensure that the expansion of intercity access does not overwhelm local resources or exacerbate inequities in service delivery.

Fig. 6
Fig. 6
Full size image

Coefficients of yearly regression on cities of varying sizes.

Unfortunately, the coefficients for small cities remain insignificant. This suggests that transportation advancements alone are insufficient to address the scarcity of medical resources in small cities. Instead, targeted efforts are needed to decentralize healthcare resources through political or economic incentives aimed at enhancing the medical infrastructure and services in these areas. Furthermore, comprehensive regional healthcare planning is crucial to establish a hierarchical healthcare system across cities of varying sizes, so as to enable small cities to play an integral role within the broader network.

Robustness check

Placebo test: What is the effect of “fake” treatments?

To validate the assumption of no pre-treatment trends and to rule out spurious relationships, we run DID regressions on four groups of placebo samples (‘fake’ treatments) to reinforce the causal interpretation of our findings. As expected, when 100% and 75% of the testing samples were assigned as ‘fake’ treatments, HSR implementation exhibits an insignificant effect. However, the effect become significant at the 0.1 significance level when the proportion of ‘fake’ treatments was reduced to 50% and 25% (Table 3). These findings demonstrate the robustness of our modeling results and provide additional confidence in the validity of our conclusions.

Table 3 Placebo test results.
Full size table

2SLS regression: What is the causal effect?

To solve the potential endogeneity problem of the DID and OLS model and to estimate the causal effect of HSR on intercity patient mobility, we apply IV regressions with 2SLS specification as a robustness check (Table 4). In the first stage of the 2SLS, the distance to the nearest HSR hub city has significant and negative correlations with both the HSR implementation and HSR travel speed increase, indicating that the further away from hub cities, the likelihood that two cities have HSR in operation between them decreases, and the increase in HSR travel speed is less prominent. In the second stage, we find that both HSR and HSR travel speed improvement significantly result in more intercity patient flows. These results are consistent with the DID and OLS regression results, indicating the robustness of our findings. The coefficient of HSR implementation in the 2SLS regression is notably greater in magnitude than in the DID regression, indicating that the HSR effects in increasing intercity patient mobility would be more profound if HSR is implemented in random order. However, the coefficient of HSR travel speed improvement in the 2SLS regression is smaller in magnitude, which suggests that the OLS model might have overestimated the effect of HSR travel speed improvement.

Table 4 2SLS regression results.
Full size table

Subsample test: Is the effect consistent across different samples?

To address potential selection biases due to the potential non-random HSR deployment, subsample tests are conducted to examine whether the DID results hold across cities with different baseline GDP. According to the results (Table 5), the effect of HSR is significant and positive in both groups, which is also consistent with the overall HSR effects as represented in Table 1. Therefore, the results are robust within different subsamples.

Table 5 DID regression results in different sample groups.
Full size table

Discussions and concluding remarks

To bridge the research gap in quantifying the actual effects of advanced transportation technology on intercity patient mobility, our study uses HSR as a proxy for such technology and investigates its impact on the intercity mobility of CKD patients. Our findings demonstrate that HSR significantly increases intercity patient mobility. Notably, the effect of HSR in megacities turns negative three years after its implementation, whereas it remains positive and continues to grow in large cities. Our results suggest that HSR can reshape patterns of healthcare utilization, potentially improving access for some populations, while raising concerns about burden concentration in larger cities and patient outflows from smaller ones. By bridging transportation and health, this study highlights the societal implications of transportation advancements and underscores the need for collaborative frameworks between transportation and healthcare planning. Aligning infrastructure development with long-term health outcomes and equity, it sets a foundation for interdisciplinary research at the nexus of mobility and public health.

First and foremost, our findings suggest that HSR increases the amount of intercity patient mobility, which highlights the broader significance of transportation infrastructure, not just as a facilitator of mobility, but as a potential strategic tool for improving healthcare accessibility (Monzón, Ortega and López, 2013; Zhang et al., 2020a). This is particularly crucial in developing countries, where healthcare resources are relatively scarce and spatially concentrated (Wang and Nie, 2024). However, it is worth mentioning that although our estimates show statistically significant increases in patient mobility following HSR implementation, the effect sizes are modest.

Second, the evidence that HSR enhances intercity patient mobility to megacities and large cities, suggests that HSR may result in the centralization of healthcare-seeking behaviors, favoring large urban centers with advanced medical facilities. This centralization could exacerbate resource strain and further compromise care quality in urban centers that are already burdened by high patient volumes (Jiao et al., 2020; Zhou, Yang and Li, 2018), thus highlight the importance of coordinated policies that expand healthcare capacity of these cities and facilitate the hospitalization process for intercity patients (Jiang et al., 2017; Meyer and Elrahman, 2019). These policies could include an intercity healthcare insurance system to provide adaptable and comprehensive coverage (Ao et al., 2022; Kornelsen et al., 2021; Zhang et al., 2022), digital referral and online appointment systems to provide convenience to intercity hospitalization (Jia et al., 2020; Xu, Xu and Matalimanja, 2022), and relevant urban facilities to support intercity patients, who often require tailored accommodations, intracity transportation, and meal options.

Third, the redistribution of patient mobility driven by HSR also emphasizes the need for strategic allocation of medical resources. In the HSR era, large cities are becoming increasingly attractive to patients, positioning them as regional healthcare hubs that should be equipped with advanced medical facilities and specialized care for chronic diseases. However, the insignificant effects on patient flows to small cities—often already underserved—may exacerbate spatial disparities. Small cities often serve as a localized role in the healthcare system that serve local residents and complement the specialized capabilities of larger centers (Chen et al., 2021). Therefore, in the HSR era, policymakers must ensure that small cities remain competitive by improving their healthcare infrastructure, thus to maintain the hierarchical healthcare supply system that optimizes resource utilization and reduces the overburdening of megacities (Cavallaro, Buzzone and Nocera, 2023; Zhou, Yang and Li, 2018; Kim and Sultana, 2015).

Fourth, whether the change of intercity patient mobility driven by HSR will influence healthcare disparity also worth further investigation. Although patient mobility could potentially exacerbate healthcare inequalities by reinforcing disparities between large and small cities, it also has the potential to increase the efficiency of healthcare resource utilization, particularly in LMICs where healthcare infrastructure is highly imbalanced. This dual nature of patient mobility—both as a possible driver of inequality and a tool for improving healthcare access—highlights the need for deeper investigation, especially in the context of LMICs where resource allocation remains a critical challenge (Ding et al., 2023; Zhang et al., 2023; Zhong, Wu and Tao, 2024). Therefore, understanding the complexities of patient mobility and its impacts on healthcare equity is essential to formulating policies that both address inequality and optimize resource distribution in these regions.

Last but not least, by bridging the fields of transportation and health, this study sets a precedent for bridging the interconnected domains of transportation and health, both of which are vital to the well-being of the population (Giles-Corti et al., 2016). The findings call for the partnership between transportation and healthcare policymakers to create frameworks that align transportation advancement with long-term health outcomes, ensuring sustainable and equitable improvements in public well-being (Wan et al., 2021; Sallis et al., 2016). Additionally, transportation planning should involve comprehensive assessments of healthcare capacity and demand forecasts to prevent overwhelming services in megacities and neglecting medical needs in smaller communities (Shetty, 2011).

Admittedly, this study has some limitations that implies potential future research directions. Firstly, this study only focused on CKD patients’ mobility behavior due to data limitation. Although CKD is a suitable case for our study due to its prevalence and resource disparities, we acknowledge that healthcare-seeking behaviors and mobility patterns may differ across disease types, and future studies are needed to assess the generalizability of our findings using data from patients with other chronic or acute conditions. Secondly, while the IV strategy helps address potential endogeneity, the exclusion restriction may not be fully satisfied as the distance to nearest HSR hub might be correlated with unobserved regional characteristics. Though the results of balance test could validate the IV we used, we still recommend carefully interpreting the causal estimates of the IV regression. Thirdly, our analysis does not evaluate downstream impacts such as health outcomes, system capacity, or cost-effectiveness. Future work should integrate data on health outcomes, economic burden, and system-level responses to fully assess the broader implications of patient mobility facilitated by HSR (Wang et al., 2020).