Satisfaction Evaluation and Sustainability Optimization of Urban Medical Facilities Based on Residents’ Activity Data in Nanjing, China

Abstract

Research on public service facility evaluation has mainly focused on spatial accessibility and facility quality from a supply perspective, but has rarely focused on the evaluation of service facilities from a usage perspective. Researchers can observe the service quality, frequency of use, functional connotation, satisfaction level, and other aspects of facilities from the user’s perspective, effectively compensating for the disadvantage of insufficient precision in traditional macro statistical data. This study proposes a new method for evaluating medical facility usage and service quality based on residents’ activity data. We established an evaluation index system from the perspective of residents’ activity by measuring intensity, frequency, and service satisfaction through network grading data and online comments context. Then, we evaluated the supply and demand relationship of medical facilities, identifying the influencing factors of medical satisfaction. We have also proposed specific strategies for sustainable optimization of medical services. The results show that (1) the service attitude (0.024, 0.002**), service efficiency (0.133, 0.001*), and service quality (0.017, 0.001**) of medical institutions are the core factors that affect medical satisfaction, followed by the convenience of residents in accessing medical resources, showing a significant positive correlation (p < 0.01, Total R² = 0.9061); and (2) the medical service level in Nanjing City is generally balanced, although spatial heterogeneity exists in the Qixia and Jiangning boroughs.

1. Introduction

During the rapid urbanization of China over the last 20 years, the socioeconomic level and living standards of residents have developed significantly. However, public service facilities still need improvement. The large population and construction scale of megacities have led to many ‘urban diseases’, such as urban traffic congestion and unequal resource distribution. The further development of Chinese cities must transition from quantity accumulation to quality improvement. Functional improvement and allocation optimization of public service facilities have become the primary tasks for the development of megacities in China in the new era. The optimal allocation of public service facilities is an important way to realize social fairness and justice, because it balances the spatial distribution between different regions and satisfies the demands of different groups in their daily lives. In this generation, residents with different social backgrounds should have relatively fair access to balanced facility services considering the various population densities in different regions. Therefore, the most important task is to reasonably assess existing public service facilities.

In 1963, Cooper studied the location of industrial spaces and the location–allocation model of public service facilities, which initiated research in this field. Current research focuses on the three aspects of public service facility allocation [1,2,3,4], the spatial structure of facility allocation [5,6,7,8], and facility accessibility [9,10,11,12]. Many spatial models have also been proposed to continue empirical research on facility allocation evaluation, such as (1) the minimum distance model [13,14,15,16,17], (2) the maximum coverage model [18], (3) the gravity model [19], and (4) the improved gravity model [20]. In general, existing research that measures the quality of facilities based on spatial allocation has two flaws. First, service quality evaluation elements are inadequate; the researches only focus on the perspective of facility hardware and accessibility measurement, ignoring the subjective feelings of residents such as usage intensity, frequency, and satisfaction. Second, against the developmental background of rapid population migration, static index data cannot effectively reveal actual resident usage efficiency of public service facilities after allocation. Furthermore, reflecting the network connections among facilities and radiation effects on surrounding areas is also difficult.

Research on the evaluation of urban public service facilities has mainly been conducted from the following three aspects: supply, demand, and supply–demand matching. The evaluation based on the supply dimension mainly relies on building indicator data such as facility level, scale, and performance to evaluate service fairness, accessibility, equalization, regional differentiation, and related planning and configuration strategies [21]. The research methods include kernel density analysis [22], cluster analysis [23], reachability analysis [24], etc. The evaluation based on the demand dimension mainly uses methods such as satisfaction surveys, facility usage frequency surveys, and walking tolerance time surveys [25] to study the needs of different age, gender, and occupational groups [26] for facilities. The evaluation based on the supply–demand matching dimension is based on different scale spaces, combining objective evaluation of facilities with subjective feelings, and exploring the matching relationship between facility accessibility and regional population size. However, the existing standards for the configuration of public service facilities lack precise descriptions of facility types and supply–demand matching relationships, and the evaluation results reflect differences in the subjective cognitive level of researchers [27]. In addition, existing evaluation methods are limited by various factors such as facility spatial layout, service scope, and personalized needs of residents, and a systematic evaluation system has not yet been provided [28].

Satisfaction is a type of mental state and facility satisfaction is the quantitative assessment of residents’ psychological feelings about the facilities’ service quality through an evaluation index system. Currently, satisfaction research in the field of geography and planning focuses primarily on community spaces and public service facilities. The researchers obtained data mostly from questionnaire surveys and household interviews. An index system was built to explore specific satisfaction factors from the perspectives of spatial location, accessibility, and facility quality [29,30,31,32]. Traditional questionnaires and interviews have many shortcomings such as long research cycles, high labour costs, and limited sample sizes and coverage. Based on smart cities and mobile social networks, the activity data of large-scale residents can be effectively identified and collected. These types of data contain more elaborate spatiotemporal information, and rich activity attributes information that covers a large population. Activity data can effectively reflect the usage intensity and frequency between residents and facilities. It also provides a new research perspective and data source for studying urban public service facilities [33,34,35,36].

Previous studies have suggested that the impact of a living space environment on satisfaction includes the following two levels: spatial attributes and non-spatial attributes. The spatial attributes include the differentiation characteristics of community location and living space [35], urban built environment [37], resource allocation and accessibility [33], and other aspects. Meanwhile the non-spatial attributes mainly include the socioeconomic attributes of residents, community type [37], social capital [38], technological updates [38], and characteristics of resident groups [38], all of which have a strong impact on resident satisfaction [39]. From the perspective of spatial elements, various types of public service facilities constitute the core content affecting resident satisfaction. Previous studies have explored the impact of basic public service facility configuration on resident satisfaction [38], confirming the direct impact of objective environmental attributes and elements around residential areas on resident satisfaction. Taking medical service facilities as an example, research has found that the transportation accessibility [32], facility scale [33], service quality [33], and other attributes of facilities have a significant direct or indirect positive impact on resident satisfaction. From the perspective of non-spatial attributes, there is significant heterogeneity in the relationship between individual socioeconomic attributes and satisfaction among residents. Different attributes such as age, family income, education level, migration situation, and nature of daily work may all have an impact, and these factors are usually included as controlling factors in the satisfaction evaluation model. There are significant differences in the element system, degree of influence, and mechanism of action among different scenarios [35]. Some studies suggest that there is an indirect impact between subjective resident perception and expectations, which is controlled by individual differences [35]. Community communication has also been proven to have a significant positive impact on resident satisfaction [36]. Based on the perceived objects of residents, research has confirmed the mediating effect of subjective perception in the living system. Satisfaction is often decomposed into the following three aspects: environmental satisfaction, facility satisfaction, and social satisfaction [36]. These are then specifically decomposed into multiple dimensions of perceived content such as safety satisfaction, comfort satisfaction, convenience satisfaction, and community belonging [39].

Overall, safety and convenience are the basic needs that residents are satisfied with in the living environment [31], and these are directly influenced by the opportunities for material space access and the level of material space construction. Comfort and aesthetics are the satisfaction needs of higher-level residents, which are not only directly influenced by the elements of the living system, but also by the comprehensive influence of social attribute factors. The subjective satisfaction indicators and their mechanisms of action in the four dimensions have been partially confirmed, but the influencing pathways need to be further clarified [33]. There are two shortcomings in the research on the impact mechanism of existing resident satisfaction with public service facilities. First, the selection of facility types and levels is not comprehensive enough, and the definition of evaluation dimensions is not clear. Research mainly focuses on the accessibility indicators of facilities, while there is relatively little research on the satisfaction impact mechanism of comprehensive aspects such as facility scale supply, functional connotation, and service quality, thus making it difficult to propose targeted facility optimization and improvement strategies [38]. Second, due to the lack of targeted research theories and methodological frameworks, the mechanism and impact path of facility factor satisfaction in this study are not yet clear, and the correlation between various dimensional elements is difficult to quantify; in particular, the research on the interaction relationship between safety, convenience, comfort, and aesthetic perception systems is not sufficiently in-depth [39]. This article focuses on two issues, one of which is to analyse whether the supply and demand of medical resources have reached a balance in various towns and districts of Nanjing City, and if not, to determine what specific supply–demand contradictions exist. The second is to analyse which factors affect the satisfaction index of medical facility services, and to identify the specific mechanism of action. Based on existing evaluation methods for public service facilities, this study proposes a new method to evaluate public service facilities from the perspective of residents’ activities. We built an index system from the aspects of usage quality and evaluation to quantitatively measure residents’ satisfaction based on microblog check-in data and online comments context. Subsequently, we used medical facilities in Nanjing as an example and evaluated the service quality of the facilities from the perspectives of supply and demand. Finally, we found that the existing medical facilities in downtown areas basically satisfy the surrounding residents’ medical needs, although spatial heterogeneity also exists in some boroughs. Based on these research findings, we assessed the specific developmental requirements of medical and health facilities in the current overall urban planning of Nanjing and proposed some advanced suggestions for the promotion and allocation optimization of facility services.

2. Data and Methods

2.1. Study Area

Nanjing was selected for the case study. Nanjing is the capital city of Jiangsu Province, with a total land area of 6587.1 square kilometres and an urban built-up area of 868.3 square kilometres. In 2023, Nanjing had jurisdiction over 11 municipal districts, 94 streets, and six towns. Its total population was 9.282 million, the urban population was 8.066 million, and the urbanization rate reached 86.9%. The urban built-up area of Nanjing presents a ‘circle’ spatial pattern. The inner old city covers an area of 172.3 square kilometres, including streets in the Gulou District, Qinhuai District, and Xuanwu District. The population distribution in the old urban area of Nanjing is dense, and the city is in the late stages of urban transformation and development. The main urban area around the old city covers an area of 372.8 square kilometres, including Jianye District, Qixia District, Yuhuatai District, and other streets. The new outer urban area covers an area of 323.2 square kilometres, including Jiangbei New Area, Jiangning District, and other streets (Figure 1).

2.2. Data Collection

2.2.1. The Location and Attributes of the Medical Facilities

We obtained the medical facility location and attribute data (POI) through the Tencent Map Location Service API interface (http://lbs.qq.com/ accessed on 22 July 2023). The data included information on the facilities’ latitude and longitude coordinates, names, addresses, and categories (

Table 1. The data structure of urban medical facilities.

Field Name	Field Explanation	Sample Data
Id	The internal serial number	10018321328223
name	POI name	Nanjing Drum Tower Hospital
address	POI address	321 Zhongshan Road, Gulou District, Nanjing City, Jiangsu Province
Province	Province	Jiangsu Province
City	City	Nanjing City
Lng	Longitude	118.789
Lat	Latitude	32.063
Level	Facility level	the highest hospital division level in china
Category	Facility category	medical facility

). Using web crawler methods, we collected 297 medical facility POIs at different levels in Nanjing City. We converted the data format into Shapefile files, which were stored on a GIS platform. Because the Tencent Map coordinated system was GCJ-02, we switched all geospatial data into the WGS-84 coordinated system to maintain uniformity.

2.2.2. Residents’ Microblog Check-in Data

Sina microblog is a broadcast style social media platform based on user information exchange, sharing, and dissemination [22]. This online platform has a huge number of users. According to a survey report in 2023, the platform has 673 million active users per month, 378 million active users per day, and an average of over 1 billion posts per day [22]. The vast user base and vast amount of information can provide sufficient user samples and public opinion corpus for this study. Based on the Sina microblog check-in data interface, we collected 138,572 check-in data with the ‘medical and health facilities’ labels from April to July of 2017 in the whole Nanjing city area, including information concerning coordinates, check-in address, location label, etc. (

Table 2. The data structure of microblog ‘check-in’ data.

Field Name	Field Explanation	Sample Data
Id	The user serial number	10018321327988223
Lng	‘check-in’ longitude	118.783
Lat	‘check-in’ latitude	32.058
Place	‘check-in’ address	Nanjing Drum Tower Hospital
Label	Facilities type	Medical and health
Comments	Residents’ comments context	Today I came to the hospital for diagnosis and the doctor told me that my body is recovering well. Many thanks to all the medical staff

). Moreover, check-in data usually contain residents’ comments context on their experience of the facilities. We retrieved and extracted evaluation words and symbols from the comments context using a natural-language parsing method. Furthermore, we used the evaluation and public opinion words as the data source to evaluate satisfaction with the facilities’ services.

2.2.3. The Online Score of the Facilities Service

After comparing the professionalism degree, user coverage, and grading rules among different facility review websites, we chose the website named ‘Dazhongdianping’ (http://www.dianping.com/ accessed on 22 July 2023) as the data source to measure the online score of the facilities. We specifically collected the online scores of 226 medical facilities in Nanjing using a web crawler program. The online score data contained medical, facility, and service levels, and we summed the scores as the facilities’ final online evaluation scores (Figure 2). We then connected the facility addresses, microblog check-in data, and online scores using spatial location information. In cases of missing data in some community hospitals, we estimated the score using a spatial interpolation fitting method based on the scores of surrounding similar medical facilities.

The sample data were selected from Sina microblog and Dianping. The above two websites are representative social media platforms of China’s internet [40]. At present, the number of active users exceeds 600 million, covering a wide range of people. We searched for keywords, links to website API ports, and retrieved data once a day in the morning and evening. Each time, we crawled commenter information with no more than 100 comments and performed deduplication processing. After removing failed crawling and duplicate users, we obtained the basic user group and randomly extracted high-frequency users with a published data volume greater than 500. We collected 211,377 check-in data from these users. To unify the comparison criteria, we pre-processed the data, including deleting users with a fan base of no more than 10, as well as user samples with incomplete personal information or missing profile pages, homepage closure, and failed crawling. The final check-in data for analysis was 138,572. Considering the randomness of the failure to successfully capture personal information, this will not affect the analysis results. We conducted statistical analysis on these valid sample data and performed logarithmic function transformation with a base of 2 (Figure 3). As a result, it was found that the data on the number of followers was mostly moderate to low, and it also included extremely low-, medium-, and high-scale users. ‘Pay attention to other people’s data’ and ‘number of Weibo posts’ also show a clear distribution pattern of high in the middle and low at both ends, which conforms to the overall characteristics of normal distribution. The average number of information posted by users conforms to the power-law distribution characteristics, and the sample distribution has good coverage.

2.3. Research Framework

To unify the analysis unit of resource convenience, we divided the Nanjing urban area into a 1 km × 1 km grid, and the specific analysis framework is shown in Figure 4. At the level of facility supply, we selected the following three indicators to construct an evaluation system for medical facility supply: facility type, facility level, and spatial distribution. We calculated the comprehensive score of facility supply falling into the cell network as the indicator of facility supply within the geographic spatial unit. At the level of facility demand, we collected data on resident use of medical facilities from multiple dimensions, specifically selecting the following three indicators: facility usage intensity, usage evaluation rate, and comprehensive network evaluation. These were then used to construct a facility usage satisfaction evaluation index system. At the same time, we calculated the comprehensive score of facility evaluation that falls within the unit grid as an indicator of facility demand for that geographic unit. Based on unifying basic geographic units, we normalized the demand and supply indicators. This article identifies the supply and demand relationship of facilities at a spatial scale, and uses methods such as natural breakpoints to divide the facility supply index and evaluation index into the following three levels: low, medium, and high. The supply and demand levels within a unit are paired in a matrix to directly compare the relative differences between supply and demand levels. Obviously, supply and evaluation belong to the same level, which is most likely to result in supply–demand balance. The larger the level gap, the more imbalanced it becomes. In theory, the supply–demand relationship can be divided into nine states (

Table 3. Types of supply and demand relationships in medical services.

Supply–Demand Balance Index		Evaluation Level
		Low Level	Medium Level	High Level
Supply level	Low level	Low level balance	Low supply–medium demand	Low supply–high demand
	Medium level	Medium supply–low demand	Medium level balance	Medium supply–high demand
	High level	High supply–low demand	High supply–medium demand	High level balance

). Finally, based on the spatial supply and demand relationship of medical resources, we take administrative jurisdiction units as the statistical scope, summarize the balance of facility supply and demand matching in each jurisdiction area, and provide targeted optimization strategies for medical resource supply in different jurisdictions.

2.4. Variables and Indicators

The ultimate goal of public service facility allocation is to provide fair, convenient, and high-quality public services to nearby residents. To achieve this target, facility allocation focuses on the following three aspects: spatial accessibility, facility quality, and service quality [32]. Spatial accessibility reflects the spatial convenience from the place of residence to the facilities, while facility quality reflects the construction quality, functional level, and integrity of the facilities. Service quality reflects the specific usage levels and satisfaction with facilities among different resident groups. These three factors determined facility selection orientation when residents had access to public services (

Table 4. Existing research on medical facility evaluation from the supply perspective.

Research Perspective	Specific Indicators	Researchers
Spatial accessibility	Spatial location	[4,5,9,10,18]
	Service radius
	Traffic convenience
Facilities quality	Hospital grade	[8,13,15,23,24]
	Charges
	Patient size
	Hospitals/per capita
	Physician/per capita
	Hospital beds/per capita

). For medical service facilities, research has mainly focused on spatial accessibility and facility quality, but less on service quality from the residents’ demand perspective.

We built an evaluation index system from a demand perspective, mainly focusing on the usage quality and service evaluation of the facilities. We measured the facility usage intensity and frequency by the residents’ daily microblog check-in activities. Furthermore, we measured the online score from the review websites and measured the usage experience level, and the emotion level through semantic analysis of the microblog check-in comments. Finally, we classified all the indices and performed quantitative scoring to measure the facilities’ comprehensive service satisfaction (

Table 5. The evaluation index system of public service facilities from the demand perspective.

First Level Indicator	Second Level Indicator	Data Support	Explanation
Facility usage quality	Usage intensity	Sina microblog Microblog check-in data.	We measured the used quality using the overall strength and frequency per capita through Sina microblog ‘check-in’ data on medical facilities.
	Usage frequency
Facility service evaluation	Online score	Dazhongdianping.com	We quantified the users’ network rating on dazhongdianping.com and summarize the total score (depending on the different scoring standard on different types of service facilities.)
	Experience level	Evaluation comments context	We quantified the keyword and normalized for five satisfaction levels (‘very satisfied’, ‘slightly satisfied’, ‘neutral’, ‘slightly dissatisfied’, ‘very dissatisfied’)
	Emotion level	Emotional comments context	We extracted the emotional icon and emotional comments, quantified with five satisfaction levels (‘very positive’, ‘slightly positive’, ‘neutral’, ‘slightly negative’, ‘very negative’)

).

In response to the current situation of mismatch between supply and demand of medical service facilities, we have referred to the relevant requirements proposed in the ‘Medical Facility Service Evaluation Index System’ issued by the General Health Administration of China and established a facility service satisfaction evaluation index system [41]. The indicator system conducts satisfaction evaluation from the following five aspects: service effectiveness, service level, hardware facilities, institutional convenience, and charging situation. Each dimension includes multiple characteristic variables, forming the specific explanatory variables of the model (

Table 6. The index system for measuring medical service satisfaction.

Dependent Variable	Primary Indicators	Secondary Indicators	Basic Definition
Satisfaction evaluation of medical facility services	Service effectiveness	Service attitude	Satisfaction with the service attitude of medical institution staff
		Service efficiency	Satisfaction with the service efficiency of medical institution staff
	Service level	Service quality	Satisfaction with the service quality of medical institution staff
		Management level	Satisfaction with service provided by medical institution management personnel
	Hardware facilities	Facility integrity	Subjective perception of the quality and completeness of medical devices
		Facility progress	Subjective perception of the performance of medical devices
	Institutional convenience	Medical response speed	Satisfaction of medical institutions in responding to timely patient needs
		Medical convenience level	The convenience of residents in accessing medical resources
		Organizational level	Organizational effectiveness of medical institutions in providing medical services
	Fee situation	Medical service fees	Satisfaction with the level of medical service fees
		Drug and device charges	Satisfaction with drug and device fee levels

). The scoring method adopts the Likert scale, which is divided into 1 to 5 points [42]. A score of 1 indicates very dissatisfied, with ratings increasing in order, and a score of 5 indicates very satisfied. On October 16 and 17, 2023, two days in total, we released 355 online questionnaires on the internet and investigated the satisfaction of 38 medical institutions at all levels in Nanjing. As a result, we collected 310 valid questionnaires with complete content, with a response rate of 87.36%.

Reliability analysis is used to detect the consistency and stability between results obtained from multiple measurements [43]. The most used reliability analysis method is the Cronbach’s alpha coefficient. When the coefficient is greater than 0.7, it is considered to have strong reliability, while between 0.5 and 0.7, it is considered to have average reliability and can be further analysed [43]. From the overall sample indicators, the standard deviation is relatively balanced, reflecting that the respondents have a consistent level of understanding of the key indicators in the questionnaire. We conducted a reliability analysis on the 11 factors of facility satisfaction evaluation (

Table 7. The reliability analysis of factors influencing satisfaction with medical facilities.

Secondary Indicators	Cronbach’s Alpha Value	Standardized Cronbach’s Alpha Value
Service attitude	0.8821	0.9014
Service efficiency	0.8745	0.8901
Service quality	0.7892	0.8191
Management level	0.8135	0.8227
Facility integrity	0.8033	0.8122
Facility progress	0.7935	0.8039
Medical response speed	0.8121	0.8237
Medical convenience level	0.7731	0.7922
Organizational level	0.7362	0.7576
Medical service fees	0.7128	0.7531
Drug and device charges	0.7471	0.7918

), and used Cronbach’s alpha reliability coefficient to measure the intrinsic reliability of the questionnaire. The alpha reliability coefficients of each measurement indicator were calculated. From the results, the Cronbach’s alpha values of various factors that affect the satisfaction of medical facilities are all greater than 0.7, indicating that the internal structure stability of the data collected in the survey questionnaire is good and the internal consistency of the survey questionnaire is strong. We have good reliability in measuring the influencing factors of medical facility satisfaction using the prompts in the questionnaire.

2.5. Models and Methods

2.5.1. Natural Language Processing

During the development of online social networks, an increasing number of people published their assessments and attitudes towards the spatial environment through online social platforms such as microblogs. We collected 19,237 microblog check-in data containing comments about medical facilities in Nanjing. After data screening and cleaning, we selected 3000 high-quality comments as the training data set. First, we trained and calculated the word vector sets using the Word2Vec model (a type of natural language processing model supplied by Google Inc.) and calculated the weight of words based on the Term Frequency Inverse Document Frequency algorithm. We then trained and classified the remaining text data using the weighted word vector and the SVM-pref model. In addition, we used a supervized machine learning method to classify the evaluation and emotional words from the entire check-in comments context. Specifically, we used the Word2Vec model to train word vector sets and cluster similar concept words by calculating the cosine distance between them. We then expanded the clustered similarity domain vocabulary to an emotional dictionary and trained a high-dimensional representation of the word vectors. Finally, we performed high-dimensional vector reduction processing using principal component analysis (PCA) and classified these vectors using the SVM-pref model. Based on the above methods, we classified the evaluation and emotional words from the check-in comments (Figure 5). After the actual test, we found that the best classification results achieved more than 90% accuracy based on the Word2Vec and SVM-pref models. The results also confirmed that this method can achieve good performance in an emotional classification task.

2.5.2. Kernel Density Estimation

To present the overall spatial characteristics of resident activities in the study area, the kernel density estimation (KDE) method was used to identify certain closeness layers with different activity densities. The calculation formula for KDE is as follows in Equation (1)

$$f_{(s)}={\displaystyle \sum _{i=1}^n\frac{1}{h^2}}k(\frac{s-z_i}{h})$$

(1)

where f(s) is the kernel density calculation function of the spatial positions, h represents the distance attenuation threshold, k refers to the space-weighting function, and n refers to the points that cause the distance from the spatial positions to be less than or equal to h. The geometric meaning of this function is that the density values are the highest in each core element Z_i and then decrease for every element after Z_i; the nuclear density value decreases to zero until the distance to the core element Z_i reaches the threshold h. In addition, we determined the appropriate search radius to calculate the KDE results.

2.5.3. The Entropy Weight Method

The entropy weight method (EWM) can overcome the subjective defect in the artificial weight setting method and the information overlap among multiple indicator variables [44], and it is widely used in the social economy and other research fields [45]. The specific entropy weight formula calculates the original index data matrix as follows in Equation (2)

$$X={(x_{ij})}_{m*n},x_{ij}\ge 0,0\le i\le m,0\le j\le m$$

(2)

where m is the evaluation object quantity and n is the evaluation index quantity.

Based on the standardized processing of positive/negative indicators, we calculated the entropy of indicator i and the indicator weight $k_i$ as follows in Equation (3)

$$k_i=g_i/{\displaystyle \sum _{i=1}^n{g}_i}$$

(3)

Finally, we used the comprehensive evaluation method Equation (4) to quantitatively calculate the comprehensive service quality of medical facilities for the entire Nanjing City area

$$CSQ={\displaystyle \sum _{i=1,j=1}^n{k}_i{x}_{ij}}$$

(4)

2.5.4. Multiple Regression Analysis Method

We used the general form of static multiple linear regression models, and set the n-variable linear regression equation of the dependent variable Y with respect to the independent variables $X_1$, $X_2$, $X_3$…$X_n$ as shown in Equation (5)

$$Y=b_0+b_1{X}_1+b_2{X}_2+\cdots +b_n{X}_n+\epsilon$$

(5)

Among them, $b_1$, $b_2$…$b_n$ are referred to as partial regression coefficients, and ε is a random error term. We used the p-value to determine whether the results are significant and whether there is a correlation between variables; additionally, it was used to determine the direction of the impact of X on Y. When the regression coefficient value is greater than 0, it indicates a positive correlation between indicators; otherwise, it is a negative correlation.

3. Results

3.1. Medical Facilities Present a Central Agglomeration Peripheral Scattered Distribution Pattern

Depending on the medical facility POIs obtained through the map interface, we created a spatial visualization and hierarchical division depending on the facility scale (Figure 6a). Thereafter, we used the KDE method to determine the spatial distribution density of the medical facilities (Figure 6b). We found that the allocation of medical facilities showed significant differences between the central city and the peripheral areas in Nanjing. Specifically, the distribution shows strong agglomeration features in the central areas and relatively scattered facility allocations in the city’s peripheral region. Overall, medical facilities were mainly allocated to the Gulou, Qinhuai, and Xuanwu boroughs. The peripheral areas have little agglomeration in the district centre, although the remaining areas show a scattered status. In the central city, we found that medical facilities were mainly allocated to old urban streets, such as Fuzimiao, Xinjiekou, Hunanlu, and Zhongyangmeng Streets.

3.2. The Intensity of Medical Facility Usage Shows a Clustering Feature in the Main City Center

To perform the quantitative analysis, we built a 1 km² grid as the minimum research unit covering the entire city area. Therefore, we connected the microblog check-in data to space grids on a GIS platform. Specifically, we measured the residents’ microblog check-in cumulative number related to the medical facilities in each grid as facility usage intensity (Figure 7a). We also measured the cumulative check-in times per capita related to the medical facilities in each grid as the facility usage frequency (Figure 7b). A Likert scale was used to stratify the standard quantitative results. Depending on the results, the overall facility usage quality characteristic shows a high-value agglomeration distribution in the city centre and dispersed distribution in peripheral areas. Regarding facility usage intensity, we found that the maximum usage intensity areas were in Gulou and Qinhuai boroughs, while few areas showed high-usage intensity in Xuanwu, Jiangning, and Qixia boroughs, and other areas showed low-usage intensity. Regarding facility usage frequency, the results show a high-usage frequency mainly occurring in large medical facilities, and the usage frequency weakened successively according to the scale of the medical facilities.

3.3. The Rating Levels of Medical Facilities Exhibit Multi-Centre Distribution Characteristics

An online social networking platform is an important channel for residents to share and exchange their usage experiences in facilities. We collected online scores and comments related to medical facilities from facility review websites and microblog check-in data. Specifically, the comments contained facility usage experience and emotional attitudes among different resident groups, which also indirectly reflected residents’ satisfaction with the facilities. We built models to classify the evaluation and emotion keywords from the comments context using natural language processing and other semantic analysis methods. We also trained and recognized the evaluation and emotion keywords to measure residents’ service satisfaction with the medical facilities (

Table 8. Semantic analysis example of microblog comments context.

Original Comments Context	Semantic Segments	Semantic Analysis Results
		Second-Level Indicators	First-Level Indicators	Satisfaction Level
The hospital has perfect medical conditions. The medical staff have a good attitude. And the doctors are very responsible to us. Also the operation and postoperative nursing were excellent. The whole medical treatment process is joyful and stress-free	‘Perfect medical conditions’.	Hardware facility	Facilities quality	5
	‘The medical staff have a good attitude’	Service attitude	Service evaluation	5
	‘The doctors are very responsible’
	‘The operation and postoperative nursing were excellent’.	Medical quality
	‘The whole medical treatment process is joyful and stress-free.’	Emotional level	Activity mood	5

). Regarding the online score, we found that high-scoring facilities were concentrated in Gulou and Qinhuai boroughs; although in other boroughs, the rating score was uniformly distributed (Figure 8a). Regarding the facility usage experience level, in addition to the high-evaluation medical facilities agglomeration downtown, the other borough received some scattered high-evaluation medical facilities (Figure 8b). Regarding the aspect of usage emotion level, we found that patients’ moods were generally low in tertiary hospitals, and generally high in small hospitals. This phenomenon is related to the major comprehensive hospitals that mainly perform diagnosis and treatment of emergency and critical diseases, and the patients’ and their families’ moods correspondingly undergo great emotional fluctuation (Figure 8c). However, small- and medium-sized hospitals mainly provide healthcare services, and the patients’ moods are relatively calm.

3.4. Concentration of High-Quality Medical Resources within the Central Urban Area

After calculating each evaluation index, we measured its weight using EWM (Table 8). We also measured the comprehensive service satisfaction with medical facilities in Nanjing and the five indexes as follows: facility usage intensity, usage frequency, online score, experience level, and emotion level. Based on this, we conducted a correlation analysis between each index variable and comprehensive service satisfaction and found a significant positive correlation (

Table 9. The weight and correlation analysis of each evaluation index.

Evaluation Indexes	Usage Intensity	Usage Frequency	Online Score	Experience Level	Emotion Level
Weight (W)	0.157	0.166	0.387	0.196	0.094
Regression coefficient (R)	0.825	0.849	0.922	0.878	0.814
Significance level (sig.)	0.001	0.005	0.003	0.000	0.008

).

We performed spatial mapping (Figure 9a) and KDE analysis (Figure 9b) for the comprehensive service satisfaction values on the GIS platform. We found high satisfaction values in downtown areas and higher values in the southern areas. High-quality medical resources were distributed around the city centre. Medical facility allocation was relatively dispersed and service satisfaction values were relatively lower in the peripheral areas. In the central city, high-satisfaction medical resources were mainly concentrated on Gulou Street, Shanghai Street, and Confucius Temple Street.

Considering that medical facilities are mainly governed by administrative boroughs in China, we mapped comprehensive service satisfaction to each borough in Nanjing (Figure 10a). We found that the spatial distribution characteristics were higher in downtown areas and lower in peripheral areas. The area with the highest service quality was Gulou borough. Due to the dense distribution of various provincial organs, universities, scientific research institutions, and other social institutions, this borough provides a great deal of employment and has the highest population density. The second is Qinhuai borough, which has a large population and diversified medical facility allocation, followed by Xuanwu, Yuhuatai, Jiangning, Jianye, and Qixia boroughs. It should be noted that service satisfaction levels improved in Pukou and Luhe boroughs. With the official approval of the Nanjing Jiangbei National Development Zone in 2015, the regional development level will be greatly enhanced and the corresponding allocated medical resources will be followed up one after another. Moreover, Lishui and Gaochun boroughs have the lowest service satisfaction levels because of the long distance to downtown and the scarcity of medical resources. With the opening of a high-speed railway, medical services would be more convenient in downtown and suburban areas. Based on the equitable development demands of public service resources, medical resources will radiate actively from downtown areas to the suburbs in the near future.

We furthermore compared the resource supply and residents’ demand for medical facilities. The results generally showed asymmetrical relationships and spatial heterogeneity (Figure 10b). Medical facility allocation quality was relatively higher in the central city areas such as Gulou, Xuanwu, Jianye, and Qinhuai boroughs, but the quality was relatively lower in peripheral areas such as Jiangning, Luhe, Lishui, and Pukou boroughs. Qixia and Gaochun boroughs had the lowest facility supply levels. However, some distinctions exist between Qixia and Gaochun boroughs. The boroughs near Qixia have relatively better medical facilities than those near Gaochun. There is a convenient rail transit and fast road connection between Qixia and downtown, rather than Gaochun, where residents can enjoy relatively better medical services in Qixia borough. This is also reflected in medical service satisfaction; Qixia borough is not equipped with adequate medical facilities, but the overall service satisfaction remains at a moderate level. Because Gaochun borough was far away from downtown and had weak traffic levels, the service satisfaction level was relatively lower than that of the other boroughs.

The model takes the overall satisfaction of residents with medical facilities Y as the dependent variable and the corresponding 11 secondary indicators as independent variables. Through correlation analysis, it is found that there is a significant correlation between the dependent variable and the 11 independent variables (

Table 10. Regression analysis results of satisfaction evaluation of medical facilities. ** p ≤ 0.01; * p ≤ 0.05.

X	B	SEM	Beta	T	p	VIF	R2	F
Constant	1.545	3.172	—	0.338	0.626	—	0.807	25.64 (p = 0.000)
Service attitude	0.024	0.135	0.273	1.823	0.021 *	3.156
Service efficiency	0.133	0.116	0.182	1.762	0.018 **	2.233
Service quality	0.017	0.122	0.353	1.369	0.010 *	3.371
Management level	0.175	0.139	0.311	1.912	0.026 *	2.873
Facility integrity	0.007	0.114	0.319	2.018	0.031	6.577
Facility progress	−0.067	0.131	−0.174	−2.006	0.036 **	2.361
Medical response speed	0.021	0.151	0.277	1.723	0.022 *	2.198
Medical convenience level	0.014	0.018	0.219	1.162	0.028 **	1.221
Organizational level	0.037	0.127	0.308	1.671	0.023	7.891
Medical service fees	0.028	0.135	0.193	1.482	0.016	8.871
Drug and device charges	−0.021	0.119	0.239	1.228	0.027 **	3.577

). The fitting effect of the model is shown in Table 10. The R² value of the model is 0.807, which can explain the 80.7% change in the overall satisfaction index, indicating a high degree of model fitting. The multi-collinearity results of the model show that the VIF values of facility readiness, organizational level, and medical service fees are greater than 5 and less than 10. The results indicate that these variables have collinearity issues. After removing the three collinear variables mentioned above, we conducted a new linear regression analysis, and the results are shown in

Table 11. Regression model analysis results without collinearity indicators. ** p ≤ 0.01; * p ≤ 0.05.

Primary Indicator	Explanatory Variable	Multiple Linear Regression Model
		Coefficient	T-Statistic	p-Value
Service effectiveness	Service attitude	0.024	0.48	0.002 **
	Service efficiency	0.133	1.82	0.001 *
Service level	Service quality	0.017	0.39	0.001 **
	Management level	−0.175	2.17	0.005
Hardware facilities	Facility Progress	0.067	2.61	0.015
	Medical response speed	0.021	3.80	0.003
Institutional convenience	Medical convenience level	0.018	2.33	0.002 *
Fee situation	Drug and device charges	−0.021	−0.37	0.018
Constant		1.545	16.69	0.000
R2	Inter group	0.8073
	Within the group	0.9781
	Total R2	0.9061
Total F-statistic		25.64
Haussmann test: Total p-value		0.001

.

The Haussmann test results of the model show a p-value of 0.001 and strongly reject the null hypothesis at the 5% level, indicating that the choice of a multiple regression model is reasonable and the statistical data are robust. The total R² of the model is greater than 0.9, the F-statistic is 25.64, and the corresponding p-value is 0.001, indicating significant differences in the data and passing the significance test. The results indicate that the service attitude, service efficiency, and service quality of medical institutions are the core factors affecting medical satisfaction. Second, it is the convenience of residents in accessing medical resources that evaluates the accessibility of residents to high-quality medical resources in the surrounding area from the perspective of transportation. This is similar to the conclusions of existing research, which analysed the satisfaction and influencing factors of residents in Seattle, USA, towards urban medical facilities. It was found that the service efficiency and quality of medical facilities are the core factors, followed by the accessibility of medical facilities [46]. Due to the larger scale of cities in the United States compared to Chinese cities, residents often need to drive to seek medical treatment. Medical institutions that are too far away are not conducive to residents seeking medical treatment, which can reduce the actual experience of residents seeking medical treatment [47].

4. Discussion

4.1. Optimize and Improve Medical Facilities in the Main Urban Area, with a Focus on Enhancing Medical Facilities in the Peripheral Suburbs

Based on the relationship analysis between the supply and demand of medical facilities, as well as the relevant planning requirements of healthcare in overall urban planning, we propose some targeted suggestions to improve the allocation of medical facilities in Nanjing. Specifically, existing medical facilities in downtown areas have satisfied the surrounding residents’ medical needs, and the current planning policy does not require that tertiary hospitals in downtown areas scale up. Therefore, improving planning policy requires improving the quality and service level of medical facilities and scaling up the medical facilities (gradually promoting them to tertiary hospitals) in the Xuanwu and Jiangning boroughs, which is consistent with the analysis results. Improving the planning policy requires building a tertiary hospital separately and improving the service quality of existing medical facilities in Pukou, Lishui, and Gaochun boroughs. The results also show that the quality of medical facilities allocated to the suburbs was relatively lower than that in downtown areas. Therefore, we suggest promoting medical service quality in Qixia borough by strengthening the allocation of tertiary and special hospitals because of the relatively low quality of local medical facility allocation and service satisfaction.

We draw conclusions based on the results of data analysis. First, there is heterogeneity in the supply and demand relationship of medical resources in various regions of Nanjing. Within the main urban area, Gulou District and Qinhuai District show a high level of supply and demand balance, Xuanwu District and Yuhuatai District show a moderate level of balance, Jianye District, Qixia District, and Jiangning District show a low level of balance, while Lishui District, Gaochun District, and Liuhe District show a significant mismatch in supply and demand. Second, from the perspective of factors that affect the satisfaction of medical facility services, the service attitude, service efficiency, and service quality of medical institutions are the core factors, followed by the convenience of residents in accessing medical resources, with an overall significant positive correlation (p < 0.01). The specific mechanism of action includes the following points. First, areas with a dense layout of public service facilities are usually located in the central urban area of the city. Although there is a high density and level of facility supply, and the accessibility of facilities is generally high, service quality and efficiency are still key factors affecting satisfaction. Second, the scale of facilities has a significant positive impact on resident satisfaction through the mediating effect. Research has shown that spacious, comfortable, and large per capita public service facilities have a significant impact on the user experience of residents, and improving facility supply is an important aspect of improving resident satisfaction. Some studies also suggest that the increase in facility scale will to some extent reduce the convenience and ease of use of facilities, and it is necessary to balance facility scale indicators and accessibility indicators. Finally, the quality of public service facilities comes from the intuitive perception of residents regarding public service facilities. Some studies suggest that the quality of public service facilities is reflected in multiple aspects such as safety, convenience, comfort, and aesthetics. Improving the service quality of public service facilities is the most direct and efficient way to manage and improve satisfaction.

4.2. Provide Precise Public Services Based on Supply and Demand Relationships to Maximize Facility Utilization Efficiency

Depending on the facility service location theory, it is necessary to rationally lay out public service facilities and maximize facility efficiency and social welfare because of the government’s limited financial capacity. Public service facility allocations should also provide public services to all citizens equally [48]. Presently, medical service quality shows a dualistic differentiation, because the medical service level is relatively higher in downtown than in suburban areas. Resource allocation and service quality should be counterbalanced by strengthening the construction of high-grade medical facilities in suburban areas and transferring high-level urban medical resources to the suburbs. However, this equalization is not average, meaning that residents have the same rights to access public services, and their rights do not depend on the backward economic conditions of their living area. This implies maintaining a basic level of public service quality at no less than the average level in all regions and ensuring that residents have equal opportunities. The final public service equalization status should reflect residents’ access to efficient and convenient public service quality in different regions, between urban and rural areas, and between individual residents. On the other hand, equalization is not a fixed state or ultimate goal, but a process of dynamic adjustment, which means maintaining a dynamic balance between the quantity and quality of public service facilities in downtown and suburban areas. The equalization of public service facilities in different areas should be improved from a ‘balanced quantity stage’ to a ‘balanced quality stage’ with continuous social economic development. This development should maintain the equalisation of facilities and services, guarantee the density of the facilities, and enrich facility types in the allocation process to achieve equality of opportunity for all types of residents. Second, facility allocation should guarantee facility standardization, and the policy should focus on the full coverage and sharing of high-quality resources.

4.3. Strengthen Functional Complementarity and Network Connectivity between Medical Facilities of Different Levels

To advance the work of the government and planning departments, the traditional configuration method of public service facilities requires innovations based on the type and size of land use, population distribution, service radius, and other indicators. New methods and technologies, such as the internet, mobile phones, GPS, and other big data analysis methods, should be adopted to measure the daily living patterns and needs of residents in different regions. Service efficiency and satisfaction with public service facilities can be evaluated on this basis. Then, the facilities’ layout, scale decision, and regional collaborative planning can be implemented reasonably, as well as upgrades and reconstruction of facilities. However, attention should be paid to the layout balance and policy guidance of public service facilities at different spatial scales, such as downtown and suburban areas, old towns, newly developed areas, commercial areas, and other peripheral areas. For example, the government should increase investments in community-level medical facilities and establish a sound medical service system to promote the rational and efficient flow of medical resources. The promotion of subway network operations, intercity rail, and other rail transit, as well as the layout of public bicycles and shared bicycles, should be prioritized to strengthen traffic links between downtown cities and suburban areas. Optimizing the construction of parking lots around medical facilities, intelligent transformation, and establishing a comprehensive metropolitan area integrated transport service system would enhance the spatial accessibility of medical facilities.

4.4. Actively Guide Refinement and Improvement of Medical Facility Functions Based on the Supply–Demand Relationship of Medical Facilities

High-precision and long-term resident activity trajectories can be effectively measured and obtained through the widespread use of sensor networks, location-based services (LBS), and other kinds of ICT. An increasing number of studies have focused on the perspective of residents’ activities and demand, to identify existing city problems and seek the optimized direction of urban space updates. Meanwhile, we can grasp the overall characteristics of urban operations, such as urban land use level, traffic congestion, and material information contact, from a macroscopic perspective based on the activity data of resident groups. Therefore, the government could allocate resources and facilities, such as the layout of large targeted medical facilities. Finally, the research scale of public service facility allocation has also transformed into a refined community scale; however, an imbalance in the supply and demand for medical facilities still exists. Through the collection of residents’ individual activity data, the activity characteristics and actual needs of local residents can be effectively grasped from the street, community, and other micro levels to provide a new decision-making basis for the spatial layout and function optimization of community-based medical facilities that meet individual needs (Figure 11).

5. Conclusions

5.1. Key Findings

Based on a review of existing literature, this study constructed a facility service quality evaluation system from the perspective of activities. Based on this, an overall evaluation of the current service quality of medical facilities in Nanjing was conducted from the perspective of residents’ needs and actual use. In combination with the relevant planning requirements in the fields of traditional Chinese medicine and health in the current urban master plan, targeted development suggestions for improving facility functions and optimizing layouts are proposed as follows:

(1) Medical facilities present a centrally aggregated, permanently scattered distribution pattern. Research has found a significant gap in the layout level of medical facilities between the central urban area and peripheral areas in Nanjing, and the spatial distribution has a strong agglomeration effect. New medical resources have not strengthened the construction of medical facilities in peripheral areas with the development of the city, and the trend of resource concentration towards urban areas is more obvious.

(2) The intensity of medical facility usage presents a clustering feature in the main city centre. The overall feature of the intensity of medical facility usage in Nanjing city is ‘high centre periphery’. Gulou and Qinhuai Districts had the highest total usage of medical facilities; Xuanwu, Jiangning, and Qixia Districts had a higher usage intensity of some medical facilities, while other areas had a relatively lower usage intensity. From the perspective of facility usage frequency, the per capita usage frequency of large medical facilities in each district was relatively high, whereas the usage frequency of small- and medium-sized medical facilities decreased gradually at the facility level.

(3) The rating levels of medical facilities exhibit multi-centre distribution characteristics. High-quality medical resources were concentrated within the main urban areas, presenting a pattern in which the south was better than the north. The layouts of medical service outlets in suburban and township areas are relatively scattered and service quality is uneven. In terms of online ratings, high-scoring facilities cluster in the Gulou and Qinhuai Districts, while the distribution of facility ratings in other districts is balanced. In terms of Weibo text evaluation, in addition to presenting the characteristics of high evaluation clustering of large hospitals in the central urban area, some hospitals in the Liuhe and Lishui Districts also had high Weibo evaluations. In terms of measuring patient emotions, patients in large hospitals generally had lower emotions, whereas patients in small- and medium-sized medical institutions generally had higher emotions.

5.2. Implications

This study is a positive attempt to evaluate facility service satisfaction based on residents’ activity data. It enriches the data sources and methods for the evaluation of public service facilities from the perspective of residents’ demand. In addition to urban renewal projects, this method can also be applied to pre-investigation and current status analysis in urban master planning, which has a wide range of application values. Existing research lacks an in-depth analysis of the synergy among medical service networks in different regions. In the future, the functional optimization of medical facilities should actively build integrated functional systems and conduct targeted functional promotion and complementarity. In addition, future research should focus on the following two aspects: medical facility allocation should consider the surrounding urban residential land layout and a reasonable prediction of the future population distribution to reduce idleness or over-utilization of medical resources. As the marginal zone between city and country areas, the suburbs also provide important supportive functions for urban–rural integration. The allocation of public service resources in rural and other urban fringe areas should be strengthened to narrow the gap between them and the city centre. The disparities among boroughs can effectively absorb the labour force in suburban areas and ease pressure on urban public service facilities which need more attention.

5.3. Limitations and Future Research Directions

This paper selects a specific period of a single city to obtain internet data as part of the data source. Compared with traditional statistical data, the magnitude, scale, and fineness of data have comparative advantages, but there are also some limitations and biases. First, the samples obtained from online data still have limitations and cannot obtain full sample data. Among all the people who use medical facilities, middle-aged and elderly people account for a large proportion. These people are not skilled in the use of mobile phones and the internet. It may be that people who often use the facilities have not published microblog comments, which reduces the accuracy of model analysis. Second, in the current environment, the speed of network information update is very fast. From the perspective of application, the timeliness of internet big data is often very short. Any data are located on a continuous timeline and has a temporal attribute, which is the age of the data. Data of different ages have different value characteristics, and traditional statistical data often have the value of overall or trend analysis. Internet data are more conducive to provide feedback on the space–time distribution characteristics of specific time nodes. In the subsequent research process, we further collected data from multiple time nodes, conducted comparative analysis before and after, and cross-validated the accuracy and rationality of the research conclusions.

Currently, research on the evaluation of public service facilities in China exists mainly in the fields of planning, geography, and government management. Facility allocation research has made great progress from a beginning of fully accepting foreign theories to consideration of China’s national requirements [49]. For the planning department, urban renewal work focuses more on planning, implementation, and evaluation in the current development stage [50]. To promote functional enhancement and optimization allocation, a more comprehensive evaluation of existing facilities in terms of facility usage and service quality is required. Public service facility planning policies should change from their previous focus by relying on a single land-use population size, block level, and other static indicators to consider the needs of residents and carry out human-oriented planning.

This study mainly calculated satisfaction with medical facilities based on residents’ activity data. Users of social network platforms such as Sina microblogs and facility review websites are mainly young people, which creates a biased sample. Nevertheless, we built an evaluation index system mainly from the perspective of residents’ demands based on residents’ check-in and review data. Integrating these new evaluation indexes into the existing evaluation index system and balancing the relationship between allocation supply and resident demand is a key direction for further research.