Can Regional Eco-Efficiency Forecast the Changes in Local Public Health: Evidence Based on Statistical Learning in China

Regional eco-efficiency affects local public health through intermediaries such as economic and environmental impacts. Considering multiple factors, the implicit and uncertain relationship with regional characteristics, and the limited data availability, this paper investigated the forecasting of changes in local public health—including the number of visits to hospitals (VTH), outpatients with emergency treatment (OWET), number of inpatients (NI), number of health examinations (NOHE), and patients discharged (PD)—using calculated regional eco-efficiency with the Least Square-Support Vector Machine-Forecasting Model and acquired empirical evidence, utilizing the province-level data in China. Results: (1) regional eco-efficiency is a good predictor in both a single and multi-factor situation; (2) the prediction accuracy for five dimensions of the changes in local public health was relatively high, and the volatility was lower and more stable throughout the whole forecasting period; and (3) regional heterogeneity, denoted by three economic and demographic factors and three medical supply and technical level factors, improved the forecasting performance. The findings are meaningful for provincial-level decision-makers in China in order for them to know the current status or trends of medical needs, optimize the allocation of medical resources in advance, and enable ample time to tackle urgent emergencies, and, finally, the findings can serve to evaluate the social effects of improving regional eco-efficiency via local enterprises or individuals and adopting sustainable development strategies.


Introduction
Economic and environmental factors affect local residents' health, and environmental and medical decision-makers attach importance to the trends of changes in local public health (CLPH) [1]. Early prediction of the changes in local public health can provide sufficient time to balance the supply and demand of medical resources such as by making people prepared to respond to medical emergencies [2] and optimizing the resource allocation of medical materials in advance, thereby dynamically promoting a local medical service level [3]. Regional eco-efficiency can serve as a flexible indicator that integrates the relevant economic or environmental factors, those which scholars indicate can be chosen as good exogenous predictors for public health. Based on the organic performance of input and output factors [4], regional eco-efficiency (REE) has been treated as an explicitly important indicator due to it integrating both economic and environmental impacts and being closely related to healthcare sustainability [5]. Therefore, is there any empirical evidence to show REE can forecast CLPH with acceptable accuracy? for forecasting CLPH using REE with Chinese provincial data, and a further comparison analysis of forecasting performance is conducted. Sections 5 and 6 show the summary and conclusion.

The Inner Relationship between CLPH and REE
The inner relationship between CLPH and REE originates from their basic concepts. REE is a concise and simplified comprehensive index and emphasizes the monetary costs of various resources and the environmental changes required for economic development. Meanwhile, CLPH is also closely and directly associated with this kind of economic and environmental change, with regards to the requirement of maximum profits and minimum pollution. In addition, a variety of factors, such as the consumption of energy, pollutant emissions in waste gas, pollutant emissions in waste water, industrial solid wastes, the increased value of industrial development, and so on [11,12], originate from the measurement of REE and impact CLPH simultaneously. The above internal connection is an important foundation for building a predictive model, and the related literatures provide a basis for theoretical feasibility.
(1) In the economic dimension [13][14][15][16][17], scholars have investigated how economic activities embedded in REE affect CLPH [18]. As the most basic and critical requirements, the physical health of residents necessitates food, exercise, spiritual guarantee, and so on, which are achieved under one important premise: that personal income level and economic development trends provide the fundamental roles [19]. Residents earn income through employment to purchase energy and nutrition, for continuous life and education services, etc. [20], and to afford the expenditure for necessary medical supplies, equipment, and services [21]. Macroeconomic trends serve as leading indicators for residents' disposable incomes on the micro-level, especially for healthcare [22][23][24]. In addition, many other social or economic activities, such as city planning [25,26], immigration [27], aging [23], or housing [28], affect the changes in local public health too. In addition, REE includes the continuing impact of economic activities, and it reflects the quantitative effects from the input-output perspective. For example, producers can optimize the decision-making process of resource allocation.
(2) In the environmental dimension [29,30], there is a variety of related literature that have probed and shown evidence that the surrounding environments incorporated into local eco-efficiency have impacted local public health. Calculations of REE already encompass environmental input-output factors, including both the BADS and GOODS [4], which have led to changes in local public health to some extent. For example, beyond a certain concentration range, the BADS, such as particulate matter (PM 2.5 ) or sulfur dioxide (SO 2 ), deteriorate the living environments of residents and pose a great hidden danger to public health. In particular, excessive PM 2.5 or SO 2 have caused many diseases of the respiratory and nervous system with both short-term and long-term damage.
Firstly, as a hot topic, public health has also been suffering from the air pollutant emissions of the manufacturing industry [31][32][33][34][35]. The highly frequent appearance of haze episodes has brought huge stress to physical and psychological health and social daily operations. Yu, Wang [36] assessed this kind of negative impact in China by using satellite observations, and Gao, Woodward [37] conducted a review of the changes in haze pollution and local public health. There are many potential risks when the concentrations are big enough. PM 10 , NO 2 , O 3 , and CO are bad for CLPH [31,[38][39][40].
Secondly, solid or plastic waste is resistant to degradation, has low costs, and is rapidly growing, which squeezes living space and keeps deteriorating sanitary conditions [41][42][43][44] no matter what kind of waste, from economic activities or daily life. Using modified eco-efficiency indicators, Woon and Lo [45] focused on the public health and solid waste management of Hong Kong. Langdon, Chandra [46] pointed out that solid or plastic waste has led to contaminants entering the living environment. Solid or plastic waste has also caused public health to be exposed to heavy metals such as lead, mercury, cadmium, and arsenic [47,48]. Moreover, solid or plastic waste affects the growth of plants and changes in local public health [49,50], and it is not conducive to the effective prevention of toxic substances and infectious diseases and weakens the effects of health work on medical institutions and CLPH [41,51].
Thirdly, excessive water pollution is another important derivative that affects changes in local public health during social or economic progress [52], and, although wastewater has been purified to be utilized again [53,54], chemical compounds-toxic micropollutantshidden in the water pollutants have gradually evolved into a huge health risk [55,56]. Saha, Rahman [57] pointed out that through ingestion or dermal contact, local residents are likely to be diseased. CLPH have continued to deteriorate and have caused various diseases due to the pesticides or toxic metals in both the irrigation and drinking water systems [58].

The Keys to Forecast CLPH with REE
When forecasting CLPH using REE, there are the following issues that need to be settled. (1) There are many factors that make the relationship so complex. It is necessary to adopt a new technique (SVM) to map the linear, nonlinear, or some complex implicit relationship because CLPH are influenced by economic, environmental, and individual factors, as well as others (as discussed in the last section). Simultaneously, REE with six control variables works to add to the practical interpretability. (2) There is an implicit and uncertain relationship description when using REE to forecast CLPH, especially as this paper applied five indicators as proxy variables for CLPH and six control variables. Whether they are positive or negative impacts and linear or nonlinear, this needs more quantitative evidence. (3) Different regional characteristics require a quantitative comparison of prediction performances. Considering the regional heterogeneity, it is necessary to build or estimate a model for a single region. Moreover, since there are five proxy variables for residents' health status, there is a question worth discussing about the relatively higher prediction accuracy obtained via eco-efficiency and six other control variables. (4) There are limited sample data, and how to obtain better prediction accuracy with limited data is another question. Multiple factors and regions generally require more data to complete the fitting, obtain the optimal parameters, and further predict the data within or outside the sample on a secondary basis. It needs a strong learning ability and effective use of a small sample of information.

Data and Variables
Considering the data availability of and lack of data on Tibet, Hong Kong, Macao, and Taiwan, in the empirical Section 3, all provinces or cities in China were taken into account. The time period is from 2002 to 2016. This paper adopted "SBM (Slacks-Based Measure)" [59] and DEA-SOLVER Pro 5.0 [60] to measure the REEs. The descriptive statistics of the main variables to calculate the REEs are listed in Table S1 and Figure 1. The results are consistent with most other studies [61]. The eastern values of REE are higher than the western values. The value of the eco-efficiency of the whole nation is up to 0.51 in 2016.
The data of VTH, OWET, NI, NOHE, and PD (mainly from hospitals) are from the Chinese Medical Health Statistics Yearbooks from 2003 to 2017. The main statistics descriptions can be obtained from Figure 2. Indicators related to REE and all the control variables were mainly collected from the China Statistical Yearbooks from 2002 to 2017. All the indicators related to value were excluded because of the effect of inflation on the prices in 1998.
keep the time range of all the variables used consistent, we had to choose all data from 2002 to 2016. Although the eco-efficiency in 2017-2019 can be calculated, statistical data such as basic medical conditions (the statistical data of medical personnel, licensed doctors, or health care institutions) were scarce or had different statistical calibers, and we were limited to unifying the range of selected years. In addition, the impact of COVID-19 after 2019 can be seen as an uncertain external impact, which may need to be the focus of future research.   It can be seen that CLPH maintained a more moderate growth trend and so did the REE. However, from the theoretical explanation, they cannot be arbitrarily predicted using linear methods because there are many factors that determine the health levels of residents, for example, physical fitness, wealth, psychological factors, exercise methods, etc., which is consistent with the view in Section 2. Considering the implicit and uncertain relationship between REE and CLPH, regional characteristics, and low data availability, the following In addition, due to the limitation in the same frequency processing of the data collection of the other variables in the forecasting model, the eco-efficiency calculation period is 2002- 6 of 19 2016 [62]. First of all, it was difficult to obtain the energy-related/CO 2 -related input and output indicators (shown in Supplementary Materials: Table S1, including the main variables of the SBM to calculate the regional eco-efficiency) in some provinces and cities, such as Tibet, which limited our sample range. Secondly, there were many independent variables and dependent variables in the prediction model. This paper applied 5 indicators as proxy variables for CLPH and 6 control variables, including the development level of regional GDP, urbanization, population, and the number of local medical personnel, local licensed (assistant) doctors, and local health care institutions. In order to keep the time range of all the variables used consistent, we had to choose all data from 2002 to 2016. Although the eco-efficiency in 2017-2019 can be calculated, statistical data such as basic medical conditions (the statistical data of medical personnel, licensed doctors, or health care institutions) were scarce or had different statistical calibers, and we were limited to unifying the range of selected years. In addition, the impact of COVID-19 after 2019 can be seen as an uncertain external impact, which may need to be the focus of future research. Figure 2 displays the regional average levels of all the used indicators in China. It can be seen that CLPH maintained a more moderate growth trend and so did the REE. However, from the theoretical explanation, they cannot be arbitrarily predicted using linear methods because there are many factors that determine the health levels of residents, for example, physical fitness, wealth, psychological factors, exercise methods, etc., which is consistent with the view in Section 2. Considering the implicit and uncertain relationship between REE and CLPH, regional characteristics, and low data availability, the following section draws on the advantages of the LS-SVM-FM in mapping and identifying the relationship (even if non-linear), which can ensure the fitting effect and prediction accuracy.

Method Design
SVM performs well in building models when there are many factors or a nonlinear data pattern with small samples in many literatures, including [63][64][65][66][67]. Based on these, this paper utilized its relevant methods to ensure the fitting effect and prediction accuracy.
(1) The implicit relationship could have a much clearer mapping in the high-dimensional hyper feature space by constructing a hyper plane and finding support vectors to represent all the information, which allowed us to predict with a small sample of data.
(2) Its diverse kernel functions (linear and nonlinear) could meet the need for complex forecasting alongside the commonly used linear models, which allowed us to predict with the complex or uncertain relationship of the forecasting model. In the high-dimensional feature space, the proposed method adopted the nonlinear kernel to map the non-linear function learned by a linear learning machine, the process of which is not limited to spatial dimensionality.
(3) Compared with other methods, taking the "Structural Risk Minimization Principle" as the principle, SVM enabled our method to be equipped with an improved classification power [68], which allowed us to acquire a better forecasting accuracy for each region in China with a good fitting [67].
(4) Most forecasting based on SVM such as the Least Square-Support Vector Machine (LS-SVM) has already been applied to time series data, and this study extended it to regional panel data by constructing the LS-SVM-FM.
Based on the classic LS-SVM, LS-SVR, and LS-SVR-DS, this paper built the LS-SVM-FM with different regions and multiple factors.
With the dependence on the two parameters σ and γ, the solution of the LS-SVR can be modified as the following equation: It is better to apply the optimal method to obtain what are the true values of those main parameters by minimizing the average of squared errors. It can be displayed as In the empirical parts, the proposed LS-SVM-FM took the CLPH as Y, whose proxy variables are, separately, the VTH, OWET, NI, NOHE, and PD. x 1 , x 2 , x 3 , . . . , and x 7 represent variable values of REE, and all 6 of the control variables are represented as X. The next section introduces the relevant data and variables in detail. The continued LS-SVM-FM is written as the following: It rewrites as . . , P, where P denotes the number of regions or province or cites. Here, P is 30 and stands for the 30 provinces or cities in China. p = 1, . . . , Q, where Q denotes the number of variables. Here, Q is 7 and stands for the 7 different variables including REE and the control variables in China. k = 1, 2, · · · N, where N is equal to 15, and k stands for the specific year from 2002 to 2016. There are four kinds of kernels. The Radial Basis Function (RBF) kernels K(x, x k ) = exp − x − x k 2 /2σ 2 were chosen as the specific form, which has been unanimously recognized by scholars with the most frequent application, relatively [69].

Main Steps
Without an explicit close form on σ and γ of G, here, we provide the following algorithm of the search procedure [69].
Step 4. Replace σ 0 with σ 0 + λ σ and γ 0 with Step 5. When G(σ 0 , γ 0 ) ≤ ε or k ≥ N, the iteration can stop. Otherwise, set k ≥ k + 1 and return to Step 2. The iteration can stop either when the forecasting accuracy can be achieved or the computation is finished within an exogenously prespecified iteration number N. When the algorithm stops, it finds the 'optimal' pair of (σ 0 , γ 0 ) for the LS-SVM-FM, which minimizes the training error.
The main procedures are described as follows: (1) We applied each of the 30 provincelevel datasets in China to the LS-SVM-FM and performed in-sample learning and fitting to determine the parameter value and out-of-sample prediction and comparison to determine the prediction accuracy; (2) mean percentage error (MPE) and mean square or standard deviation of prediction error (MSE or SDE) were chosen to judge the prediction accuracy; (3) the VTH, OWET, NI, NOHE, and PD for CLPH were respectively taken as Y; (4) REE and the control variables in China were adopted as X, and the LS-SVM-FM without or with each of the 6 control variables were compared, respectively, and we obtained the best forecasting model with a lower prediction error; and (5) the forecasting accuracy with the single factor and multiple factors in China was drawn from the comparison analysis [69]. y T and y t denote the known sample values of the year t.ŷ T andŷ t denote the predicted sample values of the last year and the year of t using the LS-SVM-FM. t = 1, 2, · · · , T.

Forecasting Accuracy with the Single Factor and Multiple Factors
Chinese data were applied to the LS-SVM-FM to obtain empirical evidence. To understand the prediction better and keep a reasonable explanation, each of the five proxy indicators of CLPH for the thirty different provinces and cities in China were forecasted, and we compared the prediction errors from the following three aspects: (I) utilizing the single-factor-REE to forecast the CLPH in China   When only using the REE to forecast the CLPH in China, performance comparisons within Tables 1 and 2 provide more information. Firstly, overall, REE can better predict the health status of regional residents of the five proxy variables in China. The MPE values all fall within the acceptable interval, and in particular, the average MPE values of the CLPH are 1.295%, 1.235%, 2.961%, 16.028%, and 2.985%, although the NOHE is bigger than 10%, and 12 of the 30 regions show a bigger than 10% prediction error. As with the literature mentioned above, REE owned the impacts from both economic and environmental aspects at the same time, and it is crucial and well-behaved for describing the health conditions of residents. Change in eco-efficiency affect living conditions and thus the changes in local public health. Therefore, it can be regarded as a good predictor, helping decision-makers to quantify future changes in residents' health in advance and, finally, adjust various medical supplies and technical preparations.
Secondly, the volatility of the forecasting error appears quite differently in each province or city, as represented by the bigger values of MSE and SDE for NI, NOHE, and PD compared with the other indicators. Their bigger MSE values are in part due to a smaller statistical unit, but SDE is much more convincing, with values of 128.99, 70.52, and 33.00. Another possibility is whether the model ignores important explanatory variables or other observed factors because local public health can actually be impacted by a number of factors no matter if on the individual or environmental level, as analyzed in the previous literature review.
Thirdly, regardless of the vertical comparison of a resident's health status or the horizontal comparison of different indicators, significant differences in forecasting accuracy between provinces and cities also exist, or the influence of regional heterogeneity on prediction accuracy is very obvious. Regional decision-makers should notice the phenomenon.
The specific situations of different provinces or cities are important clues for analyzing the above differences of LS-SVM-FM (1). By adding the relevant control variables to consider the multiple factors of the LS-SVM-FM at two times, it helps to understand the aspects confirmed in the above analysis. Tables 3 and 4  and SDE for CLPH in most provinces or cities in China. For example, the minimum averages of MPE are 0.06%, 0.05% 0.12%, 13.72%, and 0.08% and as are the values of MSE and SDE. This can be attributed to the control variables to provide better information for machine learning methods in order to identify more realistic mapping relationships in high-dimensional spaces. Tables 3-6 present similar results as those in Tables 1 and 2. The other reason for the overall continuously improved prediction effect is that the radial basis kernel function better describes the above relationship, and it can take into account the linear and nonlinear relationships between multiple explanatory variables to the greatest possible extent. The 7 explanatory variables (eco-efficiency and 6 control variables) and the one-to-one regression for the 30 selected regions of China established high requirements for the sample size. Because there are only 15 years of data, the traditional panel model fitting and prediction effects were limited. However, the method proposed in this paper only needs to find the support vectors due to the advantage of the conversion of the highdimensional space, but with extra data or information still needed. At the same time, the powerful calculation and learning capabilities make up for the limited data.

Secondly, the main findings in
Thirdly, the NOHE of CLPH owns the bigger prediction errors for MPE, MSE, and DSE in the three models than the other four. Some points can explain some of the reasons. For example, the raw data of health examinations fluctuated greatly in 2007, especially in the Shaanxi Province. In addition, as well as the factors already considered, the health examinations may be related to the medical insurance system in China and medical process of medical and health institutions, and further research is required.

Forecasting Variation with the Single Factor and Multiple Factors
The previous section gave specific prediction accuracies and a corresponding direct analysis. However, when actually predicting the CLPH, in addition to the annual forecast performance and change, scholars also arouse attention to how the changes in forecast errors shift across the time dimension, including changes in averages (average degree) and changes in variance (variation degree), that is to say, how the MPE, MSE, and SDE change according to time. At the province level, it was shown that the values of the averages and standard deviations of MPE, MSE, and SDE for each of the five proxy variables of the changes in local public health levels in China. Furthermore, it can be learned that the concentration trend and degree of dispersion of the forecast error change, based on which the reliability and robustness of the models can be analyzed.
As is shown in Tables 5 and 6 By comparing the average degrees and variation degrees of prediction errors such as MPE, MSE, and SDE, from the global perspective, LS-SVM-FM (3), taking into account all six control variables, was more reliable and has a higher relative robustness than LS-SVM-FM (1) and LS-SVM-FM (2), although the other two models are also acceptable within a certain range of prediction accuracy. Meanwhile, the forecast volatility of NI, NOHE, and PD significantly expanded, so it is the best choice to make short-term or spot predictions on the above three dimensions of the CLPH.

Main Revelation
REE affects the CLPH through intermediaries that are integrated and represented by the economic and environmental impacts from the inputs or outputs of computing the REE at the provincial level in China. Specifically speaking, green sustainable development is to improve eco-efficiency and encourage consumers, producers, and managers to avoid excessive pollution [12]. The continuous increase in green behaviors in work and life has improved the living environment on which residents depend. Under the premise of ensuring environmental protection, economic achievements improve the disposable income for living standards and medical conditions and reduce pollution in living environments and guarantee a reduction in disease.
The calculated prediction results can be used as the basis for evaluating the specific social effects of adopting sustainable development strategies by local enterprises or individuals. Making residents' living or health conditions better is one of the most fundamental pursuits of a higher REE in each province or city. Local inhabitants are the ultimate maintainers and beneficiaries. Therefore, empirical findings by forecasting CLPH via REE could serve as a tool to evaluate the performance of REE-related policy formulation and activity implementation and find out the actual effects and deficiencies that need to be addressed to guide sustainable development practices.
The innate differences and respective characteristics between provinces in China are an important material for explaining the imbalance of spatial medical demand and supply distribution. For example, there are three economic and demographic factors and three medical supply and technical level factors to reflect regional heterogeneity. These factors show why the real situations of CLPH and the magnitudes of change are different across different provinces, and the above six factors explain the different effects of increasing REE promotion on local CLPH, although there are other individual, behavioral, climatic, psychological, and even political factors for CLPH such as nutrition, climate change, noise, institutional determinants, medical insurance, and so on. However, considering the quite limited data availability from micro-individual statistics, it is necessary to investigate what the interactions are between the economy, environment, and local public health at the overall macro-and meso-levels.

Policy Implications
To be more specific, this study is very helpful for decision-makers in each province of China to understand and optimize the allocation of medical resources. With the help of the early information on CLPH obtained with the right proposed model, which can forecast VTH, OWET, NI, NOHE, and PD with a high prediction accuracy in 30 provinces or cities of China, decision-makers can take this as the quantization basis to confront some urgent emergencies via a continuous supply of medical supplies. It is an important guarantee for changes in local public health.
Taking VTH as an example, in addition to the general medical supplies, different departments of VTH require independent professionals and medical resources, and more advanced forecasts provide time and a quantitative basis for the production, purchase, and storage of various medicines, disinfectants, or medical tools, from a general point of view. Combining the whole forecast for VTH with the ratios of all the sub-departments on average, there will be more evidence to distinguish the most important demand or emergency, and the expensive medical supplies to be purchased from others. As it is shown in Figures 3, S1 and S2, the Departments of Internal Medicine, Chinese Medicine, Surgery, Obstetrics and Gynecology, and Pediatrics ranked in the top five, which reflects the differentiated needs and the five most common problems in residents' health. It seems obvious that the medical resources required by the five departments vary greatly. The treatment methods of the Department of Chinese Medicine have more Chinese characteristics. The medicines are concentrated in Chinese herbal medicines. The production and use of medicines and rehabilitation training require special medical equipment (acupuncture equipment, medical Tuina, etc.). In addition, the requirements for medical equipment for testing, diagnosis, or even treatment between Departments of Internal Medicine and Surgery vary widely with a higher accuracy. The Department of Pediatrics have higher requirements for the various ingredients of drugs, which are different from those for adults, as are the mentioned medical staff and job requirements in different professional directions. Therefore, these top five need to be given enough attention according to the real conditions of the 30 selected regions in China, and plans should be made about the following aspects, including enough medical workers, prepared medical resources, and earlier cooperation with upstream and downstream enterprises.  From the perspective of risk and early warning, related forecasting results can be used as an important reference for early warning and risk identification in public health. Taking VTH as an example, more detailed comparisons of different sub-departments between regions are attached in the Supplementary Material. Obtaining the regional heterogeneity of CLPH in the figures through our predictions, based on the perspectives of local residents' eating habits, disposable incomes, and population density, decision-makers can formulate local medical material reserve methods and emergency medical incident response plans. As the figures show, for the top five sub-departments with high numbers of visits that are urgently needed in various regions, policy or tax support can be provided to promote the healthy development of the related industries in the long run. For shortterm fluctuations in individual provinces or cities, certain consultation or coordination mechanisms can be adopted between other regions to deploy medical personnel and materials to increase efficiency and reduce the waste of resources, just as that in Guangdong, Shandong, and Shanxi. As a whole, forecasting results from the other four indicators-OWET, NI, NOHE, and PD-can also be utilized as with the above analysis with some specific auxiliary information. Accurate predictions from the above five dimensions can help to detect residents' medical conditions. The primary advantage of the above results is that they can optimize medical supplies and personnel in various regions in time and grasp the overall situation of different types of medical needs. For example, according to the need changes in OWET, inpatients, health examinations, and patients discharged, the medical industry can dynamically adjust the supply and reserves of materials, reduce inventory, and minimize waste and excessive use of medical resources, and especially important medicines and instruments that are in short supply and have a long production cycle.
From the perspective of decision optimization, through findings on the control variables and how to calculate the required indicators of regional eco-efficiency, we can learn a differentiated path to improve public health in different regions. When adopting economic policies and measures for local sustainable development for improving the REE, they should consider regional differences and be possible to adjust to the most urgent and corresponding factors that affect CLPH in real-time, according to their own economic development level. Furthermore, it is helpful to strengthen regional cooperation to optimize the allocation of medical resources. These main findings can guide the industry or the From the perspective of risk and early warning, related forecasting results can be used as an important reference for early warning and risk identification in public health. Taking VTH as an example, more detailed comparisons of different sub-departments between regions are attached in the Supplementary Material. Obtaining the regional heterogeneity of CLPH in the figures through our predictions, based on the perspectives of local residents' eating habits, disposable incomes, and population density, decision-makers can formulate local medical material reserve methods and emergency medical incident response plans. As the figures show, for the top five sub-departments with high numbers of visits that are urgently needed in various regions, policy or tax support can be provided to promote the healthy development of the related industries in the long run. For short-term fluctuations in individual provinces or cities, certain consultation or coordination mechanisms can be adopted between other regions to deploy medical personnel and materials to increase efficiency and reduce the waste of resources, just as that in Guangdong, Shandong, and Shanxi. As a whole, forecasting results from the other four indicators-OWET, NI, NOHE, and PD-can also be utilized as with the above analysis with some specific auxiliary information. Accurate predictions from the above five dimensions can help to detect residents' medical conditions. The primary advantage of the above results is that they can optimize medical supplies and personnel in various regions in time and grasp the overall situation of different types of medical needs. For example, according to the need changes in OWET, inpatients, health examinations, and patients discharged, the medical industry can dynamically adjust the supply and reserves of materials, reduce inventory, and minimize waste and excessive use of medical resources, and especially important medicines and instruments that are in short supply and have a long production cycle.
From the perspective of decision optimization, through findings on the control variables and how to calculate the required indicators of regional eco-efficiency, we can learn a differentiated path to improve public health in different regions. When adopting economic policies and measures for local sustainable development for improving the REE, they should consider regional differences and be possible to adjust to the most urgent and corresponding factors that affect CLPH in real-time, according to their own economic development level. Furthermore, it is helpful to strengthen regional cooperation to optimize the allocation of medical resources. These main findings can guide the industry or the government to strengthen the close cooperation between upstream and downstream enterprises in the medical industry. Accurate forecasting guarantees that there is enough time to carry out the following work: technical cooperation that breaks through key technical bottlenecks, resource coordination that reduces overall risks, and personnel exchanges that share prevention experience.

Main Findings
REE is a highly synthetic indicator with integrated economic and environmental impacts that is associated with local CLPH. Considering that there are multiple factors affecting CLPH in addition to REE, such as the implicit and uncertain relationship between the two, regional characteristics, and low data availability, this paper investigated how to forecast CLPH using REE by utilizing the LR-SVM-FM and acquire empirical evidence utilizing the regional province-level data in China.
Taking REE as the main predictor and province-level data in China, this paper investigated how five proxy variables of CLPH were predicted, with different control variables including more economic and demographic factors and three more medical supply and technical level factors. Some interesting empirical findings were that (1) REE is a good predictor for predicting residents' health, whether in a single-factor situation or a multifactor situation. (2) The proxy indicators that measure the health status of residents have different prediction effects. The prediction accuracy of VTH, OWET, and NI is relatively high and the volatility is lower and more stable throughout the whole forecasting period.
(3) Utilizing three economic and demographic factors and three medical supply and technical level factors can improve forecasting performance. (4) The LR-SVM-FM based on machine learning meets the forecasting needs: regional heterogeneity of provinces and cities in China, limited samples, uncertain functional relationships, etc.
As explained and proposed earlier, the results show that (1) REE is a comprehensive indicator that combines the dual impacts of the economy and the environment, which are also important factors that affect residents' health conditions. (2) The proposed prediction model relying on the machine learning method can better characterize the uncertain and complex relationship between different regions and multiple influencing factors with limited samples. (3) Six control variables from economic factors, technical factors, and demographic factors improve the model with a higher degree of explanation, which is more in line with the real phenomenon.

Future Research
This article tried to conduct interdisciplinary research by forecasting CLPH using regional eco-efficiency with integrated economic and environmental impacts. Future research based on this could include searching for more micro-individuals and psychological indicators, or happiness indexes, to improve prediction models; quantifying the impacts of regional interactions on the prediction effect; and understanding how medical emergencies and responses to them can be influenced by the prediction performance.
Supplementary Materials: The following supporting information can be downloaded at https:// www.mdpi.com/article/10.3390/ijerph20021381/s1. Table S1: main variables of the SBM to calculate the regional eco-efficiency; Figure S1: the number of visits of outpatient emergency departments by sub-department in hospitals in China; and Figure