Exploring the Effect of the Number of Primary Care Health Workers on Diarrhea Morbidity and Where the Health Resources Should Go: a Community-based Geospatial Analysis Study From China

Background: Diarrhea remains a major threat to developing countries like China. Improved water, sanitation, and hygiene (WASH) have been known as cost-effective ways for reducing the morbidity of diarrhea. Primary healthcare institutions in China have been playing pivotal roles in the maintenance of WASH facilities as well as providing health education intended for improving residents’ appropriateness and safety in using those facilities. In this study, we aimed to explore the association between the number of primary healthcare workers and diarrhea morbidity at community levels in order to provide evidence-based implications for optimizing primary healthcare resource allocations. Methods: We collected annually time series of diarrhea morbidity, the number of primary healthcare workers, and relevant data of 4,321 communities in Sichuan Province with 83.41 million residents, China, from 2017 to 2019. The global and local Moran’s I were calculated to detect the spatial clustering of diarrhea morbidity each year as well as to identify areas where increased primary healthcare resources should be allocated. The spatial lag xed effects panel data model was adopted to explore the association between the number of primary healthcare workers and diarrhea morbidity. Results: Positive global autocorrelation of diarrhea morbidity was identied based on Global Moran’s I. Signicantly high-high and low-low clusters of diarrhea cases as indicated by Local Moran’s I were found to be mainly distributed in western and eastern Sichuan during the studied period years, respectively, which demonstrated consistency with regional economic development status and primary healthcare resource allocations among different regions. The diarrhea morbidity was found to be negatively associated with the number of primary healthcare workers with a coecient of -0.187 and the P-values <0.05, indicating that a 0.187 reduction of diarrhea morbidity (1/10,000) was associate with doubled amounts of primary healthcare workers. Conclusions: Our ndings highlighted the role of primary healthcare resource allocation in the process of diarrhea prevention and control which was reected by the number of primary healthcare workers among different regions. Our ndings also implied that constant efforts should be addressed at governmental and health administrative levels in order to facilitate diarrhea prevention and control, especially in the remoted area where the regional economic development status and primary healthcare multiplier tests to identify whether a spatial lag effect or a spatial error effect model should be selected. The results showed that the spatial lag effect was more signicant than the spatial error effect. Therefore, we nally used the spatial lag xed effects panel data model to examine the associations between diarrhea morbidity and the number of primary healthcare workers. detect as well as visualize the global and local spatial autocorrelation of annual diarrhea morbidities among different communities, based on which areas where increased healthcare resources should be allocated were identied. The spatial lag xed effects panel data model was then employed to explore the relationship between the number of primary healthcare workers and the morbidity of diarrhea at the community level. The Moran’s I analysis revealed the positive global autocorrelation of diarrhea morbidity and identied high-high clusters to be mainly concentrated in regions with relative fewer amounts of primary healthcare workers. As indicated by the regression outcomes, a negative relationship was identied between the number of primary healthcare workers and the morbidity of diarrhea, with a 0.187 reduction of diarrhea morbidity (1/10,000) associated with doubled amounts of primary healthcare workers.


Introduction
Diarrhea remains a major threat to public health around the world. According to the Global Burden of Disease [1], in 2016, diarrhea ranked as the eighth leading cause of death among all age groups and the fth leading cause of death for children less than ve years old. Residents mostly affected by diarrhea as a major threat to populational health, especially for children aged less than ve years old were found to be mainly distributed in underdeveloped regions including Africa, South East Asia and the Eastern Mediterranean [2,3]. Under-ve children in developing countries suffered from an average of 2.9 diarrhea onsets each year [3], with approximately one-third of total onsets being moderate or severe cases [4]. In China, more than a million diarrhea cases were reported in 2018 alone, with the morbidity of diarrhea found to be 92 cases per hundred thousand population, making it the second highest among noti able diseases in terms of incidence [5].
Diarrhea, a gastrointestinal infection, can be caused by a wide range of pathogens, including bacteria, viruses, and parasites [6]. A critical transmission route of diarrhea is the fecal-oral route, such as the consumption of fecally contaminated food, drinking water as well as via person-to-person contact due to poor hygiene [7][8][9]. The morbidity and distribution of diarrhea would be affected by various factors, including sociodemographic factors (age, education, income etc.) [10][11][12], environmental and sanitation factors (poor access to a good water source and poor sanitation) [13,14], and climate factors (rainfall, temperature and humidity) [15][16][17].
Diverse efforts have been made in attempt to reduce the morbidity of diarrhea in a worldwide range, among which improved water, sanitation, and hygiene (WASH) facilities such as piped water, protected shallow wells, and non-shared toilets have been widely accepted as cost-effective ways for reducing the morbidity of diarrhea [18][19][20]. In China, the rapid socioeconomic development has signi cantly improved the nationwide penetration of improved water, sanitation, and hygiene facilities. For example, according to National Statistical Yearbook, in urban regions, the penetration of piped water increased from 63.9% in 2000 to 98.36% in 2018 [21]. However, various factors have posed huge obstacles for Chinese residents in obtaining actual access to improved WASH [22]. For example, in 2008, 2.81 million disability-adjusted life years (DALYs) and 62,800 deaths were attributed to unsafe water and poor sanitation or hygiene in China. Water pollution in the nation was found to be inadequately controlled [23] as 44% of nationwide rural water supplies failed to meet minimum drinking water quality standards [24]. Under such circumstances, it is noteworthy that improved WASH facilities are not necessarily associated with improved hygiene behaviors such as safe feces disposal or improved handwashing procedures, which has been highlighted by researchers such as Lamichhane [25] while has received relatively inadequate investigation based on previous literature.
As improved hygiene behaviors should be addressed as an indispensable aspect in achieving improved public health outcomes in addition to the enhancement of WASH facilities, the optimization of health education programs as well as health service delivery at primary healthcare level should be emphasized as an essential strategy in improving residents' actual access to improved WASH thus further reducing the incidence of diarrhea. Speci cally, primary healthcare institutions have been playing critical roles in China as gatekeepers throughout the process of infectious disease prevention and control, including cutting off transmission routes, protecting vulnerable populations, providing treatment for infected patients as well as providing health educational programs and assisting in the maintenance of WASH facilities among communities. Under such context, it is not di cult to imagine that improved WASH maintenance and health education programs at primary healthcare level are very much likely to achieve the reduction of nationwide diarrhea morbidity via improving residents' appropriate use of WASH facilities, minimizing the infection of well water in rural areas via disinfection procedures as well as monitoring the concentration of bacterial accumulation in water pipes on a regular surveillance basis.
Despite the signi cant contributions that primary healthcare institutions have been playing in the process of infectious diseases prevention and control, studies focused on investigating the roles of primary healthcare workers in reducing diarrhea morbidity remain limited based on previous literature. Speci cally,a couple of studies [26][27][28] have veri ed the effectiveness of regular visits conducted by community healthcare workers (similar with primary healthcare workers) in reducing childhood diarrhea morbidity, while several other studies [29,30] have highlighted the signi cant role of community healthcare workers in improving residents' health literacy. However, evidences collected under the context of China's healthcare system were found to be very limited in this aspect, while none of the currently existed studies have ever been focused on evaluating healthcare resource allocation as a determinant for healthcare institutions' performances in reducing residents' diarrhea morbidity, especially at primary healthcare level where the distribution of health resource is mainly re ected by the number of healthcare professionals.
In order to bridge such gap embedded in previous literature, this study has been designed for exploring the association between the number of primary healthcare workers and residents' diarrhea morbidity in the community range in order to provide evidence-based suggestions for healthcare resource allocation at the community level. In addition, our ndings were expected to provide practical implications for other infectious diseases as diarrhea has a list of characteristics typical of communicable diseases including high incidence, diverse pathogens and transmission routes, as well as would be affected by sociodemographic factors and the construction of health infrastructures at regional levels. Diarrhea morbidity and relevant data from 2017 to 2019 in Sichuan Province, China were collected for analysis. The spatial lag xed effects panel data model was adopted to explore the relationship between the number of primary healthcare workers and diarrhea morbidity in the community range. The local indicators of spatial association (LISA, Local Moran' I) analysis was used to determine areas where increased healthcare resources should be allocated.
This study was expected to contribute to the relevant literature in two aspects. First, our study was expected to bridge the gap embedded in similar studies through exploring the association between the number of primary healthcare workers and diarrhea morbidity at community levels. Through identifying the role of primary healthcare resource allocation in infectious diseases prevention and control in China, our ndings were also expected to assist in the formulation of region-speci c policies by providing evidences on speci c locations of high incidence clusters. Second, our study was expected to add evidences to previously published studies in this eld which were conducted at county levels via providing new evidences collected at community levels, which served as a better solution to heterogeneity issue as the spatial variation of disease morbidity would be detected and analyzed at a smaller health administrative unit. This paper has been structured to contain the following sections. Speci cally, an overview of health authorities as well as their roles in the process of infectious diseases prevention and control was brie y described in the Background section. The study area, data sources and empirical strategies were described in the Methods section, which was followed by Results and Discussion sections where ndings and discussions were provided, respectively.

Background
In China, at least 23 departments are involved in the process of infectious diseases prevention and control, which can be divided into 6 categories based on their functionalities, including governments, Health Commissions, three kinds of specialist institutions, four kinds of social insurance institutions, thirteen kinds of supportive departments, and other organizations [31][32][33].
Speci cally, governments are responsible for providing surveillance on infectious diseases prevention and control from a holistic perspective via the initiation and management of projects proposed for communicable disease prevention and control. Speci c tasks related to achieving project goals are planned and managed by Health Commissions, for which social insurance institutions such as health insurance institutions will be responsible for providing support in multiple aspects including funding, human resource and substances needed for disease prevention and control [33]. Other supportive departments and organizations in various industries such as education, academic research, technology and industry will also assist in achieving project goals and deliverables within their own scope of responsibilities. However, it is three kinds of specialist institutions that have been playing pivotal roles throughout the process of infectious disease prevention and control, namely specialized public health institutions (such as Center for Disease Control and Prevention), medical institutions and primary healthcare institutions. Speci cally, technical and surveillance plans in response to disease outbreaks as well as for disease prevention and control will be proposed and managed by specialized public health institutions throughout both project development and implementation stages [34]. Medical institutions are responsible for providing treatment for patients infected by communicable diseases as well as reporting cases identi ed [33], while primary healthcare institutions serve as the gatekeepers for communicable disease prevention and control at community levels, which typically include community health service centers, township health centers, and village clinics in China where all kinds of projects and plans proposed at health administrative levels will be actually implemented [34]. Tasks expected to be accomplished at primary healthcare levels can be summarized as three aspects.
First, controlling sources of infection via monitoring, reporting, and treating patients infected by communicable diseases. Second, cutting off all transmission routes via multiple methods such as water, soil, air sampling and testing, monitoring accumulated bacterial concentration in water pipes, mosquitoes and ies control, sterilization, and disinfection. Third, protecting vulnerable populations via the provision of health education programs for residents in order to improve their knowledge about appropriate hand washing behaviors, safe water consumption tips, and appropriate use of hygiene facilities. Figure 1 summarizes the relationship between various institutions involved in the process of infectious diseases prevention and control.
At present, the lack of healthcare workers in primary healthcare institutions remains a critical problem in China. As the consequence, projects proposed for infectious disease prevention and control have been di cult to be implemented or to achieve expected goals. In Sichuan Province, there were only 22.47 primary healthcare workers per ten thousand population and only 2.34 health workers per primary healthcare facility in 2019 [35]. It was reported by governmental o cers from Sichuan Health Commission who had been in charge of communicable disease prevention and control that the lack of human resources as well as inadequate knowledge of communicable disease among primary healthcare workers have posed huge obstacles for primary healthcare providers in achieving desired quality and e ciency of healthcare delivery in spite of their full engagement in most of the infectious disease cases identi ed. Taking health education as an example, more than 90 percentage of the workers engaged in health education were found to be part-time or even multi-tasking [36][37][38], thus resulting in 45 percentage of primary healthcare facilities incapable of providing effective health education programs [38,39]. Among primary healthcare institutions where services have been actually provided for health education purposes, residents engaged in health education programs only accounted for 60% of the targeted populations [40,41].

Study Area
Our study was based on Sichuan Province, a southwestern province in China, covering 183 counties and 4,687 communities in the geographic regions of 97°21' to 108°33' east longitude and 26°03' to 34°19' north latitude. The land area and GDP per capita of Sichuan Province respectively ranked fth and nineteen among 31 provinces of Mainland China, with a population of 83.41 million reported in 2018 [21]. The socioeconomic and topographical characters vary signi cantly across the province, where eastern Sichuan is characterized by plains, dense population, and high-level economic development, while western Sichuan is in the opposite situation [42].

Data sources
Data retrieved from Sichuan Center for Disease Control and Prevention provided detailed information on both county-level and community-level yearly diarrhea cases between January 1, 2017 and December 31, 2019. According to National Health and Family Planning Commission [43], diarrhea is de ned as a group of infectious diseases caused by various pathogens including bacteria, viruses, and parasites, with diarrhea as the typical symptom. In our study, diarrhea cases induced by dysentery, cholera, paratyphoid, and typhoid were not included due to data inaccessibility. As a noti able infectious disease in China, all diarrhea cases should be reported via an online standardized form within 24 hours of diagnosis [44].
Data related to primary healthcare institutions from 2017 to 2019 were administrative data extracted from annual reports published by primary healthcare institutions and provided by Health Commission of Sichuan Province, which included the number of health workers and subsidy per staff. The subsidy per staff was adjusted for in ation rates, and was measured in 2019 RMB. We pooled these values at community-level according to the unique community code. We then used the unique community code to match the three kinds of sources of data. However, 366 communities were missed due to the different statistical gauge, accounting for 7.8% of total communities.

Spatial autocorrelation analysis method
To capture the spatial distribution of diarrhea morbidity and thus provide evidence-based implications for optimizing primary healthcare resource allocations, Moran's I and corresponding graphic tools are used to detect and visualize the global and local spatial autocorrelation of the annual diarrhea morbidity [45]. In this study, we employed the row standardized rst-order contiguity Rook neighbors as the criterion to identify neighbors. Given the rule, if regions i and j are adjacent, the spatial matrix w ij = 1, otherwise, w ij = 0. The global Moran's I, ranging from 1 to -1, was used to the overall spatial autocorrelation of all communities in Sichuan Province. Positive spatial autocorrelation values indicate regions with similar (low-low or high-high) diarrhea morbidity are clustered together, while negative values indicate the opposite (cluster of dissimilar morbidity, low-high and high-low) [46]. Monte Carlo randomization (9999 permutations) was used to assess the statistical signi cance of Moran's I, with the null hypothesis being that the distribution of diarrhea morbidity in Sichuan province is completely random distributed [47]. Subsequently, local indicator of spatial association (LISA, local Moran's I) and Moran scatterplots were used to detect the spatial autocorrelation of each communities in Sichuan Province and to identify the clusters' location [48]. Statistically signi cant spatial clusters (high-low, region with high morbidity surrounded by region with low morbidity; and vice versa for low-low, low-high, high-high) were visualized using univariate LISA cluster map with community boundaries.

Spatial panel analysis method
In the study, annual diarrhea morbidity and annual primary care facilities' data pooled at community-level were used to explore the relationship between the number of primary healthcare workers and the morbidity of diarrhea. Because diarrhea is a kind of infectious disease, the spatial heterogeneity and spatial dependence of the morbidity in different communities need to be captured in the analysis model [49,50]. A spatial panel data model is able to deal with the spatial heterogeneity and spatial dependence simultaneously, which is typically used to t data containing repeated observations in different spatial units [51][52][53]. Spatial panel data possesses signi cant advantages compared to traditional cross-sectional or time series models, i.e. containing more variation and less collinearity among variables [53,54], thus leading to the increase in estimation e ciency [53,55]. Given the results of model speci cation (shown in Table 2), the spatial lag xed effects panel data model was ultimately employed to explore the relationship between the number of primary healthcare workers and the morbidity of diarrhea at community-level. The formula was set as follows: (1) Where Y it denotes the diarrhea morbidity for community i at year t. The parameter ρ is the spatial autoregression coe cient, indicating the impact degree of spatial factors on the dependent variable. W ij represents the element of a (N × N) spatial weighting matrix. As previously mentioned, row standardized rstorder contiguity Rook neighbours was employed to de ne the spatial weighting matrix. H it indicates the number of primary care health workers, which was pooled at community-level. S it is the covariate representing the subsidy per staff. This variable is de ned as the subsidy from governmental nance, superior departments, and other sources, divided by the number of staff in each primary care facility. To a certain degree, these subsidies can be used independently, and are typically used to purchase equipment, carry out public health programs, and pay staffs for bonus et al. This variable could be tightly related to the performance of primary healthcare workers in the prevention and control of diarrhea, thus we included it in the model to better control for the potential confounding effect. The term u i is the community speci c effect and η t represents time speci c effect. ε it is the error term and is assumed to be i.i.d. N(0,σ 2 ) distribution.
All analyses were conducted using ArcGis 10.2 and R 3.6.3. P < 0.05 is used to determine statistical signi cance.

Results
Annually, the diarrhea morbidities in Sichuan Province were found to be 4.20, 4.36, and 4.62 per 10,000 people from 2017 to 2019. As shown in Fig. 2, the high morbidity clustered areas were mainly distributed in western Sichuan, while low morbidity clustered areas were mainly distributed in the middle and eastern regions in Sichuan where the economic development status and primary healthcare resource allocations were relatively superior than those in western Sichuan. It can be seen that this kind of trend had become more pronounced during the changing period of time, with the constant emergence of high diarrhea morbidity communities in western Sichuan. Diarrhea morbidities at the community level in Sichuan Province from 2017 to 2019 were descripted in Table 1, with the average morbidity of each year found to be 3.5, 3.8, and 4.4, and the maximum morbidity of each year found to be 53.0, 71.9, and 149.1, respectively. In addition to diarrhea morbidity, the number of primary healthcare workers and subsidy per staff also demonstrated signi cant variation across communities in Sichuan Province. Note: The unit of subsidy per staff is thousand Yuan; the unit of population is a hundred person.

Spatial autocorrelation analysis of diarrhea morbidity
As shown in Fig. 3, the values of global Moran's I between 2017 and 2019 at community-level were high, ranging from 0.38 to 0.42 with all P-values < 0.01. The results identify the positive global autocorrelation of diarrhea morbidity, namely the communities with high diarrhea morbidity tend to be adjacent to the communities with high morbidity, and vice versa. LISA analysis revealed four types of spatial clusters in terms communities. As shown in Fig. 4, the signi cantly high-high and low-low clusters were mainly concentrated in western and eastern Sichuan across years, which corresponds to the distribution of the regional economic development and the primary healthcare resources (eastern Sichuan possess better primary healthcare resources while western Sichuan is in the opposite situation). Besides, this trend tended to be pronounced as the time went on, with more high-high clusters appear in western Sichuan.

Spatial panel analysis
We conducted several model speci cation tests to obtain the appropriate model. Hausman test robust to spatial autocorrelation was rstly employed to determine whether a xed effect or a random effect model should be chosen. As a result, a signi cant xed effect of each community was found. Subsequently, based on a xed effect model, we conducted Lagrange multiplier tests to identify whether a spatial lag effect or a spatial error effect model should be selected. The results showed that the spatial lag effect was more signi cant than the spatial error effect. Therefore, we nally used the spatial lag xed effects panel data model to examine the associations between diarrhea morbidity and the number of primary healthcare workers.  Table 3 show that after controlling for subsidy per staff, the relationship between number of primary healthcare workers and the diarrhea morbidity was signi cantly negative. The coe cient of the number of primary care health workers was − 0.187, indicating the diarrhea morbidity (1/10,000) would decrease 0.187 when doubling the number of primary care health workers. Besides, the coe cient of subsidy per staff was − 0.456, representing the diarrhea morbidity (1/10,000) would decrease 0.456 when doubling the subsidy per staff. The spatial autocorrelation coe cient was 0.434 and statistically signi cant, meaning that a spatial spillover phenomenon existed. Table 3 Results of spatial lag xed effects panel data model for the diarrhea morbidity (1/10,000)

Robust tests
In order test the robustness of the analysis at community level using datasets with missed information from a couple of communities, we further employed the data pooled at the county level for analysis. Table 4 reports the results of the robust tests. The results at county level were found to be similar with the results at community level in terms of the signs, effect size, and the statistical signi cance. Table 4 Robust test results of spatial lag xed effects panel data model for the diarrhea incidence per 10,000 population at county level The results of the distribution and LISA signi cance and cluster map of diarrhea morbidity at county-level are encompassed in the Appendix, shown in Figure  S1 to S3. It is noteworthy that the diarrhea morbidity of communities within each county could have signi cant heterogeneity, i.e. a few communities have high morbidity. The community-level analysis conducted in our study was capable of dealing with the heterogeneity issue which couldn't be solved by countylevel analysis conducted in previous studies.

Discussions And Conclusions
In this study, the number of healthcare workers was adopted as an indicator re ective of healthcare resource allocations among different primary healthcare institutions, which serves as an essential determinant of healthcare institutions' performances in providing health services at the primary healthcare level.
Moran's I and its corresponding graphic tools were used to detect as well as visualize the global and local spatial autocorrelation of annual diarrhea morbidities among different communities, based on which areas where increased healthcare resources should be allocated were identi ed. The spatial lag xed effects panel data model was then employed to explore the relationship between the number of primary healthcare workers and the morbidity of diarrhea at the community level. The Moran's I analysis revealed the positive global autocorrelation of diarrhea morbidity and identi ed high-high clusters to be mainly concentrated in regions with relative fewer amounts of primary healthcare workers. As indicated by the regression outcomes, a negative relationship was identi ed between the number of primary healthcare workers and the morbidity of diarrhea, with a 0.187 reduction of diarrhea morbidity (1/10,000) associated with doubled amounts of primary healthcare workers.
Similar with previous studies [16,[56][57][58], a positive correlation in the values of diarrhea morbidities was found between different regions, indicating that the distribution of diarrhea cases across Sichuan Province demonstrated obvious spatial clusters instead of being random. Through LISA analysis, signi cantly high-high and low-low clusters were found to be mainly distributed in western and eastern Sichuan during the studied period, which demonstrated consistency with regional economic development status as well as primary healthcare resource allocations among different regions. It should be noted that these ndings might have been induced by various factors such as regional economic development status, the penetration of WASH facilities, meteorological factors as well as primary healthcare resource allocations among different regions, thus should not be considered as evidences potent enough for indicating the direct association between the number of primary healthcare workers and diarrhea morbidity at community levels. However, based on our ndings as shown in Fig. 4, with an increased amount of high-high clusters emerging during the studied period, the prevention and control of diarrhea cases in western Sichuan remains a critical issue that should be constantly addressed at both governmental and health administrative levels.
The Spatial autocorrelation analysis suggested that a spatial statistical model instead of the classical linear regression should be adopted as a more appropriate method for exploring the relationship between the number of primary healthcare workers and diarrhea morbidities at the ecological level [53]. As the result, a spatial lag xed effects panel data model was adopted in this study which was capable of taking both spatial autocorrelation and individual effects into consideration thus leading to higher statistical e ciency [52,53]. Speci cally, biased outcomes produced by unmeasured potential confounders would be reduced by the adoption of spatial individual effects while the spatial correlation would be taken into account via the incorporation of a spatial weight matrix [55].
Based on the regression results, a negative relationship was found between the number of primary healthcare workers and diarrhea morbidity at community levels, with a 0.187 reduction of diarrhea morbidity (1/10,000) associated with doubled amounts of primary healthcare workers. Our ndings highlighted the pivotal role of primary healthcare resource allocation in the process of nationwide infectious diseases prevention and control in China. As previously mentioned in the Background section, the lack of primary healthcare workers in China has posed huge obstacles for primary healthcare institutions in achieving desired quality and e ciency of health service delivery in the process of infectious disease prevention and control. As the result, it is not di cult to predict that an increased amount of healthcare providers at primary healthcare level would signi cantly contribute to the reduction of nationwide diarrhea morbidity via substantially improving the quality and e ciency of primary healthcare delivery. On the one hand, an increased amount of primary healthcare workers would made it more feasible for primary healthcare institutions to improve the whole team's productivity and e ciency via assigning different tasks to particular staffs such as mosquitoes and ies control, sterilization, and disinfection. All kinds of tasks related to infectious disease prevention and control are more likely to be accomplished in a more e cient manner under more clearly de ned jobs and responsibilities for different primary healthcare workers. On the other hand, it is not di cult to imagine that staffs assigned to speci c tasks in a long-term are more likely to accomplish those tasks with improved quality and e ciency as they have obtained better knowledge through rich work experiences in those particular elds than general primary healthcare workers.
Besides, from a general perspective, an increased amount of primary healthcare workers would substantially reduce the workload imposed on each individual staff thus improving the productivity and e ciency of the whole team. As previously presumed, a negative relationship was identi ed between subsidy per staff and diarrhea morbidity, for which a reasonable explanation is that reduced subsidy is signi cantly associated with decreased motivation and productivity thus leading to higher diarrhea morbidity as the result of poor disease prevention and control.
Through respectively comparing Fig. 2 with Figure S1, Fig. 3 with Figure S2 and Fig. 4 with Figure S3, it can be concluded that signi cant heterogeneity might be embedded in community-level diarrhea morbidities within each county. While diarrhea morbidity was found to be low in most of the communities, a couple of communities demonstrated high diarrhea morbidities. It is beyond controversy that the county-level analysis conducted in previous studies were not capable of identifying this issue thus leading to biased outcomes. However, it should be noted that biased outcomes would also be induced by communitylevel analysis due to increased amount of missing data if those data were not missing completely at random. In attempt to nd an optimum solution for dealing with both heterogeneity and missing data issues, results produced at community levels were considered as the referential outcomes while county-level outcomes were used for robust analysis. Similar results were produced by these two kinds of analysis in terms of the signs, effect size, and the statistical signi cance, which indicated the rationality of our study.
In summary, our study provided empirical evidences on the relationship between primary healthcare resource allocation and diarrhea morbidity at community levels which was re ected by the number of primary healthcare workers among different regions. The study also identi ed areas where increased primary healthcare resources should be predisposed towards. Such ndings were expected to provide evidence-based implications for policy-makers in the formulation of region-speci c policies and strategies aimed at improving infectious diseases prevention and control in China.
Several limitations should be noted in this study. First, the study established a correlation rather than causality. As the result, biased regression outcomes might have been induced by endogenous problems potentially embedded in the established associations. For example, a reversed causality might exist between diarrhea morbidity and the number of primary healthcare workers. Speci cally, an increased amount of primary healthcare workers might be allocated in areas with higher incidences of diarrhea after disease outbreaks by governments, thus leading to underestimated impacts of primary healthcare workforce on diarrhea morbidity. Second, as the 2020 Sichuan Statistical Yearbook has not been published yet, variables re ective of regional sociodemographic factors could not be added into the regression model for analysis such as GDP per capita. Although these factors would typically change very slowly across years and xed effect has been handled in the regression model, the outcomes could still have been biased to some extent.

Declarations
The data that support the ndings of this study are available from the Center for Disease Control and Prevention and Health Commission of Sichuan Province, but restrictions apply to the availability of these data. Data are can be made available from the authors upon reasonable request and with permission of the Center for Disease Control and Prevention and Health Commission of Sichuan Province. Figure 2 The geographical distribution of diarrhea morbidity at community-level from 2017 to 2019 Note: The designations employed and the presentation of the material on this map do not imply the expression of any opinion whatsoever on the part of Research Square concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. This map has been provided by the authors.