The distribution structure of medical and care resources based on regional characteristics throughout Japan in 2020

This section provides a summary of recent research exploring the allocation of medical and care resources, which is the focal point of the present this study. First, we review previous investigations on a global scale. Second, we provide a detailed review of research conducted in Japan. In particular, we investigate studies that explore the distribution of medical and care resources individually, as well as their allocation in combination.

Numerous studies investigating the allocation of medical resources have been carried out on a global scale. International comparative studies have indicated that Japan boasts an equitable dispersion of nurses, and its healthcare system garners high levels of public satisfaction [9, 10]. By contrast, a comparison with the USA reveals that physicians in the USA are inclined to gravitate toward areas with higher income, unlike in Japan, where no such pattern is seen; these distinguishing features are attributed to the existence of universal public health insurance in Japan [11, 12]. In addition, China exhibits regional variability in nursing homes and disparities in the diversity of facilities around the aged [13, 14]. Nevertheless, these studies are confined to analyzing medical and care resources in isolation, and thus, the relationship between the two resources remains poorly understood.

The following two Japanese studies have explored the relationship between regional characteristics and the distribution of medical resources in recent years. The first study, conducted by Hara et al. [15], analyzed changes in physician distribution trends across SMAs nationwide from 2000 to 2014. The results indicated a decrease in physician supply in all regions except urban areas. The second study, by Seo et al. [16], analyzed regional differences in the performance of medical delivery systems and factors impacting medical costs in municipalities in Tokyo (urban areas) and Chiba Prefecture (suburban areas) in 2018. The results revealed that while the medical delivery system was well-developed in urban areas, it remained inadequate in suburban areas. These studies suggested that medical delivery systems differed due to regional characteristics. The following two Japanese studies examined the estimated number of hospital beds in the regional medical visions. The first study, conducted by Takizawa et al. [17], surveyed medical social workers working at medical institutions with convalescent-phase hospital beds in Aomori Prefecture in 2019. The results indicated that increasing the number of convalescent-phase hospital beds up to the number estimated based on the regional medical vision was challenging because of the underdeveloped treatment environment in community and home care. The second study, by Miyazawa [18], compared the estimated number of hospital beds based on regional medical visions with the real number in 2005, targeting SMAs nationwide. The results suggested that the regional imbalance in medical supply cannot be resolved. These studies indicated the discrepancy between regional medical visions and the actual situation. In summary, the four studies introduced so far indicate that the uneven distribution of medical resources is related to regional characteristics. Moreover, these studies suggest that it is difficult to change the distribution of medical resources in some regions, even with interventional policies such as regional medical visions, and that even if such changes are made, medical resources are still unevenly distributed in some regions.

The following three Japanese studies have explored the distribution of care resources and the current state of care delivery systems in recent years. The first study, conducted by Ikeda et al. [19], analyzed the relationship between the number of deaths at home and the distribution of medical and care resources in municipalities nationwide in 2014 and 2017. The second study, by Jin et al. [20], analyzed the factors contributing to regional differences in the cost of care, targeting long-term care insurance users and municipalities nationwide in 2016. The results showed that there were considerable regional differences in care costs, and that municipalities in urban areas with more “welfare facilities for older adults requiring long-term care” (one type of nursing home that provides end-of-life care) per population were strongly associated with higher care costs. The third study, by Nishino et al. [21], estimated the number of users of long-term care insurance facilities nationwide in 2010, comparing it with the MHLW target in 2025. That study also examined the applicability of the MHLW target for the provision of long-term care insurance facilities in 2025 for a specific city in Ishikawa Prefecture. The results suggested that it is difficult to transition from nursing home care to home care when the older adults receiving care are living in single-person households. In summary, these three studies suggest the presence of regional differences in the distribution of care resources and care delivery systems. Moreover, these studies suggest that the policy of providing home medical care and home care through “community-based integrated care systems” will be promoted nationwide, but the distribution of care resources and the care provision system differ from region to region, suggesting that the promotion is difficult depending on regional characteristics.

The following two Japanese studies addressed variables related to regional characteristics and medical and care resources in an integrated manner. The first study, conducted by Ishikawa et al. [22], identified the characteristics of SMAs nationwide from 2008 to 2011, utilizing principal component analysis (PCA) with variables pertaining to regional characteristics and medical and care resources. The findings revealed that medical care adequacy was deficient in urban and remote areas. The second study, conducted by Kusunoki et al. [23], analyzed the interrelation between regional characteristics and medical and care resources using factor analysis and structural equation modeling (SEM) for SMAs nationwide in 2015. The results indicated that suburban areas with a higher concentration of older individuals tended to have better medical care adequacy, whereas urban areas tended to have higher acute care adequacy via universities with medical schools. Although these two studies provide some perspective on the structure of the relationship between regional characteristics and medical and care resources, both methods used integrated information on regional characteristics and the allocation of care resources. Consequently, the difference in the structure of the relationship between medical and care resources due to differences in regional characteristics remains unclear.

The summary of the studies discussed in the preceding section indicates that the associated structure of the distribution of medical and care resources worldwide has not been thoroughly elucidated. Even in Japan, where collaboration between medical and care services is encouraged, many studies separate medical and long-term care resources or investigate limited target regions. In addition, in studies using multivariate analyses such as PCA or SEM, the difference in the structure of the relationship between medical and care resources due to differences in regional characteristics remains unclear. This is due to the limitations of the analysis methods, which have not been able to clarify the complex structure of relationships among variables. Therefore, the present study attempts to group all the SMAs of Japan using regional characteristics, and to compare the associated structure of the distribution of medical and care resources in each group so as to clarify the complex associated structure of variables. This study also aims to fill the existing research gap in which the difference in the structure of the relationship between medical and care resources due to differences in regional characteristics is unclear, thereby contributing to the future coordination of medical and care services.

This study was divided into three phases. Initially, in the first phase, PCA was applied to group all SMAs throughout Japan according to regional characteristics. Subsequently, we compared the associated structure of the distribution of medical and care resources for each group obtained in the first phase, the comparison consisted of the second and third phases as follows. In the second phase, SEM was conducted to build the associated structure of the distribution of medical and care resources in all of Japan. This is because a comparison of groups requires a model for the integrated groups. Finally, in the third phase, multigroup analysis was conducted to compare the models, where the model obtained in the second phase was applied to the groups obtained in the first phase.

By dividing the methodology into these three phases, the first phase deals with only variables of regional characteristics, while the second phase deals with only variables of the distribution of medical and care resources. The third phase integrates these results. Therefore, this study attempts to resolve the limitation in previous studies, i.e., the inability to clarify the complex associated structure among variables caused by its analysis methods that treated variables in an integrated manner. Thus, we expect our findings to fill the above-mentioned research gap.

In the following sections, we first define the target regions to be analyzed, and then provide a detailed description of the methodology of these three phases.

Japan’s medical administrative areas consist of, primary medical areas, tertiary medical areas, and SMAs. Primary medical areas are set up in units of municipalities and provide short-term hospitalization and outpatient visits. Tertiary medical areas are generally established on a prefectural basis and provide advanced acute-phase and specialized medical care. SMAs, which are located between these areas, consist of several municipalities, and are defined by each prefecture based on the ability to provide medical care. Additionally, the MHLW requires welfare areas for older adults (areas determining the expected volume of long-term care insurance service provision) to coincide with SMAs [24]. Whereas previous studies have primarily focused on analyzing the distribution of care resources at the municipal level, the target regions in this study are 334 nationwide SMAs, considering the distribution of medical and care resources simultaneously. The summary statistics of the populations and areas of the SMAs are presented in Table 1.

The regional characteristics used in the first phase consisted of 28 variables selected in reference to the study by Miyake et al. [26], which classified SMAs based on various regional characteristics. These variables are presented in Table 2 [25, 27‐31] and the summary statistics are presented in Table 3. In addition, the reasons for the selection of each variable are presented in Table 4. The variables are classified into the following four categories: population, which indicates the age composition and population size; household category which indicates the composition of households; industry and work, which indicates the industrial composition, mode of transportation, and economic scale, and land utilization, which indicates the land use status and spatial characteristics. As most of these variables are available for the year 2020, it was used as the base year for the present study.

The method of grouping SMAs by PCA was as follows. Initially, a PCA was performed to summarize the variables in Table 2. The number of PCs was determined from a parallel analysis, and a promax rotation was used to facilitate interpretation of the solution. Finally, the PC scores for all 334 SMAs were sorted in descending order and the SMAs were divided into top- and bottom-half groups. In the present study, a preferable method of analysis may have been to perform a detailed grouping based on a cluster analysis utilizing PC scores, as done in the study by Miyake et al. [26]. However, due to the smaller sample size in this method, multigroup analysis and SEM with such a detailed grouping often yields errors such as negative estimates of error variance. Therefore, grouping into two was assumed to be the maximum limit. Thus, the sample size for each group was 167. These PCA and parallel analyses were conducted using the “fa.parallel” and “principal” functions from version 2.2.9 of the “psych” package in R version 4.2.1 [32, 33]. The maximum likelihood method was used to estimate the PC loadings.

In total, there were 15 variables of the distribution of medical and care resources used in the second phase. These variables can be aggregated by SMA, with reference to the study by Kusunoki et al. [23], and are presented in Table 5 [34‐36]. The summary statistics are presented in Table 6. The variables are classified into the following five categories: hospital re-sources, clinic resources, care resources, home medical care and home care resources, and other medical resources. Among these valuables, in Table 5, “Hospital beds for long-term hospitalization (per 10,000 people),” “Capacities of nursing homes (per 10,000 people),” “Workers in nursing homes (per 10,000 people),” and “Workers in home medical care and home care (per 10,000 people)” are aggregated resources that have similar characteristics regarding medical and care provision. An unaggregated data analysis by SEM failed to yield goodness-of-fit indexes within the acceptable range because of the large number of observed variables and multicollinearity. Therefore, the aggregation of the aforementioned variables is the maximum limit. Population data, which were obtained from the “Population Census” data for 2020 [25], were used to calculate the variables presented in Table 5.

The method for building the model of the associated structure of the distribution of medical and care resources in the 334 SMAs nationwide by SEM was as follows. Initially, to formulate a hypothesis about this associated structure, an exploratory factor analysis (EFA) using the variables in Table 5 was performed. The number of factors was determined from a parallel analysis, and promax rotation was used. Subsequently, based on the results, we identified latent factors behind the variables in Table 5 from the obtained factors and formulated hypotheses regarding the relationships among them. Finally, SEM was used to test this hypothesis and to model the associated structure. The factor analysis and parallel analyses were conducted using the “fa.parallel” and “fa” functions from version 2.2.9 of the “psych” package in R version 4.2.1 [32, 33]. The maximum likelihood method was used to estimate the factor loadings. The SEM and multigroup analysis were conducted using the “sem” function from version 0.6.12 of the “lavaan” package in R version 4.2.1 [33, 37]. To estimate the population for these analyses, a diagonal weighted least squares method was used. This method employs a highly robust diagonal matrix to ensure accurate estimation results for non-multivariate normal data, rendering them suitable for utilizing the variables in Table 5 for medical and care resources. While the SEM and multigroup analysis were also conducted using the maximum likelihood method, which used PC and factor analyses, the values of the goodness-of-fit indexes utilized in this study were outside the tolerance limits. For the SEM and multigroup analysis described below, the goodness-of-fit indexes adopted in this study were the root mean square error of approximation (RMSEA), the comparative fit index (CFI), and the Tucker–Lewis index (TFI), the standardized root mean square residual (SRMR), and the p-value(χ²), which can be given by the “sem “ function. The RMSEA shows a high goodness-of-fit when the value is less than 0.05, and a low fit when the value is above 0.10. The CFI and TLI show a high goodness-of-fit when the values are more than 0.95. The p-value (χ²) indicates the probability of the test of the null hypothesis by the χ² value that the model constructed fits the sample in SEM; thus, a higher probability indicates a higher goodness-of-fit. However, because the null hypothesis is more likely to be rejected as the sample size increases in SEM, the value of χ² divided by the degrees of freedom (df), introduced by Schermelleh-Engel et al. [38], was used in this study. This value, “χ²/df “ shows a high goodness-of-fit when the value is less than 2.

A multigroup analysis was conducted using the obtained structural model with the SMAs grouped by regional characteristics as the sample. However, when a sample size is halved, as in this case, a solution may not be obtained owing to the deterioration of the model’s goodness-of-fit index in the multigroup analysis. In such cases, a multigroup analysis is conducted after adding constraints to fix the error variance of specific variables to the obtained structural model. It is also expected that, even with these constraints, a solution to the multigroup analysis may not be obtained. In these cases, SEM is performed for each group, with the same arrangement of variables, but with constraints to fix the error variance of particular variables.

Initially, the parallel analysis suggested that four PCs were suitable, as shown in Fig. 1. Subsequently, we interpreted the outcomes of the PCA in the following order of the contribution ratio, as shown in Table 7.

Because promax rotation was used in the PCA, correlations were found between the obtained components. A slight negative correlation was seen between “urbanicity” and “middle-density regionality.” This could be because of the substantially higher population concentrations in urban areas and the middle concentrations in middle-density regions. A slight positive correlation was found between “middle-density regionality” and “childcare regionality.” This could be because middle-density regions are more suitable than urban areas for rearing children and have a higher proportion of households with children. A slight negative correlation was found between “workplace regionality” and “middle-density regionality,” and a slight positive correlation between “urbanicity” and “workplace regionality.” This could be due to commercial areas and business districts being biased toward locations with large population concentrations.

Based on the PC scores, the SMAs nationwide were categorized into different groups. The results are presented in Figs. 2, 3, 4, 5, and the characteristics of each grouping are summarized below.

Initially, the results of the parallel analysis suggested that five factors were suitable, as shown in Fig. 6. Subsequently, we interpreted the outcomes of the EFA in the following order of the contribution ratio, as shown in Table 8.

Table 8 shows the correlation coefficients of 0.5 or higher among some of the factors. The factor “Clinic care abundance (without beds)” was positively correlated with the factor “Acute care abundance,” both of which affect the provision of medical ser-vices related to short-term hospitalization and outpatient visits. “Nursing care abundance” was positively correlated with both “Clinic care abundance (with beds)” and “Long-term hospitalization care abundance,” which affect the provision of medical and care services related to long-term hospitalization and nursing home care.

Based on the results discussed above, we formulated hypotheses regarding the associated structure of the distribution of medical and care resources in this study. Our first hypothesis concerns the latent factors that cause the distribution of medical or care resources. Local demand determines the distribution of both resources, but with different constraints on the supply of each. The distribution of hospital and clinic beds, classified as medical resources, is subject to strong restrictions because of the standard number of beds being determined based on the population by age group and bed utilization in a specific SMA to prevent excessive resource supply. Additionally, the MHLW induces prefectures to enhance home medical care, home care, and care in nursing homes when they formulate regional medical visions [39]. While nursing homes have personnel standards and other requirements, which are classified as care resources, they have fewer restrictions on their expansion than on medical resources. Therefore, the degree of enhancement of care resources is influenced by not only the degree of enhancement of medical resources, but also regional characteristics such as land prices and the number of workers needed to operate nursing homes. By contrast, the degree of medical resources is more directly influenced by the degree of care resources, which affects bed utilization, than by these regional characteristics, as the standard number of hospital beds is constrained. Based on these inferences, and in light of the fact that the SEM in this study did not deal with regional characteristics, we constructed a model that assumes “Nursing care abundance” as the cause.

Our second hypothesis concerns the causal relationship among the latent factors related to medical resources. The results of the EFA indicated a correlation between the two latent factors related to short-term hospitalization as well as outpatient visits, and the two latent factors related to long-term hospitalization, respectively. Therefore, we investigated whether the causal latent factor is related to short-term hospitalization and outpatient visits, or to long-term hospitalization. To formulate this hypothesis, we referred to Innami [40], who discussed social hospitalization. Innami considered that there is a trend for older adults to be admitted to a hospital with long-term care beds as a substitute for fully occupied nursing homes, and to be accepted to a hospital with general beds as a substitute for fully occupied long-term care beds. From this perspective, we hypothesized that “Long-term hospitalization care abundance” and “Clinic care abundance (with beds)” related to long-term hospitalization affect “Acute care abundance” and “Clinic care abundance (without beds)” related to short-term hospitalization and outpatient visits.

The third hypothesis is related to the relationship between latent and observed variables, where each latent variable is defined based on the findings of the EFA. To facilitate the interpretation of the SEM solution, we assumed that only one of the latent variables was responsible for the observed variables, namely, the number of hospital beds, number of clinic beds, and nursing home capacity. However, in the case of the observed variables related to medical professionals, such as physicians, nurses, and assistant nurses, multiple latent variables were assumed to be involved, even when the factor loadings on the factors considered as latent variables were low. In addition, we assumed that “Nursing care abundance” and “Clinic care abundance (without beds)” were behind the observed variable “Workers in home medical care and home care (per 10,000 people)” based on the results of the EFA. However, the observed variable “Universities with a faculty of medicine” was not expected to be affected by the latent variables assumed in this study. Therefore, we assumed that this variable exerted an influence on “Clinic care abundance (without beds)” and “Acute care abundance” based on the findings of the EFA. Finally, Fig. 7 depicts the resulting SEM used to test the above hypotheses. Based on the goodness-of-fit indexes, the model showed an acceptable level of fit. Note that in addition to the hypothesis of this study, by employing SEM, the assumption that latent factors related to medical resources are the cause could be tested. Therefore, analyses were conducted assuming that both “Acute care abundance” and “Long-term hospitalization care abundance” were causes. However, these analyses generated cases that were not acceptable from the goodness-of-fit indexes used in this study and cases in which the solutions did not converge.

The results of the multigroup analysis for each of the four groupings shown in the "Division of SMAs" are presented below.

First, a multigroup analysis was performed with two groups divided by “urbanicity.” However, in the analysis, the error variances of the observed variables “Hospital nurses (per 10,000 people)” and “Workers in nursing homes (per 10,000 people)” were negative. Therefore, we added a constraint to fix these error variances at 0. However, when this constraint was applied, no solution was obtained because the covariance matrix of the residuals of the observed variables was not positively definite. As a result, SEM was performed for each of the two groups with a common arrangement of variables, but with constraints to fix the error variances of the variables in question in each group; the results are shown in Fig. 8. Based on the goodness-of-fit indexes shown in this figure, the model showed an acceptable level of fit. Strictly speaking, the two models obtained had different structures and should be interpreted with this in mind.

The standardized partial regression coefficients for the same path differed by more than 0.3 between the two models at the following locations.

Second, a multigroup analysis was performed with two groups categorized by “middle-density regionality.” However, in this analysis, the error variances of the observed variables “Hospital nurses (per 10,000 people),” “Workers in nursing homes (per 10,000 people),” “Clinic physicians (per 10,000 people),” and “Clinic assistant nurses (per 10,000 people)” were found to be negative. Thus, constraints were added to fix the error variances of these variables to 0. The results are displayed in Fig. 9. Based on the goodness-of-fit indexes shown in this figure, the model showed an acceptable level of fit.

The standardized partial regression coefficients for the same path differed by more than 0.3 between the two models at the following locations.

Third, a multigroup analysis was performed with two groups divided by “workplace regionality.” However, in the analysis, the error variances of the observed variables “Workers in nursing homes (per 10,000 people)” was negative. Therefore, we added constraints to fix these error variances at 0. However, this resulted in no solution because of an optimizer error. As a result, SEM was performed for each of the two groups with a common arrangement of variables, but with constraints to fix the error variances of the variables in question in each group. The results are shown in Fig. 10. Based on the goodness-of-fit indexes shown in this figure, the model showed an acceptable level of fit. Strictly speaking, the two models obtained had different structures, and should be interpreted with this in mind.

The standardized partial regression coefficients for the same path differed by more than 0.3 between the two models at the following locations.

Fourth, a multigroup analysis was performed with two groups based on “childcare regionality.” However, in this analysis, the error variances of the observed variables “Hospital nurses (per 10,000 people)” and “Clinic physicians (per 10,000 people)” were found to be negative. Thus, constraints were added to fix the error variances of these variables to 0. The results are displayed in Fig. 11. Based on the goodness-of-fit indexes shown in this figure, the model showed an acceptable level of fit.

The standardized partial regression coefficients for the same path differed by more than 0.3 between the two models at the following locations.

The distribution structure of medical and care resources in SMAs nationwide is shown in Fig. 7. Our analysis suggests that “Nursing care abundance” has a slightly negative effect on “Acute care abundance” and “Clinic care abundance (without beds).” A previous study by Suzuki et al. [41] reported a weak negative correlation between medical and care expenditures. Although their study was limited to a few regions and used data from 2003 to 2007, our results confirm the validity of the model used. Our findings are also consistent with the MHLW’s policy to expand the use of nursing homes to reduce the number of general hospital beds. However, the results also indicate that the enhancement of nursing homes has only a slight effect on the reduction of general hospital beds and reduces the number of clinics without beds, which promotes home medical care services. Thus, our findings suggest that the MHLW’s goal of reducing hospital beds is difficult to achieve through nursing home enhancements alone.

Moreover, our results indicate that “Nursing care abundance” tends to increase “Long-term hospitalization care abundance” and “Clinic care abundance (with beds),” which implies that nursing homes alone are inadequate to meet the high demand for medical and care services, and therefore, must be supplemented with long-term hospitalization beds in hospitals and clinics. Additionally, “Long-term hospitalization care abundance” tends to increase “Acute care abundance,” and “Clinic care abundance (with beds)” tends to increase “Clinic care abundance (without beds).” Hence, our results support the consideration of Innami [40], previously mentioned in the section “Hypotheses based on EFA results and validation by SEM.”

“Long-term hospitalization care abundance” and “Clinic care abundance (with beds)” significantly increase only the number of assistant nurses. This suggests that associate nurses play a larger role in long-term hospitalization and clinic beds.

Moreover, “Universities with a faculty of medicine” tend to increase “Acute care abundance” and “Clinic care abundance (without beds).” A study by Kusunoki et al. [23] reported that the number of universities with a faculty of medicine increases the adequacy of acute care medical resources. Although their study used data on the distribution of medical resources in 2015, our results support the validity of the model used. Furthermore, our findings suggest that the location of universities with a faculty of medicine leads to increased fulfillment related to short-term hospitalization and outpatient visits in the surrounding areas.

As previously stated, the subdivision of SMAs was based on PC scores for regional characteristics such as “urbanicity,” “middle-density regionality,” “workplace regionality,” and “childcare regionality,” which were determined through a PCA. “Workplace regionality” showed a slight correlation with “urbanicity,” while “childcare regionality” showed a slight correlation with “middle-density regionality.” Consequently, the differences in path coefficients for the distribution structure of medical and care resources across groups divided by the PCs with slight correlations were similar. Moreover, “workplace regionality” and “childcare regionality” showed low contribution rates. Therefore, the following discussion focuses on differences in the distribution structure of medical and care resources among each group divided by “urbanicity” and “middle-density regionality.”

Figure 8 shows insights into differences in the distribution structure of medical and care resources between the urbanicity and non-urbanicity groups. In the urbanicity group, “Nursing care abundance” tended to increase “Long-term hospitalization care abundance” and “Clinic care abundance (with beds).” This observation suggests that in urban areas, the demand for nursing homes is met by the supply of long-term hospitalization and clinic beds. The reason for this could be the unavailability of land required for expanding nursing homes in urban areas due to the limited land size and high land price.

Furthermore, in the urbanicity group, “Nursing care abundance” tended to reduce “Clinic care abundance (without beds),” whereas in the non-urbanicity group, it tended to increase it. This finding suggests that in the urbanicity group, the supply of nursing homes and clinics without beds is in a trade-off relationship with the demand for medical and care services among older adults. Conversely, it suggests that in the non-urbanicity group, the demand for medical and care services for older adults is met by clinics without beds instead of nursing homes. In Japan, primary care is often provided by the nearest medical institution from the patient’s home. The non-urbanicity group does not have an incentive to locate a large medical institution such as a hospital because of its small population size, so clinics that are relatively easy to establish play the primary care role. As a result, in the non-urbanicity group, primary care by clinics without beds and home care can meet the demand for nursing care. In addition, the non-urbanicity group may make demands for medical and care flow out to neighboring SMAs because their vast distance from large cities creates an area unsupplied with medical and care resources. However, the present study could not test this possibility because we did not use data on patient transfers between SMAs. For this reason, it is desirable to include data such as the proportion of outflow and inflow of patients per SMA published by the MHLW.

“Universities with a faculty of medicine” in the urbanicity group tended to increase “Clinic care abundance (without beds).” This trend may be attributed to the tendency for universities with a faculty of medicine to be located in urban areas, as indicated by the study by Kusunoki et al. [23].

Additionally, in the urbanicity group, “Long-term hospitalization care abundance” tended to decrease “Hospital physicians (per 10,000 people),” whereas “Acute care abundance” tended to increase “Hospital physicians (per 10,000 people).” Thus, in the urbanicity group, hospital physicians tended to be more available in acute care sections. This could be because physicians are concentrated in acute care owing to the existence of more universities with a faculty of medicine in the urbanicity group, where the supply of advanced acute care is greater compared with the non-urbanicity group.

By contrast, in the non-urbanicity group, “Clinic care abundance (without beds)” was less likely to increase “Clinic nurses (per 10,000 people),” but showed a strong tendency to increase “Clinic assistant nurses (per 10,000 people).” The path coefficients suggest that in the non-urbanicity group, the tendency of “Clinic care abundance (without beds)” to increase “Clinic nurses (per 10,000 people)” and “Clinic assistant nurses (per 10,000 people)” was similar, which suggests that associate nurses play a larger role in clinics without beds in non-urbanicity groups. Associate nurses have more limited duties than do nurses because they provide nursing care under the direction of physicians and nurses. Therefore, the quality of nursing care may be lower in the non-urbanicity group of clinics without beds than in the urbanicity group. However, the present study could not test this possibility because the discussion of the quality of medicine used only data such as the number of medical workers per population in each SMA. For this reason, it is desirable to include data such as the number of medical accidents as indicators of the quality of medicine in future work.

Figure 9 provides insights into the distribution structure of medical and care resources between the middle and non-middle-density regionality groups. In the middle-density regionality group, “Nursing care abundance” appeared to increase “Acute care abundance” slightly. This suggests that the MHLW’s plan to reduce the number of general hospital beds by enhancing nursing homes may not produce favorable results in middle-density regionality areas. Here, we focus on two reports concerning the distribution of occupied nursing homes. The first report, by Oguro et al. [42], focused on sub-urban rural areas in Niigata city within an SMA in the middle-density regionality group. That study reported a slight excess of nursing homes compared with demand. The second report, by Asano et al. [43], concerns Toyohashi city within an SMA in the middle-density regionality group. That study reported an increasing trend of nursing homes in urbanization control areas as defined by the Japanese City Planning Act, where development is strictly restricted. Although those studies were based on historical data and specific areas smaller than SMAs, the results suggested that similar distribution patterns of nursing homes exist in municipalities within SMAs with high middle-density regionality areas nationwide. From the urban planning perspective, middle-density regionality areas are of concern with regard to urban sprawl. Thus, the expansion of nursing homes in these areas should be closely monitored not only for the reduction of hospital beds and the efficient provision of care, but also from the perspective of preventing overdevelopment.

In the non-middle-density regionality group, “Nursing care abundance” tended to reduce “Acute care abundance” and “Clinic care abundance (without beds).” Furthermore, “Nursing care abundance” tended to increase “Long-term hospitalization care abundance.” Thus, in non-middle-density regionality areas, compared with middle-density regionality areas, the demand for care is met by the supply of long-term hospitalization and clinic beds instead of nursing homes. According to the results of the PCA, non-middle-density regionality areas are typical urban or remote areas. In typically urban areas, this result is consistent with the associated structure of medical and care resources in the urbanicity group. A study by Nakazono et al. [44, 45] in remote areas reported fewer users of nursing homes due to a decline in the older population. A related study conducted by Tokito et al. [46] estimated that the amount of care service demand in remote areas tends to decrease with future changes in the older population. Although those studies were based on specific areas smaller than SMAs and used historical data, their results suggest a nationwide decline in demand for nursing homes in remote areas, which can lead to a disproportionate care supply. Consequently, supplemental medical and care supply with long-term hospitalization and clinic beds may meet the uneven demand for care in different locations.

In the non-middle-density regionality group, we observed strong tendencies for “Long-term hospitalization care abundance” to increase “Acute care abundance” and for “Nursing care abundance” to increase “Long-term hospitalization care abundance.” This result may be attributable to the consideration of Innami [40], previously mentioned in the section “Hypotheses based on EFA results and validation by SEM.” In addition, in the non-middle-density regionality group, “Long-term hospitalization care abundance” tended to reduce “Hospital physicians (per 10,000 people)” and “Hospital nurses (per 10,000 people).” This suggests that the quality of medicine related to long-term hospitalization is lower in the non-middle-density regionality group. In a typical urban area among the non-middle-density regions, this result is consistent with the associated structure of medical and care resources in the urbanicity group.

This study has some limitations in addition to those previously mentioned. Primarily, because this study was conducted in a cross-sectional setting, causal relationships could not be identified. A time series analysis, for example, would be desirable to resolve this issue. Second, this study was limited to discussing regional differences in medical and care resources solely based on supply-side factors using data on medical and care resources. However, studies by Ibuka et al. [47] and Jin et al. [48] reported that regional differences in medical and care resources are also dependent on demand-side factors. Therefore, it is desirable to include data generated by demand-side factors, such as medical and care costs based on receipt data. Based on the perspective of the related structure of medical and care resources according to regional characteristics obtained in the present study, these findings are expected to provide insights into factors that contribute to unnecessary medical and care provision, such as the physician-induced demand hypothesis [49] discussed in health economics.