Each municipality’s volume of clinics was calculated by dividing the number of home care clinics in the municipality by the population of individuals aged ≥75 years in that municipality (data obtained from the 2010 Japan National Population Survey [6] ).

Clinics’ geographical accessibility was quantified by calculating the proportion of elderly people living near home care clinics in each municipality. Accessibility could range between 0 (no elderly people in the municipality lived near a clinic) to 1 (all elderly people in the municipality lived near a clinic). We divided each municipality into 0.5-square-kilometre areas (termed “meshes”), each of which was subsequently associated with the number of people living in the area aged ≥75 years (data obtained from the 2010 Japan National Population Survey [6] ). We then located all clinics on this map of divided municipalities and established each clinic’s “reachable area” using the ArcGIS program (Esri, Redlands, CA, USA). Clinics’ reachable area covered the geographical area ≥10 minutes’ travel from the agency by car, not using highways. Figure 1 shows clinics with black triangles and reachable areas with gray shadows. Meshes fully within a reachable area were coded as “reachable meshes”; the population in these meshes was coded as “reachable population.” We calculated each municipality’s total reachable population. Finally, we calculated each municipality’s “reachable proportion” (i.e., the reachable population divided by the total elderly population. Reachable proportion = reachable population (number)/all of population in a municipality). These methods were developed through discussion with four public health nursing researchers and one geographic science researcher.

Only disabled individuals (as identified by Japanese long-term care insurance level [7] ) aged ≥75 years were included. A care level of 1 (requires partial support) to 5 (requires maximum support) indicates that the individual requires support for activities of daily living. We analyzed medical insurance claim data from October 2012; the sample included 22,662 persons living in 13 municipalities regarding whom we could obtain data for all municipality characteristics.

Data regarding participants’ hospitalization duration were obtained as numerical data from health care claim data. This data was not normally distributed; therefore, it was converted to categorical data (0: ≤10 days in a single month, 1: ≥10 days hospitalized in a month). An additional category was constructed (0: ≤25 days, 1: ≥25 days) to allow this procedure’s effect on sensitivity to be monitored.

Examined individual variables included age, sex, municipality of residence, and certified care level in October 2012. Age, sex, residence, and certified care level were collected from long-term care insurance claim data. Data on the presence of conditions or impairments in October 2012 and the following six months were collected from health care insurance claim data to account for these conditions’ or impairments’ subsequent development. Data on ten types of disease were collected, following the 10th version of the International Statistical Classification of Diseases and Related Health Problems; these were as follows: cancer (C00-97), cerebral vascular disorder (I60, I61, I63, I69.0, I69.1, I69.3), arthropathy (M15-19), fracture (S02, S12, S22, S32, S42, S52, S62, S72, S82, S92, T02, T08, T10, T12), pneumonia (J12-18), chronic obstructive pulmonary disease (J41-44), dementia (F01, F03, G30), psychiatric disorder (F20-48), neurological disorder (G00-29, G31-99), and ischemic heart disorder (I20-25).

We examined the quality of collaboration between physicians and care managers in each municipality. Care managers are professionals in Japanese home care management; collaboration between care managers and physicians was considered important to preventing long-term hospitalization and promoting smooth discharge from hospitals. We measured the quality of collaboration between physicians and local care managers using the existing prefecture public survey question “Do physicians and care managers in your municipality share their clients’ goals well?” Responses used a 5-point Likert scale (1 = not at all; 5 = completely); higher scores indicated better collaboration. The question was answered by public health center officers in 13 municipalities [8] .

Municipality and individual demographic information were collated; hospitalization duration’s bivariate relationship with individual variables was subsequently analyzed. We then tested the fit of multilevel logistic models that included random intercepts for municipalities and individual- and municipality-level independent variables. Odds ratio was calculated for participants’ hospitalization duration (0: ≤10 days in a single month, 1: ≥10 days hospitalized in a month). Individual variables that were significant at p < 0.1 and all municipality variables were included in this model. Regression analysis was conducted using two models with independent variables set as clinic volume and clinic accessibility. All statistical analysis was performed using SPSS v. 21.

Clinic volume ranged between 0.0 and 9.53 per 10,000 elderly people; clinic accessibility ranged between 0.0 and 0.83. These indexes’ relationship was nonlinear; therefore, greater clinic volume did not entail greater clinic accessibility. Figure 1 illustrates that clinics were often concentrated in parts of each municipality where the population was dense. This trend supports reports on long-term care facility allocation stating that long-term

Note: Quality of coordination: Quality of coordination between physicians and care managers.

Note: Numbers are N (%).

care service providers are concentrated in urban areas due to greater market opportunities [9] . The clinics examined in this study may be distributed similarly to long-term care service providers, as both treat a similar population of disabled community-dwelling people. And it could relate to better appropriateness of accessibility

Note: OR = odds ratio, CI = confidence interval.

that people can receive treatment from outside of their own municipality; nevertheless they lived municipality with poor volume, they can use services from near clinics in neighbor municipality. Home care clinics are not homogeneously geographically distributed and arrowed to provide care for people who lived in other municipalities; therefore, clinics’ geographic accessibility more accurately indicates available clinic resources than clinic volume does.

Clinic volume and clinic accessibility were both significantly negatively correlated with hospitalization duration. This supports the Andersen-Newman model [2] , which proposes that increasing either service volume or service accessibility promotes service use and improves outcomes. Increased clinic volume or accessibility might be correlated with improved treatment and then reduced hospitalization duration among participants. This association has two likely causes: prevented hospitalization and reduced hospitalization duration. The former may reflect an increase in hospitalization avoidance through maximizing treatment quality; the latter may reflect a direct reduction in hospitalization duration. Continuity of care with the same physician has been found to prevent hospitalization and discharge delays [10] ; home care clinics may provide better continuity of care as they provide services to clients at all times of the day, every day.

The odds ratio of home care clinic volume was close to 1.0, whereas that of clinic accessibility was around 0.7. Nonetheless, due to clinics’ non-homogeneous geographic distribution and contribution in the regression analysis, we concluded that geographic accessibility better explains reductions in hospitalization duration. The implication for home care system development is that only focusing on volume of home care clinics within a community might lead to misunderstanding resource readiness. Communities constructing new home care clinics should prioritize geographic accessibility rather than clinic volume.

This study has the following limitations. Our analysis did not address caregivers’ needs or detailed medical conditions; previous research has found that these factors are related to hospitalization duration [11] . Additionally, one rural prefecture containing 13 municipalities was analyzed; the present results therefore may not generalize to urban municipalities. Because of data limitation, we cannot include usage of home care clinics in analysis.