Data

In 2004, Niigata University Hospital had 810 inpatient beds, 23 clinical departments and medical staff of 818. The service area of the hospital as a tertiary care hospital covers all districts in Niigata Prefecture, which has a population of 2,400,000. There are 34 paediatric inpatient beds in ward for children. These beds do not include beds for paediatric surgery patients. The size of the hospital is typical for a Japanese university hospital.

Patient data were analyzed for all children and adolescents under fifteen years of age who had received inpatient care at Niigata University Hospital during the period from April 2004 to March 2005. Information on paediatric inpatients under 15 years of age was collected. Data on patient age, sex, residential address, route of referral to Niigata University Hospital, discharge diagnosis classified by Japanese case mix classification for inpatients (diagnosis procedure combination, DPC) [5, 6], length of hospital stay (LOS), outcome of admission and route of referral from Niigata University Hospital were obtained from the hospital information system. In the DPC system, patient discharge is classified into 16 major diagnostic categories and 1,727 case-mix groups. Data on average and standard deviation (SD) of LOS of the patients in Japan, who were assigned to each DPC, were obtained from the Ministry of Health, Labour and Welfare. Data on age-structured population per district were obtained from National Census of Japan conducted by the Ministry of Internal Affairs and Communications in 2000.

Data on population per district were converted to data on population per 2 km mesh using the GIS software ArcView GIS 3.1a and ArcGIS 9 (Environment Systems Research Institute Inc., CA, USA). Data on locations of paediatric clinics and hospitals with paediatric wards were obtained from lists of the Japan Paediatric Association, Japanese Society of Paediatrics, Japan Medical Association and their local organizations. Road network data also contain codes for different classes of roads, average car speed and rules of the roads in order to calculate how long it would take to drive along a particular road segment [7].

Referral from clinics and other hospitals

To study the geographic distribution of patients admitted to Niigata University Hospital, empirical Bayesian estimates for the rate of inpatients per childhood population at risk were plotted on a GIS map with 2 km meshes [8]. To study the relationship between spatial accessibility to Niigata University Hospital and the distribution of patients, the number of inpatients who lived within an area that is a certain time by car from Niigata University Hospital was calculated by using Network Analyst (ESRI Japan). The same procedure was performed for other hospitals in Niigata Prefecture in order to compare the inpatient service areas of Niigata University Hospital and general hospitals. The general hospitals had 177 to 312 inpatient beds and one or two paediatricians. Residence records for the patients admitted to the five hospitals in Niigata City were obtained from annual reports published by the hospitals. The distribution of patients referred from other medical facilities was plotted on GIS maps using the kernel density method [9, 10] and was compared with the distribution of paediatricians. For the kernel density method, we computed a density layer using a one-kilometre service area radius and a 200-meter-square area as a cell size, which yields a smooth map resolution. To differentiate the patients who have more complex conditions and those who have not, the patients were subdivided into two groups according to the information on DPC of each patient: one group of patients who had severe complications and one group of patients who did not have severe complications. The patients were also subdivided into those who admitted through emergency or taken to the hospital by ambulance and those who were not.

Referral to aftercare

For the purpose of this study, aftercare was defined as referral to a paediatric aftercare program, such as outpatient care and transfer to paediatric wards of other hospitals, not including foster care, a group home or a nursing home [11]. Distributions of patients referred to the Outpatient Department of Niigata University Hospital, patients referred to outpatient care of other medical facilities and patients transferred to paediatric wards of other hospitals were plotted on GIS maps. Since delay in referral of a patient to aftercare prolongs LOS of the patient, spatial differences in LOS of inpatients might reflect the spatial differences in accessibility to aftercare. To estimate the spatial differences in accessibility to aftercare, LOS of each patient was scored and plotted on a GIS map. The SD scores for LOS (LOSSDS) of a patient who was assigned to a DPC were calculated as follows:

$\frac{((actualLOSofthepatient)-(averageLOSoftheDPC))}{(SDoftheLOSoftheDPC)}.$

Contour maps of LOSSDS of the patients were produced by interpolation using the ordinary Kringing method [12].

Referral from clinics and other hospitals

During the study period, a total 8,177 patients were admitted to Niigata University Hospital. Among them, 1,323 patients were under 15 years old of age. The outline of the inpatient children and adolescents less than 15 years of age are shown in Table 1. The geographic distributions of all inpatients and inpatients less than 15 years of age were geocoded by residential addresses to a GIS map as shown in Fig. 1. The number of patients who visited Niigata University Hospital without referral from other hospitals and clinics was 800 (60.4%). The empirical Bayesian estimates for the rate of inpatients per childhood population at risk were plotted on a map with 2 km mesh (Fig. 2). Although these patients were predominantly distributed around Niigata City, the capital of Niigata Prefecture, there were some meshes with large numbers of inpatients far from Niigata City (Fig. 2A). This indicated that patients visited Niigata University Hospital from broad area of the prefecture without referral from other medical facilities.

Characteristics	N or mean	%
Age at admission (mean ± SD, years)	4.6 ± 4.9
Gender, male	709	53.6
SD scores for LOS (mean ± SD)	-0.46 ± 3.66
Admissions	1,323
Number of actual patients	1,096
Referral from other medical facilities	513	38.8
Referral on discharge	1,275	96.4
Outpatient Department of Niigata University Hospital	1,096	82.8
Return to the primary paediatrician	153	11.6
Transfer to paediatric wards of other hospitals	26	2.0
Death during admission	17	1.3

The relationship between spatial accessibility and distribution of residences of patients in Niigata University Hospital and five other hospitals in Niigata City also supports this finding (Fig. 2C). Eighty-three percent of patients admitted to general hospitals lived in an area 30 min from the hospital by car, whereas only 19% of patients admitted to Niigata University Hospital lived in a 30-min area. On the other hand, there were marked spatial differences in the distribution of residences of inpatients that lived in or near Niigata City (Fig. 2B). This suggested that competition with other hospitals to get patients existed in the case of patients who were directly admitted to the hospital without referral from other medical facilities.

Of the 1,323 patients under 15 years of age, 513 (38.8%) were referred from clinics and other hospitals. Results obtained by using the kernel density method indicated that inpatients that lived near Niigata City tended to be admitted directly to Niigata University Hospital, whereas inpatients who lived far from Niigata University Hospital tended to be referred from other medical facilities (Fig. 3A). the distribution of the inpatients who had more complex conditions was biased in the region far from Niigata University Hospital (Fig. 3B). The distribution of patients who were referred from other medical facilities had a tendency to agree with that of patients who had more complex conditions (Fig. 3C). On the other hand, the distribution of patients admitted through emergency showed a pattern different from that of the distribution of referred patients and that of patients with complex conditions (Fig. 3D).

Referral to aftercare

Almost all of the inpatient children (1275 out of 1323, 96.4%) received referral to aftercare, and 1096 (82.8%) of them received care in Outpatient Department of Niigata University Hospital after discharge. One hundred and fifty-three patients (11.6%) were referred to the clinic or outpatient departments of other hospitals, and 26 patients (2.0%) were transferred to other hospital wards after discharge from Niigata University Hospital. These patients lived far from Niigata City and in the Sado Island, which is located 40 km away from Niigata City (Fig. 4B)

Next, we investigated spatial difference of LOSSDS in the inpatients in order to ascertain whether there were barriers to referral to aftercare in each district. As shown in Fig. 5A, there were some foci with high LOSSDS near Niigata City (Fig. 5A). We paid special attention to one of the foci (indicated by an arrow) and investigated why the patients who lived in that focus showed high LOSSDS. There were no problems in spatial or physical accessibility in the focus, because the focus was located near arterial roads (Fig. 5B). The focus was located within area 30 min by car from two hospitals (a and b, Fig. 5C). Although these hospitals had beds for children, each hospital had only one paediatrician. Therefore, shortage of paediatric manpower is one of the reasons for the clustering of patients.