The regional discrepancies of physician supplyhave been a growing concern in Japan. To find outhow hospitals are responding in terms of physicianpayment (by monthly salaries and additional bene-fits), we conducted a survey of acute care hospitalsin Yamagata, Japan. We asked about the salary andadditional benefits of full-time physicians and thestructural and functional characteristics of healthcare service provision. From these data we set outto assemble a model that can explain effectively thevariability of physician payment in acute care hos-pitals within the prefecture. We found that physi-cian payment was associated with variables suchas type of management, staff employed per bed,full time doctors employed per bed and averagelength of stay. Hospital location was found tohave a significant effect on payment. Variablesexpressing workload, like number of in-patients perdoctor and number of surgical operations perdoctor were inversely related. Our results suggestthat hospitals may have adapted to physicianpreferences of workplace in terms of physician pay-ment. To further address the problems of unbal-anced geographic distribution of physicians inrural areas, work-sharing and educational andtechnical support schemes may also help.

Key words: physician salary, physician demand,physician supply, regional analysis

Introduction

In the country of Japan, regional discrepancies of thesupply of physicians have been a growing concern inrecent years. There are a growing number of vacanciesof hospital physician positions especially in the field ofpediatrics, obstetrics and gynecology. Rural hospitalsare having difficulty in finding new doctors to occupythese seats, and some are shutting down the special-ties, giving rise to serious problems of access to care.

More Japanese doctors prefer to work in urban hos-pitals than rural hospitals. Several reasons for thesepreferences have been elucidated by past surveys.1,2

Doctors tend to favor the new technology and machin-ery and prefer “the latest treatment”. They think thatthese are more available in urban hospitals. Doctorsanticipate that there is exposure to excessive hard workin rural hospitals and that they would be deprived oftheir time for leisure and academic activities.1,2 Theseare not necessarily true. However these images seemto be keeping away new doctors from participating inrural heath care. As a result, those working in rural hos-pitals are having a higher and higher workload. There isa fear of being the only doctor responsible for the healthof the residents in the area, where immediate problemswould arise when the doctor may get sick or may wantto attend an academic meeting.2

Some other reasons that doctors choose to work inurban hospitals are more related to family lifestyle.Some feel that there are substantial differences in qual-ity of education for their children between the rural and

Original Article

What factors are affecting physician payment by acute care hospitals in rural Japan?

Kazushi Yamauchi1, Takao Funada2, Hiroshi Shimizu2 and Kazuo Kawahara1

1) Tokyo Medical and Dental University Department of Health Policy Science 2) Yamagata University School of Medicine Department of Health Policy Science

J Med Dent Sci 2007; 54: 1–149

Corresponding Author: Kazushi YamauchiTokyo Medical and Dental University Department of Health PolicyScience1-5-45 Yushima Bunkyoku Tokyo Japan 113-8519email: ky.hcm@tmd.ac.jp, kk.hcm@tmd.ac.jp Received October 13; Accepted December 1, 2006

urban areas.1 There may be a persistent pursuance forthe “sophisticated” urban lifestyle, compared to rural lifewhich may be imaged as boring.

Although not apparent in the surveys, it has beenshown that economic incentives do have an impact onthe doctor’s choice of occupation and place of work.2,3

Despite the fact that physicians are less in numbers inrural areas, the way hospitals (and governments) areresponding in terms of physician payment is not actu-ally clear. It could be understood that hospitals thathave more patients, operate efficiently and haveurgent need to employ physicians would tend to paymore to their employees. However, the labor market ofphysicians is known to be far from perfect. Local med-ical schools have had large control over the clinicalpositions of local hospitals leaving little room forchoosing workplace by the individual physician.4

There are usually a small number of employers (hospi-tals) within a certain area, limiting the choice of work-place for doctors.

We have done a survey of 51 acute care hospitals inYamagata prefecture, Japan. We asked the salaryand additional benefits of full-time physicians. Wealso questioned about the structural and functionalcharacteristics of health care service provision. Usingthis survey we set out to assemble a model that canexplain effectively the variability of physician paymentworking in acute care hospitals within the prefecture.Our object was to find the factors that are affectingphysician payment, workload and regional variabilitybeing some of our main interests.

Methods

Yamagata prefectureYamagata prefecture is located in the northern part of

Honshyu Island, with an area of about 9,300 km2, sur-rounded by mountains and the Japan Sea (Figure 1).In 2001, Yamagata had a population of 1.24 millionpeople. The percentage of the elderly (over 65 yearsold) in the total population was 24.1%. The prefecturalincome per capita in Yamagata was 2.5 million yen(national income per capita was 2.9 million yen).

There are four medical service areas (MSA) inYamagata prefecture. These zones are set by thegovernment, based on the everyday movement of res-idents living in Yamagata prefecture. Health care isplanned based on these regional zones by the localprefectural government. The prefectural capital, theinland city of Yamagata is located in MSA 1, which is

the largest city in Yamagata with a population of about250 thousand people. There are two other majorcities near the coastline (Sakata and Tsuruoka) in MSA4 with an added population of about 260 thousand. Theother MSAs 2 and 3 are less populated, having ruralmountainous regions. MSA 2 is an exceptionallydepopulated area in Yamagata where health careresources are sparsely distributed.

The weather is basically warm in spring and summer,but there are large amounts of snowfall in the winterseason.

The SurveyThe Survey of Trends in Hospital Patients and

Health Care Providers in Yamagata Prefecture wasconducted during the period of January 24, 2005 toFebruary 25, 2005. Questionnaires were sent to all the51 hospitals in Yamagata prefecture with acute carebeds (we call these hospitals “acute care hospitals”

K. YAMAUCHI et al. J Med Dent Sci2

Fig. 1. A map of Yamagata prefecture. Yamagata is split into 4 med-ical service areas, Murayama (MSA 1), Mogami (MSA 2), Okitama(MSA 3), and Shounai (MSA 4) by the government based on theeveryday movement of its residents. Health care provision isplanned by the local government based on these regional zones. Thelocation of the provincial capital, Yamagata city and two other majorcities, Sakata and Tsuruoka are shown. (Note that the borderlines ofSakata and Tsuruoka have recently changed due to merging with thesurrounding municipalities.)

here). Some hospitals had a mixture of acute care andchronic care beds. Detailed questions were askedabout hospital structure and management, staffinglevels, patient numbers and characteristics, salary,physician demand and medical equipment. Some of theresults of the questionnaire are summarized in Table 1.

Some items need special notice. The mean annualsalary for full-time working physicians was asked toeach hospital (Table 1). The shape of their distributionwas positively skewed. The zone of hospital location

was split into 3 groups. We collapsed MSA 2 and 3 intoone group for the analysis, because of the small num-ber of hospitals in each of these zones (especially inMSA 2 where there were only four hospitals). The zonewhere the hospital resides was one of our main inter-ests in this analysis, because we wanted to know if thelocation and surrounding environment affected theirdecision on amount of salary, especially where healthcare resources are relatively scarce.

The number of newly demanded physicians by thehospital was inquired and each hospital was asked torank the severity of this need from grade A to C, Abeing the strongest need. We counted the number ofpositions that hospitals are in urgent need for a physi-cian (grade A). Since salary of physicians is decided byhospitals, we thought that this need might have greatinfluence on physician salary.

Hospitals were also asked about the number ofphysicians who resigned from his or her job in the pre-vious year. Physicians who belong to the medicaldepartments of local medical schools usually switchworkplaces with their colleagues every 2 to 3 yearsupon instructions of the head of the department. Wecalculated the number of physicians who left their jobsthat are not attributed to these routine transfers, but tomore personal decisions (e.g. opening his or her ownclinic). We thought that hospitals might change theamount of salary to influence the doctor’s decision ofworkplace.

At the end of the survey period a total of 45 out of 51questionnaires could be recovered. Four of them hadno input about physician payment and were excludedfrom the analysis. One was a hospital where the hos-pital did not have data on mean annual benefits. Theother three hospitals left both salary and benefitsblank for unknown reasons. We had a final total of 41(80%) hospitals for construction of the model. For thefull details of the survey see reference5.

The ModelUsing the survey data we processed and picked up

21 candidate variables for our model. All the variableswere checked for linearity with income by the likelihoodratio test, including type of management and area oflocation in all models.6 Dummy variables were usedwhen the variable was categorical or there was evi-dence of non-linear relationship. The variables relatedto staffing and structure were: number of staff (per 100beds), number of fulltime physicians (per 100 beds),number of part-time physicians (per 100 beds), per-centage of physicians from local (Yamagata) medical

3PHYSICIAN PAYMENT BY HOSPITALS IN RURAL JAPAN

Table 1. The results in brief of The Survey of Trends in HospitalPatients and Health Care Providers in Yamagata Prefecture. Dataconcerning hospital structure and management, staffing levels,patient numbers and characteristics, physician payment anddemand are shown. Mean and medians are shown for numericalitems and proportions are shown for categorical items. Needed doc-tors: the number of urgently needed doctors for employment by thehospital, SD: standard deviation, ALOS: average length of stay, MSA:medical service area.

school (%), the average age of fulltime physicians(years), the average clinical experience of fulltimephysicians (years), number of beds, and the number ofspecialty services provided. The variables related toworkload were: number of average in-patients in a day(per physician), number of out-patients in a day (perphysician), number of surgical operations in a year (perphysician), number of patients for regular medicalcheckups in a year (per physician), at least one emer-gency patient per day on average (no=0, yes=1), and atleast one resident trainee (no=0, yes=1). The variablesrelated to operation and management were: type ofmanagement (public=0, private=1), average percentageof bed occupancy (%), and average length of stay(days). The variables expected to have an associationwith hospital physician demand were: number ofdemanded physicians (per 100 beds), hospital with atleast one physician who resigned from his or her job theprevious year (no=0, yes=1). And finally there were,located in MSA 2 or 3 (MSA 2 or 3=1, other=0), locatedin MSA 4 (MSA 4=1, other=0) and physician payment(1000 yen).

Physician payment was the dependent variable in ourmodel. The other 20 independent variables werechecked head to head for co-linearity, and pairs withstrong correlations were singled out. Specifically, clini-cal experience was strongly correlated (R= 0.96) withage and age was dropped out. Bed number was alsostrongly correlated with number of different specialtyservices provided (R=0.89), so the number of special-ty services provided was dropped out. The strongestcorrelated remaining pair was out-patient number perphysician and bed number (R=-0.68).

Using the remaining 18 variables we attempted toconstruct a multi-variable linear regression model.Regression was conducted by least squares maximumlikelihood method. For the selection of significant vari-ables we used backwards-stepwise algorism.7 Oursignificant removal level was 0.15. Our estimatedmodel was:

y=ςixi+b1x1+b2x2+b3x3+c

where b1 is the coefficient for the variable type ofmanagement (x1), b2 is the coefficient for the dummyvariable localized in area 2 or 3, b3 is the coefficient forthe dummy variable localized in area 4, Çi is the coeffi-cient for the variable xi, c is the constant term. After ourselection of variables was done, we checked for inter-actions for all the remaining variables with type of man-agement, since the salary decision process is likely to

be very different between public and private hospitals.Our estimated model was:

y=ςixi+ςjx1xj+b1x1+b2x2+b3x3+b4x1x2+b5x1x3+c

where Çj is the coefficient for the interaction of type ofmanagement and xj, b4 is the coefficient for the inter-action of type of management and area 2 and3(x1x2), b5 is the coefficient for the interaction of typeof management and area 4 (x1x3). We conducted aglobal likelihood ratio test to assess the fit of themodel between the one with interactions and the onewithout.(8) All calculations were done with STATA ver-sion 8.2 (StataCorp, Texas, USA) .

Results

Single variable linear regression was conductedwith all variables. All of the coefficients are shown inTable 2. We found a significant relationships (p<0.05)with number of full-time physicians, years of experi-ence, mean number of out-patients, number of peoplecoming for yearly medical check-ups, number of beds,having at least one resident trainee, location of the hos-pital in MSA 2 or 3, and private type of management.Having a doctor who resigned from his job, and numberof doctors in immediate need did not turn out to be sig-nificant variables.

The variable selection for our final model left us withfive variables excluding type of management and areaof location. They were number of hospital staff per bed,number of fulltime physicians per bed, mean number ofin-patients per doctor, number of surgical proceduresper doctor per year, and average length of stay. Weconstructed a model including interaction terms withtype of management assuming effect modification,and compared the fit of the data with the model withoutthem using the likelihood ratio test. The results told usthat the model without the interactions fit the data bet-ter (p=0.31). The coefficients of our final model areshown in Table 3A. The degree of freedom adjusted R2

for our multi-variable linear regression model was0.68, suggesting that the model fairly explains the vari-ability of doctor payment by Yamagata hospitals.

We found that an increase in the number of hospitalstaff increased physician payment while the increase inthe number of physicians per bed tended to decreasefulltime physician payment. An increase in the meannumber of in-patients per doctor, number of surgicaloperations per doctor and average length of stay also

K. YAMAUCHI et al. J Med Dent Sci4

tended to decrease physician payment. Hospitalsresiding in MSA 2 or 3 paid doctors more on averagethan hospitals in MSA 1, and public hospitals tended topay fulltime doctors more on average than privatehospitals. The coefficients of the number of staff perbed, the number of physicians per bed and the meannumber of in-patients per doctor were more than dou-bled in the multi-variable model compared to that of thesingle variable model, suggesting confounding by theother included variables.

There was insufficient input in one hospital where itwas unclear how many full time doctors in the hospitalhad graduated from the local medical school, so weused the number of doctors dispatched from the med-ical department of local medical schools. We conduct-ed a sensitivity analysis by moving the number from thelargest amount (all fulltime doctors are from the localmedical school) to the smallest amount (no doctors arefrom the local medical school). It turned out that afterthe whole process of data analysis, the final model wasthe same in either extreme.

Using the same methods, we also created a modelwithout the MSA dummy variables, so that we may findwith better reliability the variables affecting physiciansalary among the hospitals in Yamagata prefecture. Itturned out that the variables included in the final

model and the coefficients were similar (Table 3B).

Discussion

Based on our model, we found that hospitals thatemployed less staff per bed and more physicians perbed, had more in-patients per physician to take care of,that conducted more surgery per physician, that had alonger average length of stay, located in medical ser-vice area 1 (not 2 or 3), and were private tended to paydoctors less. The degrees of freedom adjusted R2 was0.68, indicating that our model could explain the vari-ability of physician payment by acute care hospitals fair-ly well. However, these results should be interpretedwith caution, since there may still be other unknownconfounding factors, and substantial residual con-founding between the existing variables.

The tendency for public hospitals to pay physiciansmore than private hospitals on average was a surprise,since past reports showed that average salary of doc-tors tended to be higher in private hospitals.9 Thebreakdown analysis of physician payment showedthat while salary of private hospitals was actuallyslightly higher than public hospitals, benefits besidesmonthly salary for public hospitals was ~8 times theamount of private hospitals on average. A large part of

5PHYSICIAN PAYMENT BY HOSPITALS IN RURAL JAPAN

Table 2. The results of single variable linear regression. Each of thevariables was fit into a linear regression model including location andtype of management. The coefficient and its 95% confidence intervalfor each of the variables are shown. See the text for the details of thevariables. MSA: medical service area.

Table 3. A: The final multi-variable linear regression model after vari-able selection. The model included these 5 variables not includinglocation and type of management. The coefficient and its 95% confi-dence interval for all the variables are shown. See the text for thedetails of the variables and the model. B: The final multi-variable lin-ear regression model not including MSA. The variables included andthe coefficients were similar. MSA: medical service area.

the benefits were for engaging in special duties (34%)and bonuses (20%). It may be that public hospitals tendto be more lenient on their payment of benefits, whilesome private hospitals pay physicians by an annuallyfixed amount being stricter on their payments.However, doctors employed by public hospitals mayface a different workload compared to doctors workingin private hospitals. For example, public hospitals his-torically have undertaken the task of critical andemergency care in Yamagata prefecture. Our surveyshows that in fact, 93% of emergency patients weretaken by public hospitals in Yamagata prefecture.

We found that variables that we thought mightdecide hospital physician payment, such as physiciandemand and clinical experience did not show up in thefinal model. This may be due to the following reasons.1) The employment of doctors traditionally has beendecided by the local medical schools and there was lit-tle discussion about treatments of the employee.Working doctors themselves were quite indifferentabout their own working conditions. The hospitals in thisway could obtain a stable supply of doctors while themedical schools could gain prestige and power over thelocal medical community.4 2) The medical insurancereimbursement rates have been declining allowinghospitals a smaller amount to adjust physician salarydespite the demands that they have.10 Reports do showthat the salary paid to doctors have been relatively sta-ble over the past few years.9 3) The assignment of vari-ables could not capture the relationship betweenthese variables and doctor payment. Some of thesevariables were made into dummy variables during thedata processing due to non-linear relationships and thesmall numbers of samples and information may havebeen lost. It has been reported that physician paymentsystems in a large proportion of Japanese hospitals arenot performance-based but rather based on a fixedannual plan.11

The observation that hospitals located in MSA 2 and3 tended to pay doctors more compared to those inMSA 1, may be due to the difficulties in fulltimeemployment of doctors in these rural areas despite theirdemands. This may be associated with the perceiveddisadvantages of attaining the latest medical technolo-gy, workload, lifestyle, and high-quality children’s edu-cation in rural areas. This cannot be generally con-cluded though, because of the small number of privatehospitals located in area 2 and 3 (actually only one hos-pital). Public hospitals were paying doctors more byadding benefits (mean amount 11416 thousand yen forMSA 2 or 3 versus 9246 thousand yen for MSA 1)

besides the salary, which is not expected to changefrom area to area in public hospitals by large amount.

The puzzling inverse relationship between the num-ber of in-patients per physician, the number of surgicaloperations and physician payment may be related tothe fact that traditionally the Japanese health insurancesystem has reimbursed more generously primarycare, drugs and laboratory tests compared to acutehigh-tech in-patient care and surgical procedures.12,13

Hospitals with relatively more in-patients and surgicaloperations per doctor may have less to afford forincreases in physician payment. It is known that inJapan physician specialists working in large hospitalsthat practice the latest treatment and medical technol-ogy earn less than their counterparts practicing prima-ry care. It is assumed that there is a tradeoff betweenthe social prestige that the physicians gain by workingin these hospitals and their foregone income.13

Moreover, Japanese patients have a preference forhigh-tech care and tend to concentrate in larger hospi-tals, in search of high quality care.10 Our analysis sug-gests that doctors working in hospitals with more in-patients and more surgical operations per physicianmay be facing a heavier workload by treating morepatients, some of them difficult to manage needinghigh-tech intensive care, but are paid less, compared tothe physicians working in the relatively smaller lesshigh-tech, less prestigious hospitals.

Average length of stay was also inversely related tophysician payment. This may also be related to thehealth insurance system. Since present health insur-ance reimbursement rates decline as admission peri-ods get longer, there may be less incentive toincrease physician salary in hospitals with patients thatstay long. Historically some Japanese hospitals havetaken the role of nursing facilities for the elderly.4,13

However, patients that stay longer may be patients thatare medically difficult to manage and have little choiceabout the time and place of receiving medical care,which may further disadvantage physician payment bythe hospital. We could not adjust for differences incase-mix between hospitals, which may have significantimpact on this relationship between workload andpayment.

The interesting and contrasting result of full timephysicians per bed and staff per bed indicates a possi-ble tradeoff relationship between the two. More full timephysicians employed may have resulted in smallerpayments simply because one doctor’s share out of thehospital’s salary pool diminished if the number ofpatients and intensity of care were the same, while

K. YAMAUCHI et al. J Med Dent Sci6

replacement of physicians by other co-medical staffmay have left room for larger physician payment due totheir lower employment costs.

Our study is based on the survey conducted to all thehospitals with acute care beds in Yamagata prefecture.The hospitals that were excluded for the analysis,were all private, five of them residing in MSA 1, two ofthem in MSA 2 and 3 and two of them in MSA 4. Thesize of these hospitals ranged from small size with onlya selected variety of services to medium size with awide variety of specialties. The exclusion of thesehospitals from our analysis may have introduced biasinto our results. However, since there is no reason tobelieve that there are any unseen characteristics thatthis group of hospitals shares in common, we assumethat the private hospitals left in the analysis may serveas a representative of private hospitals in Yamagataprefecture.

Medical school clinical departments may not beable to enjoy the power they have exercised on theemployment of physicians in local hospitals for long.There has been a drastic change in the education sys-tem of medical school graduates in 2004, where grad-uates are allocated to hospitals by a matching systemand not by intentions of the medical school.10 The totaleffect of this new policy on the choice of workplace isnot yet clear, but it can be naturally expected that grad-uates would tend to concentrate in large urban hospi-tals. Yamagata has recently organized a committee(Zao Kyogikai) consisting of members from the localgovernment, medical school, and health care facilitieswithin the prefecture to discuss a fair, efficient andtransparent geographical positioning of physicianswithin the prefecture.

In summary, we found that physician payment tend-ed to be associated with variables such as type of man-agement, staff employed per bed, full time doctorsemployed per bed and average length of stay.Hospital location was found to have a significanteffect. Variables expressing workload, number of in-patients per doctor and number of surgical operationsper doctor, were inversely related. Variables that wereexpected to have an association with the hospital’sdecision for payment (physician demand, clinicalexperience) were excluded from the final model.These results suggest that hospital payment hasadapted to physician preferences on workplace tosome extent. Further research is needed to evaluate ifthese variations in payment are sufficient to alleviate

the choice of workplace by physicians. To furtheraddress the problem of unbalanced distribution, 1) sub-sidizing scheme to increase physician incomeengaged in rural care should be considered, and ifphysician choices are indeed inelastic, 2) a work shar-ing scheme to relieve the anxiety of excessive hardwork attached to rural care, and 3) a medical educationand technical support system arranged by medical cen-ters of the area for physicians to keep in touch with thelatest treatments, may also be helpful.

Acknowledgement

This study has been funded by Grant-in-Aid forScientific Research for Health Policy Science by theMinistry of Health, Labour and Welfare.

References1. Japan Municipal Hospital Association Committee for

Ensuring the Stable Supply of Physicians. Report on Action forEnsuring Stable Supply of Physicians (2004). [in Japanese]

2. Matsuura Y, Takeuchi K, Hatano F. Report on the committeefor training and securing the supply of health care workers.Hiroshima Igaku 2004;57(12):941-68. [in Japanese]

3. Akagi H. Regional disparities in health care costs. Grant-in-Aidfor Scientific Research for policy science by the Ministry ofHealth, Labour and Welfare. 1999. [in Japanese]

4. Ikegami N, Campbell JC. The art of balance in health policy:maintaining Japan’s low-cost egalitarian system. New York:Cambridge University Press, 1998;53-86.

5. The report on The Survey of Trends in Hospital Patients andHealth Care Providers in Yamagata Prefecture. ShougaiKyouiku To Chiiki Iryou 2005;Suppl 1:2-107. [in Japanese]

6. Holford TR. Mulitvariate methods in epidemiology. New York:Oxford University Press, 2000;163-204

7. Hosmer DW, Lemeshow S. Applied logistic regression. NewYork: Wiley-Interscience,. 2001;91-142.

8. Kleinbaum DG, Klein M. Logistic Regression. New York:Springer, 2003; 191-226.

9. Shiraga M. The salary of physicians and ensuring their stablesupply. Byoin 2004;63(2):122-30. [in Japanese]

10. Ikegami N, Campbell JC. Japan’s health care system: con-taining costs and attempting reform. Health Aff2004;23(3):26-36.

11. All Japan Hospital Association Committee of Health CareWorkers. Survey on the Salary of Health Care Workers.2004. [in Japanese]

12. Ikegami N, Campbell JC. The art of balance in health policy:maintaining Japan’s low-cost egalitarian system. New York:Cambridge University Press, 1998;145-74.

13. Yoshikawa A, Bhattacharya J, Vogt WB, et al. Health eco-nomics of Japan. Tokyo: University of Tokyo Press, 1996;3-37.

7PHYSICIAN PAYMENT BY HOSPITALS IN RURAL JAPAN