Evaluating the effect of Japan’s 2004 postgraduate training programme on the spatial distribution of physicians

Optimizing physician supply and distribution are vital ingredients of sustaining equitable access to essential health services. However, there may be a tension between expanding access to physician supply versus improving the quality of their training, as our case study from Japan will illustrate.

In 2004, the Japanese Ministry of Health, Labour and Welfare (MHLW) sought to improve the quality of medical residency training nationwide by instituting a new postgraduate medical education (PGME) programme [1]. Prior to 2004, medical graduates commenced their specialty training immediately after graduating from medical school. Since medical graduates ended up being trained only in their chosen specialty, they had little or no experience in other fields. Since 2004, a 2-year general residency has been required of all medical graduates, ensuring that residents rotate through different specialties to gain hands-on experience at teaching hospitals designated by MHLW. These hospitals would include both university-affiliated and non-university-affiliated hospitals, encompassing both public and private institutions [2]. After 2004, new physicians commenced their specialty training following the mandatory 2-year residency training programme.

Before 2004, a majority of new graduates entered their specialty training programme at the university hospital associated with the medical school where they graduated. In those times, physicians were dispatched to non-university hospitals by the department of their specialization in the university hospital, and individual choice was not considered.

As part of the 2004 reforms, the Ministry permitted medical graduates to choose their preferred residency location directly through a national matching system for the first time [2]. Then, medical graduates tended to select independent urban hospitals not affiliated with a university for their training rather than university hospitals [3, 4]. Even though the matching system only applied to the 2-year residency training programme, young physicians now had the freedom to choose a residency location that could also provide them the specialty training, in terms of targeted mentors and professional recognition, when the residency was completed. As a result, young physicians tended to stay at the same location and continue their specialty training after the residency training. Subsequently, the number of incoming physicians at university hospitals decreased. Medical specialty departments in university (teaching) hospitals normally dispatch physicians at all levels to other affiliated hospitals. Because the number of physicians at university hospitals has decreased, university hospitals find it increasingly difficult to send physicians to affiliated hospitals, which are often located in rural and underserved areas [5]. Furthermore, physicians working in rural areas return to university hospitals to fill the void left by the lack of new medical graduates. The Japan Medical Association Research Institute reported that almost 60% of all university hospitals have reduced the dispatch of physicians to other medical institutions since the launch of the new training programme [6].

To determine if the 2004 reform had the unintended consequence of impacting physician supply and distribution, we performed an ecological analysis using nationally available data.

In a previous study, we examined the impact of the 2004 reform on the supply of paediatricians in Japan [7]. Our previous study suggested that the 2004 medical residency scheme had the unintended consequence of making it easier for new graduates to choose residency training location, exacerbating regional inequalities in paediatrician supply. A limitation of our previous study was that we ignored the distinction between hospital-based clinicians and smaller-clinic-based physicians (that is primary care clinics with fewer than 20 beds). The 2004 reforms affected resident training programmes that take place primarily in training hospitals designated by the MHLW, as opposed to physicians’ practice locations in small clinics. Therefore, the spatial distribution for clinic-based physicians would not be expected to be affected by the new scheme.

Indeed Toyabe [8] analyzed the time trends in the number and distribution of physicians between 1996 and 2006 and showed that the increasing inequality in geographic distribution of hospital-based clinicians was markedly worse compared to clinic-based physicians; moreover, the number of physicians working at hospitals increased over time in urban areas but not in areas with low population density.

In this study, we sought to disaggregate the hospital clinician trends from clinic-based physicians. We are interested in 2 questions: first, what are the principal drivers of physicians’ practice location choice, and second, what factors affected physician distribution before and after the launch of the new programme in 2004?

Table  2 shows the aggregate level change in physician supply. Both the absolute number of physicians and the physician-to-population ratio increased in both the pre- and post-periods.

Table  3 shows descriptive statistics of dependent variables, which are differences in numbers of physicians between both 4-year time periods: 1998 to 2002 for the pre-period, and 2006 to 2010 for the post-period, and continuous independent variables for the pre- and post-periods, along with the P-value resulting from testing the Equality of Means. Overall physician density increased (P-values for overall physician density was .001, for hospital physician density was .011, and for clinic physician density was < .0001). Table  4 shows descriptive statistics of categorical independent variables for the pre- and post-periods. The number of urban, suburban, and rural centres changed slightly between the pre- and post-period, and the number of medical schools did not change during the study period.

Table  5 presents the results of our main predictors of interest from the multivariate ordinary least-squares regression models controlled by all other factors shown in Table  1 for all physicians. While the change in physician supply was not associated with baseline physician density in the pre-period, it was positively associated with baseline physician density in the post-period (P-value < .0001). In other words, in the post-2004 period, areas with already high physician supply attracted even more physicians. After adjustment for all other variables, we estimated that each unit increase in physician density (1 physician per 1,000 population) in 2006 was associated with an increase in the number of physicians of 64.80 in 2010. While the change in physician supply was inversely associated with urban centres in the pre-period (P-value .026), it was positively associated with urban centres in the post-period (P-value < .0001). On average, we estimated that urban centres lost 26.71 physicians from 1998 to 2002 and gained 68.81 from 2006 to 2010, compared to rural areas after adjustment for all other variables.

Table  6 shows the results from the LRT for all physicians. Physician density, urban/rural status, and the SES composite index did not have meaningful effects on the change in physician supply in the pre-period, while they had significant effects in the post- period.

Table  7 shows the results from the multivariate ordinary least-squares regression analyses for hospital-based clinicians. While the change in hospital physician supply was inversely associated with baseline hospital physician density in the pre-period (P-value = .047), it was positively associated with baseline hospital physician density in the post-period (P-value < .001). Urban centres and the SES composite index showed statistically-positive associations with the change in hospital physician supply in the post-period (P-values < .001 for urban centres, and .025 for SES c index), whereas the change in hospital physician supply was inversely associated with urban centres and no such effects of SES composite index during the pre-period (P-value = .015 for urban centres and .483 for SES composite index). The coefficient equality test showed significant differences in coefficients between the pre-period and post-period both for urban centres (P-value < .0001) and the SES composite index (P-value = .026).

Table  8 shows the results from the LRT for physicians working at the hospitals. Hospital physician density and urban/rural status had meaningful effects on the change in hospital physician supply both in the pre- and post-periods. However, both estimates were negative in the pre-period, and positive in the post-period. Therefore, the effects of those variables were meaningful in both periods in an opposite direction. SES composite index had significant effect in the post-period, while there was no such effect in the pre-period.

Table  9 shows the results from the multivariate ordinary least-squares regression analyses for clinic-based physicians, including clinic owners. The change in clinic physician supply was positively associated with baseline hospital physician density both in the pre- and post-period (P-values = .001 for pre-period and < .001 for post-period). The coefficient equality test suggested that there were significant differences in coefficients between the pre-period and post-period for hospital physician density (P-value of coefficient equality test = .006). The change in clinic physician supply was inversely associated with baseline clinic physician density both in the pre- and post-period (P-values = .013 for pre-period, and .022 for post-period) and there were no significant changes in the estimated impact of these factors between the 2 periods. Urban centres showed statistically positive associations with the change in clinic physician supply in the post-period (P-value = .003). Although it did not show statistically positive associations with the change in clinic physician supply in the pre-period, the coefficient equality test showed there were no significant differences in coefficients between the pre-period and post-period for urban centres (P--value = .050). The SES composite index showed statistically positive associations with the change in clinic physician supply both in the pre- and post-period (both P-values < .001).

Table  10 shows the results from the LRT for physicians working at the clinics. Urban–rural status and SES composite index, which had meaningful effects on the change in hospital physician supply between pre- and post-periods, also had meaningful effects on the clinic physician supply both in pre- and post-periods.

All the robustness checks showed the similar results. (Detailed results are available upon request).

The current study explored community-level factors that affected changes in physician supply in Japan, as well as the impact of the 2004 national training programme on physician supply. We found that the determinants of the change in physician supply differed before and after the launch of the 2004 PGME programme for the following two reasons. First, physicians tended to move to places with higher physician density after the launch of the programme. Second, physicians tended to leave urban centres before the launch of the 2004 programme, whereas they tended to move to urban centres afterwards. The LRTs supported these phenomena, that is physician density was an important variable in the post-period, while it was not in the pre-period, and urban/rural status were important variables in post-period, while they were not in the pre-period. In short, the PGME programme - which was introduced to improve the quality of medical training - had the unintended consequence of worsening geographical inequality in physician supply. The essence of this story is that our disaggregated analysis revealed that the deterioration in physician supply to rural and lower-SES areas primarily affected hospital-based clinicians. More specifically, before the launch of the 2004 programme, physicians working at hospitals tended to go to locations with lower physician density, while the opposite trend was observed after the launch of the new programme. Second, physicians working at hospitals tended to move to urban centres and higher SES communities in the post-period, whereas no such effects were found pre-period; moreover, for physicians working at the clinics, tendencies to move to urban centres and higher SES communities were observed in both pre- and post-periods. The LRTs also supported these phenomena, that is hospital physician density and urban/rural status were important variables in both pre- and post-period for physicians working at hospitals; however, the direction of the effects were the opposite, and urban/rural status and SES composite index were important variables in both pre- and post-periods for physicians working at clinics.

The public perception in the Japanese media (also acknowledged by the government) is that regional inequality in physician supply worsened after the launch of the 2004 programme because many physicians - when allowed to choose freely - prefer work in urban areas [5, 39]. However, published quantitative analyses of the 2004 programme’s impact on physician distribution have previously failed to corroborate these views [11, 40‐42]. The current study showed that our results are more in line with public perception.

There are several possible reasons why physicians favour areas with higher hospital physician density. First, in Japan, salaries of physicians at hospitals are not based on a fee-for-service structure [43]; therefore, there is not much sense of competition among hospital physicians. Rather, a colleague will complement the workload. Therefore, physicians working in hospitals with higher physician density tend to enjoy lower workloads. Additionally, when physicians need to admit patients, but hospitals do not have beds for the patients, physicians look for the beds in other hospitals. It is easier for physicians in the areas with more hospitals and physicians to find beds for patients. Also, Krishnan [15] noted that there is a tendency among physicians to prefer locations near their colleagues to benefit from professional interactions. The Japanese government raised the number of medical school admissions to increase the number of physicians until 2019. Available evidence indicates that only physician-rich geographic areas have been able to develop their hospital facilities and enhance their relative physician supply [44], providing more evidence of collegial preferences.

Another interesting result from the disaggregated analyses was that physicians working at the hospitals tended to move to urban centres and communities with higher SES post-period, whereas no such tendency was found pre-period. This suggests that residential quality has emerged as a driving force in hospital physicians’ location preference following the 2004 legislation. Our disaggregated analyses indicate that clinic-based physicians tended to move to urban centres and communities with higher SES both before and after the launch of the 2004 programme, and there were no significant differences between the two periods. Physicians at clinics had been able to decide their preferred practice location even before the launch of the 2004 programme, while the physicians at hospitals gained that freedom only after the launch. Our results showed that physicians will move to urban areas and areas with higher SES when the practice location choice is left to individual freedom. As revealed in our previous study [7], physicians are no different from the rest of the Japanese population in preferring to move to urban areas [45].

The critical challenge in most settings has been that of recruiting physicians to rural areas, where physician coverage is generally low and child health often significantly poorer compared to urban areas. Since available evidence demonstrates that students of rural origin are more likely to work in rural settings [46‐48], some rural local governments adopted a programme, which reimburses medical school tuition for those students who agree to work in the rural areas to which they are assigned for a designated number of years after graduation. The government has started discussions to establish training courses that can help rural physicians obtain their board-certified specialist status after their mandatory training in the rural areas so that more physicians will stay in rural areas even after the mandatory periods. We believe that the results of our study will contribute to the discussion. We would also like to note that the policy to reimburse medical school tuition for those students who agree to work in the rural areas for a designated number of years was first instituted on a large scale in 2006 [46]. The first graduates covered by this policy graduated in 2012; therefore, this policy does not impact the results of our study.

We have previously shown how the 2004 PGME reforms impacted the regional distribution of paediatrician supply [7, 49]. Our current analysis extends our previous conclusion by showing that the impact of the 2004 reforms has not been limited just to the supply of paediatricians. We believe that we are able to show underlying evidence that new placement schemes should be developed to achieve more equity in access to medical care in Japan.

There are some limitations to consider in interpreting the results of this study. First, publicly-available data do not indicate whether a physician works full-time or part-time. This analysis was based on an overall headcount, which might overestimate the number of physicians. In particular, the percentage of female physicians is increasing [25] and the percentage of female physicians who work part-time is higher than that of men [50]. Second, publicly-available data do not include information on physician age or gender, although previous studies have noted the effect of gender or age on differences in physicians’ practice location choices [18, 51‐53]. In reality, a majority of clinics in Japan are privately-owned, with the very few exceptions of clinics owned by the local governments in some rural areas. Therefore, it is possible that the data for physicians at clinics included physicians who do not provide clinical work, but are rather engaged in business administration. Last, some local governments have been implementing their own policies to attract physicians to specific (generally underserved) areas, and these policies could also influence physician practice location choice. Some regions use public funding to provide a better salary to physicians who work in rural areas to ensure physician supply in underserved areas [48]. Omitting this variable for this type of aid in the model would underestimate the coefficients reported here reflecting the true effect of the policy change on physician practice location choices. Therefore, the true effect would be even stronger than the estimated coefficients in the current study.

After the 2004 PGME reforms, physicians tended to move to places with higher physician density, and the shift toward urban, higher-SES locations became more accentuated. These changes were primarily driven by hospital-based clinicians. Our study suggests that new placement schemes should be developed to achieve more equity in access to medical care in Japan.