The Survey of Physicians, Dentists, and Pharmacists is a national census survey conducted by the MHLW every 2 years. The Medical Practitioners’ Act requires all medical practitioners report their status every 2 years, and the survey questionnaires are completed by the medical practitioners themselves. The response rates for the Survey of Physicians, Dentists, and Pharmacists have not been published by the MHLW but are estimated to be approximately 90% .
We used data from the survey with permission from the MHLW. Data on the registration number, year of registration as a physician, sex, workplace type (municipality type and medical institution type), area of practice, and board certification status for each physician in 2012 and 2014 were evaluated in this study. We used the physician registration number to establish a cohort dataset. Board certification status and geographical migration patterns between the two survey periods were then analyzed. The municipality borders that changed because of mergers were adjusted in the two periods. In total, 1896 geographical areas that included all municipalities were identified and used for this study. These areas were classified into three categories based on population density: first tertile (lowest population density, rural), second tertile (second-lowest density, intermediate), and third tertile (highest density, urban). The cut off points of population density were 106.0/km2 and 629.8/km2. As Japan has not rurality criteria similar to the Office of Management and Budget (OMB) standards in the United States , we used a classification based on population density. A previous study in Japan  employed population density and used quartiles, but the number of segments was arbitrary, and thus three segments were employed in the present study. Moving to a more urban area was defined as working in a rural area in 2012 and moving to an intermediate or urban area in 2014 or working first in an intermediate area and then moving to an urban area. Moving to a less urban area was defined working in an urban area in 2012 and moving to an intermediate or rural area in 2014 or working first in an intermediate area and then moving to a rural area.
To assess urban–rural movement patterns between the two survey periods, 3 × 3 tables (rural, intermediate, and urban in 2012 × rural, intermediate, and urban in 2014) were prepared for each board certification status change from 2012 to 2014: no board certification in either year; board certification in 2014 but not 2012 (newly board certified); board certification in both years (maintained board certification); and board certification in 2012, but not in 2014 (lost board certification).
To assess whether board certification status change was associated with geographical migration, logistic regression analyses were conducted. In each municipality-type tertile, we tested whether the odds of migrating to more urban municipalities were associated with board certification status change from 2012 to 2014, adjusting for sex (male/female), years since registration as a physician (0–14, 15–29, 30–44, or ≥ 45), and type of workplace in 2014 (hospital/clinic or other). Logistic regression analysis was also performed to assess whether the odds of migrating to more rural municipalities were associated with board certification status changes from 2012 to 2014, adjusting for the same covariates. A sub-analysis of the individual specialties was conducted to examine any differences from the total group.
Logistic regression was also used to assess whether practicing in a rural area was associated with maintaining board certification status, testing whether the odds of holding board certification was associated with practice location: staying in a rural area (working in rural municipalities in both 2012 and 2014), staying in an urban area (working in urban municipalities in both 2012 and 2014), and others, adjusting for sex, years since registration as a physician, and type of institution in 2012. A sub-analysis of the individual specialties was conducted to examine any differences from the total group.
In these analyses, we defined board-certified physicians as physicians with board certification in general internal medicine, surgery, pediatrics, obstetrics and gynecology, orthopedics, neurosurgery, ophthalmology, otorhinolaryngology, acute medicine, anesthesiology, dermatology, urology, plastic surgery, radiology, pathology, or rehabilitation. In Japan, the abovementioned 16 areas and psychiatry, laboratory medicine, and general practice have been defined as general areas of board certification. However, we were unable to include the latter three areas in the analysis because insufficient data were available from the Survey of Physicians, Dentists, and Pharmacists and it was not possible to calculate status changes for these area from 2012 to 2014. Data on board certification status in psychiatry was not collected in the 2012 survey, and data on board certification status in laboratory medicine was not collected in either survey. Board certification in general practice began after the study period. There were 8293 physicians board certified in psychiatry in 2014 , and 588 physicians board certified in laboratory medicine in August 2016 .
Population density was calculated based on the basic resident population register as of 01 January, 2015, by the Ministry of Internal Affairs and Communications , and municipality size was based on the Statistical Reports on the Land Area by Prefectures and Municipalities in Japan by the Geospatial Information Authority . In this analysis, 2015 population data were applied to municipalities for both 2012 and 2014 to set the urban–rural classification of the municipalities for the study period. We used the population data in 2015, because the 2014 survey was conducted on 31 December 2014.
For the statistical analyses, P-values less than 0.05 were considered significant. SPSS Version 22.0 J software (Japan IBM, Tokyo, Japan) was used for all statistical analyses.