Prostate Cancer and Prostatic Diseases (2012) 15, 391 -- 396 & 2012 Macmillan Publishers Limited All rights reserved 1365-7852/12 www.nature.com/pcan

ORIGINAL ARTICLE The impact of county-level radiation oncologist density on prostate cancer mortality in the United States

S Aneja1 and JB Yu1,2,3

BACKGROUND: The distribution of radiation oncologists across the United States varies significantly among geographic regions. Accompanying these variations exist geographic variations in prostate cancer mortality. Prostate cancer outcomes have been linked to variations in urologist density, however, the impact of geographic variation in the radiation oncologist workforce and prostate cancer mortality has yet to be investigated. The goal of this study was to determine the effect of increasing radiation oncologist density on regional prostate cancer mortality. METHODS: Using county-level prostate cancer mortality data from the National Cancer Institute and Centers for Disease Control as well as physician workforce and health system data from the Area Resource File a regression model was built for prostate cancer mortality controlling for categorized radiation oncologist density, urologist density, county socioeconomic factors and pre-existing health system infrastructure. RESULTS: There was statistically significant reduction in prostate cancer mortality (3.91--5.45% reduction in mortality) in counties with at least 1 radiation oncologist compared with counties lacking radiation oncologists. However, increasing the density of radiation oncologists beyond 1 per 100 000 residents did not yield statistically significant incremental reductions in prostate cancer mortality. CONCLUSIONS: The presence of at least one radiation oncologist is associated with significant reductions in prostate cancer mortality within that county. However, the incremental benefit of increasing radiation oncologist density exhibits a plateau effect providing marginal benefit. In order to optimize outcomes a geographically aware policy, which addresses the size and distribution of the workforce, must be in place in order prevent geographic disparities in prostate cancer mortality.

Prostate Cancer and Prostatic Diseases (2012) 15, 391--396; doi:10.1038/pcan.2012.28; published online 24 July 2012 Keywords: access to care; physician workforce; radiation oncologist density

INTRODUCTION cancer, the association between the availability of radiation As the demand for cancer care increases, equitable access to oncologists and prostate cancer mortality merits exploration. oncology services is critical to improving cancer mortality and Furthermore, the relative impact of the distribution of radiation optimizing outcomes. Though evidence indicates that prostate oncologists, urologists and primary care providers on prostate cancer mortality has been improving in recent years, the benefit cancer mortality is unknown. from modern cancer treatment may not be uniform throughout It is likely that the geographic distribution of radiation the United States.1,2 Furthermore, given the aging US population, oncologists is important to the receipt of cancer treatment, the incidence of prostate cancer is expected to increase particularly for those patients who are not candidates for surgery. dramatically over the next 20 years. Whether the current physician Alternatively, patients who are surgical candidates, but refuse workforce is adequately and equitably distributed, to meet and surgery, may choose radiation therapy as an alternative curative optimize the growing demand for cancer care, is an important treatment if a radiation oncologist is geographically accessible. area of ongoing study.3--6 External beam radiotherapy, the dominant form of radiation Several types of providers are involved in the diagnosis and treatment in the United States, typically requires multiple daily treatment of prostate cancer. Primary care providers and radiation treatments for 6--9 weeks, making the geographic urologists are typically involved in initial diagnosis, whereas distribution of radiation oncologists more important, particularly urologists, medical oncologists and radiation oncologists could for patients with limited mobility and resources for travel. Recent all potentially be involved in the primary treatment of prostate evidence has found a geographic maldistribution of the radiation cancer. It is well known that access to primary care is an oncology workforce throughout the United States, with radiation important predictor of cancer-specific mortality.7 Therefore, oncologists clustering in metropolitan locations.6 As it is known some argue that increasing the number of primary care that travel time to the nearest cancer center varies significantly physicians, rather than specialists, is the most efficient way to throughout the country, inequities in geographic access may be improve health outcomes for the entire US population.8 associated with variations in cancer mortality.9 We initiated this However, growing evidence suggests that for prostate cancer, study to elucidate whether the density of radiation oncologists specialist density does impact outcomes.5 As radiation oncolo- within in a county was related to variations in prostate cancer gists also serve as primary treatment providers for prostate mortality.

1Yale School of Medicine Department of Therapeutic Radiology, New Haven, CT, USA; 2Cancer Outcomes, Public Policy, and Effectiveness Research (COPPER) Center at Yale, New Haven, CT, USA and 3Yale Cancer Center, New Haven, CT, USA. Correspondence: Dr JB Yu, Yale School of Medicine Department of Therapeutic Radiology, 333 Cedar Street New Haven, CT 06515, USA. E-mail: [email protected] Received 28 March 2012; revised 17 May 2012; accepted 10 June 2012; published online 24 July 2012 Radiation oncologist density and prostate cancer mortality S Aneja and JB Yu 392 MATERIALS AND METHODS Statistical analysis Data sources A model was built with prostate cancer mortality per 100 000 people as the The geographic units of analysis in this study were the 3141 counties in the primary outcome variable. In an effort to better examine if incremental United States as defined by the 2000 Census. Similar to previous studies changes in physician density accompany changes in prostate cancer investigating the impact of physician density on cancer-related mortality, mortality, radiation oncologist and urologist densities were categorized rural counties were excluded from the analysis because only 0.4% of the (0, 0.1--1.0, 1.1--2, 2.1--4.0, 44.0 per 100 000 people). Different radiation 669 rural counties in the United States possessed radiation oncologists and oncologist density categories were mapped using the mapping software many of them lacked complete mortality data.4,10 Following exclusion of ArcGIS version 9.2 (Environmental Systems Research Institute, Inc., rural counties, 2472 non-rural counties composing B78.7% of the United Redlands, CA, USA). Univariate associations between physician workforce, States possessed complete mortality data and were available for analysis. health system and socioeconomic predictor variables and cancer mortality Counties were classified as rural based on 2003 Department of Agriculture were calculated using t-tests for categorical variables and linear regression Rural/Urban Continuum Codes. Physician data were obtained using the for continuous variables. Multivariate regression models were built using 2008 Area Resource File.11 Published by the Health Resources and Services backward stepwise selection with a univariate Po0.15 for inclusion into Administration of the US Department of Health and Human Services, the the models. Allergist--immunologist density was manually inserted into the ARF is a collection of data from over 50 sources, including the American final model to control for changes in cancer mortality that could potentially Medical Association, American Hospitalization Association, US Census, and be attributed to a high overall specialist density. Statistically insignificant National Center for US Health Statistics. The ARF aggregates information allergist--immunologist density in the final model would suggest changes concerning healthcare professionals, healthcare facilities and population in cancer mortality were likely unattributed to high overall specialist for each county in the United States, and includes the number of density. To account for potential geographic variations in prostate cancer specialists within each county based on data from the American Medical screening, prostate cancer incidence among counties in the United States Association Physician Masterfile. In addition to physician data, geographic, was also included in the model. Statistical significance for the final models health system and demographic data for each county were collected from was determined at Po0.05. Variance inflation factors were used to control the ARF in effort to account for ecological characteristics that could for excessive collinearity amongst variables. Percent changes in mortality potentially influence cancer outcomes. Prostate cancer mortality and were calculated for each density category using the cancer mortality incidence data were obtained from a merged data set from the National of a reference group. The reference group in all three models was a Cancer Institute’s Surveillance, Epidemiology and End Results program, county with no radiation oncologists, urologists or allergists. To evaluate National Program for Cancer Registries and United States Center for incremental benefits derived from increasing radiation oncologist density Disease Control and Prevention’s National Vital Statistics System.12 beyond the reference group, linear combination estimates were calculated comparing radiation oncologists among different density categories. Statistical analysis was performed using Stata version 9.2 (Stata, College Construction of variables Station, TX, USA). In addition to radiation oncologist densities, primary care physician and urologist densities were also obtained because of their previously described influence on prostate cancer screening and treatment out- RESULTS comes.4,13 Primary care physicians were defined as those trained in general Radiation oncologist density was heterogeneously distributed practice, family practice and general internal medicine. In an effort to throughout the United States, with 1616 (65.3%) counties lacking assess for reductions in prostate cancer mortality related non-specific the presence of a radiation oncologist. (Figures 1 and 2) Of the specialist density, a non-oncology specialty, Allergy-Immunology, was 2472 counties studied 1205 (48.7%) were without the presence of chosen as a control variable to test whether changes in cancer mortality both a radiation oncologist and urologist. The mean radiation were more specifically attributed to radiation oncologist density. oncology and urologist densities among counties were 0.65 Allergists--immunologists were chosen a priori as an ideal comparison and 2.00 per 100 000, respectively. The baseline prostate cancer specialist, because they have no known associations to prostate cancer mortality in a county without the presence of a radiation oncolo- management and possess a workforce size and geographic distribution gist, urologist or allergist was 38.68 deaths per 100 000 people similar to that of radiation oncology. Physician densities were calculated as (95% confidence interval (CI): 33.94--43.42). The presence of a 5-year means (2002--2006) of physicians per 100 000 people using annual radiation oncologist was associated with a statistically significant Census county population estimates. reduction in prostate cancer mortality, despite adjusting for Mortality and incidence data were reported as age-adjusted average variations in urologist and allergist/immunologist density, as well rates per 100 000 people from the years 2002 to 2006. Cancer incidences as socioeconomic, demographic and health system characteristics and mortalities were assigned to counties based on each patient’s (Table 1). Compared with counties without radiation oncologists, residence at the time of diagnosis and death, respectively. having 40--1, 1--2 or 2--4 radiation oncologists per 100 000 Geographic data collected from the ARF consisted of metropolitan and people, significantly reduced prostate cancer-specific mortality 2004 Primary Care Health Professional Shortage Areas classifications. (PCSM) by À3.65% (95% CI: 5.54--1.76% reduction, P ¼ 0.031), Metropolitan status was defined using 2003 Department of Agriculture 5.74% (95% CI: 7.87--3.61% reduction, Po0.001) and 1.48% (95% Rural/Urban Continuum Codes. To adjust for variations in health system CI: 2.73--0.23% reduction, P ¼ 0.045), respectively. The PCSM of resources across the United States, hospital characteristics were also patients residing in a county with four or more radiation collected. Hospital data from the ARF consisted of mean hospital beds per oncologists per 100 000 people was not significantly different 100 000 people for the years 2002--2006, and 2006 numbers of hospitals from counties where there were no radiation oncologists with radiation therapy services, general oncology services and intensity- (P ¼ 0.769). The confidence intervals of this group were wide, modulated radiation therapy. given the small number of counties with four or more radiation As socioeconomic and demographic factors have been shown to be oncologists per 100 000 residents (Figure 3). Linear combination associated with cancer outcomes, population data from each county were estimates found increasing radiation oncologist density beyond also collected.14,15 Population data from the ARF consisted of, percent 1.0 per 100 000 provided no statistically significant incremental population Caucasian, percent population aged 65 years or older based on reductions in prostate cancer mortality compared with having the Census County File, percent population with health insurance based on 0.1--1.0 radiation oncologists per 100 000 (Table 2). the 2000 Small Area Health Insurance Estimates, percent population with at Consistent with previous studies, increasing density of urolo- least a high school education based on the 2000 Census, median household gists also was associated with a reduction in prostate cancer income based on the 2007 Census Small Area Income Poverty Estimates mortality. Allergist/immunologist density was not related to and unemployment rate based on the 2007 Bureau of Labor Statistics. prostate cancer mortality (P ¼ 0.340). Residing in a more affluent

Prostate Cancer and Prostatic Diseases (2012), 391 -- 396 & 2012 Macmillan Publishers Limited Radiation oncologist density and prostate cancer mortality S Aneja and JB Yu 393

Figure 1. Average radiation oncologist density among United States counties: 2002--2006.

1800 hospitals with radiation therapy, intensity modulated radiation 1616 1600 Radiation therapy or general oncology services either did not meet Oncology 1400 the univariate Po0.15 inclusion criteria or were not significant 1197 Urology 1200 in the multivariate model. Univariate analysis is included as 1000 Supplementary Table 2. 800

Counties 510 600 465 330 DISCUSSION 400 240 202 231 99 The geographic distribution of the radiation oncology workforce 200 56 0 is associated with differences in prostate cancer mortality. The presence of a single radiation oncologist in a county was 0 per 00 00 >4.0 per00 associated with a statistically significant reduction in PCSM. The >0-1.0 per 100,000 100,0 >1.0-2.0100,000 per >2.0-4.0100,0 per 100,0 improvement in prostate cancer mortality persisted even when adjusting for geographic variations in prostate cancer incidence, Figure 2. Distribution of radiation oncologists and urologists among other primary and specialist physician densities, socioeconomic counties in the United States. factors and health system resources. Interestingly, incremental increases of radiation oncologists in a county did not yield incre- mental benefits in outcomes, suggesting a ‘plateau effect’ when a region becomes saturated with radiation oncologists. These county with a younger and more educated population was results corroborate with similar studies that found diminishing correlated with reduced prostate cancer mortality (Table 1). returns with increases in physician supply.5,16 Moreover, racial makeup of counties was found to be associated Radiation therapy is one of the most common treatments for with variations in prostate cancer mortality. Counties with higher prostate cancer. Although our study does not directly test the percentages of Caucasians were associated with reduction in clinical implications of radiation therapy and prostate treatment, prostate cancer mortality (Po0.001). Percent population insured, it does highlight a potential association between the availability and health system characteristics such as hospital bed density and of radiation therapy and improved prostate cancer outcomes.

& 2012 Macmillan Publishers Limited Prostate Cancer and Prostatic Diseases (2012), 391 -- 396 Radiation oncologist density and prostate cancer mortality S Aneja and JB Yu 394 Table 1. Predictors of prostate cancer mortality 2

0 % Change in P 95% CI prostate cancer -2 mortality

Radiation oncologists per 100 000 people -4

40--1.0 vs 0 À3.65 0.031 À5.54 À1.76 Mortality 41.0--2.0 vs 0 À5.74 o0.001 À7.87 À3.61 -6 42.0--4.0 vs 0 À1.48 0.045 À2.73 À0.23 44.0 vs 0 À1.09 0.769 À3.69 1.51 -8 % Reduction in Prostate Cancer Urologists per 100 000 people -10 0 >0-1.0 >1.0-2.0 >2.0-4.0 >4.0 40--1.0 vs 0 À11.52 o0.001 À16.43 À6.60 41.0--2.0 vs 0 À12.71 o0.001 À16.41 À9.02 Radiation Oncologist Density Categories (per 100,000) 42.0--4.0 vs 0 À11.63 0.001 À14.76 À8.51 o Figure 3. Reduction in prostate cancer mortality for given radiation 44.0 vs 0 À12.00 0.001 À15.64 À8.35 o oncology density. Allergist-- À0.59 0.340 À1.80 0.62 immunologist density Percent population 1.23 0.001 0.51 1.94 aged 65 years or older Table 2. Incremental benefit of radiation oncology density on Median household À2.97 o0.001 À4.17 À1.76 prostate cancer mortalities income (in $10 000) Percent population À0.17 o0.001 À0.01 À5.14 Radiation oncology density Prostate cancer mortality with high school categories education Percent population À0.39 o0.001 À0.47 À0.31 % Change in Caucasian mortality P OR Prostate cancer 0.05 0.007 0.01 0.09 incidence 1.1--2.0 vs 0.1--1.0 À2.1 0.221 0.45 2.1--4.0 vs 1.1--2.0 4.3 0.190 1.81 Prostate 95% CI 44.0 vs 2.1--4.0 0.4 0.912 1.16 cancer deaths Abbreviation: OR, odds ratio. per 100 000 people Reference group: 38.68 33.94 43.42 prostate cancer mortality in a county which carries a worse prognosis.17 Furthermore, when adjusting with no radiation for other factors known to influence cancer outcomes, such as oncologists, urologists or allergists socioeconomic factors (median household income, population education level) and health system resources (hospital bed Abbreviation: CI, confidence interval. density, radiation therapy-equipped hospitals, intensity-modu- *Note: Primary care physician density, HSPA service area status, lated radiation therapy-equipped hospitals and general oncology- metropolitan county status, hospital bed density, radiation therapy equipped hospitals) the presence of radiation oncologists was equipped hospital density, general oncology services equipped hospital associated with prostate cancer mortality, highlighting the relative density, IMRT hospital equipped density, percent population insured and unemployment rate either did not meet univariate inclusion criteria or uniqueness of radiation oncologist to prostate cancer manage- were not statistically significant in multivariate model. ment. Our analysis found increasing urologist density to have a more profound effect on PCSM compared with radiation oncologist density. The reasons for this are likely multifactorial. As urologists Radiation oncologist density may be a surrogate for specialized are involved in the screening and diagnosis of prostate cancer in oncology care, specifically the presence of large cancer centers addition to surgical treatment, increased density of urologists may with multidisciplinary tumor boards and a variety of non-radiation be associated with increased likelihood to be diagnosed at an prostate cancer specialists. Although improved outcomes cannot earlier stage with a more favorable prognosis. Conversely, definitively be attributed to the presence of a radiation oncologist, radiation oncologists serve only a curative role in prostate cancer our findings of improved prostate cancer mortality, despite treatment and require another clinical provider, typically an adjustment for urologist and allergist/immunologist density, speak urologist or primary care physician, to refer a prostate cancer to the robustness of the specific association between radiation patient to them. In counties with poor primary care and urologist oncologist density and prostate cancer mortality. Another possible presence, radiation oncologists may be faced the challenge of explanation for our findings is that radiation oncologists are a treating higher stage prostate cancers that were not screened and proxy for other general oncology indicators that could not be fully thus diagnosed later. adjusted for in our multivariate model. For example, the presence We confirmed a previously described geographic maldistribu- of a large cancer center with access to more advanced treatment tion of radiation oncologists across the United States.6 Large technologies and multidisciplinary tumor boards to better segments of the population live in counties without a radiation coordinate prostate cancer care between urologists and radiation oncologist, and this in turn is associated with increased prostate oncologists. Our study also highlighted previously cited racial cancer mortality. To complicate the problem, creating an disparities in prostate cancer outcomes. Counties with larger equitable distribution of radiation oncology services remains proportions of Caucasians were associated with increased prostate difficult. Unlike many other medical specialties, radiation oncol- cancer mortality reduction. This is likely because minorities have ogists require significant equipment to provide treatment, making been shown to present with more advanced prostate cancer, radiation therapy relatively insular to increasingly popular

Prostate Cancer and Prostatic Diseases (2012), 391 -- 396 & 2012 Macmillan Publishers Limited Radiation oncologist density and prostate cancer mortality S Aneja and JB Yu 395 telemedicine initiatives. Additionally, the large investment analysis is subject to a theoretical lead-time bias associated with required to start a radiation oncology practice may contribute patients in underserved areas potentially being diagnosed later to apprehension of radiation oncologists entering the field to and with more advanced disease. This is due to a lack of available establish a practice in an underserved area. county-level stage and grade data. This limitation is present in The mechanism by which radiation oncologist density is related studies of this nature4,10 and was somewhat mitigated by to prostate cancer mortality is difficult to pinpoint. Prior work has adjusting for geographic variations in prostate cancer incidence. shown that radiation oncologist and urologist density was not Finally, given the relative slow growth of prostate cancer, current predictive of whether patients receive any curative therapy. mortality is likely related to treatment options available upwards Rather, individual patient characteristics, such as marital status, are of 10-year prior. A multivariate model replacing current physician predictive of receipt curative treatment.18 However, subsequent densities with physician densities from 1995 yielded similar analysis suggests radiation oncologist and urologist density are results (Supplementary Table 1). This is likely because the predictive of whether patients initially choose radiation therapy geographic maldistribution of the radiation oncology and urology or surgery as the treatment modality for prostate cancer.19 It is workforces have remained relatively unchanged within the last plausible that regional physician density is related to aspects of 15 years.5,6 The relationship between historical health system management about which patients are less informed, such as resources and long-term outcomes of a population in the years nuanced treatment modalities, but decisions related to larger following is an interesting topic for future studies. Nevertheless, questions of whether to receive curative treatment are ultimately previously published studies of our similar nature,4 our study is personal ones. Furthermore, as geographic differences in prostate fundamentally an analysis of the current variations in healthcare cancer mortality have mainly been attributed to differences in systems across the United States. Using current physician densities disease stage related to time of diagnosis, an aspect of care with as a proxy for current general oncology infrastructure and allows which radiation oncologists are not typically involved, the density the study the relative robustness of current healthcare systems. of radiation oncologists may be of less utility.2 Perhaps it is the Our analysis ultimately comments less on the well-established radiation oncologist’s role in providing truly multidisciplinary curative relationship between urologists and radiation oncologists cancer care that most influences mortality. Where previous for prostate cancer, rather more on the relationship between investigators found an association between urologist density resource rich/poor regional healthcare systems and overall and prostate cancer mortality, we found a similar association for prostate cancer outcomes. In spite of these limitations, our radiation oncologists, even when adjusting for the presence of analysis provides a first step in understanding the relationship urologists and variations in prostate cancer incidence.5 Therefore, between variations in the radiation oncologist workforce and our two studies in combination highlight the importance of prostate cancer mortality and serves as an impetus for further multidisciplinary care in the management of patients with study of the effect of regional radiation oncologist therapy prostate cancer. resources in the management of prostate cancer. We found that the improvement in prostate cancer mortality did plateau beyond a radiation oncologist density of 1.0 per 100 000. Reasons for this diminishing return may be because incremental improvement in prostate cancer mortality when CONCLUSION comparing higher density categories was small relative to the Radiation oncologist density is related to geographic variations in large improvement in mortality from the addition of the first prostate cancer mortality among counties in the United States. radiation oncologist to a county. This plateau effect has been seen The lack of radiation oncologists among less populated areas is 10,16,20 in other studies of the physician workforce. Moreover, associated with increased rates of prostate cancer mortality. prostate cancer can be relatively indolent, in contrast to cervical Conversely, saturation of radiation oncologists within a region cancer or head and neck cancers. Patients residing in areas with provides diminishing reduction to prostate cancer mortality. an oversubscribed radiation oncologist can potentially wait until Further studies are needed to better understand the relationship prostate cancer treatment is available, perhaps mitigating the between variations in the radiation oncology workforce and need for additional radiation oncologists to improve time prostate cancer management. Policy attempting to decrease both between diagnosis and treatment. prostate cancer mortality and optimize the radiation oncology Our study was limited for several reasons. As the American workforce must be geographically aware in order to prevent Medical Association Masterfile records only the primary location of further disparities in outcomes. a radiation oncologist, we did not capture the availability of physician services at secondary ‘satellite’ locations. This limitation is somewhat attenuated in our analysis given the decreased prevalence of satellite radiation oncology centers because of large CONFLICT OF INTEREST required investment to establish satellite practices. We were The authors declare no conflict of interest. unable to assess the impact of the availability of various radiation treatment modalities (EBRT, brachytherapy, SBRT), as no informa- tion was recorded as to the type of treatment available at each practice location. Another limitation of our study is the exclusion REFERENCES of rural counties within the United States for which much of the 1 Siegel R, Naishadham D, Jemal A. Cancer statistics, 2012. CA Cancer J Clin 2012; 62: prostate cancer mortality data were unavailable. Nevertheless 10--29. even if that data were available, the paucity and relative 2 Jemal A, Ward E, Wu X, Martin HJ, McLaughlin CC, Thun MJ. Geographic patterns maldistribution of radiation oncologists and urologists among of prostate cancer mortality and variations in access to medical care in the United rural counties would make it difficult to establish any reliable States. Cancer Epidemiol Biomarkers Prev 2005; 14: 590--595. relationship between physician density and prostate cancer 3 Smith BD, Haffty BG, Wilson LD, Smith GL, Patel AN, Buchholz TA. The future of outcomes. What remains clear, however, is the variation in radiation oncology in the United States from 2010 to 2020: will supply keep pace prostate cancer outcomes in resource poor and resource rich with demand? J Clin Oncol 2010; 28: 5160--5165. 4 Odisho AY, Cooperberg MR, Fradet V, Ahmad AE, Carroll PR. Urologist density and regions. Further studies focused on rural counties will be needed county-level urologic cancer mortality. J Clin Oncol 2010; 28: 2499--2504. to better generalize the results of our analysis to rural areas. 5 Odisho AY, Fradet V, Cooperberg MR, Ahmad AE, Carroll PR. Geographic Additionally, patient-specific characteristics were unattainable distribution of urologists throughout the United States using a county level using the merged State Cancer Profiles data. Additionally, our approach. J Urol 2009; 181: 760--765; discussion 5--6.

& 2012 Macmillan Publishers Limited Prostate Cancer and Prostatic Diseases (2012), 391 -- 396 Radiation oncologist density and prostate cancer mortality S Aneja and JB Yu 396 6 Aneja S, Smith BD, Gross CP, Wilson LD, Haffty BG, Roberts K et al. Geographic 14 Gross CP, Smith BD, Wolf E, Andersen M. Racial disparities in cancer therapy: did analysis of the radiation oncology workforce. Int J Radiat Oncol Biol Phys 2012; 82: the gap narrow between 1992 and 2002? Cancer 2008; 112: 900--908. 1723--1729. 15 Du XL, Lin CC, Johnson NJ, Altekruse S. Effects of individual-level socio- 7 Starfield B, Shi L, Grover A, Macinko J. The effects of specialist supply on economic factors on racial disparities in cancer treatment and survival: findings populations’ health: assessing the evidence. Health Aff 2005; Suppl Web from the National Longitudinal Mortality Study, 1979--2003. Cancer 2011; 117: Exclusives:W5-97--W5-107. 3242--3251. 8 Phillips Jr RL, Dodoo MS, Green LA. Adding more specialists is not likely to 16 Krakauer H, Jacoby I, Millman M, Lukomnik JE. Physician impact on hospital improve population health: is anybody listening? Health Aff 2005; Suppl Web admission and on mortality rates in the Medicare population. Health Serv Res Exclusives: W5-111--W5-4. 1996; 31: 191--211. 9 Onega T, Duell EJ, Shi X, Wang D, Demidenko E, Goodman D. Geographic access 17 Pruthi RS, Tornehl C, Gaston K, Lee K, Moore D, Carson CC et al. Impact of race, to cancer care in the U.S. Cancer 2008; 112: 909--918. age, income, and residence on prostate cancer knowledge, screening behavior, 10 Aneja S, Bordeaux JS. Association of increased dermatologist density with lower and health maintenance in siblings of patients with prostate cancer. Eur Urol melanoma mortality. Arch Dermatol 2012; 148: 174--178. 2006; 50: 64--69. 11 US Department of Health and Human Services, Health Resources and Services 18 Aneja S, Makarov D, Gross C, Roberts K, Yu JB. The influence of physician densities Administration: Area Resource File (ARF): National county-level health resource and patient characteristics on the decision to treat prostate cancer patients with information database. http://www.arfsys.com. varying clinical benefit. Int J Radiat Oncol Biol Phys 2011; 81: S553. 12 McCahill LE, Single RM, Aiello Bowles EJ, Feigelson HS, James TA, Barney T. et al. 19 Aneja S, Gross C, Makarov D, Roberts K, Yu JB. The influence of regional radiation Variability in reexcision following breast conservation surgery. JAMA 2012; 307: oncologist and urologist capacities on treatment choice for prostate cancer. Int J 467--475. Radiat Oncol Biol Phys 2011; 81: S552. 13 Jang TL, Bekelman JE, Liu Y, Bach PB, Basch EM, Elkin EB et al. Physician visits prior 20 Goodman DC, Fisher ES, Little GA, Stukel TA, Chang CH, Schoendorf KS. The to treatment for clinically localized prostate cancer. Arch Intern Med 2010; 170: relation between the availability of neonatal intensive care and neonatal 440--450. mortality. N Engl J Med 2002; 346: 1538--1544.

Supplementary Information accompanies the paper on the Prostate Cancer and Prostatic Diseases website (http://www.nature.com/pcan)

Prostate Cancer and Prostatic Diseases (2012), 391 -- 396 & 2012 Macmillan Publishers Limited