| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Health Policy and Practice |
1 From the Research Department, American College of Radiology, 1891 Preston White Dr, Reston, VA 20191 (M.B., J.H.S.) and the Department of Diagnostic Radiology, Yale University School of Medicine, New Haven, Conn (J.H.S.). Received July 28, 2004; revision requested October 6; revision received November 16; accepted December 21. Address correspondence to M.B. (e-mail: mythreyib{at}acr.org).
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONReferences |
|---|
MATERIALS AND METHODS: Non–individually identified data from the American College of Radiology (ACR) 2003 Survey of Radiologists were compared with data from previous ACR surveys; all statistics were nationally representative. Workload according to individual practice characteristics, such as size, type, location, and setting, was tested for statistically significant differences from the overall average. Time trends and the independent effect on workload of practice characteristics were measured with regression analysis. Changes in average procedure complexity were calculated in physician work relative value units (RVUs) per Medicare procedure.
RESULTS: In 2002–2003, the average workload per full-time equivalent (FTE) radiologist was 13 900 procedures annually (standard error of mean, 200), an increase of 8.1% since 1998–1999 (P < .05) and 25.1% since 1991–1992 (P < .01). Academic practices performed 9900 procedures per FTE radiologist, and private radiology practices performed 15 200 procedures per FTE radiologist. Within most practice categories, radiologists at the 75th percentile of workload typically performed at least 50% more procedures than radiologists at the 25th percentile. Average physician work RVUs per Medicare procedure increased by 6.2% between 1998 and 2002 and by 21.6% between 1992 and 2003, mainly because of an increase in the share of more complex techniques such as magnetic resonance imaging and computed tomography in the procedure mix.
CONCLUSION: Workload per radiologist measured in procedures and RVUs increased steadily between 1991–1992 and 2002–2003. Because there is much unexplained variation, averages or medians should not be used as norms.
© RSNA, 2005
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONReferences |
|---|
The most recent data on workload that were previously available, dating from 1998–1999, showed that annual workload per full-time equivalent (FTE) radiologist had increased approximately 8.5% in terms of procedures and 4% in terms of relative value units (RVUs) per procedure over the preceding 3 years (1995–1996 through 1998–1999) (3). In keeping with the notion of imbalance, findings about the radiologist supply situation in that period indicated a serious shortage of radiologists (4–8). In contrast, results of the most recent study (9) about the radiologist supply situation indicate that the shortage "has considerably eased." That raises the question of whether workload has continued to increase or whether it has declined. In particular, an issue of concern for radiologists is whether an easing of the shortage implies a decrease in future work and income for radiologists.
The American College of Radiology (ACR) periodically conducts large-scale, general-purpose surveys as part of its mission of providing important and useful information to diagnostic radiologists, radiation oncologists, interventional radiologists, and medical physicists. Because of the importance of workload issues, these surveys devote careful attention to workload issues.
Thus, the purpose of this study was to measure the workload of radiologists in the United States in 2002–2003, variations in workload according to practice characteristics, and trends since 1991–1992 (3,10,11).
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONReferences |
|---|
The survey sample—a stratified random sample composed of four strata—was taken primarily from the American Medical Association Physician Masterfile (14), a reasonably complete listing of all allopathic physicians in the United States, whether or not they are American Medical Association members. The sample from the Masterfile consisted of a 16% sample of all those physicians self designated in the Masterfile as vascular/interventional radiologists, an 8% sample of all other radiologists, and an 8% sample of nuclear medicine specialists.
The percentages were selected to ensure adequate numbers of observations for statistically meaningful inferences on certain demographics of interest, given the assumption of a 65% expected response rate. For instance, we were interested in comparing academic radiologists by sex, and to obtain a sufficient number (ie, approximately 50) of responses from women at a 65% response rate, we needed an 8% sample of all diagnostic radiologists. The survey had a double-sized (ie, 16%) sample of interventional radiologists because this survey had a special focus on studying interventional radiologists after their official recognition as "a new component of the specialty of radiology" at the 2001 Annual Meeting of the ACR (15).
The sample included residents, fellows, and retirees, not merely posttraining professionally active physicians. In addition, the sample included 92 osteopathic radiologists selected at random by the American Osteopathic College of Radiology, or AOCR, from among its members. This sample constituted an approximately 6.7% sample of all osteopathic radiologists in the United States, including those who were not AOCR members (Pamela A. Smith [Executive Director, AOCR], written communication, 2003).
In March 2003, our contractor, the Center for Survey Research of the University of Virginia, mailed the survey. There was a statement on the survey cover sheet that responses would not be individually identified, and responses were processed by the Center for Survey Research for enhanced assurance of confidentiality. Nonrespondents were sent up to four remailings of the survey as necessary, at approximately monthly intervals, in addition to other reminders. The last remailing took place in mid-July; acceptance of responses ended a month later.
As in previous ACR surveys, we were interested in only nuclear medicine specialists who had major ties to radiology, and we operationalized this concept of a major tie to radiology as consisting of holding American Board of Radiology certification and/or being a member of the ACR (8). On this basis, we omitted from consideration approximately two-thirds of the original sample of nuclear medicine specialists, which left us with a sample of 53 nuclear medicine physicians to whom the survey was sent.
The total sample of interest, which was composed of four strata—interventional radiologists, osteopathic radiologists, nuclear medicine specialists of interest, and all other radiologists—consisted of 3090 physicians. From these, we received 1924 usably complete responses. In addition, we received information (not in the form of completed questionnaires) that 21 addressees were deceased, six were no longer practicing in the United States, and six were not radiologists (and, for the purposes of calculating a response rate, were equivalent to respondents who told us in the first question that they were not radiologists). The response rate was thus [(1924 + 6)/(3090 – 21 – 6)] · 100 = 63%.
Response Weighting
Responses were weighted so that the weighted statistics would be representative of the answers that would have been received if all radiologists and nuclear medicine physicians of interest in the United States had been surveyed and had responded. The weighting process, as applied to previous surveys, has been described previously (8): Logistic regression analysis was employed to determine how many different sets of response weights were to be used in each of the four strata. We identified 10 weighting categories among diagnostic radiologists, based on whether or not a physician was an ACR member and his or her age; two weighting categories for interventional radiologists, based on whether or not the physician was an ACR member; and one weighting category each for nuclear medicine specialists of interest and osteopathic radiologists. After all responses in each weighting category were given a weight equal to the reciprocal of the response rate for that category, these weights were multiplied by the reciprocal of the sampling rate to complete the process of making responses representative of the entire U.S. population of radiologists.
For example, if a weighting category had a response rate of 65% and it was part of a stratum that had been sampled at the general 8% sampling rate, then all responses in that weighting category were given a weight of (1/0.65) · (1/0.08) = 19.23.
Data Quality Improvement
Every survey has some deficient data—that is, missing items, responses not in accordance with directions given in the questionnaire, and responses that are inconsistent with other responses or have other problems. Our leading tool for minimizing data deficiencies was the designation of the 12 items on the questionnaire that were judged to be most crucial as "core questions." When questionnaires were returned, the Center for Survey Research checked that these 12 items were indeed answered and made three designated consistency checks involving them. If there were any problems with the core items, the Center for Survey Research tried to telephone the respondent to obtain the missing response(s) and/or resolve the consistency problems.
During the data entry process, the Center for Survey Research spot-checked one of every six entered questionnaires against the paper questionnaires and found an error rate of less than 0.1%. Judging this error rate to be satisfactory, we did not have the data double entered.
Data used in this report were additionally cleaned and edited to further minimize deficiencies. Items with relatively extensive cleaning and editing that are relevant to this report are as follows:
Survey respondents were asked to report the number of procedures performed per year in their main practice, the numbers of full-time and part-time radiologists in the practice, and the typical number of hours worked per week by each of those categories of radiologists. The count of FTE radiologists per practice was calculated by using the ratio of typical part-time to typical full-time hours in each practice to obtain the full-time equivalence of part-time workers and adding that number to the number of full-time radiologists in the practice. For responses in which data on typical hours were missing (4% of the sample) or in which the number of reported part-time hours exceeded the number of full-time hours (2.2% of the sample), we used the all-practice median ratio of full-time to part-time hours (calculated from the sample to be 0.5) as the ratio for "converting" part-time to FTE radiologists.
The number of procedures per FTE radiologist was calculated as the ratio of reported procedures for the practice to the calculated number of FTE radiologists. So that we could avoid distortions of results because of outliers that possibly represented erroneous responses, and so that we could compare results with findings from previous studies, we used a previously employed method (3) of eliminating responses that reported fewer procedures per FTE radiologist than one-third the median value across all responses (ie, 12 609 [all values are rounded to the nearest whole number]) or more numerous than three times the median value. This eliminated slightly more than 10% of responses with very low reported average numbers of procedures per FTE (ie, less than 4203 procedures) and slightly more than 1% of responses with very high reported average numbers of procedures per FTE (ie, more than 37 826 procedures). In ignoring these observations entirely, we implicitly assume that the distribution of the "true number of procedures" underlying these erroneous responses is similar to the distribution of the responses that we do include and that are correct.
Definition of Variables
The survey asked respondents the location of their main practice according to the following six categories: main city of a large metropolitan area (total area population of 1 million or more), suburb of a large metropolitan area, main city of a smaller metropolitan area (total area population of >50 000 but <1 million), suburb of a smaller metropolitan area, nonmetropolitan location (total area population of 50 000 or less or rural location), and "varied locations" (no one location is principal). We report location according to these six categories.
Somewhat similarly, we asked respondents "Which best describes your main practice?" and provided seven answer options: solo practice; locum tenens; primarily academic practice (owned by any agency or variety of agencies); private, multispecialty practice that is not primarily academic; government practice, not primarily academic; private radiology, nuclear medicine, and/or interventional radiology practice, not primarily academic; or other (specify). We report practice type according to these seven categories.
The definition of other variables is self-evident. Most other variables were reported on the 2003 survey, but some were self-reported to the Masterfile—for instance, age and state. States are grouped into the four census regions—Northeast, Midwest, South, and West—for reporting statistics.
Additional Data
To compare trends in workload over time, we used previously collected data on workload for 1998–1999 (from the ACR 1999 Survey of Practices) and 1995–1996 (from the ACR 1996 Survey of Hiring by Groups) and previously published data for 1991–1992 (11).
The relative complexity of procedures was measured by using a variation of a previously employed method (3,10) of comparing average RVUs per procedure in Medicare with the Medicare Physician/Supplier Procedure Summary (formerly Part B Medicare Annual Database, or BMAD) master files for 1992, 1995, 1998, and 2002 (16). These files contain summarized information for all Medicare claims for non–managed care enrollees. There is no patient-specific information in these files. The data are aggregated on the basis of claim characteristics, such as procedure code, code modifiers, place of service, and provider specialty code. The analysis was restricted to procedures performed by radiologists.
We used the American Medical Association's Current Procedural Terminology, or CPT (17), codes and an ACR algorithm that identifies imaging procedures by modality with the CPT codes (which, for the procedures of interest in this article, are equivalent to the Healthcare Common Procedure Coding System, or HCPCS codes, used by Medicare). We used physician work RVUs assigned to each procedure that were reported in Medicare's National Physician Fee Schedule (18). For interventional radiology procedures that use a supervision and interpretation, or S&I, code along with surgical code(s) to represent a procedure, we developed empirical estimates of physician work RVUs that represented a "complete procedure" on the basis of results of analysis of year 2001 claims for a 5% sample of Medicare enrollees. (The Medicare summarized data files do not identify the surgical procedures that were performed along with the radiology S&I code, but these surgical codes are part of the radiologist's work. Using the RVUs for the radiology S&I procedures alone underestimates radiologists' work on interventional procedures. Therefore, we estimated a "complete procedure" RVU for each interventional S&I code on the basis of the average RVUs per episode with each S&I code.)
RVUs vary across procedures according to the amount of work involved in interpreting the image. Although this often results in RVUs that are proportionate to the time spent on the procedure, there are some modalities for which the RVUs per unit of time spent in the procedure are larger than those for other modalities. This is deliberate. The RVUs are intended to reflect both time required for a procedure and "intensity," the number of RVUs per unit of time, which reflects such elements as mental effort and stress. So that we could compare workload per FTE radiologist more in terms of the amount of time spent per procedure, we also calculated time-adjusted RVUs per FTE by using adjustment factors that were used by Arenson et al (19)—0.58 for angiography, 0.5 for computed tomography (CT) and magnetic resonance (MR) imaging, and 1 for all other modalities. These adjusted RVUs are, in large part, a measure of time.
Statistical Analysis
Data analyses were performed with SAS software, release 9.0 (SAS Institute, Cary, NC), and SUDAAN 8.2 (Research Triangle Institute, Research Triangle Park, NC).
All analysis was restricted to posttraining respondents who reported that they were working full-time or part-time in radiology at the time the survey was conducted. Thus, all residents, fellows, retirees, or radiologists otherwise temporarily or permanently not working in radiology were excluded from all analysis.
All information presented is based on weighted data and thus is representative of what responses would have been if all radiologists in the United States had been surveyed and had responded. Reported standard errors of the mean (SEMs), standard errors of regression coefficients, and tests of statistical significance, which were based on SEMs and standard errors of regression coefficients, were calculated by taking into account not only the weighted nature of the data but also the complex survey design—that is, the fact that responses came from distinct strata. The standard errors were calculated with survey-specific SAS software procedures (proc surveymeans and proc surveyreg) and verified by using SUDAAN.
Standard errors are not only used in calculating the statistical significance of differences observed when making comparisons but are also the most common measure of sampling variability. (Sampling variability is the phenomenon that, in general, a statistic—such as a mean—from a sample will differ somewhat from the same statistic for the entire underlying population from which the sample is drawn.) There is a 95% probability that the true value of a statistic for an entire population lies within approximately 2 standard errors of the corresponding statistic for a sample drawn from that population.
We (M.B., J.S.) calculated average annual workload per FTE radiologist (and associated standard errors) for each practice type. The survey had a relatively large number of responses from radiologists in private nonacademic radiology practices and in academic practices. For each of these two practice types, we therefore also calculated average annual workload per FTE radiologist for categories based on group size, settings served (hospital only, nonhospital only, or both), region (Northeast, Midwest, South, or West), and location type (eg, main city of a large metropolitan area, suburb of a large metropolitan area). Practice characteristics tend to be correlated with each other; for example, academic practices are, on average, larger than private practices and are unlikely to be in rural locations. Calculating the average workload according to practice characteristics separately for academic and private practices in the descriptive analyses partly controlled for such correlations. So that we could explain differences in workload between academic and other practices, we calculated the average percentage of time that radiologists reported that they spend on teaching and research.
So that we could measure variations in annual workload per FTE radiologist for each subgroup used in the analyses, we also calculated quartiles (25th percentile, median, and 75th percentile).
To keep the comparison homogeneous, we based all calculations on responses for which we had information on all the relevant variables. This avoided, for example, a response being in the analysis based on census region but being omitted from the analysis based on settings served because information on the settings served was missing.
We used two-tailed tests of comparisons of means to compare averages in 2002–2003 with averages in 1998–1999, 1995–1996, and 1991–1992.
To measure imaging complexity, we calculated, for each modality, the physician work RVUs per procedure and the relative share in workload among all Medicare imaging procedures performed by radiologists. We calculated these separately for offices and hospitals to ascertain if there was a difference according to site of service. For comparison, we also calculated time-adjusted RVUs per procedure.
In addition to calculating descriptive statistics, we used multivariable least squares regression analyses to estimate the separate effect of each practice characteristic on a practice's annual workload while statistically controlling for the effects of all other practice characteristics being considered. One regression included only data from 2002–2003. In a second regression, we combined survey data from all the years for which data were available: 1995–1996, 1998–1999, and 2002–2003. This allowed us to measure the time trend in workload while controlling for changes over time in practice characteristics. Also, because this second regression analysis involved more observations, it could, in general, reveal smaller effects.
The dependent variable in the regression analysis was the natural logarithm of the annual number of procedures performed in the practice. In the first regression, the independent variables were the natural logarithm of the number of FTE radiologists in the practice and categorical variables describing practice type, census region, settings served, and practice location. The second regression involved all the independent variables considered in the first regression, as well as an independent trend variable denoting the year in which data were collected (with 1996 set to 0, 1997 to 1, and so on) was used.
The unit of measurement in the regression analyses was the practice and not the individual radiologist, whereas the unit of analysis for the descriptive tables was the individual radiologist. Therefore, for the regression analyses, each observation was weighted by using a practice weight. Because each individual represented all the radiologists in his or her practice, the weight assigned to each response was set equal to the person-weight divided by the number of radiologists in the practice. This made the responses representative of practices, not radiologists. Note that for the descriptive statistics on procedures per FTE radiologist, group size categories were defined in terms of number of radiologists in the practice (ie, head count), not number of FTE radiologists. For the regression analysis, the relevant measure of group size was FTE radiologists because the dependent variable used was procedures per practice.
In the regression analyses, all of the categorical variables had a reference category, and the effects of the other values of the variable were measured relative to the effect of the reference category. The reference categories used were the most common values of the variables: private radiology for practice type, South for census region, practicing in both hospital and nonhospital settings for settings served, and main city of a smaller metropolitan area for practice location.
In the regression analyses, we used P < .01 as the test of significance for every individual category (eg, South) and P < .05 to test the significance of the category as a whole (eg, region). For multiple simultaneous comparisons (such as the ones performed while testing the significance of each of the values of a categorical variable, such as region), a more stringent test was required for the significance of each category to guarantee significance of the entire variable. For the regressions, we used SUDAAN to ensure that standard errors of the regression coefficients were calculated with a correction for sample design.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONReferences |
|---|
|
Variation in Workload
Radiologists' workloads varied significantly according to practice characteristics in private (nonacademic, nonmultispecialty, nonsolo) radiology practices (Table 2) and in academic practices (Table 3).
|
|
Among academic practices, the average workload per radiologist for a practice in the 75th percentile was typically at least 60% higher than that in a practice in the 25th percentile of practices with the same characteristic (Table 3). Also, large academic practices performed fewer procedures per FTE radiologist than the average academic practice.
RVUs per Procedure
In 2002, the average number of physician work RVUs per Medicare radiology procedure was 0.65 (Table 4), which represented an increase of 6.2% (or an average compound annual increase of 1.5%) since 1998. Combining this fact with the above findings, we estimate that the average number of physician work RVUs per FTE radiologist in 2002–2003 was 9100 (13 900 · 0.65 = 9100) (Table 5).
|
|
There was no meaningful difference in work RVUs per procedure between hospital and nonhospital settings; the average was 0.67 RVUs per procedure in nonhospital settings and 0.65 RVUs per procedure in hospital settings.
The time-adjusted average RVUs per procedure increased from 0.47 in 1998 to 0.48 in 2002, and, unrounded, this was a 3.5% increase during the period. For the entire period of 1992–2002, time-adjusted RVUs per procedure increased from 0.43 to 0.48 (ie, by 13%).
Trends in Workload from 1992–2003
The average of 13 900 procedures performed per FTE radiologist (SEM, 200) in 2003 was 1100 procedures higher than the average in 1998–1999, 2100 procedures higher than the average in 1995–1996, and 2800 procedures higher than the average in 1992 (Table 5, Fig 1). The difference between 1998–1999 and 2002–2003 was significant at 5%, and the differences between 1995–1996 and 2002–2003 and between 1991–1992 and 2002–2003 were significant at 1%. This represents a total growth in number of procedures per FTE radiologist of 25.1% during 1992–2003 and an average annual compound growth rate of approximately 2.1% (Table 5). For nearly every analyzed category and subcategory of radiologists, there was an increase in average annual number of procedures per FTE radiologist over time (Figs 2–6).
|
|
|
|
|
|
Results of Multivariate Regression Analyses
The results of the regression analysis that involved only 2002–2003 data indicate that for every 1% increase in the number of FTE radiologists, groups performed, on average, 0.962% more procedures; 0.962% was not significantly different from 1% (Table 6). In other words, procedures performed by practices varied, on average, in the same proportion as differences in the number of FTE radiologists. With controlling for all other practice characteristics, academic practices performed 26% fewer procedures per year in 2002–2003 than did private radiology practices. Practices that functioned in only nonhospital settings performed approximately 22% fewer procedures than otherwise identical practices that functioned in both hospital and nonhospital settings.
|
In the regression analyses that included data from three ACR surveys (the 1996 Survey of Hiring by Groups, the 1999 Survey of Practices, and the 2003 Survey of Radiologists), we found that workload increased slightly less than proportionately to the number of FTE radiologists in a practice. Specifically, a 1% larger number of FTE radiologists resulted in a 0.946% larger number of procedures, and this was significantly different from a 1% increase in procedures. In other words, the number of procedures performed by the practice varied slightly less than proportionately to the number of FTE radiologists in the practice. The analysis that included all three surveys revealed that academic practices performed, on average, 25% fewer procedures than otherwise identical private radiology practices. All other things being equal, census region had a marginally significant effect on workload, with practices in the West performing 10% fewer procedures than those in the South. Other characteristics being identical, practices functioning exclusively in nonhospital settings performed 25% fewer procedures than those functioning in both hospital and nonhospital settings. Practice location (eg, large metropolitan area, small metropolitan area) did not have a significant effect on the number of procedures per practice.
With controlling for all practice characteristics, workload for practices (measured in number of procedures while controlling for the number of FTE radiologists in the practice) grew at an average compound annual rate of 3.0% per year between 1995 and 2003, which is larger than the 2.1% overall growth rate in procedures per FTE radiologist shown in Table 5, the data in which do not reflect controlling for practice characteristics.
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONReferences |
|---|
According to American Medical Association data on trends (21), the number of hours per week spent in patient care in diagnostic radiology increased from 55.1 in 1991 and 53.9 in 1992 to 58.5 in 2001. We calculated average growth per year in hours per week for 1991–2001 and 1992–2002 and used these rates to estimate total growth in hours per week for 1991–2002. These data translate into approximately 7%–10% more hours per week spent in patient care in diagnostic radiology in 2002 than in 1991. Average numbers of weeks per year spent in patient care in diagnostic radiology were practically identical—44.2 in 1991, 44.4 in 1992, and 44.2 in 2000. In all, there was an increase of less than 10% in time spent in patient care, but radiologists' work, measured in RVUs, increased 45.9%. This is an indicator that radiologists have become increasingly productive over the 11-year period from 1991 to 2002. (The standard definition of productivity is the amount of work completed per work hour.)
When practice characteristics (including the number of FTE radiologists in the practice) were controlled for, analysis revealed that the mean number of procedures per practice grew at 3.0% per year between 1995–1996 and 2002–2003; this translates to a total growth of 23.1% over the 7-year period. This is larger than the growth in the average number of procedures per FTE radiologist (without controlling for practice characteristics), which was 2.4% per year and 18.1% overall for the same period. In other words, when practice characteristics were controlled for, radiologists' productivity increased even more rapidly than the simple average annual growth rate in workload per FTE radiologist indicates.
When practice characteristics were controlled for, we found that private radiology practices that operated exclusively in nonhospital settings performed significantly fewer procedures than practices that operated in hospitals as well. Our calculations of average RVUs per procedure did not reveal any meaningful differences between hospital and nonhospital settings; therefore, differences in procedure complexity do not explain a substantial part of this difference in average workload between practice settings.
There is recent evidence that points to an easing of the shortage of radiologists (5,7,9). Because workload per radiologist has increased substantially in the past few years, the easing of the shortage is not explained by a declining workload. On the contrary, from workload figures alone, one would expect the shortage to have intensified. This study did not address all the factors that affect radiologists' productivity. It omitted possibly important factors, such as purportedly productivity-enhancing technologies and nighthawk coverage. Nonetheless, we think that it is an important finding that workload per FTE radiologist has grown rapidly in recent years and, thus, changes in the volume of procedures performed cannot account for the easing of the shortage.
This finding is also important because radiologists' share in total imaging workload is decreasing (22). Our findings show that self-referral of the kind in which nonradiologists interpret their own images is not expanding so rapidly that radiologists' workload is decreasing. Rather, as we have shown, radiologists' workload is growing relatively rapidly.
In our study there was wide variation in radiologists' workloads within each subcategory, a finding that is unexplained. Therefore, the means, medians, 25th percentiles, and 75th percentiles reported in this article should not be used as benchmarks or standards. Also limiting their relevance as standards is the fact that our analysis did not account for the other activities of practices. For instance, calculations from this survey indicate that radiologists in academic practices spend approximately one-third of their time on activities such as teaching and research rather than on pure clinical work. This does much to account for the lower average workload of such radiologists in terms of the number of procedures per FTE radiologist.
Comparison of Our Results with Those of Other Studies
Comparison of our results with those of previous ACR studies of workload (3,10) reveals strong agreement. For example, like previous investigators, we found that radiologists at the 75th percentile of workload typically performed 50% more procedures than those at the 25th percentile, and radiologists in academic practices performed approximately one-third fewer procedures than average.
Our study results are in broad agreement with estimates of trends in utilization obtained on the basis of Medicare claims. Overall, the number of imaging procedures per Medicare fee-for-service enrollee increased at a compound average annual rate of 3.1% per year between 1992 and 2001 (1). Subsequent unpublished calculations that disaggregated the data according to physician specialty revealed that Medicare imaging per (non–managed care) enrollee by radiologists grew at 2.2% per year between 1992 and 2002. The number of FTE radiologists grows at approximately 1.6% per year (2), and the population of the United States grew at approximately 1.2% per year between 1992 and 2002 (23). Combining these two pieces of information with the estimated 2.1% increase in the number of procedures per FTE radiologist per year in this survey, we would estimate the per-person increase in imaging by radiologists to be 2.5%, which is similar to the estimated 2.2% annual increase in imaging by radiologists per Medicare enrollee.
Furthermore, estimates of procedure volume derived from this study are broadly in agreement with other estimates of the volume of imaging in the United States. The following comparison demonstrates this agreement: The total number of procedures performed by radiologists as estimated from this survey is approximately 320 million (13 900 procedures per FTE radiologist times 23 000 FTE radiologists). The number of imaging services used per 1000 Medicare fee-for-service beneficiaries in 2001 was 4176 (1). The number of imaging services used by Medicare beneficiaries grew at 4.8% per year between 1998 and 2001. Extending this to 2002–2003, the number of imaging services used per 1000 Medicare fee-for-service enrollees was approximately 4500. Because the majority (approximately 85%) of the Medicare population is elderly (aged 65 years or older), we assume that this average utilization applies to all elderly persons. For just fewer than 36 million elderly persons in 2002–2003 (24), this represents a total utilization of approximately 160 million imaging procedures by the elderly in 2002–2003.
According to our calculations, which are based on Medicare claims data, radiologists perform approximately 60% of all imaging procedures, which translates into 96 million imaging procedures performed by radiologists for those aged 65 years and older. As calculated from published statistics (25), health care expenditures for the elderly constituted approximately 35% of all health care expenditures during the late 1990s. If the ratio of the number of procedures performed in the elderly to the number of procedures performed in all persons is assumed to be the same as the ratio of expenditures, the estimated total number of imaging procedures performed by radiologists in the United States in 2002–2003 was approximately 270 million. Given the uncertainties involved, this is reasonably close to (ie, within less than 20% of) our survey-based estimate of 320 million procedures for 2002–2003.
Study Strengths and Limitations
Like other studies, ours had both strengths and limitations. Major strengths include the following: The data were from a large, carefully conducted survey that achieved a high response rate through intensive follow-up. Weighting adjusted for nonresponse bias—that is, differences between respondents and nonrespondents—in the characteristics used in the weighting (age and ACR membership). Paying careful attention to the completeness of our sample, we included osteopathic radiologists, who constitute approximately 3% of all radiologists, and nuclear medicine specialists with a major connection to radiology. Multiple steps improved data quality.
Agreement of findings from the survey with data from independent information sources was generally good. For example, the survey-based estimate of the number of radiologists who have each of the American Board of Radiology's three Certificates of Added Qualifications does not differ by a statistically significant amount from the true number tabulated by the American Board of Radiology.
Nonetheless, the survey had noteworthy limitations. As with statistics drawn from almost any survey, statistics drawn from it may have inaccuracies from at least three sources: sampling variability (the likely size of these inaccuracies is measured by the standard errors), nonresponse bias (but only with respect to characteristics not considered in the weighting or logistic regressions), and incorrect or illogical responses (some of which still remain despite careful and extensive data cleaning). Most obviously, we had to delete approximately 11% of responses because of implausibly low or high numbers of reported procedures per FTE radiologist. Our cutoff for eliminating outliers was necessarily arbitrary, although it was a formula that we have used before (3). Although our data for 2002–2003 may have some inaccuracies, the same cutoff for deleting outlier responses was applied to previous surveys, so trends should be relatively reliable.
Approximately 28% of radiologists reported working in multispecialty practices in the 2003 survey. This is an unexpectedly large percentage compared with the percentage in the 1999 survey (ie, 6%). We are unable to explain this discrepancy because the wording and order of answer options were comparable between the two surveys.
Most of the definitions in the survey questionnaire were consistent across the years. There was a small change in the definition of a small metropolitan area—in 2003, it was defined as an area with a population of 50000 or more; in previous years, it had been defined as an area with a population of 100000 or more.
We assumed that survey nonresponders had the same distributions of procedures per FTE radiologist as responders. If this is incorrect and the responders did not respond because they were too busy, then our estimates of workload are too low. Again, our measurement of trends should be relatively accurate because if this is a problem, it was present with past surveys and affected them similarly.
If the mix of imaging techniques used in the non-Medicare population (which receives two-thirds of all imaging services) is significantly different from that for the Medicare population, then our use of Medicare estimates of RVUs per procedure as being applicable to all imaging services might have been erroneous. However, again, trends should be relatively unaffected by this potential problem.
In our calculations of RVU per FTE radiologist, we could not adjust for variations in case mix in each radiologist's practice. There were no questions on the survey regarding the relative proportion of various modalities in a practice; therefore, there was no means of adjusting accurately for case mix, and we could only calculate average RVUs per FTE radiologist in each year at the aggregate level.
RVUs per procedure do not reflect the amount of time spent on the procedure. We applied the adjustment factors used by Arenson et al (19) and recalculated the growth rate. With the adjusted RVUs, which approximate a measure of relative time required, the growth in mean RVUs per FTE radiologist during the period from 1992 to 2002 was 41.4%. However, to fully capture the complexity of image interpretation, RVUs are deliberately not based on time spent. Therefore, we believe that the unadjusted RVUs are a better measure of the trends in workload for the radiologist than time-adjusted RVUs.
Despite these limitations, we believe this study yielded important and useful information about the workload of radiologists, especially trends in workload.
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
Abbreviations: ACR = American College of Radiology • FTE = full-time equivalent • RVU = relative value unit • SEM = standard error of mean
Authors stated no financial relationship to disclose.
Author contributions: Guarantors of integrity of entire study, M.B., J.H.S.; study concepts/study design or data acquisition or data analysis/interpretation, M.B., J.H.S.; manuscript drafting or manuscript revision for important intellectual content, M.B., J.H.S.; approval of final version of submitted manuscript, M.B., J.H.S.; literature research, M.B., J.H.S.; statistical analysis, M.B.; and manuscript editing, M.B., J.H.S.
|
References |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONReferences |
|---|
This article has been cited by other articles:
|
|
 |
|
  M. Bhargavan, A. H. Kaye, H. P. Forman, and J. H. Sunshine Workload of Radiologists in United States in 2006-2007 and Trends Since 1991-1992 Radiology, June 9, 2009; (2009) 2522081895. [Abstract] [Full Text]
|
|
|
|
 |
|
 G. W. L. Boland Enhancing CT Productivity: Strategies for Increasing Capacity Am. J. Roentgenol., July 1, 2008; 191(1): 3 - 10. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
P. H. Ellenbogen The Radiologist Assistant: Best New Thing Since Sliced Bread or Trojan Horse? Radiology, July 1, 2008; 248(1): 4 - 7. [Full Text] [PDF]
|
|
|
|
|
|
I. M. Burger, N. E. Kass, J. H. Sunshine, and S. S. Siegelman The Use of CT for Screening: A National Survey of Radiologists' Activities and Attitudes Radiology, July 1, 2008; 248(1): 160 - 168. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
N. R. Nayak, C. Meghea, M. Bhargavan, H. P. Forman, and J. H. Sunshine Prevalence of Productivity-Enhancing Technologies in Radiology Am. J. Roentgenol., June 1, 2008; 190(6): 1445 - 1452. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
G. P. Krestin, J. C. Miller, S. J. Golding, G. G. Frija, G. M. Glazer, H. G. Ringertz, and J. H. Thrall Reinventing Radiology in a Digital and Molecular Age: Summary of Proceedings of the Sixth Biannual Symposium of the International Society for Strategic Studies in Radiology (IS3R), August 25 27, 2005 Radiology, September 1, 2007; 244(3): 633 - 638. [Full Text] [PDF]
|
|
|
|
|
|
H. M. Zafar, R. S. Lewis, and J. H. Sunshine Satisfaction of Radiologists in the United States: A Comparison between 2003 and 1995 Radiology, July 1, 2007; 244(1): 223 - 231. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
R. S. Lewis, M. Bhargavan, and J. H. Sunshine Women Radiologists in the United States: Results from the American College of Radiology's 2003 Survey Radiology, March 1, 2007; 242(3): 802 - 810. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
T. L. Ebbert, C. Meghea, S. Iturbe, H. P. Forman, M. Bhargavan, and J. H. Sunshine The State of Teleradiology in 2003 and Changes Since 1999 Am. J. Roentgenol., February 1, 2007; 188(2): W103 - W112. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
C. Meghea and J. H. Sunshine Retirement patterns and plans of radiologists. Am. J. Roentgenol., December 1, 2006; 187(6): 1405 - 1411. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. H. Sunshine, and C. Meghea, How could the radiologist shortage have eased? Am. J. Roentgenol., November 1, 2006; 187(5): 1160 - 1165. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
R. S. Lewis, J. H. Sunshine,, and M. Bhargavan A portrait of breast imaging specialists and of the interpretation of mammography in the United States. Am. J. Roentgenol., November 1, 2006; 187(5): W456 - W468. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
L. Merewitz and J. H. Sunshine A Portrait of Pediatric Radiologists in the United States Am. J. Roentgenol., January 1, 2006; 186(1): 12 - 22. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. H. Sunshine, R. S. Lewis, and M. Bhargavan A Portrait of Interventional Radiologists in the United States Am. J. Roentgenol., November 1, 2005; 185(5): 1103 - 1112. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
C. I. Meghea and J. H. Sunshine Who's Overworked and Who's Underworked among Radiologists? An Update on the Radiologist Shortage Radiology, September 1, 2005; 236(3): 932 - 938. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| RADIOLOGY | RADIOGRAPHICS | RSNA JOURNALS ONLINE |
