Note: This copy is for your personal non-commercial use only.

To order presentation-ready
copies for distribution to your colleagues or clients, contact us at www.rsna.org/rsnarights.

ORIGINAL RESEARCH
Utilization of Diagnostic Medical
Imaging: Comparison of Radiologist

HEALTH POLICY AND PRACTICE

Referral versus Same-Specialty
Referral1
G. Scott Gazelle, MD, MPH, PhD
Purpose: To retrospectively compare the frequency with which
Elkan F. Halpern, PhD
patients underwent diagnostic medical imaging proce-
Heather S. Ryan, MSc
dures during episodes of outpatient medical care accord-
Angela C. Tramontano, MPH ing to whether their physicians referred patients for imag-
ing to themselves and/or physicians in their same specialty
or to radiologists.

Materials and Institutional review board approval was not necessary for
Methods: this HIPAA-compliant study. An insurance claims database
from a large national employer-based health plan was ob-
tained. Claims data from 1999 2003 were grouped into
episodes of care for six conditions: cardiopulmonary dis-
ease, coronary and/or cardiac disease, extremity fracture,
knee pain, intraabdominal malignancy, and stroke. For
each condition, each referring physicians behavior was
categorized as either same-specialty referral or radiolo-
gist referral on the basis of that physicians entire history
of imaging referrals for the condition. The frequency with
which patients underwent diagnostic medical imaging pro-
cedures during episodes of care was compared according
to whether their physicians referred patients for imaging
to themselves and/or same-specialty physicians or to radi-
ologists. Rates were compared by using 2 tests, and logis-
tic regression was used to compare utilization rates, with
patient age and number of comorbidities as covariates.

Results: For the conditions evaluated, physicians who referred pa-

tients to themselves or to other same-specialty physicians
for diagnostic imaging used imaging between 1.12 and
2.29 times as often, per episode of care, as physicians who
referred patients to radiologists (P .005 for all compari-
sons). Adjusting for patient age and comorbidity, the like-
lihood of imaging was 1.196 3.228 times greater for pa-
tients cared forby same-specialtyreferring physicians.

Conclusion: Same-specialtyreferring physicians tend to utilize imaging

more frequently than do physicians who refer their pa-
tients to radiologists. These results cannot be explained by
differences in case mix (because analyses were performed
1
From the Institute for Technology Assessment, Massa- within six specific conditions of interest), patient age, or
chusetts General Hospital, 101 Merrimac St, 10th Floor,
comorbidity.
Boston, MA 02114-4724 (G.S.G., E.F.H., H.S.R., A.C.T.);
and Department of Health Policy and Management, Har-
vard School of Public Health, Boston, Mass (G.S.G.). From RSNA, 2007
the 2006 RSNA Annual Meeting. Received January 29,
2007; revision requested March 15; revision received
March 28; accepted April 11; final version accepted April
27. Supported in part by a grant from the American Col-
lege of Radiology. Address correspondence to G.S.G.

RSNA, 2007

Radiology: Volume 245: Number 2November 2007 517

HEALTH POLICY AND PRACTICE: Utilization of Diagnostic Medical Imaging
T
he use of diagnostic medical imag-
ing has increased rapidly during
the past several years, and rela-
tively recent studies suggest that imag-
ing by nonradiologists has increased at a
faster rate than imaging by radiologists
(1,2). In the face of these increases, the
potential for conflicts of interest associ-
ated with self referral or with same-
specialty referral to affect utilization has
garnered attention (3). Results of prior
studies investigating the effect of self
referral on the utilization of diagnostic
medical imaging have demonstrated
that physicians who do not refer their
patients to radiologists for medical im-
aging use imaging more frequently than
do physicians who refer their patients to
radiologists (46). Those studies, based
on episode-of-care analyses of health in-
surance claims data from the 1980s,
were published prior to the enactment
of legislation aimed at preventing self
referral in some settings. Studies docu-
menting the increased performance of
imaging by nonradiologists (1,2) did
not use an episode-of-care analytic ap-
proach; thus, their findings may repre-
sent, at least in part, a change in the
relative proportion of imaging proce-
dures performed by radiologists and by
nonradiologists rather than an increase
in utilization on the part of self-referring
or same-specialtyreferring physicians
per se. The purpose of our study, there-
fore, was to retrospectively compare
the frequency with which patients
underwent diagnostic medical imaging
procedures during episodes of outpa-
tient medical care according to whether
their physicians referred patients for
Advances in Knowledge
Same-specialty referral is associ-
ated with increased utilization of
diagnostic imaging.
Within the conditionimaging
procedure combinations studied,
patient age and comorbidity do
not explain differences in utiliza-
tion of imaging.
These results suggest that earlier
findings by other investigators
remain relevant, despite regula-
tory changes (eg, Stark laws).
518
imaging to themselves and/or physi-
cians in their same specialty or to radi-
ologists.
Materials and Methods
Data Acquisition and Analysis
For our Health Insurance Portability
and Accountability Actcompliant study,
we obtained access to the claims for a
large national employer-based health
plan (our Data Use Agreement with the
health plan precludes its specific identi-
fication). The approximately 4 million
enrollees include employees, retirees,
and their dependents. They are distrib-
uted across the United States and are
sociodemographically mixed. A deiden-
tified (both patients and physicians)
data extract for the years 1999 through
2003 provided more than 526 000 000
claims. Because of the deidentified na-
ture of the health plan claims database,
our study did not require institutional
review board approval or informed con-
sent. The health plan is able to make its
deidentified claims data available to re-
searchers who sign a Data Use Agree-
ment. Separate files provided informa-
tion regarding each member and pro-
vider and details of each claim for
professional and other services, inpatient
stays, and pharmacy use. These data
were merged into a single database.
The data were processed (H.S.R.,
A.C.T.) by using episodic analysis soft-
ware (ETG Grouper; Ingenix/Symmetry
Products, Phoenix, Ariz), resulting in
the identification of episodes of care.
Each episode consisted of one or more
records for a member, comprising an
illness. The software assigns each epi-
sode to an underlying Episode Treat-
ment Group. Episodes relating to each
of six conditions (cardiopulmonary dis-
ease, coronary and/or cardiac disease,
extremity fracture, knee pain, intraab-
dominal malignancy, stroke) were iden-
Implication for Patient Care
Policies focused on limiting spend-
ing on diagnostic imaging should
address the role of same-specialty
referral.
Gazelle et al
tified by means of these Episode Treat-
ment Groups.
All inpatient care claims were then
eliminated. For each episode, a refer-
ring physician was defined as the physi-
cian who billed for an office visit.
Episodes with more than one such re-
ferring physician were eliminated. One-
physician episodes comprised approxi-
mately 76% of all valid episodes in the
study data set.
Imaging procedures of interest were
identified (G.S.G.) for each condition
(Table 1). For stroke and knee pain, we
separately evaluated CT and MR imag-
ing.
For each imaging claim, the pro-
vider of professional services was
identified. That providers specialty
(as indicated in the claims database)
was contrasted with the specialty of
the referring physician. Each imaging
claim was then categorized as being by a
radiologist (RAD), a provider with the
same specialty as the referring physi-
cian (SAME), a provider with some
other known specialty (OTHER), or
none of the above (MISSING). All epi-
sodes with imaging for each referring
physician were evaluated. Each refer-
ring physician was then assigned to one
of six categories on the basis of that
physicians pattern of referral for imag-
ing procedures: all imaging by radiolo-
gists (ALL RAD), all imaging by same-
specialty providers (ALL SAME), most
(80%99%) imaging by radiologists
(MOST RAD), most imaging by same-
specialty providers (MOST SAME),
not predominantly (21%79%) either
(MIXED), and INDETERMINATE. The
INDETERMINATE group consisted of
Published online
10.1148/radiol.2452070193
Radiology 2007; 245:517522
Author contributions:
Guarantor of integrity of entire study, G.S.G.; study con-
cepts/study design or data acquisition or data analysis/
interpretation, all authors; manuscript drafting or manu-
script revision for important intellectual content, all au-
thors; manuscript final version approval, all authors;
literature research, G.S.G., H.S.R., A.C.T.; statistical analy-
sis, all authors; and manuscript editing, all authors
Authors stated no financial relationship to disclose.
Radiology: Volume 245: Number 2November 2007
HEALTH POLICY AND PRACTICE: Utilization of Diagnostic Medical Imaging
physicians whose referring behavior
could not be determined from the avail-
able data. These physicians either never
referred patients for imaging (for the
condition of concern) or all of their im-
aging referrals were categorized as
OTHER or MISSING. For most condi-
tions, such physicians accounted for less
than 4% of all referring physicians, ex-
cept in the case of evaluation of knee
pain with radiography (20.7%) and
evaluation of extremity fracture with ra-
diography (48.8%).
Analyses were performed to com-
pare ALL RAD with ALL SAME refer-
ring physicians and also to compare
80% or more RAD (MOST RAD and
ALL RAD) with 80% or more SAME
(MOST SAME and ALL SAME). For
brevity, we present only results of those
analyses contrasting ALL RAD with ALL
SAME.
Statistical Analysis
For each combination of condition, im-
aging procedure, and referring physi-
cian type, the number of referring phy-
sicians, the total number of episodes,
the total number of episodes with imag-
ing, the percentage of episodes with im-
aging, and the total number of episodes
with imaging on the same date as the
office visit were calculated. To assess
the comparability of patients seen by
ALL RAD and ALL SAME referring phy-
sicians, patient age and a comorbidity
score (the number of distinct diagnoses
in all of the claims for the episode [one,
two, three, four, or five or more] [7,8])
were also calculated for each combina-
tion of condition, imaging procedure,
and referring provider type. These mea-
sures were then compared by using t
tests and continuity-adjusted 2 tests for
contingency tables, as appropriate. Lo-
gistic regression was then used to deter-
mine whether there was a difference
between ALL RAD and ALL SAME re-
ferring physicians in the likelihood that
imaging would be performed (for each
conditionimaging procedure combina-
tion) after controlling for patient age
and comorbidity (ie, age and comorbid-
ity were included as covariates). P val-
ues of .008 or less were considered to
indicate statistically significant differ-
Radiology: Volume 245: Number 2November 2007
ences (to adjust for the multiple com-
parisons in our study of six conditions).
To investigate the variability among
referring physicians of each type re-
garding the likelihood that an episode
would have imaging, the fraction of epi-
sodes with imaging was calculated for
each referring physician. The standard
deviation was then calculated for all
physicians of each type with at least six
episodes in the condition of concern.
(The fraction of episodes with imaging
among referring physicians who were
classifiable [ie, not INDETERMINATE]
was biased, because all physicians must
have had at least one episode with imag-
ing. Referring physicians with five or
fewer episodes were excluded to reduce
this bias). All analyses were performed
(A.C.T., E.F.H.) by using statistical
software (SAS, version 9.1; SAS Insti-
tute, Cary, NC).
Results
A total of 18 123 121 episodes of care
associated with 882 687 referring physi-
cians were analyzed. The study popula-
tion was 45% male and 55% female; 1%
of the population was 0 34 years old,
6% were 3554 years old, and 93%
were 55 or older. Table 2 summarizes
the raw numbers and percentages of
referring physicians, episodes, episodes
with imaging (as a percentage of all epi-
sodes), episodes with imaging on the
same day as the office visit (as a per-
centage of episodes with imaging), and
the standard deviation across referring
Table 1
Conditions and Imaging Procedures Evaluated
Condition
Cardiopulmonary disease
Cardiac or coronary artery disease
Extremity fracture
Knee pain or injury
Knee pain or injury
Known or suspected abdominal malignancy
Known or suspected stroke
Known or suspected stroke
Note.CT computed tomography, MR magnetic resonance.
Gazelle et al
physicians within each type (ALL
SAME, ALL RAD) for the percentage of
episodes with imaging (for referring
physicians with six or more episodes).
Within the conditionimaging pro-
cedure combinations evaluated, physi-
cians who referred their patients to
themselves or to other physicians in the
same specialty for diagnostic imaging
used imaging between 1.12 and 2.29
times as often, per episode of care, as
did physicians who referred their pa-
tients to radiologists. The variation in
imaging utilization among physicians
with at least six episodes within a
given conditionimaging procedure
combination was substantially larger
among same-specialtyreferring physi-
cians than among radiologist-referring
physicians. In seven of the eight condi-
tionimaging procedure combinations,
patients seen by same-specialtyrefer-
ring physicians were more likely to un-
dergo imaging on the same day as an
office visit (range, 0.932.5 times as of-
ten).
Table 3 summarizes the age and
comorbidity score (number of distinct
diagnoses) for each combination of
condition, imaging procedure, and re-
ferring provider type. Mean patient
age was greater among patients seen
by same-specialtyreferring physicians in
six of eight conditionimaging proce-
dure combinations, while the comorbid-
ity score was greater for patients seen
by radiologist-referring providers in all
eight combinations.
Table 4 contains the results of the
Imaging Procedure
Chest radiography
Nuclear myocardial perfusion or blood pool imaging (planar
imaging, single photon emission computed tomography,
positron emission tomography)
Radiography
Radiography
MR imaging
CT
CT
MR imaging
519
HEALTH POLICY AND PRACTICE: Utilization of Diagnostic Medical Imaging Gazelle et al

Table 2

Numbers of Physicians and Episodes, Frequencies of Imaging and Same-Day Imaging, and Variation in Imaging Utilization
Condition and Variable SAME* RAD* SAME/RAD Ratio P Value

Cardiopulmonary disease
No. of physicians 21 156 (25.6) 51 000 (61.7) 0.41
No. of episodes 612 339 (15.9) 1 196 495 (50.9) 0.31
No. of episodes with imaging 127 190 (20.8) 179 236 (9.1) 2.29 .001
No. of same-day imaging procedures 108 083 (84.9) 85 508 (47.7) 1.78 .001
Standard deviation of percentage of episodes with imaging 0.13 0.06 2.17 .001
Cardiac and/or coronary disease
No. of physicians 6737 (33.0) 11 827 (58.0) 0.57
No. of episodes 469 175 (27.0) 783 213 (45.2) 0.60
No. of episodes with imaging 55 014 (11.7) 36 213 (4.6) 2.54 .001
No. of same-day imaging procedures 6315 (11.5) 1659 (4.6) 2.5 .001
Standard deviation of percentage of episodes with imaging 0.16 0.07 2.29 .001
Extremity fracture
No. of physicians 13 622 (42.8) 13 449 (42.3) 1.01
No. of episodes 51 992 (29.7) 47 331 (27.0) 1.10
No. of episodes with imaging 43 585 (83.8) 34 926 (73.8) 1.14 .001
No. of same-day imaging procedures 38 466 (88.3) 25 475 (72.9) 1.21 .001
Standard deviation of percentage of episodes with imaging 0.35 0.56 0.63 .001
Knee pain
No. of physicians 14 979 (39.9) 17 267 (46.0) 0.87
No. of episodes 139 216 (28.1) 94 429 (19.0) 1.47
No. of episodes with imaging 84 430 (60.6) 43 999 (46.6) 1.30 .001
No. of same-day imaging procedures 77 251 (91.5) 30 878 (70.2) 1.30 .001
Standard deviation of percentage of episodes with imaging 0.26 0.36 0.72 .001
Knee pain#
No. of physicians 668 (8.6) 6781 (86.9) 0.10
No. of episodes 17 066 (5.9) 228 707 (79.5) 0.07
No. of episodes with imaging 2206 (12.9) 22 885 (10.0) 1.29 .001
No. of same-day imaging procedures 1617 (73.3) 13 462 (58.8) 1.25 .001
Standard deviation of percentage of episodes with imaging 0.50 0.15 3.33 .001
Abdominal cancer**
No. of physicians 299 (5.1) 5448 (93.0) 0.05
No. of episodes 12 226 (4.0) 270 639 (88.9) 0.05
No. of episodes with imaging 667 (5.5) 10 812 (4.0) 1.37 .001
No. of same-day imaging procedures 108 (16.2) 937 (8.7) 1.86 .001
Standard deviation of percentage of episodes with imaging 0.59 0.13 4.53 .001
Stroke**
No. of physicians 235 (5.2) 4250 (93.7) 0.06
No. of episodes 2561 (6.2) 37 758 (90.7) 0.07
No. of episodes with imaging 467 (18.2) 6131 (16.2) 1.12 .005
No. of same-day imaging procedures 228 (48.8) 2375 (38.7) 1.26 .001
Standard deviation of percentage of episodes with imaging 1.36 0.30 4.53 .001
Stroke#
No. of physicians 313 (10.3) 2682 (88.0) 0.12
No. of episodes 3270 (9.4) 29 509 (84.6) 0.11
No. of episodes with imaging 628 (19.2) 4185 (14.2) 1.35 .001
No. of same-day imaging procedures 80 (12.7) 571 (13.6) 0.93 .536
Standard deviation of percentage of episodes with imaging 1.14 0.30 3.8 .001

* Data in parentheses are percentages. P values are for SAME versus RAD. Evaluated with chest radiography. Evaluated with nuclear medicine. Evaluated with radiography. # Evaluated with
MR imaging. ** Evaluated with CT.

520 Radiology: Volume 245: Number 2November 2007

HEALTH POLICY AND PRACTICE: Utilization of Diagnostic Medical Imaging Gazelle et al

Table 3

Age of and Number of Diagnoses for Patients Seen by SAME versus RAD
Mean Patient Age (y) Mean No. of Diagnoses
Condition Imaging Procedure Physician All Episodes Imaging Episodes* P Value All Episodes Imaging Episodes* P Value

Cardiopulmonary disease Chest radiography SAME 60.7 61.9 .001 2.22 2.64 .001
RAD 63.3 62.0 2.54 3.81
Cardiac and/or coronary disease Nuclear medicine SAME 69.0 67.0 .001 2.62 3.74 .001
RAD 68.3 66.0 2.66 4.07
Extremity fracture Radiography SAME 53.8 53.1 .001 2.54 2.57 .001
RAD 47.9 45.5 3.56 3.8
Knee pain Radiography SAME 56.9 57.6 .001 2.15 2.28 .001
RAD 57.8 58.1 2.56 3.13
Knee pain MR imaging SAME 58.0 51.9 .003 2.37 3.53 .104
RAD 57.5 50.9 2.39 4.02
Abdominal malignancy CT SAME 70.9 70.2 .001 2.68 4.01 .005
RAD 67.1 64.0 2.73 4.26
Stroke CT SAME 74.8 74.1 .002 2.85 3.51 .002
RAD 73.9 74.2 2.95 4.23
Stroke MR imaging SAME 73.5 70.7 .022 2.71 3.63 .001
RAD 73.0 68.7 2.98 4.24

* For only those patients in episodes where imaging was performed.

P values are for SAME versus RAD, considering all episodes (ie, not just episodes with imaging).

logistic regression analysis of the chance

Table 4
that an episode would include imaging.
The odds ratio is for ALL SAME versus Results of Logistic Regression Analysis of Likelihood of Imaging
ALL RAD, controlling for patient age
Condition Imaging Procedure Odds Ratio* P Value
and number of diagnoses. P values are
imprecise, because no adjustment was Cardiopulmonary disease Chest radiography 3.228 (3.201, 3.255) .001
made for clustering of episodes for the Cardiac and/or coronary disease Nuclear medicine 3.004 (2.962, 3.048) .001
same patient or same referring physi- Extremity fracture Radiography 2.753 (2.659, 2.850) .001
cian. However, given the strength of the Knee pain Radiography 2.092 (2.056, 2.129) .001
P values and the related narrowness of Knee pain MR imaging 1.913 (1.840, 1.990) .001
the confidence intervals, we can be con- Abdominal malignancy CT 1.494 (1.375, 1.623) .001
fident that these effects would remain Stroke CT 1.260 (1.127, 1.409) .001
Stroke MR imaging 1.196 (1.012, 1.413) .036
significant even if there were much
greater intercorrelation of clustered * Odds ratios are for SAME versus RAD, controlling for patient age and comorbidity score. Data in parentheses are 95%
data than is likely. confidence intervals.

Discussion
For the conditions and imaging proce-
dures evaluated, physicians who re-
ferred their patients for diagnostic im-
aging to themselves or to other physi-
cians in the same specialty used imaging
1.122.29 times more frequently than
did physicians who referred their
patients to radiologists. The variation
in imaging utilization among same-
specialtyreferring physicians was also
significantly greater than among radiol-
ogist-referring physicians. These differ-
ences cannot be attributed to differ-
ences in case mix, because analyses
were performed within specific Episode
Treatment Groups. Similarly, the logis-
tic regression analysis with patient age
and comorbidity as covariates showed
that the differences could not be attrib-
uted to differences in age or comorbid-
ity. After controlling for patient age and
comorbidity, the likelihood of undergo-
ing an imaging procedure was 1.196
3.228 times greater for patients cared
for by same-specialtyreferring physi-
cians than for patients cared for by ra-
diologist-referring physicians. In seven
of eight comparisons, the odds ratios fa-
voring increased imaging by same-spe-
cialtyreferring physicians were greater
than the unadjusted ratios.
The results of previous studies that
used a similar episodic analytic ap-
proach have generally been similar (4
6), although the magnitude of the in-
creased utilization was less in the cur-
rent study than in those of Hillman (4)
and Hillman et al (5). Differences may
be due, at least in part, to the enact-
ment of legislation aimed at limiting self-
referral in some settings. Compared

Radiology: Volume 245: Number 2November 2007 521

HEALTH POLICY AND PRACTICE: Utilization of Diagnostic Medical Imaging
with previous studies, our study used a
larger claims data set, evaluated the ef-
fects of same-specialty referral rather
than strictly self-referral, assigned re-
ferring physicians to referral categories
by using a more restrictive definition,
and, through logistic regression, at-
tempted to control for differences in pa-
tient age and comorbidity. Our study
population, derived from a large na-
tional employer-based health plan, in-
cludes employees, retirees, and their
dependents. It is geographically and so-
ciodemographically diverse and may be
more representative of the U.S. popula-
tion than the data sets used by Hillman
(4) and Hillman et al (5).
We chose to look at same-specialty
referral instead of just self-referral be-
cause self-referral may represent re-
ferral to ones partners or colleagues.
This broader definition may have re-
sulted in the inclusion of some physi-
cians with no financial or other relation-
ship with the referring physician and
may thus have reduced the magnitude
of the observed increase in utilization
(ie, if same-specialtyreferring physi-
cians without financial relationships did
not demonstrate increased utilization of
diagnostic imaging, their inclusion would
have diluted the effect). Compared with
Hillman (4) and Hillman et al (5), who
defined self-referring physicians as indi-
viduals who charged at least once for an
imaging procedure, we categorized re-
ferring physicians on the basis of their
entire referral history for the condition
of concern. Same-specialty referrers
must have always referred patients to
themselves or to others in the same spe-
cialty. Finally, our logistic regression
analysis, which controlled for patient
age and comorbidity, may provide a
more accurate estimate of the effect of
same-specialty referral on the utiliza-
tion of diagnostic imaging procedures.
We based our methods on those of
prior studies and attempted to improve
on those methods where possible. Nev-
ertheless, our study had several poten-
522
tial limitations. First, categorization of
referring physician behavior was not
possible in some instances, where ei-
ther physicians only referred patients to
an imaging provider with some other
known specialty or the specialty of the
imaging provider could not be identi-
fied. These physicians accounted for a
small percentage of all referring physi-
cians. Our results are based on those
referring physicians whose referral be-
havior could be accurately categorized.
Second, because we categorized refer-
ral behavior on the basis of the specialty
associated with the professional (rather
than technical) component of the imag-
ing claim, we may have missed cases of
same-specialty referral where the pro-
fessional component (ie, interpretation
only) was subcontracted out to a radiol-
ogist. Finally, we did not look at imaging
facility ownership because this informa-
tion was not available in the claims data
set.
Our findings should be of interest to
regulatory and reimbursement agencies
and other groups that are trying to man-
age trends in imaging utilization. It is
not possible to determine from our re-
sults if the increased utilization of same-
specialtyreferring physicians represents
better care or results in better patient
outcomes. It is also not possible to de-
termine the extent to which financial
incentives may explain the greater utili-
zation. However, the magnitude of our
findings and their consistency with
those of other studies suggest that finan-
cial incentives may play a role. It may
also be that patients are more likely to
adhere to recommendations to undergo
imaging procedures if the procedures
can be performed on the same day as
office visits, and our study results did
demonstrate that patients seen by same-
specialtyreferring physicians more of-
ten underwent imaging on the same day
as an office visit. Finally, there may be
some selection bias involved if physi-
cians who for some reason are more
likely to order imaging procedures tend
Gazelle et al
to acquire imaging equipment them-
selves or to affiliate with others in the
same specialty who do so.
In conclusion, our results demon-
strate that same-specialtyreferring phy-
sicians tend to utilize imaging more fre-
quently than physicians who refer their
patients to radiologists. These findings
were consistent across the eight combina-
tions of conditions and imaging proce-
dures evaluated and cannot be ex-
plained by differences in case mix, pa-
tient age, or comorbidity.
References
1. Maitino AJ, Levin DC, Parker L, Rao VM,
Sunshine JH. Practice patterns of radiologists
and nonradiologists in utilization of noninva-
sive diagnostic imaging among the Medicare
population 19931999. Radiology 2003;228:
795 801.
2. Maitino AJ, Levin DC, Parker L, Rao VM,
Sunshine JH. Nationwide trends in rates of
utilization of noninvasive diagnostic imaging
among the Medicare population between
1993 and 1999. Radiology 2003;227:113117.
3. Thompson DF. Understanding conflicts of in-
terest. N Engl J Med 1993;329:573576.
4. Hillman BJ. Frequency and costs of diagnostic
imaging in office practice: a comparison of
self-referring and radiologist-referring physi-
cians. N Engl J Med 1990;323:1604 1608.
5. Hillman BJ, Olson GT, Griffith PE, et al. Phy-
sicians utilization and charges for outpatient
diagnostic imaging in a Medicare population.
JAMA 1992;268(15):2050 5054.
6. Referrals to physician-owned imaging facili-
ties warrant HCFFAs scrutiny: report to the
chairman. Subcommittee on Health. Commit-
tee on Ways and Means. House of Represen-
tatives. Washington, DC: U.S. General Ac-
counting Office, 1994; 1 61.
7. Melfi C, Holleman E, Arthur D, Katz B. Select-
ing a patient characteristics index for the pre-
diction of medical outcomes using administra-
tive claims data. J Clin Epidemiol 1995;48:
917926.
8. Rochon PA, Katz JN, Morrow LA, et al. Co-
morbid illness is associated with survival and
length of hospital stay in patients with chronic
disability: a prospective comparison of three
comorbidity indices. Med Care 1996;34(11):
10931101.
Radiology: Volume 245: Number 2November 2007