The American Journal of Surgery xxx (2017) 1e4

Contents lists available at ScienceDirect

The American Journal of Surgery

journal homepage: www.americanjournalofsurgery.com

The diagnosis and staging of pancreatic cancer: A comparison of

endoscopic ultrasound and Computed Tomography with pancreas
protocol
Tianli Du a, *, Katherine A. Bill d, Jennifer Ford a, Mohammed Barawi b,
Richard D. Hayward a, Amer Alame a, Richard N. Berri c
a
St. John Hospital and Medical Center, Department of Surgery, Detroit, MI, USA
b
St. John Hospital and Medical Center, Department of Gastroenterology, Detroit, MI, USA
c
St. John Hospital and Medical Center, Department of Surgery, Section of Surgical Oncology, Detroit, MI, USA
d
School of Medicine, Wayne State University, Detroit, MI, USA

a r t i c l e i n f o a b s t r a c t

Article history: Background: Pancreatic cancer is the fourth leading cause of cancer-related death in United States. We
Received 6 July 2017 compared Computed Tomography (CT) with pancreas protocol and Endoscopic Ultrasound (EUS) in
Received in revised form terms of mass detection, mass size, vascular involvement and lymph node involvement.
14 November 2017
Methods: We retrospectively evaluated 93 patients. Concordance between CT and EUS, and accuracy of
Accepted 14 November 2017
CT and EUS were assessed using a retrospective chart review and statistical analysis.
Results: CT and EUS agreed on mass detection in 88% of the cases and mass size in 67% of the cases. They
agreed in 74% of cases about the presence or absence of vascular involvement and 82% in lymph node
involvement. Cohen's kappa indicated that the concordance between two tests was moderately reliable.
Conclusion: CT and EUS agree moderately well in identifying characteristics of pancreatic masses, but
discrepancies between the two modalities are common, particularly with respect to involvement of
specific blood vessels and lymph nodes. Clinicians should use caution in relying on a single modality to
make decisions.
© 2017 Elsevier Inc. All rights reserved.

1. Introduction
Pancreatic cancer is the fourth leading cause of cancer-related
death in United States. Most patients who present for initial eval-
uation have advanced stage disease, and only 15e20% are candi-
dates for resection.
CT and EUS with or without fine needle biopsy (FNA) have been
widely used for diagnosing and staging of pancreatic cancer. There
have been extensive studies comparing CT and EUS in diagnosing
and staging of pancreatic cancer. These studies showed variable
results and conflicting conclusions as to which modality is better.1,2
In recent years, CT with pancreas protocol was created to eval-
uate the pancreas. It consists of a pre-contrast scan, multiple post-
contrast phases including arterial phase (20e30s after injection of
contrast), venous phase (60e70s after injection of contrast), with or
* Corresponding author.
E-mail address: tianlidu@gmail.com (T. Du).
without delayed phase (90s after injection of contrast). The arterial
phase of enhancement provides excellent opacification of the celiac
axis, superior mesenteric artery, and peripancreatic arteries. The
portal venous phase provides better enhancement of the superior
mesenteric vein, splenic and portal veins. CT with pancreas pro-
tocol also has high resolution and thin sections. There are limited
number of studies regarding the accuracy of CT with pancreas
protocol in detecting a pancreatic mass and evaluating vascular
involvement and resectability.3e5 In addition, the correlation be-
tween EUS and CT with pancreas protocol in detecting and staging
of pancreatic cancer has not been studied to our knowledge.
2. Methods
Concordance and accuracy between CT with pancreas protocol
and EUS were assessed using a retrospective chart review. CT with
pancreas protocol was done with cross section thickness 1e4 mm.
It was also multiphasic, including an arterial phase, venous phase
and delayed phase, which were 30, 60, 90s after injection of

https://doi.org/10.1016/j.amjsurg.2017.11.021
0002-9610/© 2017 Elsevier Inc. All rights reserved.

Please cite this article in press as: Du T, et al., The diagnosis and staging of pancreatic cancer: A comparison of endoscopic ultrasound and
Computed Tomography with pancreas protocol, The American Journal of Surgery (2017), https://doi.org/10.1016/j.amjsurg.2017.11.021
2 T. Du et al. / The American Journal of Surgery xxx (2017) 1e4
contrast. EUS was performed by a single gastroenterologist per-
forming on average 700 EUS procedures every year.
The sample consisted of 93 patients who underwent both CT
with pancreas protocol and EUS for clinically suspected pancreatic
cancer from 6/2012 to 10/2016. There was no chemotherapy or
surgery done in between the two tests. Because assessments for
each of the outcomes examined in the study were not necessarily
recorded for every image, the sample size available for each anal-
ysis differs somewhat. Patients were excluded from the analysis if
more than 90 days elapsed between the two imaging sessions.
Mass detection for both CT and EUS were coded as positive,
negative, or indeterminate. A case was counted as in agreement if
both CT and EUS indicated the same category of result. It was
counted as in disagreement if both images indicated different cat-
egorical results.
Mass size estimates as indicated by CT and EUS in centimeters
were transcribed from patient charts. In 15 cases, size was indicated
only as “small” or “large,” and these cases were excluded from
analyses of mass size. If the size difference in size estimates given
by CT and EUS was less than 1 cm, the case was coded as in
agreement. If the difference in size estimates was 1 cm or greater,
the case was coded as in partial agreement.
Vascular involvement was also recorded. Readings for CT and
EUS indicated involvement or encasement of the pancreatic mass
with specific major blood vessels. The recorded blood vessels
included superior mesenteric artery and vein, celiac axis, portal
vein, common hepatic artery, splenic artery and vein. Agreement
was assessed in two ways. First, the presence of vascular involve-
ment of any type was recorded. A case was determined to agree if
both imaging methods indicated the presence or absence of any
form of vascular involvement. Second, the specific blood vessels
involved were compared. Because a single image could indicate
involvement of multiple blood vessels, agreement on this measure
could be either partial (if both images indicated at least one blood
vessel involvement in common, but not identical involvement), or
full (if both CT and EUS indicated identical vascular involvement).
Disagreement was coded when the vessels indicated by each mo-
dality shared no common blood vessels.
Lymph node abnormality readings were coded for agreement in
a similar fashion to that described above for vascular involvement.
Lymph node abnormalities were defined by lymph nodes with
enlarged size over 1 cm. The recorded lymph nodes included per-
ipancreatic lymph nodes, porta hepatic lymph nodes, and celiac
lymph nodes. First, agreement was assessed for detection of any
lymphatic abnormalities. Second, the specific lymph nodes identi-
fied were assessed for full, partial and no agreement. Patient
gender, patient age, and time between CT and EUS were also
collected by chart review to help assess potential confounding
factors.
Approval for the project was granted by the Institutional Review
Board of St. John Hospital and Medical Center. The primary focus of
the statistical analyses was on agreement between CT with
pancreas protocol and EUS measures, as evaluated with Cohen's
kappa. Descriptive statistics were also conducted, and chi-square
and t-tests were used to screen for differences in CT-EUS agree-
ment categories associated with demographic and clinical back-
ground characteristics. All statistical analyses were conducted in
SAS 9.4.
3. Results
The sample included 46 men and 47 women with a mean age of
68.4 years (SD ¼ 11.6). The mean time between CT and EUS was 26.7
days (SD ¼ 34.6). The median time between CT and EUS was 11.3
days (SD ¼ 3.5). All 93 patients had pathologically confirmed
Please cite this article in press as: Du T, et al., The diagnosis and staging of pancreatic cancer: A comparison of endoscopic ultrasound and
Computed Tomography with pancreas protocol, The American Journal of Surgery (2017), https://doi.org/10.1016/j.amjsurg.2017.11.021
pancreatic tumors. Most of them (75) had adenocarcinoma, nine
patients had neuroendocrine tumors, six patients had mucinous
neoplasms, three had other tumors. Twenty-four patients went for
surgery. Five of them were found to have unresectable disease and
surgery aborted. 5 patients had neoadjuvant chemotherapy prior to
resection. Fourteen patients had definitive resection without neo-
adjuvant chemotherapy.
To address potentially confounding variables, associations be-
tween each agreement of each categories and the following de-
mographic and design factors were assessed: gender, age,
assessment order (CT or EUS first), and the time between CT and
EUS. A significant relationship was detected on only one of these
pairs of variables, such that older age was associated with a higher
rate of agreement in lymphatic abnormality (t[64] ¼ 2.699,
p ¼ 0.009).
Agreement statistics for estimates of mass detection, mass size,
vascular involvement, and lymphatic abnormalities are presented
in Table 1. Most of cases (88%) showed agreement on mass detec-
tion. Both tests showed agreement on mass size in 67% of the cases.
CT and EUS agreed in 74% of cases about whether any vascular
involvement was indicated. However, agreement was poor with
respect to identification of involvement of specific blood vessels,
with only 12% of cases in full concordance, and 50% indicating
involvement of completely different blood vessels from one
another. Overall agreement was somewhat better for lymph node
involvement, with 82% of cases in agreement on their presence or
absence. When it comes to the involvement of specific lymph node
regions, the agreement was poor, with 29% full agreement and 29%
partial agreement.
Cohen's kappa statistics provide an index of the magnitude of
association between categorical variables, and are used to assess
inter-rater reliability. Significant Cohen's kappa statistics for all four
dimensions of agreement (see Table 2) indicated that CT and EUS
agreed at rates greater than that which could be explained by
chance. Although the magnitude of the kappa statistics reported
here fall into the range conventionally identified as moderately
reliable, considerable caution is needed when comparing these
statistics across dimensions.6,7
4. Discussion
Computed tomography (CT) and endoscopic ultrasound (EUS)
with or without fine needle biopsy (FNA) have been widely used for
diagnosing and staging of pancreatic cancer. There have been
extensive studies comparing the accuracy, sensitivity and speci-
ficity of single phasic, helical CT and EUS in terms diagnosing and
Table 1
Agreement results between CT and EUS for mass detection mass size, vascular
involvement, and lymphatic abnormality.
N (%)
CT not read a
0 (0.0%)
EUS not reada 1 (1.0%)
Neither reada 2 (2.2%)
Both reada 90 (96.8%)
No Mass on CT or EUSb 3 (3.3%)
Indeterminate on CT and EUSb 2 (2.2%)
Indeterminate EUS, negative CT 1 (1.1%)
Indeterminate CT, negative EUS 0 (0.0%)
Mass on CT onlyb 5 (5.6%)
Mass on EUS onlyb 5 (5.6%)
Mass on CT and EUSb 74 (82.2%)
Cohen's kappa 0.45, p < 0.001
a
Percent is out of all patients.
b
Percent is out of patients with both CT and EUS read.
 T. Du et al. / The American Journal of Surgery xxx (2017) 1e4 3

Table 2
Agreement results between CT and EUS for mass detection mass size, vascular involvement, and lymphatic abnormality.

Mass Size Vascular Involvement Lymphatic Abnormality

CT not reada 3 (3.2%) 6 (6.5%) 3 (3.2%)

EUS not reada 7 (7.5%) 17 (18.3%) 18 (19.4%)
Neither reada 0 (0.0%) 5 (5.8%) 6 (6.5%)
Both reada 83 (89.2%) 65 (69.9%) 66 (71.0%)

EUS positive but not quantified with positive CTb 22 (26.5%) N/A N/A
EUS positive but not quantified with negative CTb 1 (1.2%) N/A N/A
Nothing detected CT or EUSb 4 (4.8%) 14 (21.5%) 40 (60.6%)
Detection on CT onlyb 1 (1.2%) 10 (15.4%) 6 (9.1%)
Detection on EUS onlyb 3 (3.6%) 7 (10.8%) 7 (10.6%)
Detection on CT and EUSb 52 (62.7%) 34 (52.3%) 14 (21.2%)

No Agreementc N/A 17 (50.0%) 6 (42.9%)

Partial Agreementc 17 (32.7%) 13 (38.2%) 4 (28.6%)
Full Agreementc 35 (67.3%) 4 (11.8%) 4 (28.6%)

Cohen's kappa 0.70, p < 0.001 0.42, p < 0.001 0.54, p < 0.001

NOTES: For mass size, partial agreement indicates a discrepancy  1 cm. For vascular involvement and lymph abnormality partial agreement indicates that CT and EUS
identified at least one element in common but differed on at least one other; no agreement indicates each method identified elements with no overlap. Kappa statistics
computed based on non-detection vs. any detection among cases with valid readings on both CT and EUS.
a
Percent is out of all patients.
b
Percent is out of patients with both CT and EUS read.
c
Percent is out of patients with detection on both CT and EUS.

staging of pancreatic cancer. In a systemic review involving 30
studies and 1554 patients, EUS is superior to conventional CT in
identifying vascular involvement. The sensitivity of EUS and CT was
72% and 63% and the specificity of EUS and CT was 89% and 92%.8 A
metanalysis showed EUS has a sensitivity of 91% and the specificity
of 94%.9 However, other studies have showed variable results and
conflicting conclusion. Currently, there is no consensus in which
image modality is better at diagnosis and pre-operative evaluation
of pancreatic cancer.
In recent years, a special CT protocol was created to evaluate the
pancreas. It consists of a pre-contrast scan, multiple post-contrast
phases including arterial phase 20e30s, venous phase 60e70s,
with or without delayed phase. It also has high resolution and thin
sections, specifically axial section thickness of about 1e4 mm. In
theory, it should be superior to conventional CT and EUS in evalu-
ation of the vasculature around a pancreatic mass. However, there
has been a limited number of studies investigating this. In a study
including 25 patients, was shown that CT with pancreas protocol
has a sensitivity of 84%, specificity of 98%, positive predictive value
of 95% and negative predictive value of 93%.4 In recent studies
including 25 patients, multiphasic CT has been shown to have a
negative predictive value of 87% (20/23 patients) for overall
resectability.3 These studies are limited by small sample size. In
addition, the concordance between EUS and CT with pancreas
protocol in detecting and staging of pancreatic cancer has not been
evaluated in those studies.
Our study evaluated the concordance between CT with pancreas
protocol and EUS in terms of pancreatic mass detection, mass size,
vascular involvement and lymph node involvement. The mean time
between CT and EUS was 26.7 days and median time was 11.3 days.
Some patients had both studies done up to 85 days apart, which
increased the average time significantly. Those patients all had less
aggressive histological type of pancreatic cancer (eg. Neuroendo-
crine tumors). Supplemental analyses also found no significant
relationship between the length of time between CT and EUS and
the rate of agreement between the two modalities.
Both tests agreed on the mass detection in 88% and mass size in
67% of the cases. In terms of vascular and lymph node invasion, CT

Fig. 1. The agreement rate between CT and EUS for mass detection, mass size, vascular involvement, and lymphatic abnormality. For vascular involvement and lymph node
involvement category, black column shows the agreement on presence or absence of involvement, while the white column demonstrates the agreement on involvement of specific
blood vessels and lymph node areas. Kappa statistics was computed based on the cases with valid readings on both CT and EUS.

Please cite this article in press as: Du T, et al., The diagnosis and staging of pancreatic cancer: A comparison of endoscopic ultrasound and
Computed Tomography with pancreas protocol, The American Journal of Surgery (2017), https://doi.org/10.1016/j.amjsurg.2017.11.021
4 T. Du et al. / The American Journal of Surgery xxx (2017) 1e4
and EUS agreed in 74% of cases about presence or absence of
vascular involvement and 82% of cases in presence or absence of
lymph node involvement. However, agreement was poor with
respect to identification of specific forms of vascular or lymph node
involvement, only 12% and 29%. The concordance of these cate-
gories was statistically analyzed. Cohen's kappa was calculated and
indicated the substantially reliable concordance of mass size, and
the moderately reliable concordance of mass detection, vascular
and lymph node involvement (Fig. 1). Our results indicated that EUS
and CT with pancreas protocol did not correlate with each other in
all categories. Thus, each test may provide additional information
when the other test is equivocal. The discordance between the two
modalities can potentially result from the following reasons1: EUS
has a short optimal focal range, about 2e4 cm from the transducer,
which leads to the inconsistent visualization of pancreas tail.2 It is
difficult to distinguish inflammation and neoplasm in CT as well as
EUS. The presence of chronic pancreatitis decreases the diagnostic
accuracy of both tests.
We use EUS in almost all cases because EUS is highly performer
dependent and we are fortunate to a very experienced gastroen-
terologist to do all of the EUS at our institution. While our study did
not show superiority of EUS, we recommend adherence to NCCN
guidelines, using CT with pancreas protocol as the first line test, and
EUS when additional information about the diagnosis or staging of
pancreatic cancer is needed, and especially if you have a gastro-
enterologist who is experienced in performing the test. One
advantage of EUS is that it can establish cytological diagnosis
through EUS guided final needle aspiration (FNA). While EUS-FNA
is not needed for patients with clearly resectable pancreatic can-
cer, it is needed to provide a tissue diagnosis in those patients with
unresectable tumors and in those patients for whom neoadjuvant
therapy is planned.
Some limitations in our studies should be considered. First, the
sample size to determine the concordance of both tests was
modest. Larger case series are needed to evaluate more precise
elements of concordance, for example with respect to differences in
mass position and involvement of specific vascular and lymphatic
areas. Second, the radiographic resectability of the pancreatic
masses based on vascular involvement was not evaluated. This
information is important as part of the pre-operative evaluation
and directly drive surgeons to different directions of treatment.
Future research should seek to address these limitations, and to
examine CT and EUS concordance in comparison with objective
Please cite this article in press as: Du T, et al., The diagnosis and staging of pancreatic cancer: A comparison of endoscopic ultrasound and
Computed Tomography with pancreas protocol, The American Journal of Surgery (2017), https://doi.org/10.1016/j.amjsurg.2017.11.021
pathology findings and intraoperative findings.
In summary, our results showed that CT with pancreas protocol
and EUS correlate moderately well in terms of mass detection, mass
size, vascular involvement and lymph node involvement. Agree-
ment between two tests was somewhat lower for more specific
details, including which specific blood vessels and lymph nodes
were involved. Each test may provide additional information about
diagnosis and staging of pancreatic cancer. Clinicians should use
caution in relying on a single modality to make decisions. With the
aggressive presentation of most pancreas cancers and the impor-
tance of proper staging, it would seem that CT with pancreas pro-
tocol and if available a skilled gastroenterologist who can perform
EUS together can provide complimentary information regarding
pancreatic masses. While the intent of this study was to only
compare CT and EUS as preoperative staging modalities, future
direction from our group will include steps to investigate how CT
and EUS correlate with final surgical and pathologic findings.
Conflicts of interest
We have no conflict of interest.
References
1. Agarwal B, Abu-Hamda E, et al. Endoscopic ultrasound-guided fine needle
aspiration and multidetector spiral CT in the diagnosis of pancreatic cancer. Am J
Gastroenterol. 2004:04177.
2. Mertz HR, Sechopoulos P, Delbeke D, Leach SD. EUS, PET and CT scanning for
evaluation of pediatric adenocarcinoma. Gastrointest Endosc. 2000;52:367e371.
3. Sheridan MB, Ward J, Guthrie JA, et al. Dynamic contrast-enhanced MR imaging
and dual-phase helical CT in the preoperative assessment of suspected
pancreatic cancer: a comparative study with receiver operating characteristic
analysis. Am J Roentgenol. 1999;173:583e590.
4. Lu DS, Reber HA, Krasny RM, et al. Local staging of pancreatic cancer: criteria for
unresectability of Major vessels as reviewed by Pancreatic phase, thin section
helical CT. Am J Roentgenol. 1997 Jun;168(6):1439e1443.
5. Bronstein YL, Loyer EM, Kaur H, et al. Detection of small pancreatic tumors with
multiphasic helical CT. Am J Roentgenol. Volume 168 number 3.
6. Landis JR, Koch GG. The measurement of observer agreement for categorical
data. Biometrics. 1977;33:159e174.
7. Gwet KL. Handbook of Inter-rater Reliability: The Definitive Guide to Measuring the
Extent of Agreement Among Raters. Advanced Analytics, LLC; 2014.
8. Yang R, Lu M, Qian X, et al. Diagnostic accuracy of EUS and CT of vascular in-
vasion in pancreatic cancer: a systematic review. J Cancer Res Clin Oncol.
2014;140:2077e2086.
9. Hewitt MJ. McPhail M.J.W. Possamai L, et al. EUS-guided FNA for diagnosis of
solid pancreatic neoplasms: a meta-analysis. Gastrointest Endosc. Volume 75,
Issue 2.