| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
1 University of Medicine and Dentistry of New Jersey, Newark, NJ.
2 Department of Radiology, Mayo Clinic, 200 First St. SW, Mayo East-2B,
Rochester, MN 55905.
3 Mayo Medical School, Mayo Clinic, Rochester, MN 55905.
4 Division of Biostatistics, Mayo Clinic, Rochester, MN 55905.
5 Department of Internal Medicine, Division of Gastroenterology, Mayo Clinic,
Rochester, MN 55905.
Received August 5, 2003;
accepted after revision September 30, 2003.
Address correspondence to J. G. Fletcher
(fletcher.joel{at}mayo.edu).
Abstract
 Top Abstract IntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
SUBJECTS AND METHODS. Two radiologists interpreted in a blinded manner 62 CT scans in 28 pancreatic cancer patients that were obtained before histologic diagnosis and 89 CT scans in 89 control subjects and noted specific CT findings. The presence of pancreatic cancer was characterized as definite, suspicious, low probability, or normal. The scans of the pancreatic cancer patients were divided into four groups on the basis of the time interval preceding cancer diagnosis (0–2, 2–6, 6–18, or > 18 months), and one scan (closest to 18 months) was selected per patient per time interval. Sensitivity and specificity for pancreatic cancer and interobserver agreement for CT findings were calculated.
RESULTS. Radiologists agreed that CT findings definite or suspicious
for pancreatic cancer were present in 50% of the scans obtained 2–6 and
6–18 months before the diagnosis of pancreatic cancer (3/6 and 4/8
scans, respectively), but they noted such CT findings in only 7% (1/15) of the
scans obtained more than 18 months before diagnosis. Pancreatic duct
dilatation and cutoff were early CT findings identified by both radiologists
and were associated with near-perfect and substantial interobserver agreement
(
= 0.84 and 0.76, respectively). Ninety-five percent confidence
intervals of specificity for tumor absence ranged from 92% to 100%.
CONCLUSION. CT can detect a significant proportion of asymptomatic incident pancreatic cancers before the clinical diagnosis of pancreatic cancer. CT should be considered in screening at-risk patient populations. Pancreatic duct dilatation and cutoff are early findings associated with the development of pancreatic cancer and can be detected on CT with a high degree of reproducibility.
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
Early detection of small pancreatic tumors will require screening of asymptomatic subjects for pancreatic cancer. Several genetic syndromes such as familial pancreatitis, Peutz-Jeghers syndrome, and familial atypical multiple mole melanoma are associated with an increased risk of pancreatic cancer, in addition to mutations in the second breast cancer gene (BRCA2) and the DNA mismatch repair genes [4–8]. Currently, genetic syndromes with a high incidence of pancreatic cancer are being targeted for screening [9–11]. A strategy of surveillance using endoscopic sonography and ERCP has been successfully used to detect precancerous lesions in one family with autosomal dominant familial pancreatic cancer with high penetrance [10]. Mapping of a susceptibility locus for pancreatic cancer in this family to bands 4q32–34 [12] adds further impetus to this approach to pancreatic cancer screening.
CT has traditionally been used to diagnose and stage pancreatic adenocarcinoma once tumor of the pancreas is suspected clinically— that is, after a patient has developed symptoms such as jaundice, weight loss, or abdominal pain [13–16]. However, abdominal CT, a non-invasive imaging technique, has generally not been considered as useful for screening asymptomatic individuals for pancreatic cancer because of the belief that CT is less sensitive than endoscopic sonography [4, 17]. Nevertheless, recent reports have highlighted the improved performance of MDCT or thin-section helical CT compared with endoscopic sonography or single-detector CT [16, 18, 19].
The purpose of our study was to ascertain whether imaging findings diagnostic of or suspicious for pancreatic duct adenocarcinoma could be seen on abdominal CT of pancreatic cancer patients before the clinical diagnosis of cancer.
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
Using electronic medical records, we identified subjects with a diagnosis of pancreatic duct adenocarcinoma who underwent abdominal CT at the Mayo Clinic Rochester. The charts of the subjects were reviewed, and only patients with a histologic diagnosis of pancreatic cancer were included in the study. For the purposes of this study, the date of diagnosis of pancreatic adenocarcinoma was defined as the date that histologic confirmation of pancreatic adenocarcinoma was obtained. Twenty-four of these pancreatic cancer patients underwent CT at our institution before clinical diagnosis. Four additional patients who underwent CT examinations at outside institutions before clinical diagnosis were included in the study. CT examinations were performed between August 1994 and June 2001. The study group was composed of 18 men (64%) and 10 women (36%), with a mean age of 71 years (range, 45–94 years). A total of 62 CT examinations were performed in the 28 patients.
A control group of patients without pancreatic cancer with an age and sex distribution similar to the patient group was chosen. Eighty-nine patients who underwent CT for nonpancreatic indications were selected as control subjects. The control group consisted of 63 men (71%) and 26 women (29%) with a mean age of 71 years (range, 42–88 years). These patients were selected from a logbook that lists the names of patients who underwent CT on one of our abdominal CT scanners. Control patients with incidental pancreatic findings noted on their radiology reports were excluded because we did not have clinical follow-up in these patients. All control patients consented to the use of their medical records for research purposes.
Hard-copy films of all CT scans from patients and control subjects were presented to two radiologists in random order for interpretation. The radiologists independently interpreted these images while unaware of clinical history and results of all other imaging studies. They noted scanning technique (slice thickness and use of IV contrast material), pancreatic duct dilatation (present, absent, or borderline or indeterminate), pancreatic duct cutoff (present, absent, or borderline or indeterminate), focal attenuation difference (present, absent, or heterogeneous attenuation difference or indeterminate), common bile duct dilatation (present or absent), and pancreatic mass (present, with size in centimeters, or absent). Scanning technique varied among patients and was determined by the attending radiologist at the time scans were obtained.
CT findings were used to characterize the presence of pancreatic cancer as follows: Definite, indicating near 100% probability of pancreatic cancer, was defined by the presence of a pancreatic mass with or without duct dilatation. Suspicious, indicating highly suspicious findings leading to invasive techniques such as endoscopic sonography, imaging-guided biopsy, or surgical exploration, was defined when either duct dilatation and cutoff or a focal attenuation difference was present, in the absence of a definite pancreatic mass. Low probability, indicating high likelihood for the absence of pancreatic cancer, was defined by indeterminate radiographic findings such as indeterminate pancreatic duct dilatation or cutoff or heterogeneous focal attenuation difference. Negative findings were defined as normal.
For the purposes of statistical analysis, CT scans of pancreatic cancer patients were grouped according to the time interval between CT examination and clinical (histologic) diagnosis of pancreatic cancer as follows: 0–2 months, 2–6 months, 6–18 months, and more than 18 months. To avoid bias caused by review of multiple scans of a single patient within any time interval, we included only one scan per patient in each time interval, which was based on the date of the CT examination, for statistical analysis. If multiple CT examinations were performed during any of the three intervals before 18 months, the first CT examination in the interval was used in the interval-specific result. When multiple CT scans were obtained in a single patient more than 18 months before diagnosis, the CT examination closest to 18 months was used in the interval-specific result, given the low likelihood of pancreatic abnormalities on older scans in a malignancy with such a rapid clinical course.
For each time period, sensitivity and 95% confidence intervals (CIs) were estimated for both observers. Agreement of the CT interpretations by the two observers was calculated for each of the four time intervals using the same CT examinations used in the sensitivity estimation. Because marginal totals in the contingency tables of agreement were non-symmetric, the kappa statistic was not used as a measure of agreement; instead, the kappa statistic was calculated to measure the proportion of scans in which the two observers agreed and the proportion of scans in which the two observers agreed and correctly identified a positive CT finding. In this manner, the observed agreement was estimated, rather than the chance-corrected agreement.
In addition, agreement between observers for specific CT findings was estimated for all 151 CT scan results: the 62 scans from the 28 pancreatic cancer patients and the 89 scans from the 89 control subjects. In this estimation of agreement, a kappa statistic and 95% confidence interval were calculated. The timing of the CT scan relative to a possible subsequent pancreatic cancer diagnosis was ignored, and the potential correlation of CT scans within a subject was also ignored. As described by Landis and Koch [20], a kappa statistic of 0.8–1.0 was considered almost perfect agreement; 0.6–0.79, substantial agreement; 0.40–0.59, moderate agreement; 0.2–0.39, fair agreement; 0–0.19, slight agreement; and 0 to –1.0, poor agreement. The specificity for the absence of pancreatic cancer was estimated using the control group.
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
Table 1 summarizes the indications for the CT examinations in the patients with pancreatic cancer and the number of examinations interpreted as definite or suspicious for pancreatic cancer at the time of scanning. With the exception of scans obtained less than 2 months before the diagnosis of pancreatic cancer (when nearly half the scans were obtained to rule out the possibility of pancreatic cancer), patients underwent CT examinations for a variety of conditions, with abdominal pain, surgical planning or follow-up, and follow-up of non-pancreatic cancer, accounting for the majority of the indications. Twenty-four of the 28 cancer patients had their first CT scan interpreted as showing negative findings. One patient had a CT scan obtained 9 months before clinical diagnosis of pancreatic cancer; at that time, the CT scan was interpreted as showing a pancreatic mass, but the patient deferred further workup because he was recovering from repair of an abdominal aortic aneurysm. In three of the patients, the initial scan revealed indeterminate pancreatic findings. Twelve of the 15 patients with scans obtained less than 2 months before their clinical diagnosis had at least one scan obtained later (i.e., during another time interval).
|
The technical parameters used in scanning the pancreatic cancer and control populations were similar: 56 (90%) of 62 CT scans of pancreatic cancer patients and 84 (94%) of 89 scans of control subjects were obtained using IV contrast material. In patients who eventually developed pancreatic cancer, the reasons for the attending radiologist not giving IV contrast material were allergy (n = 2), concurrent use of metformin (n = 1), elevated level of creatinine and contrast material not needed to answer clinical question (n = 1), contrast material not needed to answer clinical question (n = 1), and unknown (n = 1). Five control subjects did not receive IV contrast material for the following reasons: elevated level of creatinine (n = 2), allergy (n = 1), no IV access (n = 1), and IV contrast material not needed to answer the clinical question (n = 1). The slice thicknesses used in the two groups of CT examinations were also similar, with the mean slice thickness (± standard deviation [SD]) in the cancer patients and control subjects being 7.0 ± 1.7 mm and 6.9 ± 0.4 mm, respectively. Biphasic examinations were performed for three (5%) of the 62 examinations in the patient group and two (2%) of the 89 examinations in the control group. Biphasic examinations consisted of contrast-enhanced scanning in the portal venous phase preceded by either arterial or pancreatic phase scanning.
Table 2 summarizes the sensitivity for radiologic findings definite or suspicious for pancreatic cancer in the patients who subsequently developed cancer. As we expected, both observers identified pancreatic cancer on CT scans obtained 0–2 months before clinical diagnosis in a high percentage of the 15 patients with pancreatic cancer (i.e., 93% and 100%, respectively). Almost half of these scanning examinations were ordered because of clinical suspicion of a pancreatic mass and likely led to the diagnosis of pancreatic cancer (Table 1). More surprising, however, is that 67–83% of the selected scans obtained 2–6 months before clinical diagnosis of pancreatic cancer had CT findings diagnostic of or suspicious for pancreatic cancer (Table 2). Furthermore, 63% of the scans obtained 6–18 months before clinical diagnosis also showed findings diagnostic of or suspicious for pancreatic cancer (Table 2).
|
Even in scans obtained more than 18 months before clinical diagnosis, the two observers identified findings definite or suspicious for pancreatic cancer in 33% and 27% of the scans, respectively. In this group of patients, however, the results were susceptible to interobserver variability, with radiologists agreeing on the presence of tumor in only a single case. Within this time interval, the radiologists agreed that seven of the 15 cases did not show tumor, and disagreed on the presence of tumor in the other seven cases.
Figure 1 shows the radiologic findings identified by both radiologists on CT scans of pancreatic cancer patients within each time interval. Pancreatic duct dilatation was the only CT criterion that both radiologists identified on scans obtained more than 18 months before the diagnosis of pancreatic cancer (i.e., 2/15 patients [Fig. 1]). As time intervals approached the date of clinical diagnosis, duct dilatation was observed with increasing frequency in each time interval (Fig. 2A, 2B, 2C, 2D, 2E). With the exception of the loss of fatty marbling, the remaining CT criteria for pancreatic cancer were observed on scans obtained 6–18 months before clinical diagnosis (Fig. 3A, 3B, 3C, 3D). Of note is that none of the six scans reviewed in the 2- to 6-month category showed a mass that was identified by both radiologists; nevertheless, both radiologists agreed that three of these six scans revealed findings suspicious for pancreatic cancer because pancreatic duct cutoff was present.
|
|
|
|
|
|
|
|
|
|
When all scans in patients with pancreatic cancer were considered, observer 1 identified a mass in 27 (44%) of 62 scans, with a median size of 2.2 cm when present (mean ± SD = 3.0 ± 1.9 cm). Observer 2 identified a mass in 32 (52%) of 62 scans, with a median size of 2.0 cm when present (mean ± SD = 2.6 ± 1.1 cm).
In the control population, observer 1 had a specificity of 99% (88/89; 95%
CI: 94–100%) for the absence of pancreatic tumor, and observer 2 had a
specificity of 98% (87/89; 95% CI: 92–100%).
Table 3 indicates the observer
agreement for specific CT findings in all 151 interpreted scans included in
our study: the 62 scans in the 28 patients who developed pancreatic cancer and
the 89 scans in the 89 control patients. Radiologists most reliably agreed on
the presence of pancreatic duct dilatation, which showed almost perfect
interobserver agreement ( = 0.84). There was substantial agreement for
the presence of pancreatic duct cutoff ( = 0.76) and moderate agreement
for the presence of a pancreatic mass and vascular invasion ( = 0.51
and 0.58, respectively). Agreement regarding the loss of fatty marbling and
focal attenuation difference was slight and moderate ( = 0.12 and 0.41,
respectively).
|
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
The earliest finding consistently identified by both radiologists was
pancreatic duct dilatation, followed by pancreatic duct cutoff and a
pancreatic mass (Fig. 1).
Radiologists identified duct dilatation with near-perfect agreement ( =
0.84) and abrupt duct cutoff with substantial agreement ( = 0.76), so
these CT findings appear to be identified with a high degree of
reproducibility (Table 3).
Pancreatic duct dilatation as an early CT finding in pancreatic cancer
patients is consistent with the theory that pancreatic cancer arises from
pancreatic intraepithelial neoplasia
[21] as small intraductal
tumors that obstruct and dilate the duct without creating a visible pancreatic
mass. Early dysplasia not obstructing main or side-branch ducts is likely
occult on CT, as Brentnall et al.
[10] observed in their
examination of seven patients with familial pancreatic cancer found to have
dysplasia at surgery.
Ishikawa et al. [22] retrospectively examined all the minute (< 1 cm) pancreatic cancers reported in the Japanese literature up to 1994. Those researchers found that nearly 60% of the small adenocarcinomas in their study showed pancreatic duct dilatation without a mass on CT or sonography, whereas 15% or fewer showed a mass. In addition, the absence of a mass was significantly correlated with improved 5-year survival. In a case control study, Tanaka et al. [23] found pancreatic duct dilatation to be a sign indicating high risk for pancreatic cancer, but duct dilatation more than 2 mm was also present in 5% of the older control subjects. Radiologists should be aware that pancreatic duct dilatation accompanied with duct cutoff is often seen in early pancreatic cancer without a mass being present. The detection of duct dilatation and cutoff is associated with a high degree of reproducibility by attentive observers.
Our findings suggest that cross-sectional screening of patients with suspected pancreatic cancer should consequently be optimized to delineate the pancreatic duct (i.e., narrow collimation coupled with pancreatic and portal venous phase in CT, and MR cholangiopancreatography sequences such as single-shot fast spin-echo in MRI). In these patients, visualizing the duct to the ampulla without interruption is imperative. Because MRI and endoscopic sonography are sensitive to pancreatic duct cutoff and do not require ionizing radiation, an alternating regimen of CT with endoscopic sonography or MRI may prove beneficial by exploiting the strengths of each technique. Nevertheless, our study also shows that a minority of pancreatic tumors may be occult on CT—even if a patient is scanned less than 6 months before clinical presentation.
In our study, the median size of tumors presenting as a discrete mass was 2.2 and 2.0 cm measured on hard-copy CT images, by observers 1 and 2, respectively. The true size of tumors in our patient population is of course unknown because tumors were not removed near the time of each CT examination, but these findings do suggest that detection of cancer before clinical presentation may detect smaller tumors, which are more likely to be resectable [24, 25].
Regarding our control population, few false-positive interpretations for the presence of tumor were recorded. The impact of false-positive diagnoses on patient morbidity in a screening population would, of course, depend on the prevalence of disease and the evaluation algorithm (e.g., endoscopic sonography and biopsy vs laparotomy) after a positive CT finding.
We acknowledge several weaknesses in our study. It is retrospective, and small numbers of patients were scanned in each time interval. Of 151 scans reviewed, 62 were of pancreatic cancer patients. This high prevalence of pancreatic cancer in our study may have overestimated radiologists' sensitivity for the detection of pancreatic cancer before clinical diagnosis. In addition, the rereview of films by subspecialized radiologists is known to improve the estimate of tumor resectability [26] and may also improve the estimate of tumor detection. On the other hand, most CT scans in our study were not optimized for delineation of pancreatic abnormalities because most indications for the examinations were unrelated to screening for pancreatic cancer (Table 1). Six (10%) of the 62 scans in patients who developed pancreatic cancer were obtained without IV contrast material, seriously compromising the ability to detect pancreatic tumor in these patients and potentially causing an underestimation of the ability of CT to detect asymptomatic cancers. Finally, we excluded potential control patients with incidental pancreatic imaging abnormalities noted on their radiology reports because we did not have clinical follow-up to exclude the diagnosis of pancreatic cancer in these patients. Focal chronic pancreatitis can have imaging characteristics identical to pancreatic adenocarcinoma, and our control population did not include such patients. Our selection of control patients in this manner may have artifactually increased our specificity estimates.
Despite these weaknesses, our study shows that CT abnormalities are present before clinical presentation in a significant proportion of patients who develop pancreatic cancer. Both radiologists independently identified CT findings diagnostic of or suspicious for pancreatic adenocarcinoma in approximately 50% of cases more than 2–18 months before the clinical diagnosis. In addition, some asymptomatic pancreatic cancers can potentially be detected earlier if radiologists are vigilant in their assessment of the pancreas. In agreement with the results of earlier retrospective studies, our results indicate that pancreatic duct dilatation and cutoff are early findings in pancreatic cancer and that imaging protocols should be optimized to examine the pancreatic duct. With currently available MDCT scanners using dedicated pancreatic scanning with a narrow slice thickness and biphasic technique (as opposed to 7-mm slice thickness and generally monophasic examinations in this study), the accuracy of CT for the detection of pancreatic cancer before development of symptoms can likely be significantly improved.
Acknowledgments
We thank Karen A. Madsen for her assistance in the preparation of this
manuscript.
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
