ORIGINAL ARTICLE: Clinical Endoscopy
Comparison of carcinoembryonic antigen and molecular analysis in pancreatic cyst fluid Mandeep S. Sawhney, MBBS, MS, Shiva Devarajan, Paul O’Farrel, MD, Marcelo S. Cury, MD, Rabi Kundu, MD, Charles M. Vollmer, MD, Alphonso Brown, MD, MS, Ram Chuttani, MD, Douglas K. Pleskow, MD Boston, Massachusetts, USA
Background: Pancreatic-cyst fluid carcinoembryonic antigen (CEA) levels and molecular analysis are useful diagnostic tests in differentiating mucinous from nonmucinous cysts. Objective: To assess agreement between CEA and molecular analysis for differentiating mucinous from nonmucinous cysts. Design: Retrospective analysis. Setting: Academic medical center. Methods: Patients who underwent EUS-guided FNA for evaluation of pancreatic cysts were identified. The following information was used to designate a cyst mucinous: the CEA criterion was CEA level R192 ng/mL and the molecular analysis criteria were DNA quantity R40 ng/mL and/or k-ras 2-point mutation and/or R2 allelic imbalance mutations. Pathologic analysis of cysts served as the criterion standard. Results: From 2006 to 2007, 100 patients met the study criteria. The average age of the patients was 63 years, 65% were women, and 30% were symptomatic. The mean diameter of pancreatic cysts was 2.5 cm. The median CEA value was 83 ng/mL (range 1-50,000 ng/mL), the mean DNA content was 16 ng/mL (range 1-212 ng/mL), 11% had K-ras mutations, and 43% had R2 allelic imbalance mutations. When using prespecified criteria, there was poor agreement between CEA and molecular analysis for the classification of mucinous cysts (kappa Z 0.2). Poor agreement existed between CEA and DNA quantity (Spearman correlation Z 0.2; P Z .1), K-ras mutation (kappa Z 0.3), and R2 allelic imbalance mutations (kappa Z 0.1). Of the 19 patients for whom a final pathologic diagnosis was available, CEA had a sensitivity of 82% compared with 77% for molecular analysis. When CEA and molecular analysis were combined, 100% sensitivity was achieved. Limitations: Retrospective analysis and small sample size. Conclusion: There was poor agreement between CEA levels and molecular analysis for diagnosis of mucinous cysts. Diagnostic sensitivity was improved when results of CEA levels and molecular analysis were combined. (Gastrointest Endosc 2009;69:1106-10.)
The widespread use of cross-sectional imaging has resulted in a large increase in the number of patients diagnosed with pancreatic cysts.1 To help guide evaluation
Abbreviations: CEA, carcinoembryonic antigen; EUS-FNA, EUS-guided FNA. DISCLOSURE: All authors disclosed no financial relationships relevant to this publication. Copyright ª 2009 by the American Society for Gastrointestinal Endoscopy 0016-5107/$36.00 doi:10.1016/j.gie.2008.08.015
1106 GASTROINTESTINAL ENDOSCOPY Volume 69, No. 6 : 2009
and treatment, pancreatic cysts may be categorized into mucinous or nonmucinous cysts.2 Mucinous cysts, like mucinous cyst neoplasms and intraductal papillary mucinous neoplasms, are premalignant or malignant at the time of diagnosis. Such lesions require either immediate resection or the need to be carefully observed with serial imaging or sampling. Nonmucinous-cyst–like serous cystadenomas and pseudocysts have low or no malignant potential, and resection is necessary only when these lesions cause symptoms. Although imaging and conventional cytology are unable to differentiate mucinous from nonmucinous cysts, recent studies showed that carcinoembryonic antigen www.giejournal.org
Sawhney et al
(CEA) levels and molecular analysis of cyst fluid may be useful.3,4 However, neither cyst-fluid CEA levels nor molecular analysis achieve a high diagnostic accuracy. The aim of our study was to compare CEA levels and molecular analysis of cyst fluid with each other and to determine whether combining them would improve overall accuracy. We hypothesized that high concordance between CEA levels and molecular analysis in classifying cysts would suggest that these 3 methods measured the same aspect of cyst biology and that combining them would not provide additional information. A high discordance for the classification of cysts, however, would allow for the possibility of combining results of CEA levels and molecular analysis for classifying pancreatic cysts.
Carcinoembryonic antigen and molecular analysis
Capsule Summary What is already known on this topic d
Imaging and conventional cytology cannot differentiate mucinous from nonmucinous pancreatic cysts.
What this study adds to our knowledge d
d
d
There was poor agreement between carcinoembryonic antigen (CEA) levels and molecular analysis for the classification of mucinous pancreatic cysts. CEA level had a sensitivity of 82% compared with 77% for molecular analysis. The combination of the 2 studies resulted in 100% sensitivity.
PATIENTS AND METHODS Methods We searched our institution’s EUS database from 2006 to 2007 for all patients referred for EUS for evaluation of pancreatic cysts. Only those who underwent EUS-guided FNA (EUS-FNA) of at least 1 pancreas cyst were included. Those patients referred specifically for endoscopic cystgastrostomy were excluded from the study. Intravenous antibiotics were administered to all patients before cyst aspiration. EUS was performed by using the standard technique.5 FNA of cyst fluid was performed by using a 22gauge or a 25-gauge needle. The cyst aspirate obtained was evaluated for CEA level, cytology, and molecular analysis. The CEA level was measured by a standardized sandwich technique using 2 incubations of CEA monoclonal antibodies on a Roche-Hitachi modular system (Roche Diagnostics Corporation, Indianapolis, Ind). Based on the results of the Cooperative Pancreatic Cyst Study, a CEA level of R192 ng/mL was considered diagnostic for a mucinous cyst.3 Molecular analysis on cyst fluid was performed by a commercial laboratory (Redpath Integrated Pathology, Inc, Pittsburgh, Pa) by using methods described by Khalid et al.6 Molecular analysis of cyst fluid consisted of DNA quantification, K-ras 2-point mutation analysis, and allelic imbalance analysis. The presence of any one of the following was used to designate a cyst mucinous by molecular analysis criteria: DNA quantity R40 ng/mL, the presence of a K-ras 2-point mutation, or the presence of R2 specified allelic imbalances. Cytology specimens were analyzed by a staff cytopathologist. Specimens that were interpreted as atypical or suspicious for malignancy were considered benign for purposes of analysis. When histologic specimens of pancreatic cysts were available, World Health Organization criteria were used to categorized cysts as mucinous or nonmucinous.7 Cysts associated with main duct intraductal papillary mucinous neoplasms were considered mucinous. The Cohen kappa was used to quantify agreement between cyst categorization by CEA level and molecular analysis. A kappa value of 1.0 represents perfect agreement www.giejournal.org
and 0 represents the agreement expected from chance. On this scale, values greater than 0.75 are considered excellent agreement, values from 0.40 to 0.75 indicate fair to good agreement, and values less than 0.40 indicate poor agreement. The Spearman coefficient of correlation was used to assess monotonic association between continuous variables. We decided a priori that data would be analyzed once 100 patients had undergone EUS-FNA of pancreatic cysts. Our institutional review board approved this study.
RESULTS During the study period, 111 patients were referred for EUS-FNA for evaluation of pancreatic cysts. In 11 patients, FNA was not performed because cysts could not be identified, an avascular path for needle aspiration could not be found, or patients were on anticoagulants. Baseline characteristics for study subjects who underwent EUS-FNA are shown in Table 1. Of note, most patients were asymptomatic, and pancreatic cysts were discovered at imaging that was performed for other reasons. On EUS imaging, the mean long-axis diameter was 2.5 cm (range 0.8-14 cm) and the mean short-axis diameter of pancreatic cysts was 1.9 cm (range 0.5-12 cm). In 23 patients, more that 1 cyst was noted in the pancreas, and, in 19 patients, communication between the cyst and the pancreatic duct was noted during EUS. Results of cytology and molecular analysis were available for all 100 patients who underwent EUS-FNA. Results of CEA were unavailable for 16 patients because of insufficient fluid quantity (n Z 13) or very high fluid viscosity (n Z 3). Of the 84 patients for whom results of both pancreas-cyst–fluid CEA and molecular analysis were available, the median CEA level was 83 ng/mL (mean [SD] 1514 5734 ng/mL, range 1-50,000 ng/mL), the median DNA content was 4.9 ng/mL (mean [SD] 15.9 33.56 ng/mL, range 0.6-212.2 ng/mL), 11% had K-ras mutations, and 43% had R2 allelic imbalance mutations. Cytology was reported as acellular or benign in 90 patients, atypical in 7 patients, Volume 69, No. 6 : 2009 GASTROINTESTINAL ENDOSCOPY 1107
Carcinoembryonic antigen and molecular analysis
Sawhney et al
TABLE 1. Baseline characteristics Variable
N Z 100
TABLE 2. Correlation between CEA and molecular analysis for classification of cysts into mucinous and nonmucinous
Patient Mean (SD) age (y) Women History of chronic pancreatitis Symptomatic* (n Z 81)
Association 63 14 65 5 24 (30%)
Cyst Long-axis diameter (mm), mean (SD)
25 22
Short-axis diameter (mm), mean (SD)
19 19
Presence of nodule or mass
17
Presence of septations
44
Presence of debris Location pancreas heady Uncinate pancreas
9
95% CI
P value
DNA concentration
r Z 0.17
N/A
.11
K-ras mutation
k Z 0.27
0.03-0.50
N/A
Allelic imbalance
k Z 0.1
0.10 to 0.30
N/A
Age
r Z 0.21
N/A
.06
Cyst diameter
r Z 0.08
N/A
.45
N/A, Not applicable.
TABLE 3. Test characteristics for differentiating between mucinous and nonmucinous cysts when using CEA and molecular analysis
34 9
Sensitivity (%)
Specificity (%)
Accuracy (%)
Pancreas body
39
CEA*
82
100
84
Pancreas tail
18
Molecular analysisy
77
100
79
K-ras mutation
11
100
21
Allelic imbalance
70
100
73
DNA concentrationz
29
100
37
*Assessment of symptoms was possible for 81 patients. yLocation of largest cyst when O1 cyst was present.
suspicious for malignancy in 2 patients, and diagnostic of adenocarcinoma in 1 patient. When using a CEA cutoff of 192 ng/mL, 33 cysts were designated as mucinous and 51 as nonmucinous. When using prespecified molecular analysis criteria, 49 cysts were designated as mucinous and 35 as nonmucinous. Agreement between CEA and molecular analysis was noted in 59.5% of patients, whereas agreement simply from chance was expected in 48.2%. The kappa statistic that compared agreement between CEA and molecular analysis was computed at 0.22 (95% CI, 0.02-0.42), which suggests that there was poor agreement between these 2 methods for classification of cysts as mucinous or nonmucinous. Agreement between CEA and K-ras mutation analysis is shown in Table 2; also, CEA and allelic imbalance was poor. CEA levels were then modeled as a continuous variable, and the Spearman correlation coefficient was used as a test for monotonic association. No statistically significant association was noted between the CEA level and DNA quantity, cyst diameter, or patient age. A final histologic diagnosis for pancreatic cysts was available for 19 patients (Table 3). Two patients were noted to have serous cystadenomas (nonmucinous cysts). Of the 17 patients with mucinous cysts, 5 had malignant cysts (adenocarcinoma), 4 cysts were associated with main-duct intraductal papillary mucinous neoplasm, 2 cysts were side-branch intraductal papillary mucinous neoplasms, and 6 cysts were mucinous-cyst neoplasms. Both CEA and molecular analysis correctly classified the 2 patients with
CEA is a glycoprotein found in the embryonic endodermal epithelium. GI cancers, eg, colon cancer and
1108 GASTROINTESTINAL ENDOSCOPY Volume 69, No. 6 : 2009
www.giejournal.org
*For CEA level of 192 ng/mL. yWhen using criteria specified in the Methods section. zFor DNA concentration O40 ng/mL.
nonmucinous cysts. CEA correctly classified 14 of 17 patients with mucinous cysts. CEA had a sensitivity of 82.4%, specificity of 100%, positive predictive value of 100%, and negative predictive value of 40% for the diagnosis of mucinous cyst. Molecular analysis correctly classified 13 of 17 patients with mucinous cysts. Molecular analysis had a sensitivity of 76.5%, specificity of 100%, positive predictive value of 100%, and a negative predictive value of 33% for the diagnosis of mucinous cyst. If CEA and molecular analysis were combined in that cysts with either CEA level R192 ng/dL or meeting molecular analysis criteria were classified as mucinous, then all mucinous cysts were correctly identified (sensitivity 100% and specificity 100%). Results of CEA level, molecular analysis, and other characteristics for patients with malignant and nonmalignant cysts are shown in Table 4. Malignant cytology was noted in only 1 of 5 patients with malignant cysts.
DISCUSSION
Sawhney et al
Carcinoembryonic antigen and molecular analysis
TABLE 4. Comparison of malignant and benign cysts Malignant cyst (n Z 5) Mean (SD) age (y) Short-axis diameter (mm), mean (SD) Long-axis diameter (mm), mean (SD) Nodule/mass (%)* CEA (ng/mL)
69.8 5.8 60 48.3 30.9 28 100
Benign cyst (n Z 14) 60.116.1 49.6 42.8 24.9 24.2 21.20
2130 2665.9 2893.8 3354
DNA R 40 ng/mL (%)
40
21.20
K-ras mutation (%)
20
7.10
100
50
20
0
Allelic imbalance (%) Malignant cytology (%)
*Intracyst nodule or mass noted on EUS imaging.
pancreatic cancer, acquire and express cell-surface determinants that recapitulate structures expressed during fetal development.8 The rational for using CEA levels to differentiate mucinous from nonmucinous cysts is that mucinous cysts are lined by endoderm-derived columnar epithelium capable of secreting CEA, whereas nonmucinous cysts, eg, serous cysts adenomas, are lined by simple cuboidal epithelium (not derived from endoderm) and should contain little or no CEA. The basis for molecular analysis of cyst fluid relies on amplification and analysis of DNA from whole or lysed cells lining the cyst that have been shed into the cyst fluid. We used 3 aspects of molecular analysis in our study as recommended by other investigators: DNA concentration, K-ras mutation, and allelic imbalance (loss of heterozygosity).6,9 The concentration of DNA present in the cyst fluid reflects cell division in the lining cells.6 A high level of DNA suggests rapid malignant multiplication of cells but could also be seen when there is hemorrhage into the cyst or from acute pancreatitis that causes cellular breakdown and release of DNA into cyst fluid. K-ras is a proto-oncogene that plays a critical role in the transition of normal pancreatic-duct epithelium to intraductal adenocarcinoma and is the most commonly mutated gene in pancreatic cancer.10 We assessed for Ras mutations that occur in codons 12 and 13, in which glycine is substituted by another amino acid. Allelic imbalance or loss of heterozygosity measures genomic deletions that inactivate tumor suppressor genes. The panel used in our study targeted tumor suppressor genes located at 1p, 3p, 5q, 9p, 10q, 17p, 17q, 21q, and 22q.6 Mucinous cysts demonstrate higher CEA levels than nonmucinous cysts, with significant overlap between the two. Sperti et al11 analyzed 48 pancreatic cysts and found that the median CEA level in mucinous cysts was 1558 ng/ mL (range 2.7–6400 ng/mL) versus 1.9 ng/mL (range 1–15 www.giejournal.org
ng/mL) in serous cyst adenomas. A CEA value of 500 ng/mL had a sensitivity of 38% and a specificity of 100% for the diagnosis of mucinous cysts. In a large multicenter trial of EUS-FNA of pancreatic cysts, Brugge et al3 found a CEA level of 192 ng/mL to have a sensitivity of 73%, specificity of 84%, and accuracy of 79% for the diagnosis of mucinous cysts. The addition of cyst morphology, cytology, or other tumor markers to the CEA value did not improve diagnostic accuracy. We used a CEA cutoff of 192 ng/mL and found a sensitivity of 82%, specificity of 100%, and accuracy of 84% for the diagnosis of mucinous cysts. In 16% of patients, CEA analysis could not be performed because of insufficient quantity or high viscosity of cyst fluid. Molecular analysis, however, can be performed even on small quantities of aspirate (!1 mL) but, when compared with CEA analysis, is more expensive and can only be done at specialized laboratories. Further, the overall accuracy of molecular analysis was similar to that of CEA. The recently concluded pancreatic-cyst fluid molecular analysis study (PANDA study) found that the area under the curve for differentiating mucinous from nonmucinous cysts was 0.67 for the number of DNA mutations, 0.79 for mean allelic loss amplitude, and 0.74 for CEA level.4 There was no statistically significant difference between mucinous and nonmucinous cysts with regard to concentration of DNA. K-ras mutations were noted in 1 of 25 nonmucinous cysts (4%) compared with 40 of 88 mucinous cysts (45%) (P ! .001), which resulted in a sensitivity of 45% and a specificity of 96%. In our study, we found that DNA concentration, K-ras mutation, and allelic imbalance had accuracies of 37%, 21%, and 73%, respectively. When all aspects of molecular analysis were combined, the overall accuracy increased to 79%. When compared with each other, we found poor agreement between CEA level and molecular analysis for classification of pancreatic cysts into mucinous and nonmucinous. CEA levels also correlated poorly with DNA concentration, K-ras mutations, and allelic imbalance. This was probably because CEA levels depend upon the mucinouscyst lining secreting tumor markers into the cyst fluid, whereas molecular analysis depended upon cells lining the cyst acquiring mutations known to be associated with pancreatic neoplasms. Although this high degree of discordance calls into question the accuracy of CEA and molecular analysis, it also suggests that these 2 methods may be complementary. In a small number of patients for whom we had final pathologic diagnosis, we found that the combination of CEA level and molecular analysis achieved a 100% accuracy for the diagnosis of mucinous cysts. When combining CEA levels and molecular analysis, we chose to designate a cyst as mucinous if it had a CEA level R192 ng/mL or met any of the molecular analysis criteria. This was done to maximize the sensitivity of the combined test for the diagnosis of a mucinous cyst. Once a pancreatic cyst is deemed mucinous, the next question is: has it undergone malignant transformation? Volume 69, No. 6 : 2009 GASTROINTESTINAL ENDOSCOPY 1109
Carcinoembryonic antigen and molecular analysis
Sawhney et al
Some authorities suggest that this distinction is clinically unimportant because, in general, all mucinous cysts should be considered for surgical resection. However, for patients who are at increased risk for postsurgical complications, this information can be extremely valuable. Several studies, including ours, demonstrated that CEA levels do not reliably differentiate benign from malignant mucinous cysts.4 Results of cyst-fluid molecular analysis are more encouraging. In a study of 36 patients, early K-ras mutation, followed by allelic imbalance, had a sensitivity of 91% and a specificity of 93% for the diagnosis of malignant cysts.6 In a larger study, the same group reported sensitivity of 36% and a specificity of 94%.4 We found K-ras mutations in only 1 of 5 patients with malignant cysts. All of our patients with malignant cysts demonstrated allelic imbalance; however, so did 50% of those with benign cysts. Several investigators suggested that cysts that are %3 cm in size are unlikely to be malignant and can be followed-up safely with close observation.12,13 Two of 5 malignant cysts in our study were !30 mm in size. One cyst had a CEA level of 863 ng/mL, no K-ras mutation, and 2 allelic imbalance mutations. The other cyst had a CEA level of 5 ng/mL, no K-ras mutation, and 2 allelic imbalance mutations. This again suggests that the addition of cyst-fluid analysis to cyst size and other morphologic details may improve our diagnostic ability. Several limitations of our study warrant further discussion. First, a final diagnosis was available for only 19 patients. This small sample size resulted in imprecise estimates of test accuracy and limited our ability to analyze test characteristics of different levels of CEA and different variations of the molecular analysis criteria. Second, the retrospective nature of our analysis did not allow us to fully measure the impact CEA levels or DNA analysis had on clinical decision making. In summary, we found poor agreement between the CEA level and molecular analysis for differentiating mucinous from nonmucinous cysts. When used singly, CEA levels and molecular analysis have limited accuracy in differentiating mucinous from nonmucinous cysts; when results of these tests were combined, high sensitivity and specificity were achieved. Larger studies are needed to
confirm these results and to assess their utility in patient management.
1110 GASTROINTESTINAL ENDOSCOPY Volume 69, No. 6 : 2009
www.giejournal.org
REFERENCES 1. Spinelli K, Fromwiller T, Daniel R, et al. Cystic pancreatic neoplasms: observe or operate. Ann Surg 2004;239:651-7. 2. Brugge W, Lauwers G, Sahani D, et al. Cystic neoplasms of the pancreas. N Engl J Med 2004;351:1218-26. 3. Brugge W, Lewandrowski K, Lee-Lewandrowski E, et al. Diagnosis of pancreatic cystic neoplasms: a report of the cooperative pancreatic cyst study. Gastroenterology 2004;126:1330-6. 4. Khalid A, Finkelstein S, Leblanc J, et al. Pancreatic cyst fluid DNA analysis detects malignant cysts: final report of the PANDA study [abstract]. Gastrointest Endosc 2007;65:AB102. 5. Erickson R. EUS-guided FNA. Gastrointest Endosc 2004;60:267-79. 6. Khalid A, McGrath K, Zahid M, et al. The role of pancreatic cyst fluid molecular analysis in predicting cyst pathology. Clin Gastroenterol Hepatol 2005;3:967-73. 7. von Kloppel G, Solcia E, Longnecker D, et al. Histological typing of tumours of the exocrine pancreas. Berlin: Springer-Verlag; 1998. 8. Carethers J, Boland C. Neoplasia of the gastrointestinal tract. In: Yamada T, Alpers D, Laine L, et al, editors. Textbook of gastroenterology, vol. 1. Philadelphia: Lippincott, Williams and Wilkins; 1999. 9. Schoedel K, Finkelstein S, Ohori N. K-Ras and microsatellite marker analysis of fine-needle aspirates from intraductal papillary mucinous neoplasms of the pancreas. Diagn Cytopathol 2006;34:605-8. 10. Gansauge S, Gansauge F, Beger H. Molecular oncology in pancreatic cancer. J Mol Med 1996;74:313-20. 11. Sperti C, Pasquali C, Guolo P, et al. Serum tumor markers and cyst fluid analysis are useful for the diagnosis of pancreatic cystic tumors. Cancer 1996;78:237-43. 12. Tanaka M, Chari S, Adsay V, et al. International consensus guidelines for management of intraductal papillary mucinous neoplasms and mucinous cystic neoplasms of the pancreas. Pancreatology 2006;6:17-32. 13. Sahani D, Saokar A, Hahn P, et al. Pancreatic cysts 3 cm or smaller: how aggressive should treatment be? Radiology 2006;238:912-9.
Received May 7, 2008. Accepted August 11, 2008. Current affiliations: Divisions of Gastroenterology (M.S.S., S.D., P.O., M.S.C., R.K., A.B., R.C., D.K.P.) and General Surgery (C.M.V.), Beth Israel Deaconess Medical Center and Harvard Medical School, Boston, Massachusetts, USA. Reprint requests: Mandeep S. Sawhney, MBBS, Division of Gastroenterology, Beth Israel Deaconess Medical Center 330 Brookline Ave, RABB-ROSE 101, Boston, MA 02215. If you want to chat with an author of this article, you may contact him at
[email protected] or
[email protected].