Diagnostic evaluation of cystic pancreatic lesions

Abstract

Background. Cystic pancreatic neoplasms (CPNs) present a unique challenge in preoperative diagnosis. We investigated the accuracy of diagnostic methods for CPN. Material and methods. This retrospective cases series includes 70 patients who underwent surgery at a university hospital for presumed CPNs between 1997 and 2003, and for whom a definitive diagnosis was established. Variables examined included symptoms, preoperative work-up (including endoscopic retrograde cholangiopancreatography (ERCP) in 22 cases and endoscopic ultrasound (EUS) in 12), and operative and pathological findings. Preoperative computed tomography (CT) and magnetic resonance imaging (MRI) scans (n=50 patients; CT=48; MRI=13) were independently reviewed by two blinded GI radiologists. Results. The final histopathologic diagnoses were mucinous cystic neoplasm (n=13), mucinous cystadenocarcinoma (10), serous cystadenoma (11), IPMN (14), simple cyst (3), cystic neuroendocrine tumor (5), pseudocyst (4), and other (10). Overall, 25 of 70 were malignant (37%), 21 premalignant (30%), and 24 benign (34%). The attending surgeon's preoperative diagnosis was correct in 31% of cases, incorrect in 29%, non-specific “cystic tumor” in 27%, and “pseuodcyst vs. neoplasm” in 11%. Eight had been previously managed as pseudocysts, and 3 pseudocysts were excised as presumed CPN. In review of the CT and MRI, a multivariate analysis of the morphologic features did not identify predictors of specific pathologic diagnoses. Both radiologists were accurate with their preferred (no. 1) diagnosis in <50% of cases. MRI demonstrated no additional utility beyond CT. Conclusions. The diagnosis of CPN remains challenging. Cross-sectional imaging methods do not reliably give an accurate preoperative diagnosis. Surgeons should continue to err on the side of resection.

Introduction

Cystic pancreatic neoplasms (CPNs), whether congenital, inflammatory, or neoplastic, are diagnostically challenging. They account for 10–15% of cystic lesions of the pancreas and 5% of primary pancreatic neoplasms ¹,². Many reports show that the prevalence is rising. They are commonly found incidentally on computed tomography (CT) scans ³ and represent a growing indication for resection at referral centers ¹,⁴. The diagnosis relies principally on CT and magnetic resonance imaging (MRI). Efforts to differentiate among these tumors from imaging tests have met with mixed success ⁵,⁶. Although certain features have been emphasized as being classic for specific kinds of CPN, the diagnostic power of these features has not been subjected to objective analysis. Endoscopic ultrasound (EUS) is now being used to investigate cystic pancreatic lesions, particularly as a means of cyst aspiration ⁷. How analysis of cyst fluid will strengthen the diagnostic algorithm remains unsettled ⁸. The aim of this study was to investigate the preoperative diagnostic evaluation of cystic pancreatic lesions in determining how imaging and clinical factors should guide management.

Methods

Using the University of California San Francisco (UCSF) Department of Pathology database, we retrospectively identified 70 patients who between 1997 and 2003 underwent surgical exploration for presumed CPNs, or who were ultimately found to have a CPN at operation (e.g. the preoperative diagnosis was pseudocyst but the lesion proved to be a cystic neoplasm). The histopathology of all lesions was established by resection (n=60) or surgical biopsy (n=10). Patients were excluded who had transcutaneous or endoscopic fine-needle aspiration (FNA) alone (without subsequent surgical exploration) or if the pathology results did not give a specific diagnosis. A pathologic diagnosis of malignancy was based on invasion of the pancreatic parenchyma or metastases. Mucinous cystic neoplasms and intraductal papillary mucinous tumors (IPMTs) that did not have conclusive evidence of carcinoma were considered to be premalignant ⁹,¹⁰. Increasingly, pathologists use the term ‘mucinous cystic neoplasm’ for all mucinous tumors in order to emphasize the overlap in malignant potential of these tumors. Alternatively, they are divided into three categories: adenoma, borderline, and adenocarcinoma. We prefer the standard classification, dividing mucinous tumors into just cystadenoma and cystadenocarcinoma with few exceptions, which more accurately reflects how the patients are managed clinically and their outcome ¹¹. Though there are instances in which patients with completely resected mucinous cystadenomas have developed metastatic disease, these are rare ⁹.

The preoperative data included age at diagnosis, presenting symptoms, and the results of diagnostic tests. Details of the operations were obtained from the operative notes and pathology reports, including the extent of excision, perioperative course, complications, and outcome.

Preoperative CT and MRI scans that were available for review (n=50 patients; CT = 48; MRI = 13) and scans from patients with pathologically proven pseudocysts (CT = 5, MRI = 5) were analyzed. Fourteen of these scans had been performed at other institutions. Two abdominal radiologists, blinded to all demographic and clinical data, independently reviewed the CT and MRI images at separate sessions. Each reader recorded 22 separate morphologic tumor characteristics, the probability of specific diagnoses, and the overall likelihood of malignancy. Multivariate analysis (manual stepwise regression using backward elimination) was used to investigate morphologic characteristics as predictors of malignancy and specific diagnoses (using ANOVA for continuous variables and the chi-squared test for discrete characteristics). The statistical analyses were performed using the software packages Stata Version 8.0 (Stata Corp, College Station, Tx., USA) and StatXact (Cytel Software, Cambridge, Mass., USA).

This study was reviewed and approved by the UCSF Committee on Human Research.

Results

Seventy patients underwent laparotomies for presumed CPN or were found to have a CPN at an operation where the preoperative diagnosis was something other than CPN (e.g. pseudocyst). There were 48 women and 22 men and the mean age was 60 (±16) years.

Seventy-nine operations were performed: distal pancreatectomy (open or laparoscopic) (n=28), Whipple pancreaticoduodenectomy (21), biopsies with or without palliative bypass (10), enucleation (6), total or subtotal pancreatectomy (5), and cystenterostomy (7). Six patients had more than one operation. Five patients with cystic neoplasms initially had cystenterostomies (twice in one patient) for presumed pseudocysts before the lesion was resected. A patient who had previously undergone a distal pancreatectomy for a cystic neoplasm underwent a Whipple completion pancreatectomy for a presumed recurrence in the head of the gland. On pathologic examination, the second lesion proved to be a pseudocyst.

The final pathologic diagnoses for all patients are listed in Table I. Overall, 25 (36%) of 70 were malignant, 21 (30%) were premalignant, and 24 (34%) were benign. Fourteen (19%) were discovered incidentally in asymptomatic patients; of these, 3 (21%) were malignant, 6 (43%) were premalignant, and 5 (36%) were benign.

Table I. Histopathologic diagnoses of patients preoperatively diagnosed with cystic pancreatic neoplasms.

Diagnosis	All	Premalignant	Malignant
Mucinous cystadenoma	13	13
Mucinous cystadenocarcinoma	10		10
Intraductal papillary mucinous tumor	14	7	7
Serous cystadenoma	11
Neuroendocrine tumor	5		3
Pseudocyst	4
Simple cyst	3
Other	10	1	5
Cystic dilation due to duct obstruction
Mucin-producing ductal adenocarcinoma
Anaplastic pancreatic carcinoma
PanIN 3 (carcinoma-in-situ), chronic pancreatitis
Lymphoepithelial cyst
Lymphangioma
Solid pseudopapillary neoplasm
Cystic gastric duplication
Total	70	21	25

Laboratory values contributed little to the preoperative diagnosis. Hyperamylasemia (>2× the upper limit of normal) was present in three patients: one had an islet cell tumor and two had mucinous cystic neoplasms. Four patients had elevated bilirubin levels. Their diagnoses were as follows: mucinous cystadenoma of the pancreatic head, IPMN with adenocarcinoma, mucin-producing ductal adenocarcinoma, chronic sclerosing pancreatitis with PanIN-3. No laboratory tests were predictive of any individual pathologic lesion or the presence of carcinoma.

Preoperative evaluation included CT and/or MRI scans in all patients and endoscopy in 34. Endoscopic retrograde cholangiopancreatography (ERCP) performed in 22 patients demonstrated mucin at the ampulla and dilatation of the pancreatic duct in all of the 6 patients with IPMT who underwent ERCP. In two cases, mucinous cystic neoplasms were found to communicate with the pancreatic duct, and this finding was interpreted erroneously by the endoscopist as being diagnostic of pancreatic pseudocyst. The other studies were normal (4 patients), technically unsuccessful (2 patients), demonstrated non-specific irregularity or ectasia of the pancreatic duct (4 patients), or showed dilatation of the bile duct with a normal pancreatic duct (4 patients).

Twelve patients underwent EUS. The correct diagnosis was suggested in two of these cases by their imaging features and in four by cyst aspiration (for cytology in one and carcinoembryonic antigen (CEA) in three). EUS proved non-diagnostic in four cases, and the interpretation of the findings was incorrect in two other cases.

All patients had preoperative cross-sectional imaging, including CT scans in 69 and MRI in 13. The preoperative radiology reports were available for 47 of the scans and, among these, the radiologist's diagnosis was correct in 10 (21%) and incorrect in 15 (32%). In 18 (38%) reports the final pathologic diagnosis was within the radiologist's differential diagnosis, while no diagnosis was offered in 4 (9%) cases.

CT and MRI images were available for re-review for 50 of the 70 patients (CT = 48; MRI = 13). Scans from patients with pathologically proven pseudocysts were included in the blinded analysis (CT = 5, MRI = 5) as controls.

With respect to the readers’ descriptions of the specific morphologic characteristics of each scan or image, inter-observer agreement for continuous and binary outcomes was good (Pearson's correlation coefficient: 0.73–0.90). Overall diagnostic accuracy and likelihood of predicting malignancy were not significantly different between the two readers (p=0.86 and 0.63, respectively; McNemar's chi-squared test).

Multivariate analysis was performed using a model that included 22 separate variables (size, locularity, lesion wall thickness, presence of solid nodules, calcification, number of locules, septal thickness, etc.; see Appendix I). Individual morphologic characteristics proved to be poor predictors of malignancy and of specific histopathologic diagnoses. Overall, the heterogeneity among cystic lesions and significant overlap in morphologic characteristics between the different diseases made it impossible to identify features that consistently predicted specific diagnoses or reliably distinguished benign from malignant and premalignant lesions.

Combining the CT and MRI data, Reader 1 was correct in 46%, 61%, and 66% of cases with respect to his number 1 diagnosis, top 2 diagnoses, and top 3 diagnoses, respectively. Reader 2's accuracy was comparable (43%, 64%, and 67%). The results of CT and MRI diagnoses were of similar accuracy for both readers (Table II). To investigate the possibility that a subset of the lesions was causing diagnostic confusion, we looked at those for which both readers had the same first diagnosis, presumably the more typical lesions. The readers’ first diagnosis was the same in 28 (40%) of 70 scans, but these diagnoses were correct in only 16 (57%) instances. In cases where Reader 1 gave the correct diagnosis, Reader 2 agreed in only 53%; and if Reader 2 was correct, Reader 1 agreed in 55%. We also determined the accuracy of diagnosis when stratified according to the readers’ certainty of their leading diagnosis. The accuracy of their number 1 diagnosis improved to just 55% and 48% (readers 1 and 2, respectively) as the reader's certainty increased to ≥90%. This questions the notion that certain typical features are diagnostic.

Table II. Accuracy of CT and MRI for Readers 1 and 2.

	Reader 1			Reader 2
Correct re:	CT	MRI	p- value	CT	MRI	p- value
No. 1 diagnosis	44%	50%	ns	40%	50%	ns
Top 2 diagnoses	62%	62%	ns	60%	78%	ns
Top 3 diagnoses	65%	67%	ns	62%	83%	ns

It is the surgeon who synthesizes the clinical and imaging findings to make the best possible diagnostic estimate. We investigated the accuracy of these preoperative diagnoses, which were available for 65 of the 70 patients. The surgeon's preoperative diagnosis was correct in 21 (32%) cases, incorrect in 20 (31%), “cystic tumor” (non-specific) in 16 (25%), and “neoplasm versus pseudocyst” in 8 (12%).

Eight patients were managed inappropriately because the preoperative diagnosis was incorrectly thought to be pseudocyst (Table III). Five had cystenterostomies before the correct diagnosis was made and the cystic neoplasm was removed (one at UCSF, 4 at community hospitals). Two patients were managed unsuccessfully with pancreatic duct stents for lengthy periods by gastroenterologists on the assumption that they were treating pseudocysts. One patient with abdominal pain and a cystic lesion in the head of the pancreas was observed for three years. After losing 35 pounds, an operation was performed and unresectable mucinous cystadenocarcinoma was found.

Table III. Patients misdiagnosed as pseudocysts.

n	Age	Initial management	Final operation	Final path
1	36	1. Cystgastrostomy 2. Cystjejunostomy	Distal pancreatectomy	MCN
2	59	Cystjejunostomy	Whipple	Neuroendocrine tumor
3	56	Cystgastrostomy	Distal pancreatectomy, partial gastrectomy and colectomy	Anaplastic carcinoma
4	41	ERCP w/ stent into ‘cyst’	Ex-lap, biopsy (unresectable)	Mucinous cyst-adenocarcinoma
5	46	Cystjejunostomy	Distal pancreatectomy	MCN
6	77	“Observation”×3 years	Biopsy, double bypass (unresectable)	Mucinous cyst-adenocarcinoma
7	45	ERCP w/ stent×2 years	Whipple	MCN
8	44	Cystjejunostomy	Distal pancreatectomy	MCN

MCN = mucinous cystadenoma; ex-lap = exploratory laparotomy.

In four patients, lesions thought preoperatively to be cystic tumors proved to be pseudocysts. Two underwent distal pancreatectomies. In one case, the lesion was recognized to be a pseudocyst during the operation, and a cystgastrostomy was performed. One patient underwent distal pancreatectomy for a solid and papillary epithelial neoplasm. On a follow-up CT scan he was found to have another cystic lesion in the head of the gland, and a Whipple procedure was performed for a presumed recurrence. The lesion was a pseudocyst.

Discussion

In 1959, L. Henry Garland delivered a lecture at the New York Academy of Medicine entitled “The Problem of Observer Error”, carefully outlining the diagnostic errors found in every aspect of the clinical evaluation of patients ¹². Garland's research focused on the rate of error in the interpretation of radiographs, as a result of both inter-observer variability and intra-observer variability (i.e. a single radiologist has a fixed rate of inconsistency on re-reading a set of films) ¹³. Garland argued that a measurable degree of human error was inherent to these diagnostic tests due to the “human equation”.

Now, nearly 50 years later, the clinical evaluation of cystic pancreatic lesions appears to suffer from the same limitations despite the improvement in the diagnostic tools. Distinguishing CPNs from pseudocysts and discriminating among the various CPNs is important to appropriate management. The consequences of mistaking a cystic tumor for a pseudocyst can be serious ¹⁴,¹⁵. Many asymptomatic lesions are premalignant or malignant ¹⁶. Given the morbidity of pancreatic resections, however, it is important to single out the lesions that may be managed expectantly.

We included all patients who had surgery for a suspected CPN or were ultimately found to have cystic neoplasms, rather than confining the analysis to proven CPN. Fourteen pathologic diagnoses were included (Table I). When performed in this broad context, the analysis more accurately shows the utility of each study. A test that could distinguish a serous from a mucinous cystadenoma is not sufficient, given the myriad possibilities encountered in referral centers.

In this series, no laboratory studies helped predict individual pathologic lesions or the presence of carcinoma. Serum tumor markers (CEA, Ca 19-9, Ca 125, etc.), though widely studied for pancreatic ductal adenocarcinoma ¹⁷,¹⁸, have demonstrated little value for the assessment of cystic neoplasms ¹⁹. Though hyperamylasemia suggests a diagnosis of pancreatic pseudocyst, it does not rule out a CPN. In this series, hyperamylasemia was seen in only three patients, all who had neoplasms. Other authors have similarly documented patients with cystic neoplasms presenting with hyperamylasemia or a clinical history of pancreatitis ²,⁹,²⁰.

Endoscopy was used selectively. ERCP proved most useful in patients with IPMN, demonstrating mucin at the ampulla and diffuse dilatation of the pancreatic duct ²¹. Otherwise, ERCP contributed little. Neither the appearance of the duct, nor evidence that the duct communicates with the cyst differentiated among the possibilities ²². Though pseudocysts often communicate with the duct, this finding also occurs in many cases of mucinous cystic neoplasms ²,¹⁴. This was seen in two of our patients with mucinous cystic neoplasms. The ERCP reports in both cases listed pseudocyst as the principal diagnosis. ERCP sampling of pancreatic secretions for cytology and tumor markers (e.g. K-ras) has been reported, but its validity is still unproved ²³.

EUS offers two means of diagnosis of cystic pancreatic lesions, i.e. morphologic imaging and guidance for FNA ²⁴,²⁵,²⁶,²⁷,²⁸,²⁹,³⁰. Twelve patients in this series underwent EUS, which suggested the correct diagnosis in two cases because of morphologic characteristics and in four by FNA. Characterization of morphology has shown little promise for differentiating cystic pancreatic lesions due to the overlap in features ³¹ and to wide inter- and intra-observer variability ³⁹. The sensitivity of cytology of pancreatic cyst aspirate is low ²⁹,³²,³³,³⁴. Cyst fluid analysis – for tumor markers (CEA, CA 19-9, CA-125), amylase, lipase, viscosity, and mucin – has been investigated with the aim of finding one or a panel of these tests that would differentiate among cystic pancreatic lesions. CEA appears to be the most useful, though the diagnostic cut-off varies ⁷,³⁴,³⁵. Some authors have reported considerable overlap in the results of cyst fluid analysis among CPN and between CPN and pseudocysts ³⁶,³⁷. False negatives can occur when the tumor communicates with the pancreatic duct and pancreatic juice dilutes cyst contents ¹⁴. Complications of FNA include the possibility of spilling malignant cells into the peritoneum ³⁸, hemorrhage, and injury to adjacent organs. We generally reserve FNA for lesions with a low risk of cancer, a poor risk patient with an indeterminate cystic mass in the head of the pancreas that would require pancreatoduodenectomy, or a patient with a probable mucinous tumor who desires additional evidence prior to resection. There is some evidence that EUS with FNA increases diagnostic accuracy over cross-sectional imaging alone ³⁹. Although in this series (through 2003) just 12 patients had EUS, this technology is now used frequently.

Thus, the evaluation of cystic lesions relies heavily on CT and MRI. Much of the available literature is comprised of descriptions of the archetypal imaging findings accompanying case reports ¹⁴,⁴⁰,⁴¹. Little systematic analysis has been done to validate these claims on the sensitivity or specificity of particular findings or on the broader diagnostic accuracy CT and MRI. Blinded analyses of CT scans from CPN have given mixed results, and they typically only included serous cystadenomas and mucinous cystic neoplasms ⁵,⁶,⁴². This does not mimic the diagnostic problem of the clinician, who must choose among a wider spectrum of possibilities.

Given that CT and MRI are the most useful diagnostic tests, we investigated their accuracy in our patients. The preoperative radiology reports were unreliable: correct in 21%, within the differential in 38%, incorrect in 32%, and no diagnosis in 9%. Blind interpretations of the scans by a pair of highly experienced abdominal radiologists were only slightly better. The readers’ first diagnoses were accurate in less than 50% of cases. Using a three-item differential diagnosis, both radiologists were correct in two-thirds of cases. MRI offered no additional utility over CT. Even when the readers agreed on the first diagnosis, and when the readers’ confidence was ≥90%, the accuracy improved to only 60%. This questions the widely held belief that there are “classic” features that possess great diagnostic weight.

We also looked for individual objective findings on the imaging studies (size, calcification, etc.) or a combination of findings that would prove more accurate. There was little inter-observer variability in the recording of morphologic features. However, the heterogeneity among cystic lesions and overlap between diagnoses prevented reliable predictors from emerging that consistently predicted the diagnosis or distinguished benign from malignant lesions. Of note, we did not find a clear size criterion that excluded malignancy.

In this series, 2 of 15 (13%) unilocular, thin-walled cysts =3 cm were malignant. This is somewhat discrepant with the results of a recent study from Massachusetts General Hospital in which 35 of 36 unilocular pancreatic cysts smaller than 3 cm were benign ⁴³. The reason for this discordance is unclear. In any case, the end-point of frank malignancy is not sufficient for clinical management because even though entities such as mucinous cystadenomas are benign, they are considered to be premalignant. Although criteria for selective observation are emerging ⁴⁴,⁴⁵,⁴⁶, these are hotly debated, and many experts still recommend resection rather than observation of suspected mucinous cystadenomas of the pancreas for patients who are reasonable candidates for surgery ⁴⁷.

Though the radiographic studies standing alone demonstrate significant shortcomings, the surgeon is in a position to integrate the radiographic findings with the clinical findings into a (presumably) more informed assessment. The surgeon's preoperative diagnosis was correct in roughly one-third of cases, incorrect in another third, and non-specific (e.g. cystic tumor; neoplasm vs pseudocyst) in the remainder. This is similar to what others have reported ²,⁴¹,⁴²,⁴³. For example, in a multicenter study by Le Borgne and colleagues of 164 serous cystadenomas and 222 mucinous tumors, the correct diagnosis was made preoperatively in 20% of serous cystadenomas, 30% of mucinous cystadenomas, and 29% of mucinous cystadenocarcinomas ².

The most important preoperative distinction is between CPN and pseudocysts. The epithelial lining of CPN may be partly denuded and misdiagnosed as a pancreatic pseudocyst even when examined by frozen section ⁴⁸, and the resulting error in management may lead to recurrent or metastatic disease ¹⁴,¹⁵,⁴⁹. Eight of our patients with CPN (12%) were initially managed as pseudocysts. In two cases, prolonged non-surgical management based on assumptions that the lesions were benign allowed cystadenocarcinoma to metastasize. Four out of five patients who mistakenly had cystenterostomies had benign tumors that were eventually cured by resection. The fifth eventually succumbed to metastatic disease.

Pathologic confirmation of the diagnosis is necessary to rigorously test the accuracy of each diagnostic modality. This may introduce a selection bias. Perhaps all of the benign lesions that have “classic” radiographic appearances (e.g. serous cystadenomas or pseudocysts) are being managed expectantly without surgery. Even if this is the case, it does not change the implications of this study, because the remaining lesions cannot be differentiated.

Conclusions

Cystic lesions of the pancreas often impersonate one another – cystic neoplasms as pancreatic pseudocysts, peripancreatic cystic lesions as pancreatic cysts, serous adenomas as mucinous neoplasms, IPMN as chronic pancreatitis. To some degree, they can be differentiated by clinical presentation and the diagnostic modalities discussed above. The accuracy of each of these diagnostic procedures is subject to errors in technique of examination and interpretation. As L. Henry Garland stated: “The former are correctable with care, and the latter partly correctable with training and experience” ¹². This series illustrates the persistent diagnostic dilemma presented by cystic pancreatic lesions. All symptomatic patients should undergo resection. A large portion of even asymptomatic lesions are premalignant or malignant. Therefore, while asymptomatic lesions with the radiologic appearance of serous cystadenomas may be followed with serial imaging, any change in appearance should prompt resection. Ultimately, the final recommendation must remain: in uncertain cases, err on the side of resection.

Appendix I. Morphologic features recorded for each CT and MR.

Imaging findings
Location
Size
Wall thickness
Loculation
No. of locules
Septal thickness
Average locule size
Largest locule size
Presence of solid nodules
Central scar
Calcification and location
Vascular encasement
Lesion communicated with pancreatic duct
(Predominantly) exophytic
MR signal intensity of cyst fluid (on T1)
Upstream pancreatic atrophy
Upstream pancreatic duct diameter
Biliary diameter
Evidence of acute pancreatitis
Pancreatic calcifications
Locoregional adenopathy
Hepatic or other metastases