Over the past decade, the standardization of error classification in anatomic pathology has become an important issue. The objective of the current study was to assess the extent of errors occurring in the cytopathologic diagnosis of neuroendocrine lesions of the pancreas, and to classify these errors and their associated harm.
Information on all cases diagnosed as a neuroendocrine neoplasm either by endoscopic ultrasound-guided fine-needle aspiration (EUS-FNA) in cytology or by surgical pathology between 2000 and 2012 was collected. Using standardized error and harm classification, the authors reviewed the cytology and surgical pathology material and evaluated the type and the cause of diagnostic errors and their impact on the patient.
A total of 177 patients who underwent EUS-FNA were diagnosed with a neuroendocrine neoplasm either by cytology or surgical pathology. Eighty of these cases had surgical follow-up available at the study institution. Of these 80 cases, 56 had an adequate cell block and immunohistochemistry was performed. There were 14 discrepancies noted between cytologic and surgical pathologic diagnoses. There were 9 false-negative cases, consisting of 3 interpretation errors and 6 cytology sampling errors. There were 5 misclassifications, including 4 cases of solid pseudopapillary neoplasm and 1 case of neuroendocrine carcinoma (diagnosed as adenocarcinoma on cytology). There were no surgical pathology errors noted. All errors were associated with no or minor harm.
Neuroendocrine lesions are the second most common neoplasm diagnosed in the pancreas.[1-3] The prognosis of these lesions varies depending on their level of differentiation, but most are well differentiated and are associated with better survival than adenocarcinoma. For this reason, it is imperative that the cytopathologist recognizes a neuroendocrine lesion when it is present, and appropriately rules it out when it is not. Endoscopic ultrasound-guided fine-needle aspiration (EUS-FNA) has been shown by many studies to be a safe, inexpensive, and effective modality for the diagnosis of solid pancreatic masses.[4-7] The objective of the current study was to assess the extent of error occurring in the cytopathologic diagnosis of neuroendocrine lesions of the pancreas, classify these errors using the system proposed by Zarbo et al, and determine their impact on clinical outcomes.
MATERIALS AND METHODS
In this retrospective study, information was collected regarding all cases diagnosed as neuroendocrine neoplasm either by EUS-FNA in cytology or by surgical pathology from 2000 to 2012 at the University of Alabama at Birmingham. The study protocol was reviewed and approved by the University of Alabama at Birmingham Institutional Review Board. Charts were thoroughly reviewed for all patients with a diagnosis of neuroendocrine neoplasm on surgical pathology, and we determined whether EUS-FNA had been performed before surgery. If EUS-FNA had been performed before surgery, these cases were included. Cases of neuroendocrine neoplasms diagnosed on cytology only (without surgical pathology confirmation) were not used in the analysis of EUS-FNA diagnostic performance because there was not adequate follow-up to ascertain the accuracy of the cytologic diagnosis. Data collected from the chart review include diagnosis, age, sex, imaging characteristics of the tumor both on EUS and any prior procedures, the number of passes made at the time of EUS-FNA, any special stains or immunohistochemistry, and any tumor characteristics noted (such as necrosis and mitoses).
EUS-FNA was performed under conscious sedation using a curvilinear echoendoscope (UC-30P; Olympus, Melville, NY). FNAs were performed using a 22-gauge needle, with a 19-gauge or 25-gauge needles used in a few cases. A cytopathologist was present for the on-site evaluation of adequacy and the diagnosis of all pancreatic masses during EUS-FNA. For on-site examination, only Diff-Quik staining was used. For final pathological interpretation, additional Papanicolaou-stained slides were prepared, as well as hematoxylin and eosin stains for cell block. We used the following diagnostic categories, as are standard in our department for pancreatic EUS-FNA: positive for endocrine neoplasm, suspicious for endocrine neoplasm, atypical with a comment, benign, and unsatisfactory. Cases diagnosed as “atypical” by cytology were considered to be negative and those diagnosed as “suspicious” were considered positive. “Atypical” diagnoses were considered negative because the pretest probability did not significantly differ from the posttest probability and therefore did not add diagnostic value. “Suspicious” cases were considered positive because they increased the likelihood ratio and imparted a higher risk than an “atypical” diagnosis, as previously reported by our institution. Often during rapid on-site evaluation, a positive diagnosis was rendered with endocrine neoplasm in the differential diagnosis, because cell block preparation and immunohistochemical analysis were frequently used to distinguish endocrine neoplasms from their mimics. If the final cytology report differed from the rapid diagnosis given during the procedure, the final cytology report was the diagnosis used in the current evaluation because no clinical decisions were made for the patient until after the final report was issued.
With the surgical specimen histology as the gold standard, all cases were reviewed to determine false-positive, false-negative, and misclassified cases. The discrepant cases were first reviewed by a cytopathologist in training who was blinded to the cytology diagnosis given to any 1 case of endocrine tumor noted on surgical pathology. Once the cytopathologist in training determined the type of error responsible for the discrepancy, a review by an experienced cytopathologist was conducted. The experienced cytopathologist was blinded to the type of discrepancy and any diagnosis given by cytology or surgical pathology, but did know that these cases were pulled in a study of errors in the diagnosis of neuroendocrine neoplasms. Any disagreement between the 2 was resolved using a multiheaded microscope. Using the standardized error and harm classification described by Zarbo et al, we reviewed the cytology and surgical pathology material and evaluated the type and cause of diagnostic errors and their impact on the patient. Accordingly, both cytology and surgical pathology diagnostic errors were classified as 1) false-negative, or an undercall of a lesion; 2) false-positive, or an overcall of a lesion; and 3) misclassification, in which there was neither an undercall nor an overcall but the lesion was placed in the wrong category of disease. Harm, conversely, was divided into minimal, minor, moderate, and major, depending on the morbidity or mortality inflicted. Minimal harm resulted in no morbidity and included errors such as a delay in diagnosis or therapy (< 6 months) and unnecessary further noninvasive efforts (eg, blood tests or computed tomography scans) or therapy. Minor harm was associated with minor morbidity and included a delay in diagnosis or therapy (> 6 months) and unnecessary further invasive procedures. Moderate harm encompassed all diagnostic or therapeutic efforts that resulted in moderate morbidity. Major harm was reserved for cases of dismemberment or loss of an organ or function of an organ system due to unnecessary diagnostic or therapeutic errors, or any cause of mortality.
During the study period, 177 patients underwent EUS-FNA and were diagnosed with a neuroendocrine neoplasm. Cytology diagnosis of a neuroendocrine neoplasm from pancreatic EUS-FNA or a surgical pathology diagnosis of a neuroendocrine neoplasm with a history of pancreatic EUS-FNA comprised these 177 cases (91 of the patients were male, and ranged in age from 15 years-90 years). Of these 177 cases, 80 (45%) had a cytologic diagnosis from pancreatic EUS-FNA as well as surgical pathology at the study institution, with both of these specimens available for review. The surgical pathology cases consisted of pancreatic resections in all 80 cases. In 56 of these 80 cases (70%), an adequate EUS-FNA cell block was obtained on which immunohistochemistry was performed. All 56 cases were stained with chromogranin A and synaptophysin on the cell block. Other stains used, depending on the differential diagnosis, were neuron-specific enolase, CD56, CD45, pan-cytokeratin, CA 19.9, α-1-antitrypsin (A1AT), E-cadherin, vimentin, CD10, and β-catenin. All 80 cases demonstrated the classic histologic appearance of a neuroendocrine neoplasm on surgical pathology. In 50 of the surgical pathology cases (63%), the diagnosis was supported by immunohistochemistry.
There were 14 discrepancies found between the cytologic and surgical pathologic diagnoses (18% of EUS-FNA cases with surgical follow-up). There were no false-positive cases, 9 (64% of total errors) false-negative cases, and 5 (36%) misclassifications (Table 1).
Table 1. Classification of Errors Associated With EUS-FNA of Neuroendocrine Neoplasms of the Pancreas
The false-negative cases consisted of 3 interpretation errors (20%) and 6 cytology sampling errors (40%). The 3 false-negative interpretation errors consisted of 2 cases diagnosed as atypical and 1 diagnosed as chronic pancreatitis. The 2 atypical cases were undercalled and on second review should have been given a suspicious or positive designation. One of these 2 undercalled cases demonstrated atypical single cells consistent with a neuroendocrine neoplasm, which was the final surgical pathology diagnosis given at the time of resection (Fig. 1). The other of the 2 atypical cases was called neuroendocrine tumor versus intraductal papillary mucinous neoplasm versus solid pseudopapillary neoplasm (SPN) at rapid diagnosis, but was revised as only atypical cells present on final evaluation; immunohistochemistry was noncontributory in this case. The final surgical pathology revealed adenocarcinoma with a florid proliferation of neuroendocrine cells. On review, the adenocarcinoma was present in the cytologic material, but was difficult to call in the background of increased neuroendocrine cells (Fig. 2).
The 6 cases that were false-negative due to cytology sampling error either consisted of cyst fluid, necrosis, benign pancreatic parenchyma, benign gastrointestinal epithelial contaminant, or chronic pancreatitis. Four of these patients underwent surgical resection despite the benign FNA specimens, and the histology demonstrated neuroendocrine neoplasms in all 4. The fifth error was in a patient found to have chronic pancreatitis on EUS-FNA who underwent repeat EUS-FNA that demonstrated benign pancreatic parenchyma, and this individual eventually underwent surgical resection. The histology demonstrated chronic pancreatitis with biopsy site changes and a 0.5-cm neuroendocrine microadenoma. The last error was in a patient who was appropriately diagnosed as having benign pancreatic parenchyma on cytology. The patient had previous bouts of pancreatitis, with EUS-FNA performed only when a possible mass was discovered. Final surgical pathology in this case indicated neuroendocrine carcinoma, metastatic to several lymph nodes.
The 5 misclassifications consisted of 4 cases of SPN diagnosed as neuroendocrine neoplasm on cytology and 1 case of neuroendocrine carcinoma diagnosed as adenocarcinoma on cytology. All 4 of the cases incorrectly classified as neuroendocrine neoplasms that were found to be SPN on surgical pathology had cell blocks, and all used immunohistochemistry. Errors in interpretation of the stains were to blame for 3 of these misclassifications. In the fourth case, only chromogranin A and synaptophysin were ordered, and those slides were not available for review at the time of the current study. These cases originally were interpreted as positive, but these same stains were found to be negative on surgical pathology, which also used neuron-specific enolase, A1AT, and CD10 positivity to confirm the histologic impression of SPN. One of these cases is shown in Figure 3. The case of neuroendocrine carcinoma that was given an adenocarcinoma diagnosis on cytology did demonstrate significant atypia and pleomorphism rarely encountered in neuroendocrine neoplasms, although the overall impression by both reviewing pathologists was a neuroendocrine neoplasm (Fig. 4). The patient had a history of T3N0M0 colon adenocarcinoma, and it is probable that this created some bias on the pathologist's part. FNA passes for cell block were not performed, and therefore no stains were performed on this case until the surgical pathology resection specimen was received.
The clinical characteristics of the patients with discrepant cases were varied. They ranged in age from 10 years to 71 years, 6 were males, the tumors ranged in size from 0.5 cm to 8.3 cm, and all had previous computed tomography scans that demonstrated masses. All of the masses were hypoechoic on EUS and were solid, except for 1 cystic neuroendocrine neoplasm. The site of the masses in the pancreas ranged from the head to the tail, including the neck, uncinate, and body (Fig. 5). There were no significant differences noted between this group and the remainder of the patients with accurate diagnoses.
There were no surgical pathology errors identified.
Of the 14 discrepant cases, 2 were associated with minor harm (14% of errors; 2.5% of all cases with surgical follow-up) and 1 case with minimal harm (7% of errors; 1% of all cases with surgical follow-up) (Table 2). The remaining 11 cases were associated with no harm to the patient (79% of errors; 12.5% of all cases with surgical follow-up). No instances of moderate or major harm were found.
Table 2. Classification of Harm Associated With Diagnostic Errors of EUS-FNA of Neuroendocrine Neoplasms of the Pancreas
Of the cases determined to be false-negative due to interpretation error, 2 were associated with no harm and 1 caused minor harm. In the 2 cases in which misinterpretation caused no harm, the patients underwent definitive surgical resection soon after EUS-FNA. The third patient experienced minor harm as he underwent palliative resection soon after EUS-FNA but was found to have pancreatic adenocarcinoma as well as liver metastases, and might have received neoadjuvant chemotherapy or chosen not to be treated at all. Of the 6 false-negative cases resulting from sampling error, 4 caused no harm to the patients, as they all underwent surgical resection despite having benign FNAs, and the histology demonstrated neuroendocrine neoplasms in all 4 cases. The fifth sampling error resulted in minor harm, because the patient had an invasive further diagnostic effort. The last sampling error resulted in a delay in treatment of 9 months, which is consistent with minimal harm.
None of the instances of misclassification error were associated with any harm to the patient because all 5 patients underwent surgical resection for their tumors soon after the diagnosis was made on EUS-FNA.
There is not only a need to measure error rates, but also to analyze their impact in a standardized way so that error reduction and prevention methods can be developed. Before the article by Zarbo et al, many publications sought to provide error rates in FNA of various organs, but no agreement was put forth regarding how to classify or the taxonomy of error, making comparisons and meta-analyses weak at best; the discrepancy error rates varied from 0.1% to 10%.[8, 10]
Although EUS-FNA is widely accepted as the preferred method for obtaining diagnostic material in patients with pancreatic masses, it is not without error. The results of the current study reiterate that EUS-FNA is a safe and effective method for the evaluation and diagnosis of pancreatic endocrine neoplasms. Furthermore, we have shown that harm done to patients through errors in this method is at most minor, with a delay in the correct diagnosis being the most common type of harm. No harm is the most common outcome of the error found in EUS-FNA, most likely because of high clinical suspicion for the procedure in the first place, which is often reflected in the high positive predictive value of EUS-FNA. The lack of any moderate or major harm is most likely due to the lack of any false-positive cases in the current review because none of the patients were found to have benign pancreatic specimens at the time of surgical excision.
It has not gone unnoticed that 80% of our misclassifications were SPNs. SPN is a well-known entity in the differential diagnosis of pancreatic neuroendocrine neoplasms and must always be considered when an increased population of single cells with plasmacytoid, round to oval nuclei is encountered on a cytology smear. We found that all 4 misclassified SPN cases demonstrated similar morphology that was distinct from our cases of neuroendocrine neoplasm, although this was a subtle observation. These smears were much more cellular, demonstrated even less distinct cytoplasm, had larger nuclei that were more oval than round, showed small bits of metachromatic material, and tended to have more clustering than other smears diagnosed as neuroendocrine neoplasm. They did not demonstrate the classic features of SPN as described previously, such as pseudopapillae, capillary cores with metachromatic myxoid material, coffee bean/grooved nuclei, clear vacuoles in the cytoplasm, slender cytoplasmic processes, or hyaline globules.[11-13] Difficult differential diagnoses such as SPN from pancreatic neuroendocrine neoplasms are one of the main reasons to obtain an adequate cell block at the time of EUS-FNA so that immunohistochemistry may be performed to elucidate the correct diagnosis, which has been published previously. In 2009, Burford et al eloquently outlined which immunohistochemical stains could best distinguish pancreatic neuroendocrine neoplasms from SPN. Among all FNA samples tested, E-cadherin, β-catenin, and CD10 demonstrated the greatest difference between pancreatic endocrine neoplasms and SPN. It is interesting to note that the patterns of E-cadherin and β-catenin staining were highly specific for this distinction. In pancreatic neuroendocrine tumors, E-cadherin should demonstrate a decidedly membranous staining pattern, and the lack of this pattern is suggestive of SPN. SPN will demonstrate nuclear staining for β-catenin, and pancreatic neuroendocrine neoplasms will not. CD10, chromogranin A, and progesterone receptor will further help to establish the correct diagnosis, but are mostly suggestive. A CD10-positive, chromogranin A-negative, progesterone receptor-positive pattern would suggest SPN, but pancreatic neuroendocrine neoplasms also can demonstrate these results, although much less commonly.
The results of the current study highlighted other common pitfalls in pancreatic cytologic diagnosis, including chronic pancreatitis and benign gastrointestinal contaminant. However, we would be remiss if we did not note that the differential is much larger for pancreatic neuroendocrine neoplasms than what we have demonstrated herein. Acinar cell carcinoma, lymphoma or plasmacytoma, well-differentiated ductal adenocarcinoma, serous cystadenoma, and pancreatoblastoma must all be considered when evaluating tumors with a monomorphic appearance, as pancreatic neuroendocrine tumors typically have. Metastases, such as renal cell carcinoma and neuroendocrine carcinoma, must also be considered.
The current study was limited by the nature of its retrospective approach. The reviewers were blinded to specific diagnoses made on cytology and surgical pathology, but were aware that the cases were pulled based on a finding of neuroendocrine tumor in one or the other modality. This study was also limited by the finding that the cases were pulled from the medical record, and often relied on past reports. The information obtained by the cytopathologist from the endoscopist during a rapid on-site interpretation was often left undocumented. There were also cases for which some of the slides could not be located.
Moving forward in improving the EUS-FNA diagnosis of pancreatic neuroendocrine neoplasms will no doubt mean incorporating new techniques with which to assess tumor characteristics that are currently limited in EUS-FNA, such as prognostic indicators. Current guidelines evaluate for grade and stage on surgical pathology only, but DNA microsatellite loss analysis performed on alcohol-fixed, Papanicolaou-stained slides prepared from EUS-FNA samples has demonstrated that fractional allelic loss correlates with pancreatic neuroendocrine tumor recurrence, progression, and mortality. Exciting findings such as these indicate that minimally invasive EUS-FNA is not only the gold standard, but can possibly tell us more about tumors than we originally believed possible.