Cytological criteria of high-grade epithelial atypia in the cyst fluid of pancreatic intraductal papillary mucinous neoplasms
Martha B. Pitman MD,
Department of Pathology, Massachusetts General Hospital, Harvard Medical School, Boston, Massachusetts
Corresponding author: Martha Bishop Pitman, MD, Department of Pathology, Harvard Medical School, Massachusetts General Hospital, 55 Fruit St, Boston, MA 02114; Fax: (617) 724-6564; firstname.lastname@example.org
The recognition of epithelial cells with high-grade atypia (HGA) in the cyst fluid of an intraductal papillary mucinous neoplasm (IPMN) identifies a cyst at high risk of invasive carcinoma. To the best of the authors' knowledge, the cytological features of HGA have not been systematically analyzed to define diagnostic criteria.
Cell groups from patients with histologically confirmed branch-duct IPMNs were evaluated by 2 cytopathologists with expertise in pancreatic cytology. A consensus interpretation categorized the cell groups as having either low-grade (LG) or high-grade (HG) morphology. Characteristics regarding cell size and architecture, nuclear and cytoplasmic features, and background necrosis were analyzed. Performance characteristics were assessed using the Fisher exact test at 95% confidence intervals.
Sixty cell groups yielded 27 LG and 25 HG morphological groups. No consensus was reached for 8 groups, which were excluded from statistical analysis. Five features that were found to be significantly different between the LG and HG groups included: 1) cell size < a 12-μm duodenal enterocyte for HG and size equal for LG; 2) an increased nuclear-to-cytoplasmic (N/C) ratio; 3) marked nuclear membrane abnormalities; 4) abnormal chromatin pattern; and 5) background necrosis. The 3 most accurate features for the identification of HGA were background necrosis (88%), abnormal chromatin pattern (84%), and an increased N/C ratio (82%).
Conservative management of patients with branch-duct intraductal papillary mucinous neoplasms (BD-IPMNs) has become increasingly common due to the tendency of these cysts to be low grade (LG). In addition, many patients are elderly with comorbid conditions that increase the surgical risk. The optimal time to resect a BD-IPMN is before the development of invasive adenocarcinoma. Patient prognosis drops to a 5-year survival rate of 40% once an invasive carcinoma arises. The majority of BD-IPMNs are lined with gastric-type epithelial cells with LG dysplasia (LGD). The invasive carcinoma that arises from these cysts is of tubular type with the same dismal prognosis as conventional ductal adenocarcinoma.
Although the risk stratification of patients with BD-IPMN is based largely on imaging characteristics,[3, 4] endoscopic ultrasound-guided fine-needle aspiration (EUS-FNA) plays a significant role in recognizing high-risk cysts that may not otherwise appear to be high risk on imaging.[5, 6] The cytological characteristics of IPMN in general[7, 8] and of epithelial cells with high-grade atypia (HGA) in particular[5, 9] have been described previously, but to the best of our knowledge the criteria for HGA have not been systematically established. High-grade atypical epithelial cells represent cells that are suspicious or positive for malignancy and correspond histologically with high-grade dysplasia (HGD) or invasive carcinoma with high accuracy.[5, 9] We assessed a wide variety of cellular characteristics of cell groups from 24 histologically confirmed cases of BD-IPMNs to determine the cytological criteria for epithelial cells with HGA in cyst fluid, a designation that would indicate a cyst at high risk of invasive carcinoma.
MATERIALS AND METHODS
Hospital Institutional Review Board approval was obtained for this study.
A set of 60 cell groups were assessed from Papanicolaou-stained direct smears or cytospin preparations of EUS-FNA specimens of histologically confirmed IPMNs of various grades: 4 LGD; 7 intermediate-grade dysplasia (IGD); 6 HGD; and 7 IPMN with invasive carcinoma (ACa). These same cell groups were part of an international interobserver concordance study that demonstrated that the 2 most experienced reviewers had very good agreement (87%) and good correlation with histology with IGD classified as LG.
The cell groups represented a variety of cell types, including gastrointestinal epithelium, acinar cells, and histiocytes (collectively categorized as nonlesional [NL]), as well as mucinous epithelial cells with various degrees of cellular atypia. Two reviewers with expertise in pancreatic cytology (M.B.P. and B.A.C.) classified the cell groups as either LG morphology, which included NL cells, LGD, and IGD, or HG morphology, which included HGD and ACa. If there was not complete agreement regarding the exact classification of the cell groups but agreement that the cells were LG, the cell group was classified as LG atypia (LGA). Similarly, if there was not complete agreement regarding the exact classification of the cell group as HGD or ACa, the classification was then HGA. Only cell groups with concordant interpretations in the LG and HG groups were used in assessing cytological criteria and included in the statistical analysis.
Establishing Cytological Criteria
Cytological features assessed included the following: cell architecture (single cells; small [2-6] cell clusters; medium [7-20] cell clusters; or large [> 20 cells] sheets, groups, or papillary formations); cell size (relative to a 12-μm duodenal enterocyte [Fig. 1]: smaller than [<] or equal to and larger than [>]); nuclear-to-cytoplasmic (N/C) ratio (increased: yes/no); cytoplasmic mucin (visible vacuoles: yes/no); nuclear envelope irregularity (none/minimal or marked); chromatin pattern changes (hypochromatic/hyperchromatic or none); nucleoli (prominent: yes/no); and background cellular necrosis (yes/no).
Analysis of the cell features was calculated for LG versus HG morphology using the Fisher exact test at 95% confidence intervals and considered to be significant at a P value < .05. Performance characteristics were calculated for the features with significant differences between the 2 groups.
Cytological Consensus Diagnosis
Of the 60 cell groups, there were 27 LG morphological groups (15 LGA, 11 NL, and 1 LGD) (Fig. 2) and 25 HG morphological groups (10 HGA, 12 HGD, and 3 ACa) (Fig. 3). There was no consensus for 8 of 60 cell groups (13%) (Table 1).
Table 1. Consensus Cytological Diagnosis of Cell Groups
Low-grade atypia (nonspecific)
High-grade atypia (nonspecific)
The evaluated cytological features and performance characteristics stratified by LG and HG morphology are presented in Table 2. The cytological features that were found to be significantly different between LG and HG cysts and the features that supported a cytological interpretation of HGA included cell size smaller than an enterocyte (P = .002), increased N/C ratio (P = .001), marked nuclear membrane abnormalities (P = .002), abnormal chromatin pattern (P = .001), and background cellular necrosis (P = .001). Cell size equal to an enterocyte was found to be a significant feature for LG morphology. Size larger than an enterocyte, cell architecture, the presence of cytoplasmic mucin, and prominent nucleoli were not found to be significantly different between the 2 groups.
Table 2. Cytological Features and Performance Characteristics of Low-Grade Versus High-Grade Cells in IPMN Cyst Fluids
The arrangement of cells as single cells or cell groups of any size was not significantly different between LG and HG groups. Single cells (P = .7), small cell groups (P = .3), medium cell groups (P = .4), and large cell groups (P = .5) were found to be relatively evenly distributed between the 2 groups.
Cell size relative to a 12-μm duodenal enterocyte was found to be statistically significant for smaller cell size (P = .002) but not for larger cell size (P = .08), although as many HG cells were larger than an enterocyte as were smaller. Nonlesional and LG dysplastic cells may be larger than the typical enterocyte due to the variable amount of cytoplasmic mucin. As such, large cell size was not found to be significantly different between the 2 groups. Cell size equal to an enterocyte was significant for LG morphology (P = .001). Small cells were a specific (92%) but not a sensitive (46%) finding for HG morphology, with an overall accuracy of 69%.
An increased N/C ratio was found to be significantly associated with HG morphology (P < .0001), and was a more sensitive (92%) than specific (76%) feature, with an overall accuracy of 82%. LG cells rounding up in solution can yield a high N/C ratio, thereby impacting specificity of this feature.
Cytoplasmic mucin was visible in cells of both LG and HG morphology and was not a distinguishing feature for grade (P = .07).
Nuclear Envelope Abnormality
Marked nuclear envelope abnormality distinguished LG and HG morphology (P = .002). This feature was a specific (96%) feature of HG morphology compared with no or minor nuclear membrane abnormalities, but was insensitive (38%) and therefore the overall accuracy was only 67%.
Chromatin Pattern Changes
An abnormal chromatin pattern of both hypochromasia or hyperchromasia was found to be significantly different between LG and HG cells (P = .001). The sensitivity, specificity, and accuracy rates were all 84%.
Nucleoli were not common in either LG or HG cells. Prominent nucleoli were noted in some cells in both the LG and HG groups (P = .20), although they were more common in HG cells.
The presence of background cellular necrosis in the cyst fluid appears to be a reliable marker for HG morphology (P = 0.001). The sensitivity was 80%, the specificity was 96%, and the overall accuracy was 88%.
In the current study, we evaluated the cytological features of the cells from EUS-FNA specimens of BD-IPMNs with a consensus interpretation of either LG or HG morphology. Analysis of cell architecture, cell size relative to a 12-μm duodenal enterocyte, N/C ratio, visible cytoplasmic mucin, nuclear envelope abnormality, abnormal chromatin pattern, prominent nucleoli, and background necrosis demonstrated 5 features that were found to be significantly different between LG and HG morphological cell groups: 1) small cell size; 2) an increased N/C ratio; 3) marked nuclear membrane abnormalities; 4) an abnormal chromatin pattern; and 5) background cellular necrosis. Of these 5 characteristics, the 3 most accurate features for the identification of HGA included background necrosis (88%), chromatin pattern changes (84%), and an increased N/C ratio (82%). To our knowledge, this is the first study to systematically evaluate the cytological features of histologically confirmed IPMNs that identify cytological criteria of HGA.
Nonspecific “atypical” cytology has been described in pancreatic cyst fluids and shown to be a strong predictor of malignancy in patients with mucinous cysts.[11, 12] Better defining these atypical cells as epithelial cells with HGA (eg, atypical epithelial cells insufficient in quality or quantity for a cytological interpretation of a malignant neoplasm, eg “positive cytology”) has been shown to be predictive of HGD or ACa in mucinous cysts, with a reported sensitivity of 72% and specificity of 85%, which is a marked improvement over the 29% sensitivity of a “positive” interpretation.[5, 13] Detecting HGA by cytological analysis has been shown to add value to the imaging assessment for malignancy risk, with HGA being more sensitive albeit less specific for the detection of HGD or worse compared with a mural nodule and dilated main pancreatic duct. However, to our knowledge diagnostic criteria for HGA have yet to be defined.
The cytological features of IPMN, in general, have been described previously.[7, 8, 14-20] One of the first reports in the English literature in 1995 by Nakaizumi et al described the cytological features of 12 pancreatic neoplasms, most likely representing IPMN with HGA. Cells had an elevated N/C ratio in all cases, irregular nuclear envelopes in 50% of the cases, and hyperchromasia in all but 1 case. Four years later, Shimizu et al described the cytological features of 2 IPMNs with HGD. Representative images demonstrated overt malignancy in each case, with medium-sized clusters, large cells, an increased N/C ratio, marked nuclear envelope abnormalities, prominent nucleoli, and hyperchromasia. Stelow et al described various cytological features of 18 IPMNs, including the presence of mucin, overall cellularity, goblet cells, and cytoplasmic mucin, but there were no specific cytological features described and generic atypia was used for all cells that differed from the reference biliary, ductal, or duodenal epithelial cells. Only 4 cases had histological confirmation (1 IPMN with LGD, 2 with HGD, and 1 with invasion). Layfield et al also evaluated the cytological features in 15 mucinous cysts including 13 IPMNs and 2 mucinous cystic neoplasms with LGD. All IPMNs were histologically either invasive (7 cases) or with IGD (6 cases). There were no cases of IPMN with LGD or IPMN with HGD noted. Although their analysis did not statistically evaluate the cytological features of epithelial cells by grade, their data appear to be similar to ours in that although marked nuclear atypia, nuclear envelope irregularity, and prominent nucleoli supported a cytological diagnosis of HGA, these features were also noted in IPMN with IGD. Sole et al and Salla et al described similar findings with cellular atypia found to be absent in IPMN with LGD, and moderate in both IGD and ACa. Marked nuclear atypia and discohesiveness were usually an indication of HGD or ACa, but marked atypia was also observed in cases of IPMN with IGD. Similar to the current study, cellular architecture was not found to be a distinguishing feature of grade in their study. These studies support our observations that the cytological features of cells aspirated from IPMN are highly variable in quantity and cytological characteristics.
IPMNs are very heterogeneous neoplasms with 4 morphologically distinct cell types: gastric (null) type, intestinal type, pancreatobiliary type, and oncocytic type. Although the pancreatobiliary and oncocytic types are considered HG, intestinal-type IPMN may be either IG or HG and gastric type ranges from LGD to HGD. Although the majority of BD-IPMN are low grade, the invasive tubular-type carcinoma arising from gastric-type epithelium in patients with BD-IPMN carries the same poor prognosis as conventional ductal carcinoma, and hence the need for the early detection of HGA in these cysts.
With all these different cell types and grades, compounded by the process of cells rounding up and degenerating while floating in cyst fluid, it is a challenge to say the least for a cytopathologist to specifically grade cells of an IPMN. LGD cells can round up, lose cytoplasmic mucin, and demonstrate reactive changes with chromatin pattern alternations. IGD cells, previously classified as moderate dysplasia, demonstrate cytological features that range from LGD to HGD and, similarly, HGD cells have features that overlap with adenocarcinoma. Most cytological grading systems have moved from multitiered grading systems to more general LG and HG systems to increase interobserver agreement in classification as well as to better align the cytological interpretation with patient management. Our recent international interobserver concordance study of the epithelial cells in the current study demonstrated poor agreement using a 5 tiered grading system but very good agreement with experience using a 2-tiered LG and HG grading system.
Management recommendations for patients with BD-IPMN are increasingly conservative given their prevalence, tendency to occur in older patients with comorbid conditions, and typical LG nature. Risk profiles for malignancy are based predominantly on imaging features related to cyst characteristics such as the degree of dilation of the main pancreatic duct and the presence of an enhancing mural nodule or a cyst wall mass. Studies have shown that small BD-IPMNs are not always benign despite the absence of high-risk imaging features.[6, 22-25] A recent study by Fritz et al found that nearly 25% of cysts considered to be negative using the Sendai criteria (eg, small [< 3 cm]) without high-risk imaging features are malignant (HGD or ACa) histologically. Similarly, a study by Wong et al found that patients with small BD-IPMN had HGD in 23% of cases, with 3 of 7 cases (43%) noted in cysts measuring < 1 cm. Others have also shown that BD-IPMN measuring < 3 cm harbor malignancy with a significant incidence of HGD or invasive carcinoma.[28-30] The specificity of imaging studies to predict malignancy is high when imaging features demonstrate overt high-risk features such as a mural nodule or cyst wall mass.[31, 32] Obtaining cytological confirmation of malignancy adds little value to the decision to surgically resect such high-risk cysts in surgically fit candidates. For patients with cysts that do not demonstrate high-risk features of malignancy or who have cysts of uncertain etiology, aspiration of cyst contents is the best method with which to gather more information about the cyst for purposes of patient management. The simple confirmation that a cyst is mucinous by macroscopic cyst fluid characteristics, cytological analysis, or elevations in carcinoembryonic antigen is a step forward in aligning proper patient management. In a study by Cizginer et al, carcinoembryonic antigen has been shown to be the most accurate test for detecting a mucinous cyst and cytology the most accurate test for detecting a malignant cyst.
Despite its limitations, the collection of the cyst fluid for cytological, biochemical, and molecular analysis contributes greatly to patient care.[13, 33-41] The recognition and reporting of HGA is critical to recognizing cysts that are at high-risk of malignancy. In the current study, we evaluated histologically confirmed cases of BD-IPMN and critically evaluated the cytological features that support a cytological interpretation of HGA. The 3 most accurate features for the identification of HGA were found to be background necrosis (88%), chromatin pattern changes (84%), and an increased N/C ratio (82%). The HGA classification is a nonspecific HG category that includes ACa and HGD. However, patient care providers must keep in mind that the aspiration of cyst contents is essentially a screening test for malignancy, and therefore there will be some cysts resected with an interpretation of HGA that may not be found to have HGD or ACa on histology. To the best of our knowledge, until more specific and accurate tests are developed, cytological analysis provides the best test for detecting cysts at high-risk of malignancy before the development of high-risk imaging features. Establishing diagnostic criteria for HGA is a step toward standardized interpretation.
No specific funding was disclosed.
CONFLICT OF INTEREST DISCLOSURES
Dr. Centeno acts as a member of the Pancreatic Cancer Clinical Advisory Board for Asuragen.