Novel Cytological Diagnostic Scoring System for Pancreatic Specimens Obtained by Endoscopic Ultrasound-Guided Fine Needle Aspiration

Background: Cytological diagnosis of pancreatic specimens obtained by endoscopic ultrasound-guided ne needle aspiration (EUS-FNA) is often challenging because of the small amount of sample or well-differentiated adenocarcinoma with weak cytological atypia. Therefore, the sensitivity and specicity of cytological diagnosis for pancreatic cancer should be improved. Hence, we aimed to evaluate the indices, which are used to distinguish malignant from benign lesions for the cytological diagnosis of pancreatic EUS-FNA specimens. Methods: Seven reviewers, including 3 cytotechnologists and 4 medical doctors, evaluated 20 morphological indices in pancreatic specimens obtained by EUS-FNA (malignant, n=111; benign, n=31). Statistical analyses were performed using Fisher’s exact test, logistic regression analysis, area under the receiver operating characteristic curve, and Youden Index. Results: Among the 20 indices, there was a high incidence rate (>40%) of the following 13 indices in malignant cases: structural atypia, hyperchromatic nucleus, irregular cell polarity, unclear cell boundary, nuclear membrane thickening, anisonucleosis, overlapping, irregular nuclei, high nuclear/cytoplasmic ratio, binding decline, simultaneous appearance of malignant and benign cells, enlarged nucleoli, and background necrosis. When we diagnosed pancreatic specimens using these 13 cytological indices, the cutoff value of 8/9 showed the highest Youden index (0.950) as well as high sensitivity and specicity in distinguishing malignant from benign specimens (98% and 97%, respectively). Conclusion: Thirteen cytological indices showed high sensitivity and specicity in differentiating malignant and benign lesions using pancreatic EUS-FNA samples. Further validation or prospective studies are necessary to establish criteria for the cytological diagnosis of pancreatic cancer.


Introduction
The rates of mortality and morbidity associated with pancreatic cancer have been increasing worldwide. 1 The prognosis of pancreatic cancer remains poor, with an overall 5-year survival rate of less than 10% due to its aggressive growth and high rate of metastasis. Endoscopic ultrasound-guided ne needle aspiration (EUS-FNA) allows for repeated, minimally invasive sampling from the pancreas, 2 and has 85%-91% sensitivity for pathological diagnosis of pancreatic cancer. 3 Previous studies reported that atypical cytological ndings by pancreatic EUS-FNA highly indicated pancreatic neoplasms (55%) including cancer (36%). 4 Cytological diagnosis was also highly speci c for detection of pancreatic cystic tumors such as intraductal papillary mucinous neoplasm 5 and mucinous cyst neoplasm. 6,7 Thus, cytological diagnosis using pancreatic EUS-FNA provides important information on the indications for surgery. 5,8,9 The Japanese Pancreas Society recommends EUS-FNA as the rst choice for pathological diagnosis of pancreatic cancer. 10 Although most pathological diagnoses using EUS-FNA samples are straightforward, some inconclusive pathological ndings were reported, especially in small and/or degraded samples, and when differentiating atypical epithelium and well-differentiated cancer with weak cytological atypia, as well as pancreatic mucinous neoplasms from the contaminants of gastric mucosa. 11 Furthermore, current cytological criteria have several problems such as; 1) until recently there was no widely accepted categorization criteria for pancreatic cytology obtained by EUS-FNA, 2) cytological criteria and standard de nition of pancreatobiliary cytology have not been well established, 3) important cytological indices to distinguish malignant from benign have not been well clari ed.
Several studies have attempted to identify the de nite cytologic criteria, which can predict malignancy in bile duct and pancreatic duct brushings. [12][13][14][15][16][17] Avadhani et al. reported that 11 cytological indices could be helpful in identifying malignancy in cytological specimens obtained by bile duct brushings. 18 Robins et al. proved that nuclear overlapping, nuclear contour irregularity, and irregular chromatin distribution were the best predictors of malignancy in pancreatic CT-guided FNA specimens. 19  (1) malignant cases had a de nitive histological diagnosis of adenocarcinoma, either by biopsy (n=110) or both biopsy and resection (n=9). Of the 111 cases, 110 were diagnosed as malignant by biopsy. In one case, biopsy revealed no pancreatic epithelium; however, cytology and resected specimens were malignant. Nine of the 111 cases were diagnosed as malignant with both biopsy and resected specimens; (2) benign cases had a de nitive histological diagnosis of non-malignancy by biopsy (n=31) and a clinical uneventful follow-up of 2-70 months (median value, 35 months). In our institute, we initially diagnose the specimens with FNA using cytology specimens, and then separately diagnose using biopsy specimens (H&E staining and immunostaining). Benign cases included patients with pancreatitis, simple pancreatic cysts, and low-grade dysplasia. Malignant cases included patients with invasive adenocarcinoma and high-grade dysplasia, as it is impossible to distinguish carcinoma in situ from invasive carcinomas using EUS-FNA samples. All methods were performed in accordance with the guidelines and regulations of the Research Ethics Committee of the Tokyo Metropolitan Geriatric Hospital. Informed consent for study participationwas obtained. The present study was approved by the Research Ethics Committee of the Tokyo Metropolitan Geriatric Hospital (approval number: R17-58).

Review
The seven reviewers included two certi ed pathologists who are licensed medical specialists from the Japanese Society of Pathology and Japanese Society of Clinical Cytology, one senior trainee pathologist, and four cytotechnologists who are licensed specialists from the Japanese Society of Clinical Cytology. All reviewers were blinded to all clinical or radiologic information, histologic diagnoses, and other reviewers' diagnoses. For each case, the seven reviewers evaluated the presence or absence of 20 indices using two microscope glass slides with Papanicolaou stain.

Statistical analysis
A universal standard was developed to clarify the correlation between years of experience and quali cations of 20 indices. In accordance with this universal standard, the presence or absence of these 20 indices was determined. The agreement signi cantly associated with malignant cases was examined using Fisher's exact test. Logistic regression analysis was used to calculate the odds ratios for individual indices and for malignancy. The number of malignant indices present was calculated for each sample, and the sensitivity, speci city, and accuracy were evaluated for each cutoff point using the Youden Index. P-values of < 0.05 (two sided) were considered statistically signi cant. Then, we compared various cutoffs for high agreement. All analyses were conducted using SPSS version 20 (IBM, Armonk, NY, USA).

The 13 indices (gray boxes in
The useful indices to distinguish malignant and benign cases should have the following properties: (1) high incidence rate in malignant cases, (2) statistically signi cant difference (P < 0.05) between malignant and benign cases, and (3) reproducibility in various groups with different backgrounds and skills. Therefore, we determined that the following 13 cytological indices were useful to distinguish malignant and benign cases: irregular structure, hyperchromasia, irregular cell polarity, unclear cell boundary, nuclear membrane thickening, anisonucleosis, overlapping, irregular nuclei, high N/C ratio, decreased cell adhesion, two-cell pattern, prominent nucleoli, and necrotic background.

Cutoff values of the 13 indices indicating tumor malignancy
Each of the 13 indices was given a score of 1 point, based on the evaluation of 4 out of 7 reviewers (Table 2). Malignant cases had higher scores than benign cases (Table 2). To determine the cutoff values of the scores of all 13 malignant indices, we calculated the Youden index. 22 A score of 8 points compared with a score of > 9 points showed the highest Youden index (0.950, Table 3). The cutoff value 8/9 showed high sensitivity (98%, Table 3), speci city (97%), accuracy (98%), and high area under the receiver operating characteristic curve value (0.996).

Discrepancy between score and pathological diagnosis
At a cutoff value of 8/9, 3 cases showed a discrepancy between cytological score and pathological diagnosis (Table 2A, B and C and Supplementary Table 1). Cytological specimens of case A lacked malignant features with contamination of many red blood cells (Fig. 6A). The cytological specimens of case B contained only a small number of cells (Fig. 6B). The cytological specimen of case C showed epithelial cells with enlargement of the nucleus, nuclear groove, and yellow mucus (Fig. 6C), and case C was diagnosed as mild atypical epithelium but not malignancy.

Discussion
The present study revealed that (1) among the 20 cytological indices, 13 indices showed statistically signi cant differences between malignant and benign cases, and had high incidence rates in malignant cases; (2) our new scoring system showed high sensitivity and speci city to distinguish malignant and benign lesions in pancreatic EUS-FNA samples.
Previous reports on cytological diagnosis of pancreatic cancers have also shown the usefulness of several cytological features such as high N/C ratio, 23 nuclear membrane thickening, 14,24 anisonucleosis, 20,25 and irregular cell polarity 26,27 . In the present study, these 4 cytological features showed high odds ratios (> 100) and incidence rates (over 90%), suggesting that they are critical for pancreatic cancer diagnosis. Furthermore, we established a new scoring system for pancreatic cancer diagnosis using a combination of cytological indices.
We had previously reported that atypical pancreatobiliary glands with pink mucus (probably sulfomucin) suggested malignancy, and yellow-orange mucin (probably neutral mucin) suggested benignity. 21 In the present study, pink intracellular mucus was detected only in malignant cases and not in benign cases, but its presence in malignant cases was low (12%-44%). These results indicate that pink intracellular mucus possesses high speci city and low sensitivity to distinguish malignant and benign tumors.
The difference in hypercellularity was statistically signi cant between malignant and benign tumors, when evaluated by cytotechnologists, but not by medical doctors. This might be due to the fact that, in Japan, cytotechnologists routinely evaluate hypercellularity but medical doctors do not.
Among the 20 indices evaluated in the present study, mucus background, yellow-orange mucin in cytoplasm, cannibalism enclosed image, and keratinization did not have statistically signi cant differences between malignant and benign cases, when evaluated by all reviewers, suggesting that they were not useful to distinguish between malignant and benign tumors.
The sensitivity and speci city of our new scoring system using the 13 cytological indices were higher than those of Avadhani who reported similar diagnostic criteria for the cytological diagnosis of bile duct samples. 18 Pancreatic EUS-FNA can provide fresh samples with few degenerative changes, while the bile duct sample showed degeneration based on the results of bile juice examination. The results showed that the sensitivity and speci city of our scoring system were high, and all 13 indices were useful for diagnosing malignancy by cytology of pancreas obtained by EUS-FNA.
The present study has several limitations. We focused mainly on indices of malignant morphology and not those of benign morphology here. This was a retrospective study using malignant and benign samples obtained by pancreatic EUS-FNA. Hence, prospective and multicenter studies are needed to determine the usefulness of the diagnostic criteria. Furthermore, we need to evaluate the usefulness of the diagnostic criteria using various samples, including pancreatic juice and liquid-based cytology (LBC). [29][30][31] This was a single-institution retrospective study, so the result needs to be evaluated in other hospitals.
In conclusion, we introduced a new scoring system for cytologic diagnosis using pancreatic EUS-FNA samples. We proved that all the 13 indices were important ndings for diagnosing malignancy in the pancreatic cytology smear of EUS-FNA. A further validation study is warranted.  The gray boxes indicate the malignant scores at a cutoff value of 8/9 based on the results of Table 3. "A, B, C" indicate cases with discrepancy between scoring and pathological diagnosis (details in Supplementary Table 1).