Original Article

Atypical Cytologic Diagnostic Category in EUS-FNA of the Pancreas Follow-Up, Outcomes, and Predictive Models Evan Alston, MD1; Sejong Bae, PhD2; and Isam A. Eltoum, MD, MBA1

BACKGROUND: The objective of this study was to assess how atypical diagnostic category (ADC) is followed up, its outcomes, and the predictors that are associated with subsequent diagnosis of neoplasm/malignancy. METHODS: We reviewed pancreatic endoscopic ultrasound fine-needle aspiration (EUS-FNA) with ADC and compared the rate of detection of neoplasms after a repeat FNA, a biopsy/resection, or a clinical follow-up following ADC. Logistic regression was used to determine the factors associated with the diagnosis of a neoplastic or a malignant lesion following ADC. Predictive probability for each case was calculated on the basis of the significant predictors, and whether it improved diagnostic performance was assessed. RESULTS: Of 3832 cases that received pancreatic EUS-FNAs, 187 (4.9%) were ADC. A total of 93 neoplasms (55%), including 61 carcinomas (36%), were detected after an atypical cytologic diagnosis. Similar rates of detecting neoplasms were observed after repeat FNA or biopsy/resection but higher than after clinical follow-up. The presence of a mass, history of alcohol use, and absence of a history of pancreatitis were significant predictors of a higher rate of diagnosis of neoplasm. Weight loss and bile flow obstruction were more likely to be associated with higher rates of carcinoma. Predictive probability demonstrated a wide range of risk and changed the ambiguous diagnosis to informative in 30% of cases. CONCLUSIONS: ADC of pancreas is associated with a high risk of benign and malignant neoplasms regardless of the method of follow-up. The presences of a mass, alcohol use, and absence of a history of pancreatitis are significant predictors of a diagnosis of neoplasm, whereas weight loss and bile duct obstruction are significant predictors C 2013 American Cancer Society. of ductal carcinoma following an ADC. Cancer (Cancer Cytopathol) 2014;122:428-34. V

KEY WORDS: EUS-FNA; pancreas; guidelines; predictive modeling; diagnostic accuracy.

INTRODUCTION Endoscopic ultrasound fine-needle aspiration (EUS-FNA) has gradually become a standard procedure in the initial workup of patients suspected of pancreatic neoplasm with high rates of accuracy.1–4 In most laboratories, the diagnostic cytologic categories generally follow Bethesda-System-like categories: unsatisfactory, benign, atypical, suspicious, and positive for neoplasm.5 As in most organ systems, the atypical category is ambiguous and creates diagnostic and management challenges. In a recent meta-analysis, we provided a strong argument for the need for guidelines for the standardization and management of the atypical category of EUS-FNA of the pancreas.5 The objective of this study was to assess the types of follow-up provided for this category in a large EUS-FNA service, their outcomes, and the possible predictors of neoplastic/malignant outcomes. Our goal is to provide a predictive model to assist in the management of this diagnostic category.

Corresponding author: Isam A. Eltoum, MD, MBA, 122 HSB 619 Ave 19th St South, Birmingham, AL 35249; [email protected] 1 Department of Pathology, Division of Anatomic Pathology, University of Alabama at Birmingham, Birmingham, Alabama; 2Department of Medicine, Division of Preventive Medicine, University of Alabama at Birmingham, Birmingham, Alabama

Received: October 14, 2013; Revised: November 21, 2013; Accepted: December 2, 2013 Published online January 16, 2014 in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/cncy.21389, wileyonlinelibrary.com

428

Cancer Cytopathology

June 2014

EUS-FNA Predictive Model/Alston et al

MATERIALS AND METHODS In this retrospective study, we collected information on all cases, irrespective of lesion features, cystic or otherwise, diagnosed as atypical by EUS-FNA of the pancreas from July 2000 to 2013 at University of Alabama at Birmingham. Electronic records were reviewed for follow-up and for the risks associated with pancreatic neoplastic lesions. The final outcome was based on surgical diagnosis, repeat FNA, or clinical follow-up. Cases were considered positive for neoplasm if a neoplasm was demonstrated on repeat FNA, on histology, by progression of the disease with clinical follow-up, or by the diagnosis of metastatic disease through imaging, cytologic examination, or histologic examination. Data collected from the chart review included diagnosis, age, sex, time to last follow-up, smoking history, alcohol abuse, diabetes, jaundice, abdominal pain, history of pancreatitis, weight loss, presence of a mass, obstruction of the bile duct, and dilation of the pancreatic duct. EUS-FNA was performed under conscious sedation using a curvilinear echoendoscope (UC-30P; Olympus, Melville, NY) as described previously.6 FNAs were performed using a 22-gauge needle, with a few cases using 19-gauge or 25-gauge needles. A cytopathologist was present for on-site evaluation of adequacy and diagnosis of all pancreatic masses during EUS-FNA. For on-site examinations, only DiffQuik staining (Baxter, McGraw Park, IL) was used. For final pathological interpretation, additional Papanicolaou-stained slides were prepared, as well as hematoxylin and eosin stains for cell block. During the study period multiple pathologists and clinicians were involved in the diagnostic process and the clinical decision regarding management of these patients. We used the following diagnostic categories: positive for neoplasm, suspicious for neoplasm, atypical with a comment, benign, and unsatisfactory. Cell blocks were routinely ordered and examined for further analysis. Flow cytometry was performed for atypical lymphoid infiltrates. The management and decision on the type of follow-up of a case after the diagnosis was largely based on clinical, imaging, and cytologic findings, as we have described previously.7 In our institution, there were no defined criteria for atypical diagnosis, but in general these cytologic specimens showed changes beyond benign, but not severe enough to make a diagnosis of suspicious or positive for a neoplastic process. The suspicious category showed cytologic and architectural features that were short of features Cancer Cytopathology

June 2014

of frank malignancy/neoplasia either because of sampling issues, the nature of the lesion with few characteristic features, or unfamiliar neoplasms. The posttest probability for finding a neoplasm in the atypical category was lower than that of the suspicious for neoplasm category, as we have demonstrated previously.8 Analysis

Statistical analysis was performed using the IBM SPSS statistical analysis program, version 21.0 (IBM Corporation, Armonk, NY). The main analysis focused on the presence of a neoplasm as an end point. In some analyses, in particular as related to risk, the presence of ductal carcinoma was considered a secondary outcome. The atypical cases were cross-tabulated by the method of follow-up, and then the rates of neoplasm detection were compared among the different methods using the chi-square test. Risk factors were stratified by the outcomes and compared with univariate analysis including the Student t and chisquare tests. The risk factors and the methods of followup were entered into a preliminary binary logistic regression model. In addition to sex and age, any factors that were significant at P < .10 in the preliminary model were reentered in a final model in which the predictive factors were considered statistically significant at P < .05. The final model was validated internally using the bootstrap method, used as described previously.9–11 The predictive probability for each case was then calculated using SPSS as described previously. The predictive probability for a case was the probability, “risk,” of having a neoplasm given all the values for the significant predictive factors plus sex and age for that case as expressed in the following formula: Predictive probability 5exp ðb0 1b1 x1 1b2 x2 . . . bn xn = 11exp ½b0 1b1 x1 1b2 x2 . . . bn xn Þ where x is the significant predictor and b is the regression coefficient. To assess model discriminatory power, we constructed an receiver operating characteristic (ROC) curve for the predictive probability, and we estimated the area under the curve using the IBM SPSS procedure. Risk stratification was then performed by identifying cases that had predictive probability high enough to be treated as suspicious/positive for neoplasm or low enough to be treated as negative for neoplasm. The cut values of 429

Original Article Table 1. Frequency (% Column) of Various Neoplasms Diagnosed as Atypical on Cytology Stratified by Method of Follow-Up

Ductal carcinoma Benign Pancreatitis Lymphoma Neurondocrine neoplasm Metastasis Intraductal papillary mucinous neoplasm Mucinous neoplasm Spindle cell lesion Solid pseudopapillary Total a b

FNA

Histology/ Flow

Clinical

Total

12 (63) 4 (21) 2 (11) 1 (5) 0 (0) 0 (0) 0 (0)

24 (36) 7 (10) 11 (16) 10 (15) 5 (7) 3 (4) 2 (3)

25 (30) 35 (43) 16 (20) 1 (1) 3 (4)a 1 (1)b 1 (1)b

61 (37) 46 (28) 29 (17) 12 (7) 8 (5) 4 (2) 3 (2)

0 (0) 0 (0) 0 (0) 19 (100)

2 (3) 2 (3) 1 (1) 67 (100)

0 (0) 0 (0) 0 (0) 82 (100)

2 1 1 167

(1) (1) (1) (100)

Three cases of multiple endocrine neoplasia diagnoses based on imaging.

Table 2. Univariate Analysis of Predictive Factors of Neoplasm

Mean age (SD) Male sex Abdominal pain Weight loss Diabetes Smoking Alcohol Family history of malignancy Diabetes Obesity Obstruction of bile flow History of pancreatitis Jaundice Bile duct dilated Mass in the pancreas

Negative for Neoplasm

Positive for Neoplasm

P

62 (15) 40/94 (43%) 49/78 (63%) 38/82 (46%) 25/75 (33%) 26/76 (34%) 10/76 (13%) 3/73 (4%) 23/72(32%) 9/73 (12%) 19/73 (26%) 31/79 (39%) 28/76 (28%) 29/82 (35%) 62/92 (76%)

65 (12) 55/93 (59%) 50/89 (56%) 50/89 (56%) 26/89 (29%) 34/89 (38%) 34/89 (20%) 7/89 (8%) 26/89 (30%) 23/89 (26%) 34/87 (33%) 16/89 (18%) 34/89 (38%) 32/86 (37%) 74/91 (81%)

n.s. < .001 n.s. n.s. n.s n.s. n.s. n.s. n.s. .02 n.s. .002 n.s. n.s. .02

n.s., Not significant.

predictive probability that corresponded to the benign, suspicious, and positive for neoplasm categories, as well as of treatment and test/no test threshold probabilities were obtained from our previous probabilistic model for EUSFNA8; predictive probabilities (95% CI) for these categories were 0.06 (0.04-0.13), 0.92 (0.77-0.99), and 0.99 (0.96-1.0), respectively. After determining the predictive probability for each case based on the equation above, we reclassified the case into one of these categories based on 95% confidence intervals of these categories and the treatment threshold.8 To internally validate our model, we used a SAS software bootstrap macro (SAS software, SAS Institute Inc., Cary, NC) as described previously.10 Validation entailed using 500 random samples with replacement, running the logistic regression on these samples, and determining the bias (Harrell’s optimism). To calibrate the model, we plotted and compared the agreement between observed and predicted probabilities over the range (deciles) of probability using a regression line.12,13

RESULTS Of 3832 cases with EUS-FNAs, 187 (4.9%) were diagnosed with atypical cytology (56% male; mean age (SD), 65 [13] years). Nineteen (11%) were lost to follow-up. Follow-up included: repeat FNA, 19 (11%); biopsy/resection, 54 (32%); clinical follow-up, 82 (49%); and other types of follow-up, 13 (8%), including flow cytometry (10) and cytology or biopsy of a metastatic site (3). Overall, the median time of follow-up was 77 days; median time for repeat FNA was 75 days, for biopsy/resection was 60 days and for clinical follow-up was 30 days. The out430

comes following an atypical diagnosis were heterogeneous and included pancreatitis, ductal carcinoma, cystic lesions, lymphoma, neuroendocrine neoplasms, and other lesions (Table 1). Five of these lesions (3%) were cystic,and the rest were solid (Table 1). A total of 93 neoplasms (55%), including 61 ductal carcinomas (36%), were detected after an atypical cytologic diagnosis. Similar rates of detecting neoplasms were observed after repeat FNA, 13 of 17 (76%), or biopsy/resection, 41 of 54 (77%), which were significantly higher rates than for clinical follow-up, 31 of 82 (37%), P < .05. Nineteen of the lesions (61%) in the clinical follow-up group were considered neoplastic on clinical and imaging follow-up within the first month of detection. The classical symptoms and signs of a pancreatic neoplasm varied among patients with an atypical cytologic diagnosis. Of those who were followed up, 100 of 166 (72%) had a definitive mass, 36 of 165 (36%) were smokers, 28 of 165 (17%) consumed alcohol, 51 of 164 (31%) were diabetic, 62 of 165 (38%) were jaundiced, 99 of 167 (59%), had abdominal pain, 88 of 168 (52%), had weight loss, 17 of 168 (24%) had a history of pancreatitis, 55 of 163 (34%) had a dilated pancreatic duct, and 61 of 168 (36%) had obstructed bile duct. Univariate analysis indicated that the presence of a mass, obesity, the absence of a history of pancreatitis, and male sex were significantly more likely to be associated with the presence of a neoplasm after an atypical cytologic diagnosis (Table 2). Logistic regression analysis was used to control for confounding factors and to determine the significant Cancer Cytopathology

June 2014

EUS-FNA Predictive Model/Alston et al

Table 3. Multivariate Analysis of Predictive Factors of Neoplasm

Age Female sex Obstruction of bile duct Clinical follow-up Repeat EUS-FNA Biopsy/resection History of pancreatitis Alcohol abuse Mass present Constant

B-Coefficient (SE)

Odds Ratio

0.02 (0.02) 20.11 (0.38) 0.54 (0.40) Base 1.23 (0.60) 1.60 (0.41) 20.94 (0.42) 1.25 (0.52) 0.85 (0.42) 22.65 (1.14)

1.02 (0.99-1.05) 0.89 (0.43-1.86) 1.71 (0.79-3.71) 3.42 4.96 0.39 3.49 2.34 0.07

(1.05-11.14) (2.21-11.14) (0.17-0.89) (1.26-9.66) (1.03-5.31)

P .177 .76 .175 .041

Atypical cytologic diagnostic category in EUS-FNA of the pancreas: follow-up, outcomes, and predictive models.

The objective of this study was to assess how atypical diagnostic category (ADC) is followed up, its outcomes, and the predictors that are associated ...
203KB Sizes 0 Downloads 0 Views