Fine-Needle Aspiration Biopsy of Liver Masses:

Diagnostic Value and Reproducibility of Cytological Criteria

Rosario Granados, M.D.,1* Jose A. Aramburu, M.D.,1 Nieves Murillo, C.T.I.A.C.,1 Encarnacion Camarmo, C.T.I.A.C.,1 Miguel A. de la Cal, M.D.,2 and Pilar Fernandez-Segoviano,
M.D., Ph.D.
1

There are many helpful cytological criteria for the diagnosis of liver ne-needle aspiration biopsies (FNABs), but none of them are pathognomonic of primary or metastatic tumors. We analyzed the diagnostic value and reproducibility of 28 cytological parameters in FNABs from 140 hepatic masses, including 29 benign lesions, 49 hepatocellular carcinomas (HCCs), and 62 metastatic tumors, encompassing 48 adenocarcinomas (ACAs). Five different observers evaluated each sample, and the interobserver and intraobserver agreement was studied. Multivariable analysis showed that the criteria more closely associated with malignancy were irregular nuclear contour, three-dimensional cell groups, and atypical naked nuclei. Capillaries separating tumor cells and granular cytoplasm were strongly associated with HCCs, while eccentrically placed nuclei and necrosis were most commonly seen in ACAs and in metastatic tumors. The intraobserver and interobserver agreement was excellent for the nal cytological diagnosis, and there was fair to very good interobserver agreement for 22 of the 28 criteria studied. Architectural features were less reproducible than pure cytological criteria. Intraobserver variability was not inuenced by the years of experience in the eld. A precise and strict denition of terminology rendered a better reproducibility of the cytological criteria. Diagn. Cytopathol. 2001;25:365375.

2001 Wiley-Liss, Inc.

Key Words: liver; ne-needle aspiration; interobserver; intraobserver

The most common hepatic tumors are metastases, particularly adenocarcinomas from the gastrointestinal tract, and the main differential diagnoses are primary hepatic tumors, 90% of which are hepatocellular carcinomas (HCCs).
1 Department of Pathology, Hospital Universitario de Getafe, Getafe, Madrid, Spain 2 Intensive Care Unit, Hospital Universitario de Getafe, Getafe, Madrid, Spain Grant Sponsor: Ministerio de Sanidad y Consumo, Fondo de Investigacion Sanitaria; Grant number: 96/0185. *Correspondence to: Rosario Granados, M.D., Department of Pathology, Hopital Universitario de Getafe, Carretera de Toledo Km 12.5, Getafe, 28905 Madrid, Spain. E-mail: rgranado@teleline.es Received 10 November 2000; Accepted 18 July 2001

Some serological markers (such as alpha-fetoprotein) are useful for the diagnosis when they are markedly elevated,1 but most patients do not have high levels of these markers at the time of diagnosis. In the liver, ne-needle aspiration biopsies (FNABs) have replaced the more invasive large-needle or open biopsies because of their simplicity and near lack of complications.25 Since most patients will never undergo further diagnostic studies, cytological analysis of FNAB should render a denitive diagnosis. The main diagnostic concerns in this eld are the sensitivity of the procedure to detect malignancy, and the cytological criteria to distinguish primary from metastatic lesions. Although there are many descriptive reports of the cytological ndings of liver tumors,6 22 only a few of them offer a comprehensive analysis to evaluate the most important diagnostic criteria.6 8,12,19,21 However, most of these studies analyzed a few criteria with denitions that overlapped with each other, and they provided different results. There is also a wide range (60 100%) of diagnostic sensitivity of liver FNABs in the literature.2 6,8,18 21 These differences probably reect interpretative and methodological variations, including lack of uniformity in the terminology used for the different cytological parameters analyzed, leading to high interobserver and intraobserver variability. For any diagnostic test, accuracy is as important as reproducibility, which can be inuenced by biological, intraobserver, or interobserver variability. Although these aspects of cytological diagnosis have been analyzed for other organs,23,24 they have not been reported for liver FNAB interpretation. In this study, we tried to establish the diagnostic value of 28 cytological criteria used in FNABs of liver masses, and to evaluate the reproducibility of these parameters by analyzing intraobserver and interobserver agreement.
Diagnostic Cytopathology, Vol 25, No 6

2001

WILEY-LISS, INC. DOI 10.1002/dc.10025

365

GRANADOS ET AL.

Materials and Methods Patients

A total of 140 FNABs of liver masses from patients with histological or clinical denitive diagnosis (gold standard), or both, was the sample for this study. There was a retrospective and a prospective phase. Patients undergoing FNAB of the liver at the Hospital Universitario de Getafe from 19911998 had their histology and clinical history reviewed. There was a total of 255 FNABs, but only 140 of them had a histological and/or clinical unequivocal diagnosis.

Table I. Final Diagnosis of Cases Studied Diagnosis Hepatocellular carcinoma Lymphoma Metastatic tumor Adenocarcinoma Undifferentiated carcinoma Small-cell carcinoma Squamous-cell carcinoma Carcinoid Cholangiocarcinoma Benign lesions Ecchinoccocal cyst Abscess Cirrhosis Adenoma Normal liver Number of cases 49 2 56 44 4 4 2 2 4 29 1 6 8 1 13

Cytological Analysis
Five observers, i.e., 3 pathologists (observers number 1, 3, and 4) and 2 cytotechnologists (observers number 2 and 5) with experience in the eld ranging from 525 yr, reviewed each one of the samples individually, without knowledge of the diagnosis and with no time limit. FNABs were obtained percutaneously with a 22-gauge needle under ultrasound guidance, and the aspirates were immediately smeared onto glass slides by a cytotechnologist. One or two slides were stained with Diff-Quik (Dade Behring) for on site evaluation of the adequacy of the sample. All cases had air-dried slides stained with DiffQuik, and some of them also had alcohol-xed material for Papanicolaou stain.

Denition of Terminology
Before cytological analysis, there were four preliminary meetings among the investigators, where all cytological parameters to be used in the analysis were described in detail. Some FNABs were reviewed in a multiheaded microscope with the purpose of becoming familiar with the different features and their terminology. A total of 28 cytological parameters, 14 considered architectural and 14 classical cytological features, were evaluated in each case and registered in a form. Eight of them had three categories (no, moderate or light, and numerous or severe), since it seemed important for the correct analysis of these features to include an intermediate category. The other 20 cytological parameters were analyzed as to their presence or absence in the sample. The denitions of cytological variables are as follows. Architectural criteria. Architectural criteria include: 1) Hypercellularity, a subjective estimate after reviewing all slides. 2) Cell groups, meaning the presence of cells in cohesive groups, independent of the characteristics of the group. The presence of a single group qualies as moderate. 3) Threedimensional (three-D) groups represent a subjective estimate of the amount of groups with a three-D or ball-like cohesive arrangement. 4) Two-dimensional (two-D) groups involve the presence of cell groups in a sheet-like arrangement. 5) Singled cells are isolated cells with intact cytoplasm. 6) Microtrabecules are cell cords with a 1 4-cell thickness. 7) Macrotrab366
Diagnostic Cytopathology, Vol 25, No 6

ecules are cell cords more than four cells in width. 8) Acinar pattern means tubular arrangements with a central lumen. 9) Endothelial cell lining describes spindled cells partially covering cell groups. 10) Capillaries are seen as parallel endothelial cell arrangements crossing through cell groups. 11) Necrosis includes poorly preserved cell fragments and detritus. 12) Ductal epithelium involves benign cuboidal or columnar cells in sheet-like arrangements. 13) Mixed tumor and benign cell groups are benign cells and tumor cells closely admixed within the same cell group. 14) Fibrosis represents fragments of connective brous tissue. Classical cytological criteria. Classical cytological criteria include: 1) pleomorphism, i.e., subjective estimate of anisonucleosis, meaning variation in nuclear size and shape. 2) Polygonal cells represent a predominance of cells with polygonal cytoplasm and central nuclei. 3) Eccentric nuclei describes cylindrical cells with nuclei at the periphery or pushed by a large vacuole (signet-ring cell). 4) In welldened cell borders, the cytoplasmic edge is easy to be outlined in most cells. 5) For granular cytoplasm, the cell type under study clearly shows granular cytoplasm. 6) Cytoplasmic vacuoles are single or multiple, with a mucinous appearance. Lipid or empty-looking vacuoles should not be considered. 7) Steatosis involves cells with one or more fat (empty-looking) vacuoles. 8) Bile pigment includes intracellular or extracellular pigment, dark-green by Papanicolaou or brown-greenish by Diff-Quik stain. 9) In increased nuclear/cytoplasmic (N/C) ratio, cells with a ratio 0.5 are considered normal. Values between 0.51 are classied as mild increases, and ratios above 1 are considered severe. 10) Irregular nuclear contour is considered moderate when less than 25% of the cells contain irregularities of the nuclear membrane. When present in more than 25% of the cells or when there are monstrous forms in a few cells, irregular nuclear contour is classied as severe. 11) Macronucleolus describes a nucleolus that occupies more than

FINE-NEEDLE ASPIRATION BIOPSY OF LIVER MASSES

Table II. Cytological Parameters With Statistically Signicant Values (P Cytological criteria Hypercellularity No Yes Cell groups No Moderate Numerous Three-D groups No Moderate Numerous Two-D groups Moderate Numerous Microtrabecules No Yes Macrotrabecules No Yes Acinar pattern No Yes Endothelial lining No Yes Ductal epithelium No Yes Mixed cell groups No Yes Pleomorphism No Moderate Severe Benign lesions (%) 41 59 24 55 21 93 7 0 24 41 35 48 52 86 14 93 7 100 0 38 62 100 0 72 28 0 Malignant lesions (%) 23 77 3 46 51 13 58 29 24 65 11 68 32 61 39 43 57 86 14 70 30 69 31 2 34 64 P value 0.05 0.05) to Discriminate Malignant From Benign Hepatic Lesions Cytological criteria Polygonal cells No Yes Eccentric nuclei No Yes Dened borders No Yes Granular cytoplasm No Yes Cytoplasmic vacuoles No Yes Steatosis No Yes Increased N/C No Mild Severe Irregular nucleus No Moderate Severe Macronucleolus No Yes Atypical naked nuclei No Few Many Mitoses No Yes Benign lesions (%) 24 76 100 0 28 72 24 76 97 3 59 41 72 28 0 76 24 0 90 10 79 21 0 100 0 Malignant lesions (%) 52 48 42 58 59 41 51 49 66 34 79 21 1 33 66 1 30 69 43 57 20 44 36 69 31 P value 0.006

0.0001

0.0001

0.003

0.0001

0.007

0.006

0.0003

0.04

0.02

0.008

0.0001

0.0001

0.0001

0.02

0.001

0.0001

0.0001

0.0001

0.0001

0.0001

10% of the nuclear surface in more than 10% of the cells under study. 12) Nuclear vacuoles involve the presence of optically empty nuclear inclusions. 13) Atypical naked nuclei are those without cytoplasm and with atypical features, such as irregular shape or size. 14) Mitoses involve the presence of any mitotic gure in the sample. After classifying the cytological parameters described, there was also a nal cytological diagnosis given for each case. Finally, the observers reached a consensus diagnosis for each parameter and for the nal diagnosis in each case. A consensus diagnosis was dened as the agreement of three or more observers on a given variable, and a form with this diagnosis was lled out for each case. To study intraobserver variability, 28 cases (representing 20% of the sample) were reviewed a second time by each observer, at least 3 mo after the rst examination.

For histological diagnosis, material from cell block preparations, biopsies, or autopsies was used. It was considered clinically diagnostic of hepatocellular carcinoma when the serologic level of alpha-fetoprotein was above 1,000 ng/ml.

Statistical Analysis
Univariate analysis by chi-square or Fishers exact test was used to select the cytological criteria discriminating benign from malignant and primary from metastatic tumors. For multivariate analysis, stepwise logistic regression analysis was performed. All variables with P values equal to or under 0.1 in univariate analysis were entered into the multivariate analysis, and factors were considered signicant if the P value was equal to or under 0.05 in the multivariate analysis. For the multivariate analysis, and in order to achieve a higher precision of the regression coefcients, only two categories were considered for each variable. Therefore, in eight criteria, the three categories were reduced to two.
Diagnostic Cytopathology, Vol 25, No 6

Gold-Standard Diagnosis
Either histological diagnosis or clinical denitive diagnosis was used as the nal or gold-standard diagnosis.

367

GRANADOS ET AL.
Table III. Cytological Features Not Discriminating Benign From Malignant Liver Lesions (P 0.05) Cytological criteria Singled cells No Moderate Numerous Capillaries No Yes Necrosis No Yes Fibrosis No Yes Bile pigment No Yes Nuclear vacuoles No Yes Benign lesions (%) 24 48 28 66 34 79 21 48 52 69 31 72 28 Malignant lesions (%) 14 62 24 50 50 63 37 50 50 82 18 63 37 P value 0.28

0.11

0.07

0.53 Fig. 1. Numerous atypical naked nuclei present in the background of a smear from a hepatocellular carcinoma. Nuclei are pleomorphic and contain numerous irregular nucleoli (Diff-Quik stain, 200).

0.10

0.33

Table IV. Cytological Criteria Selected by Stepwise Logistic Regression Analysis as More Closely Associated With Malignancy When Benign and Malignant Liver Lesions Were Compareda Variable Irregular nuclei Three-D groups Atypical naked nuclei
a

OR 71.2 25.5 9.1

95% CI 5.6897 2.6254.8 1.270.9

P value 0.001 0.005 0.03

OR, odds ratio; CI, condence interval.

Kappa statistics, used to calculate the extent of the agreement between two experts above and beyond chance agreement,25 were performed to estimate the interobserver and intraobserver agreement. As reported by Landis and Koch,26 a kappa value higher than 0.75 was considered excellent; between 0.4 0.75 the agreement was fair to very good; and values below 0.40 reected poor agreement.

Results
From the 140 liver lesions studied, there were 29 benign lesions, 49 HCCs, and 62 metastatic tumors (see Table I). In order to simplify the analysis, the four cases of cholangiocarcinoma were included in the group of metastatic tumors, since their cytological distinction from metastatic adenocarcinoma is not possible. The two cases of lymphoma were also placed in this group.

Cytological Analysis
Benign vs. malignant. Twenty-two of the 28 cytological criteria studied showed statistically signicant differences when benign and malignant lesions were compared. The contingency tables with these criteria are shown in Table II. Some of these cytological parameters, including the presence of endothelial lining, mixed benign and malignant cell 368
Diagnostic Cytopathology, Vol 25, No 6

groups, eccentric nuclei, and mitoses, were only seen in malignant lesions. In addition, only malignant lesions had numerous three-D cell groups or naked atypical nuclei. However, 7% and 21% of benign lesions had a moderate number of three-D groups and atypical naked nuclei, respectively. Only 20% of malignant tumors did not show atypical bare nuclei in the smears. None of the benign lesions had a severe degree of pleomorphism, increased N/C ratio, or irregular nuclear membranes, but 28% of them showed moderate pleomorphism and mild increase in N/C ratio, and 24% showed a moderately irregular nuclear contour. These ndings, all of which are nuclear features related to malignancy, were particularly seen in cases of cirrhosis. As can be expected, samples from malignant lesions were more cellular, and had more cell groupings, macrotrabecules, acinar pattern, cytoplasmic vacuoles, and macronucleoli, than benign lesions. In contrast, the slides from the benign group had frequent microtrabecules, ductal epithelium, and polygonal cells with well-dened cell borders and granular cytoplasm. There was also more steatosis in benign lesions. The presence of singled cells, capillaries, necrosis, brosis, bile pigment and intranuclear vacuoles was not significantly different between benign and malignant lesions (Table III). Since bile pigment and intranuclear vacuoles are features of hepatic cells, there was a similar distribution among benign and malignant lesions due to the presence of primary hepatic tumors within the malignant group. Although necrosis is more commonly found in malignancies, it is also seen in hepatic abscesses, which represented six cases of our group of benign lesions. Multivariate analysis by logistic regression selected three cytological criteria as the most strongly associated with malignant tumors (see Table IV). They were: irregular nuclear contour (OR, 71.2; 95% CI, 5.6 897), three-D cell

FINE-NEEDLE ASPIRATION BIOPSY OF LIVER MASSES

Table V. Cytological Features Discriminating Primary From Metastatic Disease
Cytological criteria Cell groups No Moderate Numerous Two D groups No Moderate Numerous Microtrabecules No Yes Macrotrabecules No Yes Endothelial lining No Yes Ductal epithelium No Yes Pleomorphism No Moderate Severe Polygonal cells No Yes Eccentric nuclei No Yes Dened borders No Yes Granular cytoplasm No Yes Cytoplasmic vacuoles No Yes Steatosis No Yes Increased N/C No Mild Severe Irregular nucleus No Moderate Severe Mitoses No Yes Atypical naked nuclei No Moderate Numerous Necrosis No Yes Bile pigment No Yes Capillaries No Yes Nuclear vacuoles No Yes Primary (%) 0 35 65 6 80 14 39 61 22 78 69 31 61 39 4 51 45 12 88 78 22 27 73 4 96 84 16 55 45 2 55 43 2 45 53 80 20 22 31 47 84 16 63 37 10 90 35 65 Metastatic (%) 5 55 40 39 53 8 92 8 92 8 98 2 77 23 0 21 79 84 16 15 85 84 16 89 11 52 48 98 2 0 16 84 0 18 82 61 39 18 55 27 47 53 97 3 82 18 90 10 P value 0.0005

0.001

0.001

0.001

0.001

0.05

Fig. 2. Bidimensional group of cells from a hepatocellular carcinoma. Cells are polygonal, with centrally placed nuclei, and granular, welldened cytoplasm (Diff-Quik stain, 400).

0.001

0.001

0.001

0.001

0.001

0.001

0.001

Fig. 3. Numerous capillaries among tumor cells from a hepatocellular carcinoma (Diff-Quik stain, 200).

0.001

0.001

0.03

0.03

0.001

0.001

0.001

0.001

groups (OR, 25.5; 95% CI, 2.6 254.8), and atypical naked nuclei (Fig. 1) (OR, 9.1; 95% CI, 1.270.9). In three-D groups and in atypical naked nuclei, the categories of moderate and numerous were lumped together and compared with their absence. Likewise, for irregular nuclear contour, moderate and severe were jointly compared against lack of irregular nuclei. Sensitivity of all three selected cytological criteria was 99% (110 cases), but specicity was only 76%. Primary vs. metastatic malignant lesions. There were 21 cytological parameters with signicantly different distributions (Table V). Among the most striking differences were the presence of macrotrabecules and endothelial lining in primary liver tumors. Together with polygonal cells, granular cytoplasm (Fig. 2), intranuclear vacuoles, well-dened cell borders, steatosis, and bile pigment, they were strongly associated with HCC. The presence of capillaries in the smears (Fig. 3) was signicantly higher in primary tumors
Diagnostic Cytopathology, Vol 25, No 6

369

GRANADOS ET AL.
Table VI. Cytological Features Not Discriminating Primary From Metastatic Tumors Cytological criteria Acinar pattern No Yes Mixed cell groups No Yes Fibrosis No Yes Hypercellularity No Yes Three-D groups No Moderate Numerous Macronucleolus No Yes Singled cells No Moderate Numerous Primary (%) 47 53 73 27 43 57 16 84 16 65 19 35 65 8 74 18 Metastatic (%) 40 60 66 34 55 45 29 71 11 52 37 50 50 18 53 29 P value 0.31

0.27

0.14

0.09

0.09

Fig. 4. Tumor cells from a metastatic adenocarcinoma. Cells show eccentrically placed nuclei and cytoplasmic vacuoles (Diff-Quik stain, 400).

(90%) than in metastatic lesions (18%). On the other hand, the nuclear features commonly associated with malignancy, such as pleomorphism, irregular nuclear membrane, increased N/C ratio, and mitoses, were more marked in metastatic tumors. Necrosis, eccentric nuclei, and cytoplasmic vacuoles (Fig. 4) were associated with metastatic lesions as well. Although with a different distribution, there was ductal epithelium present in both groups and also atypical naked nuclei in both, but they were more numerous in HCCs. Seven of the 28 cytological parameters studied did not show signicant differences in their association with primary or metastatic tumors (see Table VI). Logistic regression analysis chose granular cytoplasm (OR, 0.03; 95% CI, 0.004 0.21) and capillaries among tumor cells (OR, 0.04; 95% CI, 0.006 0.30) as the two cytological variables more commonly associated with HCC (see Table VII). These two criteria were seen together in 96% (47 cases) of HCCs in our sample, and their specicity was of 89%. The two cytological features more strongly associated with metastatic tumors were eccentric nuclei (OR, 7.4; 95% CI, 1.151.6) and the presence of necrosis (OR, 8.0; 95% CI, 1.0 66.9). Both were present in 85% of metastatic tumors, and the specicity was 78%. HCC vs. adenocarcinoma. When we compared the features of HCCs with those of ACAs (n 48), the results were very similar to those obtained from comparing HCCs with all metastatic tumors (Tables VIII and IX). The presence of eccentric nuclei and necrosis were the criteria most closely related to ACA in this model as well (Table VII).

0.08

0.08

In 22 of the 28 cytological parameters studied, the observers reached fair to very good agreement (kappa values of 0.40 0.75), and in six of these criteria, the agreement was poor (Table X). The observers showed fair to very good concordance for 6 of the 7 criteria selected by the multivariate analysis as the most diagnostic of a malignant lesion (primary, metastatic, or both). However, there was poor agreement (k 0.35) for the presence of three-D groups. This variable is part of a group of architectural criteria that describe the type of cell-grouping present in smears. For all of them there was poor interobserver agreement (k 0.40). They included the amount of singled cells, two-D or three-D cell groups, and the presence of mixed benign and malignant cell groups in the slides. There was also poor agreement on the evaluation of brosis, which we found quite subjective.

Intraobserver Variability
Observers 1 and 4 had perfect agreement (k 1) for the nal diagnosis of benign vs. malignant lesions (Table XI). Observer 1 also had perfect agreement for the diagnosis of HCC vs. metastatic tumor. Observer 4 had excellent agreement (k 0.89) for this last variable. Observers 2, 3, and 5 had very good to excellent intraobserver agreement (k 0.72 0.90) for both nal diagnoses. The three observers that proved more consistent with their nal diagnoses were numbers 1, 3, and 4. These observers were pathologists. Observers 2 and 5 were cytotechnologists, and reached very good to excellent concordance for their nal diagnosis. The agreement for the 28 cytological parameters varied. The overall concordances measured by the mean kappa

Interobserver Variability
There was excellent agreement (kappa (k) 0.75) for the nal cytological diagnosis among the ve observers. This included the distinction between benign and malignant (k 0.86), as well as the classication of tumors into primary and metastatic (k 0.76). 370
Diagnostic Cytopathology, Vol 25, No 6

FINE-NEEDLE ASPIRATION BIOPSY OF LIVER MASSES

Table VII. Cytological Criteria Selected by Stepwise Logistic Regression Analysis When HCCs and Metastatic Tumors or HCCs and ACAs (Values in Parentheses) Were Compareda Variable Necrosis Eccentric nuclei Capillaries Granular cytoplasm
a

OR 8.0 (8.0) 7.4 (8.9) 0.04 (0.06) 0.03 (0.04)

95% CI 166.9 (163.7) 1.151.6 (0.983.3) 0.0060.3 (0.00863.7) 0.0040.21 (0.0050.29)

P value 0.05 (0.05) 0.04 (0.05) 0.001 (0.006) 0.0004 (0.001)

Association MT (ACA) MT (ACA) HCC HCC

OR, odds ratio; CI, condence interval; MT, metastatic tumor.

value for the 28 criteria were 0.52, 0.54, 0.57, 0.57, and 0.59 for observers 15, respectively. Therefore, the mean intraobserver agreement did not differ signicantly among the ve observers. In addition, observer 4 had excellent agreement in eight cytological criteria, observer 1 in seven, observers 3 and 5 in ve, and observer 2 in four. When cytological criteria were listed on descending order based on the mean kappa values from all ve observers (Table XII), we saw that, overall, classical cytological criteria based on nuclear and cytoplasmic features obtained better degrees of concordance than did architectural features.

Sensitivity and Specicity

We found very high sensitivity and specicity of this technique when the consensus diagnosis was analyzed. Likewise, these parameters were very high in each one of the ve individual evaluations (Table XIII). Sensitivity estimation of the second-time evaluation to study intraobserver variability showed values with 100% sensitivity, specicity, and diagnostic accuracy in all ve observers (Table XIV). These results were obtained considering the sensitivity to diagnose a malignant case as malignant. When the ability to subclassify a malignant case as primary or metastatic was analyzed, the sensitivity ranged from 95100%.

Discussion
In some instances, the distinction between benign and malignant liver lesions may be extraordinarily difcult. In addition, tumors that are readily interpreted as malignant, when poorly differentiated, may be difcult to classify. Adenocarcinomas are the main source of diagnostic discrepancies in FNABs from poorly differentiated HCCs. Features classically described in ACAs, such as acinar arrangements, are seen in many HCCs.12,14,17 Other metastatic tumors that are less common, such as lymphomas, carcinoid tumors, squamous-cell carcinomas, or small-cell carcinomas, have been reported as diagnostic dilemmas in the differential diagnosis.5,10,22 Some of these metastatic tumors may pose problems in the diagnosis of subtypes of HCCs, such as the small-cell type.27 Although there are many cytological criteria reported to be useful in the evaluation of FNABs of liver lesions, none of them are pathognomonic of HCCs, metastatic tumors, or

benign lesions. Therefore, there is a need to order these criteria, and select the most important for diagnosis. In our study, we analyzed the diagnostic value and reproducibility of 28 cytological criteria for FNAB of benign, malignant primary, and metastatic liver lesions. We also studied ACAs (cholangiocarcinomas and metastatic ACAs) as a separate subset, since they are considered the main differential diagnosis for HCC. By stepwise logistic regression analysis, we found that comparing the group of benign lesions with all malignant tumors (primary and metastatic), the presence of irregular nuclear contour, three-D groups, and atypical naked nuclei were strongly associated with malignancy. These results differ somewhat from others previously reported. While Cohen et al.6 and others11,12,21 found the presence of atypical naked nuclei to be a key cytological criterion for diagnosing HCC, Sole et al.8 reported their presence in a con siderable percentage of both benign lesions and HCCs, and considered them nonspecic. In our hands, abundant atypical naked nuclei were only seen in malignant tumors (primary and metastatic), and were signicantly more common in HCCs (P 0.03). Although benign lesions were never found to have a large number of these nuclei, 21% of them had a few dispersed bare nuclei that were interpreted as atypical. In this case, the use of three categories is necessary for a better understanding of the cytological ndings. Likewise, the presence of three-dimensional groups, a criterion usually associated with malignancy in cytology, was seen, in moderate amounts, in two (7%) of our benign cases, which corresponded to a liver-cell adenoma and a case of cirrhosis. No benign case had numerous three-D groups. The third criterion found to be strongly associated with malignancy was irregular nuclear contour. Again, although we could not nd severe nuclear irregularities in any of the benign lesions, 24% of them had a moderately irregular nuclear contour, corresponding mainly to cirrhotic lesions. Therefore, breaking down a cytological criterion into three categories proved to be important for the correct interpretation of the results. While simplication of the terminology and grading is desirable for an easier analysis in most studies, an excessive simplication may give rise to diagnostic discrepancies. Cohen et al., in a study comparing 10 cytological diagnostic criteria on FNAB from 52 HCCs and 30 nonneoplasDiagnostic Cytopathology, Vol 25, No 6

371

GRANADOS ET AL.
Table VIII. Cytological Features Discriminating HCCs From ACAs
HCC (%) 16 65 19 6 80 14 39 61 22 78 69 31 4 51 45 12 88 78 22 27 73 4 96 ACA (%) 4 56 40 33 61 6 92 8 92 8 100 0 0 19 81 83 17 8 92 85 15 88 12 P value 0.02

Table IX. Cytological Features Not Discriminating HCCs From ACAs Cytological criteria Fibrosis No Yes Macronucleolus No Yes Mixed cell groups No Yes Cell groups No Moderate Numerous Ductal epithelium No Yes Singled cells No Moderate Numerous Hypercellularity No Yes Acinar pattern No Yes HCC (%) 43 57 35 65 73 27 0 35 65 61 39 8 74 18 16 84 47 53 ACA (%) 48 52 42 58 65 35 0 54 46 73 27 21 58 21 31 69 29 71 P value 0.38

Cytological criteria Three-D groups No Moderate Numerous Two-D groups No Moderate Numerous Microtrabecules No Yes Macrotrabecules No Yes Endothelial lining No Yes Pleomorphism No Moderate Severe Polygonal cells No Yes Eccentric nuclei No Yes Dened borders No Yes Granular cytoplasm No Yes Cytoplasmic vacuoles No Yes Steatosis No Yes Increased N/C No Mild Severe Irregular nucleus No Moderate Severe Mitoses No Yes Atypical naked nuclei No Moderate Numerous Necrosis No Yes Bile pigment No Yes Capillaries No Yes Nuclear vacuoles No Yes

0.003

0.31

0.23

0.001

0.001

0.18

0.001

0.16

0.002

0.16

0.001

0.07

0.001

0.06

0.001

0.001

84 16 55 45 2 55 43 2 45 53 80 20

44 56 98 2 0 15 85 0 17 83 60 40

0.001

0.001

0.001

0.009

0.03

22 31 47 84 16 63 37 10 90 35 65

19 56 25 46 54 98 2 77 23 88 12

0.03

0.001

0.001

0.001

0.001

tic liver lesions, found that the three cytological features predictive of HCC were the increased N/C ratio, the trabecular pattern, and the presence of atypical naked hepatocytic nuclei.6 Sole et al., in a series of 102 FNAB from HCC and 28 from benign liver lesions, found that irregular arrangement of cells in clusters, irregular chromatin pattern, and uniformly smaller cytoplasm were the three most predictive features for diagnosing HCC.8 They studied 39 cytological parameters. Some of these parameters overlapped with those described by Cohen et al.,6 and others were not easy to recognize cytologically by the reader. These two studies found different parameters to be most diagnostic of HCC, and they also showed conicting results when the same variables were analyzed. In our study, granular cytoplasm and capillaries separating tumor cells were the two cytological criteria more closely related to HCC. Eccentric nuclei and necrosis were strongly associated with metastatic tumors, and with ACAs when they were studied separately. All nuclear features commonly associated with malignancy, such as pleomorphism, irregular nuclear outline, increased N/C ratio, and mitoses, were more severe in metastatic tumors. Bottles et al.7 found that polygonal cells with centrally placed nuclei, malignant cells separated by capillaries, and the presence of bile were predictive for HCC. They found granular, well-dened cytoplasm in 91.4% of HCCs and in 17.6% of metastatic tumors. In our study, we had granular cytoplasm in 96% of HCCs and in 11% of metastases.

372

Diagnostic Cytopathology, Vol 25, No 6

FINE-NEEDLE ASPIRATION BIOPSY OF LIVER MASSES

Table X. Interobserver Agreement Cytological criteria Benign vs. malignant Primary vs. metastatic Granular cytoplasm Polygonal cells Hypercellularity Macrotrabecules Capillaries Dened cytoplasmic borders Cell groups Pleomorphism Eccentric nuclei Steatosis Increased N/C Ductal epithelium Necrosis Endothelial lining Macronucleoli Nuclear vacuoles Irregular nucleus Acinar pattern Mitoses Microtrabecules Atypical naked nuclei Bile pigment Cytoplasmic vacuoles Singled cells Three-D groups Mixed cell groups Fibrosis Two-D groups Kappa value 0.86 0.76 0.71 0.69 0.63 0.63 0.61 0.58 0.57 0.55 0.55 0.54 0.54 0.52 0.50 0.49 0.49 0.49 0.48 0.45 0.44 0.44 0.43 0.41 0.39 0.39 0.35 0.33 0.32 0.20 Table XI. Kappa Values for Intraobserver Variabilitya Cytological criteria Hypercellularity Cell groups Three-D groups Two-D groups Singled cells Microtrabecules Macrotrabecules Acinar pattern Endothelial lining Capillaries Necrosis Ductal epithelium Mixed cell groups Fibrosis Pleomorphism Polygonal cells Eccentric nuclei Dened borders Granular cytoplasm Cytoplasmic vacuoles Steatosis Bile pigment Increased N/C Irregular nucleus Macronucleolus Nuclear vacuoles Atypical naked nuclei Mitoses Benign vs. malignant Primary vs. metastatic
a

OBS. 1 0.58 0.49 0.42 0.41 0.07 0.85 0.92 0.41 0.60 0.79 0.43 0.38 0.13 0.17 0.30 1.00 0.61 0.86 0.93 0.41 0.62 0.75 0.47 0.58 0.28 0.62 0.15 0.21 1.0 1.0

OBS. 2 0.73 0.64 0.26 0.59 0.44 0.71 0.70 0.44 0.70 0.48 0.61 0.27 0.30 0.29 0.50 0.77 0.50 0.63 0.85 0.29 0.83 0.64 0.80 0.74 0.47 0.23 0.33 0.41 0.76 0.72

OBS. 3 0.69 0.60 0.66 0.67 0.13 0.86 0.35 0.71 0.26 0.42 0.58 0.75 0.50 0.55 0.51 0.65 0.23 0.71 0.79 0.28 0.58 0.76 0.66 0.71 0.62 0.78 0.44 0.64 0.90 0.77

OBS. 4 0.68 0.35 0.51 0.38 0.51 0.58 0.81 0.38 0.78 0.30 0.58 0.84 0.44 0.24 0.76 0.79 0.52 0.72 0.79 0.26 0.83 0.44 0.82 0.65 0.53 0.71 0.55 0.30 1.0 0.89

OBS. 5 0.63 0.64 0.55 0.56 0.46 0.58 0.77 0.57 0.50 0.34 0.75 0.52 0.33 0.52 0.50 0.79 0.75 0.56 0.71 0.55 0.52 0.79 0.67 0.70 0.70 0.60 0.49 0.48 0.76 0.72

OBS., observer.

To our knowledge, there is no recent report in the English-language literature evaluating the reproducibility of cytological diagnostic criteria in FNAB of the liver. In this study, we analyzed the reproducibility of 28 cytological criteria by measuring the intraobserver and interobserver variability. After independent evaluation of all samples by ve different observers, 22 of the 28 criteria showed fair to very good reproducibility. We found that architectural criteria, in general, were less reproducible than the classical cytological criteria referring to the characteristics of the nuclei or cytoplasm. This is probably due to the fact that pure cytological criteria can be better dened and quantied, while the interpretation of architectural criteria is more subjective. An exception within the group of architectural features were the variables of hypercellularity and cell groups. In the rst case, there was a very good interobserver agreement (k 0.63) on calling a sample hypercellular, which probably can be explained by the fact that cytologists are trained to constantly estimate this variable in almost every sample. In the case of the presence of cell groups in the specimen (k 0.57), the denition was precise, since all samples qualied when they had any cell grouping at all. Likewise, the observers reached a good agreement in the evaluation of macrotrabecules (k

0.63) and microtrabecules (k 0.44), two variables that were perfectly dened as thicker or thinner than a four-cell width. The mean kappa values from the intraobserver variability study for those two criteria were very good (0.71 and 0.72). It was noteworthy that the overall intraobserver agreement for the presence of atypical naked nuclei was poor (k 0.39), and that the interobserver agreement in this variable was barely fair (k 0.43). It is the case of a cytological criterion with a high diagnostic value, but poor reproducibility. This may account for the reported differences regarding the importance of this criterion in FNAB of the liver. Although the presence of three categories instead of two (present or absent) is a cause for lower agreement on most criteria, we did not consider this factor sufciently important as to eliminate the third category. We found that the presence of cell groups (k 0.57), cellular pleomorphism (k 0.55), increased N/C ratio (k 0.54), and irregular nuclei (k 0.48) were reproducible cytological criteria with three categories. Our interpretation of these results is that reproducibility of diagnostic criteria is heavily dependent on their clear, precise, and strict denition at the beginning of the study.
Diagnostic Cytopathology, Vol 25, No 6

373

GRANADOS ET AL.
Table XII. Mean Kappa Values for Intraobserver Variability Mean kappa 0.84 0.82 0.81 0.80 0.72 0.71 0.70 0.68 0.68 0.68 0.68 0.66 0.59 0.59 0.57 0.55 0.54 0.52 0.52 0.52 0.51 0.50 0.48 0.47 0.41 0.39 0.36 0.35 0.34 0.32 Sensitivity Specicity False positives False negatives Positive P.V. Negative P.V. Accuracy Subclassication
a

Table XIV. Sensitivity Analysis After Second Evaluation (n OBS. 1 100 100 0 0 100 100 100 100 OBS. 2 100 100 0 0 100 100 100 100 OBS. 3 100 100 0 0 100 100 100 95 OBS. 4 100 100 0 0 100 100 100 95

28) a OBS. 5 100 100 0 0 100 100 100 100

Cytological criteria Benign vs. malignant Primary vs. metastatic Granular cytoplasm Polygonal cells Microtrabecules Macrotrabecules Dened borders Steatosis Bile pigment Increased N/C Irregular nucleus Hypercellularity Necrosis Nuclear vacuoles Endothelial lining Ductal epithelium Cell groups Two-D groups Eccentric nuclei Macronucleolus Pleomorphism Acinar pattern Three-D groups Capillaries Mitoses Atypical naked nuclei Cytoplasmic vacuoles Fibrosis Mixed cell groups Singled cells

OBS., observer; P.V., predictive value. Subclassication accounts for all malignant tumors correctly classied as primary or metastatic. All numbers are percentages.

Table XIII. Sensitivity Analysis After First Evaluation (n Consensus Sensitivity Specicity False positives False negatives Positive P.V. Negative P.V. Accuracy Subclassication
a

140) a OBS. 5 100 83 5 0 96 100 96 88

OBS. 1 99 97 1 1 99 97 99 86

OBS. 2 100 83 5 0 96 100 96 85

OBS. 3 95 97 1 5 99 85 96 77

OBS. 4 97 97 1 3 99 90 97 77

100 97 1 0 99 97 99 84

OBS., observer; P.V., predictive value. Subclassication accounts for all malignant tumors correctly classied as primary or metastatic. All numbers are percentages.

The overall intraobserver agreement, measured as the mean kappa values for the 28 criteria, was very similar for all ve observers (range, 0.52 0.59). Therefore, amount of experience in the eld did not affect intraobserver concordance. As happened with interobserver agreement, the classical cytological features related to nuclear and cytoplasmic characteristics showed better concordance than architectural features in general. We also calculated the interobserver and intraobserver agreement for the nal diagnosis broken down into two 374
Diagnostic Cytopathology, Vol 25, No 6

variables, the distinction between benign and malignant lesions, and the diagnosis of primary vs. metastatic tumors. For both diagnostic categories, the interobserver agreement was excellent (kappa values of 0.86 and 0.76, respectively), and the intraobserver concordance ranged from very good to perfect (k 0.721). These ndings show that the observers agreed better in the nal diagnosis than in the diagnostic features that led to the diagnosis. These data suggest that there is a conjunction of factors or morphological features that are not measurable in our study and that permit a consistent nal diagnosis. Although reproducibility and accuracy are not synonymous, in our experience the interpretation of the different observers was also accurate (96 99% diagnostic accuracy). This study showed very high sensitivity (100%), specicity (97%), and diagnostic accuracy (99%) when the consensus diagnosis was evaluated. We also found high sensitivities (95100%) for the diagnoses of each of the ve observers. Observers 2 and 5 had the highest sensitivity (100%), but the lowest specicity (83%), due to a larger number of false-positive diagnoses. These two observers are cytotechnologists, and the results show excellent sensitivity, which is the most desirable parameter for a screening test in a laboratory. Cytopathologists need to nd a balance of a high enough sensitivity, while maintaining high specicity. In our study, all three pathologists had 97% specicity, and the sensitivity ranged from 95100%. The ability to classify a malignant tumor as primary or metastatic was also very high in our group (77 88%). When we analyzed the sensitivity values for each observer at the rst and second evaluations, we realized that there had been a considerable improvement in all investigators, with a sensitivity, specicity, and accuracy at diagnosing malignancy of 100%. Although the sample in the second evaluation was considerably smaller (28 vs. 140), it is reasonable to assume that the improvement was due to self-training during the rst round of evaluations. In conclusion, we found that the cytological features more strongly associated with malignant liver lesions are increased N/C ratio, irregular nuclear contour, and atypical

FINE-NEEDLE ASPIRATION BIOPSY OF LIVER MASSES

naked nuclei. The two criteria more closely associated with HCC are the presence of capillaries among tumor cells and granular cytoplasm. Eccentric nuclei and necrosis were the most useful criteria for the diagnosis of metastatic tumors and ACAs. In general, architectural features were less reproducible than classical cytological features. Interobserver and intraobserver variability may be the cause of differences in sensitivity of liver FNAB among reported studies. We also found that intraobserver and interobserver agreement for the nal diagnosis surpassed that obtained for the individual criteria. Therefore, it is probably a conjunction of morphological features, rather than specic selected criteria, that leads to the correct nal diagnosis. Intraobserver variability was not inuenced by the years of experience of the observer. In view of our results, we believe that probably the most determinant factor for reproducibility of cytological criteria is a strict and precise denition of the terminology.

References
1. Sato Y, Nakata K, Kato Y, Shima M, Ishii N, Koji T, Taketa K, Endo Y, Nagataki S. Early recognition of hepatocellular carcinoma based on altered proles of alpha-fetoprotein. N Engl J Med 1993;328:1802 1806. 2. Cochand-Priolet B, Chagnon S, Ferrand J, Blery M, Hoang C, Galian A. Comparison of cytologic examination of smears and histologic examination of tissue cores obtained by ne needle aspiration biopsy of the liver. Acta Cytol 1987;31:476 480. 3. Bell DA, Carr CP, Szyfelbein WM. Fine needle aspiration cytology of focal liver lesions. Results obtained with examination of both cytologic and histologic preparations. Acta Cytol 1986;30:397 402. 4. Pinto MM, Avila NA, Heller CI, Criscuolo EM. Fine needle aspiration of the liver. Acta Cytol 1988;32:1521. 5. Whitlatch S, Nunez C, Pitlik DA. Fine needle aspiration biopsy of the liver. A study of 102 consecutive cases. Acta Cytol 1984;28:719 725. 6. Cohen MB, Haber MM, Holly EA, Ahn DK, Bottles K, Stoloff AC. Cytologic criteria to distinguish hepatocellular carcinoma from nonneoplastic liver. Am J Clin Pathol 1991;95:125130. 7. Bottles K, Cohen MB, Holly EA, Chiu S-H, Abele JS, Cello JP, Lim RC, Miller TR. A step-wise logistic regression analysis of hepatocellular carcinoma. Cancer 1988;62:558 563. 8. Sole M, Calvet X, Cuberes T, Maderuelo F, Bruix J, Bru C, Rey MJ, Serna N, Cardesa A. Value and limitations of cytologic criteria for the diagnosis of hepatocellular carcinoma by ne needle aspiration biopsy. Acta Cytol 1993;37:309 316. 9. Salomao DR, Lloyd RV, Goellner JR. Hepatocellular carcinoma: needle biopsy ndings in 74 cases. Diagn Cytopathol 1997;16:8 13.

10. Greene CA, Suen KC. Some cytologic features of hepatocellular carcinoma as seen in ne needle aspirates. Acta Cytol 1984;28:713 718. 11. Pedio G, Landolt U, Zobeli L, Gut D. Fine needle aspiration of the liver. Signicance of hepatocytic naked nuclei in the diagnosis of hepatocellular carcinoma. Acta Cytol 1988;32:437 442. 12. Das DK. Cytodiagnosis of hepatocellular carcinoma in ne-needle aspirates of the liver: its differentiation from reactive hepatocytes and metastatic adenocarcinoma. Diagn Cytopathol 1999;21:370 377. 13. Ali MA, Aktar M, Mattingly RC. Morphologic spectrum of hepatocellular carcinoma in ne needle aspiration biopsies.Acta Cytol 1986; 30:294 302. 14. Wee A, Nilsson B, Chan-Wilde C, Yap I. Fine needle aspiration biopsy of hepatocellular carcinoma. Some unusual features. Acta Cytol 1991; 35:661 670. 15. Tao LC, Ho CS, McLoughlin MJ, Evans WK, Donat EE. Cytologic diagnosis of hepatocellular carcinoma by ne-needle aspiration biopsy. Cancer 1984;53:547552. 16. Gupta SK, Das DK, Rajwanshi A, Bhusnurmath SR. Cytology of hepatocellular carcinoma. Diagn Cytopathol 1986;2:290 294. 17. Suen KC. Diagnosis of primary hepatic neoplasms by ne-needle aspiration cytology. Diagn Cytopathol 1986;2:99 109. 18. Bottles K, Cohen M. An approach to ne-needle aspiration biopsy diagnosis of hepatic masses. Diagn Cytopathol 1991;7:204 210. 19. Takenaka A, Kaji I, Kasugai H, Sasaki Y, Ishiguro S, Wada A, Horai T, Otani T, Ishikawa H. Usefulness of diagnostic criteria for aspiration cytology of hepatocellular carcinoma. Acta Cytol 1999;43:610 616. 20. Noguchi S, Yamamoto R, Tatsuta M, Kasugai H, Okuda S, Wada A, Tamura H. Cell features and patterns in ne-needle aspirates of hepatocellular carcinoma. Cancer 1986;58:321328. 21. De Boer WB, Segal A, Frost FA, Sterret GF. Cytodiagnosis of well differentiated hepatocellular carcinoma. Can indeterminate diagnosis be reduced? Cancer (Cancer Cytopathol) 1999;87:270 277. 22. Pisharodi LR, Lavoie R, Bedrossian CWM. Differential diagnostic dilemmas in malignant ne-needle aspirates of liver: a practical approach to nal diagnosis. Diagn Cytopathol 1995;12:364 271. 23. Klinkhamer PJJM, Vooijs GP, de Haan AFJ. Intraobserver and interobserver variability in the diagnosis of epithelial abnormalities in cervical smears. Acta Cytol 1988;32:794 800. 24. Renshaw AA, Friedman MM, Rahemtulla A, Granter SR, Dean BR, Cronin JA, Jiroutek M, Cibas ES. Accuracy and reproducibility of estimating the adequacy of the squamous component of cervicovaginal smears. Am J Clin Pathol 1999;111:38 42. 25. Wassertheil-Smoller S. Biostatistics and epidemiology. 2nd ed. New York: Springer-Verlag; 1995. 26. Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics 1977;33:159 174. 27. Wee A, Nilsson B, Yap I. Fine needle aspiration biopsy of small/ intermediate cell tumors in the liver. Considerations in a Southeast Asian population. Acta Cytol 1996;40:937947.

Diagnostic Cytopathology, Vol 25, No 6

375