Poorly differentiated thyroid carcinoma (PDTC) is an uncommon and aggressive malignancy. Despite the significant clinical implications of a diagnosis of PDTC, its cytomorphologic features have not been well defined. Statistical analysis was applied to a series of 40 PDTCs to identify a specific set of cytomorphologic features that characterized these tumors on fine-needle aspiration biopsy (FNAB).
In total, 40 thyroid FNABs that were highly diagnosed histologically as PDTC (19 insular carcinomas and 21 noninsular carcinomas) comprised the study group. A control group of 40 well differentiated thyroid neoplasms were selected for comparison. All FNABs were reviewed and scored for a series of 32 cytomorphologic features. The results were evaluated using univariate and stepwise logistic regression (SLR) analyses.
In univariate analysis, 17 cytomorphologic features were identified that characterized the 40 PDTCs: insular, solid, or trabecular cytoarchitecture (P < .001); high cellularity (P = .007); necrosis (P = .025) or background debris (P = .025); plasmacytoid appearance (P = .0007); single cells (P < .0001); high nuclear/cytoplasmic ratio (P < .0001); scant cytoplasm (P = .03); nuclear atypia (P < .0001), including nuclear pleomorphism (P = .0052) and anisokaryosis (P < .0001); granular/coarse chromatin (P = .026); naked nuclei (P = .01); mitotic activity (P = .0001) and apoptosis (P < .0001); endothelial wrapping (P = .0053); and severe crowding (P < .0001). In logistic regression analysis, severe crowding (P = .0008) and cytoarchitecture (P < .0001) were identified as the most significant cytomorphologic features of PDTCs, and the combination of cytoarchitecture, severe crowding, single cells, and high nuclear/cytoplasmic ratio was the most predictive of PDTC.
Poorly differentiated thyroid carcinoma (PDTC) is a rare and clinically aggressive thyroid cancer of follicular cell origin. According to the World Health Organization (WHO) definition,1 PDTC consists of 2 histomorphologic subtypes: insular and noninsular. The classic histologic form of PDTC is insular carcinoma, which is defined by its “cellular nests” or insular groups; however, noninsular forms with trabecular or solid patterns are recognized and can coexist. “Insular carcinoma” was described originally by Carcangiu et al,2 who reinterpreted the original observation made by Langhans in 1907,3 who described a locally aggressive tumor with a peculiar architecture made by tumor cells arranged in large, round-to-oval formations, the so-called “insulae.”3 The frequency of PDTC varies from country to country, but, overall it represents approximately 4% to 7% of all thyroid cancers.1, 4 PDTC can arise as a de novo entity5 or can develop from a follicular, Hurthle cell, or papillary thyroid carcinoma.1
The importance of recognizing PDTCs lies in their aggressive clinical behavior and their inherently worse prognosis relative to well differentiated thyroid carcinomas (WDTCs).4, 6, 7 Even when PDTCs are present as a minor component of an otherwise well differentiated tumor, there is an increased likelihood that the tumor will behave aggressively.8 With a mean 5-year survival rate of approximately 50%, patients with PDTC follow a clinical course that is intermediate between that of WDTCs and undifferentiated (anaplastic) thyroid carcinoma.1, 9 PDTCs often present at an advanced stage, have a propensity for local recurrence, and tend to metastasize to regional lymph nodes, lung, and bones.1, 6, 9
Over the past 2 decades, fine-needle aspiration (FNA) biopsy (FNAB) has emerged as 1 of the most important tests for the initial evaluation of thyroid nodules and for guiding their clinical and surgical management. Because of the significant clinical impact of a diagnosis of PDTC, it would be advantageous to recognize this rare subset of thyroid cancers preoperatively in FNABs. Since the original cytologic description of 6 cases by Pietribiasi at al,10 only a few small series11-13 and case reports14-25 have been published. In an effort to better define the cytomorphologic features of PDTC, we applied statistical analysis to a multi-institutional series of 40 histologically proven cases. To our knowledge, this is the largest FNAB series of PDTCs studied.
MATERIALS AND METHODS
Forty-eight cases of surgically resected PDTC with a corresponding cytology were identified from our institutions' archival files (Geneva University Hospital, Massachusetts General Hospital, Brigham and Women's Hospital, Virginia Commonwealth University, University of Pennsylvania Medical Center, and Unilabs-Cypa Lausanne). The histologic diagnosis of PDTC was made according to the histologic criteria defined by the WHO.1 To focus on the specific features of PDTC rather than those of mixed cancers, only cases in which >90% of the tumor displayed PDTC morphology were included in our series. Three cases were excluded because a well differentiated carcinoma component (papillary carcinoma, N = 2; follicular carcinoma, N = 1) comprised >10% of the tumor, and 1 case was excluded because a significant undifferentiated thyroid carcinoma component was present. Four additional FNA cases that contained insufficient material for a cytologic diagnosis because of hypocellularity also were excluded from the study. The remaining 40 FNAB cases constitute the basis of this study. The cytologic cases for review were comprised of direct alcohol-fixed, Papanicolaou-stained smears (N = 31); a combination of alcohol-fixed, Papanicolaou-stained smears and thin-layer preparations (N = 5); and air-dried, Giemsa-stained smears (N = 4). For cytomorphologic comparison, a control group of 40 thyroid FNAB cases that consisted of 15 papillary carcinomas and 25 follicular neoplasms (follicular adenoma, N = 4; follicular carcinoma, N = 21) were selected from the archival files of Massachusetts General Hospital. The pathology reports for each patient were reviewed for pertinent clinical information, including age, sex, and tumor size.
All FNAB study group cases of PDTC were reviewed retrospectively by 2 cytopathologists (M.B. and W.C.F.) who were blinded to the precise cytologic and histologic diagnoses. Each specimen was evaluated semiquantitatively for the presence of 32 cytomorphologic features: cytoarchitectural pattern, cellularity, necrosis, background debris, papillae, follicles, colloid, cell size, cell shape, cellular cohesion, single cells, nuclear/cytoplasmic (N/C) ratio, cytoplasmic characteristics (quantity, quality, color), cellular borders, nuclear atypia, nuclear pleomorphism, anisokaryosis, nuclear shape, chromatin pattern, nuclear grooves and intranuclear pseudoinclusions, stripped nuclei, micronucleoli and macronucleoli, mitoses, atypical mitotic figures, apoptosis, endothelial wrapping of the tumor cell groups, transgressing vessels, and severe crowding (Table 1).
Table 1. Cytomorphologic Features Evaluated in 40 Thyroid Fine-Needle Aspiration Biopsies of Poorly Differentiated Thyroid Carcinoma
Macro-micro-mixed-monolayer vs. insular-solid-trabecular
Follicles (any type)
For the architectural assessment, specimens were divided into 2 groups according to the predominant cytoarchitecture: 1) tumors that exhibited a follicular, papillary, or monolayer pattern versus 2) tumors with either an insular, trabecular, or solid pattern. A follicular pattern included tumors with a macrofollicular architecture (cohesive, 2-dimensional sheets of follicular cells in a background of colloid) or microfollicular architecture (small groups, usually with 6 to 12 follicular cells surrounding a central lumen). An insular cytoarchitectural pattern was defined by crowded groups of follicular cells with peripheral endothelial wrapping lacking a lumen or colloid.
Transgressing vessels were defined as blood vessels that traversed groups of follicular cells and that were detectable at an intermediate magnification (×10-20 objective).26 Severe crowding was defined as 3-dimensional groups of cohesive follicular cells with a morular-like pattern lacking defined cell borders and without a recognizable follicular or papillary pattern.
Results of immunocytochemical analysis performed prospectively in the initial cytologic evaluation of a subset of the PDTC cases were recorded. Immunocytochemistry was performed manually using a standard avidin-biotin complex technique.27 Immunocytochemical staining for thyroglobulin was used in 9 of 40 cases. Other immunocytochemical stains included calcitonin, carcinoembryonic antigen (CEA), chromogranin, synaptophysin, thyroid transcription factor 1 (TTF-1), cytokeratin cocktail, and S-100. Immunocytochemistry was performed on destained slides (5 cases), Thin Prep slides (3 cases), and on a cell block (1 case).
Univariate analysis was used to evaluate each of the 32 cytomorphologic features for statistical significance in predicting PDTC. Chi-square tests and Fisher exact tests were applied. Statistical significance was achieved at P < .05, and only those features with statistical significance (from univariate analysis) were used in stepwise logistic regression analysis (SLR). Specificity, sensitivity, positive predictive value, and negative predictive value also were calculated for statistically significant cytomorphologic features. Because there were no statistically significant differences in the cytomorphologic features of the insular and noninsular subtypes of PDTC, the univariate and multivariate findings are presented as a single combined PDTC group to allow for greater statistical power. The multivariate analysis (SLR) was a probability model that enabled us to select key variables while controlling for other variables within the model. We included the Hosmer and Lemeshow (HL) goodness-of-fit statistic to test the validity of our measurements so that the standard methods of statistical inference could be applied on our dataset. Statistical analysis was performed using SAS version 9.1 (SAS Institute, Cary, NC).
The 40 PDTCs that comprised the study group were obtained from 38 patients, including 16 men and 22 women, with an average age of 62 years (age range, 31-86 years) (Table 2). The average size of the resected carcinomas was 4.2 cm (range, 1.5-8.5 cm). Based on the histomorphologic patterns of the tumors in the thyroid resection specimens, 47.5% of tumors (N = 19) were the insular subtype, and 52.5% of tumors (N = 21) were the noninsular subtype, ie, with a solid or trabecular pattern, as defined by the WHO classification.1
Table 2. Clinicopathologic Features of Patients With Poorly Differentiated Thyroid Carcinoma
Among the FNAB cases of PDTC, 57.5% (N = 23) were diagnosed initially as “malignant,” and 32.5% (N = 13) were recognized prospectively as PDTC. In contrast, 42.5% (N = 17) of the FNAB cases in our series were diagnosed originally as “suspicious for a follicular neoplasm.” The 23 malignant cases consisted of 12 insular type PDTCs and 11 noninsular type PDTCs. The 10 cases that were diagnosed as “malignant” but were not recognized prospectively as PDTC were categorized either as “carcinoma, not otherwise specified” (N = 4) or as “papillary thyroid carcinoma” (N = 6) (Table 2).
Cytologic evaluation of the PDTC cases demonstrated that 92.5% (N = 37) exhibited either an insular, solid, or trabecular cytoarchitectural pattern (Fig. 1A,B,C) in contrast to only 5% (N = 2) of the well differentiated thyroid neoplasm (WDTN) control group (Table 3). Twenty percent of cases had well defined insular groups with endothelial wrapping (Fig. 1D). Aspirates of PDTC were cellular and comprised of a uniform-appearing population of cells at low magnification. Seventy-five percent of cases (N = 30) contained isolated cells alternating with large, solid groups of severely crowded, mitotically active, and apoptotic cells (Fig. 1E,F). The proportion of isolated cells to cohesive groups varied among cases. In the majority of PDTCs, the cells had a high N/C ratio, and the nuclei were round. At higher magnifications, variable degrees of nuclear atypia, including nuclear pleomorphism and anisokaryosis, could be observed in >50% of cases (Fig. 1G, Table 3). Background debris consisting of cytoplasmic and nuclear fragments was present in 15% (N = 6) of the PDTC cases (Fig. 1H). Papanicolaou-stained preparations were the most commonly used in our study; however, among those cases with Romanowsky-stained preparations, cells appeared larger but had cytomorphologic features similar to those observed in Papanicolaou-stained cases.
Table 3. P Values of Statistically Significant Cytomorphologic Features in Poorly Differentiated Thyroid Carcinomas Versus Well Differentiated Thyroid Neoplasms Using Univariate Analysis and P Values From Logistic Regression Analysis
In the univariate analysis, 17 of the 32 cytomorphologic features that were examined demonstrated a statistically significant difference between the PDTC group and the control group of 40 WDTNs: insular, solid, or trabecular cytoarchitecture (P < .001); high cellularity (P = .007); the presence of necrosis (P = .025) or background debris (P = .025); plasmacytoid appearance of cells (P = .0007); isolated cells (P < .0001); high N/C ratio (P < .0001); scant cytoplasm (P = .03); nuclear atypia, (P < .0001) including nuclear pleomorphism (P = .0052) and anisokaryosis (P < .0001); granular/coarse chromatin (P = .026); stripped nuclei (P = .01), mitotic activity (P = .0001) and apoptosis (P < .0001); endothelial wrapping (P = .0053); and severe crowding (P < .0001) (Table 3). Unlike the WDTNs, aspirates of PDTCs lacked either a papillary cytoarchitecture (P = .0007) or a predominant follicular-patterned arrangement of cells (P = .0002). Colloid (P < .0007), nuclear grooves (P = .0008), and pseudoinclusions (P = .0039) also were absent in most cases (Table 3). Eight PDTCs (20%) in our study exhibited a limited (<10%) component of WDTC in the corresponding thyroid resection specimen; however, the cytomorphologic features of these FNAB cases did not differ statistically from other PDTCs in the series that lacked a well differentiated histologic component.
Only features that demonstrated statistical significance in the univariate analysis were included in the multivariate SLR analysis, thus allowing for more statistical power in the SLR model. The SLR analysis was verified by using the HL goodness-of-fit test (P = .92). Therefore, the null hypothesis stating that the data did not fit the model for distinguishing PDTC from WDTN was rejected. The HL statistic is a chi-square–based test to assess the goodness of fit of our probability model. This test is devoted to validating the ‘chi-square’ component of our regression model and describes how effective our model is in describing the outcome variable (PDTC).
In the SLR model, a combination of 4 cytomorphologic features was identified as statistically significant in predicting PDTC: insular, solid, or trabecular cytoarchitectural pattern (P < .0001); single cells (P = .0147); high N/C ratio (P = .0203); and severe crowding (P = .0008) (Table 3). Table 4 describes the validity of using these 4 statistically significant cytomorphologic criteria to successfully predict the outcome (PDTC). Although some cases of PDTC lacked these features, most demonstrated these cytologic criteria. The corresponding P values reflect the relative importance of each feature in the model and suggest that the cytoarchitectural pattern was the single most important feature for correctly diagnosing PDTC. In our series of PDTCs, severe crowding had 100% specificity and positive predictive value for the detection of PDTC; however, the sensitivity was only 70% (Table 4). Other cytomorphologic features with similar high specificities were necrosis, background debris, apoptosis, and endothelial wrapping; however, low sensitivities ranging from 15% to 45% limited their usefulness in the prediction of PDTC. After conducting our analysis with SLR and HL statistics, the logistic regression model appeared to be effective in describing the outcome variable (prediction of PDTC) on the basis of statistical assumptions and on the basis of cytomorphologic interpretation.
Table 4. Four Stepwise Logistic Regression-derived Features Used in Predicting Poorly Differentiated Thyroid Carcinomas
Immunocytochemical evaluation was performed in a small subset of cases in our series (as reported in the original diagnostic evaluation). All cases in that evaluation were immunocytochemically positive for thyroglobulin (N = 9), TTF-1 (N = 2), and cytokeratin cocktail (N = 2). In addition, none of the PDTC cases were positive for calcitonin (N = 8), chromogranin (N = 2), or S-100 (N = 2). CEA was negative in 2 of 3 cases and was focally positive in 1 case.
We identified 4 statistically significant cytomorphologic features that, in combination, distinguished PDTCs from WDTNs on FNAB: an insular, solid, or trabecular cytoarchitectural pattern; single cells; high N/C ratio; and severe crowding. The cytoarchitectural pattern was 1 of the most statistically significant cytomorphologic features for predicting PDTC that was identified in the current study (P < .0001). Intuitively, the latter feature is useful because it correlates with the most important histologic feature observed in resection specimens. The hallmark of PDTC in histopathology that is required, albeit not sufficient, for a diagnosis of PDTC, is the presence of an insular, trabecular, or solid pattern. In contrast to the typical macrofollicular, microfollicular, or papillary cytoarchitecture observed in the majority of WDTNs, PDTCs lacked these patterns. It is well documented that a subset of PDTCs can have occasional microfollicles, but the latter does not represent a predominant architectural component.1, 4 A potential reason that cytoarchitecture was so successful in our evaluation of the cases may be related to the finding that we focused on the predominant architectural pattern and did not include minor patterns in the analysis. This is analogous to the approach used to diagnose most WDTNs in FNAB specimens. Further studies to evaluate the reproducibility of these cytomorphologic features will be useful to confirm their prospective application to the diagnosis of PDTCs in FNAB samples.
Aspirates of PDTC traditionally are difficult to diagnose because they are rare, their cytomorphologic features overlap with those of follicular neoplasms, and their characteristic FNA features have not been well documented. For these reasons, many PDTCs are diagnosed as “suspicious for a follicular neoplasm” in FNAB.28 This was supported in our series by the finding that approximately 43% of cases originally were interpreted as follicular neoplasms.
Although the histologic criteria for diagnosing PDTCs have been defined by the WHO1 and were modified more recently in the Turin proposal,29 the cytologic features of this uncommon carcinoma have received much less attention. One of the first cytologic descriptions of PDTCs was a report of 6 cases by Pietribiasi et al in 199010 followed by 4 cases reported in 1992 by Sironi et al,11 6 cases reported in 1999 by Guiter et al,12 and 5 cases reported in 2001 by Nguyen and Akin.13 Each of those small FNAB series identified a group of several cytologic features that included high cellularity, bloody background with scant colloid, cellular clusters (trabecular or solid), and isolated cells with scant cytoplasm and mild atypia. Occasional cases with mitoses were observed, but necrosis was rare among all of the series and was identified in only 4.7% of cases. In our series of PDTCs, many of the same cytologic features were observed, and necrosis also was infrequent (observed in only 15% of cases). However, nuclear atypia (55% of cases), nuclear pleomorphism (40% of cases), anisokaryosis (50% of cases), and mitoses (42.5% of cases) were frequent in our series; whereas features like papillae, follicles, nuclear grooves, and inclusions were observed in <10% of cases. The differences between certain cytologic features observed in our study and those observed in several previous series may be explained by the finding that we included only those PDTCs in which WDTC represented <10% of the tumor, whereas many other series included cases of mixed PDTC and WDTC. In support of this explanation and in contrast to our series of PDTCs, cytologic features often associated with WDTNs, such us microfollicles, papillary cytoarchitecture, intranuclear pseudoinclusions, and longitudinal nuclear grooves, were common in almost all previous series except the 1 by Nguyen and Akin.13 It is noteworthy that the series reported by Nguyen and Akin, like ours, also consisted of relatively “pure” PDTCs.
According to the authors of the WHO volume on Pathology and Genetics of Tumours of Endocrine Organs, the definitive diagnosis of a poorly differentiated carcinoma can be made only at the histologic level.1 However, our current results indicate that a combination of cytomorphologic features defined by SLR analysis is highly predictive of PDTC in FNAB specimens. Nonetheless, correlation with clinical and ultrasonographic findings also may be helpful: PDTCs usually are large (>4.0 cm), widely invasive, heterogeneous tumors with necrosis and extrathyroidal extension.6, 9 In fact, in our FNAB series, the average size of PDTCs in resection specimens was >4.0 cm.
In the literature, few FNAB cases have been recognized prospectively as PDTC. Among the latter are cases described by Oertel and Miyahara-Felipe24 and by Gong and Krishnamurthy23 in which an FNAB diagnosis of insular carcinoma was suggested. In our study, only 32.5% of cases were recognized originally as PDTC by FNAB, and the other cases were diagnosed as “carcinoma, not otherwise specified,”; “papillary carcinoma”; and “suspicious for a follicular neoplasm.” The potential to diagnose even a subset of PDTCs by FNAB has important consequences for the clinical management of patients. It is recognized that a definitive diagnosis of PDTC may not be possible; however, in many cases, the cytomorphologic features together with clinical findings will be sufficient to suggest PDTC. In contrast to patients who have an FNAB diagnosed as “suspicious for a follicular neoplasm,” for which only a lobectomy is performed, patients who are diagnosed prospectively with PDTC potentially may be treated more aggressively with a total thyroidectomy and directed neck dissection.7
Our series of PDTC FNABs focused on cases in which <10% of the carcinoma was well differentiated or undifferentiated to more accurately define the characteristic features of “pure” PDTCs. Although PDTCs can arise de novo,5 it is believed that many are derived from WDTC.1 Therefore, it is recognized that, in fact, many PDTCs will be comprised of a combination of WDTC and PDTC, thus potentially confounding the ability to diagnose tumors with a minor PDTC component by FNAB. Because our study did not address this issue, future investigations of this “mixed” WDTC and PDTC subset would be useful. This finding also suggests that, although approximately 33% of our PDTCs were recognized prospectively, this proportion would be smaller if the “mixed” subset were included. In our initial selection of PDTC cases, approximately 10% that were identified from our archival files were excluded, because >10% of the tumor was either WDTC or undifferentiated carcinoma.
Although the control group in our FNAB study consisted of WDTNs, which are the primary tumors in the differential diagnosis of PDTCs, several other pathologic entities also may be considered in the cytologic differential diagnosis. Two cytomorphologic features that, in our series, characterized a subset of PDTCs—a predominant, single-cell pattern and plasmacytoid cell morphology—raise the possibility of a medullary carcinoma. In contrast to medullary thyroid carcinoma, most PDTCs lack the “salt-and-pepper” neuroendocrine-type chromatin pattern, are strongly immunoreactive for thyroglobulin, and are negative for calcitonin and other neuroendocrine markers, such as chromogranin and synaptophysin. In 9 of our cases in which immunocytochemical studies were performed and in all reported FNAB cases, PDTCs were positive for thyroglobulin, whereas neuroendocrine markers were negative.10, 13, 17, 19 Occasionally, the single-cell pattern and plasmacytoid cell morphology can mimic a lymphoproliferative disorder, but PDTCs are negative for the immunomarker CD45 and for markers of B cells (eg, CD19, CD20) and plasma cells (eg, CD138). Other features of PDTCs, such as nuclear atypia, absent colloid, and background necrosis, can suggest either a metastasis from an extrathyroidal primary tumor or undifferentiated thyroid carcinoma. The positive immunoreactivity of PDTCs for thyroglobulin and TTF-1 helps to exclude a metastasis, whereas the lack of marked nuclear pleomorphism, the lack of high-grade atypia, and the absence of sarcomatoid features distinguishes PDTCs from undifferentiated carcinoma.
Currently, there are no specific ancillary markers available to reliably distinguish PDTCs from WDTCs; therefore, the cytologic and histologic diagnoses depend primarily on microscopic evaluation. Nonetheless, there is a limited role for ancillary studies, particularly immunocytochemical stains for thyroglobulin and calcitonin, in the diagnosis of a subset of these tumors when the differential diagnosis includes medullary thyroid carcinoma or metastatic disease.30 In our series, only 23% of cases were evaluated prospectively using immunocytochemical stains. Both TP53 and ras point mutations have been identified in a subset of PDTCs, but further studies are needed to evaluate the potential application of these as molecular markers in thyroid FNAB specimens.30
In conclusion, we evaluated 32 cytomorphologic features in 1 of the largest FNAB series of histologically proven PDTC. Applying univariate and SLR analyses, we demonstrated that 4 features used in combination were highly predictive of PDTC: an insular, solid, or trabecular cytoarchitectural pattern; single cells; high N/C ratio, and severe crowding.