Reproducibility of the WHO classification of thymomas: practical implications

Abstract

Background

The WHO-classification was shown to be an independent prognostic marker in some but not all retrospective studies possibly due to lack of reproducibility. We investigated the reproducibility of the WHO-classification and its prognostic implication using a large series of resected thymomas.

Methods

Four independent pathologists histologically classified a surgical series of 129 thymic tumors in a blinded fashion. Fleiss’ kappa-coefficient was used to assess the pathologists’ overall agreement, and Cohen-Kappa to assess the agreement between two observers. Disease-related-survival (DRS) and progression-free-survival (PFS) curves were generated by Kaplan-Meier method and compared by log-rank test.

Results

In 63/129 (48.8%) cases there was a complete agreement; in 43/129 (33.3%) cases 3/4 pathological diagnoses were identical; in 15/129 (11.6%) cases the diagnoses were identical by pair; in 8/129 (6.2%) cases three different pathological diagnoses were on record. The Kappa-correlation coefficient was only moderate (0.53). A following web review carried out on the 23 cases with at least two different diagnoses reached a complete consensus. The histotype showed a statistically significant impact on PFS and DRS in the classification provided by only two pathologists.

Conclusions

In this study, the agreement on WHO classification of thymomas was only moderate and this impacted on patients management. Web consensus conference on the diagnosis, more stringent diagnostic criteria or the adoption of referral diagnostic centres may substantially reduce discrepancies.

INTRODUCTION

The histological classification of thymomas underwent several revisions over many years. The first widely accepted proposal was that by Bernatz and colleagues from the Mayo Clinic in 1961 [4]. According to the proportion of epithelial cells and lymphocytes, they subdivided thymomas into four histotypes: lymphocyte-rich, epithelial-rich, mixed lympho-epithelial, and spindle cell. Since then, other schemes have been proposed but none achieved wide acceptance [5–7].

In 1999 the World Health Organization (WHO) reached the first consensus on thymic epithelial tumors’ classification [8]. The latter, updated in 2004 [3], recognizes two main categories of thymoma. Depending on whether the neoplastic cells and their nuclei have a spindle/oval or round/polygonal appearance they can be subdivided, respectively, into type A and type B thymomas. The former has uniformly bland nuclei; the latter presents several degrees of nuclear atypia and number of lymphocytes, and they are classified into three subcategories: B1, B2 and B3. Thymomas with a mixture of type A and B1-like (rarely B2-like) features are defined as AB thymomas. A third category of thymic epithelial tumors, the type C thymoma or thymic carcinoma, was also introduced to label overtly malignant tumors (i.e. those with cytological atypia, invasiveness and lack of organoid aspects). The latest WHO version also acknowledged the existence of combined thymomas, i.e. the presence of combined subtypes of thymomas, and that of rare histotypes (metaplastic, micronodular, sclerosing, microscopic, and anaplastic thymoma).

The WHO pathological classification of thymomas aimed to provide a potentially useful tool to classify patients also in terms of prognosis. However while some retrospective studies confirmed the prognostic significance of the WHO classification [9–11] others emphasized its limited clinical usefulness,- due to lack of reproducibility [12–14].

In the present study, we investigated the reproducibility of the WHO classification when performed by different pathologists at two different Institutions. Our goals were to identify the level of reproducibility of this scheme and to evaluate how it impacted on prognosis.

METHODS

Case selection

Clinical data and formalin fixed paraffin embedded samples were collected from a series of consecutive patients who underwent surgery for thymic epithelial tumors, either for a primary tumor or a local relapse, at a single institution, the Humanitas Cancer Center (Rozzano-Milan, Italy), in the period from 1996 to 2008. This series has been described previously in detail [15]. For the relapsed tumors, the date of the first surgery was taken into consideration, extending the observation period from 1976 to 2008. Case selection was based on a diagnosis of thymoma or thymic carcinoma and the availability of tumor tissue (paraffin blocks and slides) and clinical history.

Pathological diagnosis

The original diagnosis of the series under study (Table 1) had been performed by one of a group of 10 different surgical pathologists (as a whole referred as CLAS1), none of whom with a specific interest in thymomas. Complete clinical data were available at the time of original diagnosis. In detail the pathologist who was in charge of the case could evaluate laboratory, imaging and clinical features of the patient via a computer assisted device. The original pathological report was also based on the evaluation of gross features and several H/E sections (range 2–20, mean 12) of the tumor. Immunohistochemical stainings were employed only rarely and mainly for the suspicion of a thymic carcinoma. Common was the attitude to share the case, and thus the histological report, among different pathologists. A pathologist belonging to the same Institution (CLAS2), blinded to clinical data, gross pathological features and original diagnosis, reviewed all the original H/E slides of each case. Two representative H/E slides and the corresponding blocks were then selected and sent to the National Institute of Health (collaborating institution) and reviewed by two additional pathologists (CLAS3, CLAS4) blinded to clinical, gross features and original diagnosis.

Table 1.

Patient characteristics (n=129)

Sex	Male / female	65 / 64
Median age (range), years		56 (20–86)
Primary surgery		108 (81.8%)
Surgery at relapse		21 (18.2%)
Paraneoplastic Syndromes	Myasthenia Gravis	31 (24.0%)
	Other Syndromes	2 (1.6%)
Histotype*	A	21 (16.3%)
	AB	28 (21.7%)
	B1	22 (17.1%)
	B2	20 (15.5%)
	B3	27 (20.9%)
	C	11 (8.5%)
Stage	I	35 (27.1%)
	IIA	25 (19.4%)
	IIB	17 (13.2%)
	IIIA	15 (11.6%)
	IIIB	3 (2.3%)
	IVA	5 (3.9%)
	IVB	11 (8.6%)
	Na**	18 (13.9%)
Type of Resection	R0	78 (60.5%)
	R1	22 (17.0%)
	R2	10 (7.7%)
	Na***	19 (14.8%)

Na=not assessable.

^*Original diagnosis;

^**For 18 patients the stage at the diagnosis was not assessable.

^***For 19 patients there was insufficient information to assess the completeness of resection of the first surgery.

Consensus review

H&E slides of cases with more than 2 discordant diagnoses were acquired as digital images using ScanScope AT turbo (Aperio, Vista, CA) in order to obtain pictures with a magnification up-to 40×. Image files were visualized using ImageScope (Aperio) and were visualized, navigated and discussed in a web conference held between NIH (Bethesda, MD), Humanitas IRCC (Rozzano, Milan, Italy) and Seoul National University Bundang Hospital (Seongnam-si, Gyeonggi, Republic of Korea), hosted using Adobe Connect 9 (Adobe, San Jose, CA).

This study has been conducted in agreement with the Declaration of Helsinki and was approved by the institutional ethical review boards (ClinicalTrials.gov ID: NCT00965627).

Statistical analysis

The global agreement was assessed by Fleiss’ kappa coefficient while the Cohen’s Kappa was used to assess agreement between two pathologists (low agreement: <0.41; moderate: ! 0.41<0.61; substantial: ! 0.61<0.80; almost perfect: ! 0.80) [16]. For each histotype, a concordance index has been calculated dividing the number of patients which have been concordantly classified by two pathologists by the total patient population number. Disease Related Survival (DRS) was calculated from diagnosis until disease related death, considering as censored patients who died of other causes. Progression-Free Survival (PFS) was calculated from diagnosis until progression or death, whichever occurred first. Patients without event were censored on the date of last visit for PFS and of last contact for DRS. Survival curves were generated according to the Kaplan-Meier method and groups compared using the log-rank test. Hazard ratio (HR) and corresponding 95% confidence intervals (CI) were estimated using Cox proportional hazard model.

RESULTS

Patient characteristics and clinical data

The original histological diagnosis produced at the time of surgery, paraffin blocks and slides, as well as clinical history were available in 129 cases.

The main patient and tumour characteristics are summarized in Table 1. Most of the samples were collected at the primary surgery. The median age was 56 years and the male:female ratio was of 1:1. Most of the patients had an early stage of the disease (59.7% stage I and II) and underwent a radical resection (60.5% R0). Survival was assessable in all patients, and median follow-up was 84.6 months (range 0–384 months). The median progression free survival (PFS) was 94.3 months (range 0–356 months) with a 5- and 10-year PFS of 68.1% and 49.1%, respectively. The median disease related survival (DRS) was not reached, whereas the 10-year DRS was 82.3%.

Pathological features

Taking into consideration the original diagnosis, 22/129 (17%) thymomas showed combined aspects. To facilitate the statistical analysis, and since the original report did not give any percentage of different histotypes, combined histologies were re-classified according to the worst histotype (for example B2-B3, were considered B3 thymomas). Given this, the original diagnosis of the thymomas under study is that reported in Table 1. Briefly, type A were diagnosed in 21/129 (16.3%); type AB in 28/129 (21.7%); type B1 in 22/129 (17.1%); type B2 in 20/129 (15.5%); type B3 in 27/129 (20.9%) and type C in 11/129 (8.5%) cases. Paraneoplastic syndromes were observed in 33/129 (26%) cases: in 17 type B3, 5 type B2, 5 type B1, 5 type AB, and 1 type A. There were no thymic carcinomas associated with paraneoplastic syndromes.

In 63/129 (48.8%) cases there was a complete agreement among the four observers. In 43/129 (33.3%) cases three out of four pathological diagnoses were identical; in 15/129 (11.6%) cases two out of four pathological diagnoses were identical; in 8/129 (6.2%) cases three different pathological diagnoses were reported; complete disagreement (four different diagnoses) was never observed. Cases with at least two different diagnoses (n=23) are reported in Table 2. Briefly in most of the cases (9/23, 39%) the diagnostic difficulty rested in the proper recognition of type B thymomas; in this setting overlap between B1 and B2 was the most frequent occurrence (6/9). In 6/23 (26%) cases the diagnostic discrepancy involved type AB thymomas; the different diagnoses taken into consideration were mostly B1 or B2 thymomas (4/6) but in two cases a B3 thymoma was also reported. In 7/23 (30%) cases the discrepancy involved fairly different histological subtypes: in four cases thymic carcinoma vs. type B3; in two cases type B3 vs. type A; and in one case thymic carcinoma vs. type A.

Table 2.

	CLAS1	CLAS2	CLAS3	CLAS4	Consensus results	Stage diagnosis	Myastenya gravis	Resection	Change therapy
Pt n° 04	B1	B1	B2	B2	B2	IIA	no	R1	no
Pt n° 12	B2	B1	B1	B2	B2	IIB	no	R0	no
Pt n° 20	B3	B3	AB	AB	B3	III	yes	R1	no
Pt n° 25	B3	B3	B2	B2	B2	III	no	R0	no
Pt n° 28	A	TC	TC	B3	TC	I	no	R0	yes
Pt n° 56	B2	B2	B3	B3	B3	IIA	yes	R0	no
Pt n° 61	A	B3	A	B3	A	III	no	R2	no
Pt n° 67	AB	AB	B1	B2	B1	uk	yes	uk	no
Pt n° 68	TC	B3	B3	TC	TC	III	no	R2	no
Pt n° 71	B3	B3	A	A	A	III	no	R0	yes
Pt n° 75	A	TC	TC	A	TC	I	no	R0	yes
Pt n° 76	A	AB	B3	B3	B3	IIA	no	R1	yes
Pt n° 78	B2	B2	B3	B3	B3	IIB	yes	R0	no
Pt n° 87	AB	B1	B2	AB	AB	I	no	R0	no
Pt n° 89	B1	B1	B2	B2	B2	IIA	no	R1	no
Pt n° 90	B1	B1	B2	B2	B2	IIA	no	R0	no
Pt n° 91	B3	B3	TC	AB	B3	I	no	R0	no
Pt n° 102	AB	B1	B1	B2	AB	I	no	R1	no
Pt n° 108	B1	B1	B2	B2	B2	I	yes	R0	no
Pt n° 114	B1	B2	B1	B2	B1	IIA	uk	R1	no
Pt n° 122	B2	B1	AB	B2	AB	IIA	no	R0	no
Pt n° 127	A	A	Cystic	Cystic	Cystic	I	no	R0	no
Pt n° 128	B3	B3	TC	A	B3	uk	yes	uk	no

Inter-observer agreement

Table 3 summarizes histotype frequencies according to the 4 observers. Taking into account all the WHO histotypes and all the pathologists at the same time, the Kappa correlation coefficient was 0.53 (moderate strength). Table 4 reports the coefficient of agreement for all comparison pairs that ranged from moderate to almost perfect. The best agreement was observed between CLAS2 and CLAS1 (K coefficient: 0.84). A substantial agreement was also observed between CLAS3 and CLAS2 (K coefficient: 0.70) and CLAS3 and CLAS1 (K coefficient: 0.64). A moderate strength of agreement was observed between CLAS4 and each of the other three observers, with a Kappa correlation coefficient ranging from 0.52 to 0.53.

Table 3.

The frequencies of WHO thymic histotypes reported by the four pathologists.

WHO histotypes
	A	AB	B1	B2	B3	C
CLAS1	16.4%	21.1%	17.2%	15.6%	21.1%	8.6%
CLAS2	11.6%	20.2%	19.4%	13.9%	25.6%	9.3%
CLAS3	12.7%	19.8%	17.5%	15.9%	19.0%	15.1%
CLAS4	17.6%	19.2%	3.2%	33.6%	18.4%	8.0%

CLAS: pathologist

Table 4.

Kappa correlation between pathologist pairS considering the WHO histotypes.

Pathologist consensus review and therapeutic implications

A review carried out on the 23 cases with at least two different diagnoses reached a complete agreement (table 2). Based on NCCN guidelines version 2.2013, the therapeutic strategy would have been changed in 7 out of 23 patients (30%) due to pathological discrepancies. In particular, a misdiagnosed thymic carcinoma, falsely considered as a thymoma or vice versa, implicates that four patients (pt n° 28; pt n° 75; pt n° 91; pt n° 128) could have been treated in ways different from recommendations in the adjuvant setting, whereas the other three patients (pt n° 61; pt n° 71; pt n° 76) could have received or not an adequate adjuvant treatment (radiotherapy without or instead of chemotherapy), depending on the rendered diagnosis. After consensus, the therapeutic strategy could have significantly changed in 4 patients compared to the initial diagnosis (CLAS1). In particular, in three cases (pt n° 28, pt n° 75, pt n° 76) an adjuvant treatment could have been undertaken, whereas in one case (pt n° 71) it should not have been undertaken (Table 2).

Prognostic implications for PFS and DRS

Considering all the histotypes and stratifying patients for the single WHO histotypes, only 2 pathologist interpretations were prognostic for PFS (CLAS1 p=0.001; CLAS4 p<0.001) and DRS (CLAS1 p=0.039; CLAS4 p=0.027) (Tables 5 and 6). Analysing only the thymoma population, we did not observe a statistically significant difference in terms of PFS and DRS among the four pathologists.

Table 5.

The 5-year PFS in the WHO thymic histotypes.

Progression-Free Survival at 5 years
WHO histotypes
	A	AB	B1	B2	B3	C	P value**
CLAS1	77.5%	87.7%	66.8%	70%	59.8%	34.3%	0.001
CLAS2	77.5%	89.1%	74.6%	63.2%	61.2%	40.9%	0.065
CLAS3	72.7%	88.4%	60.3%	73.4%	67.3%	45.8%	0.066
CLAS4	84.8%	89.1%	100%	61.4%	60.2%	26.7%	<0.001
P value*	0.643	0.893	0.565	0.874	0.756	0.225

^*p value for the single WHO histotype;

^**p value for every single pathologist.

CLAS: pathologist

Table 6.

The 10-year DRS in the WHO thymic histotypes.

Disease Related Survival at 10 years
WHO histotypes
	A	AB	B1	B2	B3	C	P value**
CLAS1	100%	87.5%	94.5%	70.2%	77%	57.9%	0.039
CLAS2	100%	100%	87.3%	77.9%	75.2%	63.6%	0.107
CLAS3	93.3%	100%	90.9%	62.3%	86.1%	72.9%	0.201
CLAS4	100%	90.9%	100%	79.3%	80.7%	61.0%	0.027
P value*	0.300	0.562	0.862	0.927	0.821	0.839

^*p value for the single WHO histotype;

^**p value for all WHO histotypes.

CLAS: pathologist

DISCUSSION

Thymomas are rare tumors that cover a wide spectrum of histopathological aspects. This heterogeneity has been acknowledged by the WHO classification which identifies six main histotypes and includes also combined forms. Few studies reported data on the reproducibility of this scheme [12, 21–23]. In a large series of 200 thymomas [21], a properly trained but not expert pathologist achieved >90% concordance with the final diagnosis (reached by himself plus two expert pathologists, during a consensus meeting). Similar findings were reported by Park et al [22] and Rieker et al [12]. The former recorded a 94% concordance in a series of 150 cases of thymomas referred to a single expert pathologist by several different institutions. The latter [12] in a series of 218 cases reviewed by two different pathologists reported a K=0.87. Despite these encouraging results, the WHO scheme showed some limits when applied to larger groups of pathologists. Indeed Verghese et al [23] observed that the reproducibility falls to K=0.45 if the scheme is applied among several pathologists with robust expertise in thoracic pathology (n=17). The WHO classification may have some areas that are more difficult to classify than others. Rieker et al [12] in their large study observed that the reproducibility in the subset of type B thymomas was only moderate (K=0.49).

In the present study, we investigated the reproducibility of the WHO scheme on a series of 129 thymomas among four different observers and the overall agreement was moderate (K=0.53). The highest agreement (almost perfect) was recorded among two independent observers that belonged to the same institution (CLAS1 and CLAS2, K=0.84). The most probable explanation to this finding is that CLAS2 observer, who was blinded to clinical data, gross pathological features and final diagnosis, could evaluate all the H&E sections of each case (and immunohistochemistry if available). Substantial agreement was also recorded among the third independent observer (CLAS3) and the two mentioned above. In particular, the agreement was lower with CLAS1 (K=0.64) and higher with CLAS2 (K=0.70). A possible explanation to this finding rests on the fact that CLAS3 observer, blinded to clinical and pathological data, could evaluate only two H&E slides, selected by CLAS2. Data herein recorded as CLAS1 seem in keeping with the above mentioned hypothesis. CLAS1 data refers to a group of different pathologists, none with specific interest in thymoma, that, at the moment of the diagnosis, could evaluate the complete set of H/E slides as well as clinical information. Interestingly, CLAS1 histotypes distribution strictly adhered to that of the WHO scheme and predicted clinical outcome.

In 23/129 (17.8%) cases of the present series, two or three different diagnoses were rendered on the same case. Based on these pathological discrepancies, the therapeutic strategy would have been changed in at least 7 out of 23 patients (30%). In particular, a misdiagnosed thymic carcinoma, implicates that several patients could receive or not an adequate adjuvant treatment, depending on the pathologist considered. When we focused on this setting we noticed that these “grey areas” of the classification were represented by one of the following: 1) sub-classification of type B thymomas (9/23, 39%); 2) recognition of type AB thymomas (6/23, 26%); 3) distinction among type A, type B3 and type C thymomas (7/23, 30%). The first point, especially the distinction between B1 and B2, represents the great majority of the discrepancies observed in the present study. A moderate agreement in type B thymomas has also been observed by Rieker et al [12] and Verghese et al [23]. B1 and B2 thymomas have to be differentiated according to the number of epithelial cells and their degree of atypia and their distinction between is relevant for the clinician ctaking into consideration more aggressive behaviour reported for B2. However, stringent diagnostic criteria, i.e. percentage of epithelial cells or an absolute number as well as definition of the proper degree of atypia, are lacking; hence the urge to have specific tissue biomarkers able to properly label cells belonging to each category. Diagnostic discrepancy on type AB thymomas were observed 6/23 cases (26%) of the present series and mainly involved B1 or B2 thymomas. This figure is in keeping with data from Verghese et al [23]. In their series at least 5 cases (among 22 AB) were classified either as B thymomas by the majority of pathologists probably due to the underestimation of short bundles of fibrous like cells which are indeed epithelial. To avoid this misclassification the use of an epithelial marker for the correct definition of the bundles seems mandatory. In addition, rather than a more precise definition of the so called B1 like and B2 like features, specific markers able to discriminate the B-like from the B-true zone can be useful. The last critical area highlighted by the present study deals with the differential diagnosis involving thymic carcinoma. Not too surprisingly in most of the cases this rested with B3 thymomas. Three cases of thymic carcinoma/type B3 thymoma vs. type A thymoma were also recorded. Obviously, the clinical management of these patients could be very different and the use of immunohistochemical markers such as CD5 and c-kit, as tools to discriminate, thymomas and thymic carcinoma reflects the diagnostic difficulties of this area.

The updated WHO classification has been shown to be prognostic [11]. A Chinese study of 200 cases of thymoma observed that types A, AB, B1, B2, and B3 display an increasingly worse prognosis [17]. However, this correlation was not confirmed by other studies [12,18]. For example, Rieker and colleagues [12] observed that type AB and B1 tumors showed the most favourable outcome, whereas types A and B2 behaved much worse and essentially showed overlapping survival curves. These discrepancies could be explained by the inconsistencies in the accurate histologic classification of the tumors among the different groups of investigators. This was also confirmed in the present series were only two sets of diagnoses correlated with PFS and DRS.

Several proposals have been presented to increase inter-observer reproducibility. Suster and Moran [19] suggested to simplify the classification discriminating between well-differentiated tumors, corresponding to conventional thymoma, poorly differentiated neoplasms, true thymic carcinomas, and tumors showing intermediate features of differentiation were designated as atypical thymoma. Rieker and colleagues [12] demonstrated excellent reproducibility (K 0.97) when WHO categories were merged in only three groups. In addition these 3 subgroups showed distinct survival curves. Marchevsky and colleagues [20] performed a meta-analysis of published studies based on the WHO schema and concluded that WHO histopathologic categories could easily be collapsed into only three categories.

Alternatively, a marked improvement on thymoma classification delegating the final diagnosis to a panel of selected pathologists with a particular expertise in the WHO classification of thymic tumors. Finally, cases could be shared by more experienced and unexperienced pathologists in an attempt to reach a consensus on the diagnosis. In the present study we adopted such approach on the 23 cases with at least two different diagnoses: it was easy to find a consensus on the classification by sharing the same material at the same time during a web conference,.

CONCLUSIONS

Our results underline the necessity to sharpen the accuracy in the allocation of these neoplasms in defined diagnostic categories to improve their classification, the delineation of significantly different prognostic subgroups, and the management of patients. While waiting for advances in understanding the biology of these rare tumors that may help in the future to refine their classification, sharing the diagnosis with expert pathologists may reduce the diagnostic discrepancies and facilitate the setting of a proper therapeutic strategy in clinical practice.

Figure 1.

A thymic tumor characterized by well demarcated lobules (A); even at scanning magnification these lobules show different amount of epithelial cells (B); in some ares neoplastic cells are characterized by a larger amount of cytoplasm and vescicular nuclei with small but distinct nucleoli (C); most of the fields, however, are characterized by few neoplastic cells intermingled among several lymphocytes. This lesion was classified as type B1 (n° 2) and B2 (n° 2) thymomas.

Figure 2.

A thymic tumor made by neoplastic cells with prominent eosinophilic nucleoli, a “syncitial” appearance, arrangend in lobules separated by lymphocytes. This lesion was classified as thymic carcinoma (n° 2), type A (n° 1) and type B3 thymoma (n° 1).