WHO Non-Hodgkin's Lymphoma Classification by Criterion-Based Report Review followed by Targeted Pathology Review: An Effective Strategy for Epidemiology Studies

Several large epidemiologic studies are investigating the increase in incidence of non-Hodgkin's lymphoma, which has occurred over the past several decades. The currently accepted, immunologically based Revised European American Lymphoma (1) and WHO (2) lymphoma classifications recognize that non-Hodgkin's lymphoma comprises several subtypes, each with different clinicopathologic features, geographic distribution, and, probably, etiology. In addition to morphology, ancillary tests, including immunophenotyping, are required for WHO classification. The morphologic Working Formulation (3) categories and broad subgroups of Working Formulation categories used previously in epidemiology do not generally translate accurately into the WHO subtypes, especially the rarer ones. Some WHO subtypes are so rare that meaningful results can only be obtained by pooling cases from different studies, as in the International Collaboration of Investigators Working on Lymphoma Epidemiological Studies (“InterLymph”) (4).

Ideally, all cases would be classified by central pathology review by a panel of expert hematopathologists, both in individual studies and when pooling data from several studies. This would be very expensive and time-consuming when large numbers of cases are involved. In addition, there may be difficulties in obtaining patient consent for access to diagnostic tissue, and archived material in pathology laboratories may not always be available.

In the New South Wales Non-Hodgkin's Lymphoma Study, an Australian population-based epidemiologic study of 717 cases and an International Collaboration of Investigators Working on Lymphoma Epidemiological Studies participant, an expert review was first conducted of all available pathology reports. From this information, a WHO classification category and the corresponding International Classification of Diseases for Oncology, Third Edition code was assigned to each case, with either a high (≥90%) or low (<90%) degree of confidence according to strict criteria that emphasized immunophenotype. The criteria used in assigning a high level of confidence in the WHO classification included a report with a confident morphologic diagnosis on an adequate tissue biopsy and demonstration of B-cell or T-cell/natural killer (NK)–cell phenotype by immunohistochemistry, flow cytometry, or molecular genetic studies. In addition, a combination of clinical information and results of ancillary studies was required for each WHO category using the criteria in Table 1 (reproduced from Annals of Oncology with permission from the European Society of Medical Oncology; ref. 5). A WHO classification was assigned in 91% of cases, but confidence was high in only 57.5% of these. The most common reasons for low confidence were insufficient immunophenotyping and inadequate material.

The present study was designed to review the original pathology of all cases previously unclassified or classified with low confidence in the report review, all of some rare subtypes, and a random sample of all other subtypes. In so doing, its aim was to increase the overall accuracy of classification of the cases in this study. Further, by estimating the accuracy of confident classification of cases by report review only, it would be possible to evaluate the effectiveness of restricting pathology review to cases unclassified or classified with low confidence on report review. If effective, use of this method in future epidemiologic studies could increase diagnostic accuracy while minimizing costs of pathology review.

The study population has been described in detail elsewhere (5, 6). Briefly, eligible cases were ages 20 to 74 years, resided in New South Wales or the adjacent Australian Capital Territory and were first diagnosed with non-Hodgkin's lymphoma between January 1, 2000 and August 31, 2001. Diagnoses of chronic lymphocytic leukemia (CLL), plasma cell myeloma, precursor B-cell and T-cell lymphoblastic leukemia, and lymphomatoid granulomatosis grade 1 and 2 were excluded from ascertainment for the study. The notifying clinician was also asked to exclude patients who had a history of transplantation or HIV infection.

Of 1,129 eligible patients ascertained and confirmed as eligible by notifying clinicians, 144 had died, 109 were too ill, and 34 could not be contacted. Of 842 (74%) patients invited to participate in the study, 717 (85%) participated and were subsequently included in the pathology report review (5). After report review, 13 cases were excluded because of low confidence in the diagnosis of non-Hodgkin's lymphoma: 2 with Hodgkin's lymphoma, 3 with probable Hodgkin's lymphoma, 3 with probable carcinoma, 1 with lymphomatoid granulomatosis grade 2, 3 with fine-needle aspiration cytology only and no immunophenotype, and 1 for which no report of a diagnostic biopsy was able to be located. A decision was made to retain in the study 31 cases of CLL and 4 cases of acute lymphoblastic leukemia found in the report review. Thus, 704 cases (410 males, 294 females) remained for further analyses.

Of the 704 cases remaining after report review, 419 were selected for pathology review according to three criteria, which were not mutually exclusive. We included all cases (n = 292) with low confidence in (n = 235) or who could not be allocated (n = 57) a WHO classification by report review (5), all cases (n = 103) of WHO subtypes of particular interest to International Collaboration of Investigators Working on Lymphoma Epidemiological Studies (mantle cell lymphoma, extranodal and nodal marginal zone lymphoma, CLL/small lymphocytic lymphoma, and T-cell lymphomas), and a 20% random sample of all remaining subtypes (n = 141); the samples from the third group were selected before removal of the low confidence cases, all of which were to be reviewed (5). No pathology material was requested for cases who refused to give written consent to allow the researchers access to their lymphoma tissue (n = 33), or died before they could do so (n = 6), or those of the remaining cases determined by the report review to have inadequate tissue specimens (n = 48).

Original stained slides and tissue blocks, or 15 unstained slides suitable for immunohistochemistry, were requested for the remaining 332 cases. Of these, original slides and/or blocks were obtained in 315 cases; 14 could not be located at the laboratory, 2 had insufficient material, and 1 was a fine-needle aspiration cytology sample only. The diagnostic biopsy specimen types included 24 tissue core biopsies, 53 incisional biopsies, 156 excisional biopsies, 18 endoscopic biopsies, 34 resections, 29 bone marrow trephines, and 1 fine-needle aspiration cytology (a case of CLL with typical immunophenotype). Original H&E-stained slides were received for 275 cases, original immunohistochemistry slides in 185, other stains in 84, blocks in 241, and unstained slides in 57. In 17 cases, only stained slides were sent.

Pathology reports relevant to the diagnosis of lymphoma were available in all 315 cases included in the pathology review; these included immunohistochemistry reports in 80% (251 of 315), flow cytometry reports in 55% (174 of 315), gene rearrangement reports in 7% (21 of 315), and conventional cytogenetic reports in 2% (7 of 315).

The pathology material was reviewed by one anatomic pathologist (J.J.T.) with a particular interest in hematopathology. H&E-stained slides were reviewed in 291 and immunohistochemistry (range, 1-27 antibodies; average, 6.1) in 289 cases. Antibodies were selected from the following list: CD20, CD79a, CD10, CD5, CD23, cyclin D1, bcl2, bcl6, CD35, CD138, κ, λ, CD43, CD3, CD4, CD8, CD7, CD56, TdT, CD45, CD30, CD15, Alk-1, EBV-LMP1, Ki-67, cytokeratin, S100 protein, melan A, CD34, myeloperoxidase, and CD68, and immunohistochemistry was done on a Ventana Benchmark autostainer (Ventana Medical Systems, Inc., Tucson, AZ) using standard protocols. The antibodies were selected for each case according to the diagnostic criteria for the WHO category which was to be confirmed or excluded (Table 1), taking into account available flow cytometry results and original immunoperoxidase-stained slides and the amount of material available. Fluorescence in situ hybridization was done in 19 cases. A LSI IgH/cyclin D1 dual-color, dual-fusion translocation probe was used in 9 cases to confirm or exclude mantle cell lymphoma, a LSI IgH/BCL2 dual-color, dual-fusion translocation probe was used in 5 cases to confirm or exclude follicular lymphoma, a LSI MYC dual-color, breakapart rearrangement probe was used in 5 cases to confirm or exclude Burkitt's lymphoma, and a LSA MALT1 (18q21) dual-color, breakapart rearrangement probe was used in 1 case to confirm or exclude a gastric MALT lymphoma with dissemination (Vysis, Abbott Laboratories, Downers Grove, IL). In situ hybridization for EBV early RNA (Ventana Medical Systems) was done in 12 cases to confirm or exclude lymphoma subtypes which may be EBV associated, specifically Hodgkin's lymphoma (3 cases), extranodal NK/T–cell lymphoma (5 cases), and angioimmunoblastic T-cell lymphoma (4 cases).

The criteria used for confident WHO classification were similar to those employed in the report review (Table 1; ref. 5), with slight modifications; CD10⁻ follicular lymphoma, CD45⁻ anaplastic large-cell lymphoma and CD8⁻ s.c. panniculitis-like T-cell lymphoma were accepted if the morphology was typical and there were no morphologic or immunophenotypic features to suggest other WHO subtypes.

The number of cases reviewed and the percentage of cases in which the report review agreed with the pathology review were tabulated and 95% confidence intervals were calculated using the β distribution (http://www.causascientia.org/math_stat/ProportionCI.htm). Specific changes to original report review categories and the nature of these changes were also tabulated.

The distribution by WHO classification category that might be expected for all cases after pathology review was estimated by distributing those sampled and available from each report review category to their pathology review category. To take account of the fact that not all slides were reviewed, the overall estimated distribution was then modified by multiplying each category by a weight that was the inverse of the fraction of slides reviewed in each category divided by all those originally allocated to that category.

The increase in accuracy of WHO classification of the study cases was estimated as the percentage of cases in which a WHO classification was made or changed by pathology review. The percentage agreement between pathology review and report review was calculated separately for high and low confidence report review classifications. These percentages were applied to all the unreviewed cases in the study to estimate the potential error remaining in the WHO classifications of the whole study population after pathology review, and to all the cases in the study to estimate the likely error by report review alone. Similar calculations were used to assess the effect on accuracy of complete pathology review, because all 285 cases not selected for review had high confidence WHO classifications; it was assumed that material would not be able to be obtained in the same percentage of cases as in the high confidence cases that were selected. The effect of not reviewing any high confidence report review classifications, and changing the criteria for high confidence report review classifications [by assigning a low level of confidence to all cases diagnosed as CD30⁺ diffuse large B-cell lymphoma (DLBCL), “high-grade” lymphoma, anaplastic large-cell lymphoma, diffuse follicle center lymphoma, and follicular lymphoma grade 3] was similarly calculated.

Of 419 cases selected for pathology review, material on which a review could be based was obtained for 315 (75%). The proportion obtained varied among the major report review categories of non-Hodgkin's lymphoma: 64% for DLBCL, 82% for follicular lymphoma, 79% for all other B-cell lymphomas, 85% for T-cell lymphomas, and 56% for those that were unclassifiable by report review. Further, pathology material was obtained for 112 (88%) of 128 high confidence cases selected and 203 (69%) of 291 low confidence cases. Relative to their proportions in the 315 cases that were reviewed, there were, in the 104 cases that were selected but not obtained for review, more of those unclassified at report review (24% versus 10%), classified as DLBCL on report review (22% versus 14%), and classified with low confidence or unclassified (86% versus 64%) and fewer of those classified as follicular lymphoma (24% versus 37%).

Ten cases were found not to be non-Hodgkin's lymphoma after pathology review. The pathology review diagnoses for these 10 ineligible cases were 5 reactive, 3 Hodgkin's lymphoma, 1 acute monoblastic leukemia, and 1 multiple myeloma. Nine of these 10 had low confidence in eligibility as non-Hodgkin's lymphoma after report review. Three hundred one (99%) of the remaining 305 cases were assigned a WHO classification, with high confidence in 87% (261 of 301).

Of the cases reviewed, 32 (10%) cases had not been classifiable on report review because of inability to exclude reactive lymphoid tissue (10 cases) or another tumor type (2 cases), no or insufficient immunophenotyping (9 cases), morphologic uncertainty in the pathology report (8 cases), or insufficient clinical information (3 cases). Of these 32 cases, disproportionate numbers were classified on pathology review as lymphoplasmacytic lymphoma/Waldenstrom's macroglobulinemia (4 of the 10 ultimately classified to this category, 40%), splenic marginal zone lymphoma (3 of 13, 23%), nodal marginal zone lymphoma (3 of 7, 43%), and not non-Hodgkin's lymphoma (4 of 10, 40%). Three of this group and one case classified as follicular lymphoma on report review could be classified no more specifically than “B cell not otherwise specified” on pathology review.

Overall, there was 78% agreement between the WHO classification assigned by report review and that assigned by pathology review (see Table 2). Agreement was >80% for 11 of the 23 WHO categories in the report review series. Agreement was greater when there was high confidence in the report review classification (92%; 95% confidence interval, 85-96%) than when there was low confidence (69%; 95% confidence interval, 62-75%). Ignoring categories with only one or two cases from the report review, agreement was least for Burkitt's lymphoma (40%, 2 of 5) and peripheral T-cell lymphoma unspecified (43%, 3 of 7) and greatest for splenic marginal zone lymphoma (100%, 4 of 4) and mantle cell lymphoma (88%, 14 of 16). Follicular lymphoma was the largest category with high agreement (82%, 95 of 116); agreement was slightly lower for DLBCL (77%, 33 of 43), the second largest group.

The specific changes made by pathology review in WHO classification categories assigned by report review are summarized in Table 3. Whereas the numbers reassigned from one category to another were generally small, the greatest areas of diagnostic confusion appeared to be among DLBCL and follicular lymphoma and the combined category formed by the two (14 cases changing category among them), between Burkitt's lymphoma and DLBCL (3 of 5 in the former changed to the latter on review), and between follicular lymphoma and marginal zone lymphomas (9 of the former distributed equally among splenic, extranodal, and nodal marginal zone lymphomas on review).

Nine cases with high confidence in the report review WHO classification were assigned a different WHO classification after pathology review. Four initially considered to be DLBCL were reclassified as follicular lymphoma (n = 2), precursor B lymphoblastic leukemia or lymphoma (n = 1), and Hodgkin's lymphoma (n = 1); one was considered on report review to be a diffuse large B-cell and splenic marginal zone lymphoma and was reclassified as a splenic marginal zone lymphoma; one was considered to be a follicular lymphoma and was reclassified as a DLBCL; and three were considered to be extranodal marginal zone B-cell lymphomas and reclassified as CD10⁺ follicular lymphoma (n = 1) and diffuse large B-cell and extranodal marginal zone B-cell lymphoma (n = 2).

Forty (13%) of the WHO classification assignments made on pathology review were considered to be of low confidence. Notably high proportions with low confidence based on appreciable numbers of cases were 54% (7 of 13) for splenic marginal zone lymphoma and 43% (3 of 7) for nodal marginal zone lymphoma, and notably low proportions were 6% for follicular lymphoma (6 of 104) and extranodal marginal zone B-cell lymphoma (2 of 32). The main reasons for these low confidence assignments were inadequate tissue biopsy (20%), insufficient clinical information (35%, particularly important for splenic and nodal marginal zone lymphomas and T-cell lymphomas as a group), morphologic uncertainty (12.5%), and insufficient or ambiguous immunophenotype. Three cases that were morphologically typical of mantle cell lymphoma were assigned a low confidence because cyclin D1 was negative and t(11;14) was not found.

The estimated distribution of non-Hodgkin's lymphoma cases across WHO classification categories after pathology review is compared with the observed distribution after report review in Table 4. Overall, the proportion of B-cell lymphomas rose by 0.8 of a percentage point and that of T-cell, null, and NK lymphomas fell by a corresponding amount. The largest absolute changes were in the largest categories, with the proportion of DLBCL falling by 5.2 percentage points and that of follicular lymphoma by 4.1. The largest absolute increases were in splenic marginal zone lymphoma (2.5 percentage points), DLBCL and follicular lymphoma (2.2), and MALT lymphoma (1.8).

The actual working classification of the 694 eligible cases that we will use in subsite-specific epidemiologic analyses of our data lies between the report review distribution and the estimated distribution after slide review. This distribution is based partly on pathology review and partly on report review when pathology was not reviewed because the cases in question were not sampled or the necessary pathology material was not available. For reference purposes, this distribution is small lymphocytic lymphoma or CLL (27 cases), lymphoplasmacytic lymphoma or Waldenstrom's macroglobulinemia (26 cases), mantle cell lymphoma (22 cases), DLBCL (231 cases), Burkitt's lymphoma or leukemia (4 cases), splenic marginal zone lymphoma (14 cases), follicular lymphoma (227 cases), extranodal marginal zone (MALT) B-cell lymphoma (37 cases), nodal marginal zone lymphoma (7 cases), precursor B lymphoblastic lymphoma or leukemia (4 cases), hairy cell leukemia (10 cases), diffuse large B-cell and follicular (24 cases), other combined B-cell subtypes (7 cases), B cell lymphoma not otherwise classified (25 cases), T-cell lymphomas (25 cases), and non-Hodgkin's lymphoma not otherwise classified (4 cases). By applying the overall agreement between pathology review and report review of 92% to high confidence cases that were not reviewed and 69% to low confidence cases that were not reviewed, it is estimated that 52 of 694 (7.5%) of these classifications may be in error (24 of 300 high confidence cases and 28 of 89 low confidence cases). The estimated error for report review alone is 18% (128 of 717, 33 of 412 high confidence cases and 95 of 305 low confidence cases). Complete pathology review would have required review of 249 of 285 high confidence cases not originally selected, assuming that 12.5% would not be able to be obtained [16 of 128 (12.5%) of high confidence cases selected were not obtained]. It is estimated that the classification would be changed in 20 of these cases, an estimated reduction in the overall error of 3% (20 of 694). If 112 cases with high confidence report review classifications had not been reviewed, the error would potentially be increased by 1.3% (9 of 694 cases). Modifying the criteria for high confidence report review classifications would have required review of 31 more cases, with a potential change in classification of up to 3 cases and a maximum reduction in the error by 0.4%.

Until recently, epidemiologic studies of non-Hodgkin's lymphoma have used the morphologic Working Formulation to define subtypes of the disease, and broad groups of subtypes have been considered sufficiently specific for analytic purposes. One study (7) concluded that expert panel review of H&E slides was probably not useful in epidemiology studies relative to time and cost. The agreement of the expert panel with the diagnosis of non-Hodgkin's lymphoma in abstracted reports of local pathologists was 93% but was only 55% with the Working Formulation non-Hodgkin's lymphoma subtype (range, 14-100% for different subtypes). Agreement among the panelists as to subtype was itself only 60%. Similar results were obtained in a more recent study (8) in which one expert pathologist reviewed original H&E-stained slides in a cancer registry–based case-control study and found 98% agreement with the registry diagnosis of non-Hodgkin's lymphoma, but agreement with the assigned International Classification of Diseases for Oncology, Second Edition subtypes (modified Working Formulation) was only 59% (range, 5-100% for different subtypes).

Non-Hodgkin's lymphoma subtyping using the WHO classification is more stringent but is reproducible in clinical settings. In The Non-Hodgkin's Lymphoma Classification Project (9), five expert hematopathologists obtained interobserver and intraobserver reproducibility of at least 85% for most of the major lymphoma subtypes in 1,403 cases from eight countries. Average agreement of 82% between individual panel members and the consensus diagnosis of the same panel of five experts was obtained for lymphomas contributed to a lymphoma register from all parts of India, and the agreement of the local host pathologist with the consensus diagnosis was not different from that of the individual experts (10). Similar results were obtained for agreement between a single, expert community pathologist and a panel of three in an academic center (89%; ref. 11) and several community pathologists and, generally, two members of a four-member expert panel (83%; ref. 12). Central pathology review of all cases by a panel of expert hematopathologists is generally considered to be the gold standard for classifying non-Hodgkin's lymphoma. These studies suggest that diagnoses will be changed in some 11% to 18% by panel review instead of review by a single hematopathologist. We were unable to find any previous expert pathology review of non-Hodgkin's lymphoma using the WHO classification in an epidemiologic study.

On the face of it, our study achieved its objective of increasing the accuracy of WHO classification of our non-Hodgkin's lymphoma cases. Ten cases were found that were not non-Hodgkin's lymphoma. A WHO non-Hodgkin's lymphoma subtype was assigned in 99% of the remaining reviewed cases, with high confidence in the classification in 87%. The four cases that were not assigned a WHO non-Hodgkin's lymphoma subtype could be specified as a B-cell phenotype. Altogether, this can be estimated to represent a 12.5% increase in accuracy of diagnostic classification of the 704 cases, because a WHO diagnosis was made or changed by pathology review in 88 of them.

Sampling and unavailability of pathologic material meant that we did not review slides of 389 (55%) of our cases. By applying the calculation described previously to all the unreviewed cases, we can anticipate errors of classification in 7.5% of the 694 cases now considered eligible in the study. This is a substantial improvement compared with a potential error rate of 18% using report review alone, estimated in the same way. Complete pathology review would have reduced the 7.5% error by only 3% (the residual inaccuracy being due to unavailability of pathology material for review) at a cost of 251 extra pathology reviews. If no high confidence cases had been reviewed, the error would have increased by only 1.3% for a saving of 113 pathology reviews. We also assessed whether the 92% accuracy of a high confidence report review classification could be improved by changing some of our criteria Adopting these changes would have led to review of an additional 31 low confidence cases with a change in classification of up to three of them (i.e., a maximum increase in accuracy of 0.4%). These incremental gains or losses in accuracy are comparatively small compared with the estimated 12.5% gain in accuracy from 315 pathology reviews.

Follicular lymphoma has traditionally been considered a relatively easy pathologic diagnosis but, for our report review, strict immunophenotypic criteria were required for a confident classification. After pathology review, nine further cases of follicular lymphoma were found among other report review WHO categories. In addition, 18% of cases originally classified as follicular lymphoma by report review were reclassified by pathology review, and only one of these had been classified with high confidence in the report review, thus justifying the use of strict criteria in the report review, particularly if there is an interest in identifying rare but epidemiologically interesting subtypes that might be confused with follicular lymphoma. The criteria used for confident classification of DLBCL in the report review (5) were less strict than for follicular lymphoma, as it was expected that morphologic diagnosis of DLBCL, together with a positive B-cell marker, would be accurate. After pathology review, however, 23% of cases originally classified as DLBCL were reclassified, including 4 cases classified with high confidence after report review. Sixteen further cases of DLBCL were found among other report review WHO categories after pathology review. Lack of familiarity of pathologists with the new criteria used to diagnose DLBCL in the WHO classification would have been involved in some of these cases. This comparatively high error rate in the original diagnosis of the two most common subtypes, follicular lymphoma and DLBCL, suggests that in past epidemiologic studies translation from the Working Formulation to the WHO classification without pathology review is likely to have been inaccurate.

The pathology review strategy used in this study restricted it to 45% of cases. This proportion could have been reduced to 29% if no high confidence diagnoses had been reviewed, without much loss in accuracy of classification, and could be reduced even further as pathologists become more familiar with the WHO classification and are educated to immunophenotype lymphomas. Notification of lymphoma immunophenotype, including flow cytometry, to cancer registries would improve WHO/International Classification of Diseases for Oncology, Third Edition coding of non-Hodgkin's lymphoma and should be encouraged. We recommend pathology report review using the previously published criteria (5) and review of pathology of cases with a low confidence report review WHO classification as an effective and cost-saving approach for epidemiologic studies. Resources of studies could then be conserved for central expert panel review of rare or difficult WHO subtypes.

We thank Naomi Crain, Melisa Litchfield, and Justine Simard-Lebrun for tracking down the slides and blocks from pathology laboratories.