EurJ Cmm, Vol. 28A, No. 11, pp. 185&1862,1992. Printedin Grea Blitarn
096&1947l92$S.W + 0.W 0 1992 l’erganwnPressLrd
An Inter-observer and Intra-observer Variability Study on the Diagnosis of Lymph Node Biopsy Specimens A.M. Hanby, P.A. Hall, N. Rooney, l? Dennis, P. James, P. Richman, S. Buk, D.A. Levison and W.M. Gregory One hundred lymph node biopsy specimens were examined on two separate occasions by seven pathologists differing in experience in lymphoreticular pathology. Neither history nor immunohistochemistry was provided and the study, therefore, focused on morphological interpretation alone. The participants evaluated each case using a constructed response form in which the confidence with which they entered each response was also entered. Agreement on various points, between pathologists, between the two rounds, and with the referring centre was assessed. Whilst there was a high level of agreement over a diagnosis of benign vs. malignant and non-Hodgkin lymphoma vs. Hodgkin’s disease, there was considerably less agreement over both T vs. B cell phenotype and high vs. low grade. The lack of agreement over grade, an evaluation which is usually made independent of immunohistochemistry, is particularly important, because of the relevance to selection of treatment. Proliferation markers may be more appropriate determinants of treatment choice. EurJ Cancer, Vol. 28A, No. 11, pp. 1858-1862,1992. INTRODUCTION THE HISTOLOGICALdiagnosis of lymph
node disease has a profound influence on patient management. Both the initial diagnosis of malignancy, and the subsequent classifications into Hodgkin’s disease (HD), non-Hodgkin lymphoma (NHL), and further subclassifications of these diseases determine whether and what combinations of cytotoxic chemotherapy will be used. Consequently, it is important to know both how closely pathologists agree on their diagnoses and whether they are consistent. There have been several studies looking at agreement on histopathological diagnosis, including reviews by a single pathologist [l] or several pathologists [2-4]. More recently, multiobserver studies have been conducted into the diagnosis of lymphoreticular disease but most have concerned themselves with specific aspects, such as the histological subclassification of advanced high grade lymphoma [5], follicular lymphoma [6] and the reproducibility of the major lymphoma classification systems [7]. None of these studies has concerned itself with the full range of lymphoreticular disease, including reactive and neoplastic conditions and thus reflecting the range of nodal disease encountered in surgical pathology. We have performed this study in such a way that it serves as an audit of diagnostic accuracy and reproducibility based on the interpretation of morphology alone. Correspondence to W.M. Gregory. A.M. Hanby and W.M. Gregory are at the ICRF Clinical Oncology Unit, Guy’s Hospital, London SE1 9RT; P.A. Hall is at the Department of Histopathology, Royal Postgraduate Medical School, Hammersmith Hospital, London W12 OHS; N. Rooney is at the Department of Histopathology, Bristol Royal Infirmary, Bristol BS2 8HW; R. Dennis is at the Department of Pathology, Peterborough District Hospital, Peterborough PE3 6DA; P. James is at the Department of Histopathology, University Hospital, Queen’s Medical Centre, Nottingham NG7 2JH; P. Richman is at the Department of Cytopathology, St Bartholomew’s Hospital, West Smithfield, London EClA 7BE; and D. A. Levison and S. Buk are at the Department of Histopathology, UMDS, Guy’s Hospital, London SE1 9RT, U.K. Revised 28 Feb. 1992; accepted 14 Apr. 1992.
Using the updated Kiel classification of non-Hodgkin lymphomas [8] and current concepts of other nodal entities [9] we have attempted to gain insight into areas of both group and individual difficulty in diagnostic lymphoreticular pathology. METHODS
Five centres (Guy’s and St Bartholomew’s Hospitals, London; Peterborough District Hospital; Queen’s Medical Centre, Nottingham; Bristol Royal Infirmary) each identified from their files 20 nodal biopsies. A block from each case was sectioned at Guy’s Hospital with the preparation of eight sets of haematoxylin plus eosin-stained slides. Case selection at each of the centres was left entirely to the selector and was unknown to any of the other participants. However, any case including even “normal” nodal tissue was suitable for inclusion in this study. The cutting of the sections was performed by a non-participant. The sets were sent to each of seven participating diagnostic histopathologists within the study with recirculation after renumbering, by a nonparticipant 4 months later. All the participants reviewed their set of slides within 6 weeks of receipt on each occasion. For each circulation of each case a constructed response form was completed with categorisations into benign or malignant, NHL or HD, HD subtype, NHL T cell or B cell and high grade or low grade, plus, where possible, a final diagnosis. Against each positive response a confidence score was given (range l-10) to gauge the responding pathologist’s confidence in that categorisation. All the pathologists were instructed to use the updated Kiel classification [8] with which they were familiar, when a diagnosis of Iymphoma was made. The participants ranged in seniority from Professor to Senior Registrar/Research Fellow. Four of the participants claimed a special interest in lymph node pathology, while the other three did not. Individual diagnoses were compared both with those of the reporting hospital and also the mode diagnosis, the latter being the majority response for any particular case. To represent the levels of agreement observed, group and individual agreement
1858
Lymph Node Biopsies: Inter-observer Variability in Diagnosis with a mode diagnosis was given for each parameter, expressed as a percentage. The percentage agreement on each case and in each diagnostic category between rounds was a measure of consistency. The average degree of confidence scores given by the participants were calculated for each parameter. To give an additional measure of the strength of the agreements on the various classifications, and to determine whether these were greater than might be expected by chance, kappa [ 10, 1l] statistics were calculated. Since these compare only two raters, they were calculated for all pairs of raters, and an average kappa was reported (see Table 2). Ninety-five per cent confidence limits were also calculated for each kappa. For all the kappas reported the median of the lower 95% limits for a particular categorisation was greater than 0, implying that this level of agreement was very unlikely to have occurred merely by chance. There are 91 possible combinations of 14 raters taken in pairs. Seven of these represent comparisons of the same rater on two different occasions. Thus the mean kappa for agreement between pathologists is based on 84 separate kappas, while the between rounds kappas are based on a mean of seven values. To provide a meaningful kappa statistic for each of the main categorisations only the cases where there was agreement on the previous classification in the hierarchy were considered. For example, to evaluate NHL grade, only those cases where the two raters had both classified the slide as malignant NHL were used. Thus, for example, the mean kappa for the HD/NHL classification is less than that for distinguishing between NHL grade, suggesting that the distinction between high and low grade for a slide already categorised as NHL is more difficult than deciding whether a slide already thought to be malignant is HD or NHL. The values for kappa shown in Table 2 are uniformly lower than the percentage agreements also reported. This occurs because the latter take no account of agreements which could be expected by chance. However, the percentage agreement is still a useful statistic, in that it relates to a uniform agreement about a particular slide among all the pathologists, whereas the kappa statistics only relate to agreement between pairs of pathologists, and different pairs may agree on different diagnoses for the same case. In the analysis of individual performance the diagnoses given were separately compared with both the mode and referring hospital (actual) diagnoses. This analysis excluded any cases submitted by the participant being analysed. RESULTS Benign vs. malignant (Tables 1 and 2)
There was a high level of agreement (92%, mean kappa = 0.67) in the categorisation of nodes as benign or malignant. There was more agreement on average for a diagnosis of malignant (93%) than benign (78%). The relative ease of discrimination between each group was also reflected in the confidence scores; the average confidence for a benign diagnosis (7.3) being lower than for a malignant diagnosis (9.2). Pathologists were relatively consistent between rounds (92% agreement). In five cases the actual diagnosis of the referring centre was at variance with the mode diagnosis. The mode disagreed towards malignant in three of these cases and towards benign in two. Hodgkin’s disease vs. non-Hodgkin lympkoma (Tables 1 and 2)
The agreement for this categorisation of 85% (mean kappa = 0.69) was similar to that for the separation between benign and malignant, with consistency between rounds (91%) remaining
1859
Table 1. Levels of agreement and confidence of assignment
Average Average agreement agreement between Distinction
between pathologists rounds (%) (%)
Number in group* Average (defined by confidence mode score agreement)
Benign Malignant
100 90
78 93
7.4 9.2
18 91
HDt NHL*
100 80
84 87
8.3 8.3
23 54
70 30 90 50
57 36 86 54
6.4 6.4 8.4 6.9
NHL grade Low grade High grade
80 70
79 70
8.3 8.6
30 23
NHL lineage B cell T cell
90 50
79 59
7.5 5.5
35 12
HD subtype LP nodular LP diffuse NS MC LD (none present in study)%
HD, Hodgkin’s disease; NHL, non-Hodgkin lymphoma. *Those classified as uncertain are not included in this column. tThose dissenting: 81% to NHL, 12.5% to benign, 6.2% unsure. $Those dissenting: 23% to HD, 38.5% to benign, 46% unsure. §One pathologist responded once to this diagnosis with a confidence score of 7. It was in fact metastatic melanoma.
high. The agreement on a diagnosis of Hodgkin’s disease (84%) was similar to that for NHL (87%), as were the confidence scores (8.3 and 8.2, respectively). Of dissenting opinions away from a mode diagnosis of Hodgkin’s disease, 81% were towards a diagnosis of NHL. The converse was not, however, true with 84.5% of the dissenting opinions in the NHL group being either in favour of benign diagnoses or unsure. Hodgkin’s disease subtype (Tables 1 and 2)
The inter-observer agreement over subtyping was 66% (mean kappa = 0.44) with a between rounds concordance of 81% (mean kappa = 0.53) indicating rather better intra-observer consistency. Nodular sclerosing Hodgkin’s disease showed high inter-observer agreement (86%), all the other subtypes scoring Table 2. Overall levels of agreementfor particular categorisations Between rounds
Distinction Benign vs. malignant HD vs. NHL HD subtype NHL grade NHL lineage
Between pathologists
Agreement (%)
Mean Kappa
Agreement (%)
Mean Kappa
92 91 81 95 75
0.79 0.77 0.53 0.62 0.47
92 85 66 73 68
0.67 0.69 0.44 0.55 0.37
HD, Hodgkin’s disease; NHL, non-Hodgkin lymphoma.
A.M. Hanby et al.
1860
Table 3. Comparison of nodes diagnosed by referring centre as having Hodgkin’s disease with the mode diagnosis by subtype Referring centre diagnosis
Table 5. Agreement of lineage, actual vs. modal lineage assignment
Mode diagnosis
Modal lineage assignment
Hodgkin’s subtype Lymphocyte predominant Nodular = 3 Lymphocyte predominant Diffuse = 3
1 agreed 1 NHL 1benign 1 agreed 1 NHL 1 lymphocyte predominantnodular
Actual
B
T
Uncertain lineage
B T Uncertain lineage
26 1 5
3 5 0
2 3 0
Nodular sclerosing = 11
8 agreed 2 mixed cellularity 1 NHL
The agreement between actual and modal grading with dissent in only 3/45 cases of NHL.
Mixed cellularity = 7
6 agreed 1 NHL
Individual performance (Table 6)
relatively low (range 3657%). The between-rounds agreement for nodular sclerosing was also high, the average agreement being 90°h, the other subtypes ranging between 30 and 70% (excluding cases where subtype was not given by the participant). Again the confidence score reflected these differences, with an average score of 8.4 for nodular sclerosing and values between 6.4 and 6.9 for the other subtypes. Only two out of six mode diagnoses agreed with the actual referring centre diagnosis of lymphocyte predominant Hodgkin’s disease (both nodular and diffuse subtypes). Agreement between actual and modal diagnoses for the subtypes nodular sclerosing and mixed cellularity was better (Table 3). Non-Hodgkin lymphoma (Tables 1,2 and 5)
For NHL, agreement over lineage was comparable to that for Hodgkin’s subtyping (68%, mean kappa = 0.37), but observers were less consistent between rounds (75%, mean kappa = 0.47). There was also a high level of disagreement over lineage between the actual and modal diagnosis. Discrepancies were more marked for cell lineage than for grade (see below). Agreement between observers (79%) and rounds (90%) over a B lineage was high, while for T it was not (59 and 50%, respectively). Likewise, the confidence scores of 7.5 for B and 5.5 for T indicate that this is another area of difficulty. Non-Hodgkin lymphoma (Tables 1,2 and 4)
There was a low level of agreement (73%, mean kappa = 0.55) in the grading of NHL, though with a higher level of agreement between rounds (95%, mean kappa = 0.62). Agreement over low grade (79%) was slightly greater than for high grade (70%) though confidence scores were very similar (8.3 and 8.6, respectively). The agreement between rounds for these two groups was 80 and 70%, respectively.
was high
This was assessed on the percentage agreement of each pathologist with the modal diagnosis. Due to the gradual reduction in agreement as the categorisations become more specific, it was decided to analyse only the four main categorisations (Table 6). Overall, the levels of agreement were very similar between pathologists. However, two (4 and 7) differed markedly in their agreement when the mode diagnosis was benign. Pathologist 4 agreed with a mode diagnosis of benign in 43%, though in most cases where he dissented it was only to the extent of being unsure. This pathologist also showed a low level of agreement with the diagnosis of Hodgkin’s disease. Pathologist 7 was at variance with the diagnosis of benign but otherwise was similar in performance to the bulk of the group. This pathologist professed an interest in lymph node disease and it is thus difficult to explain this discrepancy as due to lack of experience; however, greater experience may have resulted in extra caution and an unsure response. DISCUSSION The overall agreement between benign or malignant diagnoses and between the referring centre diagnosis and the mode may superficially appear to be quite high. However, for every 10 lymph nodes looked at by the seven pathologists, this implies one disagreement on this crucial parameter. Although this is higher than one would hope for in working practice, in this study a number of problematic cases were known to have been submitted and neither history nor immunostaining results were provided. In addition, the diagnosis of malignant is a positive decision, whilst that of benign, particularly in this sort of study, is made almost by exclusion. The feeling that there might be a catch in the benign group may also be reflected in the lower Table 6. Individual performance on selected groups Pathologist (1-7) (Percentage agreement with mode diagnosis)
Table 4. Agreement over grade, actual vs. modal grade (45 cases) Mode diagnosis
1
2
3
4
5
6
7
94 98.4 79 91.2
83 89 97 92
87.5 91.2 83.8 87
43.3 97.7 63.6 97
78 95 87 97.5
84.5 97 82 93.5
69.5 94.5 90 95
Modal Actual Low grade High grade
Low grade
High grade
23 2
1 19
Benign Malignant Hodgkin’s disease Non-Hodgkin lymphoma
Lymph Node Biopsies: Inter-observer Variability in Diagnosis
confidence score. A number of benign nodal conditions are known to mimic malignancy and have been discussed in the literature [12]. This knowledge may account for the caution with which a benign diagnosis was made. The inter-round concordance was high and suggests that those features used by any one observer for a benign/malignant distinction are relatively constant. Turning to the HD vs. NHL distinction, levels of agreement were only slightly poorer than for the benign vs. malignant decision. It is interesting that participants dissenting from a mode diagnosis of Hodgkin’s disease nearly all did so in the direction of NHL, but only a small number opted for Hodgkin’s disease when the mode diagnosis was NHL, despite the fact that cells resembling Reed-Sternberg cells may be seen in NHL [13]. These trends were also seen when the referring centre and mode diagnoses were compared. The mistake of calling Hodgkin’s disease NHL was partly the result of observers erring in the direction of T cell lymphoma, knowing that these two diagnostic entities can mimic each other. Only a minority of dissent from the mode diagnosis of Hodgkin’s disease was towards a benign diagnosis. Miller et al. found a similar low figure (3/36). Again this is perhaps surprising as cells resembling Reed-Sternberg cells may also be seen in benign disease [13, 141. Hodgkin’s disease subtyping appears to be fraught with difficulty. Only nodular sclerosing Hodgkin’s disease, with its distinct hyaline fibrosis and lactmar Reed-Sternberg cells, was diagnosed with a high level of agreement, confidence and consistency. This difficulty in subtyping may be because the morphological features of the subtypes all merge with each other and with forms of NHL. For example, mixed cellularity can often be very similar to cellular phase nodular sclerosing, and lymphocyte predominant Hodgkin’s disease mimics both benign lesions and follicular lymphoma. Similar findings were uncovered by Miller et al. [3]. This, no doubt, accounted for the poor results in these groups. Although treatment choice is rarely based on HD subtype, this may well change. The international Hodgkin’s disease database [ 151, comprising data from over 14 000 patients, shows significant differences between all the subtypes, in a multivariate analysis, with lymphocyte depleted histology faring especially badly. Alternative strategies may be developed for these subtypes in the future, possibly involving more intensive treatments. Agreement on NHL lineage was also poor. However, the practical implication of these results is minimal, as immunohistochemistry is routinely used to help define lineage. The disparity on this point between actual (immunohistochemically supported) and modal (without immunohistochemical information) diagnosis is not surprising. Frizzera et al. [ 161 and later Jaffe et al. [ 171noted a high degree of disagreement between morphological and immunohistochemical determination of lineage. In addition, current therapies do not rely on cell lineage in most centres, though this may change. Lymphoma grading is of paramount importance and is the basis upon which most modern therapy for NHL is structured. Since the introduction of the CHOP regimen [ 181for high grade NHL, which yields long-term survival rates of 3OXt%, ever more intensive regimes have been developed in an attempt to improve on these results [19, 201. In contrast, for patients with disseminated low grade NHL, oral chlorambucil appears to give very similar results to those achievable with more intensive drug combinations, but with less toxicity [21, 221. The reasons for these differences in treatment choice and outcome appear to
1861
relate directly to the tumour’s growth rate [23, 241 and it seems likely that treatment choice can be tailored even more precisely by the use of proliferation indices such as Ki67 [25]. It is, therefore, disappointing that there was only 73% agreement over NHL grade. However, the high confidence scores given and the inter-round agreement suggest that observers make this distinction using their own relatively stable criteria. Unfortunately, these criteria appear to be different for different pathologists. Lymphoma grading is based both on cell size and on the proportion of blast cells present. The subjective nature of this assessment is a problem noted by other authors [6]. Other systems, such as Bloom and Richardson grading for breast carcinoma [26], may be more consistently applicable [27], as they are the result of a composite assessment of cytological and architectural features. However, epithelial neoplasms are more readily assessed in this manner and even then interobserver agreement in our experience is sometimes still poor. More objective and hopefully, more reproducible grading methods for lymphomas have used proliferation indices such as Ki-67 [25] and PCNA [28, 291 or other measures of cell proliferation [29]. The use of these measures should be encouraged in view of the significance of grade for determination of treatment. As mentioned, the diagnosis of NHL was made with a greater level of agreement than Hodgkin’s disease, with dissent on the latter group being mainly in the direction of NHL. When comparing the seven different observers, one stood out in his low level of agreement with a mode diagnosis of Hodgkin’s disease, though many of his responses were for the unsure category. The low degree of agreement of this pathologist in two out of the four categories analysed for individual performance, may reflect the non-lymphoma bias of his routine work and, in line with this, many of his responses were cautious as reflected by low confidence scores. A positive diagnosis of benign lymph node disease is often a worrying diagnosis to make. In many of the categories defined in this study there was good agreement between the actual (referring centre) and modal diagnosis. Where there was disagreement, however, it was not always clear which was the correct diagnosis. In certain areas, such as lineage determination, the use of immunohistochemistry by the referring centre should have led to a more correct diagnosis. Grogan and Thomas [30] found a high level of agreement in the immunophenotyping of NHL. With grading, however, where subjective parameters are involved, it is possible that a mode diagnosis, being the composite result of seven pathologists, would be more correct, especially where there was a high level of agreement on the mode. Coppelson et al. [2] advocated the advantages of such a panel of pathologists in the diagnosis of Hodgkin’s disease. Inevitably where there appears to be a morphological continuum, such as in Hodgkin’s disease subtyping or in the subtyping of small cell lymphomas [30, 311 definition of subtypes will remain arbitrary. In the separation of benign and malignant categories it is more difficult to say whether the referring centre diagnosis, with the aid of immunohistochemistry, provides a more correct diagnosis. The interpretation of such staining in difficult cases where this discrimination is made is often as subjective and difficult as the original morphology, and thus may add little. Perhaps a clearer advantage in favour of the referring centre diagnosis is provided by the clinical history, which was not available to the panel. This study was carried out in a somewhat artificial environment in which the diagnosis of lymph node disease was made without the aid of history or immunohistochemistry, and is thus a difficult case for accurate diagnosis. It has however, allowed
1862
A.M. Hanby et al.
attention to be paid to the morphology of lymph node disease. The study has been a valuable exercise in confirming some suspicions regarding lymph node diagnosis held by the participants. In particular it has demonstrated areas where there is a lack of either agreement, consistency or both. Whilst some of these difficulties can be resolved by immunohistochemistry, others, notably grading, cannot, unless perhaps some of the latest proliferation markers fulfill their apparent potential. The study also turned out to be a valuable educational and quality control exercise for all the participants. For many, it has drawn attention to individual black holes of diagnostic difficulty. The need for quality control in histopathology has been previously advocated [4, 32, 331 and, in this respect, further such audit exercises have much to recommend them and would, no doubt, give insight into other difficult areas. particular
1. Symmers W St C. Survey of the eventual diagnosis in 600 cases
referred for a second histological opinion after an initial biopsy diagnosis of Hodgkin’s disease._7 Clin Path01 1968,21,650-653. 2. Coppleson LW, Factor RM, Strum SB, Graff PW, Rappaport H. Observer disagreement in the classification and histology of Hodgkin’s disease._7 Nat1 Cancer Inst 1970,45,731-740. 3. Miller TP, Byrne GE, Jones SE. Mistaken clinical and pathologic diagnoses of Hodgkin’s disease: a Southwest Oncology Group Studv. Cancer Treat Reb 1982,66,645-651. 4. Woodruff 0. Reviewing histologic diagnosis of lymphoma. Arch Path01 Lab Med 1981,105,573-576. 5. Armitage JO, Dick FR, Platz CE, Corder MP, Leuniert JT. Clinical usefulness and reproducibility of histological subclassitication of advanced diffuse histiocytic lymphoma. Am J Med 1979, 67, 929-934. 6. Metter GE, Nathwani BN, Burke JS, et al. Morphological sub-
classification of follicular lymphoma: Variability of diagnoses among hematopathologists. A collaborative study between the Repository Center and Patholoxv Panel for Lvmohoma Clinical Studies. 7 Clin . _
Oncoll985,3,25-g. 7. NC1 Non-Hodgkin’s
Classification Projects Writing Committee. Classification of Non-Hodgkin’s lymphomas-reproducibility of major classification systems. Cancer 1985,55,91-95. 8. Stansfeld AG, Diebold J, Noel H, et al. Updated Kiel classification for lymphomas. Lancet 1988, i, 292-293. 9. Stansfeld AG (ed.). Lymph Node Biopsy Interpretation. Edinburgh, Churchill Livingstone, 1985. 10. Holman CDJ, James IR, Heenan PJ, er al. An improved method of analysis of observer variation between pathologists. Histopathology 1982,6,581-589.
11. Svanholm H, Starklint H, Barlebo H and Olsen S. Histological evaluation of prostatic cancer-l. Reproducibility of tumour type. Acta PatholMicrobiol Immunol Stand 1989,97,69%704. 12. Dorfman RF, Warnke R. Lymphadenopathy simulating the malignant lymphomas. Human Path01 1984,5,519-550. 13. Strum SB, Park JK, Rappaport H. Observations of cell resembling
Sternberg-Reed cells in conditions other than Hodgkin’s disease. Cancer 1970,26,176-190. 14. Tindle BH, Parker JW and Luker RJ. “Reed-Sternberg cells” in infectious mononucleosis. AmJ Clin Patholl972,58,607-617. 15. Somers R, Henry-Amar M, Meerwaldt JK, Carde P (eds). Treatment Strategy in Hodgkin’s Disease. London, John Libby & Co., 1990.
16. Frizzera CR, Grajl-Peczalska KJ, Blootield CD, Kersey JH. Predictability of immunologic phenotype of malignant lymphomas by conventional morphology. Cuncer 1979,43,1216-1224. 17. Jaffe ES, Strauchan JA, Berard CM. Predictability of immunologic phenotype by morphologic criteria in diffuse aggressive non-Hodgkin’s lymphoma. AmJ Clin Path01 1982,77,46-49. 18. McKelvey EM, Gottlieb JA, Wilson He, etal. Hydroxodaunomycin (adriamycin) combination chemotherapy in malignant lymphoma. Cancer 1976,38,1484-1493. 19. Fisher RI, DeVita VT Jr, Hubbard SM. Diffuse aggressive lymphomas: increased survival after alternating flexible sequences of ProMACE and MOPP chemotheram. __ Ann Int Med 1983,, 98., 304-309. 20. Skarin AT, Canellos GP, Rosenthal DS. Improved prognosis of diffuse histiocvtic and undifferentiated lvmohoma bv use of hixh dose methotrexate alternating with standard agents (M-BACON). J Clin Oncoll983,1,91-98. 21. Lister TA, Cullen MH, Beard MET, etal. Comparison of combined and single agent chemotherapy in non-Hodgkin’s lymphoma of favourable histological type. Br Med.7 1978,1,533-537. 22. Hoppe RT, Kushlan P, Kaplan HS, etal. The treatment ofadvanced stage favourable histology non-Hodgkin’s lymphoma; a preliminary report of a randomised trial comparing single agent chemotherapy, a combination chemotherapy and whole body irradiation. Blood 1981,58,592-598.
23. Shackney SE, McCormack GW, Cuchural GJ. Growth patterns of solid tumours and their relation to responsiveness to therapy. Ann ZntMed 1978,89,107-121.
24. Gregory WM. Cell proliferation and the principles of cancer chemotherapy. In Hall PA, Levison DA, Wright NA, eds. Assewnenr of Cell Proliferation in Clinical Practice. London, Springer-Verlag, 1992,193-205. 25. Hall PA, Richards MA, Gregory WM, D’ardenne AJ, Lister TA, Stansfeld AG. Prognostic value- of Ki67 immunostaming in nonHod&in’s Ivmohoma. 7Patholl988.154.223-232. 26. Bloom HJG, R&hard&n WW. Histological grading and prognosis in breast cancer. Br_7 Cancer 1957,11,359-377. 27. Elston CW. The assessment of histological differentiation in breast cancer. Aust NZ3 Surg 1984,54,11-15. 28. Woods AL, Hall PA, Shephard NA, et ol. The assessment of proliferating cell nuclear antigen (PCNA) innnunostaining in primary gastrointestinal lymphomas and its relationships to histological grade, S+GZ+M-phase fraction (flow cytometric analysis) and prognosis. Histopathologv 1991,19,21-27.. 29. Hall PA. Levison DA, Woods AL. et al. Proliferating cell nuclear antigen (PCNA) immunolocalisation in paraffin sections: an index of cell proliferation with evidence ofderegulated expression in some neoplasms.3 Path01 1990,162,285-294. 30. Grogan TM and Tubbs RR. A double-blind comparative immunotypic study between two institutions phenotyping non-Hodgkin’s lymphomas. Am-7 Clin Patholl987,87,478-484. 31. Dardick I, Caldwell DR. Follicular centre cell lymphoma. Cancer 1986,58,2477-2484. 32. Langley FA. Quality control in histopathology and diagnostic cytology. Histopathology 1978,2,3-18. 33. Gilchrist KW, Harrington DP, Wolf BC, Neiman RS. Statistical and empirical evaluation of histopathologic reviews for quality assurance in the Eastern Co-operative Oncology Group. Cancer 1988,62,861-868. Acknowledgements--We would like to thank Dr M. Richards, ICRF Department of Clinical Oncology, Guy’s Hospital for his advice and Miss J. Lunan for typing the numerous drafts.
096&1947l92$S.W + 0.W 0 1992 l’erganwnPressLrd
An Inter-observer and Intra-observer Variability Study on the Diagnosis of Lymph Node Biopsy Specimens A.M. Hanby, P.A. Hall, N. Rooney, l? Dennis, P. James, P. Richman, S. Buk, D.A. Levison and W.M. Gregory One hundred lymph node biopsy specimens were examined on two separate occasions by seven pathologists differing in experience in lymphoreticular pathology. Neither history nor immunohistochemistry was provided and the study, therefore, focused on morphological interpretation alone. The participants evaluated each case using a constructed response form in which the confidence with which they entered each response was also entered. Agreement on various points, between pathologists, between the two rounds, and with the referring centre was assessed. Whilst there was a high level of agreement over a diagnosis of benign vs. malignant and non-Hodgkin lymphoma vs. Hodgkin’s disease, there was considerably less agreement over both T vs. B cell phenotype and high vs. low grade. The lack of agreement over grade, an evaluation which is usually made independent of immunohistochemistry, is particularly important, because of the relevance to selection of treatment. Proliferation markers may be more appropriate determinants of treatment choice. EurJ Cancer, Vol. 28A, No. 11, pp. 1858-1862,1992. INTRODUCTION THE HISTOLOGICALdiagnosis of lymph
node disease has a profound influence on patient management. Both the initial diagnosis of malignancy, and the subsequent classifications into Hodgkin’s disease (HD), non-Hodgkin lymphoma (NHL), and further subclassifications of these diseases determine whether and what combinations of cytotoxic chemotherapy will be used. Consequently, it is important to know both how closely pathologists agree on their diagnoses and whether they are consistent. There have been several studies looking at agreement on histopathological diagnosis, including reviews by a single pathologist [l] or several pathologists [2-4]. More recently, multiobserver studies have been conducted into the diagnosis of lymphoreticular disease but most have concerned themselves with specific aspects, such as the histological subclassification of advanced high grade lymphoma [5], follicular lymphoma [6] and the reproducibility of the major lymphoma classification systems [7]. None of these studies has concerned itself with the full range of lymphoreticular disease, including reactive and neoplastic conditions and thus reflecting the range of nodal disease encountered in surgical pathology. We have performed this study in such a way that it serves as an audit of diagnostic accuracy and reproducibility based on the interpretation of morphology alone. Correspondence to W.M. Gregory. A.M. Hanby and W.M. Gregory are at the ICRF Clinical Oncology Unit, Guy’s Hospital, London SE1 9RT; P.A. Hall is at the Department of Histopathology, Royal Postgraduate Medical School, Hammersmith Hospital, London W12 OHS; N. Rooney is at the Department of Histopathology, Bristol Royal Infirmary, Bristol BS2 8HW; R. Dennis is at the Department of Pathology, Peterborough District Hospital, Peterborough PE3 6DA; P. James is at the Department of Histopathology, University Hospital, Queen’s Medical Centre, Nottingham NG7 2JH; P. Richman is at the Department of Cytopathology, St Bartholomew’s Hospital, West Smithfield, London EClA 7BE; and D. A. Levison and S. Buk are at the Department of Histopathology, UMDS, Guy’s Hospital, London SE1 9RT, U.K. Revised 28 Feb. 1992; accepted 14 Apr. 1992.
Using the updated Kiel classification of non-Hodgkin lymphomas [8] and current concepts of other nodal entities [9] we have attempted to gain insight into areas of both group and individual difficulty in diagnostic lymphoreticular pathology. METHODS
Five centres (Guy’s and St Bartholomew’s Hospitals, London; Peterborough District Hospital; Queen’s Medical Centre, Nottingham; Bristol Royal Infirmary) each identified from their files 20 nodal biopsies. A block from each case was sectioned at Guy’s Hospital with the preparation of eight sets of haematoxylin plus eosin-stained slides. Case selection at each of the centres was left entirely to the selector and was unknown to any of the other participants. However, any case including even “normal” nodal tissue was suitable for inclusion in this study. The cutting of the sections was performed by a non-participant. The sets were sent to each of seven participating diagnostic histopathologists within the study with recirculation after renumbering, by a nonparticipant 4 months later. All the participants reviewed their set of slides within 6 weeks of receipt on each occasion. For each circulation of each case a constructed response form was completed with categorisations into benign or malignant, NHL or HD, HD subtype, NHL T cell or B cell and high grade or low grade, plus, where possible, a final diagnosis. Against each positive response a confidence score was given (range l-10) to gauge the responding pathologist’s confidence in that categorisation. All the pathologists were instructed to use the updated Kiel classification [8] with which they were familiar, when a diagnosis of Iymphoma was made. The participants ranged in seniority from Professor to Senior Registrar/Research Fellow. Four of the participants claimed a special interest in lymph node pathology, while the other three did not. Individual diagnoses were compared both with those of the reporting hospital and also the mode diagnosis, the latter being the majority response for any particular case. To represent the levels of agreement observed, group and individual agreement
1858
Lymph Node Biopsies: Inter-observer Variability in Diagnosis with a mode diagnosis was given for each parameter, expressed as a percentage. The percentage agreement on each case and in each diagnostic category between rounds was a measure of consistency. The average degree of confidence scores given by the participants were calculated for each parameter. To give an additional measure of the strength of the agreements on the various classifications, and to determine whether these were greater than might be expected by chance, kappa [ 10, 1l] statistics were calculated. Since these compare only two raters, they were calculated for all pairs of raters, and an average kappa was reported (see Table 2). Ninety-five per cent confidence limits were also calculated for each kappa. For all the kappas reported the median of the lower 95% limits for a particular categorisation was greater than 0, implying that this level of agreement was very unlikely to have occurred merely by chance. There are 91 possible combinations of 14 raters taken in pairs. Seven of these represent comparisons of the same rater on two different occasions. Thus the mean kappa for agreement between pathologists is based on 84 separate kappas, while the between rounds kappas are based on a mean of seven values. To provide a meaningful kappa statistic for each of the main categorisations only the cases where there was agreement on the previous classification in the hierarchy were considered. For example, to evaluate NHL grade, only those cases where the two raters had both classified the slide as malignant NHL were used. Thus, for example, the mean kappa for the HD/NHL classification is less than that for distinguishing between NHL grade, suggesting that the distinction between high and low grade for a slide already categorised as NHL is more difficult than deciding whether a slide already thought to be malignant is HD or NHL. The values for kappa shown in Table 2 are uniformly lower than the percentage agreements also reported. This occurs because the latter take no account of agreements which could be expected by chance. However, the percentage agreement is still a useful statistic, in that it relates to a uniform agreement about a particular slide among all the pathologists, whereas the kappa statistics only relate to agreement between pairs of pathologists, and different pairs may agree on different diagnoses for the same case. In the analysis of individual performance the diagnoses given were separately compared with both the mode and referring hospital (actual) diagnoses. This analysis excluded any cases submitted by the participant being analysed. RESULTS Benign vs. malignant (Tables 1 and 2)
There was a high level of agreement (92%, mean kappa = 0.67) in the categorisation of nodes as benign or malignant. There was more agreement on average for a diagnosis of malignant (93%) than benign (78%). The relative ease of discrimination between each group was also reflected in the confidence scores; the average confidence for a benign diagnosis (7.3) being lower than for a malignant diagnosis (9.2). Pathologists were relatively consistent between rounds (92% agreement). In five cases the actual diagnosis of the referring centre was at variance with the mode diagnosis. The mode disagreed towards malignant in three of these cases and towards benign in two. Hodgkin’s disease vs. non-Hodgkin lympkoma (Tables 1 and 2)
The agreement for this categorisation of 85% (mean kappa = 0.69) was similar to that for the separation between benign and malignant, with consistency between rounds (91%) remaining
1859
Table 1. Levels of agreement and confidence of assignment
Average Average agreement agreement between Distinction
between pathologists rounds (%) (%)
Number in group* Average (defined by confidence mode score agreement)
Benign Malignant
100 90
78 93
7.4 9.2
18 91
HDt NHL*
100 80
84 87
8.3 8.3
23 54
70 30 90 50
57 36 86 54
6.4 6.4 8.4 6.9
NHL grade Low grade High grade
80 70
79 70
8.3 8.6
30 23
NHL lineage B cell T cell
90 50
79 59
7.5 5.5
35 12
HD subtype LP nodular LP diffuse NS MC LD (none present in study)%
HD, Hodgkin’s disease; NHL, non-Hodgkin lymphoma. *Those classified as uncertain are not included in this column. tThose dissenting: 81% to NHL, 12.5% to benign, 6.2% unsure. $Those dissenting: 23% to HD, 38.5% to benign, 46% unsure. §One pathologist responded once to this diagnosis with a confidence score of 7. It was in fact metastatic melanoma.
high. The agreement on a diagnosis of Hodgkin’s disease (84%) was similar to that for NHL (87%), as were the confidence scores (8.3 and 8.2, respectively). Of dissenting opinions away from a mode diagnosis of Hodgkin’s disease, 81% were towards a diagnosis of NHL. The converse was not, however, true with 84.5% of the dissenting opinions in the NHL group being either in favour of benign diagnoses or unsure. Hodgkin’s disease subtype (Tables 1 and 2)
The inter-observer agreement over subtyping was 66% (mean kappa = 0.44) with a between rounds concordance of 81% (mean kappa = 0.53) indicating rather better intra-observer consistency. Nodular sclerosing Hodgkin’s disease showed high inter-observer agreement (86%), all the other subtypes scoring Table 2. Overall levels of agreementfor particular categorisations Between rounds
Distinction Benign vs. malignant HD vs. NHL HD subtype NHL grade NHL lineage
Between pathologists
Agreement (%)
Mean Kappa
Agreement (%)
Mean Kappa
92 91 81 95 75
0.79 0.77 0.53 0.62 0.47
92 85 66 73 68
0.67 0.69 0.44 0.55 0.37
HD, Hodgkin’s disease; NHL, non-Hodgkin lymphoma.
A.M. Hanby et al.
1860
Table 3. Comparison of nodes diagnosed by referring centre as having Hodgkin’s disease with the mode diagnosis by subtype Referring centre diagnosis
Table 5. Agreement of lineage, actual vs. modal lineage assignment
Mode diagnosis
Modal lineage assignment
Hodgkin’s subtype Lymphocyte predominant Nodular = 3 Lymphocyte predominant Diffuse = 3
1 agreed 1 NHL 1benign 1 agreed 1 NHL 1 lymphocyte predominantnodular
Actual
B
T
Uncertain lineage
B T Uncertain lineage
26 1 5
3 5 0
2 3 0
Nodular sclerosing = 11
8 agreed 2 mixed cellularity 1 NHL
The agreement between actual and modal grading with dissent in only 3/45 cases of NHL.
Mixed cellularity = 7
6 agreed 1 NHL
Individual performance (Table 6)
relatively low (range 3657%). The between-rounds agreement for nodular sclerosing was also high, the average agreement being 90°h, the other subtypes ranging between 30 and 70% (excluding cases where subtype was not given by the participant). Again the confidence score reflected these differences, with an average score of 8.4 for nodular sclerosing and values between 6.4 and 6.9 for the other subtypes. Only two out of six mode diagnoses agreed with the actual referring centre diagnosis of lymphocyte predominant Hodgkin’s disease (both nodular and diffuse subtypes). Agreement between actual and modal diagnoses for the subtypes nodular sclerosing and mixed cellularity was better (Table 3). Non-Hodgkin lymphoma (Tables 1,2 and 5)
For NHL, agreement over lineage was comparable to that for Hodgkin’s subtyping (68%, mean kappa = 0.37), but observers were less consistent between rounds (75%, mean kappa = 0.47). There was also a high level of disagreement over lineage between the actual and modal diagnosis. Discrepancies were more marked for cell lineage than for grade (see below). Agreement between observers (79%) and rounds (90%) over a B lineage was high, while for T it was not (59 and 50%, respectively). Likewise, the confidence scores of 7.5 for B and 5.5 for T indicate that this is another area of difficulty. Non-Hodgkin lymphoma (Tables 1,2 and 4)
There was a low level of agreement (73%, mean kappa = 0.55) in the grading of NHL, though with a higher level of agreement between rounds (95%, mean kappa = 0.62). Agreement over low grade (79%) was slightly greater than for high grade (70%) though confidence scores were very similar (8.3 and 8.6, respectively). The agreement between rounds for these two groups was 80 and 70%, respectively.
was high
This was assessed on the percentage agreement of each pathologist with the modal diagnosis. Due to the gradual reduction in agreement as the categorisations become more specific, it was decided to analyse only the four main categorisations (Table 6). Overall, the levels of agreement were very similar between pathologists. However, two (4 and 7) differed markedly in their agreement when the mode diagnosis was benign. Pathologist 4 agreed with a mode diagnosis of benign in 43%, though in most cases where he dissented it was only to the extent of being unsure. This pathologist also showed a low level of agreement with the diagnosis of Hodgkin’s disease. Pathologist 7 was at variance with the diagnosis of benign but otherwise was similar in performance to the bulk of the group. This pathologist professed an interest in lymph node disease and it is thus difficult to explain this discrepancy as due to lack of experience; however, greater experience may have resulted in extra caution and an unsure response. DISCUSSION The overall agreement between benign or malignant diagnoses and between the referring centre diagnosis and the mode may superficially appear to be quite high. However, for every 10 lymph nodes looked at by the seven pathologists, this implies one disagreement on this crucial parameter. Although this is higher than one would hope for in working practice, in this study a number of problematic cases were known to have been submitted and neither history nor immunostaining results were provided. In addition, the diagnosis of malignant is a positive decision, whilst that of benign, particularly in this sort of study, is made almost by exclusion. The feeling that there might be a catch in the benign group may also be reflected in the lower Table 6. Individual performance on selected groups Pathologist (1-7) (Percentage agreement with mode diagnosis)
Table 4. Agreement over grade, actual vs. modal grade (45 cases) Mode diagnosis
1
2
3
4
5
6
7
94 98.4 79 91.2
83 89 97 92
87.5 91.2 83.8 87
43.3 97.7 63.6 97
78 95 87 97.5
84.5 97 82 93.5
69.5 94.5 90 95
Modal Actual Low grade High grade
Low grade
High grade
23 2
1 19
Benign Malignant Hodgkin’s disease Non-Hodgkin lymphoma
Lymph Node Biopsies: Inter-observer Variability in Diagnosis
confidence score. A number of benign nodal conditions are known to mimic malignancy and have been discussed in the literature [12]. This knowledge may account for the caution with which a benign diagnosis was made. The inter-round concordance was high and suggests that those features used by any one observer for a benign/malignant distinction are relatively constant. Turning to the HD vs. NHL distinction, levels of agreement were only slightly poorer than for the benign vs. malignant decision. It is interesting that participants dissenting from a mode diagnosis of Hodgkin’s disease nearly all did so in the direction of NHL, but only a small number opted for Hodgkin’s disease when the mode diagnosis was NHL, despite the fact that cells resembling Reed-Sternberg cells may be seen in NHL [13]. These trends were also seen when the referring centre and mode diagnoses were compared. The mistake of calling Hodgkin’s disease NHL was partly the result of observers erring in the direction of T cell lymphoma, knowing that these two diagnostic entities can mimic each other. Only a minority of dissent from the mode diagnosis of Hodgkin’s disease was towards a benign diagnosis. Miller et al. found a similar low figure (3/36). Again this is perhaps surprising as cells resembling Reed-Sternberg cells may also be seen in benign disease [13, 141. Hodgkin’s disease subtyping appears to be fraught with difficulty. Only nodular sclerosing Hodgkin’s disease, with its distinct hyaline fibrosis and lactmar Reed-Sternberg cells, was diagnosed with a high level of agreement, confidence and consistency. This difficulty in subtyping may be because the morphological features of the subtypes all merge with each other and with forms of NHL. For example, mixed cellularity can often be very similar to cellular phase nodular sclerosing, and lymphocyte predominant Hodgkin’s disease mimics both benign lesions and follicular lymphoma. Similar findings were uncovered by Miller et al. [3]. This, no doubt, accounted for the poor results in these groups. Although treatment choice is rarely based on HD subtype, this may well change. The international Hodgkin’s disease database [ 151, comprising data from over 14 000 patients, shows significant differences between all the subtypes, in a multivariate analysis, with lymphocyte depleted histology faring especially badly. Alternative strategies may be developed for these subtypes in the future, possibly involving more intensive treatments. Agreement on NHL lineage was also poor. However, the practical implication of these results is minimal, as immunohistochemistry is routinely used to help define lineage. The disparity on this point between actual (immunohistochemically supported) and modal (without immunohistochemical information) diagnosis is not surprising. Frizzera et al. [ 161 and later Jaffe et al. [ 171noted a high degree of disagreement between morphological and immunohistochemical determination of lineage. In addition, current therapies do not rely on cell lineage in most centres, though this may change. Lymphoma grading is of paramount importance and is the basis upon which most modern therapy for NHL is structured. Since the introduction of the CHOP regimen [ 181for high grade NHL, which yields long-term survival rates of 3OXt%, ever more intensive regimes have been developed in an attempt to improve on these results [19, 201. In contrast, for patients with disseminated low grade NHL, oral chlorambucil appears to give very similar results to those achievable with more intensive drug combinations, but with less toxicity [21, 221. The reasons for these differences in treatment choice and outcome appear to
1861
relate directly to the tumour’s growth rate [23, 241 and it seems likely that treatment choice can be tailored even more precisely by the use of proliferation indices such as Ki67 [25]. It is, therefore, disappointing that there was only 73% agreement over NHL grade. However, the high confidence scores given and the inter-round agreement suggest that observers make this distinction using their own relatively stable criteria. Unfortunately, these criteria appear to be different for different pathologists. Lymphoma grading is based both on cell size and on the proportion of blast cells present. The subjective nature of this assessment is a problem noted by other authors [6]. Other systems, such as Bloom and Richardson grading for breast carcinoma [26], may be more consistently applicable [27], as they are the result of a composite assessment of cytological and architectural features. However, epithelial neoplasms are more readily assessed in this manner and even then interobserver agreement in our experience is sometimes still poor. More objective and hopefully, more reproducible grading methods for lymphomas have used proliferation indices such as Ki-67 [25] and PCNA [28, 291 or other measures of cell proliferation [29]. The use of these measures should be encouraged in view of the significance of grade for determination of treatment. As mentioned, the diagnosis of NHL was made with a greater level of agreement than Hodgkin’s disease, with dissent on the latter group being mainly in the direction of NHL. When comparing the seven different observers, one stood out in his low level of agreement with a mode diagnosis of Hodgkin’s disease, though many of his responses were for the unsure category. The low degree of agreement of this pathologist in two out of the four categories analysed for individual performance, may reflect the non-lymphoma bias of his routine work and, in line with this, many of his responses were cautious as reflected by low confidence scores. A positive diagnosis of benign lymph node disease is often a worrying diagnosis to make. In many of the categories defined in this study there was good agreement between the actual (referring centre) and modal diagnosis. Where there was disagreement, however, it was not always clear which was the correct diagnosis. In certain areas, such as lineage determination, the use of immunohistochemistry by the referring centre should have led to a more correct diagnosis. Grogan and Thomas [30] found a high level of agreement in the immunophenotyping of NHL. With grading, however, where subjective parameters are involved, it is possible that a mode diagnosis, being the composite result of seven pathologists, would be more correct, especially where there was a high level of agreement on the mode. Coppelson et al. [2] advocated the advantages of such a panel of pathologists in the diagnosis of Hodgkin’s disease. Inevitably where there appears to be a morphological continuum, such as in Hodgkin’s disease subtyping or in the subtyping of small cell lymphomas [30, 311 definition of subtypes will remain arbitrary. In the separation of benign and malignant categories it is more difficult to say whether the referring centre diagnosis, with the aid of immunohistochemistry, provides a more correct diagnosis. The interpretation of such staining in difficult cases where this discrimination is made is often as subjective and difficult as the original morphology, and thus may add little. Perhaps a clearer advantage in favour of the referring centre diagnosis is provided by the clinical history, which was not available to the panel. This study was carried out in a somewhat artificial environment in which the diagnosis of lymph node disease was made without the aid of history or immunohistochemistry, and is thus a difficult case for accurate diagnosis. It has however, allowed
1862
A.M. Hanby et al.
attention to be paid to the morphology of lymph node disease. The study has been a valuable exercise in confirming some suspicions regarding lymph node diagnosis held by the participants. In particular it has demonstrated areas where there is a lack of either agreement, consistency or both. Whilst some of these difficulties can be resolved by immunohistochemistry, others, notably grading, cannot, unless perhaps some of the latest proliferation markers fulfill their apparent potential. The study also turned out to be a valuable educational and quality control exercise for all the participants. For many, it has drawn attention to individual black holes of diagnostic difficulty. The need for quality control in histopathology has been previously advocated [4, 32, 331 and, in this respect, further such audit exercises have much to recommend them and would, no doubt, give insight into other difficult areas. particular
1. Symmers W St C. Survey of the eventual diagnosis in 600 cases
referred for a second histological opinion after an initial biopsy diagnosis of Hodgkin’s disease._7 Clin Path01 1968,21,650-653. 2. Coppleson LW, Factor RM, Strum SB, Graff PW, Rappaport H. Observer disagreement in the classification and histology of Hodgkin’s disease._7 Nat1 Cancer Inst 1970,45,731-740. 3. Miller TP, Byrne GE, Jones SE. Mistaken clinical and pathologic diagnoses of Hodgkin’s disease: a Southwest Oncology Group Studv. Cancer Treat Reb 1982,66,645-651. 4. Woodruff 0. Reviewing histologic diagnosis of lymphoma. Arch Path01 Lab Med 1981,105,573-576. 5. Armitage JO, Dick FR, Platz CE, Corder MP, Leuniert JT. Clinical usefulness and reproducibility of histological subclassitication of advanced diffuse histiocytic lymphoma. Am J Med 1979, 67, 929-934. 6. Metter GE, Nathwani BN, Burke JS, et al. Morphological sub-
classification of follicular lymphoma: Variability of diagnoses among hematopathologists. A collaborative study between the Repository Center and Patholoxv Panel for Lvmohoma Clinical Studies. 7 Clin . _
Oncoll985,3,25-g. 7. NC1 Non-Hodgkin’s
Classification Projects Writing Committee. Classification of Non-Hodgkin’s lymphomas-reproducibility of major classification systems. Cancer 1985,55,91-95. 8. Stansfeld AG, Diebold J, Noel H, et al. Updated Kiel classification for lymphomas. Lancet 1988, i, 292-293. 9. Stansfeld AG (ed.). Lymph Node Biopsy Interpretation. Edinburgh, Churchill Livingstone, 1985. 10. Holman CDJ, James IR, Heenan PJ, er al. An improved method of analysis of observer variation between pathologists. Histopathology 1982,6,581-589.
11. Svanholm H, Starklint H, Barlebo H and Olsen S. Histological evaluation of prostatic cancer-l. Reproducibility of tumour type. Acta PatholMicrobiol Immunol Stand 1989,97,69%704. 12. Dorfman RF, Warnke R. Lymphadenopathy simulating the malignant lymphomas. Human Path01 1984,5,519-550. 13. Strum SB, Park JK, Rappaport H. Observations of cell resembling
Sternberg-Reed cells in conditions other than Hodgkin’s disease. Cancer 1970,26,176-190. 14. Tindle BH, Parker JW and Luker RJ. “Reed-Sternberg cells” in infectious mononucleosis. AmJ Clin Patholl972,58,607-617. 15. Somers R, Henry-Amar M, Meerwaldt JK, Carde P (eds). Treatment Strategy in Hodgkin’s Disease. London, John Libby & Co., 1990.
16. Frizzera CR, Grajl-Peczalska KJ, Blootield CD, Kersey JH. Predictability of immunologic phenotype of malignant lymphomas by conventional morphology. Cuncer 1979,43,1216-1224. 17. Jaffe ES, Strauchan JA, Berard CM. Predictability of immunologic phenotype by morphologic criteria in diffuse aggressive non-Hodgkin’s lymphoma. AmJ Clin Path01 1982,77,46-49. 18. McKelvey EM, Gottlieb JA, Wilson He, etal. Hydroxodaunomycin (adriamycin) combination chemotherapy in malignant lymphoma. Cancer 1976,38,1484-1493. 19. Fisher RI, DeVita VT Jr, Hubbard SM. Diffuse aggressive lymphomas: increased survival after alternating flexible sequences of ProMACE and MOPP chemotheram. __ Ann Int Med 1983,, 98., 304-309. 20. Skarin AT, Canellos GP, Rosenthal DS. Improved prognosis of diffuse histiocvtic and undifferentiated lvmohoma bv use of hixh dose methotrexate alternating with standard agents (M-BACON). J Clin Oncoll983,1,91-98. 21. Lister TA, Cullen MH, Beard MET, etal. Comparison of combined and single agent chemotherapy in non-Hodgkin’s lymphoma of favourable histological type. Br Med.7 1978,1,533-537. 22. Hoppe RT, Kushlan P, Kaplan HS, etal. The treatment ofadvanced stage favourable histology non-Hodgkin’s lymphoma; a preliminary report of a randomised trial comparing single agent chemotherapy, a combination chemotherapy and whole body irradiation. Blood 1981,58,592-598.
23. Shackney SE, McCormack GW, Cuchural GJ. Growth patterns of solid tumours and their relation to responsiveness to therapy. Ann ZntMed 1978,89,107-121.
24. Gregory WM. Cell proliferation and the principles of cancer chemotherapy. In Hall PA, Levison DA, Wright NA, eds. Assewnenr of Cell Proliferation in Clinical Practice. London, Springer-Verlag, 1992,193-205. 25. Hall PA, Richards MA, Gregory WM, D’ardenne AJ, Lister TA, Stansfeld AG. Prognostic value- of Ki67 immunostaming in nonHod&in’s Ivmohoma. 7Patholl988.154.223-232. 26. Bloom HJG, R&hard&n WW. Histological grading and prognosis in breast cancer. Br_7 Cancer 1957,11,359-377. 27. Elston CW. The assessment of histological differentiation in breast cancer. Aust NZ3 Surg 1984,54,11-15. 28. Woods AL, Hall PA, Shephard NA, et ol. The assessment of proliferating cell nuclear antigen (PCNA) innnunostaining in primary gastrointestinal lymphomas and its relationships to histological grade, S+GZ+M-phase fraction (flow cytometric analysis) and prognosis. Histopathologv 1991,19,21-27.. 29. Hall PA. Levison DA, Woods AL. et al. Proliferating cell nuclear antigen (PCNA) immunolocalisation in paraffin sections: an index of cell proliferation with evidence ofderegulated expression in some neoplasms.3 Path01 1990,162,285-294. 30. Grogan TM and Tubbs RR. A double-blind comparative immunotypic study between two institutions phenotyping non-Hodgkin’s lymphomas. Am-7 Clin Patholl987,87,478-484. 31. Dardick I, Caldwell DR. Follicular centre cell lymphoma. Cancer 1986,58,2477-2484. 32. Langley FA. Quality control in histopathology and diagnostic cytology. Histopathology 1978,2,3-18. 33. Gilchrist KW, Harrington DP, Wolf BC, Neiman RS. Statistical and empirical evaluation of histopathologic reviews for quality assurance in the Eastern Co-operative Oncology Group. Cancer 1988,62,861-868. Acknowledgements--We would like to thank Dr M. Richards, ICRF Department of Clinical Oncology, Guy’s Hospital for his advice and Miss J. Lunan for typing the numerous drafts.
