Anatomic Pathology / HISTOLOGIC CLASSIFICATION OF TUMOR-INFILTRATING LYMPHOCYTES IN MALIGNANT MELANOMA
Histologic Classification of Tumor-Infiltrating Lymphocytes
in Primary Cutaneous Malignant Melanoma
A Study of Interobserver Agreement
Klaus J. Busam, MD,1 Cristina R. Antonescu, MD,1 Ashfaq A. Marghoob, MD,2
Kishwer S. Nehal, MD,2 Dana L. Sachs, MD,2 Jinru Shia, MD,1 and Marianne Berwick, PhD3
Key Words: Malignant melanoma; Statistical analysis; Tumor-infiltrating lymphocytes; Interobserver agreement
Abstract
The density and distribution of lymphocytes
infiltrating the vertical growth phase of primary
cutaneous melanomas has been suggested by several
studies to be of prognostic significance. However, few
pathologists comment on tumor-infiltrating lymphocytes
(TILs), and there is the perception that the assessment
of TILs is subject to great interobserver variability. We
studied interobserver agreement on the categorization
of TILs; 20 cases of primary cutaneous malignant
melanoma with a vertical growth phase component
were circulated among 3 pathologists and 3
dermatologists. For each case, TILs were classified as
brisk, nonbrisk, or absent according to Clark. Only 1
pathologist (a dermatopathologist) was familiar with
the classification of TILs. Observers were given written
guidelines and a brief tutorial before their examination
of the slides. Our results show that with little
instruction, overall agreement among observers was
good (kappa values, 0.6 or more), especially among
pathologists (kappa values, > 0.7). Three observers had
excellent agreement among each other (kappa values,
> 0.75). These findings suggest that the categorization
of TILs can be easily taught and can be applied with an
acceptable level of reproducibility in routine diagnostic
practice.
856
Am J Clin Pathol 2001;115:856-860
Histologic prognostic parameters of primary cutaneous
malignant melanoma have been the subject of numerous
studies.1-9 Multivariate analyses have demonstrated that
tumor thickness as measured according to Breslow1 is the
most powerful predictor of outcome (measured by recurrence, disease-free survival, and overall survival).4,5 Tumor
thickness is the fundamental component for the T classification of melanomas in the melanoma staging system of the
American Joint Committee on Cancer,10,11 and it is recommended that measurement of thickness be included in every
pathology report of primary cutaneous melanoma. Many also
view ulceration as an important prognostic indicator,4,5 at
least for thin melanomas, and the newly proposed American
Joint Committee on Cancer classification for melanoma uses
ulceration as a modifying component for T classification.11
A number of additional histologic parameters have been
suggested, including the density and pattern of distribution of
tumor-infiltrating lymphocytes (TILs) involving the vertical
growth phase (VGP) of melanoma 3,9,12 or metastatic
melanoma deposits in lymph nodes. 13 Although several
studies have suggested that a so-called brisk infiltrate of TILs,
as defined by Clark et al,3 is associated with a more favorable
clinical outcome, this parameter is currently viewed as
optional for the reporting of melanomas,14,15 and few pathologists comment on TILs (K.J.B., personal observations).
For any histologic parameter to become acceptable as a
standard feature for reporting, reproducibility of its assessment needs to be demonstrated. Some have suggested that
there may be poor interobserver agreement on TILs. 16
However, no study has formally investigated concordance
among various observers of this histologic parameter, and it
is not known to what extent the evaluation of TILs can be
taught and applied in a consistent manner.
© American Society of Clinical Pathologists
Anatomic Pathology / ORIGINAL ARTICLE
It was the goal of the present study to examine interobserver agreement in the classification of TILs. The group of
observers deliberately included nonexperts in the field: if
nonexperts can agree on TILs, this would strengthen the
candidacy of TILs as a broadly applicable histologic parameter. Since only TILs as categorized by Clark et al3 have
repeatedly been shown to be of prognostic significance,9,12
we focused our analysis on the reproducibility of the criteria
of Clark et al.3 We analyzed interobserver agreement not
only for the recognition of brisk vs nonbrisk vs absent but
also for brisk vs nonbrisk and absent grouped together. The
reason for including the latter category was that the recognition of “brisk” seems to be of greatest prognostic significance.3 In the study by Clark et al,3 the adjusted odds ratio
for survival for clinical stage I, VGP melanoma, was 11.3 for
brisk TILs, 3.4 for nonbrisk TILs, and 1.0 for absent TILs.
tau-b using the PROC CORR procedure (SAS Institute,
Cary, NC). This is a measure calculated from values of
paired observations as to whether they vary together (in
concord) or differently (in discord). The kappa statistic is a
measure that accounts for the agreement that might be
expected by chance. When a composite measure of agreement across categories is desired, a weighted estimate of
kappa is used, the interpretation of which is the same as for
the unweighted kappa. To compare agreement among all 6
observers and 2 groups of 3 observers, the formulas of
Fleiss17 were followed. Originally described by Cohen,18 the
higher the kappa value, the better the agreement. Kappa
values greater than 0.75 indicate excellent agreement beyond
chance; those between 0.40 and 0.75 indicate fair to good
agreement beyond chance; and values less than 0.40 indicate
poor agreement beyond chance.
Materials and Methods
Results
Twenty H&E-stained slides of malignant melanoma
were circulated among 3 pathologists and 3 dermatologists.
One of the pathologists was a board-certified dermatopathologist (K.J.B.) who selected the slides. Each slide contained a
primary cutaneous melanoma in VGP. Melanomas in radial
growth phase (RGP) were excluded. For each tumor, the
lymphocytic infiltrate was typed as brisk, absent, or nonbrisk
by each observer. The criteria formulated by Clark et al3
were used to classify the lymphocytic infiltrate.
Briefly, lymphocytes needed to disrupt and surround
tumor cells of the VPGP to qualify as TILs ❚Image 1A❚. If
TILs diffusely infiltrated the entire invasive component
❚Image 1B❚ or were present and infiltrating across the entire
base of the VGP ❚Image 1C❚, they qualified as brisk. TILs
were absent if no lymphocytes were present ❚Image 2❚ or if
lymphocytes were present but did not infiltrate the tumor at
all. If lymphocytes infiltrated melanoma only focally or not
along the entire base of the VGP, the term nonbrisk was used
❚Image 3❚.
With the exception of the dermatopathologist, none of
the other observers was familiar with the Clark model of
TILs. Therefore, a handout outlining the criteria was given to
each individual by the dermatopathologist, who also illustrated each category with 2 representative cases. None of the
didactic cases was included in the series of slides, which
were circulated among the 6 observers for evaluation. Before
circulating the slides, the dermatopathologist classified the
slides into brisk, nonbrisk, and absent, and this served as the
“correct” answer with which all other observer’s answers
were compared. Of the 20 melanoma cases, 7 tumors had
brisk, 6 had nonbrisk, and 7 had absent TILs.
The data were analyzed for concordance and by the
kappa statistic. Concordance was evaluated by the Kendall
The results are summarized in ❚Table 1❚ and ❚Table 2❚.
Table 1 illustrates that the overall agreement in distinguishing a brisk infiltrate of TILs from both nonbrisk and
absent together was good. The overall agreement for brisk vs
nonbrisk vs absent was moderate. When pathologists were
analyzed separately, the agreement was good for the classification of brisk vs nonbrisk vs absent and excellent for the
separation of brisk from nonbrisk and absent grouped
together. Likewise the agreement among dermatologists was
better for the distinction of brisk from the combined group of
nonbrisk and absent than for the distinction of brisk vs
nonbrisk vs absent, although the agreement was still good for
the latter.
When individual observers were compared with each
other (Table 2), the agreement was highly variable. On the
distinction of brisk vs nonbrisk vs absent, poor agreement
was found between only 1 pathologist and 1 dermatologist.
Excellent agreement was found between 3 observers (Table
2), which included 2 pathologists and 1 dermatologist. The
kappa values for these 3 observers for the distinction of brisk
vs nonbrisk vs absent ranged from 0.78 to 0.95. The majority
of observer pairs showed good agreement.
© American Society of Clinical Pathologists
Discussion
There are conflicting reports about the prognostic significance of lymphocytes associated with clinical stage I
primary cutaneous malignant melanoma.3,9,12,19-21 However,
comparison of results from various studies is difficult owing
to the lack of a standardized approach in the evaluation of
lymphocytic infiltrates.21 Many authors have failed to specify
in their analysis the extent of the infiltrate or its presence
within or around the tumor. Some investigators have
Am J Clin Pathol 2001;115:856-860
857
Busam et al / HISTOLOGIC CLASSIFICATION OF TUMOR-INFILTRATING LYMPHOCYTES IN MALIGNANT MELANOMA
A
B
❚Image 1❚ “Brisk” tumor-infiltrating lymphocytes. A, A dense
population of lymphocytes is interposed between melanoma
cells (H&E, ×200). B, The entire invasive component is
diffusely infiltrated by lymphocytes (H&E, ×40). C, The
advancing portion of the invasive melanoma is completely
surrounded by a dense band-like lymphocytic infiltrate (H&E,
×40).
C
❚Table 1❚
Agreement in Assessment of Tumor-Infiltrating Lymphocytes*
Brisk vs nonbrisk vs
absent
Brisk vs (nonbrisk and
absent)
*
Overall
Pathologists
Dermatologists
0.60
0.71
0.63
0.67
0.77
0.74
Significance of kappa values is as follows: > 0.75, excellent agreement; 0.6-0.75,
good agreement; 0.4-0.59, fair agreement; and <0.4, poor agreement.
compared the lymphocytic response at the base of the
melanoma with that at the tumor’s periphery or the lymphoid
response in RGP with that of the VGP.19-21 Others have
distinguished tumor-infiltrating from noninfiltrating lymphocytes.2,3,9,12,22
Many have found little, if any, prognostic significance
of lymphocytes associated with primary cutaneous
858
Am J Clin Pathol 2001;115:856-860
melanoma.18-20 Those, however, who have analyzed TILs
according to the method of Clark et al3 have persistently
found that prognosis of patients whose melanoma is involved
by a so-called brisk lymphocytic infiltrate was significantly
better than for patients whose tumors had little or no inflammation.3,12 In the study by Clark et al,3 the 8-year survival
rate for patients with melanoma was 88.5% when TILs were
brisk, 75% when TILs were nonbrisk, and 59.3% when TILs
were absent. In the study by Clemente et al,12 the 5-year
survival rates for patients with melanomas stratified
according to TILs were as follows: 77% for brisk, 53% for
nonbrisk, and 37% for absent. Even the presence of a brisk
infiltrate of lymphocytes involving metastatic melanoma
deposits in lymph nodes has been suggested to be associated
with longer survival.23
Our results clearly demonstrate that there is overall good
agreement among several observers in the assessment of
© American Society of Clinical Pathologists
Anatomic Pathology / ORIGINAL ARTICLE
❚Image 2❚ “Absent” tumor-infiltrating lymphocytes. No
lymphocytes infiltrate the vertical growth phase of this
melanoma (H&E, ×40).
❚Image 3❚ “Nonbrisk” tumor-infiltrating lymphocytes.
Lymphocytes infiltrate invasive melanoma only focally (H&E,
×100).
TILs. Except for 1 dermatopathologist, none of the participants was familiar with the classification of TILs. Furthermore, 3 of the participants were dermatologists who do not
routinely review melanocytic lesions. The overall good interobserver agreement among these observers indicates that the
criteria for classification of TILs can be easily taught and
readily learned by nonexperts in the field. Two observers
reached excellent agreement (95%). One of them had never
practiced assessment of TILs before.
In teaching the classification by Clark et al3 of TILs, it
became apparent that the choice of terms for the variations in
patterns of TILs was suboptimal. The meaning of the terms
brisk and nonbrisk is not intuitive and does not express a
morphologic pattern, which is why it needs to be described
and illustrated repeatedly. Terms like diffuse or focal would
be preferable. Their meaning is evident and needs little clarification. A revision of the terminology used by Clark et al3 to
clearer language would likely facilitate reproducible evaluation of TILs in melanoma.
Our results show that TILs can be assessed with defined
criteria and minor educational efforts as reliably as other
histologic parameters of primary cutaneous melanoma. In a
❚Table 2❚
Interobserver Agreement in Assessment of Tumor-Infiltrating Lymphocytes*
Brisk vs Nonbrisk vs Absent
Observer Pairs
1 vs 2
1 vs 3
2 vs 3
2 vs 4
2 vs 5
2 vs 6
3 vs 4
3 vs 5
3 vs 6
1 vs 4
1 vs 5
1 vs 6
4 vs 5
4 vs 6
5 vs 6
Concordance (%)
95†
71†
69†
78†
74†
82†
66†
48‡
50‡
81†
77†
86†
70†
75†
80†
Brisk vs (Nonbrisk and Absent)
Kappa (wt)
Concordance (%)
0.95
0.65†
0.60†
0.71†
0.65†
0.78‡
0.60†
0.38†
0.46†
0.77‡
0.70‡
0.83
0.66‡
0.70‡
0.75‡
90†
79†
71†
71†
61†
58‡
73†
58‡
38
79†
68†
66†
87†
63†
76†
Kappa (wt)
0.89
0.77
0.67‡
0.67‡
0.55†
0.57‡
0.73
0.57
0.38
0.77
0.63‡
0.66
0.86
0.63
0.74
*
The classification by observer 1 served as the “correct” answer. Observers 1-3 are pathologists; 4-6, dermatologists.
Significant at P < .01.
‡ Significant at P < .05.
†
© American Society of Clinical Pathologists
Am J Clin Pathol 2001;115:856-860
859
Busam et al / HISTOLOGIC CLASSIFICATION OF TUMOR-INFILTRATING LYMPHOCYTES IN MALIGNANT MELANOMA
study of interobserver agreement on melanoma, the kappa
values of the overall level of agreement among pathologists
were 0.71 for measurement of Breslow thickness and 0.56
for the assessment of growth phase.23
The latter observation that the pathologists’ ability to
discriminate between the so-called radial and vertical growth
phases is less than perfect may limit the potential of TILs as
a histologic prognostic parameter. TILs have been shown to
be of predictive value only in series, which limited their
analyses to melanomas said to be in VGP.3,9,12 This issue
needs to be considered for future studies designed to reevaluate the prognostic significance of TILs. Inclusion of
melanomas in RGP would likely dilute the predictive value
of TILs, as it would for other prognostic parameters, at least
in thin tumors.
Our results revealed a high level of agreement for TILs
among observers. This augurs well for the potential use of
TILs as a prognostic indicator. If TILs prove to be a robust
prognostic parameter, they can likely be assessed accurately,
which may contribute to our attempts to better predict the
outcome for patients with malignant melanoma.
From the 1Department of Pathology, 2Dermatology Service of the
Department of Medicine, and 3Department of Epidemiology and
Biostatistics, Memorial Sloan-Kettering Cancer Center, New York,
NY.
Address reprint requests to Dr Busam: Dept of Pathology,
Memorial Sloan-Kettering Cancer Center, 1275 York Ave, New
York, NY 10021.
References
1. Breslow A. Thickness, cross-sectional areas, and depth of
invasion in the prognosis of cutaneous melanoma. Ann Surg.
1970;172:902-908.
2. Elder DE, Guerry D IV, VanHorn M, et al. The role of lymph
node dissection for clinical stage I malignant melanoma of
intermediate thickness (1.51-3.99 mm). Cancer. 1985;56:413421.
3. Clark WH Jr, Elder DE, Guerry DG IV, et al. A model
predicting survival in stage I melanoma based upon tumor
progression. J Natl Cancer Inst. 1989;81:1893-1904.
4. Balch CM, Murad TM, Soong S, et al. A multifactorial
analysis of melanoma: prognostic histopathological features
comparing Clark’s and Breslow’s staging methods. Ann Surg.
1978;188:732-742.
5. Balch CM, Soong SJ, Shaw HM, et al. An analysis of prognostic factors of 8500 patients with cutaneous melanoma. In:
Balch C, Houghton A, Milton G, eds. Cutaneous Melanoma.
2nd ed. Philadelphia, PA: Lippincott; 1992:165-187.
6. Cook MG, Clark TJ, Humphreys S, et al. The evaluation of
diagnostic and prognostic criteria and the terminology of this
cutaneous malignant melanoma by the CRC melanoma
pathology panel. Histopathology. 1996;28:497-512.
860
Am J Clin Pathol 2001;115:856-860
7. Barnhill RL, Fine JA, Roush GC, et al. Predicting five-year
outcome for patients with cutaneous melanoma in a
population-based study. Cancer. 1996;78:427-432.
8. Buttner P, Garbe C, Bertz J, et al. Primary cutaneous
melanoma: optimized cutoff points of tumor thickness and
importance of Clark’s level for prognostic classification.
Cancer. 1995;75:2499-2506.
9. Tuthill RJ, Unger JM, Liu PY, et al. Risk assessment in
localized primary cutaneous melanoma: a Southwest
Oncology Group Study evaluating nine factors and a test of
the Clark Logistic Regression Prediction Model [abstract].
Mod Pathol. 2000;13:69A.
10. Fleming ID, Cooper JS, Henson DE, et al (American Joint
Committee on Cancer), eds. Manual for Staging of Cancer. 5th
ed. Philadelphia, PA: Lippincott-Raven; 1997:163-167.
11. Balch CM, Buzaid AC, Atkins MB, et al. A new American
Joint Committee on Cancer Staging System for Cutaneous
Melanoma. Cancer. 2000;88:1484-1491.
12. Clemente CG, Mihm MC Jr, Bufalino R, et al. Prognostic
value of tumor-infiltrating lymphocytes in the vertical growth
phase of primary cutaneous melanoma. Cancer. 1996;77:13031310.
13. Mihm MC Jr, Clemente CG, Cascinelli N. Tumor infiltrating
lymphocytes in lymph node melanoma metastases: a
histopathologic prognostic indicator and an expression of
local immune response. Lab Invest. 1996;74:43-47.
14. Cochran AJ, Bailly C, Cook M, et al. Recommendations for
the reporting of tissues removed as part of the surgical
treatment of cutaneous melanoma. Am J Clin Pathol.
1998;110:719-722.
15. Sober AM, Chuang T-Y, Duvic M, et al. American Academy
of Dermatology: guidelines of care for primary cutaneous
melanoma. Dermatol World. May 2000:2-14.
16. Barnhill RL. Prognostic factors in cutaneous melanoma. In:
Barnhill RL, ed. Pathology of Melanocytic Nevi and Malignant
Melanoma. Boston, MA: Butterworth-Heinemann; 1995:269284.
17. Fleiss JL. Statistical Methods for Rates and Proportions. 2nd ed.
New York, NY: John Wiley & Sons; 1981.
18. Cohen J. Weighted kappa: nominal scale agreement with
provision for scaled disagreement or practical credit. Psychol
Bull. 1968;70:213.
19. McGovern VJ, Shaw HM, Milton GW, et al. Lymphocytic
infiltration and survival in malignant melanoma. In:
Ackerman AB, ed. Pathology of Malignant Melanoma. New
York, NY: Mason Publishing; 1981:341-344.
20. Hansen MG, McCarten AB. Tumor thickness and lymphocytic infiltration in malignant melanoma of the head and
neck. Am J Surg. 1974;128:557-561.
21. Larsen TE, Grude TH. A retrospective analysis of 669 cases of
primary cutaneous malignant melanoma in clinical stage I.
Acta Pathol Microbiol Scand. 1978;86:523-530.
22. Barnhill RL, Mihm MC Jr. Histopathology and precursor
lesions. In: Balch CM, Houghton AN, Sober AJ, et al, eds.
Cutaneous Melanoma. 3rd ed. St Louis, MO: Quality Medical
Publishing; 1998:103-133.
23. McDermott NC, Hayes DP, Al-Sader MH, et al.
Identification of vertical growth phase in malignant
melanoma: a study of interobserver agreement. Am J Clin
Pathol. 1998;110:753-757.
© American Society of Clinical Pathologists