Tumor Mitotic Rate Is a More Powerful Prognostic Indicator than Ulceration in Patients with Primary Cutaneous Melanoma

Transcription

1 1488 Tumor Mitotic Rate Is a More Powerful Prognostic Indicator than Ulceration in Patients with Primary Cutaneous Melanoma An Analysis of 3661 Patients from a Single Center Manuela F. Azzola, M.D. 1 Helen M. Shaw, Ph.D. 1,2 John F. Thompson, M.D. 1,2 Seng-jaw Soong, Ph.D. 3 Richard A. Scolyer, M.B.B.S. 4 Geoffrey F. Watson, M.B.B.S. 4 Marjorie H. Colman, B.Sc. 1 Yuting Zhang, M.Sc. 3 1 Sydney Melanoma Unit, Royal Prince Alfred Hospital, Camperdown, New South Wales, Australia. 2 Department of Surgery, University of Sydney, Sydney, New South Wales, Australia. 3 Department of Biostatistics, University of Alabama at Birmingham, Birmingham, Alabama. 4 Department of Anatomical Pathology, Royal Prince Alfred Hospital, Camperdown, New South Wales, Australia. BACKGROUND. The current study was performed to determine whether tumor mitotic rate (TMR) is a useful, independent prognostic factor in patients with localized cutaneous melanoma. METHODS. From the Sydney Melanoma Unit database, 3661 patients with complete clinical information and details of primary tumor thickness, ulcerative state, and TMR were studied. TMR was expressed as mitoses per mm 2 in the dermal part of the tumor in which most mitoses were seen, as recommended in the 1982 revision of the 1972 Sydney classification of malignant melanoma. To determine which was the more prognostically useful method of grouping TMR, two separate methods (A and B) were used. Factors predicting melanoma-specific survival were analyzed using the Cox proportional hazards regression model. RESULTS. Patients with a TMR of 0 mitoses/mm 2 had a significantly better survival than those with 1 mitosis/mm 2 (P ) but no significant survival differences were recorded for the stepwise increases from 1 2, 2 3, 3 4, and 4 5/mm 2. Tumor thickness, ulceration, and TMR were closely correlated, whether TMR was grouped using Method A (0, 1 4, 5 10, and 11 mitoses/mm 2 ) or Method B (0 1, 2 4, and 5 mitoses/mm 2 ). However, Cox regression analysis indicated that the TMR was a highly significant independent prognostic factor, particularly when grouped according to Method A, in which it was second only to tumor thickness as the most powerful predictor of survival (P ). CONCLUSIONS. TMR is an important independent predictor of survival for melanoma patients. If confirmed by studies from other centers, it has the potential to further improve the accuracy of melanoma staging, as well as to define more rigidly the risk categories for patients entering clinical trials. Cancer 2003;97: American Cancer Society. DOI /cncr KEYWORDS: primary melanoma, tumor mitotic rate, thickness, ulceration, prognosis, staging. The financial support of the Melanoma Foundation of The University of Sydney and The Melanoma and Skin Cancer Research Institute is gratefully acknowledged. Address for reprints: Helen M. Shaw, Ph.D., Sydney Melanoma Unit, Royal Prince Alfred Hospital, Missenden Road, Camperdown, New South Wales, 2050, Australia; Fax: ; helen@mel.rpa.cs.nsw.gov.au Received June 11, 2002; revision received October 21, 2002; accepted October 24, The prognosis for a patient who presents with a primary cutaneous melanoma depends on a number of variables, principally tumor thickness but to a lesser extent on other features of the primary tumor such as ulceration, Clark level and anatomic site, and the patient s age and gender. The relative importance of these prognostic variables has been the subject of many studies over the past 25 years using progressively more sophisticated statistical techniques. Methods of analyzing multivariate data with the logistic model were introduced by Walker and Duncan in In 1970, Cox 2 further illustrated how this model could be used and subsequently 2003 American Cancer Society

2 Mitotic Rate and Prognosis in Melanoma/Azzola et al introduced the proportional hazards regression model in Beginning in the late 1970s and continuing into the 1980s, both models were used to study outcomes in patients with melanoma, particularly to assess the importance of a range of prognostic factors. These models helped, to varying extents, to clarify the relative roles and interdependencies of these various factors. Nevertheless, in the opinion of Vollmer in his seminal review of this subject published in 1989, 4 these models had at that time failed to provide optimal understanding of prognosis for patients with this disease. These reservations are still applicable. Indeed, when the recent literature on this topic is perused, one is struck by the remarkable discrepancies between centers that exist as to which variables have independent predictive value. Although it is apparent that tumor thickness is the principal determinant of prognosis for localized cutaneous melanoma in nearly all large studies, there is a failure of other factors to have consistent independent value, mainly because of their secondary correlation with tumor thickness. 4 Using the Cox proportional hazards regression model, Balch et al. 5 analyzed factors predicting melanoma-specific survival among 17,600 patients with cutaneous melanoma to validate proposed major revisions to the American Joint Committee on Cancer (AJCC) staging system. It was found that in the T category, tumor thickness and ulceration were the most powerful predictors of survival. Based on this large analysis, both these features were incorporated into the new staging system. 6 However, the impact of tumor mitotic rate (TMR) was not assessed in that study. Mitotic rate has been variously reported as either having or not having independent prognostic significance, but no major study to date has considered in detail its complex interdependency on tumor thickness and ulceration. Accordingly, the records of a large cohort of patients with invasive cutaneous melanoma treated at the Sydney Melanoma Unit (SMU) were analyzed to evaluate the possible independent prognostic importance of TMR, mindful of its close relationship with these other two dominant prognostic variables. The primary objective of this study was to determine whether TMR was a useful independent prognostic factor. This question has obvious relevance not only for individual patients, but for assignment of risk status in groups of patients entering clinical trials. MATERIALS AND METHODS The SMU database contains extensive clinical and pathologic details of nearly 20,000 patients treated for melanoma from 1960 to 2002 at Royal Prince Alfred, St. Vincents, and Sydney Hospitals in Sydney, Australia. First, the patients selected from this database for the current study were those who had been treated definitively (i.e., by a wide excision with either primary closure or a skin graft) from the outset at the SMU. Patients initially treated elsewhere and presenting to the SMU later in the course of their disease because of recurrence were excluded from this study to minimize referral bias. Second, patients had only one primary, invasive, cutaneous melanoma. Third, patients had localized disease (i.e., pathologic Stage I or II of the American Joint Committee on Cancer (AJCC)/International Union Against Cancer (UICC)pTNM pathologic classification). Finally, primary lesion pathology was reviewed by histopathologists from the Department of Anatomical Pathology at the Royal Prince Alfred Hospital, who were experienced in the diagnosis and reporting of melanocytic skin lesions. Pathologic details, assessed at the time of first definitive treatment of the patient s primary lesion, included the thickness, ulcerative state, and the TMR of the primary tumor, with the latter being expressed as the number of mitoses/mm 2 in the dermal area of the tumor in which most mitoses were noted. This method of assessing TMR was adopted after 1982, upon the recommendation of the participants of the 1982 Pathology Workshop 7 which revised the 1972 Sydney Classification of Malignant Melanoma. 8 In this original classification, the number of mitoses in at least 10 high power fields (HPF) over the entire lesion was counted and then expressed as the average number of mitoses/hpf. The 1982 Workshop participants considered that their revised method of assessing TMR, with results expressed as mitoses/mm 2, would be independent of the particular microscope and magnification used and therefore results obtained in different parts of the world would become more comparable. A total of 3661 patients satisfied the selection criteria outlined above. In patients with intermediate to high-risk lesions (i.e., lesions 1.5 mm in thickness), elective lymph node dissection usually was performed before Since then, a sentinel lymph node biopsy procedure has been offered to 645 patients with tumors thicker than 1.0 mm in the context of various clinical trials. We specify the criteria used for the three key pathologic variables with their codings, codings for four other putative prognostic variables, and methods of statistical analysis. Tumor Thickness This was measured according to the method of Breslow 9 as the vertical distance between the upper level of the granular layer of the epidermis and the deepest part of the tumor. Deeper tumors growing laterally from hair follicles were ignored. When ulcer-

3 1490 CANCER March 15, 2003 / Volume 97 / Number 6 ation was present, the thickness of the tumor was measured to the level of the base of the surface exudate. Thickness groupings were those recently recommended for the revised staging system of the AJCC 6 : Group 1, 1.0 mm or less; Group 2, mm; Group 3, mm; and Group 4, 4.00 mm or greater. Ulceration Ulceration in melanomas is not a homogenous phenomenon, with some types of ulceration having more prognostic significance than others. Three types of ulceration are recognized: infiltrative (erosive), attenuative (tumorigenic), and traumatic. The first two types of ulceration have been recognized by our pathologists to be prognostically significant. This is in contrast to traumatic ulceration, which has been generally accepted as prognostically irrelevant since being first described as such in 1953 by Thompkins 10 and by Peterson et al. 11 in Infiltrative ulceration is seen in melanomas displaying epidermal invasion. Although epidermal invasion is seen in melanomas of the superficial spreading histogenetic type, it is also commonly seen in the central region of melanomas of other histogenetic types, particularly as they progress. The invading cells permeate the epidermal cells, disrupting spinous layer cohesion, to eventually result in ulceration. Attenuative ulceration tends to occur in large nodular melanomas that lack epidermal invasion, but their expansile growth stretches the overlying epidermis, thinning and eventually ulcerating it. Ulceration was coded as 0, absent; or 1, present. Traumatic ulceration was coded as 0. Mitotic Rate To accept a mitotic figure, hairy extensions of chromatin (condensed chromosomes) had to be unequivocally present, extending from a condensed chromatin mass, and had to correspond to either a metaphase or a telophase figure. 12 The method used to count the TMR was as follows: All hematoxylin and eosin 5 mthick histologic sections of each specimen were evaluated to determine the dermal area of the tumor in which the TMR was greatest. Generally, the TMR was more prevalent at the invading front (usually the deep border) of the tumor. The number of mitoses was counted in a 1-mm 2 area (approximately 5 HPF), beginning in the field with the greatest numbers of mitoses and counting in successive fields, as recommended by the 1982 Pathology Workshop. At the time the study was planned, it was not clear which was the most valid and appropriate method of determining the prognostic significance of the TMR. Therefore, the TMR was first analyzed as a continuous variable without grouping, then by forming two separate groups. Method A The original 1972 Sydney Classification of Malignant Melanoma recommended that TMR be expressed as the average number of mitoses in at least 10 HPF over the entire lesion and that tumors be classified as having Grade I TMR when there was less than one mitosis per five HPFs, Grade II when there were one to four mitoses per five HPF, and Grade III when there were five or more mitoses per five HPF. Although these original groupings were used in the current study, the original method for counting TMR was not employed. Rather, TMR was determined using the recommendations of the 1982 Pathology Workshop. Five HPF were regarded as approximately equivalent to 1 mm 2. Coding using Method A was 1: 0 mitoses/mm 2 ; 2: 1 4 mitoses/mm 2 ;3:5 10 mitoses/mm 2 ; and 4: 11 or more mitoses/mm 2. Note that in the original recommendations, Groups 3 and 4 were combined. Method B According to the 1982 revision of the 1972 Sydney Classification, coding was 1: 0 1 mitoses/mm 2 ;2:2 4 mitoses/mm 2 ; and 3: 5 or more mitoses/mm 2. The grouping 0 1/mm 2 was recommended by the two histopathologists (R.A.S. and G.F.W.) because they considered that identification of the presence or absence of a single mitotic figure would depend on the number of sections (tumor area) of each tumor being evaluated. This varied considerably between cases. Other Putative Prognostic Variables Age was coded in deciles as 1: years; 2: years;...7: years; and 8: 80 years or older. Gender was coded as 0: male; and 1: female. Anatomic site was coded as 0: extremities; and 1: trunk, head and neck. Finally, Clark s level was coded as 2 5. Although details of tumor thickness, ulceration, TMR, age, gender, and anatomic site were known for each of the 3661 patients, a few patients ( 3%) had missing data concerning Clark level of invasion. However, because this study was designed specifically to evaluate the interrelationships among the three former dominant pathologic factors, there was negligible systematic bias due to incomplete data. Statistical Analyses The procedures for statistical analysis used in the current study were the same as those used in the recent study undertaken to validate the proposed revision of the AJCC Melanoma Staging System. 5 These were primarily based on the methods of survival data analysis. Survival times were calculated from the primary mel-

4 Mitotic Rate and Prognosis in Melanoma/Azzola et al FIGURE 1. Survival curves of 2871 patients with localized, cutaneous melanoma stratified by tumor mitoses/mm 2. anoma diagnosis date and considered censored for patients who were alive at the time of last follow-up or who died without evidence of melanoma. Melanomaspecific survival curves were generated according to the Kaplan Meier product-limit method and were compared using the log-rank test. Multivariate analyses of prognostic factors were based on the Cox proportional hazards model and the relative importance of prognostic factors was measured by the chi-square values, based on the Wald test of the coefficient associated with each prognostic factor in the Cox model. RESULTS In the current study, 386 patients had a positive result (uncensored in the Cox analysis), making the ratio of uncensored to total patients Since 1982, the mean follow-up for the uncensored patients has been 4.3 years. Figure 1 demonstrates a decline in the proportion of patients surviving according to a stepwise increase in the number of tumor mitoses/mm 2.Itis interesting to note that patients with tumors recorded as having 0 mitoses/mm 2 had significantly better survival than those with 1 mitosis/mm 2. (P ) However, when the TMR was analyzed as a continuous variable without grouping, no significant differences in survival were observed in the stepwise increases from 1 2/mm 2,2 3/mm 2,3 4/mm 2,or4 5/ mm 2. Mitotic Rate: Method A Figure 2 shows the influence of TMR on the proportion of patients surviving, irrespective of tumor thickness or ulcerative state. There is a highly significant difference in survival between each of the TMR groupings. Figure 3 shows the 10-year survival rates for patients according to tumor thickness and TMR for either nonulcerated (Fig. 3A) or ulcerated (Fig. 3B) tumors. There is a less clearly defined relation among survival rate, tumor thickness, and TMR in ulcerated, thick tumors than in nonulcerated, thinner ones. The Cox regression analysis was performed by including TMR either as a continuous variable without grouping or as a grouped variable. We found that the predictability of the model with TMR included as a grouped variable was far better than that of the model with TMR treated as a continuous variable. The Cox regression analysis for these 3661 patients (Table 1) indicates that TMR was the second most important independent prognostic factor. This is followed by the age of the patient, ulcerative state of the tumor, gender, and to a lesser extent, anatomic site of the primary lesion. Clark level did not reach statistical significance as an independent prognostic factor. Therefore, older patients, patients with ulcerated melanomas, male patients, and patients with melanoma located on the trunk or head and neck had worse prognoses than younger patients, patients with nonulcerated melanomas, female patients, and patients with melanomas located on the extremities, respectively. Mitotic Rate: Method B Figure 4 shows the influence of TMR on the proportion of patients surviving, irrespective of tumor thickness or its ulcerative state. Once again, there is a highly significant difference between each of the TMR groupings. Figure 5 shows the 10-year survival rates for patients according to tumor thickness and TMR in patients with nonulcerated (Fig. 5A) and ulcerated (Fig. 5B) tumors. There is a less clearly defined relation

5 1492 CANCER March 15, 2003 / Volume 97 / Number 6 FIGURE 2. Survival curves of 3661 patients with localized, cutaneous melanoma when tumor mitoses/mm 2 were grouped into four categories (Method A). between tumor thickness and TMR in ulcerated, thick tumors than in nonulcerated, thinner ones. The Cox regression analysis shown in Table 2 indicates that the age of the patient was the second most important independent prognostic factor followed by the ulcerative state of the tumor, TMR, gender, and, to a lesser extent, anatomic site of primary lesion. Clark level did not attain statistical significance as an independent prognostic factor. DISCUSSION The results of the current study have shown that TMR is a powerful independent prognostic factor. If TMR is grouped according to the method recommended by the international committee (Method A), it will be the second most powerful predictor of prognosis (even more predictive than ulceration). If the alternative grouping (Method B) as suggested by our two histopathologists (R.A.S. and G.F.W.) is used, it is still a powerful prognostic predictor but marginally less significant than age of the patient or ulceration. These differing findings may be accounted for by the observation that patients with 0 mitoses/mm 2 fared significantly better that those with 1 mitosis/mm 2 (Fig. 1). Therefore, grouping patients with tumors having 0 mitosis/mm 2 with those having 1 mitosis/mm 2 greatly reduced the significance of TMR as a prognostic variable. This emphasizes the importance of Vollmer s observation 4 that selection (and a clear statement of) groupings is critical. It is noteworthy that the methods used to assess TMR and ulceration in the current study differed slightly from the 1972 recommendations. As we have pointed out earlier, the 1972 consensus suggested that TMR be determined by taking an average number of mitoses over at least 10 HPF without endeavoring to find the area with the highest TMR. In addition, there was a tacit assumption that there was no wide variation in the area of a HPF in the equipment that was used. This has been shown to be incorrect, with TMR expressed as the number of mitoses/10 HPF varying by as much as 600%, depending on the microscope used. 13 The policy of histopathologists from the Department of Anatomical Pathology at the Royal Prince Alfred Hospital is to count the total number of mitoses/mm 2 in the dermal area of the tumor with the highest TMR. This latter method is in accordance with the recommendations of the 1982 Pathology Workshop. We believe that this method is more accurate and scientifically valid and it is particularly useful in the assessment of cases in which only rare (or a single) dermal mitosis is present in the entire specimen. Using the 1972 classification, such cases would defy subgrouping, as the average TMR may be greater than 0 but less than 1 per 10 HPF. Furthermore, this 1982 consensus method is similar to that recommended for the counting of TMR in other solid tumors. 12 With regard to ulceration, traumatic ulceration is often related to the anatomic site or to the architectural configuration of the tumor (i.e., whether it is polypoid or pedunculated) and is generally considered to be prognostically irrelevant unless it is seen in association with another pattern (most commonly attenuative with traumatic). Whether there is a prognostic difference between infiltrative and attenuative ulceration is unknown, although infiltrative ulceration may be a marker of the tumor s infiltrative capacity and the latter a marker of tumor bulk in the dermis. To clarify this uncertainty in prognostic potential, it

6 Mitotic Rate and Prognosis in Melanoma/Azzola et al FIGURE 3. (A) Ten-year survival rates for 2835 patients with nonulcerated, localized, cutaneous melanoma according to tumor thickness and tumor mitoses/ mm 2, the latter being grouped into 4 categories (Method A). (B) Ten-year survival rates for 826 patients with ulcerated, localized, cutaneous melanoma according to tumor thickness and tumor mitoses/mm 2, the latter being grouped into 4 categories (Method A). would be useful for histopathologists to document the ulcerative pattern in their reports. In 1984, attempts were made to document interobserver variation in the classification of various melanoma parameters by specialist melanoma histopathologists in Western Australia. 14 When TMR was assessed by counting the average number of mitoses/ HPF, congruence levels were low (i.e., 66% for lesions with 1 mitosis/5 HPF, 36% for those with 1 5 mitoses/5 HPF, and 57% for those with 5 mitoses/5hpf). The authors concluded that the more objective method of counting TMR as mitoses/mm 2, as recommended by the 1982 Pathology Workshop, should be more reproducible. Therefore, further studies are needed to assess the objectivity and reproducibility in the determination of TMR. Whatever method is used, the counting of TMR relies on the diligence of the assessment by the reporting histopathologist and the number of sections evaluated (particularly if only small numbers of mitoses are present). Despite these TABLE 1 Cox Regression Analysis for 3661 Melanoma Patients with Localized Disease According to Method A a Variable DF Chi-square value (Wald) P value Risk ratio 95% CL Thickness Mitotic rate b Age Ulceration Gender Site Level DF: degree of freedom; CL: confidence limits. a Coding: Thickness was coded as 1: 1.00 mm; 2: mm; 3: mm; and 4: 4.00 mm. Mitotic rate was coded as 1: 0/mm 2 ;2:1 4/mm 2 ;3:5 10/mm 2 ; and 4: 11/mm 2. Age was coded as deciles1:10 19; 2: years;...8: 80 years. Ulceration was coded as 0: absent; and 1: present. Gender was coded as 0: male; and 1: female. Site was coded as 0: extremities; and 1: trunk and head and neck. Level was coded as 2, 3, 4, and 5. b Stratified into four categories: 0, 1 4, 5 10, and 11/mm 2.

7 1494 CANCER March 15, 2003 / Volume 97 / Number 6 FIGURE 4. Survival curves of 3661 patients with localized, cutaneous melanoma when tumor mitoses/mm 2 were grouped into 3 categories (Method B). limitations, our data show that TMR is a powerful prognostic indicator, even in patients with a low TMR. In the Western Australian study, the congruence level for ulceration (without regard to its pattern) was considerably higher than for TMR (i.e., 81% for the presence of ulceration and 96% for its absence). In the current study, Clark level of invasion failed to have significant prognostic impact by multivariate analysis after adjusting for TMR. This is in contrast to the results of the multivariate analysis for the AJCC Melanoma Staging System, 5 in which Clark level emerged as an independent prognostic factor, particularly for melanomas with a thickness of 1.0 mm or less. In the current study, we attempted to replicate the AJCC findings on the prognostic value of Clark level for patients with melanomas of this thickness (n 1259). Patients with Clark Level II melanomas (n 325) were found to have a significantly better survival rate than patients with Level III (n 652) or Level IV (n 276) melanomas, but there was no significant survival difference for patients with Level III or IV melanomas. (The sample size for Level V melanomas [n 4] was too small for an adequate comparison.) The AJCC study had a much larger number of patients (n 5480) with tumor thickness 1.0 mm or less but TMR was not assessed in the AJCC database. We believe it would be useful to include this variable in the AJCC melanoma database for the next revision of the AJCC Melanoma Staging System. A previous attempt to use SMU data to delineate the prognostic significance of various histologic factors did not demonstrate that TMR was an independent prognostic factor. 15 A close correlation was found between TMR and tumor thickness. However, when attempts were made to compensate for this correlation, no clear effect of TMR could be discerned, particularly for thin lesions. Several reasons could account for this previous failure to demonstrate an independent significance of TMR. First, TMR was assessed according to the 1972 recommendations, i.e., recording the average number of mitoses per 10 HPF over the entire lesion. Second, only 694 patients were available for analysis, resulting in some groups being too small to provide meaningful prognostic information. Third, no attempt was made to eliminate referral bias. Fourth, 5-year, not 10-year, survival rates were used as end points. Finally, multivariate statistical analysis using the Cox proportional hazards model was not employed. In a number of other studies, TMR has also failed to emerge as a useful prognostic variable. However, the majority of these studies have also had serious methodologic deficiencies and their conclusions therefore must be interpreted with similar caution. In 1989, Vollmer 4 was the first to attempt to explain in detail the differing conclusions gleaned from these analyses. He extracted from the literature 54 multivariate studies reported in 48 articles from 18 centers. He summarized the results of 31 multivariate studies that included TMR. Fourteen of these showed that TMR did have some prognostic significance, the remainder failing to assign prognostic significance to TMR. Vollmer considered that, in general, the composition of a dataset was the most important factor to be considered when assessing the validity of a study. He suggested that knowledge of which patients were deleted was critical because these deletions could produce biases, some subtle, some major. Such deletions are usually due to a simple lack of data and great caution in the interpretation of results is required if

8 Mitotic Rate and Prognosis in Melanoma/Azzola et al FIGURE 5. (A) Ten-year survival rates for 2835 patients with nonulcerated, localized, cutaneous melanoma according to tumor thickness and tumor mitoses/ mm 2, the latter being grouped into 3 categories (Method B). (B) Ten-year survival rates for 826 patients with ulcerated, localized, cutaneous melanoma according to tumor thickness and tumor mitoses/mm 2, the latter being grouped into 3 categories (Method B). more than 10% of patients are excluded. Referral patterns can also produce bias, so that prognostic conclusions may relate either to melanoma patients in general or to patients with more complex problems who reach a specialized clinic. Although Vollmer drew attention to these critical issues, few subsequent studies have attempted to address them. A more detailed consideration of Vollmer s analysis is instructive. He noted that the 54 studies used 12 different end points of interest, which could be separated into two major categories: binary or time-related categories. Binary events have only two possible outcomes (i.e., the patient did or did not develop a recurrence or alternatively did or did not die of melanoma). A time-related end point is time to failure, expressed as a numeric value (weeks, months, or years) whether it be a local, lymph nodal, or distant recurrence, or ultimately, death. All these end points require long term follow-up to be of any practical value. Because at TABLE 2 Cox Regression Analysis for 3661 Melanoma Patients with Localized Disease According to Method B a Variable DF Chi-square value (Wald) P value Risk ratio 95% CL Thickness Age Ulceration Mitotic rate b Gender Site Level DF: degree of freedom; CL: confidence limits. a Coding: Thickness was coded as 1: 1.00 mm; 2: mm; 3: mm; and 4: 4.00 mm. Mitotic rate was coded as 1: 0 1/mm 2 ;2:2 4/mm 2 ; and 3: 5mm 2. Age was coded as deciles 1: 10 19; 2: 20 29,...8, 80 years. Ulceration was coded as 0: absent; and 1: present. Gender was coded as 0: male; and 1: female. Site was coded as 0: extremities; and 1: trunk and head and neck. Level was coded as 2, 3, 4, and 5. b Stratified into three categories: 0 1, 2 4, 5/mm 2.

9 1496 CANCER March 15, 2003 / Volume 97 / Number 6 TABLE 3 Summary of Recent Multivariate Studies on Mitosis Studies No. of patients Mean follow-up (yrs) Factors studied TMR cutoff End point Significance of TMR Vossaert et al , 1, 2, 3/mm 2 10-yr survival MacKie et al , 1 5, 5/10HPF 5-yr survival Barnhill et al , 0.1 6, 6/mm 2 5-yr survival Ostmeier et al (median) 16 3, 3/mm 2 First recurrence Wagner et al , 3 5, 5/HPF SNB status Massi et al , 0.1 6, 6/mm 2, 10, 10/mm 2 10-yr survival Schmidt-Wendtner et al PI, mod PI 10-yr survival TMR: tumor mitotic rate; HPF: high power field; SNB: sentinel lymph node biopsy; PI: prognostic index (tumor thickness TMR); modified PI: tumor thickness 2 TMR. least 60% of patients who present with Stage I and II melanoma survive for more than 10 years, studies should ideally be long term and large enough to define accurately the required end point of interest. However, most of the 54 studies reviewed by Vollmer were neither long term nor large. For example, a follow-up interval was reported for only 30 of the 54 studies. For those studies in which it was reported, the mean was about 5 years. The number of patients studied ranged from an unacceptably low number of 44 to 3445 (average 529). The majority of studies (i.e., 47 of 54 studies) used Cox proportional hazards survival analysis. The remainder used logistic regression or multivariate discrimination analyses. Only 21 of the 54 studies reported the number of uncensored patients (patients with a positive result) and the proportion of uncensored to total patients ranged from a low of 3% to 66%. The number of factors incorporated into the Cox analysis ranged from 4 to 25 and often key factors such as ulceration and particularly TMR were omitted, the latter being analyzed in only 31 of the 54 studies. Most important, the method of coding for the analyses was not specified or was only suggested in 52 of the 54 studies and the tumor thickness break points used were inconsistent. Vollmer considered that the selection of break points for an otherwise continuous variable was artificial and arbitrary and he favored a continuous function of thickness. With regard to TMR, some studies used a prognostic index (PI) in their analyses, defined as tumor thickness multiplied by TMR. This, therefore, becomes a continuous function of thickness. It is a second-order factor and because TMR is related to tumor thickness, it is nearly like multiplying tumor thickness by itself. It is noteworthy that Vollmer suggested that TMR might need further study using standardized methods of coding and reporting such as the number of mitoses/mm 2. Even simple binary factors such as ulceration were coded differently in the studies reviewed. Sometimes they were recorded as present or absent, but were coded by others as or 3 mm in greatest dimension. Similarly, the anatomic site of the primary lesion was recorded variously as extremity versus axial (i.e., as a binary factor) or as head and neck versus trunk versus extremity versus acral (i.e., as a four-level factor). Vollmer considered that the age of the patient should probably be regarded as a continuous parameter and suggested that any subdivision would be arbitrary. Since Vollmer s seminal review of this subject was published in 1989, several other studies have attempted to assess the prognostic importance of TMR (Table 3). Six of the seven analyses attached biologic importance and prognostic significance to TMR. Unfortunately, however, some of these studies failed to avoid all the pitfalls previously highlighted by Vollmer, making it difficult to draw firm conclusions from them. In its initial analysis of intermediate thickness melanomas ( mm) in 1987, the New York University Melanoma Cooperative Group 23 found that TMR had no prognostic significance although ulceration achieved independent significance (P 0.015). However, when a PI was derived (which was the product of tumor thickness and TMR), a significant cutoff was achieved at PI less than 19 versus PI greater than or equal to 19. This group concluded that significant predictability should be contributed to Breslow thickness by multiplying it by TMR and suggested that mitoses/mm 2 may be a correlate of the rate of growth of a subgroup of melanoma cases which in turn is related directly to the metastatic potential of melanoma. In a later New York University analysis by Vossaert et al. 16 (Table 3), which included 832 melanomas of all thicknesses, TMR itself again failed to reach prognostic significance although ulceration strengthened its significance (P ). The results of the population-based Scottish Melanoma Group study 17 (Table 3) indicated that TMR, considered in isolation (i.e., in a univariate analysis),

10 Mitotic Rate and Prognosis in Melanoma/Azzola et al was the third most important prognostic factor after ulceration and tumor thickness. However, when corrected for ulceration, TMR fell to seventh place (behind gender, anatomic site, Clark level, and age). Therefore, the authors stressed the folly of considering features of putative prognostic significance in isolation, i.e., by univariate analysis. Surprisingly, they found ulceration to be a far more significant independent factor than tumor thickness (chi-square test 188 vs. 52, respectively) in their relatively small (n 289) study. This, together with their poor 5-year survival rate (30% less than that reported by Soong et al. 24 ) led them to conclude that there may be geographic variations in prognostic factors due, for example, to extremes in ultraviolet radiation. This suggested that there might be a need for each center to derive its own model. Barnhill et al. 18 (Table 3) conducted the first comprehensive population-based (Connecticut Tumor Registry) analysis of prognostic factors in the U.S. They evaluated the significance of TMR in 548 patients using the cutoff points proposed by Clark et al. 25 These were 0, 0.1 6, and 6 mitoses/mm 2. They found that only two variables (tumor thickness and TMR) had independent effects. Ulceration of the lesion failed to reach significance as a prognostic indicator, as did age, gender, and anatomic site. Their reasons for these discrepancies between multifactorial analyses were similar to those proposed by Vollmer. However, they added that for histologic parameters, interobserver variation due to subjectivity may be just as important, with perhaps some effect of geographic variability. They concluded that more objective methods of assessing proliferation may provide better indications of prognosis. A German study of 691 patients by Ostmeier et al. 19 (Table 3) found that ulceration no longer demonstrated independent significance when TMR was included in their prognostic model. They cited reasons similar to those proposed by Barnhill et al. 18 for conflicting results from multivariate analyses. Using a different approach to assess the prognostic significance of TMR, Wagner et al. 20 (Table 3), from the Indiana University Interdisciplinary Melanoma Program, tested the ability of this parameter to predict lymph node positivity after sentinel lymph node biopsy in 279 patients. They found that TMR and ulceration achieved similar levels of prognostic significance (P 0.04 and P 0.03, respectively). On the basis of these results, they suggested performance of sentinel lymph node biopsy where ulceration is present and TMR is high, irrespective of tumor thickness. However, a major concern about this study is that the TMR groupings used were extraordinarily high (i.e., low: 3 mitoses/hpf; intermediate: 3 5 mitoses/hpf; and high: 5 mitoses/hpf). These groupings approximately correspond to less than 15 mitoses/mm 2, mitoses/mm 2, and greater than 25 mitoses/ mm 2. In their small study of 140 patients with thick tumors ( 3 mm) from Florence, Massi et al. 21 (Table 3) found that the cutoff point of 10 mitoses/mm 2 versus 10 mitoses/mm 2 proved to be significantly predictive of 10-year survival rates, with rates of 56% and 24%, respectively (P ). Ulcerative state also was found to be significantly predictive but less so (60% 10-year survival for patients with nonulcerated lesions vs. 40% for those with ulcerated lesions; P 0.006). They also concluded that TMR, as a continuous variable, acts as an independent risk factor. For practical purposes, the categorization of continuous prognostic factors (e.g., tumor thickness and TMR) is always needed. Their explanations for the differences in prognostic significance of TMR cited in the literature included variations in recording this feature (particularly the scoring system used), the number of fields evaluated, and the quality of the histologic material available for analysis. In what to our knowledge is the largest study published to date, from the University of Munich by Schmidt-Wendtner et al. 22 (Table 3), TMR was specifically analyzed both as a PI and a modified PI, the latter being the product of the tumor thickness squared and the number of mitoses/mm 2. In their 2715 patient series, they found modified PI to be a slightly more accurate determinant of prognosis than the PI as proposed by Schmoeckel and Braun-Falco. 26 The authors stated that using the square of tumor thickness (which approximately represents the surface of the tumor) may result in greater prognostic predictability. In our view, such an assumption most likely is incorrect because it does not reflect our experience in reviewing the histopathology of large numbers of primary melanomas. A major deficiency in this study is that ulceration was not documented from the beginning and, therefore, was not part of their analysis. No account was taken of the probable interrelation among tumor thickness, TMR, and ulceration. Some centers attributed importance to PI. Nevertheless, the critical values derived for PI vary considerably among centers. For example, values ranged from less than 12 versus 12 in the Schmoeckel and Braun-Falco study, 26 to 19 versus 19 in the New York University Melanoma Cooperative Group study. 23 In the current study, we attempted to minimize the pitfalls inherent in multifactorial analyses so that the informed reader may assess the validity of our findings on the prognostic significance of TMR in a primary cutaneous melanoma. Less than 3% of SMU

11 1498 CANCER March 15, 2003 / Volume 97 / Number 6 patients were excluded because variables germaine to this analysis were missing. Any referral bias was minimized by selecting only patients treated from the outset at the SMU, although we cannot infer from this that the final study dataset was truly population based. Large numbers of patients were studied with sufficient follow-up for an adequate proportion of them to provide valid estimates of survival. The number of uncensored patients was high and coding for all key prognostic factors was stipulated clearly. Using these rigorous methods to study TMR and its complex interrelationships with tumor thickness and ulceration, a powerful and clear-cut prognostic significance of TMR was found. We believe that this is the largest and most comprehensive study of its kind published to date. Therefore, this finding has important practical implications. If confirmed by similarly vigorous studies from other centers, it should allow more accurate prognostic information to be determined for individual patients with primary melanomas and more precise stratification of groups of patients entering clinical trials of adjuvant therapy. REFERENCES 1. Walker SH, Duncan DB. Estimation of the probability of an event as a function of several independent variables. Biometrika. 1967;54: Cox DR. The analysis of binary data. London: Chapman and Hall, Cox DR. Regression model and life tables (with discussion). J Stat Soc [B]. 1972;34: Vollmer RT. Malignant melanoma. A multivariate analysis of prognostic factors. Pathol Annu. 1989;24: Balch CM, Soong S-J, Gershenwald JE, et al. Prognostic factors analysis of 17,600 melanoma patients: validation of the American Joint Committee on Cancer Melanoma Staging System. J Clin Oncol. 2001;19: Balch CM, Buzaid AC, Soong S-J, et al. Final version of the American Joint Committee on Cancer Staging System for Cutaneous Melanoma. J Clin Oncol. 2001;19: McGovern VJ, Cochran AJ, Van Der Esch EP, Little JH, MacLennan R. The classification of malignant melanoma, its histological reporting and registration: a revision of the 1972 classification. Pathology. 1986;18: McGovern VJ, Mihm MC, Bailly C, et al. The classification of malignant melanoma and its histologic reporting. Cancer. 1973;32: Breslow A. Thickness, cross sectional areas and depth of invasion in the prognosis of cutaneous melanoma. Ann Surg. 1970;172: Thompkins VN. Cutaneous melanoma: ulceration as a poor prognostic sign. Cancer. 1953; 6: Petersen NC, Bodenham DC, Lloyd OC. Malignant melanoma of the skin: a study of the origin, development, aetiology, spread, treatment and prognosis. Br J Plast Surg. 1962;15: Hilsenbeck S, Allred D. Improved methods of estimating mitotic figures in solid tumours. Hum Pathol. 1990;21: Ellis PSJ, Whitehead R. Mitotis counting a need for reappraisal. Hum Pathol. 1981;12: Heenan PJ, Matz LR, Blackwell JB, et al. Inter-observer variation between pathologists in the classification of cutaneous malignant melanoma in Western Australia. Histopathology. 1984;8: McGovern VJ, Shaw HM, Milton GW, Farago GA. Prognostic significance of the histological features of malignant melanoma. Histopathology. 1979;3: Vossaert KA, Silverman MK, Kopf AW, et al. Influence of gender on survival in patients with stage I malignant melanoma. J Am Acad Dermatol. 1992;26: MacKie RM, Aitchison T, Sirel JM, McLaren K, Watt DC. Prognostic models for subgroups of melanoma patients from the Scottish Melanoma Group database , and their subsequent validation. Br J Cancer. 1995;71: Barnhill RL, Fine JA, Roush GC, Berwick M. Predicting fiveyear outcome for patients with cutaneous melanoma in a population-based study. Cancer. 1996;78: Ostmeier H, Fuchs B, Otto F, et al. Can immunohistochemical markers and mitotic rate improve prognostic precision in patients with primary melanoma? Cancer. 1999;85: Wagner JD, Gordon MS, Chuang T-Y, et al. Predicting sentinel and residual lymph node basin disease after sentinel lymph node biopsy for melanoma. Cancer. 2000;89: Massi D, Borgognoni L, Franchi A, Martini L, Reali UM, Santucci M. Thick cutaneous malignant melanoma: a reappraisal of prognostic factors. Melanoma Res. 2000;10: Schmidt-Wendtner MH, Baumert J, Schmidt M, Plewig G, Volkenandt M, Holzel D. Prognostic index for cutaneous melanoma: an analysis after follow-up of 2715 patients. Melanoma Res. 2001;11: Kopf AW, Gross DF, Rogers GS, et al. Prognostic index for malignant melanoma. Cancer. 1987;59: Soong S-J, Shaw HM, Balch CM, McCarthy WH, Urist MM, Lee JY. Predicting recurrence and survival in localized melanoma: a multivariate approach. World J Surg. 1992;16: Clark WH, Elder DE, DuPont G, et al. Model predicting survival in stage I melanoma based on tumor progression. J Natl Cancer Inst. 1989;81: Schmoeckel C, Braun-Falco O. Prognostic index in malignant melanoma. Arch Dermatol. 1978;114: