SEARCH

SEARCH BY CITATION

Summary

  1. Top of page
  2. Summary
  3. Patients and methods
  4. Results
  5. Discussion
  6. Acknowledgments
  7. References

Background  Survival and prognostic factors for thin melanomas have been studied relatively little in population-based settings. This patient group accounts for the majority of melanomas diagnosed in western countries today, and better prognostic information is needed.

Objectives  The aim of this study was to use established prognostic factors such as ulceration, tumour thickness and Clark’s level of invasion for risk stratification of T1 cutaneous melanoma.

Methods  From 1990 to 2008, the Swedish Melanoma Register included 97% of all melanomas diagnosed in Sweden. Altogether, 13 026 patients with T1 melanomas in clinical stage I were used for estimating melanoma-specific 10- and 15-year mortality rates. The Cox regression model was used for further survival analysis on 11 165 patients with complete data.

Results  Ulceration, tumour thickness and Clark’s level of invasion all showed significant, independent, long-term prognostic information. By combining these factors the patients could be subdivided into three risk groups: a low-risk group (67·9% of T1 cases) with a 10-year melanoma-specific mortality rate of 1·5% (1·2–1·9%); an intermediate-risk group (28·6% of T1 cases) with a 10-year mortality rate of 6·1% (5·0–7·3%); and a high-risk group (3·5% of T1 cases) with a 10-year mortality rate of 15·6% (11·2–21·4%). The high- and intermediate-risk groups accounted for 66% of melanoma deaths within T1.

Conclusions  Using a population-based melanoma register, and combining ulceration, tumour thickness and Clark’s level of invasion, three distinct prognostic subgroups were identified.

Since 1990, the Swedish Melanoma Study Group has prospectively gathered data concerning diagnosis, treatment and follow-up on all Swedish cases of cutaneous malignant melanoma (CMM). The data are collected and registered at the six regional oncology centres in Sweden, and transferred to the national database, the Swedish Melanoma Register (SMR), in Linköping. A first report on CMM in Sweden (1990–1999) has been published.1 As a majority of melanomas diagnosed in many countries are thin (≤ 1 mm), there is motivation for further studies concerning this large subgroup of patients with melanoma. In the American Joint Committee on Cancer (AJCC) 2009 report, the mitotic index was added as a prognostic factor,2 and research into other new factors is in progress.3,4 However, there are a few studies from Germany, Australia and the U.S.A.4–10 that indicate that the previous AJCC 2002 melanoma staging11 could be further refined for thin melanomas by simply combining the traditional prognostic factors in alternative ways. Due to its size and being population based, the SMR is ideal when exploring new risk stratifications in thin melanomas. The aim of this study was to use established prognostic factors such as ulceration, tumour thickness and Clark’s level of invasion for risk stratification of T1 cutaneous melanoma.

Patients and methods

  1. Top of page
  2. Summary
  3. Patients and methods
  4. Results
  5. Discussion
  6. Acknowledgments
  7. References

Between 1990 and 2008, 30 157 patients with primary CMMs were reported to the SMR. The mean annual coverage rate was 97% (range 93–99%) compared with the Swedish National Cancer Register, to which reporting of all new malignancies is mandated by Swedish law.12 The current registration is population based and nationwide. All cases are registered with a unique patient identification number, which facilitates record linking.

Of the 30 157 patients reported, 13 488 (44·7%) had T2–T4 tumours and 1183 (3·9%) had no associated information about tumour thickness; both sets were both excluded. The remaining 15 486 patients (51·4%) had T1 melanoma (≤ 1 mm thick) as their first invasive diagnosis. Patients with previous CMMs diagnosed before 1990 were identified from the Swedish Cancer Register and were excluded (231 patients). From the T1 patient group, 2229 patients were excluded because they were either not in clinical stage I at diagnosis (127 patients) or because information about clinical stage at diagnosis (2102 patients) was lacking, as these data were not regularly registered in one region. Data on the remaining 13 026 patients in clinical stage I were used in a univariate Kaplan–Meier analysis of melanoma-specific 5-, 10- and 15-year mortality rates. Further, a Cox analysis for prognostic factors was undertaken on data from 11 165 patients aged 15–89 years having types of melanoma other than acral lentiginous melanoma (ALM), and with the tumour not subungual or localized on the palm. The ages 15–89 were chosen because children with melanoma seem to have different outcomes from adults,13 and the cause of death of patients aged 90+ could be uncertain due to comorbidities. ALM-type melanomas and those that were subungual or localized to the palm were excluded in order to avoid potential errors in the measure of tumour thickness.

The following clinical variables were studied: age at diagnosis, sex and tumour site. Histopathological characteristics of the primary tumour, including tumour thickness, Clark’s level of invasion, presence or absence of ulceration and histogenetic type were registered. Histopathological data were reported directly from the pathology department’s database. Data on cause of death until 31 December 2009 were obtained from the Swedish Cause of Death Register, and were used to calculate the melanoma-specific mortality rate. Patients having a second CMM during the study period were censored at the second diagnosis.

The excision margin used was 1 cm, except for a small cohort of CMMs with tumour thickness ≥ 0·8 mm, operated on between 1990 and 1991, and with 2-cm or 5-cm excision margins due to an ongoing study at that time.14 According to the Swedish National Guidelines for CMM, sentinel node biopsy was not recommended for melanomas ≤ 1 mm thick before 2007. However, sentinel node biopsies were performed in 91 cases, with positive results in five. Shave biopsies and other similar procedures are not recommended by the Swedish National Guidelines for CMM.

Statistical analysis

Survival time was calculated from the date of diagnosis (date of diagnostic biopsy or date of pathology report) to the date of the event or to the date of censoring. In the analysis, death from melanoma is considered to be the primary event. Censoring was made at the time point of: the patient emigrating, the patient being diagnosed with a second CMM during the study period, the patient dying from causes other than melanoma, or end of the follow-up. Melanoma-specific mortality rates and confidence intervals (CIs) were estimated using the method of Kaplan and Meier.15 Cox’s proportional hazard regression was used to assess the independent prognostic contribution of clinical variables, either alone, or after adjustment for other clinical variables. The prognostic impacts of clinical variables were expressed as hazard ratios (HRs) with 95% CIs.

To compare multivariate Cox models, the likelihood ratios and degrees of freedom (df) between models were evaluated with χ2-tests. Statistical significance was indicated by P-values < 0·05. In order to simplify the Cox analysis, some categories that were not significantly different were merged. All potential interactions between different variables were investigated.

The results from the Cox analyses combined with the results of melanoma-specific mortality rates from Kaplan–Meier analysis were used to identify prognostic subgroups. In the text and the tables, 95% CIs are shown in parentheses. All statistical analyses were performed using SAS v9·2 (SAS Institute Inc., Cary, NC, U.S.A.).

Results

  1. Top of page
  2. Summary
  3. Patients and methods
  4. Results
  5. Discussion
  6. Acknowledgments
  7. References

During a median follow-up time of 6·6 years (range 0–20 years), 311 patients (2·4%) died from their melanoma. Within the follow-up period, 1731 patients (13·3%) died of causes other than melanoma. There were 314 patients (2·4%) with a second melanoma diagnosis within the study period. An additional 27 patients (0·2%) were censored due to emigration. The median age at diagnosis was 58 years, the median melanoma thickness was 0·6 mm and 55% of the patients were women.

Melanoma-specific mortality rate data from the SMR are presented in Table 1. The overall 10- and 15-year melanoma-specific mortality rates were 3·4% (3·0–3·8%) and 4·4% (3·9–5·0%), respectively. The melanoma-specific mortality rates for the very thin melanomas (≤ 0·5 mm) compared with the thickest (> 0·75 mm) varied from 1·9% (1·0–3·7%) to 6·6% (5·6–7·8%) at 10 years, and from 3·2% (1·6–6·4%) to 8·9% (7·5–10·6%) at 15 years.

Table 1. Melanoma-specific mortality rates (1990–2008)
Variable n Number of melanoma deaths (%)Mortality rate, % (95% CI)
5 years10 years15 years
  1. CI, confidence interval; SSM, superficial spreading melanoma; LMM, lentigo maligna melanoma; NM, nodular melanoma; ALM, acral lentiginous melanoma; AJCC, American Joint Committee on Cancer.

All patients13 026311 (2·4)1·7 (1·4–1·9)3·4 (3·0–3·8)4·4 (3·9–5·0)
Sex
 Male5890157 (2·7)1·9 (1·5–2·4)3·8 (3·2–4·6)5·1 (4·1–6·2)
 Female7136154 (2·2)1·3 (1·0–1·7)2·9 (2·4–3·6)3·8 (3·1–4·6)
Age group
 0–44328862 (1·9)1·1 (0·8–1·6)2·3 (1·7–3·1)3·1 (2·4–4·0)
 45–645233125 (2·4)1·5 (1·2–1·9)3·1 (2·6–3·8)4·5 (3·6–5·5)
 65–79339991 (2·7)1·9 (1·4–2·5)4·4 (3·5–5·5)5·6 (4·4–7·2)
 ≥ 80110633 (3·0)4·1 (2·8–5·9)5·2 (3·7–7·5)5·2 (3·7–7·5)
Tumour site
 Head/neck140148 (3·4)3·5 (2·5–4·9)5·5 (4·0–7·6)6·7 (4·8–9·2)
 Upper extremities213740 (1·9)1·3 (0·8–1·9)2·6 (1·8–3·7)3·7 (2·6–5·3)
 Lower extremities302153 (1·8)1·0 (0·7–1·5)2·5 (1·8–3·3)3·1 (2·3–4·1)
 Trunk6336159 (2·5)1·6 (1·3–2·0)3·5 (2·9–4·1)4·7 (3·9–5·6)
 Palm/subungual9410 (10·6)7·3 (3·3–15·6)11·1 (5·6–21·4)20·2 (9·4–40·3)
 Missing371 (2·7)3·1 (0·4–20·2)3·1 (0·4–20·2)3·1 (0·4–20·2)
Histogenetic type
 SSM10 256217 (2·1)1·4 (1·1–1·6)3·0 (2·6–3·5)3·9 (3·4–4·6)
 LMM109421 (1·9)1·6 (0·9–2·8)2·4 (1·4–4·1)4·8 (2·7–8·6)
 NM40325 (6·2)5·4 (3·5–8·5)8·5 (5·7–12·7)9·3 (6·2–13·8)
 ALM907 (7·8)3·7 (1·2–11·0)7·4 (3·1–17·4)16·2 (6·8–36·0)
 Other types110136 (3·3)2·7 (1·7–4·0)5·1 (3·5–7·2)6·0 (4·2–8·7)
 Missing825 (6·1)5·4 (2·1–13·9)5·4 (2·1–13·9)5·4 (2·1–13·9)
Tumour thickness, mm
 ≤ 0·2575311 (1·5)0·8 (0·3–2·0)1·9 (1·0–3·7)3·2 (1·6–6·4)
 0·26–0·50508059 (1·2)0·9 (0·7–1·3)1·4 (1·1–1·9)2·0 (1·4–2·7)
 0·51–0·75356575 (2·1)1·3 (0·9–1·7)3·1 (2·4–3·9)3·8 (3·0–4·9)
> 0·753628166 (4·6)3·2 (2·6–4·0)6·6 (5·6–7·8)8·9 (7·5–10·6)
Clark’s level of invasion
 II683688 (1·3)0·8 (0·6–1·1)1·6 (1·2–2·0)2·5 (1·9–3·2)
 III5022164 (3·3)2·3 (1·9–2·9)4·9 (4·2–5·8)6·4 (5·3–7·6)
 IV–V103653 (5·1)3·7 (2·6–5·2)7·2 (5·4–9·6)8·5 (6·4–11·4)
 Missing1326 (4·5)4·5 (1·9–10·4)6·0 (2·7–13·3)6·0 (2·7–13·3)
Ulceration
 No10 948226 (2·1)1·4 (1·2–1·7)3·0 (2·6–3·4)4·0 (3·5–4·7)
 Yes55149 (8·9)7·2 (5·2–10·1)13·0 (9·7–17·2)13·6 (10·2–17·9)
 Missing152736 (2·4)1·6 (1·0–2·4)2·6 (1·8–3·9)4·1 (2·8–6·2)
T-stage (AJCC 2002)
 T1a10 019186 (1·9)1·2 (1·0–1·5)2·6 (2·2–3·1)3·7 (3·1–4·3)
 T1b150889 (5·9)4·5 (3·4–5·8)8·5 (6·9–10·6)9·6 (7·7–12·0)
 Unclassifiable149936 (2·4)1·6 (1·1–2·5)2·9 (2·0–4·2)4·1 (2·7–6·0)

In the univariate Cox analysis of mortality, the three most dominant factors were tumour thickness (χ2 = 103·2, df = 2, P < 0·0001), ulceration (χ2 = 83·2, df = 1, P < 0·0001) and Clark’s level of invasion (χ2 = 74·6, df = 2, P < 0·0001) (Table 2). The largest effect estimates were observed for melanoma thickness > 0·75 vs. ≤ 0·5 mm (HR 4·7; 95% CI 3·4–6·5), ulceration vs. nonulceration (HR 4·5; 95% CI 3·3–6·2) and Clark level IV–V vs. Clark level II (HR 4·4; 95% CI 3·0–6·3).

Table 2. Univariate and multivariate Cox analysis, 1990–2008 (= 11 165)
Variable n Number of melanoma deaths (%)UnivariateMultivariate
HR (95% CI)χ2 P HR (95% CI)χ2 P
  1. HR, hazard ratio; CI, confidence interval; SSM, superficial spreading melanoma; LMM, lentigo maligna melanoma; NM, nodular melanoma.

Sex
 Male5078133 (2·6)1·007·50·0061·003·00·09
 Female6087120 (2·0)0·71 (0·55–0·91)0·79 (0·61–1·03)
Age group
 15–44286755 (1·9)1·0015·8< 0·0011·008·60·01
 45–644492101 (2·2)1·30 (0·94–1·81)1·19 (0·86–1·66)
 65–89380697 (2·5)1·91 (1·37–2·66)1·63 (1·15–2·30)
Tumour site
 Extremities442882 (1·9)1·0013·80·0011·0010·90·004
 Head/neck115937 (3·2)2·09 (1·41–3·08)2·03 (1·33–3·09)
 Trunk5578134 (2·4)1·28 (0·98–1·69)1·28 (0·96–1·71)
Histogenetic type
 SSM9058185 (2·0)1·0028·8< 0·00011·008·70·03
 LMM90918 (2·0)1·18 (0·73–1·92)0·91 (0·53–1·56)
 NM33621 (6·3)3·06 (1·95–4·80)1·64 (1·04–2·60)
 Other types/missing86229 (3·4)1·77 (1·19–2·61)1·55 (1·05–2·30)
Tumour thickness, mm
 ≤ 0·5502352 (1·0)1·00103·2< 0·00011·0029·4< 0·0001
 0·51–0·75307260 (2·0)1·86 (1·28–2·69)1·49 (1·00–2·21)
 > 0·753070141 (4·6)4·71 (3·43–6·47)2·76 (1·86–4·08)
Clark’s level of invasion
 II601374 (1·2)1·0074·6< 0·00011·0012·30·002
 III4272134 (3·1)2·79 (2·10–3·71)1·59 (1·14–2·23)
 IV–V88045 (5·1)4·37 (3·02–6·33)2·13 (1·38–3·27)
Ulceration
 No10 637208 (2·0)1·0083·2< 0·00011·0046·7< 0·0001
 Yes52845 (8·5)4·48 (3·25–6·19)3·16 (2·27–4·39)

In the multivariate Cox analysis, ulceration (χ2 = 46·7, df = 1, P < 0·0001), tumour thickness (χ2 = 29·4, df = 2, P < 0·0001) and Clark’s level of invasion (χ2 = 12·3, df = 2, = 0·002) were still the most important prognostic factors (Table 2). The highest HRs were observed for ulceration vs. nonulceration (HR 3·2; 95% CI 2·3–4·4), melanoma thickness > 0·75 vs. ≤ 0·5 mm (HR 2·8; 95% CI 1·9–4·1) and Clark level IV–V vs. Clark level II (HR 2·1; 95% CI 1·4–3·3). In a further multivariate Cox analysis of all potential interaction effects (data not shown), there was a significant interaction between tumour thickness and ulceration (χ2 = 6·9, df = 2, P = 0·03). This interaction indicates that tumour thickness was a less important prognostic factor when the melanoma was ulcerated (HR 0·8 for > 0·75 vs. ≤ 0·5 mm; 95% CI 0·3–1·9) compared with nonulcerated melanomas (HR 3·5 for > 0·75 vs. ≤ 0·5 mm; 95% CI 2·3–5·4).

The multivariate analysis confirmed the great influence of ulceration, tumour thickness and Clark’s level of invasion on melanoma-specific mortality. Thus, we estimated the 10-year melanoma-specific mortality rate in each patient group, defined according to these three tumour characteristics (Table 3).

Table 3. Ten-year melanoma-specific mortality rates by ulceration, Clark’s level of invasion and tumour thickness, 1990–2008 (= 11 165)
Clark level/tumour thicknessUlceration
NoYes
n (%)Mortality, % (95% CI) n (%)Mortality, % (95% CI)
  1. CI, confidence interval.

Clark level II
 ≤ 0·5 mm4072 (36·5)0·9 (0·6–1·4) 93 (0·8)7·0 (2·9–16·6)
 0·51–0·75 mm1302 (11·7)1·6 (0·9–2·9) 46 (0·4)5·8 (1·5–21·7)
 > 0·75 mm452 (4·1)5·0 (2·8–9·0) 48 (0·4)13·3 (3·5–43·6)
Clark level III
 ≤ 0·5 mm744 (6·6)2·0 (1·0–4·1) 26 (0·2)17·4 (4·2–57·2)
 0·51–0·75 mm1463 (13·1)3·0 (2·0–4·4) 73 (0·7)12·5 (5·7–26·2)
 > 0·75 mm1798 (16·1)6·2 (4·8–7·9) 168 (1·5)14·7 (8·8–23·9)
Clark level IV–V
 ≤ 0·5 mm78 (0·7)4·0 (1·0–16·0) 10 (0·1)11·1 (1·6–56·7)
 0·51–0·75 mm180 (1·6)6·5 (2·9–14·1) 8 (0·1)25·0 (6·9–68·5)
 > 0·75 mm548 (4·9)6·6 (4·4–10·0) 56 (0·5)24·0 (11·6–45·6)

Using the results presented in Table 3 we divided the T1 melanomas into three risk groups: (i) a low-risk group of nonulcerated, Clark level II–III, ≤ 0·75 mm melanomas (67·9% of T1 cases), with a 10-year melanoma-specific mortality rate of 1·5% (1·2–1·9%); (ii) an intermediate-risk group of melanomas that were nonulcerated and > 0·75 mm, or nonulcerated and Clark level IV–V, or ulcerated, Clark level II and ≤ 0·75 mm (28·6% of T1 cases), with a 10-year melanoma-specific mortality rate of 6·1% (5·0–7·3%); and (iii) a high-risk group consisting of all ulcerated T1 melanomas except Clark level II and ≤ 0·75 mm (3·5% of T1 cases), with a 10-year melanoma-specific mortality rate of 15·6% (11·2–21·4%). Figure 1 shows the cumulative, melanoma-specific mortality rates for the three risk groups.

image

Figure 1.  Cumulative melanoma-specific mortality by group. Low-risk group: nonulcerated, ≤ 0·75 mm and Clark II–III. Intermediate-risk group: nonulcerated and > 0·75 mm; or nonulcerated and Clark IV–V; or ulcerated, ≤ 0·75 mm and Clark II. High-risk group: ulcerated and Clark III–V; or ulcerated, > 0·75 mm and Clark II. CI, confidence interval.

Download figure to PowerPoint

Discussion

  1. Top of page
  2. Summary
  3. Patients and methods
  4. Results
  5. Discussion
  6. Acknowledgments
  7. References

The authors analysed all melanomas registered in the SMR between 1990 and 2008, a large unselected dataset representing the total Swedish population. The benefits of using large, population-based materials are evident from a review of published survival data concerning thin melanomas (Table 4). We have transformed and used these data as 10-year melanoma-specific mortality rates. The data from the current study (SMR), the Queensland Cancer Registry (QCR),5 Surveillance, Epidemiology, and End Results (SEER)6,7 and a New South Wales (NSW) study10 can be categorized as population based. Other cohorts are either multi- or single-institution data: AJCC 2002,11,16 AJCC 2009,2 the German Central Malignant Melanoma Register (CMMR),8 Pennsylvania Pigmented Lesion Group (PLG)6 and Sydney Melanoma Unit (SMU).10

Table 4. Ten-year melanoma-specific mortality rates from previously published studies
StudyPopulation-basedSingle-institutionalMulti-institutional
Current studyQCR5SEER6,7NSW10PLG6SMU10CMMR8AJCC 200211,16AJCC 20092
  1. QCR, Queensland Cancer Registry; SEER, Surveillance, Epidemiology, and End Results; NSW, New South Wales; PLG, Pennsylvania Pigmented Lesion Group; SMU, Sydney Melanoma Unit; CMMR, Central Malignant Melanoma Register; AJCC, American Joint Committee on Cancer. aAccording to AJCC 2009 classification. bCalculated value from T1a and T1b.

Period1990–20081982–20061998–20011983–981972–20011979–981976–20001976–20001976–2008
No. patients13 02626 73626 29118 0882389274612 728589011 841
Median age, years5852·7 (mean)5445·950 (mean)
Median thickness, mm0·600·46 (mean)0·500·53 (mean)0·600·57
% T1a (AJCC 2002)87%90%84%77%80%a
All T13·4%2·6%3·3%b3·6%3·5%b7·3%3·5%13·0%b8·0%
T1a (AJCC 2002)2·6%2·6%2·8%12·1%7–8%a
T1b (AJCC 2002)8·5%9·8%7·4%16·9%12–13%a
≤ 0·75 mm1·9–3·1%1·3–2·5%2·3–2·7%2·3–3·6%
> 0·75 mm6·6%6·8%7%7%8%6·1%
Nonulcerated3·0%7·7%3·5%14%5%
Ulcerated13·0%16%4·1%24%12%

The 10-year melanoma-specific mortality rates of T1a and T1b melanomas are much lower in the population-based datasets of SMR, QCR and SEER compared with the high mortality rates of compiled or institutional-based materials as reported in the AJCC 2002 and AJCC 2009 data (although using other criteria for T1a and T1b). Our findings thus confirm the findings of Gimotty et al.,7 who validated the AJCC 2002 classification and found that it was suitable for constructing stages, but was imperfectly calibrated to the U.S. population, as the survival rates in the AJCC material were much lower than in the U.S. population.7 This could have been due to the fact that large, specialized institutions contribute to the AJCC database, with a higher proportion of patients with poor prognosis, as patients with T1 melanomas and recurrence are referred to those institutions for further treatment. Similar conclusions were drawn in a European, population-based study from the Netherlands.17 The data from T1 melanomas in the population-based NSW study have an equally low mortality rate as the SMR data, in contrast to the institutional SMU data, which have high mortality rates that are similar to those of AJCC 2002. The latter fact is not surprising, considering that the SMU contributed 8667 of the 17 600 cases in the AJCC 2002 database.

SMR data indicate that the original thickness cut-off value (0·76 mm) defined by Breslow18 still seems to have a significant impact on the prognosis of T1 melanoma. Clark et al.19 demonstrated that almost all melanomas ≤ 0·75 mm thick or of Clark level II were in the radial growth phase, and that these melanomas rarely showed mitoses. They found no metastases in that group compared with melanomas with a vertical growth phase.19 A thickness of ≤ 0·75 mm and Clark level II might thus be a good approximation of the radial growth phase. The 10-year melanoma-specific mortality rates for melanomas in SMR, QCR, NSW, SEER, CMMR and PLG for melanomas > 0·75 mm are very similar (6·1–8·0%).

In all the reviewed studies except that of CMMR, ulceration is a poor prognostic sign. Unfortunately the large, population-based QCR had no information on ulceration.

To further validate our data, we divided the data into nine groups from the expanded AJCC described by Gimotty et al.6 (Table 5) Only minor differences were found in comparison with the SEER data: head and neck tumours in SEER have higher mortality rates than in the SMR (SEER groups 3–4), and also SEER finds sex differences that are not apparent in the SMR (SEER groups 5–6).

Table 5. Ten-year melanoma-specific mortality rates using expanded American Joint Committee on Cancer (AJCC) and Surveillance, Epidemiology, and End Results (SEER) data
Expanded AJCC groupMortality rate, %
Current studySEER
No ulceration, ≤ 0·78 mm, Clark level II, age < 60 years0·91·0
No ulceration, ≤ 0·78 mm, Clark level II, age ≥ 60 years1·42·5
No ulceration, ≤ 0·78 mm, Clark level III, other sites2·63·2
No ulceration, ≤ 0·78 mm, Clark level III, head/neck4·37·9
No ulceration, > 0·78 mm, Clark level III, women5·74·4
No ulceration, > 0·78 mm, Clark level III, men6·39·4
No ulceration, Clark level IV–V6·38·6
Ulceration, Clark level II–III11·411·1
Ulceration, Clark level IV–V22·430·2

We have identified three prognostic groups; a low-risk group (67·9% of T1 cases), an intermediate-risk group (28·6% of T1 cases) and a high-risk group (3·5% of T1 cases). The latter had a slightly higher mortality rate than the T2a tumours in a national report from the SMR, 1990–2008.20 Further subdivision of the SMR data would be possible but we chose to stay with our divisions, based on the most important factors from the multivariate analysis. We are aware that in the AJCC 2009 classification, mitoses have been added as an important prognostic factor,2 and other studies support this.4,6 A limitation in the present study is that we could not directly compare our results with AJCC 2009 data. From 2009, the SMR contains information on the presence or absence of mitosis, but before 2009 information on mitosis was not reported regularly. In the future, these data will be analysed against long-term mortality in order to determine whether adding mitosis as a prognostic factor can further improve risk stratification in this patient population.

In all the analyses from the SMR, patients were censored at the time point of a new melanoma diagnosis, thus eliminating the risk of a second melanoma being the underlying risk for cause of death from melanoma. However, this methodology was not used in the other studies. Nevertheless, omitting censoring at the second melanoma diagnosis changed the SMR results only marginally; the 10-year melanoma-specific mortality rate increased by 0·3% (from 3·4% to 3·7%) in the total study cohort, and there was no alteration regarding the outcome with respect to prognostic factors. In the SMR, eight cases of Clark level V melanoma were used in the multivariate analysis, and this may indicate that these patients could have had thicker melanomas than were actually registered. Only two of these were classified into the high-risk group, one of whom died from the melanoma. Excluding the cases with Clark level V did not change the results in the multivariate Cox analysis. Five cases with positive sentinel node studies were identified, but none of those patients was classified into the high-risk group. In the multivariate Cox analysis for T1, we tested two alternative ways of including tumour thickness. We tested whether the Cox model was improved by using four different tumour groups or by using a continuous thickness variable. Neither of these models added significant additional information (P-values 0·32 and 0·04, respectively) and we therefore chose to stay with the three-group model. Tumour thickness with three groups is also the most common way of dealing with this variable in other studies, and this facilitates comparisons.

It is important to be able to give patients with thin melanomas more accurate prognostic information. In Sweden thin melanomas constitute > 50% (even higher for women and younger patients) of all melanomas. The prognosis is very good on average compared with other stages of melanoma and other cancer types, but we found a mortality range of 1–25%, showing the importance of using prognostic factors. In the SMR, the high-risk group (3·5% of T1) accounts for 15% of all melanoma deaths, and the intermediate group (28·6% of T1) accounts for 51% of all melanoma deaths. Using the old AJCC 2002 classification on the SMR data, 12% of the patients were classified into T1b, accounting for 31% of melanoma deaths. Our findings thus predict 66% of all melanoma deaths compared with 31% using conventional T1b.

In conclusion, using a population-based melanoma register and combining ulceration, tumour thickness and Clark’s level of invasion, three distinct prognostic subgroups were identified. Meta-analyses of population-based melanoma registers may be a method to improve further the risk stratification in this patient group.

Acknowledgments

  1. Top of page
  2. Summary
  3. Patients and methods
  4. Results
  5. Discussion
  6. Acknowledgments
  7. References

The work was supported by the regional cancer centre Southeast in Linköping. Further contributing oncological centres were Southern region (Lund), Western region (Göteborg), Stockholm-Gotland region (Stockholm), Uppsala-Örebro region (Uppsala) and the Northern region (Umeå). We especially thank Professor John Carstensen, Monika Dufmats, Angela Musimiire and Professor Sture Falkmer for their valuable contributions.

References

  1. Top of page
  2. Summary
  3. Patients and methods
  4. Results
  5. Discussion
  6. Acknowledgments
  7. References