The long-term clinical course of patients with cutaneous melanoma


  • John W. Gamel M.D.,

    Corresponding author
    1. Department of Surgery, Veterans Administration Medical Center, Louisville, Kentucky
    2. Department of Ophthalmology and Visual Sciences, University of Louisville School of Medicine, Louisville, Kentucky
    • Department of Ophthalmology, University of Louisville School of Medicine, 301 East Muhammad Ali Boulevard, Louisville, KY 40202
    Search for more papers by this author
    • Fax: 502-852-7349

  • Stephen L. George Ph.D.,

    1. Department of Biostatistics, the Duke Comprehensive Cancer Center, Duke University Medical Center, Durham, North Carolina
    Search for more papers by this author
  • Michael J. Edwards M.D.,

    1. Department of Surgery, J. Graham Brown Cancer Center, University of Louisville School of Medicine, Louisville, Kentucky
    Search for more papers by this author
  • Hilliard F. Seigler M.D.

    1. Department of Surgery, Duke University Medical Center, Durham, North Carolina
    Search for more papers by this author



The clinical course of cutaneous melanoma is associated with pathologic and clinical factors, such as thickness, ulceration, and location of tumor and gender of the patient. The authors used a parametric survival model that incorporated a cured fraction of patients to translate these factors into specific estimates of long-term outcome.


A cohort study was conducted of 5837 patients who were treated for localized cutaneous melanoma between 1978 and 1990 at the Duke Comprehensive Cancer Center. Of these, 495 patients were excluded because the survival status or one or more of the prognostic factors was unknown. Maximum follow-up was 22 years. The primary outcome measures examined were cured fraction (probability of cure), median tumor specific survival (i.e., median time to death from tumor), and the probability of tumor-related survival at fixed intervals after treatment.


For an example of a class of patients with a relatively good prognosis, consider women with nonulcerated lesions measuring 0.5 mm thick on an extremity. The probability of cure (± standard error) for these patients was estimated at 80.8% ± 2.0%, and the median tumor specific survival was 10.0 years ± 0.8 years. This suggests that, in these patients, half of the deaths from melanoma will occur more than 10 years after treatment, barring death from other causes. Conversely, men with ulcerated lesions measuring 8.00 mm thick on the trunk have a relatively poor prognosis. The probability of cure for these patients was 16.8% ± 2.4%, and the median tumor specific survival was 2.7 years ± 0.2 years. Despite this poor initial prognosis, the conditional probability of cure increased to 90%; after 15 years of recurrence free survival.


Parametric statistical analysis provides quantitative measures of long-term survival. These measures show that late recurrence—longer than a decade after treatment—is to be expected in a significant portion of patients, although the probability of cure increases with progressively longer recurrence free survival. Cancer 2002;95:1286–93. © 2002 American Cancer Society.

DOI 10.1002/cncr.10813

Certain well-defined prognostic factors correlate highly with the biologic behavior of cutaneous melanoma. For example, one study showed that tumor thickness (i.e., total vertical height) was the most important factor for 8500 patients at the Universities of Alabama and Sydney.1 Other prognostic factors include ulceration,2–4 anatomic location of the primary lesion,5, 6 and gender.4, 7 An especially poor prognosis is found in males with thick, ulcerated lesions located on sites other than the extremities.

The objective of this article is to translate these known prognostic factors into specific values that measure the long-term clinical course of patients after they undergo treatment. To achieve this goal, we first examined the basic biologic parameters that govern response to treatment—the cured fraction (or probability of cure) and the median tumor specific survival among uncured patients. The role of these parameters is well illustrated by comparing pediatric patients who have solid tumors with adult patients who have prostate carcinoma, two specific but very different examples of human malignancies. Certain pediatric patients with solid tumors are cured readily by modern therapy; however, when therapy fails, death usually ensues rapidly. In contrast, adult patients seldom are cured of advanced prostate carcinoma, but they may live and function for many years, even decades, after diagnosis. With such contrasting outcomes, it is difficult to determine which of these lesions would be considered more malignant.

These examples demonstrate that both the probability of cure and the time to death among uncured patients are useful factors for characterizing the long-term clinical course of patients with cutaneous melanoma. It also is noteworthy that there is an independent correlation of these two parameters with various prognostic factors, such as ulceration, thickness, and location of the tumor as well as the gender of the patient. For such an analysis, we must use parametric survival models, a statistical method that represents the probability of cure as the asymptote to survival from the tumor under study. For clinical data, we examined a total of 5342 patients with a maximum follow-up of 22 years. Such extensive information provided exceptional insight into the long-term clinical course of patients with cutaneous melanoma. The parametric methods also allowed us to estimate the improvement in prognosis that occurs with continued recurrence free survival.


Patients were derived from an original data set of 5837 patients with localized cutaneous melanoma at the Duke Comprehensive Cancer Center. This set included all Caucasian patients with primary melanoma of the skin or mucous membrane (patients with intraocular tumors were excluded) who underwent surgical excision between 1978 and 1996, inclusive. During this interval, chemotherapy and immunotherapy were used rarely as primary therapy for patients with Stage I disease. Non-Caucasian patients were excluded from the data set because of the rarity of melanoma in this population and because of the possibility that such patients may follow a clinical course different from that of Caucasians. For this article, follow-up was terminated early in 2000; thus, the maximum possible duration was slightly more than 22 years. Of the 5837 available patients, 495 patients were excluded for the following reasons: thickness of lesion unknown (n = 223 patients), status or cause of death unknown (n = 103 patients), ulceration status of the lesion unknown (n = 127 patients), or a combination of two or more of the preceding reasons (n = 42 patients).

For the 5342 patients who were included in the analysis, the following information was available: gender of the patient (coded as 0 or 1), thickness of the primary lesion (in mm), duration of follow-up (in days), status of the patient (died of tumor, withdrawn alive, died of other causes), and the presence or absence of ulceration (coded as 0 or 1). The tumor site was classified initially as upper extremity, lower extremity, trunk, or other: The category other included subungunal lesions and lesions on the palm, sole, mucous membrane, or head and neck. Preliminary exploratory analyses revealed that there was no significant prognostic distinction between lesions on the upper and lower extremities. Similarly, there was no significant prognostic distinction between lesions on the trunk and those classified as other (data not shown). Thus, for the final analysis, site was coded as a single binary variable: 0 for extremity and 1 for nonextremity. A total of 1486 patients died of melanoma during available follow-up. It was assumed that the 216 deaths from causes other than cutaneous melanoma were causes independent of melanoma; thus, for the survival analysis, those patients were considered as withdrawn at the time of death.

The survival analysis was performed using a parametric model with the cured fraction incorporated as the asymptote to cause specific, melanoma-related survival. This model contained three parameters: ρ represented the cured fraction (i.e., that proportion of patients presumed cured by treatment), whereas μ and σ characterized the distribution of time to death from cutaneous melanoma.8 The underlying distribution can be represented in a variety of ways. We used log-normal distributions, so that μ and σ represented the mean and standard deviation of log survival, whereas exp{μ} represented the median tumor-related survival among uncured patients. We developed a multivariate extension of the Boag model that allowed the expression of ρ, μ, and σ as regression functions of prognostic covariates.9 The resulting parametric model was more effective than available nonparametric methods for measuring the impact of covariates on cured fraction.10 Software for deriving this model currently is being adapted for general use by the National Cancer Institute.9 It has been shown that the log-normal distribution provides a good fit to survival data for a variety of malignancies.11–13 Preliminary analysis revealed a skewing effect, such that the logarithm of (thickness [in mm] +1) provided a much better fit to observed data than thickness. Thus, this transform was used instead of thickness itself to estimate the impact of thickness on the parameters μ and σ.


Table 1 presents a summary of the follow-up data for the patients studied. Table 2 presents the distribution of patients by age and tumor thickness. Table 3 shows the statistically significant correlation among various clinical factors. Specifically, ulceration was associated with thick lesions, male gender, and sites other than an extremity. Men were somewhat older than women when they developed melanoma, and they were approximately twice as likely to develop lesions on the trunk and other nonextremity sites. Women were especially prone to lesions of the lower extremity.

Table 1. Follow-up of 3563 Patients with Skin Melanoma
Years of follow-upNo. of patients at riskPatients who died of melanomaPatients withdrawn from studyPatients who died of other causes
Total 534214863640216
Follow up (yrs)    
 Standard deviation3.295.105.08
Table 2. Summary of Clinical Findings
Finding (range)Patients
Age of patient (yrs)  
Total 11–925342100.0
Thickness of lesion (mm)  
 ≥ 4.0056810.6
Total 0.00–9.865342100.0
Table 3. Correlation of Clinical Factors
FactorNo. of patients (%)Mean valuesaPercenta
Thickness (mm)Age (yrs)NoYesMaleFemale
  • a

    Differences within all groups significant at P < .001 by analysis of variance (thickness and age) or chi-square test.

 Upper extremity904 (16.9)2.1047.383.116.945.254.8
 Lower extremity1044 (19.6)1.9545.979.420.624.375.7
 Trunk2336 (43.7)2.1145.
 Other1058 (19.8)2.4549.876.223.866.433.6
 Total5342 (100.0)2.1546.6    
 Male2945 (55.3)2.2747.976.423.6
 Female2388 (44.7)1.9945.080.719.3
 Total5342 (100.0)
 No4185 (78.3)1.8645.9
 Yes1157 (21.7)3.1848.9
 Total5342 (100.0)

Analysis revealed that the parameter σ was not associated significantly with any variable. The parameter ρ (cured fraction) was associated significantly (P < 0.001) with thickness, gender, and site, whereas the median tumor specific survival (exp{μ}) was associated significantly (P < 0.001) with thickness, site, and ulceration. Ulceration status, as a prognostic covariate, demonstrated marginal association with ρ (P = 0.06), whereas gender demonstrated inadequate association with μ (P = 0.4). Thus, in the results presented below, estimates of cured fraction were unaffected by ulcer status, whereas estimates of median tumor-related survival were unaffected by gender.

Figure 1 shows the fit of the log-normal model to observed tumor-related survival for patients subgrouped by the thickness of their lesion. Chi-square analysis revealed an adequate fit between the model and the data (P = 0.6).

Figure 1.

Survival from cutaneous melanoma as a function of thickness. Continuous lines represent the actuarial survival of patients with cutaneous melanoma. Solid symbols represent the log-normal parametric cure model based only on thickness stratified by interval, as shown in Table 2. The fit of the model to survival data was acceptable (P = 0.6; chi-square goodness-of-fit test).

Table 4 shows estimates of cured fraction and median tumor specific survival for patients with various risk factors. The table also shows two estimates of long-term clinical course after 5 years, 10 years, and 15 years of follow-up: tumor-related survival (predicted survival) and the conditional probability of cure (POC). Predicted survival represents the probability that a patient will survive their melanoma (note that these probabilities make no allowance for death from causes other than melanoma). The conditional POC represents the probability that a patient is cured of cutaneous melanoma, given that he or she has survived free of disease for a specified time.

Table 4. Clinical Implications of Log-Normal Analysis
Site thickness (mm)Cured fractionTumor-related median survivalPredicted survival (conditional POC)a
%SEYrsSE5 yrs10 yrs15 yrs
  • POC: probability of cure; SE: standard error.

  • a

    The conditional POC is the probability of cure given that the patient has survived for 5, 10, or 15 years without sign of recurrence.

No ulceration       
   0.5080. (84)90 (89)87 (93)
   1.0074. (80)86 (87)81 (92)
   2.0063. (72)77 (83)71 (89)
   3.0055. (66)69 (79)63 (88)
   4.0048. (61)63 (77)56 (86)
   6.0037. (54)52 (73)45 (85)
   8.0030. (48)44 (70)37 (83)
   0.5075.12.710.00.895 (79)88 (86)83 (90)
   1.0067. (74)82 (83)76 (89)
   2.0055. (65)72 (77)65 (86)
   3.0046. (58)64 (74)56 (84)
   4.0040. (53)57 (70)49 (82)
   6.0030. (45)46 (66)38 (80)
   8.0024. (40)39 (62)31 (78)
   0.5072. (79)84 (87)79 (92)
   1.0065. (74)77 (84)72 (91)
   2.0052. (65)66 (80)60 (89)
   3.0043. (59)57 (77)50 (87)
   4.0037. (54)50 (74)43 (86)
   6.0027. (47)39 (71)33 (84)
   8.0021. (43)32 (68)26 (83)
   0.5065. (73)79 (83)74 (89)
   1.0057. (67)72 (79)65 (87)
   2.0044. (58)60 (74)53 (85)
   3.0035. (51)51 (70)43 (83)
   4.0029. (46)44 (68)36 (82)
   6.0021. (39)34 (63)27 (80)
   8.0016. (35)28 (60)21 (78)
With ulceration       
   0.5080. (87)87 (93)84 (96)
   1.0074. (83)82 (91)78 (95)
   2.0063. (77)72 (89)68 (94)
   3.0055. (73)63 (87)59 (94)
   4.0048. (69)56 (86)52 (93)
   6.0037. (63)45 (84)41 (92)
   8.0030. (59)37 (82)33 (92)
   0.5075. (82)83 (90)80 (94)
   1.0067. (78)77 (88)73 (93)
   2.0055. (71)66 (85)61 (92)
   3.0046. (66)57 (83)51 (91)
   4.0040. (61)49 (81)44 (91)
   6.0030. (55)39 (78)34 (90)
   8.0024. (50)32 (77)27 (89)
   0.5072. (83)80 (91)76 (96)
   1.0065. (79)72 (90)68 (95)
   2.0052. (73)60 (88)56 (94)
   3.0043. (68)51 (86)47 (94)
   4.0037. (64)44 (85)40 (93)
   6.0027. (59)33 (83)30 (93)
   8.0021. (55)27 (82)24 (93)
   0.5065. (78)74 (88)70 (94)
   1.0057. (73)66 (87)61 (93)
   2.0044. (66)53 (84)48 (92)
   3.0035. (60)44 (82)39 (92)
   4.0029. (56)37 (80)33 (91)
   6.0021. (51)28 (78)24 (91)
   8.0016. (47)22 (77)19 (90)

To understand conditional POC, it is helpful to examine a specific example. For example, consider a woman with a nonulcerated lesion measuring 0.500 mm in thickness on an extremity. For this patient, the estimated cured fraction is approximately 81%, and the predicted survival at 5 years is 96%. These figures allowed us to estimate that, among patients with such lesions who are free of recurrence at 5 years, approximately 84% (81 of 96 patients) are cured of their disease. That is, the 5-year conditional POC is 84%. At 10 years and 15 years, this value increases to 89% and 93%, respectively. Note that, for all lesions, conditional POC increases with the duration of recurrence free survival. This increase tends to be significantly greater for patients with a poor initial prognosis than for patients with a good initial prognosis.

Table 5 demonstrates the survival impact of each risk factor. For reference, a baseline low-risk lesion was selected—a nonulcerated lesion measuring 0.500 mm in thickness on the extremity of a woman. When the gender risk factor was changed from female to male, holding all other risk factors constant, the estimated cured fraction was reduced by an absolute percentage of 5.7% from baseline (i.e., from 80.8% to 75.1%). When the site risk factor was changed from extremity to nonextremity, the estimated cured fraction was reduced from the baseline by 8.1%. The major changes in the estimated cured fraction, however, were related to thickness. By increasing thickness from the baseline of 0.50 mm to 8.00 mm, the estimated cured fraction was reduced by 50.1%.

Table 5. Summary of Clinical Impact
VariableRelative impact of risk factors
Percent cured fraction (decrease from baseline)aTumor-related median survival (years) (decrease from baseline)Ten year survival (%) (decrease from baseline)
  • a

    Negative values show a decrease produced by substituting alternative risk factors one at a time. All decreases in percentage represent absolute decreases from baseline.

  • b

    Baseline (minimal) risk is for women with a nonulcerated lesion on an extremity that measures 0.50 mm in thickness.

Risk factor   
 Male75.1 (−5.7)10.0 (−0.0)88 (−2)
 Nonextremity72.7 (−8.1)8.1 (−1.9)84 (−6)
 Ulcerated80.8 (−0.0)6.9 (−3.1)87 (−3)
  1.0074.6 (−6.2)8.9 (−1.1)86 (−4)
  2.0063.8 (−17.0)7.5 (−2.5)77 (−13)
  3.0055.1 (−25.7)6.7 (−3.3)69 (−21)
  4.0048.1 (−32.7)6.1 (−3.9)63 (−27)
  6.0037.8 (−43.0)5.3 (−4.7)52 (−38)
  8.0030.7 (−50.1)4.8 (−5.2)44 (−46)

Note that, in Tables 4 and 5, the estimated cured fraction is not correlated with the presence of ulceration, and the median tumor-related survival is not correlated with gender. This is because these correlations did not satisfy the statistical threshold necessary for inclusion in the final parametric model that was used to generate these data.


Our findings are consistent with those from previous nonparametric studies of patients with primary cutaneous melanoma.1–6 The thickness of the primary lesion was identified as the major predictor of clinical course, whereas ulceration status, anatomic site, and patient gender played important secondary roles. Of particular interest in our current study, however, are insights into the basic mechanisms of long-term survival as well as the biologic forces that govern metastasis. These mechanisms are understood more easily by considering the two primary determinants of long-term survival—probability of cure and median survival among uncured patients.

For example, a high cured fraction is found in patients with lesions measuring 0.500 mm thick, which supports the low malignant potential generally attributed to such lesions. Despite a good prognosis, however, patients with cutaneous melanoma may develop recurrent disease many years after surgery.14 One study found that, among patients who had survived melanoma for 10 years, 25% eventually developed a recurrence.15 It is especially interesting that many of these late recurrences developed in patients with thin lesions.

Such late metastases are to be expected, given the results from parametric analysis. Patients with the most favorable prognosis (i.e., women with nonulcerated lesions measuring 0.500 mm thick on an extremity) had a cured fraction of 80.8% and a median tumor specific survival among uncured patients of 10.0 years. Thus, although most of these patients were cured surgically, approximately half of all deaths from melanoma occurred a decade or more after treatment, barring death from other causes. In fact, thin lesions are especially likely to yield late recurrences, because lesion thickness is correlated strongly but inversely with the median tumor specific survival (Tables 4, 5). This trend, although it often is underestimated by clinicians, also has been confirmed by studies that relied on conventional nonparametric methods.15

Parametric insights also help explain another common observation among patients with cutaneous melanoma. For those who have survived 10 or more years without developing a recurrence, the prognosis is correlated less with thickness than at the time of initial treatment. In other words, as survival increases, the prognosis tends to equalize among patients with thick lesions and patients with thin lesions. To understand this leveling effect, it is useful to consider the conditional POC. This is the probability that a patient is cured, given recurrence free survival for a specific time. Note that, at the time of treatment (i.e., at time t = 0), the conditional POC is equal to the cured fraction. However, as follow-up increases, the conditional POC increases steadily, with a maximum potential value of 1.0. However, this maximum value is unlikely to occur in reality, because it implies that the probability of survival among uncured patients has declined to 0, thereby assuring that only cured patients remain alive.

Note that, after 15 years of follow-up, the conditional POC is similar among all risk groups (Table 4). For example, after 15 years of follow-up, patients with thin lesions (0.500 mm) and patients with thick lesions (8.00 mm) share a similar conditional POC, despite marked differences in the initial cured fraction and the median tumor-related survival. This apparent anomaly results from the extremely short survival associated with thick lesions. Consequently, uncured patients with thick lesions tend to die quickly, leaving only a small proportion still alive after 15 years. Thus, as follow-up advances, the few cured patients constitute a progressively greater proportion of all survivors.

In conclusion, our parametric approach provides novel insights into the mechanisms that govern survival from cutaneous melanoma. These insights help us to conceptualize phenomena that otherwise may appear anomalous. In addition, however, we hope that Tables 4 and 5 will prove to be practical guides for clinicians, allowing them to estimate more precisely the long-term clinical course of their patients after initial treatment.


The authors thank Wilma E. Stanley for her technical assistance with data acquisition and Maggie Abbey for her editorial assistance.