There is still a crucial need for prognostic markers in melanoma (MM) to guide the follow-up strategies and to indicate the adjuvant therapies. For primary MM, the best currently available markers are still tumor thickness and presence of ulceration.1 Recently, the histologic status of the sentinel node has been shown to be a strong prognostic indicator, but it requires an invasive procedure.2 A lot of potential biologic markers were explored in an attempt to assess the aggressiveness of MM: histologic markers such as mitotic rate and vascularity, histochemic markers such as cell activation markers, adhesion molecules and integrins and molecular markers such as oncogene expression and PCR detection of circulating melanoma cells.3, 4, 5, 6 However, these investigations did not result in the identification of reliable and convenient prognostic indicators independent of tumor thickness.
There is evidence that all primary MMs do not develop in the same way, from very slowly growing to fast-increasing tumors. These different kinetics probably may account for the lack of strong relationship between the tumor thickness and the patient delay in seeking medical advice.7
Clinicians know that a simple interview of the patient, at the moment when the primary MM is resected, is often a useful source of information about the growth phase of the tumor. Indeed, many patients can accurately describe how, when and how fast their MM developed. Curiously, this information has never been explored in terms of relationship with prognosis, probably because it has not been considered as reliable. From the observations of our daily experience, we had the impression that patients who described apparently a fast-growing primary MM tend to recur more often and sooner than the others. Assuming that the clinical description of the tumor growth by the patient had some relevance, we hypothesized that the initial kinetics of the primary MM growth could reflect its biologic aggressiveness.
Herein, we assessed the link between the kinetics of the visible growth of primary MMs and the risk of recurrence. Our first objective was to show a relationship between MM kinetics and MM aggressiveness. Our second objective was to test whether patient responses to a simple questionnaire could be a useful and reliable prognostic marker in MM.
PATIENTS AND METHODS
We designed a prospective multicenter survey in primary MM patients to study whether the kinetics of the visible growth of the MMs was predictive of risk and rapidity of relapse, using a patient questionnaire to estimate this kinetics.
In 1996, consecutive patients with MM were enrolled in 19 dermatologic centers in France. In situ MM (Clark I) were excluded because the risk of relapse was null. As patients with MM coincidentally detected by doctors were generally unable to provide precise information about dates “t1” and “t2”, only self-detected MMs were retained.
Quantification of the kinetics of the visible growth
Using a standardized questionnaire, each patient was asked to recall 2 dates: the date “t1” when he or she had first noticed a lesion on the place where the MM developed later, and the date “t2” when he or she first felt that this lesion changed and/or became curious or suspicious. The date of resection “tR” was recorded.
The definition of the kinetics (MK) of the growth of a primary MM is the ratio of the volume of the MM at resection by the time period between the starting point of the tumor growth and its resection: volume of MM at resection/time from initial growth. However, the starting point of the MM growth is unknown and the MM volume is very difficult to measure. Using the clinical events that the patients can describe and the Breslow thickness, we assumed that the ratio MM thickness/duration of the MM visible growth could be used as a surrogate value for MK. Because t1 was the most direct assessment of the starting point of the visible growth of the MM, we built up the melanoma kinetics index (MKI) = MM thickness/t1–tR, which was expected to be the best surrogate value for MK, at least in MMs that were developing de novo (Fig. 1). Indeed, in MMs growing on a preexisting nevus, the t1–tR period may encompass not only MM growth but also nevus history (Fig. 1). MKI was thus used as a surrogate value for MK in patients with a t1–t2 delay under 5 years, a subgroup in which MMs were more likely to be de novo.
In order to have a surrogate for MK that could be applied to MMs on preexisting nevi as well as to de novo MMs, we assumed that the date t2, i.e., the first time the patient detected a change, could also be an approach to the starting point of the clinically visible MM growth. We thus defined another index: MKI* = MM thickness/t2–tR.
Data from the initial interview were recorded in the coordinating center and follow-up data were collected in the year 2000. Follow-up was conducted in the same way8 and provided the date of the first relapse and the dates of the last visit and death.
Univariate and multivariate analyses were conducted on SPSS version 10 to assess the link between MKI and relapse-free survival. Only relapse risk was studied since number of deaths was too limited to assess the link between MKI and overall survival.
Using MKI as a surrogate value for MM kinetics, we accepted 2 major assumptions. First, the tumor volume is more or less proportional to the tumor thickness.9 Second, the shorter is t1–tR, the clinically visible “growth period” of the MM according to the patient description, the shorter is the actual MM “growth duration.” The proportion between MKIs (Breslow/t1–tR) in different patients was thus expected to be linked to the proportion between their corresponding kinetics (volume/growth duration). However, we did not expect that absolute values of MKIs were closely linked to absolute values of their corresponding MKs, since the t1–tR period and the MM thickness were definitely not true measures of the MM growth duration and the tumor volume, respectively. In other words, MKIs were likely to be much more representative of MKs on a log scale than on an arithmetic scale. Log (MKI) and log (MKI*) were thus used in univariate and multivariate analyses with Cox model10 for relapse-free survival, and logistic regression for relapse at 1 year. These analyses were also conducted using MKI or MKI* expressed as a continuous variable or as quartiles.
In multivariate analyses, the prognostic variables were Breslow thickness used as a continuous variable, ulceration, Clark level (II to V), age, sex and periods t1–tR and t2–tR. Analyses were also performed with Breslow thickness expressed as a discontinuous variable (D-Breslow) based on the newly proposed AJCC classification:1 ≤1 mm, >1 mm and ≤2 mm, >2 mm and ≤4 mm, >4 mm. Node status was not included as a prognostic variable. Indeed, sentinel node dissection was not recognized as a standard procedure in France at time of inclusion and was thus not performed. Furthermore, no patient underwent elective node dissection. Adjuvant treatment was prescribed in a minority of patients and was thus not taken into account.
We addressed 3 main questions using MKI in the subgroup of supposedly de novo melanoma (t1–t2 < 5 years). First, we assessed the link between MKI and relapse-free survival in the 2 univariate models. Second, in order to determine whether MKI provided information independent of tumor thickness, thickness was forced in the 2 multivariate models, and log (MKI) was proposed as a candidate variable. Third, to address the question of whether MKI could still provide prognostic information when most usual prognostic markers were available, the following candidate variables were proposed to the 2 multivariate models: log (MKI), Breslow continuous and as discontinuous variable, ulceration, Clark level, sex, age and t1–tR period as candidate variables in an ascending stepwise.
In the unselected group (341 patients), the same univariate and multivariate analyses were performed. However, because t1–tR is not really representative for MM growth period in patients with an MM developed on a preexisting nevus that appeared many years before (Fig. 1), log (MKI*) and t2–tR period were used instead of log (MKI) and t1–tR delay, respectively.
Among the 507 patients with non-in situ MM who filled out the questionnaire, 362 were self-detected and were thus retained for our study. Final follow-up data were collected in the year 2000 and could not be obtained in 21 cases.
Finally, 341 MM cases of the 362 were thus analyzed, 201 of which were considered as likely to be de novo (t1–t2 period < 5 years). The characteristics of the population and MMs were as follows: sex distribution (130 male, 211 female), mean age = 49.58 (median = 49, min = 14, max = 90), mean Breslow thickness = 2.69 mm (median = 1.25, min = 0.15, max = 99) and anatomic subtype (67 nodular MMs, 248 superficial spreading MMs, 6 lentigo maligna, 13 acrolentiginous MMs and 7 others. With a median follow-up of 4 years, there were 70 relapses (20.5%) and 35 deaths (10.3%).
Kinetics of visible growth and risk of recurrence: MKI and relapse
Log (MKI) was predictive of relapse using Cox model and relative risk (RR) for recurrence at 1 year significantly increases with log (MKI) in patients with supposedly de novo MMs (Table I).
Table I. Surrogates of Primary Melanoma Kinetics [Log (MKI) and Log (MKI*)] as Risk Factors for Relapse: Crude Assessment and After Adjustment for Breslow Thickness
Surrogate of melanoma kinetics
OR for relapse (Cox)
RR for relapse at 1 year
Cox model and logistic model for relapse at 1 year by univariate analyses are given in roman type.
Cox model and logistic model for relapse at 1 year by multivariate analyses testing independent from Breslow are given in italics. (Breslow thickness is forced in the models.)
When the primary MMs were divided into 4 quartiles according to MKI, Kaplan-Meier curves (Fig. 2) showed that primary MMs with the highest MKI tended to relapse earlier than the tumors with intermediate MKI, which in turn tended to relapse earlier than tumors with lowest MKI (p < 0.001).
Independence from Breslow.
When tumor thickness was forced into the multivariate models, log (MKI) was still a significant prognostic marker both by logistic model for relapse at 1 year (odds ratio [OR] =3.34 [1.92–5.82]) and Cox model (hazard ratio [HR] = 1.79 [1.43–2.22]) (Table I). The results remained significant when the tumor thickness was expressed as a discontinuous variable: OR = 1.67 (1.14–2.45) and HR = 1.29 (1.02–1.61), respectively.
Assessment of prognostic models.
Log (MKI) was retained at the second step after the discontinuous expression of tumor thickness (T-Breslow) in the logistic model (OR = 1.66 [1.14–2.43]) (Table II) and in the Cox model (HR = 1.29 [1.02–1.61]) (Table II).
Table II. Assessment of Log (MKI) in Prognostic Models: Logistic Regression for Relapse at 1 Year in the 201 Patients with t1–tR <5 Years (Self-Detected Melanoma Likely to be De Novo)
p to enter
95% CI of exp (coefficient)
Candidate variables: log (MKI), Breslow as a continuous variable, Breslow as a discontinuous variable (D-Breslow) (≤1 mm; >1 mm ≤2 mm; >2 mm ≤4 mm; >4 mm), ulceration, Clark, age, sex and delay t1–tR.
Logistic regression for relapse at 1 year
Cox regression for relapse
It is noteworthy that all the significant results obtained in univariate and multivariate analyses with log (MKI) were also significant or borderline significant using MKI as continual variable or quartiles (data not shown)
Kinetics of visible growth and risk of recurrence: MKI* and relapse
Log (MKI*) was also found to be a significant risk factor for recurrence in all self-detected primary MMs, using Cox model or logistic model (Table I).
When the primary melanomas were divided into 4 quartiles according to MKI*, Kaplan-Meier curves (Fig. 3) showed that primary MMs with the highest MKI* tended to relapse earlier than the tumors with intermediately high MKI*, which in turn tended to relapse earlier than tumors with low or intermediate-low MKI* (p < 0.001).
Independence from Breslow.
When tumor thickness was forced in the models, log (MKI*) remained a significant prognostic marker in the logistic model (OR = 2.10 [1.59–2.76]) and in the Cox model (HR = 1.44 [1.26–1.64]) in all self-detected melanoma. When Breslow thickness was expressed as a discontinuous variable, the results were similar for logistic model OR = 1.49 (1.12–1.97), but log (MKI*) was not any more retained in Cox model.
Assessment of prognostic models.
In the logistic model, log (MKI*) was kept at the third step after discontinuous Breslow and sex (OR = 1.44 [1.08–1.92]) (Table III). Using Cox model, discontinuous expression of Breslow and sex were retained, but log (MKI*) did not enter the model (Table III).
Table III. Assessment of Log (MKI*) in Prognostic Models: Logistic Regression for Relapse at 1 Year and Cox Regression for Relapse in the 345 Unselected Self-Detected Patients
p to enter
95% CI of exp (coefficient)
Candidate variables: log (MKI*), Breslow as a continuous variable, Breslow as a discontinuous variable (D-Breslow) (≤1 mm; >1 mm ≤2 mm; >2 mm ≤4 mm; >4 mm), ulceration, Clark, age, sex and delay t2–tR.
Logistic regression for relapse at 1 year
Cox regression for relapse
The significant results obtained in univariate or multivariate analyses with log (MKI*) were also significant or borderline significant using MKI* as continual variable or quartiles (data not shown).
Kinetics and Breslow
Although significant, the correlation between log (MKI) or log (MKI*) and Breslow were weak, with Pearson correlation at 0.285 (p < 0.001) and 0.377 (p < 0.001) respectively, thus confirming that growth kinetics carries independent prognostic information.
We report the first prospective study giving evidence that the kinetics of the visible growth of a primary MM is predictive of the relapse-free survival, independently of the tumor thickness. This is the first confirmation that the initial growth rate of an MM reflects its aggressiveness. More generally, our study gives a unique example showing that apparently “subjective” and unreliable data provided by a simple interview of patients about their tumor history can provide an “objective” and strong prognostic tool.
Unlike other cancers, melanoma is often accessible to the view from the very early growth of the tumor. This specificity offers a unique opportunity to have direct access to the natural history of the tumor by a simple interview of patients. Combining this information with histopathologic assessment of the size of the tumor (tumor thickness) is a way to approach a very simple intuitive concept: a cancer that is growing fast in its early phase is likely to have a poor prognosis, not only because it is more likely to be large at time of removal but also because this high growth rate reflects a high biologic aggressiveness.
The results of our study raise several questions about what is really measured by the MKI, about the precision of the measure and finally about the potential biases that may interfere with the results. Although the indexes MKI and MKI* were used as surrogates for MK, they are not a measurement of the true MK (tumor volume at resection/time from initial growth to resection). First, they are based on clinically detectable events, i.e., the visible MM growth. Second, the detection of these events is certainly influenced by many factors such as the tumor location and features, the patient attention and anxiety, which, in turn, is probably affected by age, sex, psychosocial factors and many others. However, the only consequence of using such a surrogate is probably a dilution of the information, which may lead to underestimating the strength of the relationship between MM kinetics and the prognosis. In this regard, among the data available through a patient interview, the duration of the t1–tR period was expected to be the most meaningful and accurate value to assess the MM growth period, although it was only relevant in supposedly de novo MMs. The t2–tR period was expected to be a more indirect and imprecise approach to growth period but was applicable to all MMs, including those on preexisting nevi. We thus considered the index MKI as the best surrogate value for melanoma kinetics, and the index MKI* only as a less relevant but more widely applicable alternative. Our results confirm that MKI has a higher prognostic value and is more Breslow-independent than MKI*. In other words, the more the index is representative of the true tumor growth kinetics, the better prognostic marker it is.
It is unlikely that a systematic bias accounted for an artificial link between MKI and disease-free survival. No information bias could be suspected in the delay assessment, since dates t1 and t2 were collected immediately after primary MM resection and were not influenced by the relapse date. Although we did not document the impact on overall survival, we know that in MM patients relapse-free survival has a major influence on overall survival.
Although rapidly growing MM are more likely to be thick when they are detected, our study showed that the predictive value of the kinetics is not only linked to the tumor thickness. Log (MKI) and to a lesser degree log (MKI*) carry information independently of tumor thickness. For a similar tumor mass, primary MMs with a higher MKI had also a more rapid and aggressive course after resection, which is indirect evidence that the growth kinetics is linked to the biologic aggressiveness of the tumors. Although sentinel node status is an excellent prognostic marker, it is unlikely that the results would have been different if this factor had been available in this analysis. Indeed, the status of the sentinel node is closely linked11 to MM thickness and ulceration, which have been taken into account.
The information from the patient is “subjective,” but subjective does not mean inaccurate or irrelevant. When a patient describes a rapidly growing tumor, he is more likely to be right than wrong. A series of articles recently highlighted that the so-called “narrative medicine”12 is a necessary complement to “evidence-based” medicine. Narrative medicine takes into account the individual “stories” of patients. Our study illustrates how interpretative data, i.e., the “individual story” of each primary MM described by the patient himself, can be informative when they are collected through a standardized questionnaire.
From a clinical point of view, our study suggests that listening to a patient able to describe his tumor history may provide better prognostic information than many complex biologic tests. Our data demonstrate that a high MKI is associated with a high risk of relapse and with a short remission time from primary MM resection to detection of nodal disease. A simple questionnaire such as the one used in our study combined with the assessment of Breslow thickness can permit the classification of patients into a few prognostic classes (Figs. 2 and 3). In clinical practice, simple and informative markers like MKI and MKI* could be useful, although they would not be applicable to one-third of MMs that are still discovered coincidentally by the doctors. To use these indexes in a practical setting, further studies are needed to optimize, standardize and validate a simple questionnaire.
There were many attempts to find markers of biologic aggressiveness in MM. Although the techniques used to measure these biologic parameters are very precise, “markers” such as mitotic index, vascularisation, expression of adhesion or proliferation molecules, circulation of tumor cells or antigens, regulation of melastatin (MLSN-1 mRNA) and many others3, 4, 5, 6, 13 are only dissociated fragments of a very complex puzzle, which gives a very limited insight into the biologic aggressiveness. The correlation of these markers with the prognosis is generally limited. Conversely, although the patient recall of the way his MM grew is often imprecise, MKI probably gives a more global and subsequently more relevant approach to this aggressiveness. In this regard, one has to keep in mind that unspecific and global markers such as fever, sedimentation rate or fibrin level are often more informative for the course of cancers or connective tissue diseases than separate dosages of specific molecules involved in a single pathogenic process.
In search for repeatable and objective tests for diagnosis or prognosis in all medical fields, we have concentrated our efforts on surgical, radiologic and biologic approaches, which are often costly and/or invasive and which were often not as reliable and useful as expected. Very rich information, which can be drawn from patient self-assessment, has been neglected or underused for fear it was too subjective and not repeatable. Our study emphasizes that the simple description of the visible disease process by the patient himself recorded using a standardized questionnaire is a relevant, safe and cost-free source of information on prognosis. We may have a lot to learn from this approach in a variety of fields.
We thank Dr. J.C. Roujeau for fruitful discussion.