Nevi, family history and fair skin increase the risk of second primary melanoma

Abstract

While risk factors for primary cutaneous melanoma are well defined, relatively little is known about predictors for second primary melanoma. Given the rising incidence of this cancer, coupled with improvements in survival, there is a prevalent and growing pool of patients at risk of second primary melanomas. To identify the predictors of second primary melanoma, we followed a cohort of 1083 Queensland patients diagnosed with incident melanoma between 1982-90 and who completed a baseline questionnaire. During a median follow-up of 16.5 years, 221 patients were diagnosed with at least one additional primary melanoma. In multivariate analyses, second primary melanomas were associated with high nevus count (HR 2,91, 95%CI 1,94 - 4.35), high familial melanoma risk (2.12, 1.34-3.36), fair skin (1.51, 1.06-2.16, inability to tan (1.66, 1.13-2.43), an in situ first primary melanoma (1.36, 0.99-1.87) and masculine sex (1.49, 1.12-2.00) Patients whose first primary was lentigo maligna melanoma (1.80, 1.05-3.07) or nodular melanoma (2.13 , 1.21-3.74) had higher risks of subsequent primaries than patients whose first primary tumor was superficial spreading melanoma. These characteristics could be assessed in patients presenting with first primary melanoma to assess their risk of developing a second primary.

Introduction

During the past several decades, there have been marked increases in the incidence of cutaneous melanoma in fair-skinned populations around the world, particularly for thin melanomas (Coory et al., 2006; de Vries et al., 2003; Garbe et al., 2000; Jemal et al., 2001; MacKie et al., 2007). Whereas for previous generations, a diagnosis of melanoma carried a grim prognosis, the vast majority of melanoma patients diagnosed today can expect to survive their disease. However, the improved survival rates coupled with population-wide increases in life expectancy mean that the risk of developing a subsequent melanoma will inevitably rise. This poses clinical challenges, for while guidelines exist to treat cutaneous melanoma and to manage recurrence, little attention has been paid to manage the risk of subsequent invasive melanoma.

Registry-based linkage studies suggest that the cumulative incidence of second primary melanomas varies across populations, estimated at 1.5% at 10 years in Switzerland (Levi et al., 2005), 5.3% in the United States ; (Goggins and Tsao, 2003) and 6.4% in Queensland, Australia (McCaul et al., 2008). More detailed analyses of registry and clinical case series demonstrate that the risk of subsequent melanoma is highest in the first year following initial diagnosis (Ferrone et al., 2005; Goggins and Tsao, 2003; McCaul et al., 2008), and that the annual rate of new diagnoses of primary melanoma appears to be relatively constant thereafter (McCaul et al., 2008). Most studies, but not all, indicate that the risk of subsequent melanoma increases with the age at which the first melanoma was diagnosed, but that sex and anatomic site do not appear to influence the risk of subsequent lesions (Goggins and Tsao, 2003; McCaul et al., 2008).

Phenotypic or other host factors that could be ascertained in the clinic and which might predict future risk of subsequent melanomas have been examined only in two prospective studies (Ferrone et al., 2005; Titus-Ernstoff et al., 2006) from which some consistent features have emerged. A verified family history of melanoma and the presence of either very large numbers of banal nevi or at least one atypical nevus have been associated with increased risks of subsequent melanomas in both studies. Evidence for a role of pigmentary characteristics (such as freckling, hair color and skin type) which are strongly associated with the risk of developing primary melanoma is inconsistent owing to the paucity of data.

Here we document the distribution of second and subsequent melanomas in a large prospective cohort of Queensland patients diagnosed with primary cutaneous melanoma, and identify the factors associated with their development.

Results

The age and sex distribution of the 1083 probands in the QFMP follow-up study is given in Table 1. There were 27% more females than males, and the latter were on average almost 4 years older (47.8 years among males, 43.9 among females). Only two probands were related to one another. The mean interval from index diagnosis to follow-up contact was 16.8 years (median 16.5 years) with a range of 12.3 – 23.0 years. There was little variation in this respect between males and females or by age.

Table 1.

Distribution by sex of characteristics of participants (probands) in the QFMP follow-up study

	Males		Females		Total
	N = 477		N = 606		N = 1083
	N	%	N	%	N	%
Age (years)
< 40	135	28.3	245	40.4	380	35.1
40 – 49	122	25.6	154	25.4	276	25.5
50 – 59	123	25.8	122	20.1	245	22.6
≥ 60	97	20.3	85	14.0	182	16.8
Site
Head/neck	66	13.8	56	9.2	122	11.3
Trunk	217	45.5	133	22.0	350	32.3
Upper limbs	92	19.3	157	25.9	249	23.0
Lower limbs	72	15.1	237	39.1	309	28.5
Unspecified	30	6.3	23	3.8	53	4.9
Clark level
1	110	23.1	136	22.4	246	22.7
2	191	40.0	256	42.2	447	41.3
3	85	17.8	110	18.2	195	18.0
4/5	52	10.9	64	10.6	116	10.7
Missing	39	8.2	40	6.6	79	7.3
Familial risk group
High	30	6.3	33	5.4	63	5.8
Intermediate	109	22.8	170	28.1	279	25.8
Low	338	70.9	403	66.5	741	68.4
Morphology
Nodular	29	6.1	30	5.0	59	5.4
SSM	318	66.7	435	71.8	753	69.5
LMM	26	5.4	28	4.6	54	5.0
NOS	104	21.8	113	18.6	217	20.0

Occurrence and characteristics of second primary melanomas

There were 221 persons in the sub-cohort with at least one primary melanoma other than the index lesion and a total of 375 such lesions (118 in situ and 257 invasive). Among these the index melanoma was in situ or invasive in 63 and 158 persons respectively. Excluding those with only synchronous melanomas reduced these numbers to 208 individuals (61 with in situ and 147 with invasive index melanomas) with at least one subsequent metachronous primary melanoma and a total of 362 lesions (116 in situ and 246 invasive). The 158 individuals with invasive index lesions had 152 invasive and 105 in situ additional melanomas. In total, over 20% of the overall sample had more than one primary melanoma, and one individual had nine (Table 2).

Table 2.

Number of persons with at least one synchronous melanoma and by number of metachronous melanomas by behavior of index melanoma

		In situ		Invasive		Total
		N	%	N	%	N	%
Number of additional melanomas
	0	199	76.0	663	80.8	862	79.6
Synchronous melanomas only
		2	0.8	11	1.3	13	1.2
Metachronous melanomas
	1	33	12.6	93	11.3	126	11.6
	2	14	5.3	28	3.4	42	3.9
	3	8	3.1	13	1.6	21	1.9
	4	3	1.1	7	0.9	10	0.9
	5	2	0.8	6	0.7	8	0.7
	≥6	1	0.4	0	0.0	1	0.1
Total		262		821		1083

Compared with those with in situ index lesions, participants whose first primary melanoma was invasive (Clark level 2 or greater) were somewhat more likely to have invasive second melanomas but the difference was not statistically significant (Table 3). There was little difference in mean and median thickness from index to second metachronous melanoma (index melanoma: mean 0.82± 0.03 mm, median 0.55 mm; second metachronous melanoma mean 0.85 ± 0.09 mm, median 0.55 mm). There was only weak evidence of specific anatomical concordance between the body site of the index and second metachronous primary melanomas (unweighted Kappa 0.21) although the association was highly significant (Table 3).

Table 3.

Behavior and anatomic site of index melanomas versus metachronous second primary melanomas.

Behavior	Second primary melanoma
Index	In situ		In situ		Invasive		Invasive
melanoma		(N)		(%)		(N)		(%)	Total
In situ		34		55.7		27		44.3	61
Invasive		64		43.5		83		56.5	147
Total		98		47.1		110		52.9	208
Anatomic site1	Second primary melanoma
Index	Head/neck		Trunk		Upper limb		Lower limb		Total
melanoma	N	%	N	%	N	%	N	%
Head/neck	15	50.0	5	16.7	6	20.0	5	16.7	30
Trunk	10	15.4	28	44.6	14	21.5	12	18.5	65
Upper limb	15	30.0	13	26.0	15	30.0	7	14.0	50
Lower limb	7	12.3	10	17.5	15	26.3	28	43.9	57
Total	47	23.3	56	27.7	50	24.7	49	24.3	202

¹6 index melanomas or second metachronous melanomas did not have anatomical site recorded

Predictors of multiple primary melanomas - univariate analyses

Mean numbers of subsequent melanomas increased by age overall and in females, whereas in males the only major difference was between those under 40 years of age and those in older age groups (Table 4). Sex, age, morphology, familial risk, numbers of nevi, red phenotypeskin color, skin type, tanning ability, and a history of solar keratoses or non-melanoma skin cancers were all significantly related to numbers of subsequent primary melanomas, some highly significantly; invasive behavior, showed a marginal association (Table 4). Associations were generally weaker in females. We found no overall association between numbers of subsequent primary melanomas and any of anatomic site, freckling, Clark level, history of sunburns, hair or eye color. Findings from failure-time analyses of time to first metachronous primary melanoma were broadly similar, except that the association with invasive behavior was stronger, Of 40 individuals with at least 4 primary melanomas, 11 were from the high risk group, almost 5 times expectation assuming no association between familial history and risk of multiple primaries, while 18 were from the low risk group, 30% less than expectation.

Table 4.

Non-parametric (Kruskal-Wallis) analysis of variance of numbers of melanomas (excluding index primary).

		Males		Females		All
Factor	d.f.	χ 2	p	χ 2	p	χ 2	p
Sex	1					8.67	0.003
Age category	3	23.9	<0.001	1.69	0.64	20.8	<0.001
Morphology	3	5.82	0.12	4.95	0.18	8.77	0.03
Familial risk category	2	18.9	<0.001	4.78	0.09	20.2	<0.001
Nevus category	3	24.6	<0.001	7.78	0.05	28.1	<0.001
Red phenotype	1	7.57	0.006	0.29	0.59	4.15	0.04
Skin color	2	8.10	0.017	3.10	0.22	8.92	0.012
Skin type	3	11.9	0.008	4.72	0.19	11.5	0.009
Tanning ability	3	9.79	0.020	7.09	0.07	10.5	0.015
Solar keratoses	1	10.9	0.002	0.01	0.97	6.15	0.013
Skin cancers	1	7.69	0.006	3.34	0.07	12.9	<0.001
Behavior (in situ vs invasive)	1	3.76	0.053	0.32	0.57	3.36	0.070
Freckling category	2	13.7	0.001	0.41	0.81	4.12	0.13
Anatomic site	5	5.80	0.33	1.09	0.95	5.29	0.38
Clark level	4	3.76	0.44	1.69	0.80	1.25	0.87
Sunburns	3	0.77	0.86	0.85	0.84	1.82	0.61
Hair color	5	2.46	0.79	2.07	0.84	1.91	0.86
Eye color	2	1.25	0.53	0.51	0.78	0.18	0.91

Non-proband relatives had slightly more multiple primary melanomas than probands despite being on average 3.5 years younger (p=0.04 after correcting for age, data not shown).

We observed no association between time to second metachronous primary and either the thickness of index melanoma or with continuous measures of UV exposure in childhood, adolescence or adulthood (data not shown). All non-parametric correlations were small, between −0.05 and 0.04, and non-significant (data not shown).

Predictors of second primary melanomas - multivariate analyses

Persons with a large number of nevi, those in the familial high risk group or whose index melanoma was nodular had significant two-fold and greater increased risks for subsequent primary melanomas. Males, individuals with a moderate number of nevi, whose index melanoma was an LMM or in situ or who had fair skin or an inability to tan were also at elevated risk of acquiring a second primary lesion (Table 5).We repeated the analysis by strata of familial risk group (moderate/high; low); there was some suggestion that risk group membership may be a modifying factor, but the differences are likely to be due to chance.

Table 5.

Hazard ratios and 95% confidence intervals (95% C.I.) of time to first subsequent primary melanoma, in toto and within strata of familial risk group. (Synchronous melanomas excluded).

Factor	Categories	Hazard Ratio1	95% C.I.	Hazard Ratio1	95% C.I.	Hazard Ratio1	95% C.I.
Sex	Female	1.00	ref	1.00	ref	1.00	ref
	Male	1.49	1.12–2.00	2.49	1.47-4.23	1.17	0.81-1.68
Histological type	SSM,unspecified	1.00	Ref	1.00	ref	1.00	ref
	LMM	1.80	1.05–3.07	1.81	0.74-4.43	1.47	0.70-3.10
	Nodular	2.13	1.21–3.74	1.10	0.38-3.17	2.63	1.35-5.15
Behavior	Invasive	1.00	ref	1.00	ref	1.00	ref
	In situ	1.36	0.99–1.87	1.38	0.77-2.44	1.30	0.87-1.95
Nevus count	None or few	1.00	Ref	1.00	ref	1.00	ref
	Moderate	1.92	1.40–2.65	2.46	1.41-4.29	1.63	1.08-2.47
	High	2.91	1.94–4.35	2.81	1.35-5.85	2.98	1.82-4.89
Tanning ability	Able to tan	1.00	ref	1.00	ref	1.00	ref
	Unable to tan	1.66	1.13–2.43	1.79	0.88-3.61	1.77	1.11-2.83
Skin color	Dark/medium	1.00	ref	1.00	ref	1.00	ref
	Fair or pale	1.51	1.06–2.16	2.22	1.09-4.53	1.33	0.87-2.04
Familial risk	Low / intermed	1.00	ref	n/a	n/a	n/a	n/a
	High	2.12	1.34 – 3.36

¹Hazard ratio and 95% CI derived by Cox's proportional hazards regression in four strata of age, adjusted for all factors in the table.

Discussion

We have identified a number of characteristics that are associated with significantly elevated risk of developing a second primary melanoma. The strongest predictor was the number of melanocytic nevi reported at baseline; patients reporting large numbers of nevi had 3-fold higher risks of developing a subsequent melanoma than those reporting only small numbers of nevi. Other factors associated with significantly elevated risks of second primary melanomas included having a high familial risk of melanoma, having a sun-sensitive skin type and having a melanoma of nodular or lentigo maligna subtypes. Of note, close to half of participants with four or more primary melanomas were from the low familial risk group, the clinical implication being that patients with no or weak family history of melanoma may still be at risk of multiple primary melanomas.

There are few studies with which to compare our findings. While some earlier studies have explored the incidence and determinants of second primary melanoma in large populations using record linkage techniques (Giles et al., 1995; Goggins and Tsao, 2003; Levi et al., 2005; McCaul et al., 2008), such investigations have been limited to using routinely-collected data and thus have been constrained by the absence of data relating to the phenotype of the patient. A larger number of studies has documented the occurrence of multiple primary melanomas arising in historical cohorts of patients treated at single institutions (Ariyan et al., 1995; DiFronzo et al., 1999; Ferrone et al., 2005; Johnson et al., 1998; Savoia et al., 1998). Such studies typically have rich information describing histopathological characteristics of the tumors, but phenotypic data relating to the patient has seldom been collected in a systematic fashion. Three epidemiologic studies of multiple primary melanomas have been reported, and each gathered systematic data on phenotype and family history of melanoma (Begg et al., 2006; Burden et al., 1999; Titus-Ernstoff et al., 2006). The Scottish Melanoma Group compared the characteristics of patients with multiple versus single primary melanomas, and found strong associations with high nevus counts, family history of melanoma and ‘non-use of sunscreen’ (Burden et al., 1999). That study also reported a high prevalence of germline CDKN2A mutations among patients with multiple primary melanomas. The GEM study utilized a novel design comparing 1210 patients with incident multiple primaries to 2470 patients with a single primary melanoma (Begg et al., 2006). That study also reported similar findings to ours; in particular, that patients with high nevus counts were almost 3-fold more likely than those with low nevus counts to develop multiple primaries. The investigators also reported positive associations with family history of melanoma, freckling in childhood and light hair color. Similar to the Scottish group, the GEM investigators found a strong and statistically significant 4-fold increased relative risk of multiple primary melanoma associated with the mutations in CDKN2A. (Berwick et al., 2006). The New Hampshire epidemiologic study followed-up 354 cases from a case-control study, 27 of whom developed a subsequent primary within 2 years (Titus-Ernstoff et al., 2006). While limited in statistical power, this study reported a strong positive association with atypical nevi, and observed an inverse association between lifetime sunburns and risk of second primary melanoma.

While we found some evidence of an association between the anatomical sites of the first and second melanoma, the concordance within specific sites was modest, with a kappa value of only 0.23. A large, registry-based study in Australia reported kappa statistics for body site concordance 0.41 for synchronous melanomas and 0.29 for metachronous lesions (Giles et al., 1995). Other studies have also reported that anatomical concordance between first and second primary melanomas exceeds that expected by chance, but falls far short of a strong relationship (Johnson et al., 1998).

Strengths of this study include the large sample, prospective design and the systematic collection of phenotypic and family history data at baseline. In addition, study participants were followed-up through self-report and linkage to population registers; the latter ensuring close to complete ascertainment of subsequent melanomas. Thus our study is likely to have high internal validity. We used standard measures for assessing phenotype which we have previously demonstrated to possess moderate to high levels of repeatability (Baxter et al., 2008). Moreover, as all such measures were collected prior to outcomes, misclassification of these exposures must have been non-differential with respect to outcome and any resulting bias would likely be towards the null.

A potential limitation of the current study was the fact that the study sample was not representative of the population of people having a first primary melanoma. This occurred for two reasons. Firstly, the parent study (QFMP) was designed to identify genetic factors associated with melanoma development and thus intentionally over-sampled patients from families with higher than average risk of developing a primary cutaneous melanoma (Aitken et al., 1996). Secondly, we were required to obtain new consent from study participants in 2003-5 to enable linkage to the cancer registry which necessarily restricted the cohort to those who were alive and contactable at that time. Supplementary analyses showed significant differences in age and tumor characteristics between those QFMP participants who were followed up with those who were not, and thus the cumulative risk of developing a second melanoma in this cohort was higher (13% at 10 years) than has been reported previously in the Queensland population (6.4%) (McCaul et al., 2008). For these reasons, our findings with respect to the incidence of second primary melanoma may not be generalizable to other populations. However, we have no reason to believe that the associations we observed between patient characteristics and the risk of developing a second primary melanoma would be biased.

In conclusion, patients diagnosed with a first primary melanoma have a high risk of developing a second primary tumor, and the risks are highest for those with large numbers of nevi or who have a higher than average family history of melanoma. Other factors, including red hair phenotype and fair skin, appear to modestly increase the risk of subsequent melanoma. These factors can be assessed in all patients at the time of diagnosis with a first primary melanoma; those at highest risk for developing second melanomas can be counseled appropriately.

Materials and Methods

The analyses presented here were embedded within the Queensland Familial Melanoma Project (QFMP), a family-based study of melanoma patients for which the full details have been described elsewhere (Aitken et al., 1996; Baxter et al., 2008; Siskind et al., 2002). Briefly, the QFMP comprised a sample of melanoma patients diagnosed with first primary cutaneous melanoma in Queensland between January 1, 1982 and December 31, 1990 and registered with the Queensland Cancer Registry (QCR; notification of cancers became mandatory in Queensland in 1982). Diagnoses had to be confirmed by histology, and could be in situ or invasive. Patients with acral lentiginous melanoma (ALM) were not eligible for the QFMP. The QFMP intentionally oversampled patients with a known family history of melanoma to facilitate future genetic research; the algorithm used to stratify patients by family history is described below. In total, 1,897 probands (i.e. patients meeting the eligibility criteria above) completed a detailed, self-administered questionnaire in 1991-1993 requesting information on family history, risk factors and medical and residential history.

A subsample of 1,083 QFMP probands was re-contacted between 2002 and 2005 to elicit further diagnoses of melanoma and authorize a confirmatory search in the QCR (Baxter et al., 2008). All participants gave their consent to take part, and the study was approved by the human research ethics committee of the Queensland Institute of Medical Research.

Records were sought for all consenting respondents who reported melanoma diagnosed in Queensland since 1982 (the first year of mandatory cancer registration in that state). Where a respondent was known by a name other than their legal name, or changed their names, a record for each known alias was submitted to the QCR to attempt to capture all possible records for an individual. For completeness of records, information pertaining to all reported cancers (not only melanomas) was requested. Data were received from QCR in the form of a de-identified electronic data file.

Data were received from QCR in the form of a de-identified electronic data file. Where a record could not be located at the QCR, or diagnosis was reported in another state, a request was submitted to a doctor nominated by the respondent (generally the diagnosing doctor) for a copy of the histopathology reports. If the doctor did not reply within two weeks, a telephone interviewer called the practice to request a copy of information. Where the practice advised that the doctor had died, moved or sold the practice, the interviewer then contacted the family of the doctor, the new practice owner or associated record management company in an attempt to locate the missing records. Where this was unsuccessful, an attempt was made to locate the information through subsequent treating doctors nominated by the respondent. Once received, the histopathology reports were then coded by a QCR-trained medical coder using ICD-O (3^rd edition) and double-entered into a database.

Pathology reports were re-examined and corrected, and the file correspondingly updated, in 2009. For each primary melanoma diagnosed, trained study nurses abstracted salient details from the pathology report including the date of diagnosis, anatomic site, behavior (in situ or invasive), and Clark level or Breslow thickness.

Variables for analysis

All variables used in analyses were derived from the original baseline questionnaire. Age was defined as age at diagnosis of the first primary melanoma. Skin color, self- assessed on unexposed sites such as the inner upper arm, was recorded in two categories (dark/medium, fair), eye color in three categories (blue/gray, green/hazel, brown), and early adult hair color in six categories (fair/blonde, light brown, light red/ginger, dark red/auburn, dark brown, black). Propensity to burn in the sun was categorized as never, sometimes or always burn, and tanning ability after prolonged sun exposure as none, slight, moderate or deep tan. Self-reported nevus density was recorded in four categories by comparison with diagrammatic representations (none, few, moderate or many) and density of freckling in summer was categorized as none, ≤100 freckles or >100. Number of sunburns during life was recorded as none, one, 2 to 5, or >5. Self-reported history of solar keratoses or keratinocyte skin cancers (squamous cell carcinomas and basal cell carcinomas) was also obtained.

We derived a “red phenotype” variable, defined as individuals with red hair and/or more than 100 freckles. Participants were further classified as belonging to high, intermediate or low risk families according to a previously derived measure of familial melanoma risk. described in detail elsewhere (Aitken et al., 1994). Briefly, this index was based on the number of cases of melanoma among all relatives in the family in excess of those predicted from the age-, sex- and birth cohort-specific cumulative incidences of melanoma among all relatives in the sample. Participants in the top 2.5% of the cohort were placed in the high-risk category, those between the median and the 97.5 percentile were assigned to the intermediate category and those below to the low-risk category. Familial risk categories were determined at entry to the cohort. High-risk individuals were over-sampled for re-contact.

Ultra-violet (UV) radiation exposure - total in childhood (5 to 12 years), adolescence (13 to 19 years) and average per year from 20 years of age on – was estimated from the lifetime residence and sun exposure calendars completed by the participants at baseline (Siskind et al., 2002)

Data analysis

Our primary aim was to identify predictors of second primary melanomas within the cohort. Our secondary aim was to describe the distribution of second and subsequent melanomas, but generalization of the latter findings to other populations should be pursued with caution given the selected nature of our cohort.

Second melanomas were deemed to be synchronous with the index lesion if they were diagnosed within 30 days of the latter. These have been included in analyses of total numbers of lesions, but not when time from first to the second histological diagnosis of primary melanoma was the outcome measure. In comparisons of characteristics of first (index) with second melanomas, persons with only synchronous lesions have been excluded. In those with both synchronous and metachronous lesions, the first melanoma occurring more than 30 days after the index melanoma served as the “second” lesion in the above analyses.

Univariate analysis of numbers of subsequent melanomas was performed using Kruskal-Wallis non-parametric tests for categorical or ordinal factors. As a check failure time analysis of time to first subsequent metachronous melanoma, with log-rank tests, was also employed. For continuous factors (thickness, UV exposure) Kendall's non-parametric correlation coefficient (tau) was used.

Multivariate analysis was by means of Cox proportional hazard regression on time to first subsequent melanoma across four age strata (< 40 years, 40 – 49 years, 50 – 59 years and ≥ 60 years). Based on the results of the univariate analyses, levels in several of the phenotypic variables were combined before inclusion in the proportional hazards models. Phenotypic variables are highly inter-correlated; the minimal included subset of this group of variables was established by sequential elimination.

As well as analyses of the entire file, males and females were analyzed separately by univariate methods. In the multivariate analysis, sex is included as a predictor.