Divergent cancer pathways for early-onset and late-onset cutaneous malignant melanoma

Abstract

BACKGROUND:

Emerging data suggest that cutaneous malignant melanomas (CMM) may arise through divergent cancer pathways that are linked to intermittent versus accumulated sun exposure. However, numerous questions remain regarding the timing and/or age of exposure.

METHODS:

The authors systematically examined the effect of aging on CMM incidence in data from the Surveillance, Epidemiology, and End Results Program of the National Cancer Institute. Standard descriptive epidemiology was supplemented with mathematical models. The impact of advancing age on CMM incidence was assessed by sex, histopathologic classification (superficial spreading melanoma [SSM] or lentigo maligna melanoma [LMM]), and anatomic site (face, head, and neck [FHN] or lower extremity [LE]).

RESULTS:

Sex, histopathology, and anatomic site were age-specific effect modifiers for CMM that indicated divergent (bimodal) early-onset and late-onset cancer pathways. Early-onset melanomas were associated predominantly with women, SSM, and LE. Late-onset melanomas were correlated with men, LMM, and FHN. Early- and late-onset melanoma populations were confirmed with age-period-cohort models that were adjusted for period and cohort effects. Two-component mixture models also fit the data better than a single cancer population.

CONCLUSIONS:

The current results were consistent with a divergent and age-dependent solar hypothesis for CMM. Early-onset melanomas may represent gene-sun exposure interactions that occur early (and/or intermittently) in life among susceptible individuals. Late-onset melanomas may reflect accumulated, lifelong sun exposure in comparatively less susceptible individuals. Future analytical studies should be powered adequately to account for this age-dependent effect modification both for acknowledged melanoma risk factors (sex, histopathology, and anatomic site) and for suspected melanoma risk factors, such as constituent genetic variants. Cancer 2009. Published 2009 by the American Cancer Society.

Solar irradiation generally is considered a direct risk factor for cutaneous malignant melanoma (CMM), although the relation between sun exposure and melanoma risk is complex.1, 2 For example, in North America, increased CMM incidence is associated positively with a north-south gradient in ultraviolet (UV) light exposure and with higher levels of UV exposure, as measured by climatic, atmospheric, and recreational exposure.3, 4 CMM incidence also generally is higher among individuals with fair skin. Conversely, some observations suggest an indirect relation between sun exposure and CMM. That is, CMM is more common among indoor workers than among outdoor workers and in intermittently sun-exposed skin.5, 6 Early-onset melanomas also are affected more by intermittent sun exposure than by continuous sun exposure.7

These observations suggest that CMM may arise through at least 2 different cancer pathways, 1 associated with intermittent sun exposure and the other associated with chronic sun exposure.8, 9 However, many uncertainties remain regarding the dual effects of sun exposure and melanoma risk, such as the timing and/or age of exposure.10 Indeed, unlike many epithelial tumors with peak incidence rates at older ages, CMMs commonly arise in younger individuals. Therefore, to assess whether descriptive data are consistent with an age-dependent, 2-pathway CMM model, we examined the impact of age on CMM incidence and the variability of age affects with other factors, such as sex, histopathologic classification, and anatomic body site. We used data from the National Cancer Institute's Surveillance, Epidemiology, and End Results (SEER) Program to obtain large numbers of CMM cases with population-based incidence rates.

We supplemented standard descriptive epidemiology with mathematical models. Poisson regression models were used to provided formal significance tests of age interactions (or effect modifications) by sex, histopathologic classification, or body site; in those models, statistical interaction is consistent with biologic interactions because of cancer heterogeneity.11 Two-component mixture models were used to identify heterogeneity in age distributions at diagnosis by determining whether 2 or more melanoma populations fit the data better than a single cancer population. Age-period-cohort (APC) models were used to assess “fitted” age-specific incidence rates adjusted for calendar-period effects (ie, screening, case ascertainment, etc) and/or for birth-cohort effects (ie, risk factor and exposure).12 Variations in the shape of the APC fitted age-at-onset curves across different CMM populations with differential exposure patterns suggested etiologic heterogeneity.

MATERIALS AND METHODS

We used the SEER 9 Registries Database (SEER 9) from the National Cancer Institute's SEER Program to examine cutaneous invasive malignant melanoma that was diagnosed during the years 1975 to 2004.13 In situ lesions were excluded from analysis. The SEER 9 tumor registries are located in Atlanta, Connecticut, Detroit, Hawaii, Iowa, New Mexico, San Francisco-Oakland, Seattle-Puget Sound, and Utah and cover approximately 10% of the US population. We analyzed data for all races combined, although whites accounted for 96.4% (n = 93,378 of 96,900) of all melanoma cases.

Demographic and Tumor Characteristics

Demographic and tumor characteristics that were analyzed included sex, age at diagnosis, race, histopathologic classification, anatomic body site, and tumor thickness in millimeters (ie, Breslow measurement). Descriptive data also were stratified by 3 10-year periods (1975-1984, 1985-1994, and 1995-2004). SEER collected all case-related data for the entire study period except for tumor thickness, which was recorded from 1988 onward.

Histopathologic classification and anatomic body site were defined using the third edition of the International Classification of Diseases for Oncology (ICD-O-3).14 Histopathologic classifications included superficial spreading melanoma (SSM) (ICD-O-3 code 8743), nodular melanoma (NM) (ICD-O-3 code 8721), lentigo maligna melanoma (LMM) (ICD-O-3 code 8742), acral melanoma (ACM) (ICD-O-3 code 8744), unclassified and/or not otherwise specified melanoma (UCM) (all remaining ICD-O-3 codes), and unknown. Anatomic body sites were categorized as the face, head, and neck (FHN) (topography codes 440-444), upper extremity (UE) (topography code 446), trunk (topography code 445; including the back, abdomen, and chest), and lower extremity (LE) (topography code 447), and all “other or unknown” body sites were combined into a single category. Anatomic body site was categorized further into chronically sun-exposed skin (FHN and UE) versus intermittently sun-exposed skin (trunk and LE).

Statistical Analysis

Age-adjusted incidence rates (2000 US standard population) with standard errors were obtained using SEER/Stat version 6.3.6 and are expressed per 100,000 person-years (or persons per year). Incidence rate ratios (IRRs) were calculated as relative risks in which a given characteristic was compared with a referent characteristic that had an assigned IRR of 1.0. IRRs were tested for statistical significance at the 95% confidence level. All statistical tests were 2-sided.

Age-adjusted incidence trends by sex were plotted on a log-linear scale by 6 5-year periods (1975-1979, 1980-1984, 1985-1989, 1990-1994, 1995-1999, and 2000-2004). Cross-sectional, age-specific incidence rates by sex were plotted on a log-linear scale by 18 age categories; ie, 17 5-year age groups (ages birth-4 years, 5-9 years, ···80-84 years) and 1 age group with all individuals aged ≥85 years. Poisson regression models were used to examine age interaction by sex. We tested for age interaction by sex (modeled as a regression spline) with the use of a likelihood ratio test.12 Statistically significant age interactions could be either quantitative (noncrossover) or qualitative (crossover). Qualitative age interactions were particularly strong surrogates for etiologic heterogeneity.11, 15

Probability density functions by sex, histopathologic classification, and anatomic site were plotted according to age at diagnosis in single years.16 Density plots revealed smoothed age distributions using a nonparametric (model-free) approach. Two-component mixture models considered the probability for a given melanoma case of being in either an early-onset or a late-onset age distribution at diagnosis, assuming a flexible parametric model for the early- and late-onset age distributions.17, 18 The key model parameter was the probability of being in the early-onset group.

APC models were used to simultaneously assess age, period, and cohort effects among men and women.19 Given the relation (year of birth) = (year of diagnosis) − (age at diagnosis), we had 22 partially overlapping 10-year birth cohorts from 1890 to 2000 (referred to by the mid-year of birth), calculated from our 6 5-year periods and the 17 5-year age groups. An important and identifiable APC parameter was the “fitted” age-of-onset curve,12 which we estimated separately for CMM in men and women. Specifically, the fitted curves provided estimates of the cross-sectional, age-specific incidence rate curves adjusted for potential period and cohort effects. For example, period effects could differ between men and women if screening trends varied by sex; whereas cohort effects might differ if the profile of CMM risk factors, such as sun exposure, changed differentially in successive cohorts of men and women.

RESULTS

Descriptive Statistics

In total, SEER 9 collected 96,900 invasive melanoma cases from 1975 to 2004 (Table 1). Melanomas were 41% more common among men than among women (ratio of men to women [IRR_MF], 1.41). Before age 40 years, age-specific incidence rates were lower for men than for women (IRR_MF < 1.0); after that age, the age-specific rates were higher for men (IRR_MF, >1.0). The IRR_MF was greater among men than among women for all tumor characteristics except for melanomas of the LE (IRR_MF, 0.39). For all melanoma cases combined, incidence rates were approximately 5-fold greater for SSM than for either NM or LMM (ie, 5.7 per 100,000 person-years compared with 1.2 or 1.1 per 100,000 person-years; P < .001 for difference). ACM was rare (0.1 per 100,000 person-years). UCM was at least as common as SSM. Melanomas were slightly more common in intermittently sun-exposed skin than in chronically sun-exposed skin (7.6 per 100,000 person-years vs 6.2 per 100,000 person-years). The most common anatomic sites were the trunk among men (6.8 per 100,000 person-years) and the LE among women (4.0 per 100,000). Greater than 80% of LMMs (80.2%; 5400 of 6735), compared with 36.7% of SSMs (14,171 of 38,574), occurred in sun-exposed skin. Malignant melanomas were more likely to be thin (≤1.0 mm).

Table 1. Descriptive Statistics for Invasive Cutaneous Malignant Melanoma Using the Surveillance, Epidemiology, and End Results 9 Registry Database, 1975-2004

	All Patients			Men			Women
Variable	No.	Rate*	SE	No.	Rate*	SE	No.	Rate*	SE	IRRMF†
SE indicates standard error; IRRM/F, incidence rate ratio of men to women; NA, not calculated and/or not applicable; SSM, superficial spreading melanoma; NM, nodular melanoma; LMM, lentigo maligna melanoma; ACM, acral melanoma; UCM, unclassified melanoma; FHN, face, head, and neck; UE, upper extremity; LE, lower extremity; Chronic FHN+UE, chronically sun-exposed skin of the face, head, neck, and upper extremities; Intermittent trunk+LE, intermittently sun-exposed skin of the trunk and lower extremities. * Rate per 100,000 person-years, man-years, or woman-years (age-adjusted to the 2000 US standard population). † For the IRRM/F, the incidence rates among men was compared with the incidence rates among women (all IRRs were statistically significant at the P < .05 level, except for ACM). ‡ Tumor thickness (or depth) was measured in mm (ie, Breslow measurement).
Total	96,900	14.6	0.05	52,333	17.6	0.08	44,567	12.5	0.06	1.41
Age, y
<20	1008	0.5	0.02	400	0.4	0.02	608	0.6	0.02	0.63
20-29	6380	5.5	0.07	2319	4.0	0.08	4061	7.0	0.11	0.56
30-39	13,476	11.9	0.10	5766	10.3	0.14	7710	13.6	0.16	0.76
40-49	17,311	18.5	0.14	8771	19.0	0.20	8540	18.0	0.20	1.05
50-59	17,931	24.9	0.19	10,383	29.6	0.29	7548	20.5	0.24	1.45
60-69	17,195	32.3	0.25	10,752	43.5	0.42	6443	22.6	0.28	1.92
70-79	14,956	41.2	0.34	9281	61.3	0.64	5675	26.9	0.36	2.28
≥80	8643	44.7	0.48	4661	73.1	1.08	3982	30.7	0.49	2.38
Race
White	93,378	16.9	0.06	50,627	20.3	0.09	42,751	14.6	0.07	1.39
Black	513	1.0	0.05	250	1.2	0.08	263	0.9	0.06	1.39
Other/unknown	3009	NA	NA	1456	NA	NA	1553	NA	NA	NA
Histopathology
SSM	38,574	5.7	0.03	19,416	6.3	0.05	19,158	5.4	0.04	1.17
NM	8098	1.2	0.01	4746	1.6	0.03	3352	0.9	0.02	1.76
LMM	6735	1.1	0.01	4260	1.6	0.03	2475	0.7	0.01	2.35
ACM	798	0.1	0.00	358	0.1	0.01	440	0.1	0.01	1.05
UCM	39,378	5.9	0.03	21,602	7.3	0.05	17,776	5.0	0.04	1.47
Other/unknown	3317	NA	NA	1951	NA	NA	1366	NA	NA	NA
Anatomic site
FHN	18,540	2.9	0.02	12,378	4.5	0.04	6162	1.7	0.02	2.67
UE	22,306	3.4	0.02	10,977	3.7	0.04	11,329	3.2	0.03	1.16
Trunk	32,157	4.8	0.03	21,144	6.8	0.05	11,013	3.1	0.03	2.22
LE	19,304	2.9	0.02	4912	1.6	0.02	14,392	4.0	0.03	0.39
Other/unknown	4593	NA	NA	2922	NA	NA	1671	NA	NA	NA
Skin sun exposure
Chronic FHN+UE	40,846	6.2	0.03	23,355	8.2	0.06	17,491	4.9	0.04	1.68
Intermittent trunk+LE	51,461	7.6	0.03	26,056	8.4	0.05	25,405	7.1	0.05	1.18
Other/unknown	4593	NA	NA	2922	NA	NA	1671	NA	NA	NA
Tumor thickness, mm‡
≤1.0 (Thin)	40,500	9.7	0.05	21,297	11.3	0.08	19,203	8.6	0.06	1.31
1.01-2.0	9328	2.3	0.02	5391	2.9	0.04	3937	1.8	0.03	1.65
2.01-4.0	5256	1.3	0.02	3185	1.8	0.03	2071	0.9	0.02	1.95
≥4 (Thick)	2772	0.7	0.01	1735	1.0	0.03	1037	0.4	0.01	2.22
Other/unknown	39,044	NA	NA	20,725	NA	NA	18,319	NA	NA	NA

For all cases combined (total), incidence rates rose during the study period (Table 2). Total rates were 84.4% (95% confidence interval, 80.9%-88.0%) higher during 1995 to 2004 than during 1975 to 1984. The rates were greater for men than for women for all periods, and this gap between the sexes widened over time (the IRR_MF rose from 1.2 during 1975-1984 to 1.5 during 1995-2004). Figure 1A also shows that age-adjusted incidence rates rose slightly faster (steeper slope) for men than for women. Table 2 indicates that the greatest percentage changes in incidence rates occurred among older individuals (ages 70-79 years; 178%), individuals with LMM (125%), and individuals with chronically sun-exposed skin (98.4%).

Figure 1. (A,B) Age-adjusted (2000 US standard population) incidence trends and age-specific incidence rates among men and women with cutaneous malignant melanoma based on data from the National Cancer Institute's Surveillance, Epidemiology, and End Results (SEER) Program (SEER 9 Registries, 1975-200413) are shown.

Table 2. Temporal Trends for Invasive Cutaneous Malignant Melanoma Using the Surveillance, Epidemiology, and End Results 9 Registry Database, 1975-2004

	1975-1984				1985-1994				1995-2004
Variable	No.	Rate*	SE	IRR†	No.	Rate*	SE	IRR†	No.	Rate*	SE	IRR†	%CH	95% CL, %
%CH indicates the percentage change in rates from 1975-1979 to 2000-2004; SE, standard error; IRR, incidence rate ratio; CL, confidence limit; NA, not calculated and/or not applicable; SSM, superficial spreading melanoma; NM, nodular melanoma; LMM, lentigo maligna melanoma; ACM, acral melanoma; UCM, unclassified melanoma; FHN, face, head, and neck; UE, upper extremity; LE, lower extremity; Chronic FHN+UE, chronically sun-exposed skin of the face, head, neck, and upper extremities; Intermittent trunk+LE, intermittently sun-exposed skin of the trunk and lower extremities. * Rate per 100,000 person-years (age-adjusted to the 2000 US standard population). † For the IRR, a given characteristic was compared with a referent characteristic that had an assigned IRR of 1.0 (all IRRs were statistically significant at the P < .05 level, with 1 exception: LE compared with FHN during 1985-1994). ‡ Tumor thickness (or depth) was measured in mm (ie, Breslow measurement).
Total cases	18,724	9.9	0.07	NA	30,881	14.0	0.08	NA	47,295	18.3	0.08	NA	84.4	80.9-88
Sex
Women	9309	9.2	0.10	1.0	14,210	12.0	0.10	1.0	21,048	15.2	0.11	1.0	65	62-68.1
Men	9415	11.0	0.12	1.2	16,671	16.9	0.14	1.4	26,247	22.7	0.14	1.5	107	98.9-115.8
Age, y
<20	272	0.4	0.02	1.0	329	0.5	0.03	1.0	407	0.6	0.03	1.0	50.4	43.4-58.5
20-29	1878	4.8	0.11	13.1	2147	5.4	0.12	10.9	2355	6.3	0.13	11.4	31	29.1-32.9
30-39	3143	10.4	0.19	28.3	5016	12.4	0.18	24.8	5317	12.5	0.17	22.7	20.2	19.4-21
40-49	3107	14.0	0.25	38.1	5577	18.4	0.25	36.8	8627	21.0	0.23	38.0	50	46.5-53.8
50-59	3813	17.5	0.29	47.8	5076	24.2	0.34	48.4	9042	30.9	0.33	56.1	76.5	68.7-85.3
60-69	3297	19.8	0.35	53.9	5730	31.2	0.41	62.4	8168	44.8	0.50	81.2	126.5	103.8-154.3
70-79	2074	21.2	0.47	57.7	4594	37.0	0.55	73.9	8288	58.9	0.65	106.6	178	124.3-254.9
≥80	1140	24.4	0.72	66.4	2412	39.0	0.80	78.1	5091	60.1	0.84	108.9	146.6	95.7-224.4
Race
White	18,158	11.2	0.09	1.0	29,714	16.2	0.10	1.0	45,506	21.8	0.10	1.0	95.3	90.6-100.3
Black	128	1.0	0.07	0.0	160	1.0	0.08	0.1	225	1.0	0.07	0.0	0	NA
Other/unknown	438	NA	NA	NA	1007	NA	NA	NA	1564	NA	NA	NA	NA	NA
Histopathology
SSM	7106	3.7	0.05	1.0	13,438	6.0	0.05	1.0	18,030	6.9	0.05	1.0	86.1	82.1-90.4
NM	2047	1.1	0.03	0.3	2757	1.3	0.03	0.2	3294	1.3	0.02	0.2	17.2	16.5-17.9
LMM	1133	0.6	0.02	0.2	1991	0.9	0.02	0.2	3611	1.4	0.02	0.2	125	113.8-137.2
ACM	0	NA	NA	NA	308	0.1	0.01	0.0	490	0.2	0.01	0.0	NA	NA
UCM	7941	4.2	0.05	1.1	11,527	5.2	0.05	0.9	19,910	7.7	0.06	1.1	83.1	79.4-87
Other/unknown	497	NA	NA	NA	860	NA	NA	NA	1960	NA	NA	NA	NA	NA
Anatomic site
FHN	3362	1.9	0.03	1.0	5808	2.7	0.04	1.0	9370	3.7	0.04	1.0	98.5	92.7-104.8
UE	4165	2.2	0.04	1.2	6906	3.1	0.04	1.2	11,235	4.4	0.04	1.2	98.3	92.7-104.2
Trunk	6053	3.2	0.04	1.7	10,451	4.7	0.05	1.7	15,653	6.0	0.05	1.6	90.7	86.1-95.5
LE	4057	2.1	0.03	1.1	6072	2.7	0.04	1.0	9175	3.5	0.04	0.9	64.6	61.8-67.6
Other/unknown	1087	NA	NA	NA	1644	NA	NA	NA	1862	0.7	NA	NA	NA	NA
Skin sun exposure
Chronic FHN+UE	7527	4.1	0.05	1.0	12,714	5.9	0.05	1.0	20,605	8.1	0.06	1.0	98.4	93.3-103.8
Intermittent trunk+LE	10,110	5.3	0.05	2.8	16,523	7.4	0.06	1.3	24,828	9.5	0.06	1.2	80.2	76.8-83.7
Other/unknown	1087	NA	NA	NA	1644	NA	NA	NA	1862	NA	NA	NA	NA	NA
Tumor thickness, mm‡
≤1.0 (Thin)	NA	NA	NA	NA	NA	NA	NA	NA	28,744	11.1	0.07	1.0	NA	NA
1.01-2.0	NA	NA	NA	NA	NA	NA	NA	NA	6315	2.5	0.03	0.2	NA	NA
2.01-4.0	NA	NA	NA	NA	NA	NA	NA	NA	3519	1.4	0.02	0.1	NA	NA
≥4 (Thick)	NA	NA	NA	NA	NA	NA	NA	NA	1923	0.8	0.02	0.1	NA	NA
Other/unknown	NA	NA	NA	NA	NA	NA	NA	NA	6794	NA	NA	NA	NA	NA

Age Incidence Patterns (Rates and Age Distributions at Diagnosis)

Consistent with Table 1, cross-sectional, age-specific incidence rates were greater for women than for men before age 40 years (Fig. 1B); after that age, the rates were greater for men. Near age 40 years, there was a qualitative (crossover or reversing) age interaction by sex (P ≈ 0 for the null hypothesis of no age interaction by sex). Crossing age-specific incidence rates reflected bimodal age distributions at diagnosis (Fig. 2A). That is, the nonparametric estimate of age distributions at diagnosis (density plots) revealed a bimodal pattern of early-onset and late-onset cancer populations in both men and women (Fig. 2A). However, men had a predominantly late-onset age distribution, whereas women had a predominantly early-onset distribution. Nonparametric age distributions also were observed by histopathologic classification (Fig. 2B) and anatomic body site (Fig. 2C). LMM and FHN had a predominantly late-onset age distribution, whereas SSM and LE had a predominantly early-onset distribution. Consequently, the combination of a man, LMM, and FHN mostly occurred among a late-onset melanoma population (Fig. 2D), whereas the combination of a women, SSM, and LE resulted in a predominantly early-onset age distribution at diagnosis.

Figure 2. Age distributions at diagnosis are shown for (A) sex, (B) histopathology, (C) anatomic site, and (D) combined variables based on the National Cancer Institute's Surveillance, Epidemiology, and End Results (SEER) Program (SEER 9 Registries, 1975-200413). The area under each density plot represents 100% of melanoma cases in which density × 100 = the percentage relative frequency distribution. SSM indicates superficial spreading melanoma; NM, nodular melanoma; LMM, lentigo maligna melanoma; ACM, acral melanoma; UCM, unclassified melanoma; Path, pathology; MLF; men + LMM + face, head, and neck; FSLE, women + SSM + lower extremity.

In all cases that we considered, 2-component mixture models fit the data better than a single melanoma population. For example, among men during the calendar period from 1975 through 1984 (Table 3), Akaike criteria values20 were larger for a mixture model (−39,596) than for a single density (−39,614), consistent with a better fit for a mixture than for a single cancer population. For men, the proportion of cases corresponding to the earlier onset mode increased during the study period, with a greater probability of being in the earlier versus later group over time, ie, the proportion of cases from the early age-at-onset distribution rose from 38.7% during 1975 to 1984, to 46% during 1985 to 1994, and to 75% during 1995 to 2004. Among women, Akaike criteria values also were larger for a mixture than for a single cancer population, although women had predominantly early-onset age distributions for all periods.

Table 3. Estimates of the Mean Age at Diagnosis (Years) for the Early-Onset Group and the Late-Onset Group Among Men and Women Using 2-Component Mixture Models

	1975-1984				1985-1994				1995-2004
Variable	Men	SE	Women	SE	Men	SE	Women	SE	Men	SE	Women	SE
SE indicates standard error; P, percent probability of being in the early-age group at the time of initial diagnosis; AIC, Akaike criteria. * AIC were used to determine the model fit (see Akaike 197320) in which the larger AIC number corresponds to the model with the best fit to the data.
Mean age, y
Early-onset group	38.9	6	43	5.9	41.8	3.6	41.7	4.7	51.6	5.7	42	6.1
Late-onset group	66.5	7.3	74.8	12.4	69.9	7.6	72.8	7.5	77.6	7.6	76.8	13.4
p early-age group	38.7%	.12	75.0%	.05	46.0%	.02	62.1%	.06	75.0%	.02	66.0%	.03
AIC*
Mixture	−39,596		−39,768		−70,028		−60,859		−97,168		−79,755
Single density	−39,614		−40,029		−70,388		−60,866		−97,622		−79,808

Age-Period-Cohort Models

Period and cohort deviations differed between men and women (P < .01 for the null hypothesis of no difference by sex). The APC “fitted” age-at-onset curves that were adjusted for period and cohort effects also were different for men and women (Fig. 3) (P ≈ 0 for the null hypothesis of no difference by sex). Similar to the cross-sectional, age-specific incidence rates (Fig. 1B), the APC fitted curves revealed crossing or reversing age-specific incidence rates near age 40 years, consistent with qualitative age interaction. The age-specific incidence rates were greater for women than for men before age 40 years, after which rates were greater for men.

Figure 3. These age-period-cohort (APC) “fitted” age-at-onset curves were adjusted for calendar-period and birth-cohort effects (Surveillance, Epidemiology, and End Results 9 Registries, 1975-200413).

DISCUSSION

More than 40 years ago, Mishima proposed 2 age-related pathways for CMMs that were linked to sun exposure (eg, melanocytic and nevocytic)21 in which chronic sun exposure was a direct risk factor for melanocytic melanomas but not for nevocytic melanomas. Melanocytic melanomas developed later in life and had slower rates of growth, metastasis, and invasion, whereas nevocytic melanomas occurred earlier in life and had faster rates of proliferation. After the publication by Mishima, Whiteman et al also suggested at least 2 melanomagenic pathways, 1 associated with chronic sun exposure and the other associated with aberrant nevi.22

Divergent CMM pathogenesis also appears to be embodied within the original histopathologic classification schemes of Clark et al and McGovern et al.23-26 Chronic sun exposure is a risk factor for LMM (ie, melanocytic melanoma) but less so for nevi-prone SSM (ie, nevocytic melanoma). LMMs develop later in life, on sun-exposed skin, and usually have slower rates of growth. SSMs occur earlier in life and commonly arise in relatively sun-protected body sites. In fact, our population-based analysis confirmed predominantly late and early age distributions at diagnosis for LMM and SSM, respectively (Fig. 2B).

However, not all dermatopathologists and clinicians acknowledge the importance of melanoma histopathologic classification.27 Unclassified melanomas accounted for 40.6% of CMMs in SEER (UCM = 39,378 of 96,900 CMMs) (Table 1). Also, melanoma subtype is not included in the current prognostic scheme of the American Joint Committee on Cancer staging system for CMM.28 Therefore, the clinical relevance of histopathology for CMM may be somewhat uncertain and/or underestimated.

Nonetheless, recent molecular studies have linked CMM histopathology with somatic and germline genetic variants, sun exposure, and anatomic body site. CMMs arising among persons with chronic sun-damaged skin are generally LMM, associated with somatic KIT mutations and p53 overexpression, and develop later in life.1, 29, 30 In contrast, CMMs that occur without sun-damaged skin are correlated with somatic BRAF or N-RAS mutations and germline MC1R variants in some populations.31-34 These lesions typically develop earlier in life among nevi-prone individuals and commonly are SSM.

In the current analysis, we used descriptive epidemiology and a structured mathematical approach to further characterize the relation between aging and CMM incidence. Results revealed a robust (reproducible and reliable) age interaction with early-onset and late-onset melanomas in the general population. Sex, histopathologic classification, and anatomic site all were age-dependent effect modifiers for CMM risk. Women and men had relative excesses of early-onset and late-onset CMMs, respectively. More specifically, early-onset melanomas were enriched by the combination of being a woman, SSM histology, and LE anatomic site; whereas late-onset patterns were enriched by the combination of being a man, LMM histology, and FHN anatomic site (Fig. 2D). Lachiewicz et al recently demonstrated a similar incidence pattern by sex and anatomic site using SEER data (2000-2004).35 Our analysis adds histopathology, a longer time frame (1975-2004), and a more comprehensive mathematical framework.

One strength of the current study was the large-scale and population-based design of the SEER database, which generally is representative of the United States population. Our descriptive observations also were supplemented with novel 2-component mixture models and “comparative” APC models. Limitations of our study included nonstandardized histopathologic typing and missing data. Indeed, nearly 41% of CMMs were not classified in SEER. However, although the lack of complete histopathologic classification was most likely because of inconsistencies in reporting and classifying CMMs, our results still revealed distinct incidence rate patterns for SSM and LMM. In fact, a systematic dermatopathologic review most likely would have enhanced our findings, which certainly reflected some “real-world” random variation among the many general pathologists contributing to SEER.

In conclusion, the current data are consistent with a large body of established clinical and emerging molecular evidence demonstrating the etiologic heterogeneity of melanoma development and tumor progression with a possible link between advancing age, solar exposure, and constitutive molecular changes. Early-onset CMMs may represent gene-sun exposure interactions that occur early (and/or intermittently) in life in more susceptible individuals and often arise from precursor lesions. The later onset melanomas may reflect the cumulative effects of lifelong (accumulated) sun exposure in somewhat less susceptible individuals. Indeed, divergent age-related cancer pathways suggest that aging is an effect modifier of melanomagenesis. Future analytical studies should be powered adequately to account for this age-dependent biologic heterogeneity for acknowledged melanoma risk factors (sex, histopathology, and anatomic site) and for suspected melanoma risk factors, such as constituent genetic variants. Arguably, the current large-scale and population-based results would have been difficult (if not impossible) to detect in smaller analytic studies; therefore, new etiologic clues still can be gleaned from descriptive and/or retrospective studies, especially when supplemented with a systematic mathematical approach.

Conflict of Interest Disclosures

Supported in part by the Intramural Research Program of the National Institutes of Health/National Cancer Institute.