Nevus density and melanoma risk in women: a pooled analysis to test the divergent pathway hypothesis

Abstract

A “divergent pathway” model for the development of cutaneous melanoma has been proposed. The model hypothesizes that melanomas occurring in people with a low tendency to develop nevi will, on average, arise more commonly on habitually sun-exposed body sites such as the head and neck. In contrast, people with an inherent propensity to develop nevi will tend to develop melanomas most often on body sites with large melanocyte populations, such as on the back. We conducted a collaborative analysis to test this hypothesis using the original data from ten case-control studies of melanoma in women (2406 cases and 3119 controls), with assessment of the potential confounding effects of socioeconomic, pigmentary, and sun exposure-related factors. Higher nevus count on the arm was associated specifically with an increased risk of melanoma of the trunk (p for trend=0.0004) and limbs (both upper and lower limb p for trends=0.01), but not of the head and neck (p for trend=0.25). The pooled odds ratios for the highest quartile of non-zero nevus count versus none were 4.6 (95% confidence interval (CI) 2.7–7.6) for melanoma of the trunk, 2.0 (95% CI 0.9–4.5) for the head and neck, 4.2 (95% CI 2.3–7.5) for the upper limbs and 3.4 (95% CI 1.5–7.9) for the lower limbs. Aggregate data from these studies suggest that high nevus counts are strongly associated with melanoma of the trunk but less so if at all of the head and neck. This finding supports different etiologic pathways of melanoma development by anatomic site.

INTRODUCTION

A “divergent pathway” model has been proposed to conceptualize the development of cutaneous melanoma.^{1, 2} The model seeks to incorporate diverse yet consistently reported phenomena relating to the development of melanoma. These include the observation that melanoma incidence rates on sun-exposed and unexposed body sites peak at different ages^3–5, the epidemiologic and experimental evidence for a ‘critical period’ of sun exposure for melanoma development^{6, 7} and the high degree of heritability of nevi, the strongest phenotypic risk factor for melanoma^{8, 9}. Thus, it is hypothesized that people predisposed to few nevi, who are assumed to have a low propensity for melanocytic proliferation, require continued high levels of sun exposure to drive the proliferation of epidermal melanocytes to cancer. In contrast, those with large numbers of nevi, who are assumed to have an inherently high propensity for melanocytic proliferation, only require sun exposure to initiate carcinogenesis, after which host factors drive melanoma development. The hypothesis predicts that the former group will be at greater risk of developing melanomas on usually sun-exposed body sites such as the head and neck, whereas the latter group will be at greater risk of developing melanoma on body sites with large numbers of melanocytic nevi, such as the back.

Several lines of evidence support this divergent etiologic pathway of melanoma development. Recent laboratory studies have typically reported consistent differences in genetic or molecular profiles of melanomas by anatomical site.^10–13 For example, Curtin et al.¹⁰ reported that the majority of melanomas on skin without chronic sun-caused damage had mutations in BRAF or N-RAS whilst the majority of melanomas on other sites had mutations in neither gene. Similarly Thomas et al.¹⁴ reported that people with large numbers of nevi on the back were more likely to have melanoma with mutations in BRAF or N-RAS than people with low nevus counts on this site. Of the epidemiological studies that have investigated the association between nevus count and melanoma risk by anatomical site, most have found differences. For example, several studies have reported that nevus counts were more strongly associated with melanomas of the trunk and legs than of the head or upper extremities.^15–19 Similarly, Bataille et al.²⁰ reported that patients with melanomas of the head and neck had fewer nevi and more solar keratoses than patients with melanoma of the trunk or legs. A recent case-control study by Randi et al.²¹ reported that the risk of melanoma of the trunk was associated not only with common nevi, but particularly with atypical nevi, which are more likely to occur on the trunk than at other anatomical sites.¹⁷ These studies contribute to the mounting evidence that nevus burden confers a greater risk of melanomas on the trunk than melanomas on the head and neck, but most have been confronted with small numbers after partitioning the cases by the anatomical site of their tumor. Thus, chance cannot be excluded as an explanation for the findings.

To test this hypothesis with greater statistical power than has hitherto been possible, we conducted a pooled analysis of ten case-control studies of melanoma in women. The pooled dataset was originally established to examine reproductive and sex hormone effects on risk of melanoma in women. Using this large data set, we investigated the consistency and strength of association between nevus burden and melanoma risk separately for each major body site. Because of emerging evidence that the MC1R genotype may modify the association between nevi and melanoma risk²², we also examined the combined effect of nevus density and hair color and freckling, phenotypic indicators of MC1R genotype, on melanoma risk.

METHODS

A more detailed description of the general methods used in our collaborative analysis is provided elsewhere^23–25 but briefly, we analyzed studies completed as of July 1994 that included melanomas diagnosed both on an inpatient and outpatient basis, collected data through a personal interview on important risk factors for melanoma (i.e., pigmentary traits and sun exposure history), and included at least 100 women with melanoma and 100 women controls. These criteria were used to minimize inter-study heterogeneity and ensure comparable study quality. Data were available for all but one study that met these conditions.²⁶ Descriptive statistics for each of the analysis variables were compared with published results and provided to the original study investigators to ensure their accuracy. Table 1 summarizes the characteristics of the ten studies that collected nevus data and that met the inclusion criteria described above.^27–36 The combined study group was predominantly white as most studies comprised over 98% white subjects. Seven of the ten studies included in the pooled analyses were population-based; three studies used hospital/clinic controls.^27–29

Table 1.

Characteristics of the ten studies included in the pooled analyses of nevus density and melanoma in women.

Study	Diagnosis Years	Age range	Geographic location	Number of:		Case Sources (Response Rate)	Histology	Control Sources (Response Rate)	Nevus Assessment
				Cases	Controls
Hospital/clinic-based
Bataille et al. (1996) 27	1989–1993	16–75	North East Thames region, UK	255	253	Hospital pathology reporting (61%)	155 SSMa 36 NMb 16 LMMc 48 Otherd	Hospital outpatients, general practice surgeries (95%)	All cutaneous nevi ≥2 mm (except genetalia, female breasts and posterior scalp); categorized 2–4 mm, 5–9 mm, ≥10 mm. (observer counted)
Elwood et al (1990) 28	1984–1986	20–79	Midlands, England	139	139	Pathology Laboratories (91%)	127 SSMa 12 NMb	Hospital inpatients and outpatients (78%)	Number of nevi ≥3 mm on each arm from wrist to shoulder; Number of raised nevi ≥3 mm; Total nevi ≥6 mm. (observer counted)
Swerdlow et al. (1986) 29	1979–1984	15–84	Scotland	113	120	Dermatology departments; Plastic surgery unit	61 SSMa 29 NMb 9 LMMc 14 Otherd	Hospital inpatients and outpatients	All melanocytic naevi >2 mm (whole body except scalp and genetalia) (observer counted)
Population-based
Green et al. (1985) 30	1979–1980	15–81	Queensland, Australia	114	114	Pathology Laboratories (97%)	79 SSMa 9 NMb 23 LMMc 3 Otherd	State electoral roll (92%)	Number of moles on the arm (observer counted)
Holly et al. (1995) 31	1981–1986	25–59	San Francisco, California, USA	452	935	Cancer Registry (79%)	355 SSMa 61 NMb 13 LMMc 23 Otherd	Random digit dialing (77%)	Number of large nevi (whole body) (self-reported)
Holman et al. (1984) 32	1980–1981	10–79	Western Australia	278	278	Pathology Laboratories (90%)	168 SSMa 14 NMb 37 LMMc 59 Otherd	Electoral roll, student roll (subjects <18 years of age) (69%)	Number of palpable nevi on the arms (observer counted)
Kirkpatrick et al. (1994) 33	1984–1987	25–65	Seattle, Washington, USA	127	145	Cancer Registry (80%)	88 SSMa 39 Otherd	Random digit dialing (80%)	Number of raised pigmented lesions on both arms from armpit to wrist (self-reported)
Langholz et al. (2000) 34	1978–1983	<65	Los Angeles, USA	341	366	Cancer Registry (79%)	251 SSMa 56 NMb 2 LMMc 32 Otherd	Neighborhood controls	Number of raised/pigmented lesions on right forearm (observer counted); Number of moles >7 mm on entire body (self-reported)
Østerlind et al. (1988) 35	1982–1985	20–79	East Denmark	280	536	Cancer Registry (92%)	187 SSMa 53 NMb 40 Otherd	Population Register (82%)	Number of nevi on the arms (large ≥5 mm, and small <5 mm) (observer counted)
Smith et al. (1998) 36	1987–1989	≥ 18	Connecticut, USA	307	233	CT Tumor Registry (76%)	195 SSMa 23 NMb 42 LMMc 23 Otherd	Random digit dialing (70%)	Number of nevi on arms and back (observer counted)

^aSSM - superficial spreading melanoma

^bNM – nodular melanoma

^cLMM - lentigo maligna melanoma

^dincludes unclassified or other melanoma

Analysis Variables

Measurement of nevus burden varied across studies; however most studies collected nevus counts on the upper limbs (Table 1). Thus our main variable for analysis was number of nevi, palpable or flat, on a single upper limb, collected in seven studies^{27–30, 34–36}. This represented the average of both arms for most studies^{27–29, 35, 36}, a single arm in one study³⁰, and doubling the lower arm count in one study³⁴. The variable was expressed both categorically (none, 1–4, 5–10, >10) and as quartiles of the distribution for the combined control group of all studies.

We also examined potential confounders including educational level, hair, eye and skin color, freckling, family history of melanoma in a first degree relative, sun exposure history, and skin sensitivity to sunlight exposure. Whenever possible we used standardized groupings.^{24, 37, 38} Described in detail in an earlier report²⁴, questions relating to sun exposure history varied considerably across studies. Therefore, we included the sunlight-related factors most strongly related to melanoma risk within each study. These variables included history of sunburns, sun exposure, and, for two studies, age at migration to Australia.²⁴

Statistical Analysis

We used a two-stage method of analysis to obtain study-specific odds ratios (ORs) and pooled odds ratios (pORs) and 95% confidence intervals (CIs).²³ In the first stage, each study was analyzed according to its original design unless otherwise specified. For pair-matched studies, we used conditional logistic regression and for frequency matched studies, we used unconditional logistic regression and stratified by age (<35, 35–44, ≥45 years). To evaluate inter-study variability, we examined the study- specific ORs and tested for statistical heterogeneity using a chi-square test. The pooled exposure effect was estimated in a second-stage linear, mixed model as the average of the study-specific ORs and standard errors (SEs), weighted by the inverse marginal variances. The marginal variance is the sum of the individual study variances and the random study effect. In the absence of heterogeneity, the average of the individual study estimators was weighted by the inverse of the study-specific variances alone.²³ In the presence of heterogeneity, the pooled estimator is the average of individual study estimators weighted by the sum of the individual study variances and the random effects as described by Stukel et al..²³ We examined the data for potential sources of heterogeneity by stratifying on type of control group (population versus hospital-based controls) and the way in which nevus counts were conducted (self versus investigator).

We evaluated the potentially confounding effects of several factors including hair, eye and skin color, freckling, family history of melanoma in a first degree relative, ethnicity, skin sensitivity to sun exposure, sun exposure history and educational level. All estimates were age-adjusted. To assess the impact of other potentially confounding factors, we examined the percent change in the age-adjusted pOR with the addition of each factor. Variables resulting in a 10% or greater change in the pooled estimate were included in the final models. To assess stratum-specific effects (e.g., age category or anatomical site), we broke the pairs and stratified on age (e.g., <35, 35–49, ≥45 years). Tests for trend were based on continuous variables.

Our main analyses were based on all melanomas combined. Additionally, we separately computed odds ratios for each of the primary exposure variables by body site (head and neck, trunk, lower limbs, upper limbs), age (<50, ≥50 years) and histologic subtype (lentigo maligna melanoma (LMM) and other melanoma). In the analysis of LMM we broke the pair-matched sets to increase statistical power and analyzed the studies adjusting for the original age categories using conditional logistic regression. We tested for heterogeneity in the association between the number of nevi and melanoma risk by anatomic site of melanoma using a two-stage test of inequality similar to the two-stage models used for other analyses.²³ In the first stage, the relationship between melanoma risk and nevi was estimated separately for each study and each stratum of anatomic site, where number of nevi was analyzed as a continuous variable; models were adjusted for age, skin type and family history. This produced a separate slope for each study-stratum combination. In the second stage, analysis of variance was used to construct an F-test for anatomic site differences in the stratum-specific slopes.

To examine the combined effect of nevus density and the MC1R genotype we created a proxy variable based on assessments of hair color and freckling density. Participants were re-categorized into three groups as follows: 1=no freckling and no red hair; 2=freckling and no red hair; 3=red hair (with or without freckling). For these analyses the reference group for comparison was women with no nevi, and without red hair or freckling. Analyses were conducted using SAS version 9.1 (SAS Institute, Cary, North Carolina, USA).

RESULTS

Distribution of cases by age and anatomical site

Table 2 describes the mean age and distribution of melanoma cases by anatomic site. For seven^{28–30, 32, 34–36} of the nine^{27–32, 34–36} studies with available data, women with melanomas of the head and neck were on average older than those who developed melanoma of the trunk. For the combined dataset, women with melanomas of the head and neck were considerably older than those who developed melanoma of the trunk or limbs (51.8 vs 45.1 years; p<0.0001). The mean age of women with melanoma of the lower or upper limbs was 47.5 and 47.3 years respectively; these women were significantly younger than women with melanoma of the head and neck, and significantly older than women with melanoma of the trunk (p<0.0001).

Table 2.

Distribution and mean age of cutaneous melanoma cases for each study included in the pooled analysis by anatomical site

Study	Head & Heck		Trunk		Lower Limbs		Upper Limbs		Total†
	n (%)	Mean age (SD)*	n (%)	Mean age (SD)*	n (%)	Mean age (SD)*	n (%)	Mean age (SD)*	n (%)	Mean age (SD)*
Hospital/clinic-based
Bataille et al. (1996) 26	59 (25.4)	49.0 (16.1)	142 (61.2)	52.3 (13.5)	12 (5.2)	53.2 (11.4)	10 (4.3)	53.9 (16.1)	255	50.5 (14.3)
Elwood et al (1990) 27	14 (10.1)	54.5 (17.2	19 (13.7)	49.9 (12.6)	80 (57.6)	44.3 (12.5)	26 (18.7)	47.3 (14.8)	139	49.1 (13.8)
Swerdlow et al. (1986) 35	10 (8.9)	54.9 (20.3)	14 (12.4)	43.6 (12.4)	71 (62.8)	51.5 (14.5)	16 (14.2)	52.6 (16.2)	113	51.2 (15.2)
Population-based
Green et al. (1985) 28	26 (22.8)	58.0 (16.1)	20 (17.5)	38.2 (13.7)	40 (35.1)	45.5 (13.6)	28 (24.6)	46.8 (15.6)	114	47.4 (15.9)
Holly et al. (1995) 29	40 (8.9)	40.4 (10.3)	115 (25.4)	38.4 (9.1)	167 (37.0)	43.3 (9.8)	122 (27.0)	42.5 (9.6)	452	41.7 (9.8)
Holman et al. (1984) 30	52 (18.7)	53.6 (14.1)	60 (21.6)	36.9 (13.7)	107 (38.5)	43.9 (17.0)	58 (20.9)	48.3 (16.0)	278	45.2 (16.5)
Kirkpatrick et al. (1994) 31 §									127	46.0 (10.8)
Langholz et al. (2000) 32	23 (6.7)	43.1 (13.5)	104 (30.5)	41.6 (12.4)	106 (31.1)	45.1 (11.7)	107 (31.4)	42.9 (12.9)	341	43.2 (12.5)
Østerlind et al. (1988) 33	20 (7.1)	55.3 (13.7)	76 (27.1)	48.5 (13.8)	148 (52.9)	51.0 (13.7)	36 (12.9)	56.2 (15.2)	280	51.3 (14.1)
Smith et al. (1998) 34	41 (13.4)	62.5 (19.7)	96 (31.3)	50.4 (14.7)	91 (29.6)	51.4 (15.8)	67 (21.8)	55.6 (17.1)	307	53.5 (16.7)

^*Standard Deviation

^†Numbers may not sum to total as for some cases site was ‘other’ or unclassified

^§Anatomic site data not available for this study

Upper limb nevus counts and melanoma risk by anatomical site

A total of 1545 cases and 1754 controls from seven studies^{27–30, 34–36} were included in the primary analysis of nevus count on the upper limb. Of the 10 studies included in the collaborative analyses that collected nevus counts (Table 1), these seven studies made an upper limb count of all nevi, both raised and flat. The distribution of melanocytic nevi (single upper limb) for cases and controls by study is presented in Table 3. High nevus counts (>10) were most common in the study conducted in Los Angeles (52.4%)³⁴ and least common in the study from Denmark (1.4%)³⁵. As shown in Table 4 increasing numbers of nevi (single upper limb) were strongly associated with melanoma occurring on the trunk and upper and lower limbs, but less strongly with those occurring on the head and neck; the pOR for risk of melanoma for women in the highest quartile of none-zero nevus count compared with women with no nevi was 4.6 (95%CI 2.7–7.6) for melanoma of the trunk and 3.4 (95%CI 1.5–7.9) and 4.2 (95%CI 2.3–7.5) for the lower and upper limbs respectively compared with 2.0 (95%CI 0.9–4.5) for melanoma of the head and neck. The test for inequality of pORs between these sites did not reach statistical significance, however (p=0.15). We observed evidence for heterogeneity between studies in the effect of nevus-count categories on risk for melanomas of the upper limbs. This heterogeneity was explained by higher risk estimates from the study by Green et al..³⁰ Inter-study heterogeneity was also evident in the effects of several strata of the quantile nevi variable on risk for melanoma of the lower limbs; this was explained by lower estimates from the study by Swerdlow et al..²⁹

Table 3.

Distribution of palpable and flat melanocytic nevi on a single upper limb for cases and controls and adjusted^a odds ratios for melanoma in relation to number of nevi by study^†

Study	Nevus category	Cases N (%)	Controls N (%)	Odds Ratio (OR)
Hospital/clinic-based
Bataille et al. (1996) 26	none	15 (5.9)	17 (6.7)	1.0
	1–4	94 (36.9)	121 (47.8)	0.9 (0.4–2.0)
	5–10	64 (25.1)	62 (24.5)	1.5 (0.7–3.5)
	>10	82 (32.2)	53 (20.1)	2.1 (0.9–5.1)
Elwood et al (1990) 27	none	46 (33.1)	74 (53.2)	1.0
	1–4	60 (43.2)	50 (36.0)	2.6 (1.3–5.2)
	5–10	20 (14.4)	9 (6.5)	3.8 (0.9–16.9)
	>10	13 (9.4)	7 (4.3)	4.8 (0.6–41.4)
Swerdlow et al. (1986) 35	none	31 (27.9)	30 (25.6)	1.0
	1–4	30 (27.0)	67 (57.3)	0.4 (0.2–0.9)
	5–10	21 (18.9)	9 (7.7)	2.9 (1.0–8.0)
	>10	29 (26.1)	11 (9.4)	3.9 (1.5–10.9)
Population-based
Green et al. (1985) 28	none	23 (20.2)	79 (69.3)	1.0
	1–4	61 (53.5)	21 (18.4)	11.7 (3.5–39.3)
	5–10	20 (17.5)	11 (9.7)	12.8 (2.0–82.2)
	>10	10 (8.8)	3 (2.6)	-
Langholz et al. (2000) 32	none	45 (13.2)	100 (27.5)	1.0
	1–4	71 (20.9)	89 (24.5)	1.8 (1.1–3.0)
	5–10	69 (20.3)	63 (17.3)	2.4 (1.5–3.9)
	>10	155 (45.6)	112 (30.8)	3.0 (2.0–4.7)
Østerlind et al. (1988) 33	none	126 (45.2)	346 (64.8)	1.0
	1–4	137 (49.1)	177 (33.2)	2.1 (1.6–2.9)
	5–10	13 (4.7)	8 (1.5)	5.0 (1.9–12.9)
	>10	3 (1.1)	3 (0.6)	2.9 (0.6–14.8)
Smith et al. (1998) 34	none	54 (17.6)	92 (39.5)	1.0
	1–4	153 (42.8)	100 (42.9)	2.1 (1.3–3.4)
	5–10	50 (16.3)	22 (9.4)	3.6 (1.7–7.5)
	>10	50 (16.3)	19 (8.2)	4.5 (2.2–9.4)

^aAdjusted for age, skin type and family history of melanoma

^†Three studies did not collect information on this nevus variable

Table 4.

Adjusted^a pooled odds ratios (95% confidence intervals) for melanoma in women in relation to number of nevi, stratified by anatomical site of melanoma

	Head & Neck b			Trunk b			Lower Limbs b			Upper Limbs b
	Cases	Controls	p OR(95% CI)	Cases	Controls	p OR (95% CI)	Cases	Controls	p OR (95% CI)	Cases	Controls	p OR (95% CI)
Any nevi, palpable or flat, single upper limb
None (ref)	48	738	1.0	74	738	1.0	148	721	1.0	64	738	1.0
1–4	83	625	1.7 (1.0–2.8)	179	625	2.1 (1.4–3.1)	225	504	2.3 (0.6–9.3) *	98	625	1.7 (0.6–4.7) *
5–10	32	167	1.8 (0.7–4.7)	90	184	4.0 (2.2–7.2)	77	122	4.0 (1.6–10.1) *	49	114	3.9 (1.9–8.3)
>10	30	204	1.9 (0.8–4.9)	127	207	5.0 (2.7–9.3)	84	154	2.8 (1.6–4.9)	78	204	4.8 (2.5–9.5)
None (ref)	37	659	1.0	71	629	1.0	144	642	1.0	58	659	1.0
Q1	35	348	1.3 (0.6–2.9)	87	317	1.6 (0.9–2.9)	79	276	1.2 (0.4–3.5) *	43	348	1.3 (0.6–2.4)
Q2	25	125	1.3 (0.5–3.1)	62	141	1.9 (1.0–3.4)	49	141	1.4 (0.8–2.5)	31	74	2.1 (1.1–4.2)
Q3	23	131	1.7 (0.6–4.3)	94	134	2.8 (1.5–5.2)	78	134	2.2 (0.6–8.2) *	37	131	1.8 (0.8–4.1)
Q4	35	252	2.0 (0.9–4.5)	152	252	4.6 (2.7–7.6)	144	196	3.4 (1.5–7.9) *	90	252	4.2 (2.3–7.5)
Per unit increase in nevus number	167	1640	1.01 (0.98–1.05)	450	1640	1.02 (1.00–1.05)	261	1640	1.01 (1.00–1.02)	506	1640	1.01 (1.00–1.02)
			trend p=0.25			trend p=0.0004			trend p=0.01			trend p=0.01

^*Significant heterogeneity, random effects model used (see text).

^aAdjusted for age, skin type and family history of melanoma

^bNumbers may not sum to total because of missing data; when there are zero cells in a stratum for a study, that study drops from the pooled analysis for that stratum

We conducted further analyses stratified by hair color and freckling (as a proxy for MC1R genotype status) using the 3-category composite variable: no red hair and no freckling; no red hair and freckling; and red hair with or without freckling. Four studies contributed data to these analyses.^{28, 30, 34, 35} A test for heterogeneity in nevus counts across strata of the red hair and freckling variable was not statistically significant for cases (p=0.53) but was for controls (p=0.02). Table 5 considers the combined effect of nevus density and hair color and freckling. The pOR for risk of melanoma associated with high nevus counts in women without red hair or freckling was 4.5 (95%CI 1.3–15.7; highest quartile vs none), 7.3 (95%CI 2.7–19.7) for women without red hair but with some freckling, and 14.4 (95%CI 2.0–102.0) for women with red hair. We also examined odds ratios for melanoma with number of nevi stratified by age (<50, ≥50) and histologic subtype of melanoma (LMM vs non-LMM) and found no important heterogeneity between strata of either variable (data not presented). The LMM subtype is widely known to be associated with cumulative sun exposure, and frequently occurs on the head and neck in older adults³⁹; 26.8% of head and neck melanomas in our pooled analyses were of the LMM subtype.

Table 5.

Adjusted^a pooled odds ratios (95% confidence intervals) for melanoma in women in relation to number of nevi, and stratified by hair color/freckling^†

	All Women b			No red hair & no freckling b			Red hair +/− freckling b			Freckling but no red hair b
	Cases	Controls	p OR(95% CI)	Cases (%)	Controls (%)	p OR (95% CI)	Cases (%)	Controls (%)	p OR (95% CI)	Cases (%)	Controls (%)	p OR (95% CI)
Any nevi, palpable or flat, single upper limb
None (ref)	340 (22.1)	738 (42.1)	1.0	74 (25.7)	294 (52.8)	1.0	38 (28.8)	54 (56.3)	2.6 (1.1–6.3)	127 (27.9)	248 (50.5)	1.9 (1.1–3.4)
1–4	606 (39.5)	625 (35.7)	1.9 (0.7–5.0)	98 (36.4)	143 (25.7)	2.6 (1.4–4.9)	57 (43.2)	25 (26.0)	7.2 (2.6–19.4)	173 (37.9)	168 (34.2)	3.9 (2.2–6.8)
5–10	257 (16.7)	184 (10.5)	2.9 (2.0–4.3)	35 (13.0)	49 (8.8)	3.2 (0.7–14.6)	16 (12.1)	11 (11.5)	5.1 (0.6–43.9)	66 (14.5)	30 (6.1)	9.2 (3.6–23.4)
>10	332 (21.6)	204 (11.7)	3.2 (2.1–4.8)	62 (23.1)	71 (12.8)	4.2 (0.9–19.9)	21 (15.9)	6 (6.3)	16.4 (5.0–54.1)	90 (19.7)	45 (9.2)	8.9 (3.1–25.4)
None (ref)	317	659	1.0	60	231	1.0	25	18	3.4 (0.3–42.7) *	119	234	1.9 (0.8–4.2)
Q1	259	348	1.2 (0.5–3.0)	39	90	1.7 (0.5–5.2)	25	11	4.8 (0.9–26.7)	61	91	2.3 (0.9–6.5)
Q2	173	194	1.6 (1.1–2.3)	18	33	2.0 (0.5–7.7)	11	8	5.8 (0.6–57.5)	43	21	3.9 (1.1–13.4)
Q3	238	187	2.2 (1.5–3.2)	24	27	4.4 (1.0–19.5)	12	4	7.3 (0.6–82.2)	30	15	5.4 (1.5–19.5)
Q4	444	252	3.7 (2.6–5.1)	68	68	4.5 (1.3–15.7)	24	8	14.4 (2.0–102.0)	127	66	7.3 (2.7–19.7)
			trend p<0.001			trend p=0.001			trend p=0.06			trend p=0.004

^†Reference group for comparison is no nevi, no red hair and no freckling

^*Significant heterogeneity, random effects model used (see text)

^aAdjusted for age, skin type and family history of melanoma

^bNumbers may not sum to total because of missing data; when there are zero cells in a stratum for a study, that study drops from the pooled analysis for that stratum

A smaller subset of data was available for other measures of nevus density including number of nevi (any size), palpable or flat, whole body^{27, 29, 34}; number of nevi (any size), palpable only, single upper limb^{28, 32, 33, 36}; large size nevi, palpable or flat, single upper limb^{27, 28, 34–36}; and large size nevi, palpable or flat, whole body^{27–29, 31}; however the results of these analyses have been previously reported by the individual studies and due to the restricted sample sizes pooled analyses were not conducted.

DISCUSSION

We tested the hypothesis that a ‘divergent pathway model’ exists whereby melanomas at different body sites arise through different causal mechanisms. The model builds upon epidemiologic⁶ and experimental⁷ observations relating to the development of melanoma and predicts that melanomas occurring in individuals predisposed to few nevi will tend to arise more commonly on habitually sun-exposed body sites such as the head and neck, whereas in nevus-prone individuals melanomas will occur most often on body sites with large numbers of melanocytic nevi, such as the back.

Our findings accord with this model. As expected, and in agreement with evidence from population-based sources^40–42, we found that patients with melanomas of the head and neck were considerably older than those who developed melanoma of the trunk or limbs, consistent with a pathway of more continuous cumulative sun exposure. Secondly, we found that high nevus counts conferred systematically higher relative risks of melanomas of the trunk than melanomas of the head and neck, consistent with the ‘nevus prone’ pathway being more common for melanomas of the trunk. A higher relative risk also was observed to a lesser extent for melanoma of the limbs, whose sun exposure pattern presumably lies between that of the head and neck and trunk, depending on sub-site. We cannot exclude the possibility, however, that this pattern arose simply by chance, even with the large numbers of subjects in this pooled analysis.

We attempted to address gene-environment interaction by re-categorizing study participants on the basis of phenotypic characteristics that are highly penetrant amongst people with MC1R polymorphisms. Thus, we assumed that people with red hair and/or many freckles were likely to be MC1R variant carriers.⁴³ The risk of melanoma associated with high nevus counts was higher for women with phenotypic features consistent with the presence of MC1R variants (red hair and freckling) compared with women who did not have these phenotypes. Thus, although high nevus counts were associated with an increased risk of melanoma for all women, our analyses would suggest that women with high nevus counts and red hair have substantially higher risks of melanoma than women who have high nevus counts but do not have red hair or freckling. Large genetic studies will be required to confirm whether the association between nevus counts and site-specific melanoma risk is modified by MC1R genotype.

The large number of cases and controls made possible by pooling data from seven individual case-control studies was a strength of our study. The large sample size increased our statistical power to examine associations between nevus density and melanoma across anatomical sites while adjusting for important potential confounders including pigmentary traits and sun exposure history. It also allowed sub-group analyses to examine the effects by age, phenotype and histologic subtype. Several limitations should be considered in interpreting the results. Due to the heterogeneity in defining and collecting information on number of nevi across studies, our primary analysis was based on nevus counts restricted to the arms as this was the measure collected by the greatest number of studies. Nevus counts on the arms have been shown to be reliable predictors of whole-body nevus counts.^{44, 45} The studies contributing to the summary estimates were vulnerable to various types of bias. Misclassification of nevus size and counts is possible, particularly taking into consideration heterogeneity in the way nevi were defined and counted by the individual studies included in the pooled analyses. Such misclassification, however, is likely to be non-differential with respect to anatomical site of the melanoma. Moreover, as the specific hypothesis being tested here was not known to study personnel or subjects, information bias cannot explain the observed associations.

Non-response is another potential source of bias; however most studies included in the pooled analysis reported response rates greater than 75% for cases and controls (Table 1) and it is therefore unlikely that non-response could have resulted in appreciable bias. Finally, our pooled analyses excluded men because the original collaborative pooling project study aims related to factors associated with female sex steroids. Both melanocytic nevi and cutaneous melanoma display different anatomic distributions in men and in women^{3, 27, 46–49}, and it is therefore important that the divergent pathway hypothesis is formally tested in men also.

Our results generally agree with the published literature including both case-control^{15–18, 20, 21} and prospective studies¹⁹ that have reported stronger associations between high nevus counts and melanomas of the trunk and limbs than of the head and neck. Understanding these concepts may be important not only for understanding melanoma etiology, but also survival⁴¹, and thus may have implications for improving patient care.

Further testing of the “divergent pathway” model is possible using epidemiologic and other approaches. Several studies have reported that melanomas arising on the trunk are more likely to contain remnants of a pre-existing nevus than melanomas of the head and neck^{50, 51}, suggesting that nevi might be ‘precursors’ for at least a subset of truncal melanomas. This finding needs to be confirmed in a large population-based study. With respect to melanoma risk, it must also be considered that anatomical site is potentially confounded by sun exposure, with the head and neck being continually exposed to the sun whereas the back and much of the limbs are intermittently exposed. It has been proposed that continuous sun exposure might lead to the elimination of nevi.⁵² One test of this alternative hypothesis would be to compare the body site distributions of nevi in adults from low versus high ambient UV environments. Sunlight is also known to exert immunological effects on the skin, and the magnitude of these effects could vary between individuals. Finally, the ‘divergent pathway’ model rests on the assumption that human melanocytes in vivo proliferate in response to ultraviolet radiation^{53, 54}, and that the degree of proliferation is predicted by nevus phenotype. This central claim remains untested.