CONTRACEPTION ORAL CONTRACEPTIVES AND SKIN NEOPLASIA
Adele Green, M.B., B.S., Ph.D.
Queensland Institute of Medical Research Bramston Terrace, Brisbane Queensland 4006 Australia
ABSTRACT
A possible association between oral contraception and the development of cutaneous melanoma has been raised largely because of the hyperpigmentation of pregnancy and the effect pregnancy may have on the outcome of established disease. Present evidence suggests there is no causal link between oral contraceptive (OC) use and melanoma (or with benign melanocytic nevi), nor has a specific subgroup of women or subtype of melanoma been consistently implicated as being at increased risk of this disease due to use of OCs. INTRODUCTION Oral contraception may be regarded as one of the more remarkable legacies of the second half of the twentieth century. Melanoma is also remarkable for its alarming increase during the same period, roughly doubling in incidence among white populations every 10 to 15 years, most probably as a result of a dramatic increase in sun exposure (1). Its etiology is not straightforward and additional factors require attention. The responsiveness of the skin's pigment cells to steroid hormones at a functional level is well known, e.g., hyperpigmentation occurs in pregnancy and with use of oral contraception (2), which raises the question of influences on pigment cell growth. This paper examines the evidence that there is a causal link between OC use and melanocytic skin tumors. To better judge this, the epidemiology of the tumors and biological background are summarized, possible limitations of the validity of study findings are discussed, and presentation of analytic studies is restricted to those which permit assessment of a dose-response effect by providing data on duration of OC use (cf. everuse). EPITHELIAL TUMORS OF THE SKIN Epithelial tumors, the commonest skin cancers, have not been included because there is little reason to believe on present knowledge that OCs are associated with their development. Nonmelanoma skin tumors, including benign solar keratoses, basal cell carcinoma (BCC) and squamous cell carcinoma (SCC) in pale-skinned populations are almost entirely due to solar ultraviolet radiation (WR), especially solar keratoses and SCC (1). In a minority of cases of SCC, ionizing
JUNE 1991 VOL. 43 NO. 6
663
CONTRACEPTION
radiation or chemicals are associated, and in dark-skinned populations in Asia and Africa, SCC may arise as a complication of trauma and burns BCC is probably also associated with another (unidentified) agent (3). in addition to UVR (1,4). However,there is no known epidemiologic evidence that risk of BCC, SCC or their precursors is associated with female steroid hormones in general, or exposure to OCs in particular, in any population. NEOPIASIA
nevi
OF CUTANEOUS
MEIANOCYTES
Tumors of the skin's pigment cells comprise (or common moles) and malignant melanoma.
Melanocytic Euidemiology
benign
melanocytic
nevi (5)
Melanocytic nevi are the major known risk factors for melanoma (6,7). Only 1% of white newborns manifest any congenital nevi, and the majority of nevi are considered to be acquired after birth, especially Incidence declines after young adulthood, and finally around puberty. numbers diminish and appearances change with increasing age. Apart from any specific genetic predisposition, the chief determinants of nevi are thought to relate primarily to sun-sensitivity and UVR. A light skin, a tendency to burn rather than tan, and solar associated with nevi WR dose, especially early in life, are positively (8,9). Sex hormones may also influence nevus prevalence although Pregnancy induces clinical evidence for this is circumstantial. changes in nevi, and in a study of 128 nevi removed from 86 pregnant white patients compared to 51 female controls, pregnancy was associated However,in a series with a mild degree of histopathologic atypia (10). of 389 pregnant Puerto Rican women no definite correlation with histologic atypia was found in 26 biopsied lesions that had reportedly Secondly, the apparent proliferation of altered during pregnancy (11). nevi at puberty reported in a number of cross-sectional studies (12,13) is so dramatic that it is unlikely to be explained by a cohort effect (5), and suggests an etiologic role of sex hormones in their development. Effect
of OCs on numbers
of nevi
In only 2 published studies (13,14), both from Scotland, has the relation between nevus counts in women and use of OCs been examined and the findings were null. The first study (13) was of a series of 124 women aged 20-59 years, and while those women using OCs for more than 6 months (33*34 nevi per person) had a higher mean nevus count than those who had never used it (19+9), the difference was not significant once Among the the younger age of OC users was taken into account (13). control group of a hospital-based case-control study of melanoma in Glasgow and Edinburgh (14), there was a slight (again nonsignificant) tendency for women who were ever-users to have higher numbers of nevi than never-users. In this study (14) (and most others subsequently reviewed) the strength of association between OC use and tumor
664
JUNE 1991 VOL. 43 NO. 6
CONTRACEPTION development was indicated by the magnitude of the relative risk (RR), the ratio of the risk of the tumor in OC users divided by the risk in nonusers. Compared to those with few nevi, RR for many nevi was 1.6 (95% confidence limits 0.5, 5.6) and RR for some nevi was 1.1 (0.4, 3.2). Numbers of nevi were reportedly unrelated to the duration of oral contraception, or the age or year when it was first started (14). Since complexion and sun exposure are among the strongest risk factors for nevi, these are potential confounders of any putative association with OCs, given that habits of sun exposure may be associated with contraceptive practice. In neither study was any adjustment made for these factors. Melanoma Classification Melanoma is classified microscopically according to its noninvasive growth component,i.e.,the component limited to the epidermis. The subtypes are termed nodular melanoma, which has no demonstrable noninvasive component; superficial spreading melanoma (SSM) which is the most common variety (usually accounting for 60-70% of all types in white populations); lentigo maligna melanoma (IMM) which commonly occurs on the face in older age groups; acral lentiginous (ACL) melanoma which occurs on glabrous skin of the extremities and is the predominant type in nonwhite populations; and melanoma of unclassifiable type (15,16). Invasion of the underlying dermis is usually seen clinically as a nodule and is the final common pathway, indistinguishable among subtypes. Melanoma can also be classified according to its anatomic site distribution regardless of pathologic subtype (17), and this may be more relevant from an etiologic standpoint. Among Caucasians, highest rates per unit area of body surface generally occur on the leg and the face among females and the back among males (18,19) in contrast to nonwhites. The sole of the foot is the most common site in Indians (20), Chinese, Japanese, the American black and the African black (15). Enidemiologv In genera&melanoma is a disease of white populations. Highest reported annual incidence is close to 40 per 100,000 in Queensland, Australia (21), compared to 10 per 100,000 and 5 per 100,000 in the United States (1985 SEER data) and England and Wales,respectively (22). In developing countries for which estimates are available, incidence rates vary from around 1 to 2 per 100,000 in Paraguay and Fiji, to 0.1 to 0.5 per 100,000 in regions of India, or Indonesia, for example (23). The death rate in Australia is about a quarter of the incidence rate, and in the United States with approximately 26,000 new cases of cutaneous melanoma diagnosed annually, there are 7,800 deaths each year due to the disease (24). The sex distribution varies between countries: for example,while similar frequencies in males and females are observed in Australia or the US, a clear preponderance of females is seen in Britain and of males in Geneva (25). Occurrence is rare
JUNE 1991 VOL. 43 NO. 6
CONTRACEPTION
before puberty, but thereafter the incidence rate rises steeply till about 50 years, especially among women, when the rate of increase slows. Indeed,age-specific incidence curves based on birth cohort data from Scotland (26), show that while risk increases monotonically with age in males, in females the shape closely resembles that for breast cancer with a "Clemmensen's Hook", and this phenomenon can also be found in published age-cohort data from Norway (27), from the SEER program in the USA (1981-1985) and from New Zealand (28). Solar UVR is believed to be the main environmental determinant of cutaneous melanoma (apart from melanomas of the palms and soles) and the association with high socio-economic status in white populations in the northern hemisphere is believed to result from increased leisuretime sun exposure among the affluent. There is no association between melanoma and social class in Queensland where everyone has high sun exposure (29). The sharp increase in incidence among young and middleaged adults that has occurred in the last few decades is thought to indicate a causal role of excessive UV exposure very early in life or a relatively short latent period from exposure to diagnosis (30). Sex hormones
and melanoma
An etiologic role for steroid hormones is supported by ecologic observations such as the high incidence of disease in women relative to men in many countries, and the age distribution of melanoma - in that it is rare before puberty, with the age-specific incidence curve for women showing an inflection point at menopause and a decreased slope after menopause (indeed an association between the development of melanoma and of breast cancer has been suggested (31)). If estrogen and/or progesterone induced proliferation of cutaneous melanocytes (besides simply increasing cellular production of pigment), it would suggest a potential influence of OCs on the risk of melanoma. However,a proliferative response such as observed following W stimulation (32) has not been clearly documented in humans or animals. The hyperpigmentation of sun-exposed areas of the face during pregnancy is due to increased synthesis of melanin but the number of melanocytes appears normal (2). When large doses of estrogen were administered to oophorectomized guinea pigs,there was a significant increase in melanocyte counts in the skin of the abdominal wall but not the ear, whereas small combined doses of estrogen and progesterone caused a significant increase in melanocyte counts in neither area (33). The dramatic increase in endogenous hormones during pregnancy can affect the occurrence or prognosis of melanoma (34). Case reports and patient series of pregnant women with melanoma show that rapidly fatal outcomes may occur (as may also occur in men and nonpregnant women) (34). On the other hand, there may be regression of the disease at delivery (35). However, when the studies of the association between pregnancy and melanoma are considered collectively, there appears to be little difference in the survival of women with melanoma that was in some way related to pregnancy, compared to those with melanoma that was not associated with pregnancy (34). In one study (36) women who had previously been pregnant were noted to have improved survival and it
666
JUNE 1991 VOL. 43 NO. 6
CONTRACEPTION
was hopothesised that sensitization to fetal antigens during pregnancy may result in antibodies that later inhibited progression of melanoma and enhanced survival. It has been suggested that the hormones of pregnancy adversely affect only a particular group of women with malenoma, whose characteristics are at present unknown (34). Site of lesion may be important: in 2 studies it was reported that pregnant women were more likely to have prognostically unfavourable trunk lesions (37,38), though differences in survival in one study (38) disappeared on controlling for site and stage. A third study found no difference in site among women who were pregnant when melanoma was diagnosed, and other women however (39). While the presence and concentration of estrogen and progesterone receptors in melanoma tissue has been well researched because of the influence of pregnancy on the prognosis of some melanoma patients and the possible therapeutic importance (37,40), the receptor status of normal melanocytes - of more relevance to etiology - has not. Benign nevi removed for cosmetic reasons from 17 normal persons (11 were women) had no estrogen receptor activity in one study (41), though 9 of 21 benign nevi from melanoma patients had cytoplasmic estrogen receptors (ERs) (41). In another study of "nonmalignant skin':again from melanoma patients (42), 4 out of 15 samples had detectable amounts of ER but in none were gestagen receptors detectable. Effect of OCs on risk of melanoma The effect of OCs on the development of melanoma has been addressed briefly in a descriptive report in which melanoma data from selected cancer registries in the US, Scandanavia, Canada and Europe were analysed (43). There was no perturbation of melanoma incidence or mortality rates among females relative to males at or 10 years after the period of introduction of OCs. The remainder of the studies reviewed are analytic, and their features and the main results are summarized in Tables I-III. An overview of their possible limitations is given below followed by the detailed findings. The validity of many of the published data (Tables I-III) is questionable from a number of standpoints. Because of small sample sizes, the 95% confidence intervals around the point estimates of risk of melanoma are wide and generally include l.O,i.e.,most of the observed associations could have arisen by chance. Selection of hospital in-patients as controls could introduce substantial bias if the use of OC is associated with an increased probability of being hospitalized as has been reported (44). Also the generalizability of findings of hospital-based studies may be limited, as only a proportion of melanoma patients are treated in hospital or hospital clinics. Differential error due to interviewer bias or recall bias would not usually be a problem as an association between OCs and melanoma is not widely entertained, and as a corollary, increased surveillance of the skin for pigmented lesions is unlikely to have occurred among OC users per se. The most common potential problem is the failure to account for confounding variables. Nevi are the strongest predictors of melanoma,
JUNE 1991 VOL. 43 NO. 6
667
CONTRACEPTION
Table I Sunwq
Source, Year
featuresof case-controlstudiesof cutaneousmelanoma and oral contraceptive use
Study population, Lxation
Age
No. of
clears)
Gases
Factors accountedfor
Populationbased, %ld
15 - 49
110
Age Marital status
E&in et al. (48) 1982
Married registered nurses, USA
30 - 35
141
Age, w-if-7 Region of residence Age at 1st pregnancy
Holly et al. (49) 1983
Populationbased, Seattle
37 - 74
87
Helntrich et al. (50) 1984
Hospitalbased, Boston
20 - 59
160
Beral et al. (51) 1984
Mixedhospital &comwitybased, SYd=Y
18 - 54
287
Holnun et al. (52) 1984
Populationbased, Western Australia
10 - 79
276
Green & Bain (53) 1985
Populationbased, @ensland
15 - 81
Gallagheret al. (54) 1985
Populationbased, Western Canada
20 - 69
333
Osterlindet al. (55) 1988
Populationbased, EZastDexxnark
20 - 79
280
Adam et al.
(47) 1981
666
Age Age at 1st birth
Age
Age
phenotype Sun exposure
91
Age
~l-%YPe
Sun exposure
Age phenotype
Sun exposure
JUNE 1991 VOL. 43 NO. 6
CONTRACEPTION
JUNE 1991 VOL. 43 NO. 6
CONTRACEPTION
but whether they should be included as a confounding factor is a moot point (45,46) depending on whether nevi are regarded as indicators of a high-risk genotype or rather as actual intermediates in the development of melanoma. What is certain is that some account should be taken of phenotype in relation to sun-sensitivity and of UV exposure. These could well explain weak associations between OC and melanoma. Moreover, the potential error in measurement of UV exposure is great (46), such that confounding may be inadequately controlled even after adjustment. There are no reproductive variables known to be risk factors for melanoma which should be considered a priori as confounding variables. Case-control
studies
The association between OC use and cutaneous melanoma has been There assessed in 9 published case-control studies (Tables I and II). was no significant trend of increasing risk with OC use reported (Table II) (results of trend tests were given in 3 studies (49,52,54): pvalues were 0.09, 0.25, 0.21,respectively). Long-term use was defined variously as OC use from 3+ years up to use for lO+ years (Table II). Among the longest-term users in each study,no significant elevation of Overall point estimates of effect for long-term risk was observed. users were above 1.0 in 4 studies (47,49,51,52), equal to 1.0 in 2 studies (50,55) and less than 1.0 in 3 studies (48,53,54). In the 3 studies (52,53,55),the point estimates of RR of melanoma which were adjusted for phenotype and sun exposure, were 1.1, 0.4 and 1.0, No combined analysis was attempted of the studies in respectively. which duration of OC use data were comparable because of the inability to control confounding. Although the current histologic classification of melanoma is many have speculated that the classification has etiologic sigarbitrary, nificance and that analyses of risk factors are best carried out for Three of the studies (53-55) each subtype separately if possible. excluded LMM on the basis that it is directly related to UVR and thus The general findings for all an etiologically distinct subtype. (other) melanomas combined largely reflect those for SSM, the commonest In one study (49) analysis by histologic type revealed a sigsubtype. nificant trend of increasing risk of SSM with increasing duration of However,a clear dose-response OC use (p
Adele Green, M.B., B.S., Ph.D.
Queensland Institute of Medical Research Bramston Terrace, Brisbane Queensland 4006 Australia
ABSTRACT
A possible association between oral contraception and the development of cutaneous melanoma has been raised largely because of the hyperpigmentation of pregnancy and the effect pregnancy may have on the outcome of established disease. Present evidence suggests there is no causal link between oral contraceptive (OC) use and melanoma (or with benign melanocytic nevi), nor has a specific subgroup of women or subtype of melanoma been consistently implicated as being at increased risk of this disease due to use of OCs. INTRODUCTION Oral contraception may be regarded as one of the more remarkable legacies of the second half of the twentieth century. Melanoma is also remarkable for its alarming increase during the same period, roughly doubling in incidence among white populations every 10 to 15 years, most probably as a result of a dramatic increase in sun exposure (1). Its etiology is not straightforward and additional factors require attention. The responsiveness of the skin's pigment cells to steroid hormones at a functional level is well known, e.g., hyperpigmentation occurs in pregnancy and with use of oral contraception (2), which raises the question of influences on pigment cell growth. This paper examines the evidence that there is a causal link between OC use and melanocytic skin tumors. To better judge this, the epidemiology of the tumors and biological background are summarized, possible limitations of the validity of study findings are discussed, and presentation of analytic studies is restricted to those which permit assessment of a dose-response effect by providing data on duration of OC use (cf. everuse). EPITHELIAL TUMORS OF THE SKIN Epithelial tumors, the commonest skin cancers, have not been included because there is little reason to believe on present knowledge that OCs are associated with their development. Nonmelanoma skin tumors, including benign solar keratoses, basal cell carcinoma (BCC) and squamous cell carcinoma (SCC) in pale-skinned populations are almost entirely due to solar ultraviolet radiation (WR), especially solar keratoses and SCC (1). In a minority of cases of SCC, ionizing
JUNE 1991 VOL. 43 NO. 6
663
CONTRACEPTION
radiation or chemicals are associated, and in dark-skinned populations in Asia and Africa, SCC may arise as a complication of trauma and burns BCC is probably also associated with another (unidentified) agent (3). in addition to UVR (1,4). However,there is no known epidemiologic evidence that risk of BCC, SCC or their precursors is associated with female steroid hormones in general, or exposure to OCs in particular, in any population. NEOPIASIA
nevi
OF CUTANEOUS
MEIANOCYTES
Tumors of the skin's pigment cells comprise (or common moles) and malignant melanoma.
Melanocytic Euidemiology
benign
melanocytic
nevi (5)
Melanocytic nevi are the major known risk factors for melanoma (6,7). Only 1% of white newborns manifest any congenital nevi, and the majority of nevi are considered to be acquired after birth, especially Incidence declines after young adulthood, and finally around puberty. numbers diminish and appearances change with increasing age. Apart from any specific genetic predisposition, the chief determinants of nevi are thought to relate primarily to sun-sensitivity and UVR. A light skin, a tendency to burn rather than tan, and solar associated with nevi WR dose, especially early in life, are positively (8,9). Sex hormones may also influence nevus prevalence although Pregnancy induces clinical evidence for this is circumstantial. changes in nevi, and in a study of 128 nevi removed from 86 pregnant white patients compared to 51 female controls, pregnancy was associated However,in a series with a mild degree of histopathologic atypia (10). of 389 pregnant Puerto Rican women no definite correlation with histologic atypia was found in 26 biopsied lesions that had reportedly Secondly, the apparent proliferation of altered during pregnancy (11). nevi at puberty reported in a number of cross-sectional studies (12,13) is so dramatic that it is unlikely to be explained by a cohort effect (5), and suggests an etiologic role of sex hormones in their development. Effect
of OCs on numbers
of nevi
In only 2 published studies (13,14), both from Scotland, has the relation between nevus counts in women and use of OCs been examined and the findings were null. The first study (13) was of a series of 124 women aged 20-59 years, and while those women using OCs for more than 6 months (33*34 nevi per person) had a higher mean nevus count than those who had never used it (19+9), the difference was not significant once Among the the younger age of OC users was taken into account (13). control group of a hospital-based case-control study of melanoma in Glasgow and Edinburgh (14), there was a slight (again nonsignificant) tendency for women who were ever-users to have higher numbers of nevi than never-users. In this study (14) (and most others subsequently reviewed) the strength of association between OC use and tumor
664
JUNE 1991 VOL. 43 NO. 6
CONTRACEPTION development was indicated by the magnitude of the relative risk (RR), the ratio of the risk of the tumor in OC users divided by the risk in nonusers. Compared to those with few nevi, RR for many nevi was 1.6 (95% confidence limits 0.5, 5.6) and RR for some nevi was 1.1 (0.4, 3.2). Numbers of nevi were reportedly unrelated to the duration of oral contraception, or the age or year when it was first started (14). Since complexion and sun exposure are among the strongest risk factors for nevi, these are potential confounders of any putative association with OCs, given that habits of sun exposure may be associated with contraceptive practice. In neither study was any adjustment made for these factors. Melanoma Classification Melanoma is classified microscopically according to its noninvasive growth component,i.e.,the component limited to the epidermis. The subtypes are termed nodular melanoma, which has no demonstrable noninvasive component; superficial spreading melanoma (SSM) which is the most common variety (usually accounting for 60-70% of all types in white populations); lentigo maligna melanoma (IMM) which commonly occurs on the face in older age groups; acral lentiginous (ACL) melanoma which occurs on glabrous skin of the extremities and is the predominant type in nonwhite populations; and melanoma of unclassifiable type (15,16). Invasion of the underlying dermis is usually seen clinically as a nodule and is the final common pathway, indistinguishable among subtypes. Melanoma can also be classified according to its anatomic site distribution regardless of pathologic subtype (17), and this may be more relevant from an etiologic standpoint. Among Caucasians, highest rates per unit area of body surface generally occur on the leg and the face among females and the back among males (18,19) in contrast to nonwhites. The sole of the foot is the most common site in Indians (20), Chinese, Japanese, the American black and the African black (15). Enidemiologv In genera&melanoma is a disease of white populations. Highest reported annual incidence is close to 40 per 100,000 in Queensland, Australia (21), compared to 10 per 100,000 and 5 per 100,000 in the United States (1985 SEER data) and England and Wales,respectively (22). In developing countries for which estimates are available, incidence rates vary from around 1 to 2 per 100,000 in Paraguay and Fiji, to 0.1 to 0.5 per 100,000 in regions of India, or Indonesia, for example (23). The death rate in Australia is about a quarter of the incidence rate, and in the United States with approximately 26,000 new cases of cutaneous melanoma diagnosed annually, there are 7,800 deaths each year due to the disease (24). The sex distribution varies between countries: for example,while similar frequencies in males and females are observed in Australia or the US, a clear preponderance of females is seen in Britain and of males in Geneva (25). Occurrence is rare
JUNE 1991 VOL. 43 NO. 6
CONTRACEPTION
before puberty, but thereafter the incidence rate rises steeply till about 50 years, especially among women, when the rate of increase slows. Indeed,age-specific incidence curves based on birth cohort data from Scotland (26), show that while risk increases monotonically with age in males, in females the shape closely resembles that for breast cancer with a "Clemmensen's Hook", and this phenomenon can also be found in published age-cohort data from Norway (27), from the SEER program in the USA (1981-1985) and from New Zealand (28). Solar UVR is believed to be the main environmental determinant of cutaneous melanoma (apart from melanomas of the palms and soles) and the association with high socio-economic status in white populations in the northern hemisphere is believed to result from increased leisuretime sun exposure among the affluent. There is no association between melanoma and social class in Queensland where everyone has high sun exposure (29). The sharp increase in incidence among young and middleaged adults that has occurred in the last few decades is thought to indicate a causal role of excessive UV exposure very early in life or a relatively short latent period from exposure to diagnosis (30). Sex hormones
and melanoma
An etiologic role for steroid hormones is supported by ecologic observations such as the high incidence of disease in women relative to men in many countries, and the age distribution of melanoma - in that it is rare before puberty, with the age-specific incidence curve for women showing an inflection point at menopause and a decreased slope after menopause (indeed an association between the development of melanoma and of breast cancer has been suggested (31)). If estrogen and/or progesterone induced proliferation of cutaneous melanocytes (besides simply increasing cellular production of pigment), it would suggest a potential influence of OCs on the risk of melanoma. However,a proliferative response such as observed following W stimulation (32) has not been clearly documented in humans or animals. The hyperpigmentation of sun-exposed areas of the face during pregnancy is due to increased synthesis of melanin but the number of melanocytes appears normal (2). When large doses of estrogen were administered to oophorectomized guinea pigs,there was a significant increase in melanocyte counts in the skin of the abdominal wall but not the ear, whereas small combined doses of estrogen and progesterone caused a significant increase in melanocyte counts in neither area (33). The dramatic increase in endogenous hormones during pregnancy can affect the occurrence or prognosis of melanoma (34). Case reports and patient series of pregnant women with melanoma show that rapidly fatal outcomes may occur (as may also occur in men and nonpregnant women) (34). On the other hand, there may be regression of the disease at delivery (35). However, when the studies of the association between pregnancy and melanoma are considered collectively, there appears to be little difference in the survival of women with melanoma that was in some way related to pregnancy, compared to those with melanoma that was not associated with pregnancy (34). In one study (36) women who had previously been pregnant were noted to have improved survival and it
666
JUNE 1991 VOL. 43 NO. 6
CONTRACEPTION
was hopothesised that sensitization to fetal antigens during pregnancy may result in antibodies that later inhibited progression of melanoma and enhanced survival. It has been suggested that the hormones of pregnancy adversely affect only a particular group of women with malenoma, whose characteristics are at present unknown (34). Site of lesion may be important: in 2 studies it was reported that pregnant women were more likely to have prognostically unfavourable trunk lesions (37,38), though differences in survival in one study (38) disappeared on controlling for site and stage. A third study found no difference in site among women who were pregnant when melanoma was diagnosed, and other women however (39). While the presence and concentration of estrogen and progesterone receptors in melanoma tissue has been well researched because of the influence of pregnancy on the prognosis of some melanoma patients and the possible therapeutic importance (37,40), the receptor status of normal melanocytes - of more relevance to etiology - has not. Benign nevi removed for cosmetic reasons from 17 normal persons (11 were women) had no estrogen receptor activity in one study (41), though 9 of 21 benign nevi from melanoma patients had cytoplasmic estrogen receptors (ERs) (41). In another study of "nonmalignant skin':again from melanoma patients (42), 4 out of 15 samples had detectable amounts of ER but in none were gestagen receptors detectable. Effect of OCs on risk of melanoma The effect of OCs on the development of melanoma has been addressed briefly in a descriptive report in which melanoma data from selected cancer registries in the US, Scandanavia, Canada and Europe were analysed (43). There was no perturbation of melanoma incidence or mortality rates among females relative to males at or 10 years after the period of introduction of OCs. The remainder of the studies reviewed are analytic, and their features and the main results are summarized in Tables I-III. An overview of their possible limitations is given below followed by the detailed findings. The validity of many of the published data (Tables I-III) is questionable from a number of standpoints. Because of small sample sizes, the 95% confidence intervals around the point estimates of risk of melanoma are wide and generally include l.O,i.e.,most of the observed associations could have arisen by chance. Selection of hospital in-patients as controls could introduce substantial bias if the use of OC is associated with an increased probability of being hospitalized as has been reported (44). Also the generalizability of findings of hospital-based studies may be limited, as only a proportion of melanoma patients are treated in hospital or hospital clinics. Differential error due to interviewer bias or recall bias would not usually be a problem as an association between OCs and melanoma is not widely entertained, and as a corollary, increased surveillance of the skin for pigmented lesions is unlikely to have occurred among OC users per se. The most common potential problem is the failure to account for confounding variables. Nevi are the strongest predictors of melanoma,
JUNE 1991 VOL. 43 NO. 6
667
CONTRACEPTION
Table I Sunwq
Source, Year
featuresof case-controlstudiesof cutaneousmelanoma and oral contraceptive use
Study population, Lxation
Age
No. of
clears)
Gases
Factors accountedfor
Populationbased, %ld
15 - 49
110
Age Marital status
E&in et al. (48) 1982
Married registered nurses, USA
30 - 35
141
Age, w-if-7 Region of residence Age at 1st pregnancy
Holly et al. (49) 1983
Populationbased, Seattle
37 - 74
87
Helntrich et al. (50) 1984
Hospitalbased, Boston
20 - 59
160
Beral et al. (51) 1984
Mixedhospital &comwitybased, SYd=Y
18 - 54
287
Holnun et al. (52) 1984
Populationbased, Western Australia
10 - 79
276
Green & Bain (53) 1985
Populationbased, @ensland
15 - 81
Gallagheret al. (54) 1985
Populationbased, Western Canada
20 - 69
333
Osterlindet al. (55) 1988
Populationbased, EZastDexxnark
20 - 79
280
Adam et al.
(47) 1981
666
Age Age at 1st birth
Age
Age
phenotype Sun exposure
91
Age
~l-%YPe
Sun exposure
Age phenotype
Sun exposure
JUNE 1991 VOL. 43 NO. 6
CONTRACEPTION
JUNE 1991 VOL. 43 NO. 6
CONTRACEPTION
but whether they should be included as a confounding factor is a moot point (45,46) depending on whether nevi are regarded as indicators of a high-risk genotype or rather as actual intermediates in the development of melanoma. What is certain is that some account should be taken of phenotype in relation to sun-sensitivity and of UV exposure. These could well explain weak associations between OC and melanoma. Moreover, the potential error in measurement of UV exposure is great (46), such that confounding may be inadequately controlled even after adjustment. There are no reproductive variables known to be risk factors for melanoma which should be considered a priori as confounding variables. Case-control
studies
The association between OC use and cutaneous melanoma has been There assessed in 9 published case-control studies (Tables I and II). was no significant trend of increasing risk with OC use reported (Table II) (results of trend tests were given in 3 studies (49,52,54): pvalues were 0.09, 0.25, 0.21,respectively). Long-term use was defined variously as OC use from 3+ years up to use for lO+ years (Table II). Among the longest-term users in each study,no significant elevation of Overall point estimates of effect for long-term risk was observed. users were above 1.0 in 4 studies (47,49,51,52), equal to 1.0 in 2 studies (50,55) and less than 1.0 in 3 studies (48,53,54). In the 3 studies (52,53,55),the point estimates of RR of melanoma which were adjusted for phenotype and sun exposure, were 1.1, 0.4 and 1.0, No combined analysis was attempted of the studies in respectively. which duration of OC use data were comparable because of the inability to control confounding. Although the current histologic classification of melanoma is many have speculated that the classification has etiologic sigarbitrary, nificance and that analyses of risk factors are best carried out for Three of the studies (53-55) each subtype separately if possible. excluded LMM on the basis that it is directly related to UVR and thus The general findings for all an etiologically distinct subtype. (other) melanomas combined largely reflect those for SSM, the commonest In one study (49) analysis by histologic type revealed a sigsubtype. nificant trend of increasing risk of SSM with increasing duration of However,a clear dose-response OC use (p
