CancerCausesand Control,3, 547- 554
Oral contraceptive use and risk of cutaneous malignant melanoma Julie R. Palmer, Lynn Rosenberg, Brian L. Strom, Susan Harlap, Ann G. Zauber, M. Ellen Warshauer, and Samuel Shapiro (Received 16July I992; acceptedin revisedform 20 August 1992) The relation between cutaneous malignant melanoma (MM) and the use of oral contraceptives (OC) was investigated in a case-control study carried out from 1979 to 1991 among patients in hospitals and clinics in the Philadelphia (PA) and New York City (NY) metropolitan areas (United States). Cases were 615 women under age 70 who recently had been diagnosed with invasive melanoma; controls were 2,107 women of the same ages who had been treated for other conditions unrelated either to OC use or to skin diseases. The cases were categorized as severe or nonsevere based on the depth of invasion of the tumor or the presence or absence of metastases. Among the severe cases, OC use was not associated with MM: the relative risk (RR) estimate for ever-use was 1.1 (95 percent confidence interval [CI] = 0.8-1.5) and the estimate for 10 or more years of use was 1.1 (CI = 0.6-2.1). Nor was risk associated with recent use, long latency, or young age at first use. Among the nonsevere cases, ever-use of oral contraceptives was associated positively with MM (RR = 1.5, CI -- 1.1-2.4) but there was no trend with increased duration of use. The findings provide evidence against the hypothesis that OC use increases the risk of malignant melanoma. The elevated estimates among the nonsevere cases most likely reflect selection bias rather than a causal relation.
Key words:Malignant melanoma, oral contraceptives, risk factors, selection bias, USA.
Introduction The hypothesis that oral contraceptive (OC) use may influence the risk of malignant melanoma (MM) has been raised by evidence of a possible effect of endogenous female hormones on incidence and prognosis. The age-specific incidence curve for MM in women resembles the curve for breast cancer: incidence rises steeply from puberty until menopause, at which time the rate of increase slows? In contrast, among men, the increasing incidence with age is fairly constant throughout the lifespan following the onset of
puberty. Hyperpigmentation of the skin is caused by OC use as well as by pregnancy;2 however, this seems to be the result of increased melanin synthesis rather than melanocytic proliferation? Finally, for some women who already have melanoma, pregnancy has been associated with a poorer prognosis? There have been a number of epidemiologic studies of the relation of OC use to risk of cutaneous melanoma, s-~9and the results have been predominantly null. However, there have been some elevations in risk
Drs Palmer, Rosenberg, and Shapiro are with the Slone Epidemiology Unit, School of Public Health, Boston University School of Medicine, Brookline, MA, USA. Authors also are affiliated with the Clinical Epidemiology Unit, University of Pennsylvania Schoolof Medicine, Philadelphia, PA, USA (Dr Strorn), the Department of Epidemiology and Biostatistics, Memorial Sloan-Kettering Cancer Center, New York, NY, USA (Drs Harlap and Zauber), and the Department of Public Health, Cornell Medical Center, New York Hospital, New York, NY, USA (Ms Warshauer).Address correspondenceto Dr Palmer,Slone Epidemiology Unit, 1371 BeaconStreet, Brookline, MA 02146, USA. This researchwas supported in part by the US National Cancer Institute (grant R01 CA 45762). Additional support was provided by the US Food and Drug Administration (FD-U~O00082);the views expresseddo not necessarily represent the views of the Food and Drug Administration. The Slone Epidemiology Unit receivesgeneral supportfrom Hoffmann-LaRoche, Inc. and Marion-Merrell Dow, Inc. © 1992 Rapid Communications of Oxford Ltd
Cancer Causes and Control. Vol 3. 1992
547
]. R. Palmer et al noted for long-term use 5,6,s,l°,ls,17and for use that began many years before and lasted for several years. 8,~°,17 Most of the previous studies were small; the largest .3 had 333 cases. We report here on the relation of OC use to MM based on data from a case-control study in two metropolitan areas in the United States which includes 615 cases. In recent years, there have been changes in the diagnosis and treatment of MM. Whereas most cases previously were seen in hospitals, many now are treated as outpatients. There is considerable variation in the stage at which melanoma is detected and the variation may be related to educational level, health consciousness, and degree of contact with the health care system ('medicalization'). These same factors may be linked to OC use. Therefore, we designed the analysis to take into account the possibility of selection bias due to differential detection of melanoma in OC users and nonusers,
Materials and methods Data collection Our Case-control Surveillance Study 2°has been in progress since 1976 in several centers in the United States. Patients with cancer and with a wide range of other disorders are interviewed while in the hospital or, in the case of melanoma, in a hospital outpatient setting. The study provides an accessible database for the assessment of drug-disease hypotheses. To date, five percent of patients approached have refused the interview. We reported previously on the association between OC use and MM in data collected in eight US centers from July 1976 through March 1982; the study included 160 cases. 11 By 1987, data collection in all study centers except New York City (NY) and Philadelphia (PA) had ceased. The present analysis was restricted to subjects from New York and Philadelphia, interviewed from January 1978 to July 1991; these patients accounted for most of the MM patients, and this restriction also served to minimize confounding by regional differences in OC use and in referral patterns for treatment of melanoma. The present analysis includes 120 cases from the first analysis. Nurse-interviewers administered structured interviews to obtain information on demographic factors, reproductive and medical history, and habits such as cigarette smoking. Histories of OC use were elicited by questions about the use of drugs for birth control, regulation of periods, menstrual problems, endometriosis, or breast conditions, as well as a direct question about use of OCs specifically. Subjects were 548
Cancer Causes and Control. Vol 3. 1992
asked about the timing of each episode of use. Books with photos of all OC packets and pills marketed in the US were shown to the subjects to aid recall. From 1983 to 1986, melanoma patients in Philadelphia and a subset of other Philadelphia patients were also asked five questions concerning their usual sun exposure and their skin type, including tendency to burn or tan. We obtained discharge diagnoses for all subjects. Pathology reports were obtained for cancer patients and were reviewed by investigators who were blind to the subjects' exposure status; the information was used to classify melanoma patients according to metastases, histologic type, Clark's level, tumor thickness, and site of lesion.
Cases. The case group comprised 615 Caucasian patients with cutaneous MM. Their disease was diagnosed no more than 12 months previously, and they had had no other primary cancer and no previous cancer. Patients with melanoma in situ were not included. A total of 269 cases had been treated at an outpatient dysplastic nevus clinic and the remainder had been treated in the hospital. The cases ranged in age from 18 to 64, with a median age of 40 years. Information on histologic type was available for 314 cases: 238 (76 percent) had superficial spreading melanoma, 58 (18 percent) had nodular melanoma, 14 (four percent) had lentigo maligna melanoma, and four (one percent) had acral-lentiginous melanoma. Information on sun exposure and skin type was available for 185 cases.
Controls. Two control groups of Caucasian patients with malignancies or nonmalignant illnesses judged to be unrelated to OC use were selected. Because the age distribution of the potential controls differed from that of the cases, we used frequency-matching to select the final control group: we randomly selected one cancer control and 2.5 noncancer controls per case in each five-year age group. The cancer control group was made up of 610 patients with cancers of the lung, pancreas, bladder, colon, rectum, or bone, or with leukemia or lymphoma. Their cancers had been diagnosed no more than 12 months previously and they had not had a previous cancer. The noncancer control group was made up of 1,497 women admitted for trauma (465 patients), disc problems and back pain (282 patients), gastrointestinal disorders (285 patients), acute infections (170 patients), kidney stones and various other conditions (295 patients); they had no history of cancer. Overall, the controls ranged in age from 18 to 64, with a median age of 41. Information on sun exposure and skin type was available for 155 controls.
Oral contraceptives and malignant melanoma Data analysis Relative risks (RR) were estimated for various categories of O C use relative to never-use, with allowance for five-year age category, geographic region (New York City or Philadelphia), and year of interview (1979-81, 1982-84, 1985-88, 1989-91), by the MantelHaenszel method. 2t Multiple logistic regression analysis22 was used to control these and additional factors, including years of education (< 12, 12, 13-15, i> 16), cigarette smoking (ever, never), body mass index (kg/m 2) (< 20, 20-26, 1> 27), menopausal status, (pre-, post-), religion (Protestant, other), and skin reaction to the sun (always or usually burn; sometimes burn but Table 1. Duration of use of oral contraceptives (OC) among cancer and noncancer controls Duration of OC use (years)
Cancer controls
Noncancer controls
No. (%)~
No. (%)'
Never used
313 (52)
800 (53)
Ever used
Oral contraceptive use and risk of cutaneous malignant melanoma Julie R. Palmer, Lynn Rosenberg, Brian L. Strom, Susan Harlap, Ann G. Zauber, M. Ellen Warshauer, and Samuel Shapiro (Received 16July I992; acceptedin revisedform 20 August 1992) The relation between cutaneous malignant melanoma (MM) and the use of oral contraceptives (OC) was investigated in a case-control study carried out from 1979 to 1991 among patients in hospitals and clinics in the Philadelphia (PA) and New York City (NY) metropolitan areas (United States). Cases were 615 women under age 70 who recently had been diagnosed with invasive melanoma; controls were 2,107 women of the same ages who had been treated for other conditions unrelated either to OC use or to skin diseases. The cases were categorized as severe or nonsevere based on the depth of invasion of the tumor or the presence or absence of metastases. Among the severe cases, OC use was not associated with MM: the relative risk (RR) estimate for ever-use was 1.1 (95 percent confidence interval [CI] = 0.8-1.5) and the estimate for 10 or more years of use was 1.1 (CI = 0.6-2.1). Nor was risk associated with recent use, long latency, or young age at first use. Among the nonsevere cases, ever-use of oral contraceptives was associated positively with MM (RR = 1.5, CI -- 1.1-2.4) but there was no trend with increased duration of use. The findings provide evidence against the hypothesis that OC use increases the risk of malignant melanoma. The elevated estimates among the nonsevere cases most likely reflect selection bias rather than a causal relation.
Key words:Malignant melanoma, oral contraceptives, risk factors, selection bias, USA.
Introduction The hypothesis that oral contraceptive (OC) use may influence the risk of malignant melanoma (MM) has been raised by evidence of a possible effect of endogenous female hormones on incidence and prognosis. The age-specific incidence curve for MM in women resembles the curve for breast cancer: incidence rises steeply from puberty until menopause, at which time the rate of increase slows? In contrast, among men, the increasing incidence with age is fairly constant throughout the lifespan following the onset of
puberty. Hyperpigmentation of the skin is caused by OC use as well as by pregnancy;2 however, this seems to be the result of increased melanin synthesis rather than melanocytic proliferation? Finally, for some women who already have melanoma, pregnancy has been associated with a poorer prognosis? There have been a number of epidemiologic studies of the relation of OC use to risk of cutaneous melanoma, s-~9and the results have been predominantly null. However, there have been some elevations in risk
Drs Palmer, Rosenberg, and Shapiro are with the Slone Epidemiology Unit, School of Public Health, Boston University School of Medicine, Brookline, MA, USA. Authors also are affiliated with the Clinical Epidemiology Unit, University of Pennsylvania Schoolof Medicine, Philadelphia, PA, USA (Dr Strorn), the Department of Epidemiology and Biostatistics, Memorial Sloan-Kettering Cancer Center, New York, NY, USA (Drs Harlap and Zauber), and the Department of Public Health, Cornell Medical Center, New York Hospital, New York, NY, USA (Ms Warshauer).Address correspondenceto Dr Palmer,Slone Epidemiology Unit, 1371 BeaconStreet, Brookline, MA 02146, USA. This researchwas supported in part by the US National Cancer Institute (grant R01 CA 45762). Additional support was provided by the US Food and Drug Administration (FD-U~O00082);the views expresseddo not necessarily represent the views of the Food and Drug Administration. The Slone Epidemiology Unit receivesgeneral supportfrom Hoffmann-LaRoche, Inc. and Marion-Merrell Dow, Inc. © 1992 Rapid Communications of Oxford Ltd
Cancer Causes and Control. Vol 3. 1992
547
]. R. Palmer et al noted for long-term use 5,6,s,l°,ls,17and for use that began many years before and lasted for several years. 8,~°,17 Most of the previous studies were small; the largest .3 had 333 cases. We report here on the relation of OC use to MM based on data from a case-control study in two metropolitan areas in the United States which includes 615 cases. In recent years, there have been changes in the diagnosis and treatment of MM. Whereas most cases previously were seen in hospitals, many now are treated as outpatients. There is considerable variation in the stage at which melanoma is detected and the variation may be related to educational level, health consciousness, and degree of contact with the health care system ('medicalization'). These same factors may be linked to OC use. Therefore, we designed the analysis to take into account the possibility of selection bias due to differential detection of melanoma in OC users and nonusers,
Materials and methods Data collection Our Case-control Surveillance Study 2°has been in progress since 1976 in several centers in the United States. Patients with cancer and with a wide range of other disorders are interviewed while in the hospital or, in the case of melanoma, in a hospital outpatient setting. The study provides an accessible database for the assessment of drug-disease hypotheses. To date, five percent of patients approached have refused the interview. We reported previously on the association between OC use and MM in data collected in eight US centers from July 1976 through March 1982; the study included 160 cases. 11 By 1987, data collection in all study centers except New York City (NY) and Philadelphia (PA) had ceased. The present analysis was restricted to subjects from New York and Philadelphia, interviewed from January 1978 to July 1991; these patients accounted for most of the MM patients, and this restriction also served to minimize confounding by regional differences in OC use and in referral patterns for treatment of melanoma. The present analysis includes 120 cases from the first analysis. Nurse-interviewers administered structured interviews to obtain information on demographic factors, reproductive and medical history, and habits such as cigarette smoking. Histories of OC use were elicited by questions about the use of drugs for birth control, regulation of periods, menstrual problems, endometriosis, or breast conditions, as well as a direct question about use of OCs specifically. Subjects were 548
Cancer Causes and Control. Vol 3. 1992
asked about the timing of each episode of use. Books with photos of all OC packets and pills marketed in the US were shown to the subjects to aid recall. From 1983 to 1986, melanoma patients in Philadelphia and a subset of other Philadelphia patients were also asked five questions concerning their usual sun exposure and their skin type, including tendency to burn or tan. We obtained discharge diagnoses for all subjects. Pathology reports were obtained for cancer patients and were reviewed by investigators who were blind to the subjects' exposure status; the information was used to classify melanoma patients according to metastases, histologic type, Clark's level, tumor thickness, and site of lesion.
Cases. The case group comprised 615 Caucasian patients with cutaneous MM. Their disease was diagnosed no more than 12 months previously, and they had had no other primary cancer and no previous cancer. Patients with melanoma in situ were not included. A total of 269 cases had been treated at an outpatient dysplastic nevus clinic and the remainder had been treated in the hospital. The cases ranged in age from 18 to 64, with a median age of 40 years. Information on histologic type was available for 314 cases: 238 (76 percent) had superficial spreading melanoma, 58 (18 percent) had nodular melanoma, 14 (four percent) had lentigo maligna melanoma, and four (one percent) had acral-lentiginous melanoma. Information on sun exposure and skin type was available for 185 cases.
Controls. Two control groups of Caucasian patients with malignancies or nonmalignant illnesses judged to be unrelated to OC use were selected. Because the age distribution of the potential controls differed from that of the cases, we used frequency-matching to select the final control group: we randomly selected one cancer control and 2.5 noncancer controls per case in each five-year age group. The cancer control group was made up of 610 patients with cancers of the lung, pancreas, bladder, colon, rectum, or bone, or with leukemia or lymphoma. Their cancers had been diagnosed no more than 12 months previously and they had not had a previous cancer. The noncancer control group was made up of 1,497 women admitted for trauma (465 patients), disc problems and back pain (282 patients), gastrointestinal disorders (285 patients), acute infections (170 patients), kidney stones and various other conditions (295 patients); they had no history of cancer. Overall, the controls ranged in age from 18 to 64, with a median age of 41. Information on sun exposure and skin type was available for 155 controls.
Oral contraceptives and malignant melanoma Data analysis Relative risks (RR) were estimated for various categories of O C use relative to never-use, with allowance for five-year age category, geographic region (New York City or Philadelphia), and year of interview (1979-81, 1982-84, 1985-88, 1989-91), by the MantelHaenszel method. 2t Multiple logistic regression analysis22 was used to control these and additional factors, including years of education (< 12, 12, 13-15, i> 16), cigarette smoking (ever, never), body mass index (kg/m 2) (< 20, 20-26, 1> 27), menopausal status, (pre-, post-), religion (Protestant, other), and skin reaction to the sun (always or usually burn; sometimes burn but Table 1. Duration of use of oral contraceptives (OC) among cancer and noncancer controls Duration of OC use (years)
Cancer controls
Noncancer controls
No. (%)~
No. (%)'
Never used
313 (52)
800 (53)
Ever used
