International Journal of Epidemiology ©International Epidemiological Association 1990
Vol. 19, No. 1 Printed in Great Britain
Oral Contraceptive Use and Risk of Invasive Cervical Cancer LOUISE A BRINTON.* WILLIAM C REEVES,** MARIA M BREINIES,** ROLANDO HERRERO.t ROSA C DE BRITTOIM,* EDUARDO GAITAN.i FRANCISCO TENORIOj MARIANA GARCIA** AND WILLIAM E RAWLSU
Initial studies examining the relationship of oral contraceptive use to cervical cancer risk were reassuring, '•* but recent investigations have raised concern, particularly for long-term users. Issues of study design and analysis, however, are complex, with questions arising about the potential for confounding, particularly by sexual behaviour, as well as other sources of bias.56. Three prospective studies recently reported trends of increasing cervical cancer incidence with extended durations of pill usage.7'9 In one investigation,8 the incidence of cervical cancer after ten years of use was more than four times that in non-users. Although this study had more information on risk factors than many other studies, questions remain regarding the possibility of 'Environmental Epidemiology Branch, National Cancer Institute, Bethesda MD, USA. **Gorgas Memorial Laboratory, Panama City, Republic of Panama tUnidad National de Cancerologia, Caja Costarricense de Seguro Social. San Jose, Costa Rica. tlnstituto Oncologico Nacional, Republica de Panama §Division de Epidemiologia, Instituto Nacional de Cancerologia, Bogota. Colombia. ||Hospital de Oncologia Nacional, Instituto Mexicano del Seguro Social. Mexico City. Mexico. ^Molecular Virology and Immunology, Department of Pathology, McMaster University, Hamilton, Ontario, Canada. Reprint requests: Louise A Brinton. Environmental Epidemiology Branch. National Cancer Institute, Executive Plaza North, Rm 443, Bethesda. MD 20892. USA.
residual confounding. Thus, a number of case-control studies that have found persistent effects of pill usage after extensive control for a variety of risk factors are noteworthy. In most studies, Pap smear screening appeared to be a more important confounder than sexual behaviour, but even after adjustment relative risks of 1.5-1.9 persisted for users of five or more years.10"13 Despite these positivefindings,the relationship of oral contraceptives to risk of cervical cancer remains unresolved, with several studies concluding that a true causal effect is unlikely.14"'9 We had the opportunity to evaluate the relationship of oral contraceptives to invasive cervical cancer in a large case-control study conducted in four Latin American countries. Unique qualities of this study as compared with previous investigations included an extremely high response rate, a large proportion of subjects with only one lifetime sexual partner, information on the behaviour of the male partners of these women, and DNA hybridization assays to detect the proposed aetiological agents for cervical cancer, genital human papillomaviruses (HPV). METHODS The case-control study included four study sites: Panama; Costa Rica; Bogota, Colombia and Mexico City, Mexico. Cases consisted of women newly diag-
Downloaded from http://ije.oxfordjournals.org/ at University of California, Santa Barbara on July 19, 2015
Brinton L A (Environmental Epidemiology Branch, National Cancer Institute, Executive Plaza North, Rm 443, Bethesda, MD 20892, USA) Reeves W C, Brenes M M, Herrero R, de Britton R C, Gaitan E, Tenorio F, Garcia M and Rawls W E. Oral contraceptive use and risk of invasive cervical cancer. Internationa/Journal of Epidemiology 1990, 19: 4-11. A case-control study of 759 invasive cervical cancer patients and 1430 controls in Panama, Costa Rica, Colombia and Mexico enabled an evaluation of risk in relation to oral contraceptive use. Overall use was associated with a 21% nonsignificant elevation in risk, with some further increases in risk for more extensive durations of use. Although risks were similar for recent and non-recent users (RRs =1.3 versus 1.2), recent long-term users were at highest risk (RR for 5+ years use = 1.7,95% Cl 1.1-2.6). Relationships were similar for women with and without a recent Pap smear, arguing against detection bias. There was little evidence that other risk factors, including smoking and detection of human papilloma viruses (HPV), altered the effects of oral contraceptives. The risk associated with oral contraceptives was significantly increased for adenocarcinomas (RR =2.2), whereas for squamous cell tumours the effect was minimal (RR = 1.1). These results provide some support for an adverse effect of oral contraceptives on cervical cancer risk, although possibly limited only to a subpopulation of cases.
ORAL CONTRACEPTIVES AND CERVICAL CANCER
ing, diet, marital and occupational factors, and family history was ascertained. Oral contraceptive information included years of usage for each episode of use, but no attempts were made to collect details on specific brands. Interviews lasted an average of 60 minutes. Pathology information from hospital records enabled the cancers to be divided into histological subcategories. A total of 667 cancers (87.9%) were classified as squamous cell tumours and 61 as adenocarcinomas (n=41) or adenosquamous tumours (n=20). Thirty-one cases were unable to be histologically classified. Material for HPV DNA assays was collected using a cotton-tipped swab to gently scrape the surface of the cervical lesion of cases or the endocervix of controls. Swabs were suspended in phosphate buffered saline and stored at -20°C until tested. HPV DNA was detected by a filter in situ hybridization method detailed elsewhere.21 All autoradiographs were examined independently by three observers, blinded as to case-control status, and specimens recorded as positive by at least two observers were considered positive. To estimate the risk of cervical cancer associated with selected exposures, we calculated odds ratios, as approximations of relative risks (RR). Multivariate logistic regression was used to adjust for potential confounding variables,22 deriving maximum likelihood estimates of combined RRs and 95% confidence intervals (CI). Tests for trend in the logistic analyses were obtained by categorizing the exposure variable, and treating the scored variable as continuous. Logistic regression was also used to test the statistical significance of interactions. The results of conditional logistic regression23 were similar to the unmatched analyses which have been chosen for presentation. In order to adjust simultaneously for effects of number of sexual partners and age at first intercourse, virgins were eliminated from analyses, resulting in a final data set of 759 cases and 1430 controls. Cases versus controls from each study area consisted of 198 versus 348 from Panama, 192 versus 372 from Costa Rica, 214 versus 416 from Bogota and 155 versus 294 from Mexico. RESULTS The mean ages of the cases and controls were both 46.5 years. Apart from contraception, major cervical cancer risk factors identified were detection of HPV types 16 or 18 (adjusted RR = 5.1), early age at first sexual intercourse (RR = 1.7 for
Vol. 19, No. 1 Printed in Great Britain
Oral Contraceptive Use and Risk of Invasive Cervical Cancer LOUISE A BRINTON.* WILLIAM C REEVES,** MARIA M BREINIES,** ROLANDO HERRERO.t ROSA C DE BRITTOIM,* EDUARDO GAITAN.i FRANCISCO TENORIOj MARIANA GARCIA** AND WILLIAM E RAWLSU
Initial studies examining the relationship of oral contraceptive use to cervical cancer risk were reassuring, '•* but recent investigations have raised concern, particularly for long-term users. Issues of study design and analysis, however, are complex, with questions arising about the potential for confounding, particularly by sexual behaviour, as well as other sources of bias.56. Three prospective studies recently reported trends of increasing cervical cancer incidence with extended durations of pill usage.7'9 In one investigation,8 the incidence of cervical cancer after ten years of use was more than four times that in non-users. Although this study had more information on risk factors than many other studies, questions remain regarding the possibility of 'Environmental Epidemiology Branch, National Cancer Institute, Bethesda MD, USA. **Gorgas Memorial Laboratory, Panama City, Republic of Panama tUnidad National de Cancerologia, Caja Costarricense de Seguro Social. San Jose, Costa Rica. tlnstituto Oncologico Nacional, Republica de Panama §Division de Epidemiologia, Instituto Nacional de Cancerologia, Bogota. Colombia. ||Hospital de Oncologia Nacional, Instituto Mexicano del Seguro Social. Mexico City. Mexico. ^Molecular Virology and Immunology, Department of Pathology, McMaster University, Hamilton, Ontario, Canada. Reprint requests: Louise A Brinton. Environmental Epidemiology Branch. National Cancer Institute, Executive Plaza North, Rm 443, Bethesda. MD 20892. USA.
residual confounding. Thus, a number of case-control studies that have found persistent effects of pill usage after extensive control for a variety of risk factors are noteworthy. In most studies, Pap smear screening appeared to be a more important confounder than sexual behaviour, but even after adjustment relative risks of 1.5-1.9 persisted for users of five or more years.10"13 Despite these positivefindings,the relationship of oral contraceptives to risk of cervical cancer remains unresolved, with several studies concluding that a true causal effect is unlikely.14"'9 We had the opportunity to evaluate the relationship of oral contraceptives to invasive cervical cancer in a large case-control study conducted in four Latin American countries. Unique qualities of this study as compared with previous investigations included an extremely high response rate, a large proportion of subjects with only one lifetime sexual partner, information on the behaviour of the male partners of these women, and DNA hybridization assays to detect the proposed aetiological agents for cervical cancer, genital human papillomaviruses (HPV). METHODS The case-control study included four study sites: Panama; Costa Rica; Bogota, Colombia and Mexico City, Mexico. Cases consisted of women newly diag-
Downloaded from http://ije.oxfordjournals.org/ at University of California, Santa Barbara on July 19, 2015
Brinton L A (Environmental Epidemiology Branch, National Cancer Institute, Executive Plaza North, Rm 443, Bethesda, MD 20892, USA) Reeves W C, Brenes M M, Herrero R, de Britton R C, Gaitan E, Tenorio F, Garcia M and Rawls W E. Oral contraceptive use and risk of invasive cervical cancer. Internationa/Journal of Epidemiology 1990, 19: 4-11. A case-control study of 759 invasive cervical cancer patients and 1430 controls in Panama, Costa Rica, Colombia and Mexico enabled an evaluation of risk in relation to oral contraceptive use. Overall use was associated with a 21% nonsignificant elevation in risk, with some further increases in risk for more extensive durations of use. Although risks were similar for recent and non-recent users (RRs =1.3 versus 1.2), recent long-term users were at highest risk (RR for 5+ years use = 1.7,95% Cl 1.1-2.6). Relationships were similar for women with and without a recent Pap smear, arguing against detection bias. There was little evidence that other risk factors, including smoking and detection of human papilloma viruses (HPV), altered the effects of oral contraceptives. The risk associated with oral contraceptives was significantly increased for adenocarcinomas (RR =2.2), whereas for squamous cell tumours the effect was minimal (RR = 1.1). These results provide some support for an adverse effect of oral contraceptives on cervical cancer risk, although possibly limited only to a subpopulation of cases.
ORAL CONTRACEPTIVES AND CERVICAL CANCER
ing, diet, marital and occupational factors, and family history was ascertained. Oral contraceptive information included years of usage for each episode of use, but no attempts were made to collect details on specific brands. Interviews lasted an average of 60 minutes. Pathology information from hospital records enabled the cancers to be divided into histological subcategories. A total of 667 cancers (87.9%) were classified as squamous cell tumours and 61 as adenocarcinomas (n=41) or adenosquamous tumours (n=20). Thirty-one cases were unable to be histologically classified. Material for HPV DNA assays was collected using a cotton-tipped swab to gently scrape the surface of the cervical lesion of cases or the endocervix of controls. Swabs were suspended in phosphate buffered saline and stored at -20°C until tested. HPV DNA was detected by a filter in situ hybridization method detailed elsewhere.21 All autoradiographs were examined independently by three observers, blinded as to case-control status, and specimens recorded as positive by at least two observers were considered positive. To estimate the risk of cervical cancer associated with selected exposures, we calculated odds ratios, as approximations of relative risks (RR). Multivariate logistic regression was used to adjust for potential confounding variables,22 deriving maximum likelihood estimates of combined RRs and 95% confidence intervals (CI). Tests for trend in the logistic analyses were obtained by categorizing the exposure variable, and treating the scored variable as continuous. Logistic regression was also used to test the statistical significance of interactions. The results of conditional logistic regression23 were similar to the unmatched analyses which have been chosen for presentation. In order to adjust simultaneously for effects of number of sexual partners and age at first intercourse, virgins were eliminated from analyses, resulting in a final data set of 759 cases and 1430 controls. Cases versus controls from each study area consisted of 198 versus 348 from Panama, 192 versus 372 from Costa Rica, 214 versus 416 from Bogota and 155 versus 294 from Mexico. RESULTS The mean ages of the cases and controls were both 46.5 years. Apart from contraception, major cervical cancer risk factors identified were detection of HPV types 16 or 18 (adjusted RR = 5.1), early age at first sexual intercourse (RR = 1.7 for
