- Email: [email protected]
Intrauterine device and cervical cancer: we need more evidence
Correspondence
Patients with PID (n=11 549) Controls (n=97 985) Number of patients with ovarian cancer*
8
Incidence per 10 000 person-years
3·13
28 1·17
Crude HR (95% CI)
2·69 (1·23–5·90)
1·00
Adjusted HR† (95% CI)
2·53 (1·15 –5·57)
1·00
PID=pelvic inflammatory disease of the upper genital tract (International Classification of Diseases, ninth revision, clinical modification 614 and 615). HR=hazard ratio. *Within 3 years of follow-up. †Adjusted for age, cardiovascular disease, diabetes mellitus, chronic liver disease, rheumatic diseases, monthly income, urbanization level, and endometriosis.
Table: Crude and adjusted risks of ovarian cancer in patients with upper genital tract PID aged 15–44 years
than expected. If we exclude patients coded as ICD-9-CM code 616 and restrict age to 15–44 years, 11 549 women would be classified as having PID (table), which equates to a 3·3% prevalence of PID. The adjusted hazard ratio (HR) of ovarian cancer in patients with upper genital tract infections was higher than in the controls (table). Although oral contraceptives reduce the risks of ovarian cancer and PID, a history of oral contraceptive use also seems to have its risks. As Cheong stated, PID usually occurs in young, sexually active women, and we believe that women with a long history of taking oral contraceptives are more sexually active than are those who have never taken them.2 Therefore, extended use of oral contraceptives could increase the risk of ovarian cancer. Oral contraceptives are one of the few hormone drugs that are, in general, available over the counter; however, the use of such drugs might not be as frequent as thought. In 2002, Rossouw and colleagues3 showed that there were risks associated with hormone use and, after publication of this article, the general public was reluctant to take hormone drugs without consulting their doctors first. Because the data in our study were obtained between 2004 and 2005, the chance of wrongly classifying patients as having no history of oral contraceptive use in our study is small. We appreciate R van de Laar and L F Massuger’s feedback. In our study,
only ICD-9-CM code 614·2 would include patients with tubo-ovarian abscess, and of the 608 patients with abscesses, only one developed ovarian cancer. Primary carcinoma of the fallopian tube and primary ovarian carcinoma rarely mimic tuboovarian abscess clinically, but the inclusion of patients with ICD-9-CM code 614·2 could still have affected our final results. After we excluded such patients, analyses resulted in an adjusted HR of 1·85 (95% CI 1·22–2·83; p=0·001), and after we excluded patients coded as 614·2 and 616, analyses resulted in an adjusted HR of 2·55 (95% CI 1·35–4·84; p=0·004). Thus, these adjustments do not affect our conclusion that there is an association between PID and ovarian cancer. We took various factors into consideration while designing this study—eg, we included only patients with two or more incidences of PID, and excluded cases of ovarian cancer that occurred within 6 months of the last episode of PID. As for the period of time for inflammation to develop into cancer, 1–3 years is still plausible.4 We understand van de Laar and Massuger’s concern about the results of the patients’ ovarian cancer treatment; however, the focus of our study was solely on the relation between PID and ovarian cancer and we do not have any data for patients’ response to treatment. We declare that we have no conflicts of interest.
www.thelancet.com/oncology Vol 12 December 2011
Hui-Wen Lin, Shiyng Yu Lin, Wei Zer Lin, Wei Li Lin, *Yuen-Liang Lai [email protected] Department of Mathematics, Soochow University, Taipei, Taiwan (H-WL); Biostatistics and Research Consultation Centre (H-WL), School of Medicine (SYL, WLL), Graduate Institute of Humanities in Medicine (Y-LL), Department of Radiation Oncology, Shuang Ho Hospital (Y-LL), Taipei Medical University, Zhonghe City, Taipei County 235, Taiwan; Department of Family Medicine, Taipei Medical University Hospital, Taipei, Taiwan (SYL); School of Medicine, Chung Shan Medical University, Taichung, Taiwan (WZL); and Mackay Medical College, Taipei, Taiwan (Y-LL) 1
2
3
4
US Centers for Disease Control and Prevention. Classification of diseases, functioning, and disability. http://www.cdc.gov/nchs/icd.htm (accessed Sept 15, 2011). Lin HW, Tu YY, Lin SY, et al. Risk of ovarian cancer in women with pelvic inflammatory diseases: a population-based study. Lancet Oncol 2011; 12: 900–04. Rossouw JE, Anderson GL, Prentice RL, et al. Risks and benefits of estrogen plus progestin in healthy postmenopausal women: principal results from the Women’s Health Initiative randomized controlled trial. JAMA 2002; 288: 321–33. Wei JJ, William J, Bulun S. Endometriosis and ovarian cancer: a review of clinical, pathologic, and molecular aspects. Int J Gynecol Pathol 2011; 30: 553–68.
Intrauterine device and cervical cancer: we need more evidence In their study on intrauterine device (IUD) use and risk of cervical cancer published in The Lancet Oncology, Xavier Castellsagué and colleagues1 claimed that their work provides evidence that IUD use might reduce the risk of developing cervical cancer. The authors pursued alternative explanations using stratified analyses with possible confounders and investigated plausible mechanisms for the reported effect. They were right to have done so, not only for scientific reasons but also because studies such as these are rapidly picked up by the media, who sometimes draw premature or unrealistic conclusions from such data. Additionally, this study has important commercial implications; therefore, 1185
Correspondence
caution in the interpretation of these results is necessary. We would like to put forward some ideas in this letter that might help to understand the results better. Our main concern is related to the ever/never classification of exposure variables. Although some of the crude analyses are presented with IUD use as a categorical variable, the main results are estimated and presented with ever/never use. Oral contraceptive use is clearly related to cervical cancer,2 and therefore its inclusion in the main analysis would have yielded better adjusted results. However, oral contraceptive use cannot be evaluated precisely without simultaneously taking duration of exposure and timing of exposure into account, because induction and latency periods are key issues in cancer epidemiology.3 Therefore, the ever/never stratification of oral contraceptive use is probably not precise enough to adjust for confounding. A lower oral contraceptive use in long-term IUD users than in short-term IUD users or those who have never used an IUD cannot be completely ruled out as an alternative explanation for the lower cervical cancer risk seen in long-term IUD users. When data were stratified by duration of oral contraceptive exposure (0, 1, 2–4, 5–9, and ≥10 years; presented in the webappendix1), most of the associations were not significant, and there was even one positive association for IUD users that used oral contraceptives for 2–4 years, which was not discussed in the main Article. Furthermore, the associations between IUD use and cervical cancer in those who have never used oral contraceptives are not adjusted for other important confounders, such as alternative family planning methods used by those not using IUDs (eg, other hormonal methods that are more prevalent in developing 1186
countries), number of sexual partners, and smoking. The imprecise measurement of a crucial confounder such as oral contraceptive use is the main weakness of this study. A better defined reference category of exposure would have been much more informative. We declare that we have no conflicts of interest.
*Cristina Lopez-del Burgo, Alfonso Osorio, Jokin De Irala [email protected] Preventive Medicine and Public Health (CL-dB, JDI), Department of Education (AO), and Institute for Culture and Society (CL-dB, AO, JDI), University of Navarra, Pamplona, Navarra 31008, Spain 1
2
3
Castellsagué X, Díaz M, Vaccarella S, et al. Intrauterine device use, cervical infection with human papillomavirus, and risk of cervical cancer: a pooled analysis of 26 epidemiological studies. Lancet Oncol 2011; 12: 1023–31. International Collaboration of Epidemiological Studies of Cervical Cancer. Cervical cancer and hormonal contraceptives: collaborative reanalysis of individual data for 16 573 women with cervical cancer and 35 509 women without cervical cancer from 24 epidemiological studies. Lancet 2007; 370: 1609–21. Bosch FX, Lorincz A, Munoz N, Meijer CJ, Shah KV. The causal relation between human papillomavirus and cervical cancer. J Clin Pathol 2002; 55: 244–65.
Authors’ reply We did our best to reasonably rule out that the reported inverse association between IUD use and cervical cancer risk1 was mainly due to confounding by oral contraceptive use, a well established cofactor of cervical cancer. In our initial bivariate analyses done to identify potential confounders, we did not identify any variables related to oral contraceptive use as confounders. Despite this, we adjusted the stratified analyses to take into account duration of oral contraceptive use. These analyses showed that this covariate did not substantially alter the inverse association, except in women who used oral contraceptives for 2–4 years (odds ratio [OR] 1·29; 95% CI 0·55–3·03). Even though the association in three of the
five categories of duration of oral contraceptive use was not significant, the inverse association was the strongest and highly significant in those taking oral contraceptives for 10 years or more (OR 0·23, 95% CI 0·09–0·62). In our view, when doing stratified analyses using ordinal categories, more importance should be placed on the overall pattern than on single point estimates. Thus, we conclude that duration of oral contraceptive use does not substantially alter the inverse association for four reasons: all but one point estimate of the ORs are below 1; their 95% CIs greatly overlap with each other; there is no exposure–response relation between duration of oral contraceptive use and the inverse association between IUD use and cervical cancer; and the interaction between IUD use and duration of oral contraceptive use is far from significant (p=0·9). Concerning the issue of inadequate adjustment, the ORs adjusted for smoking and number of sexual partners (in addition to study centre, years of schooling, age at first sex, number of Pap smears, and cervical human papillomavirus [HPV] DNA status) for each of the five categories of duration of oral contraceptive use (never, 1, 2–4, 5–9, and ≥10 years) were, respectively, 0·48, 0·70, 1·38, 0·78, and 0·27, which are very similar to the original ORs reported in webappendix table 3 (0·50, 0·79, 1·29, 0·63, and 0·23, respectively).1 To the best of our ability with the available data, we have ruled out that oral contraceptive use has a large confounding effect; however, we still conclude that caution should be used in the interpretation of these results because residual confounding by other unmeasured factors can never be completely ruled out. Finally, we would like to stress that adequate screening and most probably HPV vaccination are the best strategies to prevent cervical cancer and cope with any potential influences
www.thelancet.com/oncology Vol 12 December 2011
Patients with PID (n=11 549) Controls (n=97 985) Number of patients with ovarian cancer*
8
Incidence per 10 000 person-years
3·13
28 1·17
Crude HR (95% CI)
2·69 (1·23–5·90)
1·00
Adjusted HR† (95% CI)
2·53 (1·15 –5·57)
1·00
PID=pelvic inflammatory disease of the upper genital tract (International Classification of Diseases, ninth revision, clinical modification 614 and 615). HR=hazard ratio. *Within 3 years of follow-up. †Adjusted for age, cardiovascular disease, diabetes mellitus, chronic liver disease, rheumatic diseases, monthly income, urbanization level, and endometriosis.
Table: Crude and adjusted risks of ovarian cancer in patients with upper genital tract PID aged 15–44 years
than expected. If we exclude patients coded as ICD-9-CM code 616 and restrict age to 15–44 years, 11 549 women would be classified as having PID (table), which equates to a 3·3% prevalence of PID. The adjusted hazard ratio (HR) of ovarian cancer in patients with upper genital tract infections was higher than in the controls (table). Although oral contraceptives reduce the risks of ovarian cancer and PID, a history of oral contraceptive use also seems to have its risks. As Cheong stated, PID usually occurs in young, sexually active women, and we believe that women with a long history of taking oral contraceptives are more sexually active than are those who have never taken them.2 Therefore, extended use of oral contraceptives could increase the risk of ovarian cancer. Oral contraceptives are one of the few hormone drugs that are, in general, available over the counter; however, the use of such drugs might not be as frequent as thought. In 2002, Rossouw and colleagues3 showed that there were risks associated with hormone use and, after publication of this article, the general public was reluctant to take hormone drugs without consulting their doctors first. Because the data in our study were obtained between 2004 and 2005, the chance of wrongly classifying patients as having no history of oral contraceptive use in our study is small. We appreciate R van de Laar and L F Massuger’s feedback. In our study,
only ICD-9-CM code 614·2 would include patients with tubo-ovarian abscess, and of the 608 patients with abscesses, only one developed ovarian cancer. Primary carcinoma of the fallopian tube and primary ovarian carcinoma rarely mimic tuboovarian abscess clinically, but the inclusion of patients with ICD-9-CM code 614·2 could still have affected our final results. After we excluded such patients, analyses resulted in an adjusted HR of 1·85 (95% CI 1·22–2·83; p=0·001), and after we excluded patients coded as 614·2 and 616, analyses resulted in an adjusted HR of 2·55 (95% CI 1·35–4·84; p=0·004). Thus, these adjustments do not affect our conclusion that there is an association between PID and ovarian cancer. We took various factors into consideration while designing this study—eg, we included only patients with two or more incidences of PID, and excluded cases of ovarian cancer that occurred within 6 months of the last episode of PID. As for the period of time for inflammation to develop into cancer, 1–3 years is still plausible.4 We understand van de Laar and Massuger’s concern about the results of the patients’ ovarian cancer treatment; however, the focus of our study was solely on the relation between PID and ovarian cancer and we do not have any data for patients’ response to treatment. We declare that we have no conflicts of interest.
www.thelancet.com/oncology Vol 12 December 2011
Hui-Wen Lin, Shiyng Yu Lin, Wei Zer Lin, Wei Li Lin, *Yuen-Liang Lai [email protected] Department of Mathematics, Soochow University, Taipei, Taiwan (H-WL); Biostatistics and Research Consultation Centre (H-WL), School of Medicine (SYL, WLL), Graduate Institute of Humanities in Medicine (Y-LL), Department of Radiation Oncology, Shuang Ho Hospital (Y-LL), Taipei Medical University, Zhonghe City, Taipei County 235, Taiwan; Department of Family Medicine, Taipei Medical University Hospital, Taipei, Taiwan (SYL); School of Medicine, Chung Shan Medical University, Taichung, Taiwan (WZL); and Mackay Medical College, Taipei, Taiwan (Y-LL) 1
2
3
4
US Centers for Disease Control and Prevention. Classification of diseases, functioning, and disability. http://www.cdc.gov/nchs/icd.htm (accessed Sept 15, 2011). Lin HW, Tu YY, Lin SY, et al. Risk of ovarian cancer in women with pelvic inflammatory diseases: a population-based study. Lancet Oncol 2011; 12: 900–04. Rossouw JE, Anderson GL, Prentice RL, et al. Risks and benefits of estrogen plus progestin in healthy postmenopausal women: principal results from the Women’s Health Initiative randomized controlled trial. JAMA 2002; 288: 321–33. Wei JJ, William J, Bulun S. Endometriosis and ovarian cancer: a review of clinical, pathologic, and molecular aspects. Int J Gynecol Pathol 2011; 30: 553–68.
Intrauterine device and cervical cancer: we need more evidence In their study on intrauterine device (IUD) use and risk of cervical cancer published in The Lancet Oncology, Xavier Castellsagué and colleagues1 claimed that their work provides evidence that IUD use might reduce the risk of developing cervical cancer. The authors pursued alternative explanations using stratified analyses with possible confounders and investigated plausible mechanisms for the reported effect. They were right to have done so, not only for scientific reasons but also because studies such as these are rapidly picked up by the media, who sometimes draw premature or unrealistic conclusions from such data. Additionally, this study has important commercial implications; therefore, 1185
Correspondence
caution in the interpretation of these results is necessary. We would like to put forward some ideas in this letter that might help to understand the results better. Our main concern is related to the ever/never classification of exposure variables. Although some of the crude analyses are presented with IUD use as a categorical variable, the main results are estimated and presented with ever/never use. Oral contraceptive use is clearly related to cervical cancer,2 and therefore its inclusion in the main analysis would have yielded better adjusted results. However, oral contraceptive use cannot be evaluated precisely without simultaneously taking duration of exposure and timing of exposure into account, because induction and latency periods are key issues in cancer epidemiology.3 Therefore, the ever/never stratification of oral contraceptive use is probably not precise enough to adjust for confounding. A lower oral contraceptive use in long-term IUD users than in short-term IUD users or those who have never used an IUD cannot be completely ruled out as an alternative explanation for the lower cervical cancer risk seen in long-term IUD users. When data were stratified by duration of oral contraceptive exposure (0, 1, 2–4, 5–9, and ≥10 years; presented in the webappendix1), most of the associations were not significant, and there was even one positive association for IUD users that used oral contraceptives for 2–4 years, which was not discussed in the main Article. Furthermore, the associations between IUD use and cervical cancer in those who have never used oral contraceptives are not adjusted for other important confounders, such as alternative family planning methods used by those not using IUDs (eg, other hormonal methods that are more prevalent in developing 1186
countries), number of sexual partners, and smoking. The imprecise measurement of a crucial confounder such as oral contraceptive use is the main weakness of this study. A better defined reference category of exposure would have been much more informative. We declare that we have no conflicts of interest.
*Cristina Lopez-del Burgo, Alfonso Osorio, Jokin De Irala [email protected] Preventive Medicine and Public Health (CL-dB, JDI), Department of Education (AO), and Institute for Culture and Society (CL-dB, AO, JDI), University of Navarra, Pamplona, Navarra 31008, Spain 1
2
3
Castellsagué X, Díaz M, Vaccarella S, et al. Intrauterine device use, cervical infection with human papillomavirus, and risk of cervical cancer: a pooled analysis of 26 epidemiological studies. Lancet Oncol 2011; 12: 1023–31. International Collaboration of Epidemiological Studies of Cervical Cancer. Cervical cancer and hormonal contraceptives: collaborative reanalysis of individual data for 16 573 women with cervical cancer and 35 509 women without cervical cancer from 24 epidemiological studies. Lancet 2007; 370: 1609–21. Bosch FX, Lorincz A, Munoz N, Meijer CJ, Shah KV. The causal relation between human papillomavirus and cervical cancer. J Clin Pathol 2002; 55: 244–65.
Authors’ reply We did our best to reasonably rule out that the reported inverse association between IUD use and cervical cancer risk1 was mainly due to confounding by oral contraceptive use, a well established cofactor of cervical cancer. In our initial bivariate analyses done to identify potential confounders, we did not identify any variables related to oral contraceptive use as confounders. Despite this, we adjusted the stratified analyses to take into account duration of oral contraceptive use. These analyses showed that this covariate did not substantially alter the inverse association, except in women who used oral contraceptives for 2–4 years (odds ratio [OR] 1·29; 95% CI 0·55–3·03). Even though the association in three of the
five categories of duration of oral contraceptive use was not significant, the inverse association was the strongest and highly significant in those taking oral contraceptives for 10 years or more (OR 0·23, 95% CI 0·09–0·62). In our view, when doing stratified analyses using ordinal categories, more importance should be placed on the overall pattern than on single point estimates. Thus, we conclude that duration of oral contraceptive use does not substantially alter the inverse association for four reasons: all but one point estimate of the ORs are below 1; their 95% CIs greatly overlap with each other; there is no exposure–response relation between duration of oral contraceptive use and the inverse association between IUD use and cervical cancer; and the interaction between IUD use and duration of oral contraceptive use is far from significant (p=0·9). Concerning the issue of inadequate adjustment, the ORs adjusted for smoking and number of sexual partners (in addition to study centre, years of schooling, age at first sex, number of Pap smears, and cervical human papillomavirus [HPV] DNA status) for each of the five categories of duration of oral contraceptive use (never, 1, 2–4, 5–9, and ≥10 years) were, respectively, 0·48, 0·70, 1·38, 0·78, and 0·27, which are very similar to the original ORs reported in webappendix table 3 (0·50, 0·79, 1·29, 0·63, and 0·23, respectively).1 To the best of our ability with the available data, we have ruled out that oral contraceptive use has a large confounding effect; however, we still conclude that caution should be used in the interpretation of these results because residual confounding by other unmeasured factors can never be completely ruled out. Finally, we would like to stress that adequate screening and most probably HPV vaccination are the best strategies to prevent cervical cancer and cope with any potential influences
www.thelancet.com/oncology Vol 12 December 2011