- Email: [email protected]
HPV and circumcision: A biased, inaccurate and misleading meta-analysis
Letters to the Editor
91
6. Nolan RC, Chidlow G, French MA. Parvovirus B19 encephalitis presenting as immune restoration disease after highly active antiretroviral therapy for human immunodeficiency virus infection. Clin Infect Dis 2003;36:1191e4.
Prasad R. Koduri* Department of Hematology-Oncology, Apollo Hospitals, Jubilee Hills, Hyderabad, India 500 034 *6615 N. Lawndale, Lincolnwood, IL 60712, USA. Tel./fax: þ1 847 982 0683. E-mail address: [email protected] Accepted 17 September 2006 Available online 1 November 2006 0163-4453/$30 ª 2006 The British Infection Society. Published by Elsevier Ltd. All rights reserved. doi:10.1016/j.jinf.2006.09.014
HPV and circumcision: A biased, inaccurate and misleading meta-analysis To perform a meta-analysis exploring the association between male circumcision and HPV is a methodological challenge. Studies show a great variability in methodological issues such as circumcision status ascertainment, genital sites sampled, HPV-related outcome definition (HPV infection versus clinical lesion), procedures used to ascertain the outcome (clinical exam, penoscopy, acetoacid-staining, etc.), and, for HPV detection the collection device and the assay used to detect HPV DNA. At the statistical level both a meta-analysis using the adjusted odds ratios (OR) as published in each study and a stratified meta-analysis showing results by groups of studies that are homogenous in one of the above-listed methodological aspects is required in order to strictly adjust for potential confounders. Van Howe’s recently published meta-analysis on circumcision and prevalent genital HPV1 contains a number of biased and inaccurate statistical approaches that make the paper a misleading and scientifically invalid piece of work. In this letter, we not only highlight the key faults incurred by Van Howe, but most importantly, we contribute a totally reproducible re-analysis of Van Howe’s own selection of papers and data. We show very consistently that male circumcision is associated with a statistically significant reduced risk of penile HPV and related lesions (OR Z 0.56, 95%CI: 0.39e0.82) (Table 1). The first limitation of Van Howe’s meta-analysis is the selection of the three papers that meet his ‘‘strict’’ criteria.2e4 In the two papers by Aynaud the outcome used in all subjects is not HPV DNA detection per se, but instead diagnosis of HPV-related lesions as diagnosed by clinical examination and penoscopy. Based on Van Howe’s own DOI of original article: 10.1016/j.jinf.2006.08.005.
definition of ‘‘strict’’ criteria these two papers should also be excluded. In fact, he has excluded four other papers because the diagnosis of genital warts was made by clinical exam, when, it is well established that genital warts are virtually always caused by HPV and that the diagnosis is highly accurate. Interestingly enough, none of the three ‘‘strict’’ papers report adjusted ORs, a criterion that should have been used to classify the papers as ‘‘strict’’ in the first place. Thus, his conclusion of a lack of association based on only these three studies of which two do not use HPV detection is misleading, reflecting only a distorted part of the available data. The second serious mistake is his approach of manipulating the original published data by applying HPV detection rates from one study to other studies to compensate for supposed HPV under-detection caused by not including samples from the penile shaft. Whether or not the penile shaft and scrotum should or should not be sampled is still an open and debated scientific issue. In our New England Journal of Medicine paper on male circumcision, penile HPV and cervical cancer risk,5 the fact that, even without sampling the penile shaft a strong protective effect for cervical cancer in the partner was found, which suggests that HPV as detected in the shaft and scrotum is probably clinically irrelevant for the female partner. It is obvious that circumcision cannot directly influence the development of HPV infections and lesions in the penile shaft and scrotum, as these areas are not involved in the surgical procedure. Therefore, since HPV prevalence for these two sites should be comparable between circumcised and uncircumcised men, what our data are implying is that the protective effect must be due to differences in HPV DNA detection rates in the other sites, including glans, coronal sulcus and, for uncircumcised men, the prepuce. Even two of the three ‘‘strict’’ papers that made Van Howe conclude a lack of association admit this: the report by Weaver et al. does mention a statistically significant twofold difference in HPV positivity between circumcised and uncircumcised men when only samples from the glans and foreskin are considered4; and the 1999 paper by Aynaud3 also reports a statistically significant protective effect when only preputial lesions are considered. Furthermore, if we review how HPV studies in women are usually performed, including randomized HPV vaccination clinical trials, we all would agree that most of them use cervical HPV, and not a combination of cervical, vaginal, and vulval HPV, as the outcome variable. The third very serious and inexplicable error of Van Howe’s meta-analysis is the failure to use the adjusted OR estimates published in five of the eight selected papers. In his analysis Van Howe totally ignores the strong effects of many key covariates that may act as strong confounders within each study (see the table for a list of covariates adjusted for in each study). We have repeated the metaanalysis using the adjusted ORs published in five papers and the crude ORs of the remaining three. As shown in the table, not only does the combined adjusted OR show a strong and statistically significant protective effect between male circumcision and penile HPV (OR Z 0.56), but even the combined crude OR e Van Howe’s preferred approach e shows the same statistically significant association (OR Z 0.57). Inexplicably, Van Howe fails to report
92 Table 1
Letters to the Editor Meta-analysis of studies exploring the association between being circumcised and penile HPV and related lesions
Study and type of analysis
Non-circumcised
Circumcised
No. HPVþve/total
No. HPVþ ve/total
Crude OR Adjusted OR (95% CI)
Comments
Aynaud, 19942 Aynaud, 19993 Baldwin, 20048
383/737 93/162 46/158
119/263 20/48 46/232
0.76 0.53 0.60b
0.76 (0.58e1.01)a 0.53 (0.28e1.02)a 0.34 (0.20e0.57)b
Castellsague ´, 20025
166/847
16/292
0.24b
0.37 (0.16e0.85)b
Lajous, 20059
365/830
28/95
0.53b
0.48 (0.30e0.77)b
Shin, 200410
3/43
29/325
1.31
1.80 (0.40e8.20)
Svare, 200211
85/174
4/24
0.21b
0.20 (0.06e0.60)b
Weaver, 20044 All studies (n Z 8)
17/62 1158/3013
92/285 354/1564
1.26 0.57b
1.26 (0.69e2.32)a 0.56 (0.39e0.82)b
No adjustment was reported No adjustment was reported Adjusted for sexual frequency per month, genital warts, condom use in past 3 months, and steady partner Adjusted for age, study location, level of education, age at first sexual intercourse, lifetime no. of sexual partners, and frequency of genital washing after sex Adjusted for age, SES (socioeconomic status), and lifetime no. of partners Adjusted for age, and lifetime no. of sexual partners Adjusted for age, lifetime no. of sexual partners, no. of sexual partners in the past year, and genital warts No adjustment was reported Random effects model
Combined stratified analyses - ‘‘Strict’’ studies as classified by Van Howe (n Z 3) - Non-‘‘strict’’ studies as classified by Van Howe (n Z 5)
0.79 (0.62e1.00)c 0.41 (0.30e0.56)b,c
- Studies that included sampling of the penile shaft (n Z 6) - Studies that did not include sampling of the penile shaft (n Z 2)
0.66 (0.44e1.00) 0.35 (0.22e0.54)b,c
- Studies that used clinical exam to ascertain circumcision status (n Z 5) - Studies using self-reported circumcision status (n Z 3)
0.60 (0.38e0.93)b
- Studies reporting adjusted ORs (n Z 5) - Studies not reporting adjusted ORs (n Z 3) - Studies using HPV DNA detection methods (n Z 6) - Studies not using HPV DNA detection methods (n Z 2)
0.47 (0.31e0.72)b,c
Aynaud, 1994, 1999, Weaver, 2004
Baldwin, 2004, Castellsague, 2002
Lajous, 2005, Shin, 2004, Svare, 2002
b,c
0.41 (0.30e0.56) 0.79 (0.62e1.00)c
Aynaud, 1994, 1999, Weaver, 2004
b
0.52 (0.30e0.90) 0.72 (0.56e0.93)b,c
Aynaud, 1994, 1999
All OR estimates are for the association between being circumcised and presence of penile HPV DNA or penile HPV-related lesions. a Since adjusted ORs were not reported, crude ORs were used for the estimation of the combined OR. b Denotes statistical significance. c OR estimate derived from fixed effects models, as there was no heterogeneity between studies. Otherwise they are derived from random effects models.
also the pooled estimate of the crude OR. Instead he reports a meta-regression OR (adjusted by circumcision ascertainment and failure to sample the penile shaft) that we were unable to reproduce. To even use a stricter approach to adjust for confounding we further performed stratified meta-analyses by selected key methodological variables. As shown in the table, our results consistently indicate that regardless of the genital site sampled, circumcision status ascertainment, and the use or not of HPV DNA detection assays, a consistent protective effect is found in all strata. In conclusion, our re-analysis of Van Howe’s data using the published adjusted ORs and without fixing the data from one study to another, clearly shows, no matter how
the studies are grouped, a moderate to strong protective effect of circumcision on penile HPV and related lesions. This is consistent with the current scientific evidence from observational studies and three recent randomized clinical trials indicating that circumcision is associated with a reduced risk of HIV acquisition.6,7 For the editors and reviewers of the Journal of Infection, to receive a manuscript that concludes against the accumulated scientific evidence should have alerted them about the reported inaccuracies and systematic errors as those incurred by Van Howe. We believe that the paper by Van Howe should be retracted from the literature as it is based on inappropriate statistical methods suspiciously used to falsely show a lack of association between male circumcision and penile
Letters to the Editor
93
HPV, when we clearly prove a protective association using the same data.
References 1. Van Howe RS. Human papillomavirus and circumcision: a metaanalysis. J Infect 2007;54:490e6. 2. Aynaud O, Ionesco M, Barrasso R. Penile intraepithelial neoplasia. Specific clinical features correlated with histologic and virologic findings. Cancer 1994;74:1762e7. 3. Aynaud O, Piron D, Bijaoui G, Casanova JM. Developmental factors of urethral human papillomavirus lesions: correlation with circumcision. BJU Int 1999;84:57e60. 4. Weaver BA, Feng Q, Holmes KK, Kiviat N, Lee SK, Meyer C, et al. Evaluation of genital sites and sampling techniques for detection of human papillomavirus DNA in men. J Infect Dis 2004;189:677e85. 5. Castellsague X, Bosch FX, Munoz N, Meijer CJ, Shah KV, de Sanjose S, et al. Male circumcision, penile human papillomavirus infection, and cervical cancer in female partners. N Engl J Med 2002;346:1105e12. 6. Weiss HA, Quigley MA, Hayes RJ. Male circumcision and risk of HIV infection in sub-Saharan Africa: a systematic review and meta-analysis. AIDS 2000;14:2361e70. 7. Auvert B, Taljaard D, Lagarde E, Sobngwi-Tambekou J, Sitta R, Puren A. Randomized, controlled intervention trial of male circumcision for reduction of HIV infection risk: the ANRS 1265 Trial. PLoS Med 2005;2:e298. 8. Baldwin SB, Wallace DR, Papenfuss MR, Abrahamsen M, Vaught LC, Giuliano AR. Condom use and other factors affecting penile human papillomavirus detection in men attending a sexually transmitted disease clinic. Sex Transm Dis 2004; 31:601e7. 9. Lajous M, Mueller N, Cruz-Valdez A, Aguilar LV, Franceschi S, Hernandez-Avila M, et al. Determinants of prevalence, acquisition, and persistence of human papillomavirus in healthy Mexican military men. Cancer Epidemiol Biomarkers Prev 2005;14:1710e6. 10. Shin HR, Franceschi S, Vaccarella S, Roh JW, Ju YH, Oh JK, et al. Prevalence and determinants of genital infection with papillomavirus, in female and male university students in Busan, South Korea. J Infect Dis 2004;190:468e76. 11. Svare EI, Kjaer SK, Worm AM, Osterlind A, Meijer CJ, van den Brule AJ. Risk factors for genital HPV DNA in men resemble those found in women: a study of male attendees at a Danish STD clinic. Sex Transm Infect 2002;78:215e8.
Xavier Castellsague ´* Ginesa Albero Ramo ´n Cle `ries F. Xavier Bosch Servei d’Epidemiologia i Registre del Ca`ncer, Institut Catala` d’Oncologia, Gran via s/n, km 2.7, 08907 L’Hospitalet de Llobregat, Spain *Corresponding author. Tel.: þ34 932607812. E-mail address: [email protected] (X. Castellsague ´) Accepted 19 February 2007 Available online 11 April 2007 0163-4453/$30 ª 2007 The British Infection Society. Published by Elsevier Ltd. All rights reserved. doi:10.1016/j.jinf.2007.02.009
Reply to ‘‘HPV and circumcision: A biased, inaccurate and misleading meta-analysis’’ As Dr. Castellsague ´ has become a strong advocate for universal infant circumcision1e3 and was the lead author of a study that explored the relationship between genital HPV infection and circumcision status,4 I expected him to voice a strong negative response to my meta-analysis. Consequently, the meta-analysis I performed has already taken into account the objections I expected Dr. Castellsague ´ and his colleagues to raise. The content of their letter indicates that I had anticipated correctly. I agree that there are inherent difficulties in performing a meta-analysis on quasi-experimental studies, especially when the method of diagnosis, the thoroughness of sampling, and the method of assignment of circumcision status varied among the studies. Because of the various pitfalls in performing meta-analyses on quasi-experiments, standards have been developed and were followed.5 The first subject Castellsague ´ and associates bring up is the method of diagnostics. Methods used included patient report, identification on physical examination, identification using acetic acid and penoscopy, and identification using tissue sampling. These are listed in the order of increasing reliability. The meta-analysis included studies in which tissue sampling was performed. Castellsague ´ and his colleagues misrepresent the diagnostic methods used by Aynaud and associates6,7 by merely characterizing their diagnostic methods as ‘‘clinical examination and penoscopy.’’ As noted in Table 1 in the meta-analysis, Aynaud and associates relied on penoscopy, viral cultures, and biopsy to make their diagnosis. Because tissue samples were obtained, they were rightfully included in the metaanalysis. If these two studies were eliminated, then only one study would have met the inclusion criteria. Studies in which genital warts were identified were excluded for two reasons. Primarily, most of these studies relied on patient report or identification of warts on physical examination. Without further study, such as use of acetic acid or tissue sampling, most cases of HPV would be missed. Secondly, the types of HPV that cause visible genital warts are less carcinogenic than those identified with the aid of acetic acid or tissue sampling. One large study that identified genital warts in clients of a sexually transmitted disease clinic found noncircumcised men to be at a much lower risk for genital warts (OR Z 0.32, 95%CI Z 0.22e0.45).8 Inclusion of this study and others that looked at genital warts generates a summary effect similar to meta-analysis of studies meeting the selection criteria with a trend toward noncircumcised men being at greater risk but with no statistically significant difference (OR Z 1.21, 95%CI Z 0.91e1.61). The inclusion of reporting adjusted odds as a selection criterion was considered, but was rejected for three reasons. (1) Only a few studies reported adjusted odds ratios. If only these studies were included, all the studies with complete sampling would have been excluded. It is easy to demonstrate that sampling bias has more impact on the validity of the meta-analysis than using adjusted odds DOI of original article: 10.1016/j.jinf.2007.02.009.
91
6. Nolan RC, Chidlow G, French MA. Parvovirus B19 encephalitis presenting as immune restoration disease after highly active antiretroviral therapy for human immunodeficiency virus infection. Clin Infect Dis 2003;36:1191e4.
Prasad R. Koduri* Department of Hematology-Oncology, Apollo Hospitals, Jubilee Hills, Hyderabad, India 500 034 *6615 N. Lawndale, Lincolnwood, IL 60712, USA. Tel./fax: þ1 847 982 0683. E-mail address: [email protected] Accepted 17 September 2006 Available online 1 November 2006 0163-4453/$30 ª 2006 The British Infection Society. Published by Elsevier Ltd. All rights reserved. doi:10.1016/j.jinf.2006.09.014
HPV and circumcision: A biased, inaccurate and misleading meta-analysis To perform a meta-analysis exploring the association between male circumcision and HPV is a methodological challenge. Studies show a great variability in methodological issues such as circumcision status ascertainment, genital sites sampled, HPV-related outcome definition (HPV infection versus clinical lesion), procedures used to ascertain the outcome (clinical exam, penoscopy, acetoacid-staining, etc.), and, for HPV detection the collection device and the assay used to detect HPV DNA. At the statistical level both a meta-analysis using the adjusted odds ratios (OR) as published in each study and a stratified meta-analysis showing results by groups of studies that are homogenous in one of the above-listed methodological aspects is required in order to strictly adjust for potential confounders. Van Howe’s recently published meta-analysis on circumcision and prevalent genital HPV1 contains a number of biased and inaccurate statistical approaches that make the paper a misleading and scientifically invalid piece of work. In this letter, we not only highlight the key faults incurred by Van Howe, but most importantly, we contribute a totally reproducible re-analysis of Van Howe’s own selection of papers and data. We show very consistently that male circumcision is associated with a statistically significant reduced risk of penile HPV and related lesions (OR Z 0.56, 95%CI: 0.39e0.82) (Table 1). The first limitation of Van Howe’s meta-analysis is the selection of the three papers that meet his ‘‘strict’’ criteria.2e4 In the two papers by Aynaud the outcome used in all subjects is not HPV DNA detection per se, but instead diagnosis of HPV-related lesions as diagnosed by clinical examination and penoscopy. Based on Van Howe’s own DOI of original article: 10.1016/j.jinf.2006.08.005.
definition of ‘‘strict’’ criteria these two papers should also be excluded. In fact, he has excluded four other papers because the diagnosis of genital warts was made by clinical exam, when, it is well established that genital warts are virtually always caused by HPV and that the diagnosis is highly accurate. Interestingly enough, none of the three ‘‘strict’’ papers report adjusted ORs, a criterion that should have been used to classify the papers as ‘‘strict’’ in the first place. Thus, his conclusion of a lack of association based on only these three studies of which two do not use HPV detection is misleading, reflecting only a distorted part of the available data. The second serious mistake is his approach of manipulating the original published data by applying HPV detection rates from one study to other studies to compensate for supposed HPV under-detection caused by not including samples from the penile shaft. Whether or not the penile shaft and scrotum should or should not be sampled is still an open and debated scientific issue. In our New England Journal of Medicine paper on male circumcision, penile HPV and cervical cancer risk,5 the fact that, even without sampling the penile shaft a strong protective effect for cervical cancer in the partner was found, which suggests that HPV as detected in the shaft and scrotum is probably clinically irrelevant for the female partner. It is obvious that circumcision cannot directly influence the development of HPV infections and lesions in the penile shaft and scrotum, as these areas are not involved in the surgical procedure. Therefore, since HPV prevalence for these two sites should be comparable between circumcised and uncircumcised men, what our data are implying is that the protective effect must be due to differences in HPV DNA detection rates in the other sites, including glans, coronal sulcus and, for uncircumcised men, the prepuce. Even two of the three ‘‘strict’’ papers that made Van Howe conclude a lack of association admit this: the report by Weaver et al. does mention a statistically significant twofold difference in HPV positivity between circumcised and uncircumcised men when only samples from the glans and foreskin are considered4; and the 1999 paper by Aynaud3 also reports a statistically significant protective effect when only preputial lesions are considered. Furthermore, if we review how HPV studies in women are usually performed, including randomized HPV vaccination clinical trials, we all would agree that most of them use cervical HPV, and not a combination of cervical, vaginal, and vulval HPV, as the outcome variable. The third very serious and inexplicable error of Van Howe’s meta-analysis is the failure to use the adjusted OR estimates published in five of the eight selected papers. In his analysis Van Howe totally ignores the strong effects of many key covariates that may act as strong confounders within each study (see the table for a list of covariates adjusted for in each study). We have repeated the metaanalysis using the adjusted ORs published in five papers and the crude ORs of the remaining three. As shown in the table, not only does the combined adjusted OR show a strong and statistically significant protective effect between male circumcision and penile HPV (OR Z 0.56), but even the combined crude OR e Van Howe’s preferred approach e shows the same statistically significant association (OR Z 0.57). Inexplicably, Van Howe fails to report
92 Table 1
Letters to the Editor Meta-analysis of studies exploring the association between being circumcised and penile HPV and related lesions
Study and type of analysis
Non-circumcised
Circumcised
No. HPVþve/total
No. HPVþ ve/total
Crude OR Adjusted OR (95% CI)
Comments
Aynaud, 19942 Aynaud, 19993 Baldwin, 20048
383/737 93/162 46/158
119/263 20/48 46/232
0.76 0.53 0.60b
0.76 (0.58e1.01)a 0.53 (0.28e1.02)a 0.34 (0.20e0.57)b
Castellsague ´, 20025
166/847
16/292
0.24b
0.37 (0.16e0.85)b
Lajous, 20059
365/830
28/95
0.53b
0.48 (0.30e0.77)b
Shin, 200410
3/43
29/325
1.31
1.80 (0.40e8.20)
Svare, 200211
85/174
4/24
0.21b
0.20 (0.06e0.60)b
Weaver, 20044 All studies (n Z 8)
17/62 1158/3013
92/285 354/1564
1.26 0.57b
1.26 (0.69e2.32)a 0.56 (0.39e0.82)b
No adjustment was reported No adjustment was reported Adjusted for sexual frequency per month, genital warts, condom use in past 3 months, and steady partner Adjusted for age, study location, level of education, age at first sexual intercourse, lifetime no. of sexual partners, and frequency of genital washing after sex Adjusted for age, SES (socioeconomic status), and lifetime no. of partners Adjusted for age, and lifetime no. of sexual partners Adjusted for age, lifetime no. of sexual partners, no. of sexual partners in the past year, and genital warts No adjustment was reported Random effects model
Combined stratified analyses - ‘‘Strict’’ studies as classified by Van Howe (n Z 3) - Non-‘‘strict’’ studies as classified by Van Howe (n Z 5)
0.79 (0.62e1.00)c 0.41 (0.30e0.56)b,c
- Studies that included sampling of the penile shaft (n Z 6) - Studies that did not include sampling of the penile shaft (n Z 2)
0.66 (0.44e1.00) 0.35 (0.22e0.54)b,c
- Studies that used clinical exam to ascertain circumcision status (n Z 5) - Studies using self-reported circumcision status (n Z 3)
0.60 (0.38e0.93)b
- Studies reporting adjusted ORs (n Z 5) - Studies not reporting adjusted ORs (n Z 3) - Studies using HPV DNA detection methods (n Z 6) - Studies not using HPV DNA detection methods (n Z 2)
0.47 (0.31e0.72)b,c
Aynaud, 1994, 1999, Weaver, 2004
Baldwin, 2004, Castellsague, 2002
Lajous, 2005, Shin, 2004, Svare, 2002
b,c
0.41 (0.30e0.56) 0.79 (0.62e1.00)c
Aynaud, 1994, 1999, Weaver, 2004
b
0.52 (0.30e0.90) 0.72 (0.56e0.93)b,c
Aynaud, 1994, 1999
All OR estimates are for the association between being circumcised and presence of penile HPV DNA or penile HPV-related lesions. a Since adjusted ORs were not reported, crude ORs were used for the estimation of the combined OR. b Denotes statistical significance. c OR estimate derived from fixed effects models, as there was no heterogeneity between studies. Otherwise they are derived from random effects models.
also the pooled estimate of the crude OR. Instead he reports a meta-regression OR (adjusted by circumcision ascertainment and failure to sample the penile shaft) that we were unable to reproduce. To even use a stricter approach to adjust for confounding we further performed stratified meta-analyses by selected key methodological variables. As shown in the table, our results consistently indicate that regardless of the genital site sampled, circumcision status ascertainment, and the use or not of HPV DNA detection assays, a consistent protective effect is found in all strata. In conclusion, our re-analysis of Van Howe’s data using the published adjusted ORs and without fixing the data from one study to another, clearly shows, no matter how
the studies are grouped, a moderate to strong protective effect of circumcision on penile HPV and related lesions. This is consistent with the current scientific evidence from observational studies and three recent randomized clinical trials indicating that circumcision is associated with a reduced risk of HIV acquisition.6,7 For the editors and reviewers of the Journal of Infection, to receive a manuscript that concludes against the accumulated scientific evidence should have alerted them about the reported inaccuracies and systematic errors as those incurred by Van Howe. We believe that the paper by Van Howe should be retracted from the literature as it is based on inappropriate statistical methods suspiciously used to falsely show a lack of association between male circumcision and penile
Letters to the Editor
93
HPV, when we clearly prove a protective association using the same data.
References 1. Van Howe RS. Human papillomavirus and circumcision: a metaanalysis. J Infect 2007;54:490e6. 2. Aynaud O, Ionesco M, Barrasso R. Penile intraepithelial neoplasia. Specific clinical features correlated with histologic and virologic findings. Cancer 1994;74:1762e7. 3. Aynaud O, Piron D, Bijaoui G, Casanova JM. Developmental factors of urethral human papillomavirus lesions: correlation with circumcision. BJU Int 1999;84:57e60. 4. Weaver BA, Feng Q, Holmes KK, Kiviat N, Lee SK, Meyer C, et al. Evaluation of genital sites and sampling techniques for detection of human papillomavirus DNA in men. J Infect Dis 2004;189:677e85. 5. Castellsague X, Bosch FX, Munoz N, Meijer CJ, Shah KV, de Sanjose S, et al. Male circumcision, penile human papillomavirus infection, and cervical cancer in female partners. N Engl J Med 2002;346:1105e12. 6. Weiss HA, Quigley MA, Hayes RJ. Male circumcision and risk of HIV infection in sub-Saharan Africa: a systematic review and meta-analysis. AIDS 2000;14:2361e70. 7. Auvert B, Taljaard D, Lagarde E, Sobngwi-Tambekou J, Sitta R, Puren A. Randomized, controlled intervention trial of male circumcision for reduction of HIV infection risk: the ANRS 1265 Trial. PLoS Med 2005;2:e298. 8. Baldwin SB, Wallace DR, Papenfuss MR, Abrahamsen M, Vaught LC, Giuliano AR. Condom use and other factors affecting penile human papillomavirus detection in men attending a sexually transmitted disease clinic. Sex Transm Dis 2004; 31:601e7. 9. Lajous M, Mueller N, Cruz-Valdez A, Aguilar LV, Franceschi S, Hernandez-Avila M, et al. Determinants of prevalence, acquisition, and persistence of human papillomavirus in healthy Mexican military men. Cancer Epidemiol Biomarkers Prev 2005;14:1710e6. 10. Shin HR, Franceschi S, Vaccarella S, Roh JW, Ju YH, Oh JK, et al. Prevalence and determinants of genital infection with papillomavirus, in female and male university students in Busan, South Korea. J Infect Dis 2004;190:468e76. 11. Svare EI, Kjaer SK, Worm AM, Osterlind A, Meijer CJ, van den Brule AJ. Risk factors for genital HPV DNA in men resemble those found in women: a study of male attendees at a Danish STD clinic. Sex Transm Infect 2002;78:215e8.
Xavier Castellsague ´* Ginesa Albero Ramo ´n Cle `ries F. Xavier Bosch Servei d’Epidemiologia i Registre del Ca`ncer, Institut Catala` d’Oncologia, Gran via s/n, km 2.7, 08907 L’Hospitalet de Llobregat, Spain *Corresponding author. Tel.: þ34 932607812. E-mail address: [email protected] (X. Castellsague ´) Accepted 19 February 2007 Available online 11 April 2007 0163-4453/$30 ª 2007 The British Infection Society. Published by Elsevier Ltd. All rights reserved. doi:10.1016/j.jinf.2007.02.009
Reply to ‘‘HPV and circumcision: A biased, inaccurate and misleading meta-analysis’’ As Dr. Castellsague ´ has become a strong advocate for universal infant circumcision1e3 and was the lead author of a study that explored the relationship between genital HPV infection and circumcision status,4 I expected him to voice a strong negative response to my meta-analysis. Consequently, the meta-analysis I performed has already taken into account the objections I expected Dr. Castellsague ´ and his colleagues to raise. The content of their letter indicates that I had anticipated correctly. I agree that there are inherent difficulties in performing a meta-analysis on quasi-experimental studies, especially when the method of diagnosis, the thoroughness of sampling, and the method of assignment of circumcision status varied among the studies. Because of the various pitfalls in performing meta-analyses on quasi-experiments, standards have been developed and were followed.5 The first subject Castellsague ´ and associates bring up is the method of diagnostics. Methods used included patient report, identification on physical examination, identification using acetic acid and penoscopy, and identification using tissue sampling. These are listed in the order of increasing reliability. The meta-analysis included studies in which tissue sampling was performed. Castellsague ´ and his colleagues misrepresent the diagnostic methods used by Aynaud and associates6,7 by merely characterizing their diagnostic methods as ‘‘clinical examination and penoscopy.’’ As noted in Table 1 in the meta-analysis, Aynaud and associates relied on penoscopy, viral cultures, and biopsy to make their diagnosis. Because tissue samples were obtained, they were rightfully included in the metaanalysis. If these two studies were eliminated, then only one study would have met the inclusion criteria. Studies in which genital warts were identified were excluded for two reasons. Primarily, most of these studies relied on patient report or identification of warts on physical examination. Without further study, such as use of acetic acid or tissue sampling, most cases of HPV would be missed. Secondly, the types of HPV that cause visible genital warts are less carcinogenic than those identified with the aid of acetic acid or tissue sampling. One large study that identified genital warts in clients of a sexually transmitted disease clinic found noncircumcised men to be at a much lower risk for genital warts (OR Z 0.32, 95%CI Z 0.22e0.45).8 Inclusion of this study and others that looked at genital warts generates a summary effect similar to meta-analysis of studies meeting the selection criteria with a trend toward noncircumcised men being at greater risk but with no statistically significant difference (OR Z 1.21, 95%CI Z 0.91e1.61). The inclusion of reporting adjusted odds as a selection criterion was considered, but was rejected for three reasons. (1) Only a few studies reported adjusted odds ratios. If only these studies were included, all the studies with complete sampling would have been excluded. It is easy to demonstrate that sampling bias has more impact on the validity of the meta-analysis than using adjusted odds DOI of original article: 10.1016/j.jinf.2007.02.009.