We thank Drs Cherrie and MacCalman for their thoughtful comments. However the first step in any analysis is exploring the variables. When examining the distribution of women use of vaginal barriers world- wide, one notices that no country reports more than 5% of woman using this mode of contraception, and only 9 countries have over 1% use of vaginal barriers. Clearly this variable cannot be entered at face value. When we apply a very conservative restriction, and include only countries with over 0.2% of woman use vaginal barriers (n=45) the univariable correlation with PCa incidence (r=0.19, p=0.23) and mortality (r=0.2, p=0.23) are both insignificant.If we apply the same restriction to oral contraceptive use the correlation remains the same. We are now left with 3 modes of contraception: IUD, condoms and oral contraceptives. On univariable analysis the correlation of IUD with PCa incidence (r=-0.011, p=0.893) and mortality (r=-0.19, p=0.02) is inconsistent. It is however correlated with mortality and as we stressed earlier this is supportive of our hypothesis. Now let's consider the differences between oral contraceptives and condoms correlation with PCa incidence and mortality. On univariable analysis oral contraceptives demonstrate a higher Pearson correlation for both PCa incidence ( r= 0.5, p<0.001 vs. r=0.4, p<0.001) and mortality (r= 0.17, p=0.036 vs r=0.13 p=0.1). On multivariable analysis they are both similar for both outcomes. In summary, oral contraceptives are the only mode of contraception that on univariable analysis correlates with both PCa incidence and mortality, and remain significantly correlated on multivariable analysis with PCa incidence. We therefore still believe our hypothesis is valid. We stress again that we did not prove a causal relationship- but generated a valid hypothesis worth further exploration. This theme has guided us when we wrote the manuscript and every time we addressed it. We again thank Drs Cherrie and MacCalmans for their comments; we believe our hypothesis is valid however future studies are needed to provide definitive proof.

Conflict of Interest:

None declared

We are pleased that Margel and Fleshner recognize the data errors in their original manuscript [1], which have a dramatic impact on the outcome of the ecological analysis [2]. Using the correct dataset and extending the analysis to include all countries, which we agree is appropriate, the authors report that the correlation between oral contraceptive use and prostate cancer mortality was much reduced (Pearson correlation coefficient r=0.17 compared with r=0.53 previously). It is worth noting that this result is only marginally statistically significant and, if real, suggests that something less than 3% of the variation in national rates for prostate cancer mortality might be due to oral contraceptive use. This correlation is only a little higher than that found for condom use (r=0.13), which was dismissed as non-significant.

It is easy to explore the inter-relationships between contraception and risk a little further. Using the full dataset we have undertaken a more comprehensive multivariate regression analysis, as Margel and Fleshner did on the sample in their original paper. i.e. using the four forms of contraceptive use, country GDP and prostate cancer mortality. This analysis showed that two forms of contraception were highly significantly associated with mortality (IUD was negatively associated and barrier methods positively associated), but use of oral contraceptives was not statistically significant in the model. Carrying out the same analysis on the incidence data we found that three of the four contraception methods were significantly associated with prostate cancer incidence (the use of the pill, vaginal barrier methods and condoms were all positively associated).

This was a hypothesis-generating exercise and the univariate results presented by Margel and Fleshner, along with the multivariate results discussed here, suggest that any hypothesis on the relationship between contraception use and prostate cancer mortality and incidence should not be reduced to a simple hypothesis about oral contraception.

Our main concern about the Margel and Fleshner paper [1] stems from the extensive coverage that it has received in the lay press and the likelihood that the erroneous findings have been widely interpreted as being causal. For example, we note that in the British Daily Mail newspaper the study was reported with the headline 'Rise in prostate cancer due to use of the Pill which increases men's exposure to oestrogen' [3]. We accept that many research publications ultimately turn out to present false conclusions [4]; that there is a need for researchers to float interesting ideas; and that researchers are not responsible for over -interpretation by the press. However, we believe that it is of prime importance that scientists endeavor to frame their results and their speculations in ways that limit the potential for misinterpretation. Margel and Fleshner did not do this: the original data were wrong; the authors' analysis and interpretation of the correct data remains unconvincing; and the title of their paper is misleading.

We reiterate the point we made in our original letter [1], that there is no evidence that oral contraceptive use is associated with prostate cancer mortality and the evidence that it is associated with prostate cancer incidence is weak. As a hypothesis generating exercise these data suggest that there may be some relationships between contraception and prostate cancer; that these are not simple and are not limited to one form of contraception; and that, if real, they explain at most a small proportion of differences between countries in risks of prostate cancer mortality.

John W Cherrie Laura MacCalman

References

1. Margel D, Fleshner NE. Re:Oral contraceptive use is not associated with prostate cancer BMJ Open 2011; 1(2): e000311.

2. Margel D, Fleshner NE. Oral contraceptive use is associated with prostate cancer: an ecological study. BMJ Open 2011; 1(2): e000311.

3. Rise in prostate cancer 'due to use of the Pill which increases men's exposure to oestrogen'. Daily Mail, 16th November 2011. (http://www.dailymail.co.uk/health/article-2061253/Birth-control-pills- Prostate-cancer-rise-contraceptive-pill-mens-oestrogen-exposure.html).

4. Ioannidis, J.P.A., 2005. Why most published research findings are false. PLoS Medicine, 2(8), p.e124. References

Conflict of Interest:

None declared

We are thankful for Dr Cherrie`s astute observations. After careful review of our data it appears that we indeed miscoded several points in our data set, which have resulted in an error in the analysis. We therefore decided to make sure our hypothesis is in reality supported by the data. We re-analyzed the correlation between mode of contraceptive use and prostate cancer (PCa)- using data from all countries available on the United Nations World Contraceptive Use 2007 report (167 countries in the new analysis vs. 88 in the former).

We were able to locate 160 countries (96%) on the International Agency for Research on Cancer web and to retrieve age-standardized rates of country-specific prostate cancer incidence and mortality in 2008. We repeated the same correlation analysis and we still find a significant correlation between oral contraceptives and prostate cancer mortality (r=0.17,p=0.04). We found no correlation with condom use (r=0.13, p=0.1). In addition we also found a negative correlation between PCa mortality and IUD use (r=-0.19, p=0.02). This means that in countries where IUD is used more often men are less likely to die of PCa. We believe that strengthens our hypothesis and not vice-versa.

We also re-analyzed the correlation between PCa incidence and OC using data from all available countries. The Pearson correlation coefficient between OC use and PCa incidence is 0.5 (p<0.001) and is still significant after adjusting for gross domestic product (GDP). The correlation is even stronger in the 32 European countries included (r=0.64 p<0.001).

In conclusion, we do acknowledge our error however after our re- analysis we still feel that our hypothesis is supported by the data. As we stressed earlier this is a hypothesis-generating manuscript, we do not believe we have proved a cause and effect relationship. We hope this manuscript will encourage further research in the growing field of endocrine disruptive compounds (EDCs) and the possible effect they may have on human lives.

Sincerely,

David Margel, MD

Neil Fleshner, MD

Data deposited in the Dryad repository: doi:10.5061/dryad.ff6bd0pq (http://datadryad.org/)

Conflict of Interest:

None declared

Margel and Fleshner (1) present results from an ecological study of associations between oral contraceptive (OC) use and prostate cancer. They hypothesize that men inadvertently ingest OCs or their by-products, passed from female urine into the environment or drinking water, and that the consequent increased low-level oestrogen exposure causes prostate cancer. The authors used data from a stratified random sample of countries; they report that for each country they used age-standardised rates of prostate cancer incidence and mortality along with an estimate of the proportion of women aged 15 to 49 years using OCs and other forms of contraception. They reported a statistically significant association between OC use and prostate cancer incidence and mortality, and no significant associations between prostate cancer and other forms of birth control.

In their analysis the association was stronger for incidence than for mortality, but it is possible that this may partly be due to a bias from prostate specific antigen (PSA) testing. It is known that prostate cancer incidence rates increased gradually from 1970 and then more rapidly from the 1990s, mainly due to the increasing use of the PSA test. It could be that the use of PSA tests may be more prevalent in countries with higher levels of affluence and higher use of OC. Although the authors used GDP to attempt to control for potential bias related to affluence this may not have been completely successful. We have therefore mainly focused our comments on their mortality analysis.

With reference to the data sources cited by Margel and Fleshner we have identified errors in the data as used and reported in their paper. In Figure 2A there are several points for death rate around 40 per 100,000, but in the IARC database used by the authors (i.e. GLOBOCAN. http://globocan.iarc.fr/factsheets/cancers/prostate.asp; Prostate Cancer Incidence and Mortality Worldwide in 2008) the maximum age-standardised death rate for the group of countries studied (Appendix 1 of the paper) is 26.1 per 100,000. There are other discrepancies between the data presented and the IARC data; for example, Thailand, with high OC use of 30.9%, but prostate cancer rate <2 per 100,000, does not appear in the authors' Figure 2A.

We checked whether the authors had inadvertently used crude death rates rather than age-standardised death rates as they had reported. Crude death rates are higher than 26 per 100,000 for some countries but these still do not match the authors' plotted data; and in any case using these rates would be inappropriate for comparison because of the inherent differences in the age distribution between countries.

To some extent these errors are identifiable without reference to the original sources: the mortality and incidence data for European countries (Figure 2B and 1B) are incompatible with the data presented for the whole group (Figures 2A and 1A) of 88 countries (not 87 as stated in the paper). In particular, the prostate cancer incidence rates for European countries (Figure 1B) are partly higher than the rates for the larger dataset; and the European mortality data (Figure 2B) include points with prostate cancer mortality data above 30 per 100,000 which do not appear in Figure 2A (and which, as noted, are not at all in the underlying IARC dataset). In addition, some of the patterns of coefficients and p-values in Table 1 seem curious, e.g. in the mortality analyses reported, we would have expected the 95% CI (-0.4 to 3) to be symmetric around the reported coefficient (-0.02); and the 95% CI of 0.04 to 0.9 for GPD suggests statistical significance <0.05, rather than 0.9 as given.

Using the correct data from the original sources specified in the paper, and recalculating the Pearson correlation coefficient between OC use and age-standardised prostate cancer death rate, shows that the association is much lower and not statistically significant (r=0.17, p=0.1 rather than r=0.53, p<0.05 as quoted in the paper). Further, fitting a linear model to the correct data, using the factors listed in Table 1, did not show strong evidence that OC use was related to age-standardised mortality - OC was not statistically significant in any of a series of models with different combinations of other explanatory variables, and not even nearly significant statistically when IUD and condom use were also included. (We used unweighted linear regression, as the authors seem to have done, though the IARC data show large differences between countries in the numbers of deaths contributing to the country-specific age- standardised rate.)

Furthermore, and in contrast to the original paper, we found clear evidence that prostate cancer mortality was associated with both condom use (positive association) and intrauterine device use (negative association), both associations being clearly statistically significant separately and in combination.

These limited checks and associated analyses suggest that Margel and Fleshner did not use the data as they had described, and that their conclusions would be very different if they had done so. We therefore conclude that in this ecological analysis there is in fact no clear or strong evidence of association between OC use and mortality from prostate cancer; and these mortality findings, together with the inconsistencies in the incidence data of Figure 1, raise questions about the incidence results also. While we have not verified all of the analyses carried out by Margel and Fleshner we suggest that the main findings are unsound.

We believe the most informative evidence as to whether there is causal link between OCs and prostate cancer is likely to come from cancer incidence and mortality studies amongst male-to-female transsexuals (MtF). MtF are treated with high doses of estrogens combined with anti-androgens. While there are occasional case reports of prostate cancer amongst this group (2) there is no convincing evidence of an increased risk. For example, Asscheman and colleagues (3) studied 966 MtF individuals who had received long-term high-dose estrogen treatment with OCs and found one case of prostate cancer (grouped with other cancers in the paper the SMR was not statistically significant).

While Margel and Fleshner clearly acknowledge their ecological analysis was intended to help generate a hypothesis that could be further tested we believe that the errors in their data seriously undermine their argument. From our analysis there seems to be no basis to conclude that OC use is associated with prostate cancer, as is stated in the title of the paper. We call on the authors to review the data they have used, acknowledge the errors in their analysis and if, as we think, the main conclusions are unsound, to retract the paper because the title claims a result which seems not to be supported by the data they intended to use.

References

1. Margel D, Fleshner NE. Oral contraceptive use is associated with prostate cancer: an ecological study. BMJ Open 2011; 1(2): e000311.

2. Molokwu CN, Appelbaum JS, Miksad RA. . Detection of prostate cancer following gender reassignment [4]. BJU International 2008; 101(2): 259.

3. Asscheman H, Giltay EJ, Megens JAJ, et al. A long-term follow-up study of mortality in transsexuals receiving treatment with cross-sex hormones. European Journal of Endocrinology 2011; 164(4): 635-642.

Conflict of Interest:

None declared

The paper by Margel and Fleshner reporting a direct correlation between oral contraceptive use and prostate cancer incidence and mortality rates1 is interesting and there may be an effect of oral contraceptive use on risk of prostate cancer. However, the study was not conducted according to the general rules for such studies. While ecological studies are a very useful approach for determining links to cancer (e.g., Ref. 2), they should include known risk factors in the analysis. In the case of prostate cancer, a number of international ecological studies have found significant correlations between dietary factors and prostate cancer rates.3,4 The link to animal products has been supported in a dietary intervention study5 and a case-control study.6

Another factor that may be a risk factor for prostate cancer is prevalence of apolipoprotein E epsilon 4 (ApoE4) allele.7 When used in an international ecological study, ApoE4 prevalence was significantly correlated with prostate cancer incidence and mortality rates.8 Prevalence of ApoE4 can also explain the often found inverse correlation between prostate cancer and diabetes mellitus.9 The mechanism whereby ApoE4 might increase the risk of prostate cancer is through increased production of cholesterol in the liver. Cholesterol is a risk factor for prostate cancer.10

To further evaluate the possible association of oral contraceptive use with prostate cancer, the ecological study in Ref. 7 was extended by adding the data for pill use from Ref. 12 in (1). Ref. 8 used data only for countries with dietary supply data available from the Food and Agriculture Organization.11 The regression results are given in Table 1. The data were analyzed for all 111 countries as well as the 95 countries with ApoE4 prevalence less than 30%, which eliminated the sub-Saharan African countries. As can be seen, use of the contraceptive pill was not significantly correlated with prostate cancer incidence or mortality rate after consideration of the Bernoulli principle, which requires p<0.05/n, where n is the number of factors included in the analysis.

The hallmark of a good hypothesis is that it can be tested in part through predictions. The ApoE4 hypothesis predicts that prostate cancer rates for African-Americans would be about twice that for white-Americans, which is, indeed, the case.12 Evidently there was no difference in use of oral contraceptive pills by African-American and white-American women.13 While Ref. 1 proposed that there might be an effect of oral contraceptives through water supplies, it does not seem that the oral contraceptive hypothesis could be tested in the United States.

References 1. Margel D, Fleshner NE. Oral contraceptive use is associated with prostate cancer: an ecological study. BMJ Open 2011;1:e000311. doi:10.1136/bmjopen-2011-000311 2. Armstrong B, Doll R. Environmental factors and cancer incidence and mortality in different countries, with special reference to dietary practices. Int J Cancer 1975;15:617-31. 3. Grant WB. A multicountry ecologic study of risk and risk reduction factors for prostate cancer mortality. Eur Urol 2004;45:271-9. 4. Colli JL, Colli A. International comparisons of prostate cancer mortality rates with dietary practices and sunlight levels. Urol Oncol 2006;24:184-94. 5. Frattaroli J, Weidner G, Dnistrian AM, Kemp C, Daubenmier JJ, Marlin RO, Crutchfield L, Yglecias L, Carroll PR, Ornish D. Clinical events in prostate cancer lifestyle trial: results from two years of follow-up. Urology 2008;72:1319-23. 6. John EM, Stern MC, Sinha R, Koo J. Meat consumption, cooking practices, meat mutagens, and risk of prostate cancer. Nutr Cancer 2011;63:525-37. 7. Lehrer S. Possible relationship of the apolipoprotein E (ApoE) epsilon4 allele to prostate cancer. Br J Cancer 1998;78:1398. 8. Grant WB. A multicountry ecological study of risk-modifying factors for prostate cancer: Apolipoprotein E ?4 as a risk factor and cereals as a risk reduction factor. Anticancer Res 2010;30 189-99. 9. Grant WB. The roles of ultraviolet-B irradiance, vitamin D, apolipoprotein E ?4, and diet in the risk of prostate cancer. Cancer, Causes & Control 2011;22:157-8. 10. Iso H, Ikeda A, Inoue M, Sato S, Tsugane S. Serum cholesterol levels in relation to the incidence of cancer: The JPHC Study Cohorts. Int J Cancer 2009;125: 2679-86 11. Food and Agriculture Organization. Food Balance Sheets. 1992-94 Average. Food and Agriculture Organization of the United Nations. Rome. 1996. 12. Albano JD, Ward E, Jemal A, Anderson R, Cokkinides VE, Murray T, Henley J, Liff J, Thun MJ. Cancer mortality in the United States by education level and race. J Natl Cancer Inst 2007;99:1384-94. 13. Piccinino LJ, Mosher WD. Trends in contraceptive use in the United States: 1982-1995. Fam Plann Perspect 1998;30:4-10, 46.

Table 1. Regression results for prostate cancer incidence using data for energy derived from cereals, the square root of gross domestic product (GDP), prevalence of ApoE4, and rate of use of the contraceptive pill.

Incidence Cereals, GDP, ApoE4, Pill, Adj. R2, F, p.

111 countries -0.34,*; 0.46,*;0.17,0.03;0.15,0.04; 0.15,39,*.

95 countries -0.30,*;0.38,*;0.34,*;0.12,0.08; 0.67,48,*.

Mortality

111 countries -0.41,*;0.09,0.36;0.27,0.003;0.20,0.02; 0.40,19,*.

95 countries -0.41,*;0.02,0.85;0.44,*;0.14,0.08; 0.66,36,*

Conflict of Interest:

None declared