Thank you for your ELetter. We agree with you, this is why we took issue with the misleading media coverage and made our best efforts to rectify the message where we felt it had been distorted (both in print and in radio interviews). See our rebuttal to the (London) Times for an example of this:

"Sir
In relation to “Light drinking does no harm in pregnancy”, The Times 12/09/2017
We write to you to complain about the highly misleading front page coverage that your paper dedicated yesterday to our scientific study, and to rectify the wrong messages you have propagated.
Frustratingly, in today’s paper your columnist has said: “Alcohol […] drinking in pregnancy, which many health professionals considered a crime only a month ago, now appears to be acceptable in moderation” (from “Take health advice with a big pinch of salt” The Times 13/09/2017). This continues the misinformation that yesterday’s article started. To say that “light drinking does no harm in pregnancy” is a gross misrepresentation of our findings – detailed in the scientific paper, summarised in the press release, and distilled and interpreted in plain language by one of our lead authors in conversation with your journalist.

We went to great lengths to stress that ‘little or no evidence does not mean little or no effect’. In other words, we have little evidence that light drinking in pregnancy is harmful, but we also have little evidence that it is safe! Conversely, your bold front page headline conveys certainty around its safety, which is irresponsible.

Such misreporting may boost newspaper sales, but does not benefit public health. Providing a health message in plain contradiction to existing Department of Health guidelines, when this is explicitly not supported by the research supposedly being reported, is irresponsible. Today’s unborn babies won’t be buying copies of your paper anytime soon, but one day this misreporting could cost them, and the nation looking after them, very dearly."

Yours,

Dr Luisa Zuccolo, on behalf of the authors of “Low alcohol consumption and pregnancy and childhood outcomes: time to change guidelines indicating apparently ‘safe’ levels of alcohol during pregnancy? A systematic review and meta-analysis. BMJ Open doi 10.1136/bmjopen-2016-015410”

Despite our hard work communicating the message without ambiguity to the media and answering personally all media calls, some coverage ‘got it wrong', which was completely beyond our control. We appreciate efforts by public health and clinical colleagues like yourselves to point out which messages are misleading or plain wrong, striving for trustworthy information in this dangerous era of 'Fake news'. "

The open access journal 'Roars Transactions, a Journal on Research Policy and Evaluation' has published on 2 November 2017 a paper with reflections on the unavailability of the ICMJE disclosure form of Dr. Moylan.

The title of this paper is: 'Is partial behaviour a plausible explanation for the unavailability of the ICMJE disclosure form of an author in a BMJ journal?'.

The paper can be accessed at https://riviste.unimi.it/index.php/roars/article/view/9073 The paper is published in the section ‘Discussion notes’. The editors of 'Roars Transactions, a Journal on Research Policy and Evaluation' are encouraging readers to submit comments / responses. These comments / responses will be published alongside the paper. I am hereby inviting the readers of this eletter to submit comments / responses about this topic to both journals (BMJ Open and Roars Transactions, a Journal on Research Policy and Evaluation). Copy/pasted from the Abstract of my new paper:

'This case study about the ethical behaviour in the field of scholarly publishing documents an exception on the rule for research articles in the medical journal BMJ Open that ICMJE disclosure forms of authors must be made available on request. The ICMJE, the International Committee of Medical Journal Editors, has developed these forms for the disclosure of conflicts of interest for authors of medical publications. The case refers to the form of the corresponding author of an article in BMJ Open on retraction notices (Moylan and Kowalczuk, 2016). The corresponding author is a member of the council of COPE, the Committee on Publication Ethics. I will argue that the unavailability of the form relates to personal conflicts of interest with the corresponding author about my efforts to retract a fatally flawed study on the breeding biology of the Basra Reed Warbler Acrocephalus griseldis. I describe my attempts to get the form and I will argue that its unavailability can be attributed to partial behaviour by BMJ, the publisher of BMJ Open. This study complements other sources reporting ethical issues at COPE.'

Dear Editor:

We thank Freitas-Junior et al. for their detailed letter, which we read with great consideration, specially coming from leaders of the Brazilian Mastology Society.
We will reply starting from the points that we believe that we all agree. First, that breast cancer mortality is linked to “social and medical developments”. Better health and higher life expectancy are associated with indicators of social development, such as Human development index (IDH), higher income and education. But better health is not necessarily associated to more health care (1). People of higher income may have more cancer not because of their wealth (which was expected to protect them), but because they are subjected to higher “observational intensity”, the combined effect of factors such as the frequency of screening and diagnostic exams, the ability of the exams to detect small irregularities, and the threshhold used to label results as abnormal, which can be an undesirable consequence (overdiagnosis) of medical development. As pointed by Welch and Fisher (2017), “Excessive testing of low-risk people produces real harm, leading to treatments that have no benefit (because there is nothing to fix) but can nonetheless result in medication side effects, surgical complications, and occasionally even death” (1).
We also agree that radioprotection and radioinduced breast cancer are important issues to be addressed in a benefit-harm rationale, but we believe that the harms may have been underestimated. With outcomes similar to international studies, the careful research conduced by Corrêa et al. (2) estimated that for 100.000 women screened biannually in the State of Goiás, the number of radioinduced cancers ranged from 14.9 (50-70 years, conventional mammography) to 64.1 (40-70 years, digital mammography). In this setting, they found that the risk of radioinduced cancer incidence and mortality was higher in the private sector, and that 12% of the conventional mammography equipment and 50% of the digital ones (more common in private sector) used radiation doses above the recommendations, possibly indicating, as pointed out by the authors (2), on page 126, an “increased risk of cancer induction resulting from the introduction of digital technology in mammography”. Estimates of radioinduced cancers could be higher if considered the additional exposure of increased frequency (in 2010-2011, 44.6% women in Brazil reported an interval of one year or less since last mammography), and age range (48.8% of screened women were out of the recommended age 50-70y) (3). Both annual screening and digital mammography are more common in the private sector, leading to higher radiation exposure and higher observational intensity, which reinforces our hypothesis of underestimation of harms of intensive screening.
We also agree, as stated several times in our paper, that ecological analysis are not appropriate to affirm causality; the best research design would be randomized controlled trials, and their systematic reviews - the gold standard for causal links. In sharp contrast with public perception and Pink October campaigns, the Cochrane systematic review of clinical trials of mammographic screening shows no reduction in overall cancer mortality (4), which reinforces our hypothesis of overestimation of benefits. The ratio of lives saved and lives losts owing to mammography depends on the belief that mammography saves lives, and how many, which is controversial (6). As debated by other authors (5-9), mortality-to-incidence, survival time, and proportion of advanced lesions can be inappropriate measures to assess mortality reduction, since screening substantially increases incidence, and these measures can coexist with an increased mortality. In our paper, we dialogue with publications with results similar to ours, in Brazil and abroad, with different methodologies (ecological, spatial analysis, individual level, aggregate level, systematic reviews), reinforcing our hypothesis. Unfortunately, some of the post-graduation studies with similar findings in Brazil (2, 10, 11), are found only in university repositories and in Portuguese, contributing to publication bias.
We disagree that we need to increase mammography coverage or capacity. The authors say that the coverage of mammography is just 24% in the public system, not reaching the 70% necessary to produce an impact on mortality. But according to the Vigitel (national surveillance system of risk factors and protection for chronic diseases by telefone survey) with data from 2016, the coverage for the eligible women (aged 50-69y) in all Brazilian capitals was 78,2% for screening in the last two years (90,5% for more educated women) (12). This is higher than in the USA (71.6%) (13), Germany (52.7%), France (52,3%), the average in Europe (49.2%) (14), and Japan (36,4%) (15). In terms of mammography infraestructure, to fully atend the populational need for bianual screening, in 2012 there was already an excess of 196 mammography equipments in Brazil (irregularly distributed), and the State of São Paulo, the focus of our research, had 89 excedent equipments.(16) Annual screening is in disagreement with the Ministry of Health's recommendations and “may compromise its effectiveness and entail greater risk to women” (3).
We disagree that our analysis ignored the gross domestic product, human development index and income: as clearly described in our article, we tested models with these variables (V_68, V_69 and V_41, respectively), but they did not fit the data better than our final model. We agree with the fact that the final model is not necessarily perfect. However, this is the model, among all tested ones, with the highest probability of minimizing the estimated information loss.
Linear regression results are valid as long as model assumption are not violated. To ensure this, firstly we performed an exploratory analysis before modeling to, among other things, evaluate the presence of outliers and collinearity between the covariates and make the required adjustments (in this case, covariate transformation and elimination of highly correlated covariates). After modeling, we evaluate the behavior of the residuals, which is the way indicated to ensure that the assumptions of a linear regression model are not violated. Thus, we evaluated the normality and the mean of the residuals, their relationship with the dependent and independent variables, to ensure homoscedasticity, and finally, that the residuals were spatially independent, since we worked with a design with spatially aggregated data (municipalities).
For doing this last evaluation, we used the Moran Global Index to assess the existence of spatial autocorrelation of the residuals. Since we did not identify spatial dependency in the residuals, it was not necessary to consider more complex models, such as Simultaneous or Conditional Autorregressive models, Geographically Weighted regression, among other possibilities (16). According to the criterion of parsimony, if the model meets the basic assumptions of regression modeling, there is no reason to look for more complex models. We agree with Freitas-Jr et al. in their observation about the existence of spatial autocorrelation of the dependent variable (breast cancer mortality rates), however, this dependence was explained by the covariates considered, resulting in non-spatially autocorrelated residuals. When we work with spatially aggregate data, we have a wide range of spatial statistical tools available, but we need to be careful to choose the most appropriate ones so that our goals can be achieved. In this way, it would only make sense to use Moran Local index if we were interested in identifying clusters of municipalities with high (or low) mortality rates, which was not the case in our study.
We want to insist that the conclusion of our work is not that there is “a possible association” between increased mortality and those three variables (mammography, nulliparity and use of private sector); the association exists and is very robust, in all our models, as in the studies we present. The question is if there are causal links, and we present evidence of plausibility for this argument, from recent studies of different research designs. Our conclusion is that there is a possible overestimation of benefits and underestimation of harms in mammographic screening as a public health strategy.
This can be an outstanding opportunity to study real time outcomes of screening mammography, using randomization for informed decision and individualized risk assessment, evaluating differences in incidence, morbidity and mortality. We believe that we have enough uncertainty, and equipoise, to justify such a trial, and we would be glad to collaborate in this direction.

Best regards,

Carmen Simone Grilo Diniz
Alessandra Cristina Guedes Pellini
Francisco Chiaravalloti-Neto

References

1 - Welch HG and Elliott SF. Income and Cancer Overdiagnosis—When Too Much Care Is Harmful. N Eng J Med 2017;376.23:2208-2209.
2 – Corrêa RS, Peixoto JE, Ferreira RS, Tanaka R, Freitas-Junior R. Estimativa do risco de cancer radioinduzido em rastreamento mamográfico. In: Corrêa RS. Mamografia: infraestrutura, cobertura, qualidade e risco de cancer radioinduzido em rastreamento oportunístico no Estado de Goiás. PhD Thesis, UFG. 2012 (unpublished). Available at: https://repositorio.bc.ufg.br/tede/bitstream/tde/1534/1/Tese%20Rosangela.... Accessed on November 6, 2017.
3 - Tomazelli JG, Migowski A, Ribeiro CM, Assis M, Abreu DMF. Avaliação das ações de detecção precoce do câncer de mama no Brasil por meio de indicadores de processo: estudo descritivo com dados do Sismama, 2010-2011. Epidemiol Serv Saúde 2017;26.1:61-70.
4 - Gøtzsche PC and Karsten JJ. Screening for breast cancer with mammography. Cochrane Database of Systematic Reviews 2013;6:CD001877.
5 - Gigerenzer G. Full disclosure about cancer screening. BMJ 2016;352: h6967–2.
6 - Prasad V, Lenzer J, Newman DH. Why cancer screening has never been shown to "save lives"--and what we can do about it. BMJ 2016;352:h6080.
7 - Narod SA, Iqbal J, Miller AB. Why have breast cancer mortality rates declined? J Cancer Policy 2015;5:8–17.
8 - Welch HG, Black WC. Overdiagnosis in cancer. J Natl Cancer Inst 2010;102:605–13.
9 - Carter JL, Coletti RJ, Harris RP. Quantifying and monitoring overdiagnosis in cancer screening: a systematic review of methods. BMJ 2015;350:g7773.
10 - Bello MA. Análise estatística espacial da mortalidade por câncer de mama feminina no estado do Rio de Janeiro, 2001 a 2006. Dissertação. Rio de Janeiro: Escola Nacional de Saúde Pública Sergio Arouca, 2010. Available at: https://bvssp.icict.fiocruz.br/lildbi/docsonline/get.php?id=2331. Accessed on November 5, 2017.
11 - Silveira DP. Perfil Da incidência e da sobrevida de câncer de mama: análise a partir dos registros de câncer de base populacional e cobertura de planos privados de saúde no município de São Paulo. Escola Nacional de Saúde Pública Sergio Arouca. 2011. Available at: http:// bvssp.icict. fiocruz. br/ lildbi/ docsonline/ get. php? id= 2739. Accessed on November 6, 2017.
12 - Vigitel Brasil 2016: Vigilância de fatores de risco e proteção para doenças crônicas por inquérito telefônico. Ministério da Saúde, Secretaria de Vigilância em Saúde, Departamento de Vigilância de Doenças e Agravos não Transmissíveis e Promoção da Saúde. Brasília: Ministério da Saúde, 2017. Available at: http://portalarquivos.saude.gov.br/images/pdf/2017/junho/07/vigitel_2016.... Accessed on November 6, 2017.
13 – National Cancer Institute. Cancer Trends Progress Report. Breast Cancer Screening. 2017. USA: NIH. Available at: https://progressreport.cancer.gov/detection/breast_cancer. Accessed on November 6, 2017.
14 - International Agency for Research on Cancer. Cancer screening in the Europe Union. Lyon: IARC, 2017. Available at: https://ec.europa.eu/health/sites/health/files/major_chronic_diseases/do.... Accessed on November 6, 2017.
15 - Sano, Hiroshi, Rei Goto, and Chisato Hamashima. "Does lack of resources impair access to breast and cervical cancer screening in Japan?." PloS one 12.7 (2017): e0180819.
16 - Xavier DR, De Oliveira RAD, De Matos VP, Viacava F, Carvalho CC. Mammograms coverage, allocation and use of equipment in the Health Regions. Saúde Debate, Rio de Janeiro. 2016;40(110):20-35. Available at: http://www.scielo.br/pdf/sdeb/v40n110/0103-1104-sdeb-40-110-0020.pdf. Accessed on November 6, 2017.
17 - Bivand RS, Pebesma E, Gómez-Rubio V. Applied spatial data analysis with R. New York: Second Springer editor, 2013:405.

Dear Editor:
On reading the study conducted by Diniz et al. on the possible association between mammography and breast cancer-related mortality in the state of São Paulo, Brazil, I found two weaknesses that cannot go unnoticed. The authors have gathered information on mammography ratio and discussed the presumable effect on mortality of mammographic screening. These are two different situations. Having a mammogram is not the same thing as being screened with mammography. There are other reasons for having a mammogram, as having a palpable lump, being prepared for breast cancer surgery or systemic treatment, following up patients treated for breast cancer, orienting breast biopsies. When studying a population comprised of people who have actually died of breast cancer, one should expect to find a cluster of mammograms performed for non-screening purposes. And one should expect a higher mortality rate from breast cancer in this population.
The other weakness is perhaps easier to understand for those familiar with health care in Brazil. It is very well known that death certificates in Brazil do not always reflect the actual cause of death. It is to be expected a serious under notification in the public health sector as compared to the private health sector. I do acknowledge that the mortality information system in Sao Paulo state is extensive in its coverage and that could be perceived as a strength of the article. However, one has to make sure that the rate of adequate notification is similar in the private and public sectors before being able to adequately make this comparison. Without this certitude, the reliance on the death certificates becomes a weakness.