Résumé
Objectifs : Les objectifs de cette étude étaient d’évaluer si la distribution du stade parmi les
femmes diagnostiquées avec un cancer du sein diffère entre celles ayant reçu des implants mammaires pour fins esthétiques et celles sans implants, et évaluer si l’augmentation mammaire pour fins esthétiques précédent la détection d’un cancer du sein est un prédicateur de la survie au cancer du sein.
Méthodes : Une revue systématique d’études observationnelles avec deux méta-analyses a
été effectuée afin d’évaluer ces objectifs. Une recherche systématique de la littérature publiée avant février 2011 a été effectuée dans MEDLINE, EMBASE, Global health, CINAHL, IPAB et PsycINFO. Les publications éligibles étaient celles qui incluaient des femmes diagnostiquées avec un cancer du sein ayant reçu une augmentation mammaire pour fins esthétiques.
Résultats : Notre première méta-analyse, basée sur 12 études, a démontré aucune
différence statistiquement significative pour l’association entre les implants mammaires et un stade avancé au diagnostique du cancer du sein (Rapport de Cote (RC) global d’avoir un stade non localisé au diagnostique du cancer du sein : 1,26, IC à 95%: 0,99 à 1,60; I2 = 35,6 %). Toutefois, l’effet global pour les 5 études qui ont fournies des mesures d’association
ajustées pour d’importants facteurs confondants a démontré une association statistiquement significative entre les implants mammaires pour fins esthétiques et un stade avancé au diagnostique du cancer du sein (RC = 1,51, IC à 95%: 1,18 à 1,92). La seconde méta- analyse, basé sur 5 études, a évalué l’association entre l’augmentation mammaire et la survie au cancer du sein. Cette méta-analyse a identifiée une réduction de la survie au cancer du sein chez les femmes avec implants mammaires pour fins esthétiques (Rapport de Taux de mortalité spécifique au cancer du sein : 1.38, IC à 95%: 1.08 à 1.75).
Conclusion : La littérature publiée à ce jour suggère que les femmes avec des implants
mammaires pour fins esthétiques ont un stade avancé au diagnostique du cancer du sein lorsque les estimés sont ajustés pour d’importants facteurs confondants. De plus, cette étude démontre que l’augmentation mammaire pour fins esthétiques affecte négativement la survie parmi celles préalablement diagnostiquées avec un cancer du sein. Ces résultat doivent être interprétés avec prudence alors que certaines études inclues dans la méta-
analyse portant sur la survie n’ont pas ajusté pour des facteurs confondants. Davantage d’investigations sont nécessaire concernant le diagnostique et le pronostic du cancer du sein parmi les femmes avec des implants mammaires.
Abstract
Objectives: To evaluate whether the stage distribution among women diagnosed with
breast cancer differs between those who received breast implants for cosmetic purposes and those with no implants, and to evaluate whether cosmetic breast augmentation prior to the detection of breast cancer is a predictor of post-diagnosis survival.
Methods: A systematic review of observational studies with two meta-analyses was
undertaken to asses these objectives. A systematic search of the literature published prior to February 2011 was conducted in MEDLINE, EMBASE, Global health, CINAHL, IPAB & PsycINFO. Eligible publications were those that included women diagnosed with breast cancer and had augmentation mammaplasty for cosmetic purposes.
Results: Our first meta-analysis, based on 12 studies, suggested an association between
cosmetic breast implants and later stage at breast cancer diagnosis (Overall odds ratio of having non-localized breast cancers at diagnosis: 1.26, 95 % confidence interval (CI): 0.99 to 1.60; I2 = 35.6 %). However, the overall effect for the 5 studies that provided measures
of association adjusted for relevant confounding factors showed a statistically significant association between cosmetic breast implants and advanced breast cancer at diagnosis (Odds ratio = 1.51, 95 % CI: 1.18 to 1.92). The second meta-analysis, based on 5 studies, evaluated the relationship between cosmetic breast implantation and survival. This meta- analysis showed reduced breast cancer survival among women who received implants compared to those who did not (Overall breast cancer-specific mortality hazard ratio: 1.38, 95 % confidence interval (CI): 1.08 to 1.75).
Conclusions: The research published to date suggests that, at diagnosis, breast cancers are
at more advanced stages among those with cosmetic breast implants when adjusting for relevant covariates. Additionally, there is evidence that cosmetic breast augmentation adversely affects the survival experience of women who are subsequently diagnosed with breast cancer. These findings should be taken with caution as some studies included in the meta-analysis on survival did not adjust for potential confounders. Further investigations are warranted regarding breast cancer diagnosis and prognosis among augmented women.
Introduction
Cosmetic breast augmentation has become increasingly popular (1). In the US, for example, cosmetic breast augmentation was the most commonly performed cosmetic surgical procedure in 2010 with 296,000 surgeries performed (2) an increase of approximately 800 % compared with the early 1990’s. Although breast augmentation is popular, controversies about the long-term health effects of breast implants remain.
The weight of evidence from epidemiological studies indicates that cosmetic breast implants are not associated with increased breast cancer risk (3-29). Concern remains, however, that implants may impair the ability to identify breast cancer at an early stage by mammography because cosmetic breast implants are radiopaque, impairing the visualization of breast tissue with mammography and making it more difficult to detect breast cancer at an early stage (30-33). Specialized radiographic techniques have been developed for women with breast implants to improve visualization which involve displacing the implant posteriorly against the chest wall and pulling breast tissue over and in front of the implant (33-36). However, one-third of the breast is still not adequately visualized despite such techniques, leading to an increase of false-negative mammograms (32). It is estimated that 1 in 8 women will be diagnosed with breast cancer some time in their lives (37). Therefore, a number of augmented women will eventually develop breast cancer which raises concerns regarding possible effects of implants on breast cancer detection.
Most studies that evaluated the detection of breast cancer among women with cosmetic breast implants compared the stage distribution of breast cancer at diagnosis between augmented and non-augmented women. The findings from these studies have been inconsistent. For instance, some studies reported that women with breast augmentation may be more likely to be diagnosed with advanced cancers (31;38-40) while others have reported no such difference (4;5;9;11;14;15;17;27;29;41-49). These conflicting results may be explained by methodological issues within studies as well as the small number of incident breast cancer cases in these studies which limit statistical power to obtain
significant results. In addition to the question of breast cancer detection, no study to date has been able to establish that women with breast implants, although they may be diagnosed at a more advanced stage, have a poorer survival following breast cancer diagnosis compared with non-augmented women, but these studies were also impaired by relatively low statistical power (40;42-45;48). Better understanding of the detection of breast cancer and survival patterns following breast cancer diagnosis among augmented women will aid in giving clear information on the consequences of breast augmentation surgery to these women and their physicians. The fact that implants may interfere with the early detection of breast cancer is particularly relevant and carries with it important clinical and public health implications.
Recent reviews that summarized the evidence of the long term effects of cosmetic breast implants concluded they were not associated with advanced breast cancers nor was survival affected (12;24). Although these papers were an important step forward, they were not presented as systematic reviews and were based on a qualitative rather than quantitative analysis. Through a systematic literature search, we have identified additional papers published in the past decades that were not captured by previous reviews as well as two more recent publications (40;48) providing suitable data for a quantitative meta-analysis on the diagnosis and prognosis of breast cancer among augmented women.
Specifically, our objectives were to verify the stage distribution of breast cancer and post- diagnosis survival among cosmetic breast implant women compared to non-augmented women by means of a systematic review and meta-analyses. We also sought to identify sources of heterogeneity in risk estimates in the existing literature and identify gaps in the current state of knowledge. This investigation is important in order to consolidate the existing knowledge on the long term effects of cosmetic breast implants.
Materials and methods
Searchstrategy
To identify eligible studies published before February 1st 2012, we applied a systematic literature search strategy to the following electronic databases: MEDLINE, EMBASE, Global health, CINAHL, IPAB & PsycINFO. The Cochrane Library Database of Systematic Reviews was also searched. The following keywords and subject headings were used in combination to identify relevant articles in electronic databases: breast neoplasms AND (breast implants OR breast augmentation OR mammaplasty OR mammoplasty OR breast implantation OR breast prosthesis) AND (women without implants OR non augmented women) AND (delayed diagnosis OR prognosis OR survival OR delayed detection OR staging). Reference lists from retrieved articles (4;5;9;11;14;15;17;29;31;38- 46;48;49) and published reviews (12;16;24;26) were manually examined to identify additional manuscripts. Eligible articles were original peer-reviewed published studies. Abstracts from identified articles were reviewed to assess eligibility. Additionally, as direct contact with experts has been shown to be an effective method of retrieving relevant articles, we surveyed international experts who published papers on breast cancer detection among women with cosmetic breast implants and associated survival rate patterns to request any relevant published or unpublished scientific articles (50). The search was limited to French and English articles.
Study eligibility
Eligible publications were those that included women diagnosed with breast cancer and had antecedent augmentation mammaplasty for cosmetic purposes. The comparison group consisted of women diagnosed with breast cancer who have had other common elective cosmetic surgeries or who were from the general female population.
Eligible publications for the evaluation of the association of breast implants with the stage distribution of breast cancer had to include the number of women with breast implants diagnosed with breast cancer (exposed group) and women without implants diagnosed with breast cancer (unexposed group) per stage of breast cancer at diagnosis or per status of
nodal involvement and/or metastases. Measures of association describing the risk or the odds of having non-localized breast tumors (nodal involvement positivity and/or metastases to distant sites) comparing the exposed breast cancer cases to the unexposed cases were used if provided in the paper. Otherwise, the crude odds ratios, their respective standard errors and 95 % confidence intervals were computed from the contingency tables.
Publications eligible for the evaluation of breast implants and survival following breast cancer diagnosis provided either hazard ratios comparing the breast cancer mortality rate after diagnosis between the exposed and unexposed group or provided Kaplan-Meier breast cancer specific survival curves graphically for both augmented women with breast cancer and non-augmented women with breast cancer. For the latter studies, hazard ratios were estimated from published statistics provided in the manuscripts or by extracting data directly from Kaplan-Meier survival curves using recommended techniques for time to event meta-analysis (51;52).
When several publications were available for the same study group, the most recent one was retained for analysis. Publications without a comparison group for the implant subjects were excluded. So were those that did not provide measures of association and did not provide crude numbers in contingency tables allowing calculation of measures of association regarding stage of breast cancer at diagnosis.
A certified librarian performed the search and two authors (E.L.; S.Y.P.) excluded studies at the first stage of eligibility evaluation. Studies identified for a more detailed assessment based on the abstract and title were discussed. Study inclusion in the meta-analysis was agreed upon by co-authors without blinding to study characteristics.
Data abstraction
Study characteristics that were extracted and reviewed included the following: source of data on implant, source of data on breast cancer diagnosis, source of data on mortality, assessment of stage of breast cancer, nodal involvement, type of comparison group, number of women with implants with breast cancer, number of women without implants with breast
cancer, mean length of follow-up, average age at breast cancer diagnosis, whether statistical adjustment for potential confounders was done (e.g. adjusting for age at diagnosis, period of diagnosis) and results (cell counts, odds ratios, hazard ratios and 95 % confidence intervals). Adjusted estimates were always selected over unadjusted estimates when provided in the paper. A dichotomous variable was created for the exposure variable (presence of breast implants with breast cancer vs. no implants with breast cancer) and for the outcome (stage distribution of breast cancer). Staging of breast cancer among the studies included principally the American Joint Committee on Cancer (AJCC)’s Tumor Node Metastasis (TNM) classification (53;54) without limiting the edition of the AJCC classification that was used and the U.S National Cancer Institute’s Surveillance Epidemiology and End Results (SEER) Summary Stage System (55). For the purpose of consistency in the analyses, we excluded, when possible, ductal carcinoma in situ (DCIS), lobular carcinoma in situ (LCIS) and all other non-invasive breast cancers. This was done because several publications included in our meta-analyses excluded these cases on the basis of data quality issues related to the reporting of in situ cancers. Therefore, only invasive breast cancers were considered. Considering there is a considerable variability across studies for the classification used for stage distribution of breast cancer, especially between the AJCC’s TNM and SEER’s Summary Stage System, we dichotomized the response variable as non-localized breast cancer (regional or distant) vs. localized breast cancer. This cutoff was first chosen for clinical relevancy because localized breast tumors are potentially more curable than non-localized tumors and associated with better survival rates (56). Additionally, we chose this cut off for compatibility purposes between the two cancer staging systems. Irrespective of classification system, breast tumors that did not spread to regional lymph nodes and/or distant sites were considered localized tumors. Therefore, non-localized breast tumors were considered advanced or later stage breast cancers.
Statistical analyses
The statistical analyses, forest plots, sensitivity, meta-regression and publication bias analyses were produced with Stata software, version 11 (57). Studies had to provide sufficient data to calculate an effect-size measure in order to be included in the quantitative
analysis. Dersimonian-Laird random effects model (58) was used to derive a pooled effect across studies for the association between cosmetic breast implants and stage distribution of breast cancer. The random effects model was used because it accounts for variations between studies in addition to sampling error within studies (59). All analyses were conducted on the natural log scale. The summary odds ratio with 95 % confidence interval was calculated from study-specific adjusted odds ratios taken directly from the study or estimated as crude odds ratios from cell counts. Study-specific confidence intervals were also taken directly from the study if reported or were calculated using the corresponding standard error. A random effects model was also used to calculate the summary hazard ratio for the association between cosmetic breast implants and survival. The pooled effect was calculated from study-specific hazard ratios that were obtained directly from the study or calculated from survival curves. In order to quantify the degree of heterogeneity across studies, we used Cochrane’s Q test (60) and the Higgins’ I-squared statistic with 95 % confidence intervals (61). The latter statistic indicates the proportion of the variance attributable to between-study variability (61). We also analyzed the influence of individual studies by omitting each study one by one in order to identify studies contributing disproportionately to the observed heterogeneity. A visual inspection was also performed using the Galbraith plot (62;63) to detect possible outlier studies that have an excessive influence on the overall estimate. To identify potential sources of heterogeneity, we examined only the association between cosmetic breast implants and stage distribution of breast cancer as there were too few studies to examine sources of heterogeneity for the association between cosmetic breast implants and survival. This was done by calculating a summary odds ratio across strata of factors selected a priori as potentially related to study quality and that were present enough across studies to perform the stratification. These factors included source of comparison group (other cosmetic surgery controls vs. population-based controls), source of exposure data (plastic surgeon records vs. medical records), breast cancer staging system (TNM staging vs. SEER staging) and statistical adjustment of the OR for potential confounders (adjusted vs. unadjusted). An evaluation of the impact of these factors on heterogeneity was also done using random-effects meta- regression models. The latter investigates how a categorical or continuous characteristic at the study level is associated with the effect estimate in the meta-analysis (64). The outcome
variable in the meta-regression models in this study is the odds ratio and the explanatory variables, also called potential effect modifiers, are the factors selected a priori as potentially related to study quality. The year of publication of each study as a potential source of heterogeneity was also evaluated as a continuous variable in a meta-regression model. Assessment of publication bias for the association between cosmetic breast implants and stage distribution of breast cancer was done using a funnel plot and Egger’s test(65). There were not enough studies to examine publication bias for the association between cosmetic breast implants and survival.
Results
We identified 267 unique papers after searching MEDLINE, EMBASE, Global health, CINAHL, IPAB & PsycINFO. Of these, 22 studies (n=28,924 women) met eligibility for the evaluation of stage distribution of breast cancer and breast implants and 6 studies (n=18,026 women) met eligibility for the evaluation of survival following breast cancer diagnosis and breast implants (Figure 1). One study, Xie et al. (40), included in both the meta-analysis on stage distribution of breast cancer and the one on breast cancer survival was recently up-dated and updated results (unpublished work) are used in the meta- analyses (66). No papers were identified through further investigation of previous reviews or through manual examination of references or querying of the experts. A quality assessment scale (67) also showed that publications that met eligibility are of acceptable quality to be included in these meta-analyses (supplemental file in appendix 1).
Breast implants and stage distribution of breast cancer
Twelve studies, all being cross-sectional in their design, provided sufficient data to be included in a meta-analysis to evaluate the association between cosmetic breast implants and stage distribution of breast cancer. Characteristics of the 12 publications meeting the inclusion criteria and selected for the quantitative analysis are provided in Table 1. Most of these were published after the year 2000 and were conducted in the United-States. The remainder were conducted in northern Europe or Canada. The other 10 papers were all
excluded because they overlapped with more recent publications with an extended follow- up of the same study group.
Results of the meta-analysis are depicted in Figure 2. The size of each box indicates the relative weight of each publication in the meta-analysis and the bars show the 95 % confidence intervals (CIs). Based on the 12 studies, the summary odds ratio with the random effects model was 1.26 (95 % CI: 0.99 to 1.60) for a non-localized stage of breast cancer at diagnosis comparing augmented women with breast cancer to non-augmented women with breast cancer. Moderate heterogeneity was observed (Q = 17.07, P = 0.11, I2 = 35.6 %).
Sensitivity analysis and publication bias
Sensitivity analyses revealed that one publication, Clark et al. (41), accounted for all the