|
|
| |||||||||||||||||||||||||||||||||
Background The postulated relation between silicone breast implants and the risk of connective-tissue and autoimmune diseases has generated intense medical and legal interest during the past decade. The salience of the issue persists, despite the fact that a great deal of research has been conducted on this subject. To provide a stronger quantitative basis for addressing the postulated relation, we applied several techniques of meta-analysis that combine, compare, and summarize the results of existing relevant studies.
Methods We searched data bases and reviewed citations in relevant articles to identify studies that met prestated inclusion criteria. Nine cohort studies, nine case–control studies, and two cross-sectional studies were included in our meta-analyses. We conducted meta-analyses of the results of these studies, both with and without adjustment for confounding factors, and a separate analysis restricted to studies of silicone-gel–filled breast implants. Finally, we estimated the annual number of new cases of connective-tissue disease that could be attributed to breast implants.
Results There was no evidence that breast implants were associated with a significant increase in the summary adjusted relative risk of individual connective-tissue diseases (rheumatoid arthritis, 1.04 [95 percent confidence interval, 0.72 to 1.51]; systemic lupus erythematosus, 0.65 [95 percent confidence interval, 0.35 to 1.23]; scleroderma or systemic sclerosis, 1.01 [95 percent confidence interval, 0.59 to 1.73]; and Sjögren's syndrome, 1.42 [95 percent confidence interval, 0.65 to 3.11]); all definite connective-tissue diseases combined (0.80; 95 percent confidence interval, 0.62 to 1.04); or other autoimmune or rheumatic conditions (0.96; 95 percent confidence interval, 0.74 to 1.25). Nor was there evidence of significantly increased risk in the unadjusted analyses or in the analysis restricted to silicone-gel–filled implants.
Conclusions On the basis of our meta-analyses, there was no evidence of an association between breast implants in general, or silicone-gel–filled breast implants specifically, and any of the individual connective-tissue diseases, all definite connective-tissue diseases combined, or other autoimmune or rheumatic conditions. From a public health perspective, breast implants appear to have a minimal effect on the number of women in whom connective-tissue diseases develop, and the elimination of implants would not be likely to reduce the incidence of connective-tissue diseases.
We conducted a comprehensive series of meta-analyses of the largest group of studies to date to investigate the possible relation between silicone breast implants and the risk of autoimmune conditions or connective-tissue diseases. Our study incorporated the eight studies not included in the earlier meta-analyses and had four principal objectives: to investigate the relation between breast implants and connective-tissue diseases by incorporating all eligible studies into an unadjusted analysis; to consider the effect of potential confounding factors in an adjusted analysis; to search for sources of heterogeneity among the studies with formal statistical tests and influence analyses; and to perform a separate analysis focused exclusively on silicone-gel–filled breast implants.18 In addition, we evaluated the public health effect of silicone breast implants by estimating the annual number of new cases of connective-tissue disease that can be attributed to the presence of breast implants.
Methods
Selection of Studies
We obtained the results of studies cited in other meta-analyses and reviews,7,8,9,19 and we conducted a search of the literature that was similar to that outlined by Perkins et al.7 Sources included Medline (National Library of Medicine, Bethesda, Md.) from January 1966 through May 1998; Toxline (National Library of Medicine, Bethesda, Md.) from January 1985 through May 1998; Current Contents Search (Institute for Scientific Information, Philadelphia) from July 1997 through May 1998; and Dissertation Abstracts Online (University Microfilms International, Ann Arbor, Mich.) from January 1992 through May 1998. In searches of Medline, Toxline, and Current Contents Search, we used a combination of key words for breast implants and connective-tissue diseases. Key words for studies of breast implants included "breast implant," "breast augmentation," "breast reconstruction," "mammoplasty," and "mammaplasty," with all possible suffixes allowed (e.g., "implantation" and "implants"). Key words for connective-tissue diseases included "rheumatic diseases," "connective tissue disease," "autoimmune disease," "systemic sclerosis," "scleroderma," "lupus," "dermatomyositis," "sarcoidosis," "rheumatoid arthritis," "fibromyalgia," "Sjögren," and "polymyositis." A search of Dissertation Abstracts Online was conducted with use of a combination of key words for "breast implant" and "connective tissue disease." All searches were limited to studies of human subjects and reports published in English; they produced 757 citations. We were unable to obtain one abstract20 that had appeared in 1993 and was cited in two publications.8,19
All the potentially relevant papers were reviewed independently by the investigators. The criteria for inclusion in the meta-analyses were the presence of an internal comparison group and the availability of numbers for the construction of two-by-two tables to establish categories of disease and implants. In cases in which there was more than one published report on the same population or group of patients, the most recent article was selected for analysis. Studies reporting only information on symptoms and the frequency with which individual symptoms appeared were excluded, since individual women, not individual symptoms, were the units of analysis.
Abstraction of Data
All the data were independently abstracted by two investigators with the use of standardized data-abstraction forms. Disagreements were resolved by discussion. The following information was sought from each paper, although some papers did not contain all the information: first author's name, year of publication, geographic location of the study, source of funding for the study, type of study design (cohort, case–control, or cross-sectional), study population, sample size, source of subjects (private practice, tertiary care center, or defined population), type of implant, date of implantation, reason for implantation (cosmetic or reconstructive), disease diagnosis, case definition, date of diagnosis, method of data collection (self-report or medical-record abstraction), average time to onset of symptoms after implantation, control for confounding factors by matching or adjustment, and relative risks or odds ratios and 95 percent confidence intervals for individual connective-tissue diseases and all connective-tissue diseases combined associated with all types of breast implants and with silicone-gel–filled breast implants alone, if analyzed separately.
Diseases Studied
The following disease entities were included in the analyses: rheumatoid arthritis; systemic lupus erythematosus; scleroderma or systemic sclerosis; Sjögren's syndrome; dermatomyositis or polymyositis; all definite connective-tissue diseases combined, as defined in each study; and a category of other autoimmune or rheumatic conditions. The category of other autoimmune or rheumatic conditions included conditions, such as undifferentiated connective-tissue disease or mixed connective-tissue disease, that did not fulfill the diagnostic criteria of the classic autoimmune diseases or connective-tissue diseases; this category also included signs and symptoms of autoimmune or rheumatic conditions, such as joint pain, swelling, or both, as determined by the authors of each study.
Statistical Analysis
The disease variables were as follows: the presence or absence of any of the five individual connective-tissue diseases, the presence or absence of all definite connective-tissue diseases combined, and the presence or absence of other autoimmune or rheumatic conditions. The exposure variable was the presence or absence of any type of breast implant. Women who had had direct injections of any material into the breast, including silicone, were excluded from the analysis. A separate analysis was conducted for implants described in the individual studies as silicone-gel–filled breast implants.
We used fixed-effects models, as described by Greenland,21 as opposed to random-effects models, in our meta-analyses.
Unadjusted Analyses
The basic data used in the unadjusted analyses consisted of a series of two-by-two tables defined by the dichotomous exposure and disease variables for each study. Because the numbers in some cells of the two-by-two tables were small, exact analyses and conditional maximum-likelihood methods were used.22 Separate analyses of the associations in two-by-two tables were combined to produce summary estimates of the odds ratio with exact confidence limits.22 Summary estimates of the odds ratio and associated tests for homogeneity were calculated for all connective-tissue diseases combined, for specific diseases, and for other autoimmune or rheumatic conditions. We used stratified analyses involving three dichotomous variables (cohort design vs. other study design, year of diagnosis before 1992 vs. year of diagnosis 1992 or later, and validation of disease through medical records vs. no validation of disease through medical records), as well as influence analysis, to search for sources of heterogeneity. Summary odds ratios were calculated with the use of Exact statistical software.22
Adjusted Analyses
Only studies that provided an adjusted estimate, either through the use of appropriate methods of analysis or through matching of variables in the study design, were considered in this analysis. The data needed from each study were the estimated adjusted effect (either the adjusted relative risk or the adjusted odds ratio, the latter being a good approximation of the adjusted relative risk in the case of rare diseases) and its estimated standard error (often obtained indirectly from the confidence interval reported in the study). First, we decided whether the adjusted relative risks from each study were estimating the same underlying association between exposure and disease. We used a chi-square test for homogeneity to help us make this decision.21 If the test for homogeneity was not rejected at a P value 
0.10, we computed an estimated summary adjusted relative risk involving an inverse-variance–based weighted average of the individual natural logarithms of the values for adjusted relative risk.21 Larger studies producing estimated adjusted effects with smaller standard errors were weighted more heavily in the summary adjusted relative risks than smaller studies with correspondingly larger standard errors. Using the same methods of analysis, we produced an additional meta-analysis of silicone-gel–filled implants only. SAS statistical software (version 6.12, SAS Institute, Cary, N.C.) was used to calculate the estimates of the summary adjusted relative risks.
Results
We included nine cohort studies,10,11,15,16,23,24,25,26,27 nine case–control studies,13,14,17,28,29,30,31,32,33 and two cross-sectional studies12,34 in our meta-analyses. The majority of the studies were conducted in the United States, but some were carried out in Canada,10 Australia,30 the United Kingdom,16 and northern Europe.11,15,24 The characteristics of the cohort studies are presented in Table 1, and those of the case–control and cross-sectional studies in Table 2. The cohort studies, the two cross-sectional studies, and one case–control study17 evaluated multiple disease outcomes. The diagnosis of definite connective-tissue disease was obtained from medical records in all studies except that of Hennekens et al.,12 which used self-reported data. The diagnosis of other autoimmune or rheumatic conditions was obtained from medical records for all studies except four, which used self-reported data: Giltay et al.,24 Hennekens et al.,12 Sánchez-Guerrero et al.,25 and Wells et al.27 The 95 percent confidence intervals included 1 for all diseases in all studies, except that by Goldman et al.34 (rheumatoid arthritis and connective-tissue diseases: adjusted relative risk, 0.52; 95 percent confidence interval, 0.29 to 0.92) and Hennekens et al.12 (any connective-tissue disease: adjusted relative risk, 1.24; 95 percent confidence interval, 1.08 to 1.41; other connective-tissue diseases: adjusted relative risk, 1.30; 95 percent confidence interval, 1.05 to 1.62) (Table 2).
|
|
|
|
|
|
We used several techniques of meta-analysis to evaluate the existing studies of the association between breast implants and connective-tissue diseases. These included both exact methods (unadjusted analysis)22 and approximate, large-sample methods (adjusted analysis).21 There is no evidence in either the analysis of unadjusted odds ratios or the analysis of adjusted relative risks, excluding the results of the study by Hennekens et al.,12 of a significantly increased risk of any specific connective-tissue disease, all definite connective-tissue diseases combined, or other autoimmune or rheumatic conditions. The estimated summary relative risks for scleroderma or systemic sclerosis, rheumatoid arthritis, and systemic lupus erythematosus are close to or less than 1. Although our estimate of the summary adjusted relative risk for Sjögren's syndrome (1.42; 95 percent confidence interval, 0.65 to 3.11) was elevated, a diagnosis of Sjögren's syndrome requires salivary-gland biopsy.35 Whether biopsies were actually performed in the studies cited is unknown; and there may therefore have been bias in the size of the estimated summary adjusted relative risk due to misclassification of disease.
Our analyses showed that the summary adjusted relative risks that included the study by Hennekens et al.12 were higher than the pooled results of the other studies for all definite connective-tissue diseases combined and for other autoimmune or rheumatic conditions. The study by Hennekens et al. was subject to various methodologic problems, including the lack of validation of disease diagnosis by review of the medical records. Self-reports of connective-tissue disease are inaccurate; in one study of self-reported rheumatoid arthritis, the positive predictive value was only 20 percent.36 Furthermore, the intensive publicity about the postulated adverse health effects of breast implants is likely to have made women with implants more aware of their symptoms and to have resulted in overreporting of disease among women with implants as compared with women without implants, thus potentially biasing the estimated effects upward. These factors suggest that the summary adjusted relative risks that included the study by Hennekens et al. were probably overestimates.
In general, the summary adjusted relative risks should be more valid than the summary unadjusted odds ratios. On the other hand, some of the studies with low estimates of relative risk were selectively excluded from the adjusted analysis because they had no cases of connective-tissue disease among women with breast implants. The summary adjusted relative risks calculated on the basis of the remaining studies could be higher as a result of bias.
Information on potential confounders of the association between breast implants and connective-tissue diseases was incomplete in many studies.37 However, when potential confounding factors other than age, race, and year of study were evaluated, they had little effect on the adjusted relative risks reported in individual studies. Most of the specific connective-tissue diseases do not have strong established risk factors other than sex, age, and race.38 Specific genetic markers of susceptibility are recognized for some of the connective-tissue diseases,38,39,40,41 but no information on the basis of which to evaluate them was available in the epidemiologic studies. In addition, many studies did not report whether the indication for implantation was cosmetic or reconstructive, a difference that may have affected the signs and symptoms the subjects subsequently had. The two studies that included sufficient data to analyze the effect of long latency (10 or more years after implantation) on the incidence of connective-tissue disease suggested that the time since implantation was not predictive of the risk of connective-tissue disease.12,25 Individual studies did not provide adequate data on rupture or leakage of implants for us to include these features as possible correlates of the incidence of connective-tissue disease.
Publication bias is frequently cited as a reason for lack of validity in meta-analyses.42 Publication bias could occur if studies that found no association between exposure and disease were less likely to be submitted and accepted for publication than were studies that found a positive association. In fact, the results of the majority of the studies included in our meta-analyses were negative, as stated by the authors. Nonetheless, we examined the potential for publication bias by constructing a funnel plot in which the inverse of the estimated variance of the natural logarithm of the adjusted relative risk was plotted against the natural logarithm of the adjusted relative risk for each disease.42 Funnel plots of our data showed no evidence of publication bias for any of the disease entities we studied (data not shown).
Given the results of the individual studies and our summary adjusted relative risks, we do not have convincing evidence that the underlying summary relative risk of connective-tissue disease in the population of women with breast implants exceeds 1. On the basis of calculations in which the standard errors from the studies under consideration were used, our study had approximately 80 to 90 percent power to detect true summary relative risks of 1.5 to 2.0, when all studies were included.18 However, when the study by Hennekens et al.12 was excluded, the power of the study to detect a summary relative risk 2.0 was roughly 70 percent or less for scleroderma or systemic sclerosis, systemic lupus erythematosus, and Sjögren's syndrome.18
We calculated the population attributable risk, the proportion of cases of connective-tissue disease in a population that may be caused by breast implants, using standard formulas.43,44 To estimate the number of cases of connective-tissue disease attributable to implants, we used the summary adjusted relative risks obtained from our meta-analyses that included the study by Hennekens et al.,12 assumed the proportion of women with breast implants in the United States to be 1 percent, and multiplied the annual incidence of disease (averages obtained from reports in the literature17,25,40,45,46,47,48,49,50,51,52,53,54,55,56,57) by the population attributable fraction. Although we chose high estimates for the proportion of women with breast implants and the summary adjusted relative risks so as to maximize the possible public health effect of breast implants in our calculations, the estimated annual number of new cases of connective-tissue disease that could be attributed to breast implants was small. Among 10 million women in the United States, 4.3 of 3303 new cases of rheumatoid arthritis, approximately 0.1 of 526 new cases of systemic lupus erythematosus, 0.4 of 164 new cases of scleroderma or systemic sclerosis, 1.3 of 400 new cases of Sjögren's syndrome, and 0.2 of 54 new cases of dermatomyositis or polymyositis may be attributed to breast implants each year.
Despite the differences in the meta-analyses conducted thus far, none, including the meta-analyses reported here, have identified a significant association between breast implants and connective-tissue diseases. On the basis of the research to date, no association is evident between breast implants and any of the individual connective-tissue diseases, all connective-tissue diseases combined, or the other autoimmune or rheumatic conditions, with the possible exception of Sjögren's syndrome. The uncertainty of the diagnosis of Sjögren's syndrome makes the interpretation of the estimated summary adjusted relative risk questionable, however. The meta-analysis focusing solely on silicone-gel–filled implants produced lower summary estimates of the adjusted relative risks for all the diseases than did the analyses based on all types of breast implant. From a public health perspective, breast implants appear to have a minimal effect on the number of women in whom connective-tissue diseases develop, and elimination of implants would be unlikely to reduce the incidence of connective-tissue diseases.
Supported by a contract (USCA 79004) with the Administrative Office of the U.S. Courts.
Taken in part from a report by a National Science Panel appointed by Judge Sam C. Pointer, Jr., for the federal judiciary.
We are indebted to Drs. Betty Diamond, Nancy Kerkvliet, and Peter Tugwell for their thoughtful contributions to previous versions of the manuscript; and to Lynn Igoe for her editorial assistance.
Source Information
From the Department of Epidemiology (E.C.J., B.S.H.), and the Department of Biostatistics (L.L.K.), School of Public Health, University of North Carolina at Chapel Hill, Chapel Hill, N.C.
Address reprint requests to Dr. Janowsky at the Department of Epidemiology, CB 7400, School of Public Health, University of North Carolina at Chapel Hill, Chapel Hill, NC 27599, or at esther_ janowsky{at}unc.edu.
References
| |||||||||||||||||||||||||||||||||
This article has been cited by other articles:
HOME | SUBSCRIBE | SEARCH | CURRENT ISSUE | PAST ISSUES | COLLECTIONS | PRIVACY | TERMS OF USE | HELP | beta.nejm.org Comments and questions? Please contact us. The New England Journal of Medicine is owned, published, and copyrighted © 2009 Massachusetts Medical Society. All rights reserved. |
Previous
