Cotterchio and Others, 2000 (6)
The Cotterchio paper also describes a population-based case–control study examining the relation between antidepressant use and breast cancer. Cases were selected randomly from women listed in the Ontario Cancer Registry. Subjects were aged 25 to 74 years and had a diagnosis of primary breast cancer during 1995 and 1996. Control subjects were a random sample of women residing in Ontario who were not registrants with the Ontario Cancer Registry. They were matched in 5-year age groups with the women who had breast cancer. Mailed, self-administered surveys were used to obtain data regarding exposure to antidepressants, sociodemographics, and potential confounding characteristics. There were 701 subjects with breast cancer and 702 control subjects. The survey response rate was 83% for breast cancer patients and 67% for control subjects.
The analysis generated ORs for breast cancer risk, not only with respect to classes of antidepressants but also for specific individual agents. Potential confounders were included and adjusted for in the analysis of adjusted ORs. The investigators found no statistically significant association between breast cancer risk and any class of antidepressants. With the category “longest duration of use” (³ 25 months), there was an age-adjusted association between breast cancer risk and TCA use (OR = 2.5; 95%CI, 1.2 to 5.1). This was not significant in the multivariate analysis (OR = 2.1; 95%CI, 0.9 to 5.0). The “finding” highlighted in the Globe and Mail article, however, was based upon analyses related to individual antidepressant drugs. A striking feature of these analyses is not discussed in the newspaper article—the small sample sizes upon which they were based. Most notably, the newspaper highlighted a “significant” result related to paroxetine (Paxil): age-adjusted OR of 9.1 (95%CI, 1.2 to 72.5), but multivariate OR of 7.2 (95%CI, 0.9 to 58.3). This finding was based upon 9 cases and 1 control subject and was not, in fact, statistically significant. Thus, even the single finding highlighted in the article was not significant after adjustments were made for baseline differences between the groups.
Given our discussion of the Sharpe and others paper, the merits and shortcomings of Cotterchio’s research are clear. The measurement of potential confounders (and their inclusion in the analyses) is a relative advantage of the study. However, in this paper the probability of recall error and recall bias is high. Responding to similar qualms raised by others in the postpublication correspondence, the authors contended that the questionnaire design would not have revealed the study’s hypotheses to respondents (6). This argument is not reassuring. Even without transparent hypotheses, subjects may go to greater lengths to recall exposure to any or all factors they think might be related to breast cancer. In this case, significant recall bias would prejudice the results toward finding an association between breast cancer and antidepressant exposure. Moreover, problems with recall error may dominate concerns about recall bias, particularly because respondents were asked to recall details of drug exposures that occurred several years earlier.
A greater methodological concern exists in regard to the research paper and how it was represented in the Globe and Mail article. This concern relates again to multiple statistical comparisons and how the results of these tests were described. In our opinion, omitting methods to adjust for multiple comparisons must be seen as a serious flaw in the research. Further, we believe that the decision to publish ORs based upon 9 patients and a single control subject—quite apart from implying that the resulting ORs are significant—must be seriously questioned.
Recommendations
Postmarket surveillance studies have a vital role in informing practitioners and patients of the potential risks associated with medications—risks that may not be detected during drug licensure trials or even in typical postmarketing surveillance for adverse events. These studies are particularly important for assessing the risks for diseases such as breast cancer, since there may be an interval of many years between exposure and disease manifestations. Postmarket surveillance studies may also be an important source of information that moderates the often-relentless barrage of promotional advertizing on behalf of brand-name pharmaceuticals.
In our opinion, however, the important role of surveillance studies does not exempt them from the demand for sound scientific design, analysis, and reporting, nor does it exempt investigators from presenting their results in the mass media in a careful and considered manner. Indeed, there are well-known past examples in which methodological inadequacies in case–control studies led to erroneous calls for changes in clinical practice (10–12).
As reviewed above, the results from the Cotterchio and others study—including the publicized association of breast cancer and paroxetine exposure—do not reveal any association between antidepressants and the risk of breast cancer. As such, we feel that there is no indication to factor a risk of breast cancer into decisions about treating depression with paroxetine.
The study by Sharpe and others makes a more provocative case for the increased risk of breast cancer associated with TCAs. These researchers found an increased risk of breast cancer after a period of time. Further, a post hoc hypothesis of TCA genotoxicity yielded a significant result. This study was well-designed, the major criticism being the absence of adjustment for potential confounders. The authors of this study concluded that further studies, specifically testing the TCA genotoxicity hypothesis and controlling for potential confounding, should be conducted (1). We strongly encourage the retesting of this hypothesis. The association would gain credibility if it persisted in a retrospective cohort study format that used similar types of linked administrative databases. In a cohort study, study subjects and control subjects are determined by TCA exposure, rather than by breast cancer diagnosis, as in case–control studies. The subjects and control subjects (who have not been exposed to TCAs) are then followed over time to measure the development of breast cancer, while the study controls for potential confounders.
We feel that, because of the methodological concerns raised above, there is insufficient evidence in the 2 studies examined to guide practitioners to change clinical practice. In the absence of case–control studies that adequately control for potential confounders, we are hard-pressed to suggest changes to clinical practice based upon their findings.
Note
1. The authors of this paper presented the results of their multivariate analyses in the form of relative risk ratios (RRs). Although it is well-known that case–control designs cannot be used to calculate RRs, for clarity we have retained the authors’ term. Ratios of risks from case control studies are usually expressed in odds ratios (ORs).
References
1. Sharpe CR, Collet J-P, Belzile E, Hanley JA, Boivin J-F. The effects of tricyclic antidepressants on breast cancer risk. Br J Cancer 2002;86:92–7.
2. Picard A. Research finds some depression drugs raise cancer risk. The Globe and Mail 2002;A1.
3. Parkin DM, Pisani P, Ferlay J. Esitmates of the worldwide incidence of 25 major cancers in 1990. Int J Cancer 1999;80:827–41.
4. Brewster A, Helzlsouer K. Breast cancer epidemiology, prevention, and early detection. Curr Opin Oncol 2001;13:420–5.
5. Madigan MP, Zeigler RG, Benichou J, and others. Proportion of breast cancer cases in the United States explained by well- established risk factors. J Natl Cancer Inst 1995;87:1681–5.
6. Cotterchio M, Kreiger N, Darlington G, Steingart A. Antidepressant medication use and breast cancer risk. Am J Epidemiol 2000;151:951–7.
7. Van Schaik N, Graf U. Genotoxicity evaluation of five antidepressants in the wing somatic mutation and recombination test in Drosophila melanogaster. Mutat Res 1991;260:99–104.
8. Van Schaik N, Graf U. Structure-activity relationships of tricyclic antidepressants and related compounds in the wing somatic mutation and recombination test of Drosophila melanogaster. Mutat Res 1993;286:163.
9. Fisher LD, Van Belle G. Multiple Comparisons. In: Fisher LD, Van Belle G. Biostatistics: a methodology for the health sciences. New York: John Wiley and Sons; 1993. p 596–629.
10. Armstrong B, Stevens N, Doll R. Retrospective study of the association between use of rauwolfia derivatives and breast cancer in English women. Lancet 1974;2:672–5.
11. Reserpine and breast cancer. Lancet 1974;2:669–71.
12. Heinonen OP, Shapiro S, Tuominen L, Turunen MI. Reserpine use in relation to breast cancer. Lancet 1974;2:675–7.
Author(s)
Manuscript received August 2002, revised, and accepted October 2002.
1. Resident, Department of Psychiatry, University of Toronto, Toronto, Ontario.
2. Lecturer, Department of Psychiatry, University of Toronto, Toronto, Ontario.
3. Professor, Department of Psychiatry, University of Toronto; Director, Kunin-Lunenfeld Applied Research Unit, Baycrest Centre for Geriatric Care, Toronto, Ontario.
Address for correspondence: Dr P Kurdyak, Health Systems Research and Consulting Unit, Centre for Addiction and Mental Health, ARF Division, 33 Russell Street, Toronto, ON M5S 2S1
e-mail: paul_kurdyak@camh.net
1 | 2 | 3
|
