We found that increasing exposure to any TCAs during years 11–15 was associated with a trend towards an increasing RR for breast cancer (). Post hoc analyses based on classifying the drugs according to their genotoxicity in Drosophila suggest that the use of genotoxic TCAs was responsible for the increased risk (amoxapine, clomipramine, desipramine, doxepin, imipramine, trimipramine), while the use of the nongenotoxic TCAs was not associated with an increased risk (amitriptyline, maprotiline, nortriptyline, protriptyline).
Because our accrual of cases from the source population was nearly complete (Parkin et al, 1997
) and our exposure data were collected routinely before the index date, there was little potential for selection and recall bias concerning exposure.
Since our measures of exposure were based on outpatient prescriptions, actual consumption of TCAs probably differed from our estimates. It is unlikely that all drugs dispensed were ingested. If we overestimated exposure, then the slope of the dose-risk relationship that we observed would be less than the true slope (MacMahon and Trichopoulos, 1996
). Although we may have underestimated exposure, since we had no information about some of the TCAs dispensed (see Subjects and Methods), these amounts were probably small relative to the amounts used in calculating exposure.
Could our results be confounded by the effects of other determinants of breast cancer associated with TCA use?
Although depression was once considered to be associated with the development of cancer as the result of immunologic and endocrine dysfunction, there is now little supporting evidence (Linkins and Comstock, 1990
; Spiegel, 1996
). Although Penninx et al (1998)
recently found that the incidence of cancer was 1.9 times greater (95% CI: 1.1–3.1) among patients with chronic depression, they obtained information about antidepressant use for only the 2 weeks period preceding inception of the cohort. Although they considered the possibility that antidepressant drug use may have increased the risk of cancer and controlled for it, it is unlikely that they controlled adequately for antidepressant drug use with such limited information. Furthermore, they did not consider that some antidepressants might be more hazardous than others. Similarly, Gallo et al (2000)
obtained information on depression status at baseline and cancer at follow-up 13 years later for 2017 persons. They found no overall association between depression and the development of cancer. Among women, however, there was an association between major depression and breast cancer. The RR adjusted for age, tobacco use, and alcohol abuse was 3.8 (95% CI: 1.0–14.3); this estimate was based on 25 breast cancer cases. They reported that adjustment ‘for use of psychotropic medicines and use of health-care services at baseline’ did not change this estimate substantially. No further details of this adjustment were provided, nor were they able to study the interval between major depression and the diagnosis of cancer.
Nevertheless, if other determinants of breast cancer (Sharpe and Boivin, 2000
) were associated with TCA use, it is possible that our results might be attributable to confounding by such determinants. It is unlikely, however, that strong positive confounding associated with the use of the genotoxic TCAs led to an apparent effect stronger than the overall effect of TCA exposure (compare the right panels of and to the right panel of ), while strong negative confounding associated with the use of the nongenotoxic TCAs led to an apparent lack of effect. TCAs are usually prescribed for depression without consideration of subtle details of their chemical structures and without consideration of breast cancer as an adverse effect. Exposure to the genotoxic TCAs may have increased the risk of breast cancer, while exposure to the nongenotoxic TCAs may have had no effect on risk.
Since selection bias, recall bias, exposure misclassification, and chance are unlikely explanations for our findings, and differential patterns of confounding according to the chemical structures of the drugs seem unlikely, a biologic explanation may be required. The >10 year delay between exposure to the genotoxic TCAs and the increase in the RR for breast cancer is suggestive of tumour initiation rather than tumour promotion. Ionizing radiation, the prototypical initiator, increases the risk of breast cancer after a delay of 10 years (Land, 1987
The results of TCA genotoxicity assays involving bacteria, Drosophila
, and human lymphocytes are conflicting (van Schaik and Graf, 1991
). The consistency between the results of the assays carried out by van Schaik and Graf (1991
) and Graf (2001) in Drosophila
and our epidemiologic findings suggests that it may be the most appropriate.
Although several epidemiologic studies found no positive associations between TCA use and the development of cancer (Danielson et al, 1982
; Friedman and Ury, 1980
; Kelly et al, 1999
; Selby et al, 1989
; Weiss et al, 1998
), they were limited either by small sample size, self-reporting of use, failure to specify dosage, duration of use, or timing of use, and lack of control of confounding. Recently, Wang et al (2001)
273 women who filled a prescription for an antidepressant over a period of up to 24 months, and 32
949 women who filled a prescription for any other medication over the same period. Both groups were followed for a maximum of 7.5 years. TCA use was not associated with the development of breast cancer (RR=1.09; 95% CI: 0.92–1.31), which is consistent with our finding that TCA use was not associated with any increase in risk with until at least 10 years had elapsed.
Two other studies have reported positive associations between TCA use and the development of breast cancer. Wallace et al (1982)
carried out a case–control study of antidepressant use (TCAs or phenelzine) in relation to breast cancer incidence and obtained an adjusted RR=2.8 (P
<0.04) for use >1 month. In a similar study, Cotterchio et al (2000)
found that TCA use for
25 months was associated with an adjusted RR=2.1 (95% CI: 0.9–5.0). Neither study specified dosage or the timing of use.
Another two studies reported positive associations between antidepressant use and the development of other cancers. Harlow and Cramer (1995)
carried out a case–control study of ovarian cancer incidence and obtained an adjusted RR=2.1 (95% CI: 0.9–4.8) for any prior use of antidepressants or benzodiazepines lasting
6 months. Among women who first used these drugs before age 50 years the RR was 3.5 (95% CI: 1.3–9.2). Among those who used them
10 years before diagnosis, the RR was 9.7 (95% CI: 1.2–78.8). These findings, however, were not corroborated by Coogan et al (2000)
. Both studies were based on self-reported exposures. Dalton et al (2000)
conducted a population-based cohort study in Denmark using the nation's prescription database. They followed 30
807 antidepressant users aged
15 years for up to 7 years (mean=3.2 years) and found an increased risk of non-Hodgkin's lymphoma among subjects who received
5 prescriptions for TCAs (standardized incidence ratio=2.5; 95% CI: 1.4–4.2).