Abstract

Background: Concerns have been raised that tamoxifen may be associated with depression. To investigate this question, we examined the psychological effects of tamoxifen treatment for breast cancer prevention on women at different levels of risk for clinical depression who were enrolled in the National Surgical Adjuvant Breast and Bowel Project's Breast Cancer Prevention (P-1) Study. Methods: A total of 11 064 women were randomly assigned to receive for 5 years daily doses of 20 mg of tamoxifen or placebo in the P-1 study, a multicenter, double-blind, placebo-controlled chemoprevention trial. Each woman was prospectively assessed for depression risk on the basis of medical history items collected at the baseline examination and placed in a high-, medium-, or low-risk group. Every 6 months, for a total of 36 months, the participants were assessed for depressive symptoms by completing the Center for Epidemiological Studies—Depression (CES-D) questionnaire. Scores of 16 or higher were indicative of an episode of affective distress. Differences between the risk groups and treatment arms were analyzed by logistic regression. All statistical tests were two-sided. Results: Women in the higher risk depression groups were more likely to score 16 or higher on the CES-D (percent follow-up examinations with a score of ≥16: high-risk group = 35.7%, with 95% confidence interval [CI] = 32.5% to 38.9%; medium-risk group = 19.2%, with 95% CI = 18.1% to 20.3%; and low-risk group = 8.7%, with 95% CI = 8.3 to 9.1%) and to have these scores more frequently and for longer periods than women in the lower risk groups. Within each depression risk group, there was no difference in the proportion of women scoring 16 or higher by treatment assignment (tamoxifen versus placebo) (odds ratio = 0.98; 95% CI = 0.93 to 1.02). A post-hoc analysis indicated that the lack of a tamoxifen effect was not a result of differential missing data. Conclusions: Physicians need not be overly concerned that treatment with tamoxifen will increase the risk for or exacerbate existing depression in women. Nevertheless, physicians should continue to screen for and treat or refer potential cases of depression encountered in routine clinical practice.

Concern regarding an association between clinical depression and tamoxifen, when used as an adjuvant treatment or preventative agent for breast cancer, has been voiced by a number of investigators (1–5) and continues to be discussed in regulatory agencies, such as the U.S. Food and Drug Administration. Furthermore, the Physician's Desk Reference (6) lists “depression” as an infrequent adverse reaction to tamoxifen. Although previous studies (1–5) used breast cancer patients to address tamoxifen use and depression, the studies had a number of weaknesses, including the lack of a clear definition of depression and a failure to control for the potential confounding effects of illness diagnosis, the side effects of chemotherapy (e.g., premature menopause), or normal aging. Previously, two double-blind, placebo-controlled studies of the effects of tamoxifen in postmenopausal women (7,8) found no association of tamoxifen with depression. We believe that some of the concern over the relationship between tamoxifen and depression arises from the idea that, because hormone replacement therapy has positive effects on mood and tamoxifen has antiestrogenic activity (9–11), tamoxifen, therefore, has negative effects on mood.

The completion of the Breast Cancer Prevention (P-1) Study of the National Surgical Adjuvant Breast and Bowel Project (NSABP) provides an opportunity to investigate the association between tamoxifen and depression in greater detail. The P-1 study was a multicenter, double-blind, placebo-controlled chemoprevention trial. The primary objective of the study was to evaluate whether 5 years of tamoxifen therapy would reduce the incidence of invasive breast cancer in women at an increased risk for the disease. The secondary objectives of the study included the assessment of the incidence of ischemic heart disease, bone fractures, and other negative health events, such as depression, that might be associated with tamoxifen therapy. Eligible participants were randomly assigned to receive 20 mg daily of tamoxifen or a placebo for 5 years. Detailed reports on the rationale, planning, design, and clinical outcome of the P-1 study are available elsewhere (12–16).

In our initial publication on the health-related quality of life (HRQL) (16) of all subjects in the P-1 study, we did not find a difference between the treatment groups (tamoxifen versus placebo) on the Center for Epidemiological Studies—Depression (CES-D) Scale (17) or the SF-36 Mental Health Scale (18). It is known, however, that vulnerability to clinically identifiable forms of depression is not uniformly distributed in the general female population but, instead, clusters in high-risk groups of women (19). This vulnerability to depression may be inherited, suggesting a genetic or familial origin, or it may be related to certain psychological predispositions, such as a low self-esteem, a poor resistance to stress, or a pessimistic view of the world. We were concerned that the potential negative effects of tamoxifen for women at high risk for depression may have been masked in our previous analysis (16) because of the simultaneous inclusion of a larger group of less vulnerable (i.e., low-risk) participants.

In this study, we investigated the effects of tamoxifen on women at different levels of risk for depression. Specifically, we were interested in whether tamoxifen treatment was associated with the onset or prolongs the length of existing episodes of clinically diagnosable depression in women at high risk for depression.

Patients and Methods

Participant Cohort and HRQL Data

This article covers the baseline and first 36 months of follow-up data (collected at 6-month intervals) on the same 11 064 women used in the initial HRQL report (16) from the P-1 study. The P-1 participants ranged in age from 35 years to 79 years (mean ± standard deviation = 53.8 ± 9.2 years), were predominantly white (95.6%), were well educated (≥ some college = 64.9%), and were currently employed (full- or part-time = 64.7%) in a professional or technical field (67.9%). A detailed description of this cohort of participants and the P-1 HRQL instruments was reported previously (14,16). All investigations conducted in the P-1 study were approved by review boards at each institution and were in accord with an assurance filed with and approved by the U.S. Department of Health and Human Services (12). All of the participants provided written informed consent.

Defining Depression

Depressive disorders, as defined by the current psychiatric nomenclature in the Diagnostic and Statistical Manual of Mental Disorders: DSM- IV (DSM-IV) (20), come in a variety of forms that differ on the basis of the number, severity, and persistence of symptoms. The majority of clinically diagnosable episodes of depression involve one of three disorders—major depression, dysthymia, or bipolar disorder (19). Major depression involves an illness episode lasting at least 2 weeks that includes mood disturbance (dysphoria) and at least four of the following symptoms: sleep disturbance, change in psychomotor activity, loss of ability to experience pleasure and interest, fatigue, feelings of worthlessness or guilt, difficulty in concentrating, and a preoccupation with death or a wish to die. These symptoms must be associated with a clear impairment in social functioning. Dysthymic disorder or dysthymia is a chronic illness lasting at least 2 years. Dysthymia does not show the same levels of social impairment found in major depression, but it does involve mood disturbance (dysphoria) and a loss of the ability to experience pleasure and interest in usual activities, together with some of the other symptoms used to define major depression. Individuals diagnosed with dysthymia often experience episodes of major depression during their lifetime. DSM-IV distinguishes bipolar disorders from depressive disorders. Bipolar disorders have dramatic clinical manifestations that involve one or more episodes of hypomania during an individual's lifetime alternating with illness episodes that fit the criteria for major depression disorder.

Depression was previously defined by the Research Diagnostic Criteria (RDC) (21), a nonclinical forerunner of the current DSM-IV criteria. The RDC used similar criteria as the DSM-IV to define “major depression” but, unlike the DSM-IV, also included criteria to define “minor depression” (nonpsychotic episodes of illness characterized by a prominent and sustained dysphoria but lacking all of the symptomatic features of major depression). Although important historically, the RDC has been superseded by the DSM-IV.

One of the problems associated with the definition of depression is that, in addition to these diagnosable clinical entities, there are multiple sources of affective distress that may result in short-term or self-limiting expressions of depressive symptoms without meeting the DSM-IV criteria outlined above. The best available data on rates of clinically diagnosable depressive disorders in the U.S. general population come from the National Institute of Mental Health's Epidemiological Catchment Area (ECA) study (19). ECA study investigators found that, even though clinically diagnosable depressive disorders are relatively rare, usually affecting only 5%–6% of the general female population during any 12-month period, the reporting of depressive symptoms is reasonably frequent, with 35.7% of the women in the ECA study (19) reporting having experienced a period of dysphoria (feeling sad or blue) lasting at least 2 weeks. These expressions of affective distress, which fail to meet the clinical criteria for major depression, dysthymia, or bipolar illness, are often associated with occurrences such as uncomplicated grief, medical illness and other life events, or chronic difficulties (22). Depressive symptoms may also occur secondary to other psychiatric illnesses (i.e., anxiety disorders or phobias), chronic medical conditions, or substance abuse.

Monitoring Depressive Symptoms in the P-1 Study

The primary instrument used to monitor depressive symptoms in the P-1 study was the CES-D (17). This self-administered questionnaire was designed to be a brief, first-stage screen rather than a clinical diagnostic instrument. The CES-D is composed of 20 items, each of which is scored on a scale of 0–3. Higher scores reflect increased expression of affective distress, and a score of 16 or higher is most often used as the cutoff point for likely cases of clinical depression (17,23,24).

Two problems are associated with the use of the CES-D alone to screen for clinically diagnosable episodes of depression. First, questions on the CES-D inquire only about the past 7 days, collecting little information on the length of time that a symptom has been present. Second, the CES-D collects information only on symptoms and not the degree of social impairment experienced by the respondent. Consequently, scores above the CES-D clinical cutoff point of 16 tend to include a substantial proportion of distressed individuals—perhaps upwards of one half or more—who do not meet the clinical criteria for major depression, dysthymia, or bipolar illness (24,25).

Estimating Depression Risk in P-1 Study Participants

The eligibility criteria for the P-1 study permitted, at the discretion of the local site investigator, the inclusion of women with evidence of clinical depression. Twenty to 22% of the participants scored 16 or higher on the CES-D at least once during any 12-month period of the P-1 study. This percentage exceeds the expected general population rates [5%–6% (19)] of clinically diagnosable depressive disorders over a 12-month period by 3.5–4.0 times, indicating that it is necessary to distinguish between clinically diagnosable episodes of depression and depressive symptoms that are secondary to other types of physical and psychiatric illnesses or a consequence of social conditions that produce short-term, self-limiting expressions of affective distress. The preferred means to make such a distinction would be a standardized psychiatric interview, such as the Schedule for Affective Disorders and Schizophrenia—Lifetime Version (26) or the Diagnostic Interview Schedule (19). However, in the absence of such an interview, the best single indicator of risk for a future episode of major depression, dysthymia, or bipolar disorder in the P-1 study data is a medical history of treatment for these disorders (27–30).

The ECA study (19) found that the mean age at onset for major depressive disorders in the general population was 27 years, with approximately 89% of all first depressive episodes occurring before age 35 years, which was the lower age limit of the participants in the P-1 study. Medical history information, collected on a one-time-only basis as a part of the baseline entry and eligibility assessment of all P-1 study participants, included three self-reported items regarding depression: 1) a medical history of depression, 2) current or previous prescriptions for antidepressant medications, and 3) extended periods (≥12 months) of dysphoric mood (i.e., “depressed or sad most days”). If a participant gave a positive answer to the medical history or the medication question, the interviewer obtained dates of treatment, physicians' names, specific modalities of treatment, and date of last medication dose to assess the consistency and appropriateness of the information provided.

These three medical history items were used in the current study to prospectively estimate each participant's risk of experiencing a clinically diagnosable episode of depression. A simple three-level risk score was determined for each P-1 study participant, depending on whether they endorsed 0 (low risk), 1 or 2 (medium risk), or 3 (high risk) of the medical history items regarding depression in the Entry/Eligibility Form. We hypothesized that women with higher scores on this simple depression risk scale would experience more severe and persistent episodes of affective distress and would be more likely to receive a clinical diagnosis of depression. Moreover, if tamoxifen was associated with the onset and/or prolonged the length of depressive episodes in the high-risk (i.e., more vulnerable) group, it should be apparent from longitudinal differences in the proportion of P-1 study participants in the treatment groups (tamoxifen versus placebo) who scored 16 or higher on the CES-D.

Statistical Analysis

CES-D scores were analyzed as above or below the clinical cutoff of 16 or higher. Binary logistic regression was the primary method of statistical analysis used in this study. Estimated odds ratios (ORs), confidence intervals (CIs), and P values are provided for all inferential analyses. Cox regression analysis was used to investigate the effects of treatment and depression risk on the time to the first CES-D with a score of 16 or higher, and Kaplan–Meier curves are provided for these data. When the CES-D data were handled as a continuous variable, nonparametic equivalents to a one-way analysis of variance (i.e., Kruskal–Wallis test) were used because it is unusual for CES-D scores to be normally distributed. Graphic presentations include 95% CIs on observed proportions to provide the reader with visual criteria for the magnitude of potential variation. Reported P values are all two-sided and have not been adjusted for multiple statistical comparisons. Instead, we have chosen to focus on consistent patterns of findings rather than on individual statistical tests in forming our conclusions. We also avoided the use of statistical methods for imputation of missing data points in the primary data because the data did not meet the strong assumptions that normally underlie such procedures (e.g., MCAR [i.e., Missing Completely at Random]/MAR [i.e., Missing at Random]). Analyses were carried out with the use of Minitab (Version 13; Minitab, State College, PA) and Egret (Version 1.0; Cytel Corp., Cambridge, MA).

Results

Depression Risk

To determine whether there was an association between depression and tamoxifen treatment in participants of the P-1 study, we first calculated the depression risk score from the frequency of responses to each one of the medical history items (Table 1). The three components of this score were only moderately intercorrelated. The highest correlation occurred between a history of illness and antidepressant medications (r = .564; P<.001), followed by history of illness and persistent dysphoric mood (r = .369; P<.001) and medications and dysphoric mood (r = .269; P<.001). Overall depression risk, measured by the data from this study, was not statistically significantly related to the participants' risk of breast cancer, as measured by the Gail risk model (12,31).

The construct validity of this depression risk score was evaluated, in part, with the use of the social and demographic factors associated with clinically diagnosable depressive disorders in the ECA study (19). Table 2 shows the distribution of the P-1 study participants according to the three-level depression risk scale on seven demographic variables, which approximate those associated with clinically diagnosable depression in the ECA study (19). All of these variables, except education, showed a statistically significant dose–response relationship to the depression risk scores in terms of the direction and intensity of the association.

CES-D Data

Fig. 1, a, shows the proportion of the participants in each depression risk group who scored above the clinical cutoff of 16 or higher on the CES-D Scale at baseline and at each of the follow-up examinations. A consistent, positive dose–response relationship was seen between depression risk, as determined on the basis of the medical history items, and the proportion of participants scoring 16 or higher on the CES-D Scale at each scheduled examination. For each depression risk group, Table 3 shows the mean proportion of follow-up examinations with scores of 16 or higher and the distribution of the maximum and the overall scores on CES-D examinations above the clinical cutpoint. A positive dose–response relationship was also observed between depression risk group and proportion of respondents who scored 16 or higher on sequential CES-D examinations. In the high-risk depression group, for example, 21.2% of the respondents scored 16 or higher on three or more sequential CES-D examinations, compared with 9.7% for the medium-risk group and 3.5% for the low-risk group (data not shown). These findings confirm the expectation that participants in the higher depression risk groups (high>medium>low), on average, tend to experience more persistent and severe episodes of affective distress.

We next analyzed the CES-D data from each depression risk group by treatment group (tamoxifen versus placebo) (Fig. 1, b–d; Table 4). After adjustment for examination and risk group, the results of a logistic regression found that there was a statistically nonsignificant effect for the tamoxifen group compared with the placebo group (OR = 0.98; 95% CI = 0.93 to 1.02; P = .32). These analyses indicate that treatment group is not statistically associated with the proportion of women scoring above the CES-D clinical cutoff of 16 or higher in any of the three depression risk groups. Furthermore, after adjustment for depression risk group, an analysis of variance found that there was no difference in the mean individual proportion of follow-up examinations above the clinical cutoff in each treatment arm.

The Kaplan–Meier plot in Fig. 2 shows the relationship between assigned treatment (placebo versus tamoxifen) and depression risk group (high, medium, or low) for the time from randomization until the first CES-D examination with a score exceeding the clinical cutoff of 16 or higher. The results of Cox proportional hazards regression analysis with these data were statistically significant for depression risk group (likelihood ratio statistic [LRS] P<.001; hazard ratio [HR] = 1.88; 95% CI = 1.74 to 2.05), but they were statistically nonsignificant for both treatment arm effects (LRS P = .988; HR = 1.00; 95% CI = 0.92 to 1.09) and interaction effects (LRS P = .575; HR = 1.03; 95% CI = 0.92 to 1.16). The proportional hazards assumption for this analysis was confirmed.

Missing Data

We next assessed the association between missing data and depression risk group or sequential CES-D examination (Fig. 3, a). Logistic regression analysis based on the data in Fig. 3, a, indicated that depression risk group (OR = 1.17; 95% CI = 1.13 to 1.21; P<.001) and sequential examination (OR = 1.45; 95% CI = 1.44 to 1.46; P<.001) were both statistically significantly associated with missing CES-D data. Panels b–d in Fig. 3 show the proportion of participants completing the CES-D by depression risk and treatment groups. Logistic regression analysis by depression risk, controlling for sequential examination, indicates that, compared with placebo treatment, tamoxifen treatment was associated with higher proportions of missing data in the low-risk group (OR = 1.11; 95% CI = 1.06 to 1.16; P<.001) and the medium-risk group (OR = 1.12; 95% CI = 1.04 to 1.21; P<.001) but not in the high-risk group (OR = 0.99; 95% CI = 0.84 to 1.16; P = .91). If tamoxifen-associated depression were the primary cause of these missing data, we would have predicted a positive dose–response increase in the magnitude of the ORs from the lowest to the highest depression risk group.

We noted in our previous report (15) that it was difficult to continue to collect quality-of-life data after a participant had gone off treatment. However, participants in the P-1 study were asked about their primary reason for going off treatment, and their responses were recorded on an Off Therapy Form (OTF) that included “depression” as one of 10 specific response categories.

Of the 11 064 participants in this cohort, we collected an OTF for 3539 (80.8%) of 4382 women who missed at least one CES-D examination. The presence of an OTF showed a moderate positive correlation with the total number of missing CES-D examinations (r = .62; P<.001). The women who completed an OTF accounted for 12 693 (89.7%) of 14 149 missing CES-D examinations. Only 110 (3.1%) of these 3539 women reported that depression was the primary reason for their going off therapy. The most frequent reasons for going off therapy were nonmedical in nature (1667 women [47.1%]), perceived toxic effects (921 women [26.0%]), and various protocol and nonprotocol medical conditions (841 women [23.8%]).

Table 5 shows the distribution of women who reported that depression was their primary reason for going off treatment by treatment group and depression risk group. An analysis of these data using binary logistic regression found a statistically significant effect for depression risk group (OR = 2.37; 95% CI = 1.83 to 3.07; P<.001) and a statistically nonsignificant effect for treatment group (OR = 1.10; 95% CI = 0.75 to 1.62; P = .63), indicating that the cases of depression that lead women to quit their assigned treatment did not occur with a greater frequency in those in the tamoxifen arm.

Discussion

Tamoxifen is the most widely prescribed anticancer agent currently in use. It has been proven to be effective against breast cancer as an adjuvant treatment and in a preventative setting (12,32). Given the widespread use of tamoxifen, it is important to fully investigate all of the potential side effects that may be associated with its administration, so that women, together with their physicians, can make an informed decision regarding its potential costs and benefits and its appropriateness for their individual situations.

This study is an extension of our earlier report (16) on the HRQL data from the NSABP P-1 study. Previously, we found no evidence for an association between tamoxifen treatment and depression in the overall P-1 study cohort. In this study, we recognized that vulnerability to clinically identifiable depressive disorders is not randomly distributed in the general female population and that the effects of tamoxifen on susceptible women in the P-1 study may have previously gone undetected.

Our initial problem was the a priori identification of subgroups of women with a potential clinical susceptibility for depression. Because the self-administered depression-screening form (CES-D) used in the P-1 study provides information on short-term symptoms of affective distress and is not intended for use as a diagnostic instrument (17), we incorporated the participants' self-reported medical history of depression, use of prescription antidepressant medications, and experience of extended periods (>12 months) of dysphoric mood to assign clinical risk. On the basis of these data, women were prospectively assigned to one of three depression risk groups. We hypothesized that the higher a woman's depression risk group, the greater the likelihood that she would experience a clinically diagnosable episode of depression.

The P-1 study staff were trained to check the consistency and appropriateness of the self-reported data about prior treatment for depression and the use of antidepressant medications as a routine part of the medical screening procedure carried out during entry/eligibility interview. These procedures were designed specifically to minimize false-positive classification errors. However, there was little that the interviewer could do to detect false-negative classification errors in which a potential participant did not, for whatever reason, report the requested screening information. The overall effect of this inability to control for false-negative classification errors for the current study was to create a potential misclassification bias in which women at increased risk for depression may have been placed, at an unknown rate, in one of the lower risk groups. Although less than ideal, the effect of this bias is conservative in nature, operating to maintain the comparative validity of the most important high-risk depression group.

We found a statistically significant dose–response relationship between the level of the depression risk group (high> medium>low) and the proportion of the women in each depression risk group who scored above the clinical cutoff of 16 or higher on the CES-D at baseline and at every follow-up interview. In addition, women in the higher risk groups (high> medium>low) scored above the clinical cutoff on a greater proportion of their follow-up interviews and, on average, had higher maximum CES-D scores. Together, these data suggest that there was a dose–response effect, in which women in the higher depression risk groups (high>medium>low) were more likely to experience clinically significant episodes of affective distress and that these episodes, on average, were more persistent and more severe than the episodes in the lower risk groups. Finally, we found that the distribution of social and demographic correlates (i.e., age, marital and employment status, educational level, and use of medical services) across the three depression risk groups defined in this study followed the same general patterns of risk previously identified in the ECA study of depression among the general population (21). All of the above findings serve to support the validity of the risk assignments used in our study.

The primary test of our research question involved stratifying each depression risk group by treatment assignment (tamoxifen versus placebo) and comparing the corresponding proportions of women at each follow-up interview who scored above the clinical cutoff of 16 or higher on the CES-D. We found no effect of tamoxifen for any of the three depression risk groups.

Besides the lack of a positive association between tamoxifen use and depression, there are at least two possible alternative explanations for our negative findings: lack of statistical power and missing data. We carried out a post-hoc effect size analysis to determine the size of the difference between the treatment arms that might have been detected. For our highest risk depression group (n = 519), we had an 80% chance of detecting at least a 37% (OR ≥1.37) increase between the two study arms in the proportions of women scoring above the CES-D clinical cutoff of 16 or higher at any single examination point. When a repeated measures design was used, we had sufficient power to detect a mean increase of 24% (OR ≥1.24) in the proportion of women in either arm scoring above the CES-D clinical cutoff (33,34). We considered these to be acceptable levels of statistical power for the identification of clinically significant treatment effects in our high-risk depression group. The detectable ORs were, of course, even smaller for the low- and medium-risk depression groups.

We also assessed the contribution of missing data to explain the negative association between tamoxifen and depression in the P-1 study. An initial analysis showed that assigned depression risk was statistically significantly associated with missing data rates over the course of the study. If a tamoxifen-associated depression was the primary cause of these rates, we would have predicted that the tamoxifen treatment group in the higher depression risk groups would show a progressively greater differential off-treatment rate than the placebo group. This expectation was not confirmed by our data for the high-risk depression group.

In addition, we also examined the reasons given for going off the assigned treatment. There was a strong statistical association in the P-1 study between stopping assigned treatment and missing HRQL data (16). An analysis of the reasons for going off treatment in 81% of the women with missing HRQL data resulted in the following observations: (a) Depression was cited as a relatively infrequent reason for going off treatment; (b) the higher the depression risk group, the greater the likelihood that depression was cited as the reason for going off treatment; and (c) within each depression risk group, depression was cited as the reason for going off treatment by similar proportions of women, regardless of treatment assignment. A separate report (35) has implemented a sensitivity analysis on these data with equally negative results. The findings in our report together with this sensitivity analysis indicates that there are no clear patterns in the missing data that serve to undermine the conclusions drawn from our primary analysis.

The results of our analysis strengthen our previous conclusion regarding lack of evidence for an association between tamoxifen use and depression in the P-1 study data by provisionally extending our findings to subgroups of women at a high risk for clinically identifiable episodes of depression. Clinically, these findings have two major implications. First, the evidence from NSABP's P-1 study does not lend support to the idea that tamoxifen should be considered to be a causal risk factor for the onset of depressive symptoms and/or the prolongation of depressive episodes that occur among treated women. Second, the findings of this study suggest that physicians need not automatically disqualify women as candidates for tamoxifen treatment simply because they report a history of depressive symptoms or prior treatment for a depressive disorder. Nevertheless, it is still essential that physicians carefully screen for affective disorders and treat or refer potential cases of depression encountered in routine clinical practice.

Finally, there are two important limitations on these conclusions that require discussion, one statistical and the other methodological. Statistically, it was the large size of the P-1 study that permitted us to identify and carry out stratified analyses of groups of women with a differential risk for depression. However, we also noted that there were limits on our statistical power to detect an increase in the proportion of women reporting clinically significant levels of depressive symptoms on the CES-D, particularly in the high-risk depression group. For this reason, we cannot absolutely exclude the possibility that there may be rare cases in which women react negatively to tamoxifen treatment with potentially life-threatening depressions. Here, it is useful to recall that data on neuro-mood toxic effects were collected for P-1 study participants and periodically reviewed as part of the routine safety-monitoring procedures. Over the full course of the P-1 study, there were a total of three women who committed suicide, one woman from the placebo-treated group and two women from the tamoxifen-treated group, and there were no statistically significant differences in the distribution of women reporting suicidal ideation across the two trial arms.

The methodological limitations of this article (i.e., the lack of standardized psychiatric diagnoses and missing HRQL data) are primarily due to the fact that the goals of this study were secondary to the main clinical objectives that determined the design of the P-1 study. A more definitive analysis would require additional data from a potentially smaller, yet more focused study, in which an investigation of the relationship between clinical depression and tamoxifen treatment was the primary scientific objective. Such a study would have to have the following minimum features: (a) a double-blind, placebo-controlled, randomized design; (b) participants who are at high risk for breast cancer, rather than breast cancer patients (to avoid potential confounding due to clinical diagnosis and treatment); (c) participants who are stratified on a reliable measure of risk for affective disorder (e.g., lifetime diagnosis, Schedule for Affective Disorders and Schizophrenia—Lifetime Version); (d) periodic administration, in whole or in part, of a standardized psychiatric diagnostic instrument (e.g., Diagnostic Interview Schedule) by a trained interviewer; and (e) continued collection of the psychiatric interview data even if the participant goes off the assigned treatment for any reason, except death or consent withdrawal. Whether the additional information obtained from such a study would justify the time and the expense involved in its collection is a problematic question that is beyond the scope of this article.

Table 1.

Distribution of self-reported risk factors for clinical depressive disorders at baseline examination among participants of the National Surgical Adjuvant Breast and Bowel Project's Breast Cancer Prevention (P-1) Study

    Risk factor pattern* 0–3 risk factors† 
Risk group (items endorsed) History of depression Antidepressant medications Persistent dysphoria No. No. 
*Depression risk groups were assigned on the basis of the participants' response to three medical history questions: 1) history of depression, 2) use of antidepressant medication, and 3) persistent mood disturbance (dysphoria). Each positive answer was worth 1 point. Participants with a score of 0 were assigned to the low-risk group, those with a score of 1–2 to the medium-risk group, and those with a score of 3 to the high-risk group. 
†Number and percent of participants endorsing 0, 1, 2, or 3 depression risk factors. 
Low (0) No No No 7964 72.0 7964 72.0 
Medium (1) No No Yes 621 5.6   
 No Yes No 668 6.0 1628 14.7 
 Yes No No 339 3.1   
Medium (2) No Yes Yes 120 1.1   
 Yes No Yes 202 1.8 953 8.6 
 Yes Yes No 631 5.7   
High (3) Yes Yes Yes 519 4.7 519 4.7 
    Total    11 064 100.0 11 064 100.0 
    Risk factor pattern* 0–3 risk factors† 
Risk group (items endorsed) History of depression Antidepressant medications Persistent dysphoria No. No. 
*Depression risk groups were assigned on the basis of the participants' response to three medical history questions: 1) history of depression, 2) use of antidepressant medication, and 3) persistent mood disturbance (dysphoria). Each positive answer was worth 1 point. Participants with a score of 0 were assigned to the low-risk group, those with a score of 1–2 to the medium-risk group, and those with a score of 3 to the high-risk group. 
†Number and percent of participants endorsing 0, 1, 2, or 3 depression risk factors. 
Low (0) No No No 7964 72.0 7964 72.0 
Medium (1) No No Yes 621 5.6   
 No Yes No 668 6.0 1628 14.7 
 Yes No No 339 3.1   
Medium (2) No Yes Yes 120 1.1   
 Yes No Yes 202 1.8 953 8.6 
 Yes Yes No 631 5.7   
High (3) Yes Yes Yes 519 4.7 519 4.7 
    Total    11 064 100.0 11 064 100.0 
Table 2.

Distribution of NSABP P-1 participants on ECA study social and demographic correlates of clinically diagnosed depressive disorders by depression risk score*

 Depression risk score†   
Sociodemographic item Low, % Medium, % High, % Odds ratio‡ 95% confidence interval on odds ratio 
*NSABP P-1 = Natural Surgical Adjuvant Breast and Bowel Project's Breast Cancer Prevention (P-1) Study; ECA = National Institutes of Mental Health's Epidemiological Catchment Area study (19). 
†Depression risk groups were assigned on the basis of the participants' responses to three medical history questions: 1) history of depression, 2) use of antidepressant medication, and 3) persistent mood disturbance (dysphoria). Each positive answer was worth 1 point. Participants with a score of 0 were assigned to the low-risk group, those with a score of 1–2 to the medium-risk group, and those with a score of 3 to the high-risk group. 
‡Odds ratios were determined by binary logistic regression; P<.001 for all groups compared with referent groups, except for education, where P=.235. 
Marital status: divorced/separated 11.1 17.7 23.5 1.63 1.50 to 1.98 
Employment status: not working 4.4 7.9 12.2 1.78 1.58 to 2.01 
Visited doctor within last 3 mo 71.0 76.4 84.4 1.39 1.28 to 1.51 
Hospitalized within last 5 y 42.7 48.6 54.9 1.27 1.19 to 1.36 
Age: ≥60 y 29.9 27.4 24.1 0.87 0.81 to 0.94 
Education: >high school 66.6 66.7 70.0 1.04 0.97 to 1.12 
Income: >median 46.1 37.6 31.5 0.72 0.67 to 0.77 
 Depression risk score†   
Sociodemographic item Low, % Medium, % High, % Odds ratio‡ 95% confidence interval on odds ratio 
*NSABP P-1 = Natural Surgical Adjuvant Breast and Bowel Project's Breast Cancer Prevention (P-1) Study; ECA = National Institutes of Mental Health's Epidemiological Catchment Area study (19). 
†Depression risk groups were assigned on the basis of the participants' responses to three medical history questions: 1) history of depression, 2) use of antidepressant medication, and 3) persistent mood disturbance (dysphoria). Each positive answer was worth 1 point. Participants with a score of 0 were assigned to the low-risk group, those with a score of 1–2 to the medium-risk group, and those with a score of 3 to the high-risk group. 
‡Odds ratios were determined by binary logistic regression; P<.001 for all groups compared with referent groups, except for education, where P=.235. 
Marital status: divorced/separated 11.1 17.7 23.5 1.63 1.50 to 1.98 
Employment status: not working 4.4 7.9 12.2 1.78 1.58 to 2.01 
Visited doctor within last 3 mo 71.0 76.4 84.4 1.39 1.28 to 1.51 
Hospitalized within last 5 y 42.7 48.6 54.9 1.27 1.19 to 1.36 
Age: ≥60 y 29.9 27.4 24.1 0.87 0.81 to 0.94 
Education: >high school 66.6 66.7 70.0 1.04 0.97 to 1.12 
Income: >median 46.1 37.6 31.5 0.72 0.67 to 0.77 
Table 3.

Distribution of Center for Epidemiological Studies—Depression (CES-D) Scale variables for NSABP P-1 participants who scored above the clinical cutoff of 16 or higher by depression risk group*

 Depression risk group† 
CES-D variable Low Medium High 
*The CES-D is a self-administered questionnaire, composed of 20 items, each of which is scored on a scale of 0–3. Higher scores reflect increased expression of affective distress, and a total score of 16 or higher is used as the cutoff point for likely cases of clinical depression (17,23,24). NSABP P-1 = National Surgical Adjuvant Breast and Bowel Project's Breast Cancer Prevention (P-1) Study; CI = confidence interval. 
†Depression risk groups were assigned on the basis of the participants' responses to three medical history questions: 1) history of depression, 2) use of antidepressant medication, and 3) persistent mood disturbance (dysphoria). Each positive answer was worth 1 point. Participants with a score of 0 were assigned to the low-risk group, those with a score of 1–2 to the medium-risk group, and those with a score of 3 to the high-risk group. 
‡There is a statistically significant difference between all groups (Kruskal–Wallis and analysis of variance: P<.001). “Maximum score ≥16” represents the highest single CES-D score ≥16 reported for an individual, whereas “All scores ≥16” summarizes all of the CES-D scores ≥16 reported for an individual. 
% follow-up examinations in which participants scored ≥16‡    
    Mean 0.087 0.192 0.357 
    95% CI for mean 0.083 to 0.091 0.181 to 0.203 0.325 to 0.389 
Maximum score ≥16‡    
    Median 22 24 27 
    Mean 23.97 25.61 28.58 
    95% CI for mean 23.66 to 24.28 25.16 to 26.06 27.62 to 29.54 
All scores ≥16‡    
    Median 20 21 22 
    Mean 21.52 22.49 23.74 
    95% CI for mean 21.30 to 21.74 22.17 to 22.81 23.10 to 24.38 
 Depression risk group† 
CES-D variable Low Medium High 
*The CES-D is a self-administered questionnaire, composed of 20 items, each of which is scored on a scale of 0–3. Higher scores reflect increased expression of affective distress, and a total score of 16 or higher is used as the cutoff point for likely cases of clinical depression (17,23,24). NSABP P-1 = National Surgical Adjuvant Breast and Bowel Project's Breast Cancer Prevention (P-1) Study; CI = confidence interval. 
†Depression risk groups were assigned on the basis of the participants' responses to three medical history questions: 1) history of depression, 2) use of antidepressant medication, and 3) persistent mood disturbance (dysphoria). Each positive answer was worth 1 point. Participants with a score of 0 were assigned to the low-risk group, those with a score of 1–2 to the medium-risk group, and those with a score of 3 to the high-risk group. 
‡There is a statistically significant difference between all groups (Kruskal–Wallis and analysis of variance: P<.001). “Maximum score ≥16” represents the highest single CES-D score ≥16 reported for an individual, whereas “All scores ≥16” summarizes all of the CES-D scores ≥16 reported for an individual. 
% follow-up examinations in which participants scored ≥16‡    
    Mean 0.087 0.192 0.357 
    95% CI for mean 0.083 to 0.091 0.181 to 0.203 0.325 to 0.389 
Maximum score ≥16‡    
    Median 22 24 27 
    Mean 23.97 25.61 28.58 
    95% CI for mean 23.66 to 24.28 25.16 to 26.06 27.62 to 29.54 
All scores ≥16‡    
    Median 20 21 22 
    Mean 21.52 22.49 23.74 
    95% CI for mean 21.30 to 21.74 22.17 to 22.81 23.10 to 24.38 
Table 4.

Comparison (binary logistic regression) of the proportion of NSABP P-1 participants in each treatment group (tamoxifen versus placebo) who scored 16 or higher on the Center for Epidemiological Studies—Depression (CES-D) Scale by depression risk group and sequential examination*

 Sequential examination 
Depression risk group† Baseline 3 mo 6 mo 12 mo 18 mo 24 mo 30 mo 36 mo 
*The CES-D is a self-administered questionnaire, composed of 20 items, each of which is scored on a scale of 0–3. Higher scores reflect increased expression of affective distress, and a total score of 16 or higher is used as the cutoff point for likely cases of clinical depression (17,23,24). NSABP P-1 = National Surgical Adjuvant Breast and Bowel Project's Breast Cancer Prevention (P-1) Study; OR = odds ratio; CI = confidence interval. 
†Depression risk groups were assigned on the basis of the participants' responses to three medical history questions: 1) history of depression, 2) use of antidepressant medication, and 3) persistent mood disturbance (dysphoria). Each positive answer was worth 1 point. Participants with a score of 0 were assigned to the low-risk group, those with a score of 1–2 to the medium-risk group, and those with a score of 3 to the high-risk group. 
‡OR >1.0 indicates a greater proportion of women in the tamoxifen group. 
Low         
    OR‡ 1.22 1.04 1.01 1.02 0.88 0.96 0.93 0.86 
    95% CI 0.96 to 1.55 0.86 to 1.25 0.85 to 1.19 0.86 to 1.02 0.75 to 1.04 0.80 to 1.13 0.78 to 1.12 0.71 to 1.03 
    P  .10  .68  .91 0.86  .14  .60  .44  .11 
Medium         
    OR‡ 1.03 1.29 1.10 0.99 0.91 1.04 0.96 1.01 
    95% CI 0.81 to 1.30 1.04 to 1.60 0.89 to 1.35 0.81 to 1.22 0.74 to 1.13 0.82 to 1.30 0.75 to 1.22 0.79 to 1.29 
    P  .84  .02  .39  .95  .40  .76  .72  .94 
High         
    OR‡ 0.89 0.78 0.74 0.62 0.84 0.83 1.00 1.00 
    95% CI 0.61 to 1.30 0.54 to 1.14 0.50 to 1.09 0.41 to 0.92 0.56 to 1.26 0.54 to 1.28 0.65 to 1.54 0.64 to 1.57 
    P  .54  .21  .13  .02  .40  .40  .99  .99 
 Sequential examination 
Depression risk group† Baseline 3 mo 6 mo 12 mo 18 mo 24 mo 30 mo 36 mo 
*The CES-D is a self-administered questionnaire, composed of 20 items, each of which is scored on a scale of 0–3. Higher scores reflect increased expression of affective distress, and a total score of 16 or higher is used as the cutoff point for likely cases of clinical depression (17,23,24). NSABP P-1 = National Surgical Adjuvant Breast and Bowel Project's Breast Cancer Prevention (P-1) Study; OR = odds ratio; CI = confidence interval. 
†Depression risk groups were assigned on the basis of the participants' responses to three medical history questions: 1) history of depression, 2) use of antidepressant medication, and 3) persistent mood disturbance (dysphoria). Each positive answer was worth 1 point. Participants with a score of 0 were assigned to the low-risk group, those with a score of 1–2 to the medium-risk group, and those with a score of 3 to the high-risk group. 
‡OR >1.0 indicates a greater proportion of women in the tamoxifen group. 
Low         
    OR‡ 1.22 1.04 1.01 1.02 0.88 0.96 0.93 0.86 
    95% CI 0.96 to 1.55 0.86 to 1.25 0.85 to 1.19 0.86 to 1.02 0.75 to 1.04 0.80 to 1.13 0.78 to 1.12 0.71 to 1.03 
    P  .10  .68  .91 0.86  .14  .60  .44  .11 
Medium         
    OR‡ 1.03 1.29 1.10 0.99 0.91 1.04 0.96 1.01 
    95% CI 0.81 to 1.30 1.04 to 1.60 0.89 to 1.35 0.81 to 1.22 0.74 to 1.13 0.82 to 1.30 0.75 to 1.22 0.79 to 1.29 
    P  .84  .02  .39  .95  .40  .76  .72  .94 
High         
    OR‡ 0.89 0.78 0.74 0.62 0.84 0.83 1.00 1.00 
    95% CI 0.61 to 1.30 0.54 to 1.14 0.50 to 1.09 0.41 to 0.92 0.56 to 1.26 0.54 to 1.28 0.65 to 1.54 0.64 to 1.57 
    P  .54  .21  .13  .02  .40  .40  .99  .99 
Table 5.

Reasons cited for going off treatment by depression risk* and treatment group

 Low risk Medium risk High risk  
Reasons cited for going off treatment Placebo Tamoxifen Placebo Tamoxifen Placebo Tamoxifen Overall 
*Depression risk groups were assigned on the basis of the participants' responses to three medical history questions: 1) history of depression, 2) use of antidepressant medication, and 3) persistent mood disturbance (dysphoria). Each positive answer was worth 1 point. Participants with a score of 0 were assigned to the low-risk group, those with a score of 1–2 to the medium-risk group, and those with a score of 3 to the high-risk group. 
Depression (No. of participants) 20 27 21 24 110 
Other reasons (No. of participants) 1130 1275 416 431 83 94 3429 
Depression as % of all off-treatment reasons 1.7 2.1 4.8 5.3 9.8 8.7 3.1 
 Low risk Medium risk High risk  
Reasons cited for going off treatment Placebo Tamoxifen Placebo Tamoxifen Placebo Tamoxifen Overall 
*Depression risk groups were assigned on the basis of the participants' responses to three medical history questions: 1) history of depression, 2) use of antidepressant medication, and 3) persistent mood disturbance (dysphoria). Each positive answer was worth 1 point. Participants with a score of 0 were assigned to the low-risk group, those with a score of 1–2 to the medium-risk group, and those with a score of 3 to the high-risk group. 
Depression (No. of participants) 20 27 21 24 110 
Other reasons (No. of participants) 1130 1275 416 431 83 94 3429 
Depression as % of all off-treatment reasons 1.7 2.1 4.8 5.3 9.8 8.7 3.1 
Fig. 1.

Proportion of participants in the National Surgical Adjuvant Breast and Bowel Project's Breast Cancer Prevention (P-1) Study scoring 16 or higher on the Center for Epidemiological Studies—Depression (CES-D) Scale with 95% confidence intervals by depression risk groups (low, medium, or high) (a) and by depression risk group and treatment assignment (placebo versus tamoxifen [TAM]) (b–d). Depression risk groups were assigned on the basis of the participants' responses to three medical history questions: 1) history of depression, 2) use of antidepressant medication, and 3) persistent mood disturbance (dysphoria). Each positive answer was worth 1 point. Participants with a score of 0 were assigned to the low-risk group, those with a score of 1–2 to the medium-risk group, and those with a score of 3 to the high-risk group.

Fig. 1.

Proportion of participants in the National Surgical Adjuvant Breast and Bowel Project's Breast Cancer Prevention (P-1) Study scoring 16 or higher on the Center for Epidemiological Studies—Depression (CES-D) Scale with 95% confidence intervals by depression risk groups (low, medium, or high) (a) and by depression risk group and treatment assignment (placebo versus tamoxifen [TAM]) (b–d). Depression risk groups were assigned on the basis of the participants' responses to three medical history questions: 1) history of depression, 2) use of antidepressant medication, and 3) persistent mood disturbance (dysphoria). Each positive answer was worth 1 point. Participants with a score of 0 were assigned to the low-risk group, those with a score of 1–2 to the medium-risk group, and those with a score of 3 to the high-risk group.

Fig. 2.

Kaplan–Meier curves of time from randomization to first score of 16 or higher on the Center for Epidemiological Studies—Depression (CES-D) Scale by depression risk group (low, medium, or high) and treatment assignment (placebo versus tamoxifen [TAM]). Depression risk groups were assigned on the basis of the participants' responses to three medical history questions: 1) history of depression, 2) use of antidepressant medication, and 3) persistent mood disturbance (dysphoria). Each positive answer was worth 1 point. Participants with a score of 0 were assigned to the low-risk group, those with a score of 1–2 to the medium-risk group, and those with a score of 3 to the high-risk group. At 10 months, for the patients who received tamoxifen, in the low-risk group there were 3159 patients at risk of depression (proportion remaining = 0.864; 95% confidence interval [CI] = 0.853 to 0.875); in the medium-risk group there were 799 patients at risk (proportion remaining = 0.685; 95% CI = 0.659 to 0.711); and in the high-risk group there were 123 patients at risk (proportion remaining = 0.488; 95% CI = 0.427 to 0.549). At 30 months, for the patients who received tamoxifen, in the low-risk group there were 2233 patients at risk for depression (proportion remaining = 0.746; 95% CI = 0.732 to 0.760); in the medium-risk group there were 496 patients at risk (proportion remaining = 0.528; 95% CI = 0.499 to 0.557); and in the high-risk group there were 61 patients at risk (proportion remaining = 0.317; 95% CI = 0.258 to 0.376). At 10 months, for the patients who received the placebo, in the low-risk group there were 3190 patients at risk for depression (proportion remaining = 0.870; 95% CI = 0.859 to 0.881); in the medium-risk group there were 863 patients at risk (proportion remaining = 0.713; 95% CI = 0.688 to 0.738); and in the high-risk group there were 108 patients at risk (proportion remaining = 0.475; 95% CI = 0.412 to 0.538). At 30 months, for the patients who received the placebo, in the low-risk group there were 2326 patients at risk for depression (proportion remaining = 0.753; 95% CI = 0.738 to 0.767); in the medium risk group there were 544 patients at risk (proportion remaining = 0.535; 95% CI = 0.506 to 0.563); and in the high-risk group there were 59 patients at risk (proportion remaining = 0.316; 95% CI = 0.254 to 0.377).

Fig. 2.

Kaplan–Meier curves of time from randomization to first score of 16 or higher on the Center for Epidemiological Studies—Depression (CES-D) Scale by depression risk group (low, medium, or high) and treatment assignment (placebo versus tamoxifen [TAM]). Depression risk groups were assigned on the basis of the participants' responses to three medical history questions: 1) history of depression, 2) use of antidepressant medication, and 3) persistent mood disturbance (dysphoria). Each positive answer was worth 1 point. Participants with a score of 0 were assigned to the low-risk group, those with a score of 1–2 to the medium-risk group, and those with a score of 3 to the high-risk group. At 10 months, for the patients who received tamoxifen, in the low-risk group there were 3159 patients at risk of depression (proportion remaining = 0.864; 95% confidence interval [CI] = 0.853 to 0.875); in the medium-risk group there were 799 patients at risk (proportion remaining = 0.685; 95% CI = 0.659 to 0.711); and in the high-risk group there were 123 patients at risk (proportion remaining = 0.488; 95% CI = 0.427 to 0.549). At 30 months, for the patients who received tamoxifen, in the low-risk group there were 2233 patients at risk for depression (proportion remaining = 0.746; 95% CI = 0.732 to 0.760); in the medium-risk group there were 496 patients at risk (proportion remaining = 0.528; 95% CI = 0.499 to 0.557); and in the high-risk group there were 61 patients at risk (proportion remaining = 0.317; 95% CI = 0.258 to 0.376). At 10 months, for the patients who received the placebo, in the low-risk group there were 3190 patients at risk for depression (proportion remaining = 0.870; 95% CI = 0.859 to 0.881); in the medium-risk group there were 863 patients at risk (proportion remaining = 0.713; 95% CI = 0.688 to 0.738); and in the high-risk group there were 108 patients at risk (proportion remaining = 0.475; 95% CI = 0.412 to 0.538). At 30 months, for the patients who received the placebo, in the low-risk group there were 2326 patients at risk for depression (proportion remaining = 0.753; 95% CI = 0.738 to 0.767); in the medium risk group there were 544 patients at risk (proportion remaining = 0.535; 95% CI = 0.506 to 0.563); and in the high-risk group there were 59 patients at risk (proportion remaining = 0.316; 95% CI = 0.254 to 0.377).

Fig. 3.

Proportion of participants in the National Surgical Adjuvant Breast and Bowel Project's Breast Cancer Prevention (P-1) Study completing the health-related quality-of-life questionnaire by depression risk groups (low, medium, or high) (a) and by depression risk group and treatment assignment (placebo versus tamoxifen [TAM]) with 95% confidence intervals (b–d). Depression risk groups were assigned on the basis of the participants' responses to three medical history questions: 1) history of depression, 2) use of antidepressant medication, and 3) persistent mood disturbance (dysphoria). Each positive answer was worth 1 point. Participants with a score of 0 were assigned to the low-risk group, those with a score of 1–2 to the medium-risk group, and those with a score of 3 to the high-risk group.

Fig. 3.

Proportion of participants in the National Surgical Adjuvant Breast and Bowel Project's Breast Cancer Prevention (P-1) Study completing the health-related quality-of-life questionnaire by depression risk groups (low, medium, or high) (a) and by depression risk group and treatment assignment (placebo versus tamoxifen [TAM]) with 95% confidence intervals (b–d). Depression risk groups were assigned on the basis of the participants' responses to three medical history questions: 1) history of depression, 2) use of antidepressant medication, and 3) persistent mood disturbance (dysphoria). Each positive answer was worth 1 point. Participants with a score of 0 were assigned to the low-risk group, those with a score of 1–2 to the medium-risk group, and those with a score of 3 to the high-risk group.

Supported by Public Health Service grant NCI-U10CA37377/69974 from the National Cancer Institute, National Institutes of Health, Department of Health and Human Services; by career development award DAMD17–97–1-7058 from the Department of Defense (to R. Day); and in part by an American Cancer Society Clinical Research Professorship (to P. A. Ganz).

We thank Samuel Wieand, Ph.D, Stephanie Land, Ph.D., and Ms. Sheela Goshal of the National Surgical Adjuvant Breast and Bowel Project (NSABP) Biostatistical Center and D. Lawrence Wickerham, M.D., of the NSABP Operations Center for their help in the preparation of this article.

References

1
Cathcart CK, Jones SE, Pumroy CS, Peters GN, Knox SM, Cheek JH. Clinical recognition and management of depression in node negative breast cancer patients treated with tamoxifen.
Breast Cancer Res Treat
 
1993
;
27
:
277
–81.
2
Shariff S, Cumming CE, Lees A, Handman M, Cumming DC. Mood disorder in women with early breast cancer taking tamoxifen, an estradiol receptor antagonist. An expected or unexpected effect?
Ann N Y Acad Sci
 
1995
;
761
:
365
–8.
3
Anelli TF, Anelli A, Tran KN, Lebwohl DE, Borgen PI. Tamoxifen administration is associated with a high rate of treatment-limiting symptoms in male breast cancer patients.
Cancer
 
1994
;
74
:
74
–7.
4
Pluss JL, DiBella NJ. Reversible central nervous system dysfunction due to tamoxifen in a patient with breast cancer.
Ann Intern Med
 
1984
;
101
:
652
.
5
Duffy LS, Greenberg DB, Younger J, Ferraro MG. Iatrogenic acute estrogen deficiency and psychiatric syndromes in breast cancer patients.
Psychosomatics
 
1999
;
40
:
304
–8.
6
Physician's Desk Reference. 53rd ed. Oradell (NJ): Medical Economics Co.; 1999.
7
Love RR, Cameron L, Connell BL, Leventhal H. Symptoms associated with tamoxifen treatment in postmenopausal women.
Arch Intern Med
 
1991
;
151
:
1842
–7.
8
Fallowfield L, Fleissig A, Edwards R, West A, Powles TJ, Howell A, et al. Tamoxifen for the prevention of breast cancer: psychosocial impact on women participating in two randomized controlled trials.
J Clin Oncol
 
2001
;
19
:
1885
–92.
9
Halbreich U. Role of estrogen in postmenopausal depression.
Neurology
 
1997
;
48
(5 Suppl 7):
S16
–9.
10
Gregoire AJ, Kumar R, Everitt B, Henderson AF, Studd JW. Transdermal oestrogen for treatment of severe postnatal depression.
Lancet
 
1996
;
347
:
930
–3.
11
Murray D. Oestrogen and postnatal depression.
Lancet
 
1996
;
347
:
918
–9.
12
Fisher B, Costantino JP, Wickerham DL, Redmond CK, Kavanah M, Cronin WM, et al. Tamoxifen for the prevention of breast cancer: report from the National Surgical Adjuvant Breast and Bowel Project P-1 Study.
J Natl Cancer Inst
 
1998
;
90
:
1371
–88.
13
Fisher B, Costantino J. Highlights of the NSABP Breast Cancer Prevention Trial.
Cancer Control
 
1997
;
4
:
78
–86.
14
Ganz PA, Day R, Ware JE Jr, Redmond C, Fisher B. Base-line quality-of-life assessment in the National Surgical Adjuvant Breast and Bowel Project Breast Cancer Prevention Trial.
J Natl Cancer Inst
 
1995
;
87
:
1372
–82.
15
Ganz PA, Day R, Costantino JP. Compliance with quality of life data collection in the National Surgical Adjuvant Breast and Bowel Project (NSABP) Breast Cancer Prevention Trial.
Stat Med
 
1998
;
17
:
613
–22.
16
Day R, Ganz PA, Costantino JP, Cronin WM, Wickerham DL, Fisher B. Health-related quality of life and tamoxifen in breast cancer prevention: a report from the National Surgical Adjuvant Breast and Bowel Project P-1 Study.
J Clin Oncol
 
1999
;
17
:
2659
–69.
17
Radloff LS. The CES-D scale: a self-report depression scale for use in the general population.
Applied Psychol Measurement
 
1977
;
1
:
385
–401.
18
Stewart AL, Ware JE. Measuring functioning and well-being: the medical outcomes approach. Durham (NC): Duke University Press; 1992.
19
Weissman MM, Bruce ML, Leaf PJ, Florio LP, Holzer C. Affective disorders. In: Robins LN, Regier DA, editors. Psychiatric disorders in America: the Epidemiological Catchment Area study. New York (NY): The Free Press; 1991.
20
Diagnostic and statistical manual of mental disorders: DSM-IV. Washington (DC): American Psychiatric Association; 2000.
21
Spitzer RL, Endicott J, Robbins E. Research Diagnostic Criteria. New York (NY): Biometrics Research Division, New York State Psychiatric Institute; 1978.
22
Dohrenwend BP, Shrout PE, Egin G, Mendelsohn FS. Nonspecific psychological distress and other dimensions of psychopathology. Measures for use in the general population.
Arch Gen Psychiatry
 
1980
;
37
:
1229
–36.
23
Myers JK, Weissman MM. Use of a self-report symptom scale to detect depression in a community sample.
Am J Psychiatry
 
1980
;
137
:
1081
–4.
24
Lin N, Dean A, Ensel WM. Social support, life events, and depression. New York (NY): Academic Press; 1986.
25
Boyd JH, Weissman MM, Thompson WD, Myers JK. Screening for depression in a community sample. Understanding the discrepancies between depression symptom and diagnostic scales.
Arch Gen Psychiatry
 
1982
;
39
:
1195
–200.
26
Spitzer RL, Endicott J. Schedule for Affective Disorders and Schizophrenia. New York (NY): Biometrics Research Division, New York State Psychiatric Institute; 1978.
27
Kaelber CT, Moul DE, Farmer ME. Epidemiology of depression. In: Beckham EE, Leber WR, editors. Handbook of depression. 2nd ed. New York (NY): Guilford Press; 1995. p. 3–35.
28
Keller MB, Lavori PW, Rice J, Coryell W, Hirschfeld RM. The persistent risk of chronicity in recurrent episodes of nonbipolar major depressive disorder: a prospective follow-up.
Am J Psychiatry
 
1986
;
143
:
24
–8.
29
Keller MB, Lavori PW, Lewis CE, Klerman GL. Predictors of relapse in major depressive disorder.
JAMA
 
1983
;
250
:
3299
–304.
30
Leaf PJ, Livingston MM, Tischler GL, Weissman MM, Holzer CE 3rd, Myers JK. Contact with health professionals for the treatment of psychiatric and emotional problems.
Med Care
 
1985
;
23
:
1322
–37.
31
Costantino JP, Gail MH, Pee D, Anderson S, Redmond CK, Benichou J, et al. Validation studies for models projecting the risk of invasive and total breast cancer incidence.
J Natl Cancer Inst
 
1999
;
91
:
1541
–8.
32
Fisher B, Costantino J, Redmond C, Poisson R, Bowman D, Conture J, et al. A randomized clinical trial evaluating tamoxifen in the treatment of patients with node-negative breast cancer who have estrogen-receptor-positive tumors.
N Engl J Med
 
1989
;
320
:
479
–84.
33
Fleiss JH. Statistical methods for rates and proportions. 2nd ed. New York (NY): John Wiley & Sons; 1981.
34
Brown H, Prescott R. Applied mixed models in medicine. New York (NY): John Wiley & Sons; 1999.
35
Land S, Wieand S, Day R, Ten Have T, Costantino JP, Lang W, et al. Methodological issues in the analysis of quality of life data in clinical trials: illustrations from the National Surgical Adjuvant Breast and Bowel Project (NSABP) Breast Cancer Prevention Trial. In: Mesbah M, Cole B, Lee M, editors. Statistical design, measurement and analysis of health-related quality of life. New York (NY): Kluwer Academic Publishers. In press 2001.