Study population
Participants were breast cancer patients enrolled in a Los Angeles County population-based epidemiologic case-control study of breast cancer among Non-Hispanic/Hispanic white women who were 40 years or younger at diagnosis (referred to as the "parent study"). Details of the parent study have been published [21]. Briefly, white female residents of Los Angeles County diagnosed with first primary in situ or invasive breast cancer between July 1, 1983, and January 1, 1989, who were 40 years of age or younger at diagnosis, and who were born in the United States, Canada, or Europe, were eligible for participation in the parent study. Nine hundred and sixty nine eligible breast cancer patients were identified by the Los Angeles County Cancer Surveillance Program (LA-CSP), a population-based SEER registry. Of the 969 eligible women with incident breast cancer, 225 were not interviewed for the following reasons: death (n = 20), physician refusal (n = 54), serious mental or physical illness (n = 7), patient refusal (n = 111), no longer living in Los Angeles County (n = 12), and inability to be located (n = 21). Interviews were completed with 744 breast cancer patients (in situ: n = 68, invasive: n = 676).
The 744 participants were traced in 1998 and interviewed in 1999 and 2000 by telephone to collect the detailed information regarding herbal remedy use and concurrent QOL. Of the 744 women, 276 were known to have died by the time the baseline telephone interview for herbal remedy use was initiated; 78 women were not located; 6 subjects refused participation; and 10 women were interested but unable to schedule an interview during the study period. Three hundred seventy-four women were interviewed. The interviews were conducted, on average, 13 years after diagnosis. The cross-sectional relationship between herbal remedy use and QOL has been reported elsewhere [20].
The 374 women who participated in our baseline telephone interview for herbal remedy use were followed for survival through December 31, 2007 by linking with the LA-CSP and the National Death Index. If death occurred, information on date of death and cause of death was collected. Person-days of follow-up for each woman began on the date of baseline telephone interview and ended on her date of death (n = 44 including 33 who died of breast cancer) or December 31, 2007 (n = 330). Among 33 who died from breast cancer, 13 had metastasis of breast cancer; 2 had recurrence of breast cancer; 2 had a contralateral breast cancer; 10 had a combination of these three conditions, and the other 6 women did not have any of these three conditions by the time of our baseline telephone interview. We excluded three women with incomplete questionnaire data who were alive at the end of the follow-up period. Therefore, 371 women were eligible for survival analysis.
We conducted a follow-up telephone interview in 2002-2004 to collect additional SF-36 QOL information 2-5 years after the initial collection of herbal remedy and QOL information. Of 374 case patients participating in the baseline telephone interview, 26 were known to have died when we began the follow-up interview. Of the remaining 348 women, 17 women were not located (4.9%); 26 subjects preferred not to participate (7.5%); and 3 women were too ill to participate (0.9%). Three hundred two women were interviewed. After excluding the same three women with incomplete questionnaire data, 299 women remained for analysis of the association between herbal remedy use and QOL. The recruitment results are combined and shown in Figure 1.
Data collection
In the parent study, we obtained data on summary stage at breast cancer diagnosis from the LA-CSP. The baseline telephone interview for herbal remedy use (1999-2000) obtained information on herbal remedy use, SF-36 QOL measures, psychosocial functioning, cancer-related factors including recurrences, breast-cancer surgeries and reconstructive history, treatments received for breast cancer, and comorbidities. We also updated demographic information and information on lifestyle practices after breast cancer diagnosis.
Women were given a list of herbal remedies and asked whether they had used any of the listed remedies within the past six months. The remedies listed were bee pollen, licorice root, black cohosh, mother wort, blue cohosh, nux vomica, chaste berries, progesterone topical (vitex agnus cactii), progesterone cream (wild Mexican yam), chickweed tincture, pulsatilla, dong quai (tong Kwai), royal jelly, echinacea, sage tea, evening primrose oil, sarsaparilla, false unicorn, sepia, garlic, St. John's Wort, ginko biloba, valeriana, ginseng, wild yam root, herbal tea used as a remedy, shark cartilage, or lachesis. Use of any of the herbal remedies qualified a woman as an herbal remedy user.
The follow-up telephone interview (2002-2004) again measured SF-36 QOL and psychosocial functioning; updated demographic information and lifestyle practices and cancer-related and treatment-related factors including additional surgeries. Subjects were mailed a list of scale responses to the QOL and psychosocial instruments prior to each of their telephone interviews to facilitate the interview process.
Health-related QOL was assessed using the SF-36 questionnaire at each telephone interview. This widely used self-report measure contains 36 items in eight subscales (physical functioning, bodily pain, role limitations due to physical functioning, role limitations due to emotional problems, emotional well-being, social functioning, energy/fatigue, and general health perceptions) [22, 23]. Each subscale represents a combined score (range: 0-100) derived from questions related to a particular health concept. The SF-36 has been shown to be both valid and reliable [22]. Mental component scale (MCS) and physical component scale (PCS) summary scores account for 81.5% of the reliable variance in the eight SF-36 scales and are therefore useful in providing a summarized interpretation of QOL [24, 25]. In addition, these two scales have been scored in reference to a normal population (the 1998 general U.S. population, standard form) [26].
All participants who had previously signed an informed consent as part of their recruitment for the parent study gave a verbal informed consent prior to the conduct of each telephone interview. The study was approved by the Institutional Review Boards at the University of Southern California and the City of Hope.
Statistical analysis
We compared demographic characteristics between herbal remedy users and non-users using Pearson chi-square tests for differences in frequency distributions.
Survival time was calculated as the time from the collection of herbal use at baseline telephone interview to the first of the following events: date of death or the end of follow-up on December 31, 2007. For breast cancer specific survival, we censored women who died from other causes on their dates of death. We used the Kaplan-Meier method to calculate survival rates and applied the log-rank test to test for any difference in survival between herbal remedy users and non-users.
Multivariable Cox proportional hazards regression models were fit separately for all-cause and breast cancer mortality to assess the hazard rate ratios of mortality associated with a woman's herbal remedy use within the 6 months prior to the baseline telephone interview. Hazard rate ratios, presented as relative risks (RRs) with 95% confidence intervals (CIs) were estimated using age in days at the start and end of follow-up to define a woman's participation time in the study.
Among women who participated in the follow-up telephone interview, we used t-tests to compare the QOL summary scale measures between herbal remedy users and non-users at baseline and follow-up surveys, respectively. We further applied multivariable linear regression models to compare average SF-36 health summary scale change scores (change score = follow-up score - baseline score) between herbal remedy users and non-users.
All multivariable models adjusted for the following factors, selected a priori, as potential confounders: ethnicity (non-Hispanic origin white, Hispanic origin white), age at diagnosis (< 35, 35-38, 39-40 years), stage of cancer at diagnosis (in situ, localized invasive, non-localized invasive), type of non-surgical treatment for initial breast cancer (no treatment, only chemotherapy, only radiation therapy, only hormonal therapy, any combinations of these therapies), type of surgery (lumpectomy, mastectomy), post-diagnosis cancer-related conditions (none, 1, ≥ 2 cancer conditions that included lymphedema, contralateral breast cancer, recurrence of the primary breast cancer, diagnosis of cancer at another site, and history of any breast cancer metastasis), comorbidities within six months of baseline interview (none, 1, ≥ 2 including cardiovascular diseases [defined as high blood pressure requiring medication, angina, having had an angiogram, high cholesterol, clotting disorder, or stroke], respiratory problems [asthma or allergy disorder affecting breathing], inflammatory conditions [arthritis, gall bladder disease, diabetes], musculoskeletal conditions [osteoporosis, recent fracture], nervous system disorders [migraines, hearing loss, psychiatric problem], drug abuse [excessive use of alcohol or problem with prescription or street drug use or dependency]), interval between diagnosis and initial interview (10-11, 12-13, 14-16 years), and annual income ($30,000, $30,001-60,000, $60,001-100,000, > $100,000) at baseline interview. In addition, when comparing average SF-36 health summary scale change scores between herbal remedy users and non-users, multivariable linear regression models were adjusted for the SF-36 scores at baseline.
Our analysis of mortality had limited statistical power. In estimating what we could expect to observe with 44 deaths during follow-up and a prevalence of 58.5% of participants who had used herbal remedies, we could detect a 60% or greater decrease in risk (RR = 0.4) or a 180% or greater increase in risk (RR = 2.8) for overall mortality with 80% statistical power given a 2-sided hypothesis test with a 5% level of statistical significance [27].
Two-sided P-values of 0.05 or less were considered statistically significant. We did not adjust P values for multiple comparisons as these analyses were considered as exploratory [28]. All analyses were performed using SAS version 9.2 (SAS Institute, Cary, NC).