Participants

The WAFACS evaluated effects of a combination pill of folic acid (2.5 mg/day), vitamin B₆ (50 mg/day), and vitamin B₁₂ (1 mg/day) in prevention of major vascular events among women at high CVD risk. The trial began in 1998, when the folic acid and B-vitamin component was added to the Women’s Antioxidant Cardiovascular Study (WACS), then an ongoing 2 × 2 × 2 factorial trial of vitamins C and E and β-carotene. The design of WAFACS reflected biologically plausible synergy between homocysteine-lowering and antioxidant supplements for CVD prevention. Details of the design and the main results from the WAFACS and WACS have been published previously (www.clinicaltrials.gov Identifier NCT00000541). ^{Reference Albert, Cook, Gaziano, Zaharris, MacFadyen and Danielson13,Reference Bassuk, Albert, Cook, Zaharris, MacFadyen and Danielson15,Reference Cook, Albert, Gaziano, Zaharris, MacFadyen and Danielson16}

In the WACS, 8171 female health professionals were randomised between June 1995 and October 1996 to receive vitamin C (500 mg/day), vitamin E (600 IU every other day), and β-carotene (50 mg every other day) v. matching placebos. Eligible women were ≥40 years old, postmenopausal or had no intention of becoming pregnant and had a self-reported history of CVD (myocardial infarction, stroke, coronary revascularisation or angina) or at least three traditional CVD risk factors. In April 1998, 5442 of these women, who were willing and eligible for participation in WAFACS, were randomised to an active B-vitamin/folate pill or a matching placebo. Details of the randomisation scheme are provided elsewhere. ^{Reference Bassuk, Albert, Cook, Zaharris, MacFadyen and Danielson15} Briefly, participants were assigned to active treatment and matching placebo arms using computer-generated random permuted blocks; there were eight participants in each block and 64 strata (i.e. eight 5-year age groups × eight possible prior treatment groups (from the 2 × 2 × 2 factorial WACS groups)); this block randomisation scheme mitigated risk of unbalanced entry into the study arms during recruitment. Approval for the WAFACS and WACS, and for the current analysis, was obtained from the institutional review board of Brigham and Women’s Hospital (Boston, Massachusetts, USA). All participants gave written informed consent. The WAFACS was sponsored by the National Institutes of Health; study pills were provided by BASF Corporation (Mount Olive, New Jersey, USA).

Population for analysis

Participants with a history of depression before WAFACS randomisation (n = 1111) were excluded from this analysis, leaving 4331 women (Fig. 1). History of depression at baseline was determined by: (a) self-report of ever having physician-diagnosed depression; (b) self-reported use of select antidepressants

Fig. 1

Flow diagram of participation and incident depression in the folic acid/B₆/B₁₂ combination pill and placebo arms of the Women’s Antioxidant and Folic Acid Cardiovascular Study (WAFACS) trial.

(described further below), along with an appropriate (ICD-9-CM) ¹⁷ depressive disorder code (i.e. 296.2x, 296.3x, 300.4, 309.0, 309.1, 309.28, 311), as entered by trained data coders. Women who had ICD-9-CM codes for bipolar disorder or non-affective psychosis were also excluded. ICD-9-CM codes could be generated because participants specified physician diagnoses for which they were prescribed medications; participants did so by either writing down the diagnosis or verbally relaying this information to trained research staff over the phone.

The above exclusion process provided reassurance for assembling a baseline sample that was at risk for true incident depression. Regarding the requirement that antidepressant use be accompanied by a relevant ICD-9-CM code, this procedure was driven by clear evidence in our data of varying susceptibility to misclassification by class/type of antidepressant. For example, among all participants reporting use of selective serotonin reuptake inhibitors (SSRIs), serotonin and norepinephrine reuptake inhibitors (SNRIs), monoamine oxidase inhibitors (MAOIs, excluding low-dose (5 mg daily) selegiline), tri- or tetracyclics widely known for mood indications (for example nortriptyline, desipramine, imipramine, maprotiline) or newer/atypical antidepressants (for example bupropion, nefazodone), over 70% had a depression code and nearly 80% had a mood, anxiety or other mental health-related code. By contrast, only ∼25% of participants reporting use of certain tricyclic antidepressants (for example doxepin, amitriptyline) or trazodone had any mental health-related code – depression or otherwise; these women were frequently prescribed these medication for sleep or pain conditions. Thus, our data suggested that antidepressant use alone could not function as a reliable proxy of depression and likely reflected the shift away from tricyclic antidepressants in favour of newer agents during the post-1990s period of the WAFACS trial. ^{Reference Pirraglia, Stafford and Singer18}

WAFACS parent trial follow-up procedures

Participants were followed-up annually via mailed questionnaires to update information on occurrence of major illnesses or adverse events, numerous health and lifestyle factors and study adherence. Importantly, all participants had been administered at baseline a semiquantitative food-frequency questionnaire, previously developed and validated in highly comparable samples of health professionals, ^{Reference Willett, Sampson, Stampfer, Rosner, Bain and Witschi19,Reference Feskanich, Rimm, Giovannucci, Colditz, Stampfer and Litin20} to assess nutrient intakes; folate intake was calculated as total intake of folate from diet, including supplements and post-mandatory-fortification folate. Follow-up continued until the WAFACS’ planned end on 31 July 2005, for a total duration of 7.3 years; average adherence (i.e. taking at least two-thirds of study pills as assigned) was 83%, with no differences between active v. placebo groups. Morbidity and mortality information was complete for ≥98% of person-years of follow-up. ^{Reference Albert, Cook, Gaziano, Zaharris, MacFadyen and Danielson13}

Among a subset of participants, WAFACS measured plasma levels of relevant biomarkers to determine the influence of the active agent on nutrient levels, as well as to examine possible influences of background folate fortification among those receiving placebo (mandatory folate fortification of the US food supply commenced in 1998). ^{Reference Jacques, Selhub, Bostom, Wilson, Rosenberg and Collaboration21} Over 70% of women in the WAFACS provided a blood sample prior to the initiation of fortification. Among those adherent with study medications, 300 (150 in the active agent group and 150 in the placebo group) were randomly selected to provide blood samples at the end of randomised treatment. As detailed previously, ^{Reference Albert, Cook, Gaziano, Zaharris, MacFadyen and Danielson13} baseline median plasma folate and homocysteine levels were similar between active and placebo groups. At the end of follow-up, folate levels increased significantly in both groups, but the relative increase was greater in the active treatment group. Despite significant increases in post-fortification folate levels, there was no reduction in homocysteine levels when comparing values at the beginning v. the end of the trial among the placebo group. By contrast, a significant decrease in plasma homocysteine levels was observed in the active treatment group.

Ascertainment of incident depression

Incident depression was defined as physician/clinician-diagnosed depression or presence of clinically significant depressive symptoms. Information used to determine incident depression was obtained on the WACS 24-, 48- 72-, 84-, 96- and 120-month study questionnaires as either self-reported physician/clinician-diagnosed depression or self-reported depressive symptoms, based on the Mental Health Index (MHI). ^{Reference Cuijpers, Smits, Donker, ten Have and de Graaf22} Among participants in our sample for this depression substudy, the WACS 24-month questionnaire was the baseline/pre-randomisation survey for the B-vitamin/folate factorial arm. Thus, during follow-up, incident depression was classified as the first occurrence of self-reported: (a) physician/clinician-diagnosed depression (asked at the 72-, 84-, 96- and 120-month questionnaires) or (b) clinically significant depressive symptoms (asked at the 48- and 96-months questionnaires using the MHI).

Regarding the capture of physician/clinician-diagnosed depression, participants also had the ability to use write-in space to indicate a diagnosis, along with month/year of diagnosis, on any of the annual questionnaires during the entire follow-up period; however, physician/clinician depression diagnosis was only explicitly ascertained on the questionnaire years specified above.

Regarding clinically significant depressive symptoms, these were operationalised as follows: features of depression, by mood quality, duration and level of dysfunction, that are indicative of – at minimum – depressive disorder NOS (not otherwise specified) or minor depression. ²³ Specifically, in order to be classified as having clinically significant depressive symptoms, participants had to report feeling downhearted or blue most or all the time, for the preceding continuous 4 weeks, and endorse difficulty with work and/or social activities because of their emotional symptoms.

Finally, available data in study-event files (i.e. from phone or letter contacts with participants) were used to supplement the above end-points; these files included ICD-9-CM depressive disorder diagnoses entered by trained coders, who used participants’ self-reported physician diagnosis descriptions to perform coding. Depression event dates were calculated as the month/year for physician/clinician diagnosis and as the questionnaire return date for events determined on the basis of clinically significant depressive symptoms; if participants could be classified as depressed by more than one method, the earliest date was used.

Validity of the depression measures

Symptom data on our questionnaires came from a three-item version of the MHI, which has been validated elsewhere. ^{Reference Cuijpers, Smits, Donker, ten Have and de Graaf22} However, we applied a case definition even more rigorous than one based solely on cut-points, as we required both that participants endorsed the core feature of depressed quality of mood most or all of the time and that the emotional symptoms during that same 4-week time frame interfered with social and/or occupational activities. Our approach in using self-reported clinical data or symptoms to classify depression is consistent with published work ^{Reference Pan, Okereke, Sun, Logroscino, Manson and Willett24,Reference Li, Mirzaei, O'Reilly, Winkelman, Malhotra and Okereke25} and has yielded depression prevalence and incidence rates among community-dwelling older women in our prior studies ^{Reference Li, Mirzaei, O'Reilly, Winkelman, Malhotra and Okereke25,Reference Pan, Sun, Czernichow, Kivimaki, Okereke and Lucas26} that are similar to those that have been determined using structured, live interview methods. ^{Reference Steffens, Skoog, Norton, Hart, Tschanz and Plassman27,Reference Luijendijk, van, Berg, Dekker, van Tuijl, Otte and Smit28}

Statistical analysis

Primary analyses focused on depression incidence by randomised group. After exclusion of n = 1111 at baseline because of history of depression, data from all remaining 4331 randomised participants were analysed on an intention-to-treat basis. Participants were followed until the occurrence of the depression end-point, death or the end of the trial, whichever came first.

Baseline characteristics were compared by randomised groups using two-sample t- or Wilcoxon tests for continuous variables and χ² and Fisher exact statistics for categorical variables. Kaplan-Meier curves were used to estimate cumulative incidence for the active treatment and placebo groups; the log-rank test was used to compare the curves. The primary analysis used Cox proportional hazards models to estimate relative risks (RRs) and 95% confidence intervals for active treatment v. placebo, adjusting for the design variables of age and the other randomised agents (vitamin E, vitamin C and β-carotene). In a multivariable model, we further adjusted for baseline covariates with specific relevance: dietary intakes of folate, B₆ and B₁₂; alcohol intake (which affects folate absorption); medical comorbidity, summarised by the Charlson (Deyo) index. ^{Reference Deyo, Cherkin and Ciol29} However, of note, we did not observe significant evidence of imbalance in the distributions of these key factors by randomised treatment group among these 4331 women eligible for incident depression. In extended models, we included additional demographic, lifestyle and health factors (education, smoking, physical activity and menopausal status and hormone therapy); however, results were identical for the models above and are not detailed here. Finally, because of our specific scientific interest regarding influences of B-vitamin/folate supplementation on depression in late-life, we repeated the primary analysis (as well as secondary analyses described below) with further stratification by age at 65 years.

Subgroup analyses

We addressed whether certain subgroups of women markedly varied in depression risk with B-vitamin/folate treatment. Subgroups were based on baseline status of select factors of interest: i.e. age (<65 or ≥65 years), the other randomised treatments (yes/no), low intakes of B vitamins (yes/no), daily alcohol use (yes/no) and high medical comorbidity (Charlson <2 or ≥2 points). Regarding low nutrient intakes, we applied cut-offs as described in an earlier study: ^{Reference Kang, Cook, Manson, Buring, Albert and Grodstein30} low intake was defined as <1.9 mg/d for vitamin B₆ and as <279 μg/d for folate, using cut-offs based on intakes that were found to be significantly associated with elevated homocysteine among older adults; low intake was defined as <2.4 μg/d for vitamin B₁₂ by using the reference dietary intake for older persons, as we lacked similar information for B₁₂. We created an indicator for women with either a low intake of any one of the three B vitamins (n = 1287) v. adequate intakes of all three. In addition, we performed formal tests of effect modification using multiplicative interaction terms between the subgroup indicators and randomised assignment.

Sensitivity analyses

Sensitivity analyses were undertaken to address the robustness of findings. First, we utilised an alternative definition of depression that included as cases only participants with physician/clinician-diagnosed depression, but not those defined by depressive symptoms alone. Second, we conducted an adherence analysis in which women were censored on the date closest to when they stopped taking at least two-thirds of study pills, started to use outside (non-study) B-vitamin/folate supplements on ≥4 days per month, or were missing study pill adherence information. ^{Reference Albert, Cook, Gaziano, Zaharris, MacFadyen and Danielson13} Third, we used alternative definitions for folate intake (i.e. from food only, with fortification; from food only, without fortification; from food and supplements, without fortification) and for comorbidity (i.e. in the multivariable models, medical comorbidity burden was defined continuously as the number of points on the Charlson index, as opposed to being defined dichotomously as having a Charlson score of <2 or ≥2 points). This allowed us to assess whether results were sensitive to alternative definitions of these variables.

All statistical analyses were performed with SAS version 9.2 for Unix. Two-sided tests, with a significance level of α = 0.05 (P<0.05), were used. For Cox models, the proportional hazards assumption was confirmed analytically.