Self-perceived physical health predicts cardiovascular disease incidence and death among postmenopausal women

Details of eligibility criteria, data collection, and ascertainment of health outcomes in the WHI Study have been reported previously [15]. Briefly, the Clinical Trials (CT) included 68,132 post-menopausal women (ages 50–79 years) who were recruited at 40 clinical centers throughout the U.S. and randomly assigned to one or more clinical trials: the Dietary Modification (DM) trial, the Hormone Therapy (HT) trials, and/or the Calcium and Vitamin D (CaD) trial. Participants were ineligible if they had medical conditions with a predicted survival of 3 years or less, or had any condition that made them unable to adhere to study interventions (i.e. alcohol or drug-dependency, or dementia), or if they were enrolled in another trial. Women who were ineligible or unwilling to join either the HT or DM trials were invited to join an Observational Study (OS), which enrolled 93,676. Study procedures and protocols were approved by the Institutional Review Boards at the National Institute of Health and at all forty participating institutions; participants provided written informed consent (Additional file 1: Table S1).

These analyses were limited to the 26,515 control participants (all 3 trials) to ensure that the various trial interventions (i.e. hormone therapy, dietary modification, calcium and vitamin D) did not confound the study results. Of these, 5,965 were excluded from the analyses due to missing data at baseline: 242 participants did not fill out data collection forms, 1,271 had missing SF-36 values, and 4,694 were missing key covariate values in the statistical model. Complete data on exposure, outcomes, and specified covariates were available for the remaining 20,308 women (Figure 1).

At trial entry, participants completed the SF-36 HRQOL survey. The SF-36 assesses four physical subscales: (1) general health perceptions, (2) physical functioning (i.e. the ability to perform vigorous or moderate intensity activities that one might do in a typical day), (3) bodily pain, and (4) role limitations (i.e. problem with work and other regular daily activities due to physical health; and four mental subscales: (1) mental health, (2) vitality, (3) role limitations due to emotional problems, and (4) social functioning) [16, 17]. For each subscale, overall scores were obtained by summing responses to the individual questions linked to that particular subscale. Each subscale score ranged from 0 to 100, with 0 denoting ‘bad health’ and 100 denoting ‘good health’. These scales have been shown to be reliable and have adequate construct validity in populations with a variety of disease-specific medical diagnoses [18]. All eight subscales are included in the computation of the physical and the mental component summary scores (PCS and MCS respectively), but the weight of each individual subscale varies according to algorithms specified by the SF-36 survey developers [13, 16].

Standardized questionnaires administered at trial entry ascertained demographic characteristics including age, ethnicity (White, African-American, Asian or Pacific Islander, Hispanic/Latino, American Indian or Alaskan Native, and Other), education level (less than high-school, high-school, some college education, and college degree), marital status (never married, married, and divorced/widowed), employed (yes, no), drinking status (none, past, current), and smoking status (current, past, and non-smokers). Pack-years of cigarette smoking were computed by multiplying the midpoint of the smoking frequency interval with the midpoint of the duration interval and dividing the product by 20 [19]. Fruits and vegetables consumption was assessed with a standardized Food Frequency Questionnaire (FFQ) and categorized as <3, 3–5, and 5+ servings/day [20].

Weight and height were measured by clinic staff at baseline with the participants wearing light clothing and no shoes. Body mass index (BMI) was calculated as weight in kilograms divided by height in meters squared and was categorized as normal (< 25 kg/m²), overweight (25–29.9 kg/m²), or obese (≥30 kg/m²). Systolic and diastolic blood pressures were measured according to a standardized protocol and were expressed in millimeter of mercury.

Physical activity (PA) at baseline was assessed using a validated 9-item measure of physical activity from the ‘Personal Habits Questionnaire’ and was expressed in metabolic equivalent of task (MET)-minutes per week [21]. Participants were categorized as ‘sedentary’ (<30 minutes of physical activity weekly or <90 MET/wk), ‘insufficient PA’ (90 – 449 MET/wk), ‘meets the recommended level’ (450 – 600 MET/wk), or ‘exceed the recommended level’ (> 600 MET/wk).

Health related information came from the ‘Medical History Questionnaire’ administered at baseline. Participants were asked whether they had ever received a physician’s diagnosis of: diabetes, hypertension, cardiovascular disease, asthma, emphysema, cancer except non-melanoma skin cancer, Alzheimer’s disease, multiple sclerosis, and Parkinson’s disease; or whether they were on cholesterol-lowering medication.

Three WHI primary outcomes were selected: (1) time to incident CVD (fatal and non-fatal), (2) time to death due to CVD, and (3) time to death from any cause. A CVD event was defined as having been diagnosed with any of the following conditions during follow-up: coronary heart disease, stroke, congestive heart failure, angina, peripheral vascular disease, carotid artery disease, and coronary revascularization. The information on the outcomes was probed semi-annually by study personnel; any report of outcome was documented and corroborated with medical records. The underlying cause of death was classified on the basis of the death certificate, medical records, and other records such as autopsy reports.

The analysis focused on the WHI trial phase (1993–2005). For incident CVD, follow-up was censored at the last follow up visit, end of trial, or date of death due to any cause, whichever occurred first. For CVD-specific death, follow-up was censored at the last follow up visit, end of trial, or date of death due to any cause other than CVD, whichever occurred first. For all-cause death, follow-up was censored at last follow-up visit or end of trial, whichever occurred first.

Exposure and covariates were measured at baseline and the outcomes were measured at follow-up. A complete case analysis was employed. A comparison between those who were eligible and included in the analysis (N = 20,308) and those who were eligible but excluded in the analysis due to missing exposure or scientific model variables (N = 5,965) revealed several significant differences, when these two distributions were compared by Chi-square tests of association. Differences were not more than 2-3% for a given variable between the included and excluded participants. For example, 10.33% of those who were excluded died during follow up, compared to 9.06% of those who were included, with p-value 0.007.

For the ease of interpretation, PCS and MCS scores were categorized as quintiles. Correlation between PCS and MCS quintiles was evaluated with Spearman correlation statistics. Cumulative survival of the outcome events (CVD incidence, CVD death, and all-cause death) in each of the PCS and MCS quintiles were graphed with Kaplan Meier plots and were compared with log-rank test. Cox proportional hazard regressions were used to determine the unadjusted and the adjusted associations of PCS and MCS quintiles with each outcome. Hazard ratios (HR) and associated 95% confidence intervals (CI) were the measures of associations.

Interaction term between PCS and MCS for each outcome was tested prior to model building. A pre-specified set of variables, based on literature review, was selected for model adjustment [1, 8, 10, 11, 22]. Variables included participants’ demographic (age at screening, ethnicity, education, marital status), lifestyle (BMI, physical activity, fruit and vegetable consumption, smoking and drinking habits at study entry), and health characteristics [diabetes, hypertension, COPD (asthma or emphysema), cancer (except non-melanoma skin cancer), arthritis, medication use for high cholesterol, and systolic and diastolic blood pressure at baseline].

For each outcome, analyses were adjusted for confounding variables. Models examining the association of PCS and MCS on outcomes were sequentially adjusted for age at screening, then demographic, lifestyle and health characteristics, and finally for MCS (if PCS was the exposure) or PCS (if MCS was the exposure). A sensitivity analysis, where outcomes that occurred in the first year of follow-up were excluded, was also conducted. In addition, adjusted models were used to assess all eight subscales of SF-36 with the outcomes. The aforementioned covariates were used for the adjustment.

Participants with CVD at baseline were excluded from the incident CVD and CVD-specific death models. To examine the effect of CVD at baseline, the all-cause death model was run with and without participants who had CVD at baseline. The all-cause death model presented in the tables and graphs included those with CVD at baseline. All tests were two-sided and analyses were conducted in SAS version 9.2 (Cary, NC).

Women in the analytic sample (n=20,308) had a mean age of 62.8 years [standard deviation (SD): 6.9] and a mean BMI of 28.9 kg/m² (SD: 5.9); 37.2% were college graduates, 63.1% were married, and 38.6% were employed. Participants self-identified their race/ethnicity as: White = 82.3%, African-American = 9.9%, Hispanic/Latino = 3.5%, Asian/Pacific Islander = 2.3%, and American Indian/Alaskan Native = 0.4%.

Participants’ PCS and MCS score were strongly associated with demographic, life-style, and disease-related variables but the distributions across the quintiles differed (Tables 1 & 2). For example, whereas the mean age decreased incrementally across the quintiles of PCS, it increased across the quintiles of MCS. In the lowest PCS quintile (reference highest quintile), the proportion of participants with obesity was greater by 3 times, with diabetes by roughly 6 times, and with HTN, CVD, COPD, or arthritis by nearly 3 times. On the other hand, the proportions of participants with these conditions were similar between the lowest and the highest MCS quintiles (Tables 1 & 2). PCS and MCS quintiles were significantly correlated with one another (spearman correlation: -0.12, p-value <0.0001).

Kaplan-Meier plots of time to CVD incidence, CVD-specific death, or all-cause death were significant for both PCS and MCS. Unadjusted Cox model showed that risk estimates (for all 3 outcomes) increased in the comparison PCS quintiles and they decreased in the comparison MCS quintiles (compared to respective reference quintile, Figure 2). Interaction term between PCS and MCS was not significant for any of the outcomes (p values >0.25).

Age-adjustment substantially reduced the risk estimates of PCS quintiles; however, the HR remained significant for each comparison quintile (Q1-Q4) for CVD incidence and CVD death, and for the lowest 3 quintiles (Q1-Q3) for all cause death. Adjustment for the remaining covariates, except MCS, attenuated the risks further, and for CVD incidence the HR of Q4 and Q3 quintile lost statistical significance (Data not shown). Additional adjustment for MCS scores hardly made any difference to the risk estimates (Table 3). For all-cause death, the estimates in the fully adjusted model did not change when participants with CVD at baseline were excluded (Data not shown).

The risk estimates of the comparison MCS score quintiles changed in successive adjustment steps: first with age adjustment, then with all other selected covariates (except PCS), and finally with PCS scores. The HRs of final model for each outcome had a very modest U-shaped pattern of associations, although the comparison quintiles were not significant, except the 3rd quintile in relation to CVD incidence (Table 3).

When the events that occurred in the first year of the trial were excluded, participants’ PCS scores remained significantly associated with CVD incidence, CVD-specific death, and all-cause death. A comparison of HRs between the adjusted models, with or without the exclusions of events, indicated little change in the estimate and no change in the significance (data not shown).

Fully adjusted models showed that among the eight subscales of SF-36, physical functioning was significantly associated with all outcomes; each 10-point increase in the physical functioning score was associated with 7% reduction in CVD incidence, 13% reduction in CVD specific death, and 8% reduction in all-cause death. Additionally, general health perception was significantly associated with all-cause death. The other subscales were not significantly associated with any outcome (Additional file 2: Table S2).