This study utilizes cross-sectional data from a follow-up of women who had been enrolled in a double blind, placebo controlled, cluster randomized trial of vitamin A or β carotene supplementation provided to women for three years (1994–1997). Details of the trial were published previously . Briefly, the study was conducted in the rural, low-lying Sarlahi District of Nepal. The study area comprises 30 Village Development Communities (VDCs), each divided into 9 administrative wards. Over a study period from July 1994 to June 1997, 17,531 infants were born to women enrolled in the trial. Women were supplemented before, during and after pregnancy throughout the study period. A subsample area was selected for more intensive monitoring during the trial. The area contained roughly 10% of the study population and selection was based on ease of access to clinics and a paved road in order to facilitate biochemical sample collection and to visit women at birth. The cohort of women who became pregnant during the trial period have been followed over time along with their children to monitor long-term health outcomes . From 2006–2008, women were followed-up (n = 16,469) during a series of up to three household visits. The current analysis uses the data from this follow-up period (n = 15,934), excluding women who were currently pregnant (n = 470) or who did not know their pregnancy status (n = 65).
During the first visit, field workers conducted interviews to collect information on household socioeconomic status, literacy, ethnicity and occupation. Information on their smoking status and alcohol consumption were also collected. Blood pressure was measured four times on the right arm using an automated measurement device (BPM 300, BPTrue, Canada). The first measure was dropped and the mean of the last three was used for analysis. If the mean systolic or diastolic blood pressure (SBP and DBP, respectively) were ≥ 140/90, the measurements were repeated. Mid-upper arm circumference (MUAC) was measured on the upper left arm at the mid-point of the acromion process and the tip of the olecranon using a standard insertion tape.
During a second visit among a subsample of 1679 women, anthropometric measurements were recorded, including height (Harpenden stadiometer, Crosswell, UK), weight (Seca 881, Hamburg, Germany), and waist circumference (WC) (Seca 200, Hamburg, Germany). Women were then asked to fast overnight and were visited the following morning by a team of phlebotomists to collect venous blood. Total cholesterol (TC), HDL cholesterol, triglycerides (TG), and fasting glucose were measured in plasma specimens (LDX Analyzer, Cholestech, Hayward, CA). Of the women who consented to the blood draw (n = 1447), 251 (17.3%) had not fasted, defined as no food or drink other than water within 8 hours of the blood draw. Glucose data was only analyzed among women who were fasting (n = 1196).
Continuous variables such as age, SBP, DBP, MUAC, height, weight, BMI and WC were checked for normality, outliers and missing values. Blood pressure and lipid parameters were categorized using standard cutoffs [12, 13]. Hypertension was present if SBP and/or DBP were ≥140/90 mm Hg . Pre-hypertension was defined by SBP ≥120 mmHg but <140 mmHg and/or DBP ≥ 80 mmHg but <90 mmHg . The cutoff for high TC was ≥5.17 mmol/L (200 mg/dL); high TG was ≥1.7 mmol/L (150 mg/dL); low HDL cholesterol was <1.03 mmol/L (40 mg/dL), raised fasting glucose was ≥5.6 mmol/L (100 mg/dL) and high HbA1c was ≥6.5% . Overweight was defined as BMI ≥23 kg/m2 and abdominal obesity defined as WC ≥80 cm [15–17]. Women were also classified according to the International Diabetes Federation’s worldwide definition of the metabolic syndrome which includes a WC ≥80 cm plus ≥ 2 of the followings: 1) raised triglycerides: ≥1.7 mmol/L; 2) reduced HDL cholesterol: <1.03 mmol/L; 3) raised blood pressure: systolic BP ≥130 or diastolic BP ≥85 mm Hg; and 4) raised fasting plasma glucose: ≥5.6 mmol/L .
SES was analyzed using Principal Component Analysis (PCA) by examining 17 questions on SES, extracting three components with eigen values >1 which explained 41% of the variance. One component represented a summary measure of household quality and status, with the most heavily weighted variables including type of roof and walls, type of latrines, number of servants in the house, number of rooms in the house, having electricity, and owning a motorcycle. The second component represented household farming assets and food storage, with the most heavily weighted variables being kilograms of rice stored per household member, duration of time rice stores would last, ownership of cultivable land, and ownership of cattle, goats and bullock carts. The third component represented other household assets, with the most heavily weighted variables including bicycle, TV, radio and wristwatch ownership. Each of these three components of SES were categorized into tertiles based on their component scores.
The primary outcome measure for this analysis was blood pressure, evaluated both as a dichotomous and a continuous variable. The mean differences of BP among supplementation groups (vitamin A, β carotene or placebo), were within 0.5 mm Hg and the prevalence of hypertension and pre-hypertension did not differ between groups. Thus, all intervention groups were combined for this analysis. For the full sample, simple and multiple logistic regression analyses were done to study the relationship between each independent variable with the outcomes of hypertension or pre-hypertension. The independent variables included the three SES factors, age, smoking status (yes/no), alcohol consumption (yes/no), MUAC (in cm), ethnicity (women of Pahadi ethnicity historically originated from the hill areas of Nepal while those of Madheshi ethnicity originated from the plains) and occupation, which was grouped into two categories: 1) no reported work or study or 2) work or study outside home, including farmers, unskilled or contracted laborers, business, private or government service, and students. For both the regression models, effect modification by ethnicity and age were evaluated with likelihood ratio tests by comparing two nested multivariate models with and without the interaction term. The results were stratified if any significant interaction was found.
For the subsample, blood pressure was analyzed continuously, as the sample size was smaller and the number of women with hypertension was relatively low. In order to examine the associations of SBP and DBP with other cardiovascular risk factors (high TC, high TG, low HDL cholesterol, raised fasting glucose, high HbA1c, high BMI, high WC), simple linear regression was performed with SBP or DBP as the outcome variable and each of these risk factors as predictors. Those with normal values of the predictor variables were considered as a reference group. Multivariable regression models were adjusted for other predictors found to be significantly associated with hypertension in the full cohort. In addition, models examining the lipid profile, glucose and HbA1c were further adjusted for BMI.
The follow-up study was approved by the Institutional Review Boards at the Johns Hopkins Bloomberg School of Public Health and the Institute of Medicine in Kathmandu, Nepal.