This study utilized data from the Nepal Nutritional Intervention Project – Sarlahi, carried out in the rural Sarlahi District of Nepal from 1999 to 2001, that consisted of a cluster-randomized controlled trial of micronutrient supplementation provided to pregnant women from early pregnancy through 3 months post-partum. Sarlahi is a poor district within the Terai region of Nepal, dependent upon a rural subsistence economy. The main results have been described previously (Christian et al. 2003a
). In this trial, 426 sectors of 30 village development communities were assigned daily vitamin A alone (control) or with folic acid, iron–folic acid, folic acid–iron–zinc, or a multiple micronutrient supplement containing the former plus an additional 11 vitamins and minerals. The overall nutritional status of women in this population is poor. The women were stunted (with an average height of 150.5 cm) and thin (with an average BMI of 19.3 kg m−2
), and also suffer from multiple concurrent micronutrient deficiencies in early pregnancy (Jiang et al. 2005
To enrol pregnant women early in pregnancy, female project workers visited all married women of reproductive age. Those who were currently pregnant, breastfeeding a baby <9 months old, menopausal, sterilized or widowed were excluded. Thereafter, women were visited every 5 weeks and were asked about menses since the last visit. If the woman said that she had not menstruated since the last visit, she was offered a urine-based pregnancy test (human chorionic gonadotrophin; Clue, Orchid Biomedical Systems, Goa, India). If the woman was found to be pregnant, she was invited to participate in the study and, upon consent, was provided supplements and visited weekly. Pregnant women were interviewed at baseline, during the third trimester, at birth, and at 6 weeks postpartum.
Women were interviewed at the time of enrolment to ascertain age, pregnancy history, date of last menstrual period, alcohol and tobacco use, diet, socioeconomic status, ethnic group (Pahadi or Madeshi), caste, educational level and morbidity history. Maternal age was ascertained by questionnaire. If there was uncertainty, local and national event calendars were used to estimate the year of birth and the data were checked for digit preference. Height, weight and mid-upper arm circumference (MUAC) were also measured at baseline, and weight and MUAC were repeated at the third-trimester and 6-week postpartum interviews. Pregnant women reporting to be ≤25 years old and parity 0 (no previous live birth) or parity 1 (one previous live birth) were included in this analysis.
Most women delivered their babies at home, assisted by a traditional birth attendant, relative or neighbour. Once a birth was reported, an anthropometrist visited the home and recorded infant weight using a digital scale (Seca 727, Hamburg, Germany), recumbent length using a length board (Shorr Infant Measuring Board, Shorr Productions, Rhode Island, USA), and head and chest circumference using an insertion tape (Ross Laboratories, Columbus, OH, USA). All measurements, except weight, were recorded in triplicate and the median was used for analysis. Only liveborn singleton infants who were measured within 72 h after birth were included in this analysis.
Gestational age was calculated from the reported first date of the last menstrual period, obtained at the baseline interview, and then checked against the week of the positive pregnancy test and prospectively collected histories of menstruation. If the date of the last menstrual period was not known, the midpoint between the previous reported menstrual period and the date of the positive pregnancy test was used. Pre-term was defined as delivery before 37 weeks of gestation. No clinical parameters were used to determine gestational age. SGA infants were defined as those whose weight was below the 10th percentile of the gestational age- and sex-specific US reference for fetal growth (Alexander et al. 1996
). LBW was defined as <2.5 kg. Ponderal index (kg/m3
) was calculated by using birthweight and length measurements.
The primary outcome of the original trial was birth-weight, and it was observed that the treatment arms had a differential effect on birthweight. The iron plus folic acid and the multiple micronutrient groups reduced the incidence of LBW by 16% (95% CI = 0.72−0.99) and 14% (95% CI = 0.72−0.99), respectively (Christian et al. 2003a
). Maternal age, parity, ethnicity and other characteristics potentially associated with birthweight were found to be comparable across treatment groups. There also appeared to be no difference in the relationship between maternal age and birthweight within the different treatment arms, and thus all treatment groups were pooled for this analysis. To determine the association between maternal age and the risk of adverse outcomes, we first examined the linear associations between age and the continuous variables of birthweight and length, head and chest circumference, ponderal index and gestational age. Scatter plots with lowess curves, which employ a locally weighted regression smoothing method, were used to examine the shapes of the distributions. Crude and adjusted estimates were calculated using linear regression. Logistic regression was used to calculate crude and adjusted odds ratios for risk of LBW, SGA or preterm delivery by maternal age. P
-values and confidence intervals were calculated using robust variance estimation, accounting for the potential clustering of covariates within sectors. The interaction between age and parity was examined by including an interaction term in the regression models and testing its significance. When the interaction terms were found to be significant, stratified models are presented to show the different relationships between parity 0 and 1. Potential confounders were included in the models if they were significantly (P
< 0.05) associated with both maternal age and either birthweight or gestational age in bivariate analyses. Regression models were checked by examining plots of leverage versus the squared residual. Sensitivity analyses were conducted by removing points with high leverage or residuals and comparing the new model with the original.
To determine if there was a threshold age beyond which the risk of adverse outcomes substantially decreased among parity 0 women, a logistic regression model was fit with preterm delivery as a function of six categories of maternal age. Confounders were included as either continuous, as in the case of maternal height and BMI in early pregnancy, or binary for smoking, literacy and ethnicity. The regression equation from this model was used to estimate the probability of preterm delivery for each of the six separate age categories using the regression coefficients and the means of the confounders for all parity 0 women in the dataset. In the case of the binary variables, the ‘mean’ was the proportion of 1’s for that variable. For example, literacy was coded as 0 if illiterate and 1 if literate, so the ‘mean’ in the equation was the proportion of all the women who were literate. For the continuous variables, the mean was the actual mean for all age groups of women. These estimates were then plotted to examine the differences in incidence of preterm delivery by maternal age group. Data were analysed using STATA version 8 (StataCorp LP, College Station, TX, USA).
The study was approved by the National Health Research Council of the Ministry of Health of Nepal and the Committee for Human Research at the Johns Hopkins Bloomberg School of Public Health. Informed consent was obtained from the participants.