Study population
We used data from a randomized clinical trial conducted in urban Tanzania. Details of this study have been described elsewhere [24, 25]. Briefly, from September 2010 to October 2012, a randomized trial on iron supplements was conducted in Dar es Salaam, Tanzania. Participants were screened and enrolled at antenatal care clinics. Women were eligible if they were iron-replete, non-anemic, HIV-uninfected, primigravidae or secundigravidae, and were recruited at or before 27 weeks of gestation. Baseline gestational age (weeks) was estimated based on the reported timing of the last menstrual period (LMP). The study enrolled 1,500 pregnant women who were subsequently randomized to receive a daily oral dose of either 60mg of iron or placebo from the time of enrollment until delivery.
At baseline, women completed a sociodemographic and reproductive health questionnaire, as well as a full clinical examination. They were subsequently followed at monthly antenatal visits and at time of delivery. For our study, we excluded participants with unknown gestational age at delivery (n = 22), unknown delivery outcomes (n = 15), or twin babies (n = 27). Since GWG in the second and third trimesters was the main exposure of interest, we further excluded women with only one weight measure during that time window (n = 206), leaving us with a final study sample of 1,230 participants.
Assessment and characterization of GWG
Study participants’ weight at baseline and at monthly follow-up visits was measured by trained study nurses using a calibrated weight scale. Pre-pregnancy BMI has been suggested as an important covariate for the association between GWG and pregnancy outcomes [13]. However, information on pre-pregnancy weight was not collected in the original trial study. Further, given the distribution of the baseline gestational age at enrollment, only 196 out of 1,230 participants were enrolled during the first trimester, of which the majority were enrolled in the late first trimester (interquartile range of gestational age at enrollment among participants enrolled in the first trimester: 10–13 weeks). We therefore imputed BMI at the end of the first trimester (14 weeks of gestation) for covariate adjustment and stratification. Based on the repeated weight measurements during pregnancy, we fit mixed-effects models with polynomial terms of gestational age (weeks) and imputed individual-specific weight at 14 weeks of gestation; statistical results suggested good imputation performance (mean absolute error: 1.95 kg, concordance rate of categorical BMI among women with available first-trimester weight: 89.0%). Details on the statistical methods and the imputation results can be found elsewhere [25]. Based on the imputed weight at the end of the first trimester and the height measured at baseline, the corresponding BMI status at the end of the first trimester was derived (underweight if BMI < 18.5 kg/m2, normal if 18.5 kg/m2 ≤ BMI < 25 kg/m2, overweight if 25 kg/m2 ≤ BMI < 30 kg/m2, and obese if BMI ≥ 30 kg/m2).
The main exposure of interest was GWG during the second and third trimesters. We defined degree of appropriate GWG based on the 2009 Institute of Medicine (IOM) guidelines [13]. The IOM guidelines provided recommended ranges for total weight gain and the rate of weight gain during the second and third trimesters, based on pre-pregnancy BMI status. Weekly rate of GWG during the second and third trimesters (kg/week) was derived by calculating the difference between the first measured weight in the second trimester and the last measured weight before delivery and dividing that by the number of weeks between the two measures. For each given participant, based on the calculated weekly rate of GWG, the BMI status at the end of the first trimester, and the IOM recommended GWG range (0.44–0.58 kg/week for underweight, 0.35–0.50 kg/week for normal weight, 0.23–0.33 kg/week for overweight, and 0.17–0.27 kg/week for obese), GWG was characterized as inadequate (weekly rate of GWG below the recommended range), adequate (weekly rate of GWG within the recommended range), or excessive (weekly rate of GWG above the recommended range) [13]. We made assumptions that weight gain during the first trimester was minimal and that women stayed in the same BMI category from the start of the pregnancy until the end of the first trimester [13].
We additionally characterized GWG using other metrics, including percentage of GWG adequacy (i.e. percentage method) [26] and GWG z-score (i.e., z-score method) based on the INTERGROWTH-21st standard [27]. Building upon the IOM guidelines which grouped the extent of GWG into three categories (i.e., inadequate, adequate, and excessive GWG), the percentage method provided a percentage value to further quantify the amount of GWG relative to the guidelines with accounting for the pregnancy duration. Details on this method has been described elsewhere [26]. Briefly, percentage adequacy of GWG was calculated as the ratio of observed weight gain (kg) and expected weight gain (kg) during pregnancy. The original formula was given as follows: percent adequacy = observed weight gain during pregnancy / [expected first trimester weight gain+((week at the last weight measure – 13)*expected weekly GWG rate in the second and third trimesters)]. Given the research question of our study, we modified the formula by restricting the time period to the second and third trimesters instead of the entire pregnancy. Consistent with Adu-Afarwuah et al, we derived BMI-specific percentage cutoffs based on the IOM cutoffs and classified the GWG into three groups based on the calculated percent adequacy: inadequate, adequate, and excessive, respectively [26].
We further constructed a GWG z-score (in unit of standard deviation) for participants with a normal BMI (18.5 kg/m2 ≤ BMI < 25 kg/m2) at the end of the first trimester, using a standard reference chart developed by the INTERGROWTH-21st consortium [27]. Briefly, by applying the reference chart, a gestational age-specific GWG z-score can be derived based on the total weight (kg) gained up to a given gestational age. Since GWG was likely to follow a non-linear trajectory over the course of pregnancy, a gestational age-specific z-score could account for the natural correlation between a longer pregnancy duration and a higher rate of GWG [28], which, if unaddressed, could bias the association between GWG and gestational age-related outcome (e.g., prematurity). For our analysis, total weight gain in the second and third trimesters and gestational age at the last weight measure were used to derive the z-score. Given the potential non-linearity of the z-score with respect to risks of pregnancy outcomes and the distribution of the z-scores in our sample (only one participant had z-score > 2 units), we classified participants into one of the two following groups: inadequate GWG if z-score < -2 units (2.3th percentile), adequate GWG if z-score within +/-2 units (between 2.3th and 97.7th percentile).
Outcome assessment
At the time of delivery, on-site midwives recorded participants’ pregnancy outcomes. Data on gestational age at delivery (weeks), delivery outcome if known (miscarriage [n = 1], stillbirth [n = 47], and live birth [n = 1,182]), infant sex, and infant birth weight (kg) were determined. We derived the following outcome variables for pregnancies resulting in live births: low birth weight (LBW, birthweight < 2.5kg), preterm birth (gestational age at delivery < 37 weeks), small for gestational age and large for gestational age (SGA and LGA, gender-specific birth weight below the 10th percentile and above the 90th percentile respectively for babies of the same gestational age according to the INTERGROWTH-21st reference) [29]. Although we did not have information on the type of preterm birth (i.e., spontaneous, medically induced), we considered most of the preterm cases as spontaneous, based on conversations with on-site research staff and medically induced preterm birth being relatively uncommon in Tanzania.
Statistical analysis
In the main analyses, GWG during the second and third trimesters according to the IOM recommendations was evaluated with respect to adverse birth outcomes. GWG with three levels defined by the IOM guidelines (i.e., inadequate, adequate, and excessive GWG) was modeled as a categorical variable, and the group with adequate GWG was set as the reference group. The following dichotomous outcomes were examined: LBW, preterm birth, SGA, and LGA. We used multivariable Poisson regression with a sandwich variance estimator to estimate risk ratio (RR) and 95% confidence interval (CI) [30]. We adjusted for covariates hypothesized a priori as potential confounders in the analyses, including age, baseline gestational age, gestational age at delivery, measured or imputed BMI at 14 weeks of gestation, primigravida status, treatment status, marital status, education, occupation, and history of prior complications (history of cardiovascular disease, high blood pressure, diabetes, or weight loss in previous year, or ever had a LBW baby or non-live birth among non-primigravida).
Given the evidence of the heterogeneity by pre-pregnancy BMI status for the associations of interest [13], we further stratified the analyses by BMI status at the end of the first trimester. Due to the limited sample size, we did not examine these associations among underweight women (n = 72) and examined the questions among women with normal BMI (18.5 kg/m2 ≤ BMI < 25 kg/m2; n = 756) and overweight or obese women (BMI ≥ 25 kg/m2; n = 402 [295 and 107 for overweight and obese, respectively]), separately. Heterogeneity was evaluated by the statistical significance of the cross-product term between categorical GWG and BMI status in the analysis sample excluding underweight women.
In the sensitivity analyses, we additionally examined appropriate GWG and birth outcomes, using the percentage and the z-score methods. For the z-score method, since the INTERGROWTH-21st GWG reference chart is currently available only for women with normal pre-pregnancy BMI, we restricted the analyses to participants with normal BMI at the end of the first trimester (n = 755). All analyses were conducted using SAS statistical software (version 9.4; SAS Institute Inc, Cary, NC, USA). All statistical tests were 2-sided, with p-value less than 0.05 considered as statistically significant.