Association between prenatal provision of lipid-based nutrient supplements and caesarean delivery: Findings from a randomised controlled trial in Malawi

In populations with a high prevalence of childhood and adolescent undernutrition, supplementation during pregnancy aiming at improving maternal nutritional status and preventing fetal growth restriction might theoretically lead to cephalopelvic disproportion and delivery complications. We investigated whether the prenatal provision of small-quantity lipid-based nutrient supplements (SQ-LNS) was associated with an increased risk of caesarean section (CS) or other delivery complications. Pregnant Malawian women were randomised to receive daily i) iron–folic acid (IFA) capsule (control), ii) multiple micronutrient (MMN) capsule of 18 micronutrients (second control), or iii) SQ-LNS with similar micronutrients as MMN, plus four minerals and macronutrients contributing 118 kcal. We analysed the associations of SQ-LNS, CS, and other delivery complications using log-binomial regressions. Among 1391 women enrolled, 1255 had delivery information available. The incidence of CS and delivery complications was 6.3% and 8.2%, respectively. The incidence of CS was 4.0%, 6.0%, and 8.9% (p = 0.017) in the IFA, MMN, and LNS groups, respectively. Compared to the IFA group, the relative risk (95% confidence interval) of CS was 2.2 (1.3–3.8) (p = 0.006) in the LNS group and 1.5 (0.8–2.7) (p = 0.200) in the MMN group. We found no significant differences for other delivery complications. Provision of SQ-LNS to pregnant women may have increased the incidence of CS. The baseline rate was, however, lower than recommended. It is unclear if the higher CS incidence in the SQ-LNS group resulted from increased obstetric needs or more active health seeking and a better supply of services. Trial registered at clinicaltrials.gov, NCT01239693. This was a secondary analysis of data prospectively collected as part of a dietary intervention trial, iLiNS‐DYAD‐M, in Malawi (ClinicalTrials.gov, Identifier {“type”:”clinical-trial”,”attrs”:{“text”:”NCT01239693″,”term_id”:”NCT01239693″}}NCT01239693), in which mother–child pairs in the intervention group received SQ‐LNS whereas mothers in the control groups received either IFA or MMN. The main outcomes of the study included birth weight, newborn length, and length for age Z‐score (LAZ). In the current study, we analysed the association between maternal SQ‐LNS supplementation, compared to IFA and MMN, and the incidence of CS and other delivery complications. The enrolment in the study took place in one public district hospital (Mangochi), one semiprivate hospital (Malindi), and two public health centres (Lungwena and Namwera) in Mangochi district, Southern Malawi. In total, the clinics provided health care to approximately 190,000 people. Recruitment for the trial was open to pregnant women who came for antenatal care at any of the clinics and met the following criteria: ultrasound‐confirmed pregnancy of under 20 completed gestation weeks, at least 15 years of age, and without any chronic health conditions. Enrolled participants were randomly assigned into three groups that were provided with daily nutrient supplements. Women in the first control group, the IFA group, received standard Malawian antenatal care, including supplementation with micronutrient capsules containing 60 mg iron and 400 μg folic acid. Women in the second control, the MMN group, received capsules that contained IFA and 16 additional micronutrients. MMN was chosen as the second control because of the benefits it might have over IFA (Smith et al., 2017). Participants in the intervention group, the LNS group, received 20 g SQ‐LNS sachets containing 118 kcal, protein, carbohydrates, essential fatty acids, sucrose, and 22 micronutrients. IFA and MMN looked and tasted identical, but the SQ‐LNS sachets looked different from the control supplements. Data collectors delivered 15 supplement doses (IFA or MMN capsules or LNS sachets) fortnightly to each participant, at their home, until delivery. There was no direct observation of the consumption of the supplements. As a measure of compliance to the interventions, at each visit, the data collectors collected any leftover supplements or empty packaging from the participants. The mean adherence to the intervention (proportion of days when the supplements were consumed) was comparable and higher than 80% in all three groups (Ashorn et al., 2015). All three groups also received intermittent preventive malaria treatment. Details of the interventions can be found elsewhere (Ashorn et al., 2015). Participants were enroled between 14 and 20 gestation weeks. At the enrolment visit, trained anthropometrists measured the participating women’s weight, height, and mid‐upper arm circumference. Research nurses assessed the duration of pregnancy by measuring fetal biparietal diameter, femur length, and abdominal circumference with ultrasound imagers that used inbuilt Hadlock tables to estimate the duration of gestation. The same nurses measured the women’s peripheral blood malaria parasitemia with rapid tests (Clearview Malaria Combo; British Biocell International Ltd.) and haemoglobin concentration with a finger prick. Health facility nurses tested for HIV infection in all participants, except for those who opted out or were already known to be HIV infected, by using a whole‐blood antibody rapid test (Alere Determine HIV‐1/2; Alere Medical Co., Ltd.). All participants were invited for follow‐up visits at the study clinic at 32 and 36 gestational weeks. During these visits, standardised obstetric examinations were conducted and anthropometric measurements were taken again to examine maternal weight gain during pregnancy. The mean maternal weight gain during the second and third trimesters of pregnancy was comparable between all three groups (Ashorn et al., 2017). The delivery information was collected by a clinic data collector (trained study nurses, laboratory technicians, study monitor, or study coordinator) either at the clinic or at home within 48 h after delivery (the newborn visit). A clinic data collector filled the delivery information form based on the health passport and delivery charts. We defined CS either as a planned CS or an emergency CS. In a planned CS, the woman was informed during the antenatal period that she would have to deliver by CS due to identified complications that would make vaginal delivery unsafe. In an emergency CS, the decision of the procedure was made immediately before or during labour because of a life‐threatening situation either to the mother or the child. Any delivery complication was defined as a condition of a planned CS, emergency CS, vacuum extraction, prolonged labour, large perineal tear, or symphysiotomy. The child’s length, weight, and head circumference were measured at the first clinic visit after the birth (the postnatal visit) by trained anthropometrists. We considered newborn anthropometric measurements missing if they were collected more than 6 weeks after delivery. We calculated age‐ and sex‐standardised anthropometric indices (Z‐scores) by using the World Health Organisation (WHO) Child Growth Standards (WHO Multicentre Growth Reference Study Group 2006). We calculated the duration of pregnancy by adding the time interval between enrolment and delivery determined by ultrasound gestational age at enrolment. The sample size was originally calculated in accordance with the main objective of the iLiNS‐ DYAD‐M trial (Ashorn et al., 2015) and was based on an assumption of an effect size of at least 0.3 (difference between groups, divided by the pooled SD) for each continuous outcome, a power of 80%, and a two‐sided type I error rate of 5%. We carried out the statistical analyses with Stata 15.1 (StataCorp) based on the analysis plan written and published at ilins.org. We based the analysis on the principle of intention to treat. We excluded twin pregnancies and abortions from the analyses. We estimated the incidence of delivery complications in the three groups and calculated relative risks (RR) for the comparison of binary endpoints. To prevent inflated type I errors caused by multiple comparisons, we used a closed testing procedure. Null hypotheses for pairwise comparisons could only be rejected if the global null hypotheses of all three groups being identical had also been rejected (Cheung, 2013). We tested the global null hypotheses for binary endpoints either with Fisher’s exact test or the log‐binomial regression model. We tested quantitative endpoints with an analysis of variance. With the log‐binomial regression models for the binary endpoints, we used a set of Newton–Raphson maximisation of the log‐likelihood. If the algorithm failed to converge in the estimation, we used an alternative estimation algorithm with iterated reweighted least squares (Zou, 2004). With the same setting, we calculated RRs from bivariate analysis for single variables and we created cumulative stepwise multivariate log‐binomial regression models for an association attenuation analysis. For the first multivariable model, we included the child’s sex, gestational age, and variables with p < 0.05 from bivariate analysis, excluding the child's anthropometric measurements. For the second multivariable model, we included variables that were considered intermediate outcomes (i.e., maternal weekly gestational weight gain, LAZ, weight‐for‐age Z‐score [WLZ], and head circumference Z‐score [HCZ]) with variables from the first multivariable model. We performed likelihood ratio tests for the interaction between intervention and maternal characteristics. Maternal variables were specified in the statistical analysis plan before data analysis. Variables that were tested for interaction and analysed as stratified included maternal age, education, number of previous pregnancies, height, body mass index (BMI), alpha‐1‐acid glycoprotein, C‐reactive protein (CRP), HIV, peripheral blood malaria parasitemia, and anaemia at enrolment as well as the season of enrolment, gestational age at enrolment, food insecurity status, and child's sex. We provided stratified analyses in case of positive interaction (p < 0.10) or if either of the stratified comparisons of binary endpoints resulted in p < 0.05 and thus suggested a difference between intervention groups within a stratified subgroup. For the final analyses, each analysis was adjusted to the site of enrolment (i.e., hospitals and health clinics, to control for access to health services) and to other variables that were included in the provided stratified analyses (to control for participants with multiple classifications on the selected variables).