The impact of maternal diet fortification with lipid-based nutrient supplements on postpartum depression in rural Malawi: a randomised-controlled trial

Perinatal depression is highly prevalent in low-and-middle-income countries and has been linked to poor child health. Suboptimal maternal nutrition may be a risk factor for perinatal depression. In this randomised-controlled trial conducted in rural Malawi, we set out to test the hypothesis that women taking a fatty acid-rich lipid-based nutrient supplement (LNS) would have fewer depressive symptoms postpartum than those taking iron-folate (IFA) or multiple-micronutrient (MMN) capsules. Women were recruited from antenatal clinics and randomised to receive LNS or MMN during pregnancy and for 6 months postpartum, or IFA during pregnancy only. Maternal depressive symptoms were measured using validated translations of the Self Reporting Questionnaire (SRQ) and Edinburgh Postnatal Depression Scale (EPDS), antenatally (SRQ only) and at 6 months postpartum (SRQ and EPDS). Analysis was by modified intention to treat. One thousand three hundred and ninety one women were randomised (LNS = 462, MMN = 466, IFA = 463). The groups were similar across a range of baseline variables. At 6 months postpartum, 1078 (77.5%) had SRQ completed; mean (SD) scores were LNS 1.76(2.73), MMN 1.92(2.75), IFA 1.71(2.66), P = 0.541. One thousand and fifty seven (76.0%) had EPDS completed; mean (SD) scores were LNS 5.77(5.53), MMN 5.43(4.97), IFA 5.52(5.18), P = 0.676. There were no statistically significant differences between the groups on SRQ or EPDS scores (continuous or dichotomised) in unadjusted or adjusted models. In conclusion, fortification of maternal diet with LNS compared with MMN or IFA did not reduce postnatal depressive symptoms in this study.

The ILINS‐DYAD‐M trial was conducted in Mangochi District, a predominantly rural area situated at the southern end of Lake Malawi. Key economic activities in the district are subsistence farming, fishing and small‐scale business. The trial methodology is described in full elsewhere (Ashorn et al. 2015a). The trial is registered at the clinical trial registry at the National Institute of Health (USA) under identifier {“type”:”clinical-trial”,”attrs”:{“text”:”NCT01239693″,”term_id”:”NCT01239693″}}NCT01239693 (https://clinicaltrials.gov/ct2/show/{“type”:”clinical-trial”,”attrs”:{“text”:”NCT01239693″,”term_id”:”NCT01239693″}}NCT01239693). Participants were recruited from the population of women attending antenatal clinics in the government‐run Mangochi District Hospital, a part‐private hospital (Malindi) and two government‐run health centres (Lungwena and Namwera). Inclusion criteria were: pregnancy of no more than 20 completed gestation weeks (confirmed by ultrasound), being resident in the defined catchment area and available during the study period, and giving informed consent (signed or thumb print). Exclusion criteria were: age less than 15 years, a chronic health condition requiring regular medical attention, asthma (formally diagnosed and on treatment), a severe illness requiring referral to hospital or emergency medical care, peanut allergy, history of any serious allergic reaction, significant pregnancy complications at enrollment visit, previous recruitment to the trial (during a previous pregnancy) or current enrollment in another clinical trial. The enrolled women were randomly allocated to one of three study arms; the intervention arm (LNS) or one of two comparator arms (IFA and multiple micronutrients (MMN)). Women in the IFA comparator group received supplementation from enrollment to delivery with one capsule per day containing 60‐mg iron and 400‐µg folic acid, as is recommended in standard antenatal care in Malawi. Participants in the MMN control group received one capsule per day that contained IFA plus 16 additional micronutrients. The LNS daily dose (20 g) was designed to contain the same micronutrients as the MMN capsules, plus 4 additional minerals, protein and fat, optimised to provide high amounts of essential fatty acids thought to be important in pregnancy (Coletta et al. 2010). The fatty acids contained in the supplement were 0.59‐g alpha‐linolenic acid (ALA) (42% of RDA in pregnancy and 45% RDA in lactation) and 4.59 g of linoleic acid (LA) (35% of RDA in pregnancy/lactation). The daily LNS dose also provided 118 kcal of energy. The iron dose for participants in the MMN and LNS groups (20 mg/day) was lower than for those in the IFA group (60 mg), because the MMN and LNS supplementation was continued during the first 6 months postpartum, when the recommended iron intake for lactating women is much lower than the recommended antenatal dose (Arimond et al. 2013). Data collectors delivered supplements fortnightly to each participant. At each visit, the data collectors counted and recovered any supplement doses that were unused. During the trial period, new international guidance advised that LNS used in the management of acute childhood malnutrition be tested for the presence of Cronobacter sakazakii bacteria, with any untested or infected product being withdrawn. In response to this guidance, distribution of LNS to the iLiNS‐DYAD trial participants was suspended until testing had been completed. During this time (1 to 21 August 2012), 160 pregnant women in the LNS arm missed supplement for a period ranging from 1 to 20 days. Of these women, 127 were provided with IFA capsules instead; the other 33 were not located during the IFA distribution. The study participants attended antenatal and under‐5 clinics according to the same schedule as all other Malawian pregnant women and infants and received all normal preventive services provided by the national health system. Participants were refunded for any medical costs incurred during the study period. The IFA and MMN interventions were provided using double‐masked procedures: the capsules appeared identical, and neither participants nor the research team members were aware of the nutrient contents of the capsules. For the group receiving LNS (which was easily distinguished from the capsules), we used single‐masked procedures: the data collectors who administered the postnatal depressive symptom outcome measures were blind to group allocation and the participants were asked not to disclose information about which supplement they were taking. Researchers responsible for the data cleaning remained blind to the trial code until the database was fully cleaned. Randomisation code lists were generated by an independent statistician. At enrolment participants were randomised to trial arm using allocation codes sealed in opaque envelopes. Full details of the randomisation procedure are described in Ashorn et al. (2015a). Maternal depressive symptoms were measured using the Self Reporting Questionnaire (SRQ) and the Edinburgh Postnatal Depression Scale (EPDS). The SRQ was designed by the World Health Organisation as a screen for common mental disorders that could be used internationally and particularly in developing countries (WHO 1994). It consists of 20 questions with yes/no answers exploring symptoms of depression, anxiety and somatic manifestations of distress experienced over the previous 4 weeks. Scores are obtained by totaling the number of yes answers, with higher scores indicating higher number of depressive symptoms (possible score range 0–20). Unlike the SRQ, the EPDS was specifically designed for the postnatal period and excludes somatic items (sleep, appetite, energy or other bodily complaints) that might overlap with physical symptoms typical of the perinatal period (Cox et al. 1987). The EPDS consists of 10 questions asking about frequency of symptoms over the last 7 days, each answered from a choice of 4 options, scored 0–3. Scores are obtained by totaling the individual scores on the 10 items, with higher scores indicating higher number and frequency of depressive symptoms (possible score range 0–30). Both measures were validated in the local population (Stewart et al. 2013). In this study, we used SRQ and EPDS mean scores as primary outcome measures at 6 months postpartum (SRQ and EPDS). The SRQ and EPDS are best analysed as continuous variables to reduce loss of information (Altman & Royston 2006). We also analysed the proportion of women scoring above SRQ ≥ 8 and SRQ ≥ 5, and EPDS ≥13 and EPDS ≥9 cut‐offs. In the earlier validation study, at a cut‐off score of SRQ ≥8 (a cut‐off commonly chosen in previous studies (Harpham et al. 2003)), the SRQ Chichewa version had sensitivity 50.4%, specificity 88.4% and positive predictive value (PPV) of 41.2% for detection of Diagnostic and Statistical Manual of Mental Disorders 4th edition (DSM‐IV) major depressive episode (Stewart et al. 2013). In the same study, at a cut‐off score of EPDS ≥ 13 (the most commonly chosen cut‐off in previous studies (Cox et al. 2014)), the EPDS Chichewa version had sensitivity 33.7%, specificity 94.9% and PPV 50.0%. The test characteristics of the Chiyao translation of the SRQ and EPDS were similar (unpublished data). Data collectors were trained in the administration of the SRQ and EPDS by a trilingual clinical psychologist (EU) and given written instructions for later reference. At interview, any participant answering yes to the item about suicidal thoughts was asked further questions regarding suicidal ideation. Any participant reporting active or persistent suicidal ideas was referred to local mental health care services (nurse‐led outpatient clinics). During the study, no participants fulfilled this criterion. We measured the following baseline variables: maternal age, number of years of education completed, number of previous pregnancies, household ownership of a set of assets (combined into an index (with a mean of zero and standard deviation of one) using principal components analysis (Vyas & Kumaranayake 2006)), Household Food Insecurity Access (HFIA, a measure of food insecurity), mid‐upper arm circumference (MUAC), weight, height and BMI at recruitment, HIV status, malarial infection and haemoglobin at recruitment, season of enrolment (divided into quarters: Jan–Mar, Apr–Jun, Jul–Sept, Oct–Dec), gestational age at enrolment, antenatal SRQ score (done within 21 days of enrollment) and Multidimensional Scale of Perceived Social Support (MSPSS) score (a measure of the perception of the adequacy of support from others that was translated and locally validated (Stewart et al. 2014b)). The trial was conducted according to Good Clinical Practice guidelines and Helsinki Declaration standards. The study was approved by the College of Medicine Research and Ethics Committee, University of Malawi and the Ethics Committee of Pirkanmaa Hospital District, Finland. An independent data safety and monitoring board monitored the study for suspected serious adverse events and performed 2 interim analyses for safety. All presented analyses were pre‐specified either in the trial protocol or in the statistical analysis plan (http://www.ilins.org/ilins-project-research/data-analysis/iLiNS-DYAD-M%20Statistical%20Analysis%20Plan-%20version%2016.0%20with%20appendices%201-19-%202014-12-20.pdf/at_download/file). Analysis was conducted on the principle of modified intention to treat. All randomly allocated participants were included in the analyses, but participants with missing data on an outcome variable were excluded from the analysis of that outcome and two participants whose group allocation was incorrectly transcribed and assigned during enrollment were included in the group corresponding to the actual intervention they received. The outcome measures (SRQ and EPDS) were administered by data collectors during participant study clinic visits at 6 months postpartum. Data collectors made tracing home visits if a participant did not come for the scheduled visit within 14 days of the appointment. We used outcome data if measured between 22 and 34 weeks following delivery. Data collected outside of these limits were regarded as missing. Occasional missing item values on the SRQ (8 participants) and EPDS (11 participants) were imputed using mean substitution for the same scale and participant. Mean substitution was done if there were <50% missing data points. The sample size was calculated based on the primary study outcomes of birth size and growth at 18 months. Allowing for 20% missing values, with a sample size of 1400 participants the study had 80% power to detect an effect size of 0.24 (difference between groups, divided by the pooled SD) for each continuous outcome. For SRQ and EPDS, this is equivalent to 0.65 and 1.27 points, respectively. The group means and standard deviations for SRQ total and EPDS total at 6 month postpartum were tabulated by intervention group. The difference between the three groups was tested with ANOVA and null‐hypothesis of no difference between groups was rejected if P < 0.05. The proportions of women scoring SRQ ≥5 and ≥8 and EPDS ≥ 9 and ≥13 at 6 month postpartum were tabulated by intervention group. The difference between the three groups was tested with chi‐squared test, and null hypothesis of no differences between groups was tested with global null‐hypothesis rejected if P < 0.05. In order to adjust for covariates, for the primary continuous outcomes (SRQ total and EPDS total at 6 month postpartum) we constructed linear regression models including antenatal SRQ score, assets score, social support, height, BMI at enrolment, gestational age at enrollment, haemoglobin at enrollment, age, maternal education, number of previous pregnancies, season at enrollment, child sex and twin pregnancy. Categorical variables (study arm and season) were included as dummy variables. Study arm was analysed as two dummy variables, LNS vs. IFA and MMN vs. IFA; season was analysed as three dummy variables, Jan–March, April–June and July–September, with Oct–Dec as reference category. Each regression analysis used multiple imputation with 20 imputations per missing item of data on each covariate using the multivariate normal model based on the relevant dependent variable and all the covariates. We present the mean difference and 95% confidence intervals for the two group comparisons as described above. For the dichotomized outcome measures we adjusted for the same covariates and using multiple imputation for missing data on covariates as described above for the continuous outcome measures. We present odds ratio and 95% confidence intervals for the two group comparisons. We conducted the main analyses including women who had either singleton or twin deliveries. We repeated the analysis restricting to women who had singleton deliveries. We also conducted a sensitivity analysis that restricted the analysis to the most adherent participants (participants who received and did not return supplements for more than 80% of the antenatal follow‐up days). We tested for interaction between the intervention group and variables that could modify the effect of the nutritional intervention on depression outcomes, as per the analysis plan; these were antenatal SRQ score, assets score, social support, height, BMI at enrolment, gestational age at enrollment, haemoglobin at enrollment, age, maternal education, number of previous pregnancies, season at enrollment and child sex. If a statistically significant interaction (P < 0.1) was found, the adjusted analysis was completed as stratified by the respective predictor. All analyses were carried out using SPSS version 22, and Stata version 14.

N/A

• Dima Health