Associations of maternal nutrition during pregnancy and post-partum with maternal cognition and caregiving

Pregnant and post-partum women require increased nutrient intake and optimal cognition, which depends on adequate nutrition, to enable reasoning and learning for caregiving. We aimed to assess (a) differences in maternal cognition and caregiving between women in Malawi who received different nutritional supplements, (b) 14 effect modifiers, and (c) associations of cognition and caregiving with biomarkers of iron, Vitamin A, B-vitamin, and fatty acid status. In a randomized controlled trial (n = 869), pregnant women daily received either multiple micronutrients (MMN), 20 g/day lipid-based nutrient supplements (LNS), or a control iron/folic acid (IFA) tablet. After delivery, supplementation continued in the MMN and LNS arms, and the IFA control group received placebo until 6 months post-partum, when cognition (n = 712), caregiving behaviour (n = 669), and biomarkers of nutritional status (n = 283) were assessed. In the full group, only one difference was significant: the IFA arm scored 0.22 SD (95% CI [0.01, 0.39], p =.03) higher than the LNS arm in mental rotation. Among subgroups of women with baseline low hemoglobin, poor iron status, or malaria, those who received LNS scored 0.4 to 0.7 SD higher than the IFA arm in verbal fluency. Breastmilk docosahexaenoic acid and Vitamin B12 concentrations were positively associated with verbal fluency and digit span forward (adjusting for covariates ps <.05). In this population in Malawi, maternal supplementation with MMN or LNS did not positively affect maternal cognition or caregiving. Maternal docosahexaenoic acid and B12 status may be important for post-partum attention and executive function. We conducted an add‐on study of maternal cognition and caregiving in a randomized trial described in more detail by Ashorn et al. (Ashorn et al., 2015), designed to assess the effect of maternal and infant LNS on infant growth. Pregnant women (n = 869) who attended antenatal care at two hospitals and one health center in Mangochi district, Malawi, were enrolled from February 2011 to March 2012 and assigned to one of three intervention arms, described below. Details of randomization and inclusion criteria have been published previously (Ashorn et al., 2015). All participants provided informed consent. Ethical approval for the study procedures was obtained from the Ethics Committees at the University of Malawi College of Medicine and Tampere University Hospital District, Finland. The study was registered with the U.S. National Institutes of Health as a clinical trial (http://www.clinicaltrials.gov; {"type":"clinical-trial","attrs":{"text":"NCT01239693","term_id":"NCT01239693"}}NCT01239693). The sample size of 290 per group, allowing for 20% attrition, provided 83% power to detect a difference of 0.3 SD in continuous scores between groups, with alpha at 0.05 (Zhao & Li, 2012). At enrollment, which occurred at ≤20 weeks gestation, women were randomly assigned to one of three intervention arms; all of whom received two doses of intermittent preventive malaria treatment during pregnancy. The IFA arm received standard antenatal care, including supplementation from enrollment to delivery with one capsule per day containing 60 mg iron and 400 μg folic acid. The IFA arm received a placebo tablet containing 200 mg calcium from delivery to 6 months post‐partum. The MMN arm received one capsule per day from enrollment to 6 months post‐partum that contained a lower dose of iron (20 mg), 400 μg folic acid, and 16 additional micronutrients, shown in Table 1. The LNS arm received daily 20‐g sachets of small quantity LNS produced by Nutriset SAS (Malaunay, France) from enrollment to 6 months post‐partum, containing the same micronutrients as the MMN capsules, four additional minerals, protein, fat, and essential fatty acids (Table 1). Nutrient and energy contents of the dietary supplements Note. IFA = iron/folic acid; MMN = multiple micronutrients; LNS = lipid‐based nutrient supplements. The iron dose was lower for participants in the MMN and LNS arms (20 mg/day) than for those in the IFA arm (60 mg/day), because supplementation with MMN and LNS continued during the first 6 months post‐partum, when the recommended iron intake for lactating women is much lower than the standard antenatal dose (Arimond et al., 2015). On the basis of a literature review and our estimates of the normal dietary iron intakes among pregnant women in the study area, we considered 20 mg/day a safe and adequate dose to prevent iron deficiency anemia during pregnancy, even for women who were iron deficient at enrollment (Arimond et al., 2015). At the time of enrollment, data collectors recorded sociodemographic and maternal anthropometric data. Research nurses assessed the duration of pregnancy with ultrasonography and measured the women's peripheral blood malaria parasitemia and HIV infection with rapid tests. Maternal hemoglobin concentration (Hb; g/dl) was determined using on‐site cuvette readers (HemoCue AB; Angelholm), and zinc protoporphyrin concentration (ZPP; μmol/mol heme) was determined from venous blood samples using a hematofluorometer (Aviv Biomedical Co. NJ, USA), after red blood cells were washed three times with normal saline. Plasma soluble transferrin receptor (sTfR; mg/L) was determined from plasma by immunoturbidimetry on the Cobas Integra 400 system autoanalyzer (F. Hoffmann‐La Roche Ltd, Basel, Switzerland). We determined cut‐off values for low Hb, indicating anemia, and elevated ZPP and sTfR, indicating low iron status, following Adu‐Afarwuah et al. (Adu‐Afarwuah et al., 2016): Hb < 100 g/L, ZPP > 70 μmol/mol heme, and sTfR > 6 mg/L. Research staff delivered supplements to participants’ homes every 2 weeks and collected remaining supplements. Adherence was calculated as the percent of delivered supplements that were not returned to research staff. For a detailed description of these variables, see Ashorn et al (Ashorn et al., 2015). Maternal Hb and ZPP were assessed during a clinic visit at 6 months post‐partum in the same way as described above. A subsample of 369 women was randomly selected for assessment of additional biomarkers of nutritional status. Plasma retinol concentration (μmol/L) was assessed by high‐performance liquid chromatography, as previously described (Bieri, Tolliver, & Catignani, 1979). Breastmilk samples were collected at a home visit. Breastmilk DHA (percentage by weight of total fatty acids) was assessed by gas chromatography with flame ionization detection using a GC‐2010 (Shimadzu Corporation, Columbia, MD) equipped with a SP‐2560, 100‐m fused silica capillary column (Supelco, Bellefonte, PA; Oaks et al., 2017). Breastmilk concentrations of Vitamins B1, B2, B3, B6, and B12 were assessed at the Western Human Nutrition Research Center. Free thiamin, thiamin monophosphate, and thiamin triphosphate were measured by high‐performance liquid chromatography‐fluorescence detection after precolumn derivatization to their thiochrome esters (Hampel et al., 2016). Thiamin (B1) was calculated as free thiamin + (thiamin monophosphate × 0.871) + (thiamin pyrophosphate × 0.707). Riboflavin (B2), nicotinamide (B3), pyridoxal (B6), and flavin adenine dinucleotide were measured by UPLC‐MS/MS (Waters, Milford, MA coupled to 4000 QTRAP LC‐MS/MS, AB Sciex, Foster City, CA) as previously described (Hampel, York, & Allen, 2012). Riboflavin was calculated as free riboflavin + (flavin adenine dinucleotide × 0.479). Vitamin B12 was analyzed using an IMMULITE® E‐411 competitive binding assay (Duluth, GA, USA) as previously described (Hampel et al., 2014). For further details, see Supplemental Methods. A team of five project staff, who were blind to intervention arm, visited participants at their homes at 6 months post‐partum to assess maternal cognition and caregiving. If family members or neighbors gathered to observe the assessments, data collectors politely asked them to leave in order to create a private environment, which was generally successful. Apart from the participants, one or more other adults were present at 5% of the visits, and one or more other children were present at 12% of the visits. To assess cognition, we selected three tests that were previously adapted for use in a maternal supplementation trial in Indonesia (Prado et al., 2012). In that study, tests were selected on the basis of the following criteria: widely‐used tests that primarily tap aspects of specific cognitive functions; are tied to particular brain structures and mechanisms; may be affected by micronutrient deficiency based on previous studies; do not require special equipment; do not require literacy; are easily administered and scored; and do not require subjective judgments from the testers. For this study, of the six cognitive tests used by Prado et al. (2012), we selected the three that did not require verbal stimuli to be developed in the local languages: digit span forward and backward, category fluency, and mental rotation. Digit span forward and backward tests measure attention, verbal short‐term memory, and working memory, which are rooted areas in the right dorsolateral prefrontal cortex and bilateral inferior parietal lobule, as well as the anterior cingulate (Gerton et al., 2004). Participants were orally presented with increasingly longer sequences of digits and instructed to either repeat them (digit span forward) or repeat them backwards (digit span backward), until an error was committed on two consecutive trials. The score was the number of sequences correctly repeated. Category fluency assesses semantic memory, which is rooted in areas of the temporal lobe, and executive function, which is rooted in areas of the frontal lobe (Birn et al., 2010). Participants were asked to name as many members of a category as possible in 1 min, first for the category “food” and second for the category “girl’s names.” The score was the average of the two trials. Mental rotation measures visuospatial ability and dynamic mental imagery and activates areas in the superior parietal cortex, inferior prefrontal cortex, and other structures (Zacks, 2008). The participant was visually presented with five rows of figures and instructed to mark the figures that were rotations but not mirror images of the target figure. The score was the percent correct. All cognitive tests were audio recorded and reviewed by a data collector who did not conduct the assessment, to correct any errors. The supervisor reviewed 10% of each batch of submitted forms and recordings. If any error was found, the supervisor reviewed the entire batch and corrected any errors. Thirty‐two participants were tested twice to evaluate test–retest reliability, with a mean test–retest interval of 7 days. Test–retest reliability (Pearson’s r) was 0.62 for digit span forward, 0.50 for digit span backward, 0.66 for category fluency, and 0.63 for mental rotation. We developed a functional health literacy test to assess memory and understanding of health messages communicated in words and pictures, on the basis of a test developed in Ethiopia (Stevenson, 2011). We selected health materials that were common in Malawi and that were relevant for children’s health, such as medication instructions, breastfeeding information, and growth charts. The score was the number of questions answered correctly out of a maximum 36 points. Test–retest reliability was 0.78. Because this test was not audio recorded, we also assessed inter‐rater agreement by periodically assigning pairs of data collectors to visit 32 women. One person conducted the test whereas the other independently completed the form. Inter‐rater agreement was 94%. We assessed maternal caregiving behavior using the Infant/Toddler version of the Home Observation for the Measurement of the Environment (HOME) Inventory (Caldwell & Bradley, 2003). The HOME Inventory measures the amount and quality of stimulation that children receive from their environment. Items assess maternal responsivity, acceptance, and involvement as well as the learning materials, variety, and organization in the child’s environment. To adapt the items to the local context, we conducted a focus group discussion with 12 mothers of young children in each of the three study sites. We used this information to add locally‐appropriate examples, such as toys for making music, and to modify items to increase variance in scores. For example, we changed “At least ten books are visible in the home” to “At least one book is visible in the home.” We eliminated nine items for which we could not find an appropriate modification (e.g., “Child has a special place for toys and treasures”). The adapted tool comprised 18 items coded by observation, 12 by interview, and 6 by observation or interview, according to the standard procedure. The total score was the sum of the item scores, each of which was scored 0 or 1. Inter‐rater agreement was 89% and test–retest reliability was 0.82, using the same procedures described above. The HOME score showed expected correlations with household asset index (r = 0.29, p < .001) and maternal education (r = 0.27, p < .001), providing evidence for convergent validity. All analyses were conducted using SAS Version 9.4 (SAS Institute, Cary, NC). The primary analysis was by intention to treat. We also conducted per protocol analyses excluding women with less than 80% adherence to supplement consumption. For each score, we computed z‐scores on the basis of the distribution of our sample. All cognitive scores were normally distributed, after truncating outliers to the 1st and 99th percentile (2 digit span forward scores, 7 digit span backward scores, and 3 HOME scores). We estimated the difference between the intervention arms first in unadjusted models using analysis of variance and then in adjusted models using analysis of covariance. For each outcome, we determined a set of covariates as any of 17 prespecified covariates that independently predicted that outcome score at p < .1. All potential covariates are listed in Footnote 2 of Table 3. If the F value was significant at p < .05, we used Tukey–Kramer's test to adjust for multiple comparisons for pairwise comparisons between groups. Mean maternal cognitive and HOME z‐scores at the end of the intervention perioda Note. IFA = iron/folic acid; MMN = multiple micronutrients; LNS = lipid‐based nutrient supplements; HOME = home observation for the measurement of the environment. As potential effect modifiers, we examined baseline ZPP and sTfR, plus 12 additional effect modifiers defined a priori: baseline maternal height, BMI, Hb, malaria, education, age, and gestational age; primiparity, season at enrollment, and study site; and household food insecurity access scale score and household asset index. If any interaction between the potential effect modifier and intervention arm was significant at p < .05, we further explored the pattern of the effect at high and low levels of the effect modifier. To eliminate skewness in the biomarker variables, we applied the following transformations: log transformation for ZPP and breastmilk DHA; square root transformation for breastmilk Vitamin B3, B6, and B12; and inverse transformation for breastmilk Vitamin B2. We then entered all nine biomarkers collected at 6 months post‐partum into a principal component analysis (PCA) to reduce the number of variables examined. We retained three components with eigenvalues >1. The first component represented higher nutritional status on all variables, with all eigenvectors >0.3 except breastmilk B1 (0.19), plasma retinol (0.18), and breastmilk DHA (0.05). The second component represented higher breastmilk B1, B2, and B6, but lower Hb, iron, and Vitamin A status. The third component represented higher breastmilk DHA (eigenvector = 0.51) and Vitamin B12 (eigenvector = 0.59). For details, see Table S1. For each maternal cognitive and caregiving score, we examined the association with each of the three nutritional biomarker components, adjusting for the same set of covariates specific to that outcome.