Vitamin A and Carotenoids During Pregnancy and Maternal, Neonatal and Infant Health Outcomes: a Systematic Review and Meta-Analysis

Authors


Andrew Thorne-Lyman, Department of Nutrition, Harvard School of Public Health, 665 Huntington Avenue, Building II Room 320, Boston, MA 02115, USA. E-mail: athornel@hsph.harvard.edu

Abstract

Vitamin A (VA) deficiency during pregnancy is common in low-income countries and a growing number of intervention trials have examined the effects of supplementation during pregnancy on maternal, perinatal and infant health outcomes. We systematically reviewed the literature to identify trials isolating the effects of VA or carotenoid supplementation during pregnancy on maternal, fetal, neonatal and early infant health outcomes. Meta-analysis was used to pool effect estimates for outcomes with more than one comparable study. We used GRADE criteria to assess the quality of individual studies and the level of evidence available for each outcome. We identified 23 eligible trials of which 17 had suitable quality for inclusion in meta-analyses. VA or beta-carotene (βC) supplementation during pregnancy did not have a significant overall effect on birthweight indicators, preterm birth, stillbirth, miscarriage or fetal loss. Among HIV-positive women, supplementation was protective against low birthweight (<2.5 kg) [risk ratio (RR) = 0.79 [95% confidence interval (CI) 0.64, 0.99]], but no significant effects on preterm delivery or small-for-gestational age were observed. Pooled analysis of the results of three large randomised trials found no effects of VA supplementation on neonatal/infant mortality, or pregnancy-related maternal mortality (random-effects RR = 0.86 [0.60, 1.24]) although high heterogeneity was observed in the maternal mortality estimate (I2 = 74%, P = 0.02). VA supplementation during pregnancy was found to improve haemoglobin levels and reduce anaemia risk (<11.0 g/dL) during pregnancy (random-effects RR = 0.81 [0.69, 0.94]), also with high heterogeneity (I2 = 52%, P = 0.04). We found no effect of VA/βC supplementation on mother-to-child HIV transmission in pooled analysis, although some evidence suggests that it may increase transmission. There is little consistent evidence of benefit of maternal supplementation with VA or βC during pregnancy on maternal or infant mortality. While there may be beneficial effects for certain outcomes, there may also be potential for harm through increased HIV transmission in some populations.

It has been estimated that about 19 million pregnant women in low-income countries each year are affected by vitamin A (VA) deficiency, the majority in South and South-east Asia.1,2 Nearly 10 million women suffer from night blindness during pregnancy, a condition often caused by VA deficiency and associated with a constellation of adverse health and nutritional conditions among mothers and their infants.1,3–5

Dietary VA is found as preformed retinol in animal food sources (such as liver, eggs and milk), as retinyl esters in fortified foods, and as provitamin A carotenoids found in green leafy vegetables and yellow/orange fruits such as mangos and papaya.6 Although many provitamin A carotenoids have been identified, beta-carotene (βC) has the highest conversion rates to retinol and additionally has anti-oxidative properties. Lycopene is a carotenoid without provitamin A activity but is an efficient anti-oxidant.

The recommended safe intake of VA during pregnancy by Food and Agriculture Organization/World Health Organization (FAO/WHO) is 800 µg retinol equivalents (RE)/day [2664 international units (IU)].7 VA is known to have teratogenic effects if consumed at high doses during early pregnancy, and the FAO/WHO advise limiting intake to a maximum of 3000 µg RE/day (10 000 IU) or weekly intakes of 25 000 IU to minimise risks.7 One approach used in some trials to try to meet the needs of highly deficient populations while minimising risks of teratogenicity has been to include both VA and βC in supplements.8,9

Vitamin A deficiency is most often considered to be a problem of public health concern among preschool age children (age 6–59 months) as it increases risk of mortality from infectious disease, particularly measles, diarrhoea and malaria.6,10 Findings that high dose supplementation with VA two to three times per year reduces the risk of all cause child mortality by an average of 24% form a solid evidence base for supplementation programmes of this age group.11

In contrast, VA deficiency during pregnancy was, until recently, a largely unexplored area of public health interest.2 Early work in the 1930s by Mellanby and Green identified a potential role for VA (in the form of cod-liver oil) to reduce puerperal septicaemia, although little follow-up work continued for the next half century.12–14 Four main developments contributed to a re-awakening of interest in the role of VA supplementation during pregnancy: (1) recognition that maternal supplementation during pregnancy and lactation could be an opportunity to improve the VA status and health of infants, (2) findings from a large trial in Nepal that VA and βC supplementation during pregnancy was associated with a 40% reduction in risk of pregnancy-related mortality,15 (3) observations that low serum retinol levels were associated with increased risk of mother-to-child transmission of HIV, and the hypothesis that VA, by virtue of its role in epithelial integrity or immune-modulating properties might help to reduce transmission,16 and (4) recognition that VA might have a role in improving haematological status during pregnancy.17

The objective of our review was to consolidate knowledge about the effects of supplementation on multiple outcomes related to maternal, perinatal and infant health to inform policy in low-income countries and to identify research priorities.

Methods

We searched the literature to identify trials, intervention studies and quasi-experimental studies isolating the effect of supplementation with VA and/or carotenoids during pregnancy and outcomes shown in Table 1. We searched NLM Pubmed and the Cochrane Library using search strategies outlined in Table S1 in November 2010 and updated the search in June 2011. We also hand-searched reports from major micronutrient conferences as well as the bibliographies of relevant reviews and studies.

Table 1.  List of primary and suboutcomes covered by this review
  • a

    Indicates outcomes for which sufficient comparable outcomes could not be extracted from more than one study.

1. Low birthweight due to intra-uterine growth restriction
 1a. Small-for-gestational age
 1b. Low birthweight
 1c. Mean birthweight
 1d. Weight gain during pregnancy
 1e. Average heel-crown length at deliverya
2. Preterm birth
 2a. Preterm birth
 2b. Early preterm birth
 2c. Mean gestational age at delivery
3. Neonatal growth, morbidity and mortality
 3a. Early infant mortality (first 6 weeks)
4. Infant and child growth, morbidity and mortality
 4a. Mean weighta
 4b. Mean heighta
 4c. Low weight-for-agea
 4d. Low height-for-agea
 4e. All-cause mortality
 4f. Mortality due to measles, diarrhoea, acute lower respiratory infection or malariaa
 4g. Mother-to-child transmission of HIV
 4h. Morbidity (diarrhoea, measles, malaria, acute lower respiratory infection)a
 4i. Child anaemia and mean haemoglobin
5. Maternal mortality and pregnancy complications
 5a. Mortality (all cause)
 5b. Mortality due to haemorrhage, sepsis or obstructed laboura
 5c. Hospital admissions from complications or caesarian sectiona
 5d. Pre-eclampsia
6. Maternal nutritional status
 6a. Anaemia and mean haemoglobin

An initial screening of all titles and abstracts was undertaken to exclude (1) non-human studies, (2) studies not in English, French or Spanish, (3) reviews, case reports, commentaries and other papers not reporting primary data, (4) topics clearly not relevant to the exposure/outcome relationships, and (5) papers with research designs that clearly could not isolate effects of VA or carotenoids (e.g. studies of cod-liver oil). One reviewer (AT-L) reviewed the original full papers of studies that could not be excluded based on title and abstract review.

Data were extracted using a standard spreadsheet developed for other reviews in this supplement using best available published data. The authors of one paper provided a clarification of data on birth outcomes in response to requests (A. Coutsoudis, pers. comm., 4 November 2011).8 The quality and the potential for bias of each study was using a slightly modified version of the Child Health Epidemiology Reference Group's GRADE tool.18

Meta-analysis was used to generate effect estimates for outcomes in which there was more than one study with sufficient data appropriate to enable pooling. We excluded studies that were classified as ‘Very Low’ using the GRADE criteria. Studies that randomised VA during both pregnancy and lactation were eligible for inclusion, recognising that attribution of effects of VA supplementation specifically during pregnancy could not be made for post-partum outcomes.

Review Manager 5.1 was used for all data analysis.19 Dichotomous outcomes were expressed as risk ratios (RR), and continuous outcomes were expressed as mean differences with 95% confidence intervals (CI). Inverse variance weights were used in the pooling of data. We pooled data from cluster-randomized trials with individually randomized trials using the generic inverse variance approach, adjusting for the intracluster correlation co-efficient in cases where it was possible to estimate it from the trial data.19 Heterogeneity across studies within outcomes was explored using visual inspection of the distribution of effect estimates, I2 values, and tested for using χ2-tests for heterogeneity using a P-value of less than 0.10 to denote statistical significance. Data were pooled using a fixed-effects model except in cases in which statistically significant heterogeneity was present in which case a random-effects model was used. Recognising the potential for heterogeneity across studies, we considered baseline VA status, supplement form and dosage, HIV prevalence as a priori sources of heterogeneity. We present disaggregated effect estimates in addition to pooled effect estimates for outcomes in which trials in HIV-positive women were available in order to facilitate policy decisions for populations with high or low HIV prevalence.

A number of trials used multifactorial designs with multiple treatment arms containing VA, βC or both.9,15,20–24 For purposes of data extraction and meta-analyses we treated these studies as follows: (1) for papers in which results were presented for pooled treatment arms containing VA/βC vs. those not containing VA/βC, we also pooled data in this manner, (2) for studies using the same factorial design where two separate comparisons isolating the effects were possible (i.e. VA vs. placebo and VA + multivitamins vs. multivitamins alone), data were entered for each comparison and treated as separate studies, and (3) for studies with multiple arms testing different dosages of VA but only one placebo group, data from the two treatment arms were combined when possible.

Results

Literature search and identification of studies

After removing duplicates, our search resulted in 923 articles from which 23 eligible studies were identified with at least one outcome of relevance. Of these, six were ranked ‘Very Low’ on the GRADE criteria and were therefore excluded from meta-analyses (Table S2). Our search of the abstracts from key micronutrient conferences led to the identification of several abstracts of relevance although none contained sufficient detail to enable evaluation of study quality or inclusion in pooled analysis.

Study characteristics

Key characteristics of studies included in this review are presented in Table 2. Three large cluster randomised controlled trials (RCTs) were included: trials from Nepal and Bangladesh assessed the independent and pooled effects of both VA and βC on maternal and neonatal mortality in addition to other outcomes, while the other, from Ghana, examined the effects of VA alone.15,25,26 Of the 17 studies included in the meta-analyses, 10 were conducted in Asian settings and seven in Africa and with the exception of one study from China and one from South Africa all were conducted in low-income countries.8,21 Three RCTs were undertaken in HIV-positive pregnant women in Africa prior to the availability of anti-retroviral therapy, and had the primary objective of exploring the effect of VA supplementation on mother-to-child transmission of HIV, but also reported on a number of birth and child health outcomes.9,27,28

Table 2.  Description of key studies discussed in the paper
Study IDCountryPopulationInterventionComments
  • a

    Additional information for preterm, small-for-gestational age and birthweight indicators obtained through personal communication with Drs Coutsoudis and Kuhn (4 November 2011).

  • IU, international units; VA, vitamin A; RE, retinol equivalents; βC, beta-carotene; MV, multivitamin; RCT, randomised controlled trial.

Banerjee (2009)30India (New Delhi)Healthy primigravid women with singleton pregnancy gestational weeks 12–20Daily oral dose of (1) 2 mg lycopene, n = 77, (2) placebo tablet of same appearance, n = 82 until deliveryDouble-blind RCT. Inadequate allocation concealment, lack of intent-to-treat analysis.
GRADE quality: Low
Coutsoudisa (1999)8,27South Africa (KwaZulu-Natal)HIV-infected women recruited gestational weeks 17–39. Baseline: 30.6% serum retinol <0.7 µmolDaily oral dose of (1) VA: 5000 IU retinyl palmitate and 30 mg βC and 200 000 IU VA at delivery, n = 368, (2) placebo, n = 360Double-blind RCT. Unclear sequence generation, unclear allocation concealment.
GRADE quality: High
Cox (2005)48GhanaHealthy primigravid gestational age 24 weeksDaily oral dose of (1) VA: 10 000 IU RE, n = 48, (2) placebo, n = 50.Double-blind RCT. Small sample size.
GRADE quality: Moderate
Dijkhuizen (2004)20,37Indonesia (West Java)170 pregnant women. Rural women, gestational age <20 weeksMultifactorial trial of daily (1) 4.5 mg βC, (2) 30 mg zinc sulphate, (3) 4.5 mg βC + 30 mg zinc sulphate, (4) placebo. All women received iron/folateDouble-blind RCT.
GRADE quality: Moderate
Fawzi (1998)9,10,32,34,35,39,70,77,80–83Tanzania (Dar es Salaam)HIV-1 infected between 12 and 27 weeks of gestation, 34% had serum retinol <0.7 µmol in the second trimesterMultifactorial trial, daily dose of (1) VA: 30 mg βC plus 5000 IU preformed VA and one dose 200 000 IU at delivery, n = 272, (2) MV excluding VA + βC, n = 271, (3) MV including VA + βC and one dose 200 000 IU VA at delivery, n = 268, (4) placebo, n = 267. All received iron-folate and prophylactic chloroquine phosphate. Non VA groups received placebo at deliveryDouble-blind RCT. Adequate sequence generation, allocation concealment.
GRADE quality: High
Kirkwood (2010)25GhanaRural women aged 15–45. Serum retinol 15.4% < 0.70 µmol/L; mean serum retinol 1.18 (0.52) µmol/LWeekly oral dose of (1) VA: 25 000 IU RE, n = 104 484 women, 39 601 pregnancies, (2) placebo, n = 103 297 women, 39 234 pregnanciesDouble-blind RCT. High dropout due to movement but results accounting for movement were similar.
GRADE quality: High
Kumwenda (2002)28Malawi (Blantyre)HIV-infected, enrolled at gestational weeks 18–28. 51% serum retinol <0.7 µmol during the second trimesterDaily oral dose of (1) VA: 10 000 IU, n = 340, (2) placebo, n = 357. Both groups received 200 000 IU VA at deliveryDouble-blind RCT.
GRADE quality: High
Ma (2008)21China (rural)Anaemic (Hb < 110 g/L) pregnant women, 12–24 weeks of gestationMultifactorial trial, women assigned to (1) 60 mg iron, n = 93, (2) 60 mg iron and 2 mg retinol as retinyl palmitate (≈3636 IU), n = 91, (3) 60 mg iron and 1.0 mg riboflavin, n = 91, (4) 60 mg iron, 2000 µg retinol, 1.0 mg riboflavin, n = 83Double-blind RCT.
GRADE quality: High
Muslimatun (2001)33,38,65,84,85Indonesia (West Java)16–20 weeks pregnant at enrolment, aged 17–35 years, mean baseline serum retinol ≈ 1.01 + 0.03 µmol/LTwo arms of relevance: (1) VA: weekly dose of 4800 RE as retinyl acetate (≈14 000 IU) with 60 mg elemental iron and 250 µg folic acid, n = 122, (2) iron and folic acid as above, n = 121Double-blind RCT.
GRADE quality: High
Radhika (2003)86India (Hyderabad)Healthy women gestational age 16–24 weeks. Excluded women with recurrent pregnancy loss, prior preterm deliveryRandomised to (1) daily sachet of red palm oil containing 2173–2307 µg βC/day (≈3621–3845 IU), (2) identical appearing sachet of groundnut oil provided for 8 weeks starting from 26 to 28 weeks of gestation. All women received iron-folate for 100 days and prenatal careRandomised trial. Sequence generation and allocation concealment unclear; blinding unclear. Mean Hb at baseline higher in red palm oil group than control group.
GRADE quality: Low
Semba (2001)63Malawi (Blantyre)HIV-negative pregnant women, gestational age 18–28 weeksRandomised to daily (1) 30 mg elemental iron, 400 µg folate and 3000 µg VA (≈10 000 IU) or (2) iron and folate, as above. Both groups received prenatal care, treatment for sexually transmitted diseases, malaria prophylaxis (Fansidar)Double-blind RCT.
GRADE quality: High
Sharma (2003)31India (New Delhi)Primigravid women with gestational age 16–20 weeks with absence of medical complicationsTwice daily oral dose of (1) 2 mg lycopene (n = 116), (2) placebo tablets of same appearance (n = 135)Double-blind RCT. Lack of intent-to-treat analysis.
GRADE quality: Low
Suharno (1993)22Indonesia, West JavaPregnant women with Hb 8.0–10.9 g/dL, gestational age 16–24 weeksRandomised to daily (1) placebo, n = 62, (2) VA 2.4 mg retinol as retinyl palmitate (≈4363 IU) (n = 63), (3) 60 mg iron as ferrous sulphate and placebo (n = 63), (4) VA + iron (n = 63)Double-blind RCT. Not intent-to-treat (23 participants excluded for taking supplements <8 weeks). Inadequate allocation concealment.
GRADE quality: Low
Tanumihardjo (2002)23Indonesia, West JavaWomen in the second or third trimester, mean 17 weeks of gestationWomen randomised to (1) placebo, n = 7, (2) 8000 IU VA as retinyl palmitate, n = 7, (3) 60 mg ferrous sulphate, n = 5, (4) VA plus iron as above, n = 8Double-blind RCT. Small sample size. Unclear sequence generation/allocation concealment.
GRADE quality: Moderate
Van den Broek (2006)24MalawiSingleton pregnancies with Hb 5–10.9 g/dL enrolled at 12–24 weeks, baseline mean serum retinol 37.1 (13.7) µg/dL, 7.4% <20 µg/dLWomen were randomised to daily doses of (1) 5000 IU VA, (2) 10 000 IU VA or (3) placebo. All women received daily iron-folate and antimalarial prophylaxisDouble-blind RCT.
GRADE quality: High
West (1999)3,5,15,29,36,49,55,87–90Nepal (Sarlahi)Rural married women of reproductive age. 19% with serum retinol <0.70 µmol/LWeekly oral dose of (1) 23 300 IU VA (7000 µg RE as retinyl palmitate) (n = 6070 pregnancies), (2) 42 mg of all trans-βC 7000 µg RE (n = 5650 pregnancies), (3) placebo (n = 5653 pregnancies)Double-blind RCT.
GRADE quality: High
West (2011)26Bangladesh (Rangpur district)Rural women with known pregnancy outcome and vital status recorded at 12 weeks. 7.7% with serum retinol <0.70 µmol/LWeekly oral dose of (1) VA (7000 µg RE as retinyl palmitate) (23 300 IU) (n = 19 806 pregnancies), (2) 42 mg of all trans βC 7000 µg RE (n = 19 998 pregnancies), (3) placebo (n = 19 862 pregnancies)Double-blind RCT in women of reproductive age with three arms.
GRADE quality: High

Dosage and form of supplements varied across the studies from a weekly dose providing the equivalent of a daily dose of 3333 IU VA to 10 000 IU per day. Two of the studies in HIV-positive women included intervention arms with both a supplement of 5000 IU VA per day and 30 mg βC as well as a 200 000 IU dose of VA at delivery.8,9 Two studies explored the effects 2of lycopene supplementation, one of which provided 2 mg daily while the other provided 4 mg per day, although both of these studies were judged to be of low GRADE quality (Table S2).

Outcomes

A summary of the overall evidence for each of the reviewed outcomes and quality of evidence by outcome is presented in Table 3.

Table 3.  Summary of findings and overall assessment of quality of evidence, by outcome
Quality assessmentSummary of findings
No. studies and study designHeterogeneity of results?Consistent size of effect?Generalisable to intervention of interest?Other sources of bias (e.g. major limitations in study design)nStatistical methodEffect estimate [95% CI]
  • a

    Effective sample size for analysis is actually smaller because of inclusion of cluster-randomised trials in the estimation of variance.

  • b

    Different arms of factorial trials counted as separate trials.

  • CI, confidence interval: IU, international units; RCT, randomised controlled trial; RR, risk ratio, MD, mean difference, IV, inverse variance.

Small-for-gestational age, overall quality of evidence grade = Low
Two RCTs8,9 (both in HIV-positive women)Low (I2 = 0%, P = 0.51); Neither study showed a significant effectBoth studies nullBoth studies provided the same supplement (5000 IU VA + 30 mg βC + post-partum 200 000 IUNone1387RR0.89 [0.68, 1.17]
(IV, Fixed)
Low birthweight <2.5 kg, overall quality of evidence grade = Low
Five RCTs8,9,28,48,86 (three in HIV-positive women)Moderate (I2 = 20%, P = 0.29); one study showed significant effectAll but one study had protective direction of effectDifferent supplements: VA + βC, VA, palm oil. Iron-folate, malaria prophylaxis varied across trialsAll but one study were of high or moderate grade quality2254RR0.83 [0.67, 1.01]
(IV, Fixed, 95% CI)HIV+: 0.79 [0.64, 0.99]
HIV−: 1.11 [0.61, 2.01]
Very low birthweight <2.0 kg, HIV-positive: overall quality of evidence grade = Low
Two RCTs8,9 (both in HIV-positive women)Low (I2 = 0%, P = 0.49); neither study showed a significant effectBoth studies null, effect estimates towards protectiveBoth used the same supplement (5000 IU VA + 30 mg βC)None, all were high grade quality1486RR0.73 [0.41, 1.29]
(IV, Fixed, 95% CI)
Mean birthweight (kg): overall quality of evidence grade = Moderate
Seven RCTsb8,9,20,28,48,86 (three in HIV-positive women)High (I2 = 53%, P = 0.05); two studies showed significant effects in opposite directionsFive were null, two showed opposite effects.Supplements used included VA, VA + βC, βC and palm oil as a source of βCAll but one study were of high or moderate grade quality2417MD, IV0.03 [−0.04, 0.10] kg
RandomHIV+: 0.04 [−0.01, 0.08]
HIV−: 0.02 [−0.17, 0.21]
Lycopene supplementation on mean birthweight (kg): overall quality of evidence grade = Very Low
Two RCTs30,31High (I2 = 65%, P = 0.09);Both studies null, RR in opposite directionsOne study supplemented with 2 mg/day lycopene, the other with 4 mg/dayBoth studies had low grade quality, neither used intent-to-treat analysis410MD, IV0.03 kg [−0.12, 0.18]
Random
Preterm birth <37 weeks: overall quality of evidence grade = High
Seven RCTsb8,9,15,20,24,86Low (I2 = 8%, P = 0.37)One study protective, six nullTwo studies in HIV-positive women; five in HIV-negative; differences in supplement type/dosageAll but one study were of high or moderate quality19 799aRR 
(IV, Fixed, 95% CI)0.99 [0.88, 1.10]
HIV+: 0.93 [0.75, 1.14]
HIV−: 1.01 [0.89, 1.15]
Early preterm birth <34 weeks: overall quality of evidence grade = Low
Two RCTs8,9 (both in HIV-positive women)High (I2 = 78%, P = 0.03);No, one study protective, one nullBoth studies used the same supplement (VA + βC)Both studies of high GRADE quality1513RR, IVHIV+
Random0.65 [0.20, 2.11]
Lycopene, mean gestational age at delivery: overall quality of evidence grade = Very Low
Two RCTs30,31(I2 = 94%, P < 0.0001);One null; one positiveOne with 2 mg/day lycopene, the other with 4 mg/dayBoth low GRADE quality, neither used intent-to-treat analysis410MD, IV0.40 weeks [−1.1, 1.9]
Random
Stillbirth: overall quality of evidence grade = High
Five RCTs (two HIV-positive)9,25,28,84,87Low (I2 = 0%, P = 0.98); all studies nullYes, all studies nullVariation in supplements used; VA, VA + βC,All studies of high GRADE quality106 894aRR, IV1.03 [0.97, 1.10]
FixedHIV+: 1.07 [0.66, 1.74]
HIV−: 1.03 [0.97, 1.10]
Fetal loss: overall quality of evidence grade = High
Seven RCTs (three HIV-positive)9,25,28,53,84,87,8Low (I2 = 0%, P = 0.78); all studies nullYes, all studies nullVariation in supplements usedAll studies of high GRADE quality113 207aRR, IV0.99 [0.95, 1.04]
FixedHIV+: 0.92 [0.50, 1.67]
HIV−: 0.99 [0.95, 1.04]
Miscarriage: overall quality of evidence grade = Moderate
Four RCTs (two HIV-positive)9,24,28,87Low (I2 = 0%, P = 0.58); all studies nullYes, all studies nullVariation in supplements used; VA, βC, VA + βCAll studies of high GRADE quality62 138aRR, IV0.99 [0.95, 1.04]
FixedHIV+: 0.92 [0.51, 1.67]
HIV−: 0.99 [0.95, 1.04]
Maternal pregnancy related mortality: overall quality of evidence grade = High
Three RCTs15,25,53High (I2 = 74%, P = 0.02); one study showed protective effect, two were nullOne large study protective; others nullVariation in supplements used; VA, VA + βCAll studies of high GRADE quality160 690aRR, IV0.86 [0.60, 1.24]
Random
Lycopene supplementation on pre-eclampsia: overall quality of evidence grade = Very Low
Two RCTs30,31(I2 = 54%, P = 0.14)One protective, one null2 mg/day lycopene vs. 4 mg/dayBoth low GRADE quality, neither intent-to-treat410RR, IV0.71 [0.44, 1.14]
Fixed
Neonatal and early infant mortality: overall quality of evidence grade = High
Four RCTs8,25,28,87 (two in HIV-positive women)Moderate: overall (I2 = 40.3%. P = 0.31); in HIV-negative I2 = 47%, P = 0.17), in HIV-negative I2 = 0%, P = 0.86)All studies nullVariation in supplements used; VA, βC, VA + βCAll studies of high GRADE quality91 022aRR, IV0.97 [0.90, 1.05]
FixedHIV+: 0.68 [0.39, 1.17]
HIV−: 0.98 [0.90, 1.06]
Infant and neonatal mortality: overall quality of evidence grade = High
Five RCTs8,25,28,46,87Low (I2 = 0%)All studies nullVariation in supplements used; VA, βC, VA + βC; variation in post-partum supplementsAll studies of high GRADE quality132 903aRR, IV0.99 [0.94, 1.04]
FixedHIV+: 1.03 [0.79, 1.35]
HIV−: 0.99 [0.94, 1.04]
HIV transmission from mother to child: overall quality of evidence grade = Moderate
Three RCTs8,9,28High (I2 = 75%, P = 0.02)Two studies null, one study showed significant increase in HIV transmissionVariation in supplements used; two had VA + βC, one had VAAll studies of high GRADE quality2022RR, IVHIV+: 1.05 [0.78, 1.41]
Random
Maternal anaemia (Hb < 11 g/dL) during follow-up in pregnancy: overall quality of evidence grade = High
Eight RCTsb21,22,24,63,85,86High: overall (I2 = 54%, P = 0.03%) (anaemic women I2 = 74%, P = 0.004) (non-anaemic women I2 = 0%)All studies had effect estimates in a protective direction 3/8 significantVariation in supplements used; VA, differences in provision of iron-folate supplementsAll studies but two of high GRADE quality1587RR, IV0.81 [0.69, 0.94]
RandomAnaemic: 0.72 [0.53, 0.99]
Unscreened for anaemia: 0.83 [0.69, 0.95]
Maternal severe anaemia (<8.0 or 8.5 g/dL) during pregnancy: overall quality of evidence grade = Low
Three RCTsb21,24Low (I2 = 0%, P = 0.03)All studies nullBoth trials used VABoth studies of high GRADE quality961RR, IV0.93 [0.60, 1.46]
Fixed
Mean maternal haemoglobin during follow-up in pregnancy: overall quality of evidence grade = Moderate
Eight RCTsb22–24,48,85,86Some (I2 = 17%, P = 0.29)All studies null but one factorial trial null where both comparisons showed significantly positive effectsFive studies used VA; one used palm oil. All women except those in placebo arms of two factorial trials received iron-folateAll but one study of high grade quality1034MD, IV0.35 [0.24, 0.45]
Fixed
Change in maternal haemoglobin (g/dL) during pregnancy: overall quality of evidence grade = Moderate
Eight RCTs21,24,63,85,86Some (I2 = 15%)All studies but the factorial trial nullVariation in supplements used; all had iron-folateAll but one study of high grade quality1131MD, IV0.19 [0.06, 0.33]
Random
Mean post-partum maternal haemoglobin (4–6 weeks post-partum) (g/dL): overall quality of evidence grade = Moderate
Four RCTs20,48,85Low (I2 = 0%, P = 0.31)All studies nullOne study with two arms used βC, two studies used VAAll but one study of high grade quality365MD, IV0.01 [−0.24, 0.25]
Fixed
Infant anaemia: overall quality of evidence grade = Low
Two RCTs28,32 (both in HIV-positive women)Yes (I2 = 74%, P = 0.05)One study protective, one nullVariation in supplements usedAll studies of high GRADE quality894RR, IVHIV+ 0.71 [0.49, 1.03], P = 0.07
Random
Infant haemoglobin at 4–6 months (g/dL): overall quality of evidence grade = Low
Two RCTs20,65 (one factorial)No (I2 = 0%, P = 0.96)All three studies nullVariation in supplements used;All studies of high GRADE284MD, IV−0.13 [−2.29, 2.03]
Fixed
Infant weight-for-age z score at 6 months: overall quality of evidence grade = Low
Two RCTs20,39 (two factorial)Some (I2 = 22%)Yes, all studies nullVariation in supplements usedAll studies of high GRADE1022MD, IV0.07 [−0.07, 0.22]
Fixed
Infant height-for-age z score at 6 months: overall quality of evidence grade = Low
Two RCTs20,39 (two factorial)No (I2 = 0%)Yes, all studies nullVariation in supplements usedAll studies of high GRADE1022MD, IV0.01 [−0.12, 0.13]
Fixed

Small-for-gestational age and birthweight

The overall effects of VA on small-for-gestational age (SGA), and very low birthweight (<2.0 kg) were null, and that for low birthweight (<2.5 kg) was null but had a trend towards significance (RR = 0.83 [95% CI 0.67, 1.01]) (Figure 1). The subgroup of three studies conducted in HIV-positive populations showed a significant reduction in risk of low birth weight (RR = 0.79 [0.64, 0.99]). No significant effects of VA on mean birthweight were apparent in meta-analysis (Figure S1) or in a large substudy from the Nepal trial published only in abstract form,29 and lycopene supplementation did not have a significant effect on mean birthweight in pooled analysis (Figure S2).

Figure 1.

Forest plot of the effect of vitamin A/βC supplementation during pregnancy on small-for-gestational age, low birthweight (<2.5 kg) and very low birthweight (<2.0 kg). IV, inverse variance; CI, confidence interval.

Effect on preterm birth

No significant overall effect was found of VA on preterm birth (<37 weeks) or early preterm birth (<34 weeks) (Figure 2). There was substantial heterogeneity within the HIV-positive subgroup for preterm birth (I2 = 76%, P = 0.04), and in the overall effect estimate for early preterm birth (I2 = 78%, P = 0.03). A significant 33% reduction in the prevalence of preterm birth and a 66% reduction in early preterm birth among women was observed in the South African study, but no other individual trials showed statistically significant effects.8 No effect of lycopene supplementation was found on mean gestational age at delivery (Figure S3) although substantial heterogeneity was observed (I2 = 94%, P < 0.0001) and one included trial reported an increased risk of preterm birth compared with placebo (10.4% vs. 1.2%, P = 0.02).30,31

Figure 2.

Forest plot of the effect of vitamin A/βC supplementation during pregnancy on preterm birth (<37 weeks gestational age) and early preterm birth (<34 weeks). IV, inverse variance; CI, confidence interval. BC, β-carotene; ZN, Zinc.

Effect on stillbirth, miscarriage and fetal death

The overall effect estimates for supplementation with VA or βC on stillbirth (Figure S4) miscarriage (Figure S5), and fetal loss (Figure S6) were all null, as were each of the individual studies contributing to the estimates.

Maternal mortality, morbidity and birth complications

Three large double-blinded, cluster-randomised, placebo-controlled trials assessed the effects of VA supplementation on all-cause pregnancy-related mortality up to 42 days post-partum.15,25,26 The combined effect estimate from these studies (Figure 3) indicated a null overall effect (RR = 0.86 [0.60, 1.24]), although significant heterogeneity across the studies was apparent (I2 = 74%, P = 0.02), with the Nepal study exhibiting a 44% decrease in maternal mortality and the Ghana and Bangladesh studies showing no significant effect. The trial in Ghana reported no significant difference in risk of hospital admission with diagnosis of one or more severe pregnancy-related conditions within 6 weeks post-partum between the VA and placebo groups (RR = 0.98 [0.89, 1.09]).25 Both the Nepal and Bangladesh trials reported information on cause of death through verbal autopsy but this information was not pooled because of a lack of information about the design effect from the cluster-randomised design. Although one trial with lycopene found a significant protective effect of supplementation on pre-eclampsia risk, the summary estimate from two studies was null (Figure S7). Two studies examined risk of delivery using caesarian section as an outcome, one with VA/βC and the other with lycopene and neither found a significant effect of supplementation.8,30 One study examined the effect of lycopene supplementation in pregnancy and found no effect on incidence of traumatic or atonic haemorrhage, or both.30

Figure 3.

Forest plot of the effect of vitamin A/βC supplementation during pregnancy on maternal pregnancy-related mortality. IV, inverse variance; CI, confidence interval.

Maternal anaemia and haemoglobin

Supplementation with any form of VA or βC was found to significantly (P < 0.05) reduce the risk of anaemia (Hb < 11 g/dL) at follow-up during pregnancy by 19% (Figure 4). Three of the individual trials found a significant protective effect, and there was consistency in the direction of effect measures for all of the trials including those with null results. However, there was significant heterogeneity in the effect estimate, which completely resolved in a sensitivity analysis that excluded the intervention and comparison arms of the factorial trial by Suharno providing concomitant iron supplementation (RR = 0.86 [0.79, 0.93], I2 = 0, P for heterogeneity = 0.42).22 Pooled analysis of studies with severe anaemia as an outcome indicated no statistically significant effect of VA supplementation (RR = 0.93 [0.59, 1.45]) (Figure S8).

Figure 4.

Forest plot of the effect of vitamin A/βC supplementation during pregnancy on anaemia (Hb < 110 g/L) during pregnancy. IV, inverse variance; CI, confidence interval; B, riboflavin.

Average follow-up measures of mean haemoglobin assessed after supplementation during pregnancy were found to be significantly higher among women supplemented with VA by 0.35 g/dL [95% CI 0.24, 0.45] (Figure S9) and a change in 0.19 g/dL [95% CI 0.06, 0.33] was observed among studies reporting change in haemoglobin levels (Figure S10). However, no differences in maternal haemoglobin were observed in group comparisons of women at 4–6 months post-partum (Figure S11).

Infant and young child haemoglobin and anaemia

Two studies, both in HIV-positive populations, examined the effect of maternal VA supplementation on risk of anaemia (Hb < 11.0 g/dL) among infants and young children suggested a trend towards a protective effect (RR = 0.71 [0.49, 1.03], P = 0.07) (Figure S12), although considerable heterogeneity was evident (I2 = 74%, P = 0.05).9,28 Sensitivity analyses using data on any anaemia during an average follow-up of 28 months from the Tanzanian trial rather than time to first episode of anaemia led to an attenuation of the effect estimate (RR = 0.89 [95% CI 0.74, 1.07], P = 0.21).32 No effect of VA supplementation on mean infant haemoglobin was observed at 4–6 months (Figure S13).

Neonatal and infant mortality, morbidity and growth

We found strong evidence that maternal VA supplementation during pregnancy does not have an overall effect on mortality during infancy or the neonatal period (Figure 5). The pooled effect estimates were null and weighted strongly by the large cluster-randomised trials, none of which suggested a significant effect. An analysis from the Nepal trial stratified by maternal night blindness status during pregnancy suggested that VA supplementation might reduce risk of mortality specifically among infants born to night-blind women although βC did not appear to be as efficacious.5 The Bangladesh trial was the only one that we identified reporting cause-specific infant mortality and found no differences across treatment groups.26

Figure 5.

Forest plot of the effect of vitamin A/βC supplementation during pregnancy on neonatal and infant mortality. IV, inverse variance; CI, confidence interval.

Assessment methods for neonatal and infant morbidity differed greatly across trials limiting the utility of pooling. A trial in Indonesia reported no significant effect on maternal VA supplementation during pregnancy on incidence of diarrhoea, cough, difficulty breathing or other morbidity outcomes over the first year of life.33 In the Tanzania trial of HIV-positive women, VA + βC supplementation during pregnancy and a large dose post-partum was found to significantly reduce risk of cough and rapid respiratory rate among offspring by 31% (RR = 0.69 [0.49–0.96]), but no effect on diarrhoea was observed.34 VA + βC supplementation alone was also associated with a borderline significant 63% reduction in incidence of clinical malaria in the same study (RR = 0.37 [0.13, 1.04], P = 0.06) compared with placebo but no significant effects of supplementation on malarial mortality were observed (RR = 0.65 [0.29, 1.46].34,35 In findings from the Nepal trial, published in abstract form, no significant supplementation group differences were observed in 7-day recalls of diarrhoea, dysentery, high fever or cough for infants at 3 or 6 months of life.36 Similarly in Indonesia, no significant effect of βC was observed on first occurrence of cough or diarrhoea.37

A number of studies have reported on anthropometric measures taken during the first year of life although differences in reporting of indicators constrained our ability to incorporate findings into pooled analyses.20,28,38,39 We found little evidence of an effect of VA or βC supplementation on weight-for-age or height-for-age z score (Figures S14,S15). In the Malawi trial, infants of HIV-positive women who received VA during pregnancy had greater attained length and weight at 6 weeks, but these differences did not persist at 14 weeks or 6 months of age.28 In a subset of 833 infants from the Nepal study, published in abstract form only, VA supplementation during pregnancy was found to be associated with significantly greater average daily weight and mid-upper arm circumference gain but not length gain from 0 to 3 months of age, but did not persist at 3–6 months of age, although there was a trend towards greater attained weight in both intervention groups at 6 months.29

Mother-to-child HIV transmission

The overall effect of supplementation with VA or VA/βC on HIV transmission was null, although there was significant heterogeneity for this outcome (I2 = 75%, P = 0.02) (Figure 6). Examining the individual effects of these trials, the trial from Tanzania showed a significant 35% increased risk of transmission and the remainder were null.9 Although not eligible for inclusion in our pooled analysis because it tested the effects of maternal and/or infant VA supplementation during the early post-partum period and not supplementation during pregnancy, we also noted findings from the Zvitambo trial suggesting increased risk of HIV infection or death among offspring whose mothers received post-partum VA supplementation but who did not receive direct supplementation themselves.40

Figure 6.

Forest plot of the effect of vitamin A/βC supplementation during pregnancy on HIV transmission from mother to child. IV, inverse variance; CI, confidence interval.

Comments

Intra-uterine growth restriction and preterm birth

Some, but not all, observational studies have reported that low plasma retinol concentrations are associated with intra-uterine growth restriction.41–44 We found limited evidence to evaluate the effect of VA supplementation on intra-uterine growth restriction, and no effect on mean birthweight was observed overall. Our findings of a significant reduction in the risk of LBW specifically among HIV-positive women are in contrast to previous meta-analyses, which reported a null effect.45,46 Our analysis differed in two ways: (1) we excluded a trial of multi-micronutrient supplementation which did not isolate the effect of VA, and (2) we requested and included additional data from one of the studies in which the published denominators were unclear.8,9,47 We therefore believe our effect estimates provide the most accurate representation of the pooled effects of VA/βC supplementation during pregnancy on risk of LBW incidence among HIV-positive women to date. It remains unclear whether differences in the effect of VA supplementation on LBW incidence among HIV-positive women are the result of a reduction in intra-uterine growth restriction, preterm birth or both, as the pooled estimates of the effects on both of these components of LBW were null.8 The trial from Malawi, which showed a protective effect, did use a higher dose of retinol (10 000 IU) than other studies and did not include high dose βC present in the other two trials although contextual differences might also explain this difference.9,28,35,48

All but one study suggested that VA or βC supplementation does not influence risk of preterm birth. The one study to show a protective effect of supplementation on preterm birth also noted that this effect disappeared when excluding multiple pregnancies.8 Interestingly, the large trial in Nepal found a heightened risk of twinning associated with VA/βC supplementation, a phenomenon that might be expected to increase the risk of adverse perinatal outcomes including preterm birth, although no difference in risk of preterm birth was observed overall.49

Stillbirth, miscarriage and fetal loss

Our finding of no adverse effect of VA supplementation on outcomes related to fetal loss supports conclusions that VA supplementation appears to be safe throughout pregnancy at levels at or below the amounts tested in the large trials (23 000 IU/week and 10 000 IU/day), although we did not specifically examine birth defects as an outcome.

Maternal all-cause pregnancy-related mortality

The World Health Organization currently does not recommend VA supplementation during pregnancy for the prevention of maternal and infant morbidity and mortality, but does recommend supplementation for the prevention of night blindness in areas where VA deficiency is a serious public health problem.50 Our findings are supportive of this recommendation and are also consistent with those of the recent Cochrane review, although our analysis differed in that we also included the subsequently published trial from Bangladesh and omitted an early trial of cod-liver oil on the grounds that it did not isolate the effects of VA.12,24,26

Substantial heterogeneity was observed for the effect estimate on maternal mortality, and reflects the contrast of the protective effect observed in Nepal against null findings observed in Bangladesh and Ghana. The heterogeneity observed across these three large and well-conducted trials is difficult to explain. Differences in compliance rates are an unlikely explanation, as all studies assessed and reported high compliance.15,25,26 It has been argued that findings in the Nepal study may have arisen by chance, based primarily on observations that a large number of deaths were attributable to injury and other causes in the verbal autopsy.25

Other contextual differences related to the prevalence of VA deficiency or pregnancy-related mortality might be responsible in some way for contrasting effects.51 The trial population in Nepal appears to have been worse off than those in Ghana and Bangladesh in many ways including having substantially higher rates of pregnancy related mortality (704, 377 and 231 deaths per 100 000 pregnancies respectively), higher presumed prevalence of wasting malnutrition, and lower intake of eggs and fish compared with Bangladesh.15,25,52,53 Although there was little reported information on availability of health services in the trial reports, it was noted that a greater proportion of deliveries in Bangladesh were attended by traditional or skilled health workers than in Nepal.26

Moderate VA deficiency (plasma retinol <0.7 µmol/L) was also most prevalent in Nepal (19%) followed by Ghana (15%) and Bangladesh (8%).15,25,52 While VA, but not βC, supplementation appeared to lead to significant reductions in moderate VA deficiency in Bangladesh and Nepal, findings from Ghana were counterintuitive: a greater proportion of women in the VA arm of the trial had moderate VA deficiency (<0.70 µmol/L) compared with those in the placebo arm (25% vs. 15.4%, P = 0.048). Such findings are difficult to explain as investigators reported good compliance with assigned regimens, although serum retinol is known to be an imperfect measure of VA status as it is homeostatically regulated and affected by infection and other factors unrelated to deficiency.54

The prevalence of night blindness during pregnancy in the Nepal and Bangladesh trials was similar (about 9–10%), while night blindness was not measured in the Ghana trial but is reportedly uncommon.25 Night-blind women in the Nepal trial were observed to have a fivefold greater risk of death due to infection-related causes compared with non-night-blind women and much of this mortality risk was attenuated by supplementation with VA or βC.55 In the Bangladesh trial, all women who were identified as night blind when screened at week 28 of gestation were treated with VA according to World Health Organization protocols for ethical reasons.56 If mortality risk was concentrated among such women in Bangladesh, as it appeared to have been in Nepal, it is possible that treatment may have led to an attenuation of any beneficial effects. In any case, findings from the Nepal study underscore the importance of recommendations that all night-blind women be treated with VA.

Maternal and child haemoglobin and anaemia

Cross-sectional surveys and observational studies in women and in children often show correlations between anaemia and VA deficiency and improving VA status in deficient populations typically results in improvements in anaemia.57–59 Our finding that supplementation with VA led to an average increase in mean haemoglobin levels on the order of approximately 0.2 g/dL is consistent with previous estimates that VA supplementation of deficient populations can be expected to raise haemoglobin concentrations by 0.2–1.0 g/dL over a period greater than 2 weeks.58 While not all intervention studies included in our meta-analysis showed a statistically significant reduction in anaemia risk, the direction of effect was consistent across all studies. Most of the studies we identified were conducted in trial populations that first screened as anaemic and there appeared to be a slightly greater effect of VA supplementation on haemoglobin in those populations. All but two of the studies we identified with outcomes related to maternal haemoglobin or anaemia provided iron to all women, so it was not possible to examine potential synergistic effects of VA in the presence or absence of iron supplementation.

The mechanisms through which VA affects anaemia remain largely unclear, although a number of hypotheses have been proposed including (1) a potential role of VA in mobilising iron stores from the liver, (2) increasing erythropoiesis, (3) decreasing the ‘anaemia of inflammation’ by increasing circulating iron through reducing infection, and (4) increasing iron absorption.58,60 Studies of the effects of VA supplementation on concentrations of erythropoietin, a hormone that stimulates production of red blood cells by the bone marrow, have been mixed, with one in pregnant women in Malawi showing no effect, another in preschool aged children in a malarial area of Tanzania a showing a decrease in concentrations, and a third in a non-malarial endemic area showing an increase in erythropoietin concentrations.61–63

Although we did not observe a statistically significant improvement in neonatal/infant anaemia associated with supplementation in our pooled analysis of two trials of HIV-positive women, a trend towards statistical significance did suggest the possibility of such an effect. However, a previous study in Zimbabwe noted no effect of high dose VA supplementation (400 000 IU) to women post-partum or 50 000 IU to infants on infant haemoglobin or anaemia, and another trial from Indonesia noted a similar content of iron in breastmilk in women who were supplemented with βC vs. those who were not.64,65 It therefore seems likely that any benefits of VA supplementation during pregnancy on infant haemoglobin are more likely to be the result of improving the iron endowment at birth rather than through breastmilk transfer of iron or VA, and that these benefits probably decrease over time. In a substudy of 728 infants of mothers participating in the large Nepal trial (published in abstract form), a 2 g/L increase in infant haemoglobin at 3 months was observed in the intervention group relative to controls suggesting that it is possible that effects may last at least this long.66

Neonatal and infant mortality, morbidity and growth

Pooled estimates for neonatal and infant mortality provide strong evidence that VA/βC supplementation during pregnancy at the doses tested does not appear to reduce neonatal or infant mortality. This is consistent with a meta-analysis showing no effect of maternal post-partum supplementation on mortality, although possible benefits of neonatal supplementation have not been ruled out and several trials are underway to test this hypothesis.67,68 As with maternal mortality, findings of an effect of supplementation on infant mortality specifically within the subset of infants born to night-blind women support recommendations that this condition be treated with VA supplements.5

While VA is known to have an effect on child growth, the precise impact has been difficult to quantify, as VA deficiency often occurs in conjunction with other growth-limiting deficiencies.6 We found little evidence to suggest that VA supplementation during pregnancy affects linear growth or weight gain during infancy, although one limitation of most of the studies we identified through our search was that they examined the effect of supplementation on mean growth rather than indicators more reflective of growth faltering.

HIV transmission

Trials of VA supplementation during pregnancy on risk of HIV transmission from mother to child were initiated based on evidence from observational studies that low maternal serum retinol was associated with increased risk of infant mortality, increased transmission of HIV from mother to child, and greater concentrations of HIV-1 in breast milk and genital tract secretions.8,10,16,28,40,42,69–72 None of these trials showed a reduction in HIV transmission associated with VA supplementation and the trial in Tanzania suggested a greater increase associated with supplementation during pregnancy and lactation, a finding that was supported by a trial in Zimbabwe of high dose post-partum and/or infant supplementation.40,70

A number of plausible mechanisms have been proposed for potential increased transmission.73 Evidence from a prospective cohort study in the US suggested a U-shaped relationship between dietary intake of VA and HIV disease progression and mortality.74,75 Some in vitro studies suggest that VA could increase replication of HIV-1 and increase susceptibility of monocytes and macrophages to HIV infection through increased expression of CCR5 receptors.76 Recent evidence from the Tanzanian trial showed that VA/βC supplementation was associated with increased risk of subclinical mastitis and increased HIV viral load in breastmilk.77,78 Thus, caution is warranted before initiating VA supplementation programmes to pregnant women in HIV endemic areas.

Limitations

We searched multiple sources including conference abstract books for relevant reviews but it is possible that we may have missed relevant studies or that publication bias may have influenced the results of our meta-analysis. Our analysis assumes that pooling of different forms and dosages of VA across different settings and contexts is appropriate – it is possible that different forms of VA may have different effects depending on the extent of deficiency and other contextual factors. While there are undoubtedly many different plausible explanations for apparent heterogeneity in results for some outcomes, our ability to identify sources of heterogeneity was limited by the small number of published studies. The recent Cochrane review on maternal VA supplementation during pregnancy provides complementary disaggregated analysis to ours that may be useful in examining sources of heterogeneity including by concomitant micronutrient supplementation, dosage and duration of intervention.79

Conclusions and future research directions

Our review examined the effects of VA/βC supplementation on a broad range of adverse perinatal, maternal and neonatal outcomes prevalent in low-income countries. In pooled analysis, we found little evidence of an overall effect of VA supplementation on either maternal or neonatal/infant mortality although we found significant heterogeneity in estimates of the effect on maternal mortality. VA/βC supplementation appears to also improve haematological status of women during pregnancy in the presence of concomitant supplementation with iron and folate. Supplementation may also reduce risk of LBW among HIV-positive women, although in areas of high HIV prevalence these benefits should be weighed against the potential risk of increased risk of transmission from mother to child. Findings of no adverse effect on outcomes, including stillbirth, fetal loss, miscarriage, support the safety of supplements at levels tested. The evidence level for some outcomes remains low or moderate, and more studies are needed to be able to better understand potential sources of heterogeneity potentially attributable to dosage, HIV status, VA status, interaction with other nutrients and other contextual factors.

Acknowledgements

We gratefully acknowledge the additional data provided by Dr Anna Coutsoudis and Dr Louise Kuhn for inclusion in this analysis and the feedback received from Dr Reynaldo Martorell and Dr Beth Imhoff-Kunsch on an earlier draft of this manuscript. Andrew Thorne-Lyman is supported by US National Institute of Health training grant T32 #DK 007703 and a Julius B. Richmond Fellowship from the Center on the Developing Child at Harvard University.

Conflicts of interest

The authors declare no conflicts of interest.

Ancillary