Maternal antenatal multiple micronutrient supplementation for long-term health benefits in children: a systematic review and meta-analysis
Description of trials with follow-up reports
Follow-up reports were identified for nine of the trials in the Cochrane review. In
total, 88,057 women were recruited. The trials were spread geographically: two in
Africa 23], 24], one in the Americas 25], and six in Asia 26]–31]. All sites were rural, with the exception of Nepal Janakpur (urban and rural) and
Guinea (semi-urban). Mean ages of mothers were similar and ranged from 21.5 (Nepal
Janakpur) to 25.6 years (Indonesia). Mean maternal BMI, measured at recruitment during
pregnancy, ranged from 19.3 kg/m
2
(Nepal Sarlahi) to 24.1 kg/m
2
(Mexico). Trial characteristics, with results summarized in the way in which they
were presented in the trial papers, are shown in Table 1 and have been previously described in detail 20], 32].
Table 1. Description of trials with follow-up reports
Six of the nine trials used the UNIMMAP supplement developed by UNICEF, the United
Nations University, and WHO, and were designed to provide the recommended daily allowances.
It contained vitamin A 800 ?g, thiamine 1.4 mg, riboflavin 1.4 mg, niacin 18 mg, vitamin
B6 1.9 mg, folic acid 400 ?g, vitamin B12 2.6 ?g, vitamin C 70 mg, vitamin D 5 ?g,
vitamin E 10 mg, copper 2 mg, iodine 150 ?g, iron 30 mg, selenium 65 ?g, and zinc
15 mg 33]. The Bangladesh JiVitA trial used the same micronutrients as UNIMMAP in similar doses.
The supplement used in Nepal Sarlahi contained micronutrients in similar doses (with
60 mg iron), plus magnesium and vitamin K, but no selenium or iodine 34]. The supplement used in Mexico included iron 62.4 mg and magnesium 252 mg, and did
not include copper, iodine or selenium 25]. In some cases, a comparison group of iron 60 mg and folic acid 400 ?g was not available:
Nepal Sarlahi included additional vitamin A 34], Mexico did not include folic acid 25], Indonesia used 30 mg iron 29], and Bangladesh JiVitA used 27 mg iron and 600 ?g folate 26]. Supplement constituents are shown in Additional file 1: Table S1. Supplementation was initiated in early to mid-pregnancy, with a range
of median commencement gestation across studies of 14 weeks (Table 1).
Follow-up reports
We found 20 follow-up reports (Table 2 and Additional file 1: Figure S1). We divided the findings into five general categories: mortality 26], 27], 35]–41], anthropometry 35], 38], 39], 41]–44] and body composition 39], 44], 45], cardiovascular 39], 43], 46], 47], cognitive 37], 48]–51], and respiratory 52]. Primary publications from the Bangladesh JiVitA and MINIMat trials included follow-up
mortality data and were included in the list of follow-up reports. Meta-analyses were
conducted for mortality, weight, height, head circumference and blood pressure outcomes.
Table 2. Description of follow-up reports
Mortality
Follow-up reports from all trials systematically recorded and reported infant/child
mortality as an outcome (Table 3). Meta-analysis showed no difference between intervention and control groups (risk
difference, –0.05 per 1000 livebirths; 95 % CI, –5.25 to 5.15; I
2
, 8 %; Fig. 1). No difference by age was seen. Subgroup analysis including trials that used only
the UNIMMAP supplement showed a risk difference of 3.41 per 1000 livebirths (95 %
CI, –4.45 to 11.26; I
2
, 0 %; Additional file 1: Figure S2). Subgroup analysis for trials that used 60 mg iron in control groups
yielded a risk difference of 4.51 per 1000 livebirths (95 % CI, –2.91 to 11.94; I
2
, 0 %), and trials that used approximately 30 mg iron in the control group yielded
a risk difference of 0.41 per 1000 livebirths (95 % CI, –14.76 to 15.57; I
2
, 62 %; Additional file 1: Figure S3).
Table 3. Mortality outcomes
Fig. 1. Forest plot showing mortality rate per 1000 livebirths (meta-analysis using a random
effects model)
Anthropometry
Seven reports described anthropometry (Table 4). No differences were seen in any report at the most recent follow-up for WAZ, HAZ
or head circumference, nor in any secondary anthropometric outcomes. Differences were
seen at younger ages in two trials. In the Burkina Faso trial, greater mean WAZ (?,
0.13; 95 % CI, 0.04 to 0.23) and length-for-age z score (?, 0.13; 95 % CI, 0.02 to 0.24) were seen in the multiple micronutrient supplement
group, and a lower proportion were stunted at 1 year (hazard ratio, 0.73; 95 % CI,
0.60 to 0.87), but there was no difference at 2.5 years of age 35]. In the Nepal Janakpur trial, greater mean WAZ (?, 0.14; 95 % CI, 0.00 to 0.27) was
seen in the multiple micronutrient supplement group at a mean of 2.5 years, but there
was no difference at 8.5 years (?, 0.05; 95 % CI, –0.09 to 0.19). Small increases
were seen in head, chest, hip, and mid-upper arm circumferences at 2.5 years, but
were also not present at 8.5 years 39], 43].
Table 4. Anthropometry, body composition and cardiovascular results
Effect modification by maternal BMI or child sex was not found in any report with
the exception of the Bangladesh MINIMat trial, in which stunting was greater in boys
in the MMN group (Males, 7.8 %; 95 % CI, 2.0 to 13.6; Females, 1.8 %; 95 % CI, –3.8
to 7.3). A test for interaction was not reported.
Meta-analyses for WAZ and HAZ showed no difference between multiple micronutrient
and 60 mg iron and folic acid groups. The differences in WAZ and HAZ were 0.02 (95 %
CI, –0.03 to 0.07; I
2
, 0 %; Fig. 2) and 0.01 (95 % CI, –0.04 to 0.06; I
2
, 0 %; Fig. 3), respectively. Meta-analysis for head circumference was possible for three trials
(Mexico, Bangladesh MINIMat and Nepal Janakpur) and showed no difference (0.11 cm;
95 % CI, –0.03 to 0.26; I
2
, 0 %; Fig. 4). Subgroup analysis including UNIMMAP trials made little difference: WAZ 0.04 (95 %
CI, –0.01 to 0.09; I
2
, 0 %), HAZ 0.01 (95 % CI, –0.05 to 0.06; I
2
, 0 %), and head circumference 0.11 cm (95 % CI, –0.05 to 0.26; I
2
, 10 %; Additional file 1: Figure S2).
Fig. 2. Forest plot showing weight-for-age (meta-analysis using a random effects model)
Fig. 3. Forest plot showing height-for-age (meta-analysis using a random effects model)
Fig. 4. Forest plot showing head circumference (meta-analysis using a random effects model)
Body composition
The Bangladesh MINIMat trial found no difference in biceps, triceps, subscapular or
suprailiac skinfold thicknesses. The Nepal Janakpur trial found an increase in triceps
skinfold thickness at 2.5 years (0.20 mm; 95 % CI, 0.00 to 0.40 mm) 43], but no difference was found in any skinfold thickness at 8.5 years 39]. The Nepal Sarlahi trial found no difference in triceps or subscapular skinfold thickness
at 7.5 years of age. Neither the Bangladesh MINIMat nor the Nepal Janakpur trial found
a difference in lean mass or fat mass measured using bio-impedance 39], 45], 53], 54].
Cardiovascular risk markers
Cardiovascular outcomes were only examined in trials from South Asia (Table 4). The Bangladesh MINIMat and Nepal Janakpur trials measured blood pressure, while
the Nepal Sarlahi trial investigated metabolic syndrome (blood pressure, HbA
1c
, urine microalbumin:creatinine, cholesterol, glucose, insulin, homeostasis model
assessment of insulin resistance). The Nepal Janakpur cohort showed a reduction in
mean systolic blood pressure of 2.5 mmHg (95 % CI, 0.47 to 4.55) at 2.5 years of age,
but no difference at 8.5 years (0.02 mmHg; –1.02 to 1.05) 39], 43]. The Bangladesh MINIMat cohort showed no difference at 4.5 years compared with a
control group who received iron 30 mg and folic acid 46]. The Nepal Sarlahi trial found neither a difference in blood pressure at 7.5 years,
nor a difference in other cardiovascular risk markers compared with a control group
who received iron 60 mg and folic acid 47]. Meta-analysis of the three trials showed no difference in blood pressure: the difference
in systolic blood pressure was 0.11 mmHg (95 % CI, –0.41 to 0.63; I
2
, 0 %), and in diastolic pressure 0.47 mmHg (95 % CI, –0.01 to 0.95; I
2
, 0 %; Additional file 1: Figure S4). Subgroup analysis for trials that used iron 60 mg in the control group
was similar: systolic blood pressure difference 0.16 mmHg (95 % CI, –0.54 to 0.87;
I
2
, 0 %), and diastolic 0.37 mmHg (95 % CI, –0.35 to 1.08; I
2
, 0 %; Additional file 1: Figure S3).
Cognitive function
The Bangladesh MINIMat, China, and Indonesia trials all assessed subgroups of children:
in Bangladesh, children born in a 19-month period, at 7 months of age 48]; in China, in the middle year of the trial up to 1 year of age 49] and at 8.8 years 50]; and in Indonesia, in women assigned to blood tests and whose children were born
in a 6-month time period, at 3.5 years of age 37]. The Nepal Sarlahi study administered cognitive, motor and executive function tests
at 7–9 years of age. Mean cognitive scores were a little lower for the MMN group compared
to iron, folic acid and vitamin A (Universal Nonverbal Intelligence Test score, –2.4;
95 % CI, –4.6 to –0.2). Results of motor and executive function tests were mixed 51]. The Bangladesh MINIMat and China trials found no difference in motor or psychomotor
scores 48], 49]. The Indonesia trial found an increase in motor ability in an adjusted analysis expressed
as a fraction of the variation of the score (0.19; 95 % CI, 0.02 to 0.37) 37]. The Bangladesh MINIMat follow-up found no difference in problem solving or behaviour
48]. In the China trial, there were no differences at 3 or 6 months, but age-adjusted
scores at 1 year were higher for mental development in the MMN supplementation group
(1.20 points; 95 % CI, 0.32 to 2.08: equivalent to about 6 days of age) 49]. At 8.8 years of age, there were no differences in cognitive scores compared to folic
acid or to iron and folic acid groups 50]. The Indonesia trial follow-up found no difference in visuospatial and visual attention,
executive functioning, language ability, or socioemotional development 37].
For completeness, we mention stratified analyses. In the Bangladesh MINIMat trial,
stratification by maternal BMI, combining the early and usual food groups together,
showed an increase in Psychomotor Development Index in children whose mothers had
been allocated to MMN supplements and had BMI??18.5 kg/m
2
(0.22 z scores; 0.01 to 0.42; interaction term, P?=?0.05) 48]. In Indonesia, MMN supplementation was associated with greater motor ability (??=?0.35
z scores; 95 % CI, 0.01 to 0.69, interaction term, P?=?0.04) and visual attention/spatial ability (??=?0.35; 95 % CI, 0.08 to 0.63, interaction
term, P?=?0.01) in children of undernourished mothers (MUAC, 23.5 cm). These differences
were equivalent to approximately 5 months of age 37].
Respiratory
The Nepal Janakpur trial investigated lung function at 8.5 years of age. Spirometry
data were obtained from 836 children, with 793 (95 %) achieving optimal results using
American Thoracic Society/European Respiratory Society guidelines 55]. No difference in lung function was found between allocation groups: forced expiratory
volume in the first second, –0.01 L (95 % CI, –0.04 to 0.02 L); forced vital capacity,
–0.01 L (95 % CI, –0.04 to 0.02 L) 52].
Assessment of bias
Trials
The trials were considered high quality and bias was not thought to be important.
There was potential selection bias in the Guinea-Bissau trial as a result of inadequately
concealed allocation 24], and potential attrition bias for the Nepal Janakpur and Mexico trials, in which
exclusions prior to randomization were not reported 25], 30].
Follow-up reports
The trials were powered on the primary outcomes of gestational age and birthweight
(and mortality, in the case of the Bangladesh JiVitA Indonesia trials) 29]. Follow-up reports described power or sample size calculations before data collection,
with the exception of the Bangladesh MINIMat (cardiovascular) 46], Burkina Faso (anthropometry) 35] and Nepal Sarlahi (anthropometry and cardiovascular reports) follow-up publications
44], 47]. There was little statistical heterogeneity, with I
2
values low for primary analyses (0–8 %). Some clinical heterogeneity was present as
participants were from different countries and ages of follow-up varied. The intervention
was the same in most cases, but (as described above) the Bangladesh JiVitA, Mexico
and Nepal Sarlahi trials used slightly different multiple micronutrient formulations,
and the Bangladesh JiVitA, Indonesia, Mexico and Nepal Sarlahi trials used different
controls. Although choice of outcomes varied from one report to another, similar methods
were used to assess similar outcomes.
Selection bias
Primarily a result of inadequate randomisation and allocation concealment, this has
been covered in the 2015 Cochrane assessment 20]. An additional potential source of bias is selection of trials from the Cochrane
review that did not show an increase in birthweight associated with MMN supplementation.
The Cochrane review included 14 trials in its analysis of SGA. Of the five not considered
here 56]–60], one showed a significant reduction in SGA (Fawzi et al. 57], Tanzania; RR, 0.79; 95 % CI, 0.70–0.89). Supplement composition was substantially
different in this trial, which compared a supplement containing eight vitamins and
no minerals with a 60 mg iron and 250 ?g folic acid control 57]. Meta-analysis of the trials included in our review showed an increase in birthweight
of 30.2 g (95 % CI, 14.1 to 46.3), which is similar in magnitude to that found in
previous meta-analyses 11], 61]–63]. Similarly, three of the 11 trials included in the meta-analysis of neonatal mortality
did not conduct follow-up studies. None of these trials showed a reduction in neonatal
mortality. Meta-analysis of neonatal mortality rate for included trials produced an
RR of 1.01 (95 % CI, 0.90 to 1.16).
Performance and detection bias
Participants and data collectors in all follow-up reports remained blind to allocation,
with the exception of Guinea-Bissau, where this was not mentioned explicitly in the
report.
Attrition bias
While all reports described loss to follow-up (Table 2), attrition bias was relatively small (0–29 %), except in the Mexico trial, in which
just over half the children were seen at 24 months. The largest group lost were too
old (3 months) at the start of the follow-up. Excluding this group, follow-up rates
were similar to those of the other reports. No important differences in loss to follow-up
between allocation groups were reported, with the exception of China at 9 years, where
study groups differed by school type, recent respiratory tract infection, mother’s
occupation (farmer or other) and father’s level of education. These biases work in
opposite directions and were accounted for in the analyses 50]. Where recorded, differences between children retained and lost to follow-up were
small. Children lost to follow-up tended to have mothers with more education (Bangladesh
MINIMat, Nepal Janakpur, Mexico and Nepal Sarlahi), lower parity and younger age (Bangladesh
MINIMat and Mexico), and were more likely to live in an urban location (Nepal Janakpur),
have differences in ethnicity and assets (Nepal Sarlahi), and lower birthweight and
shorter gestation (Bangladesh MINIMat). Maternal age, weight, height and parity also
differed in Guinea-Bissau, but the directions of these effects were not reported.
Although most reports did not enumerate them, losses for individual outcomes also
occurred: (20 % loss to measurement) kidney volume and function in the Bangladesh
MINIMat trial, body composition in the Nepal Janakpur trial, and fasting glucose,
insulin and homeostasis model assessment in the Nepal Sarlahi trial.
Reporting bias
We could not make a definitive assessment of reporting bias as follow-up protocols
were unpublished, but funnel plots for mortality, HAZ, WAZ and head circumference,
using results from the most recent follow-up report, did not suggest publication bias
for the primary outcomes (Additional file 1: Figure S5).