Effects of Vitamin A or Beta Carotene Supplementation on Pregnancy-Related Mortality and Infant Mortality in Rural Bangladesh:  A Cluster Randomized Trial FREE

Maternal vitamin A deficiency appears to be widespread in low-income countries. Currently, the World Health Organization estimates nearly 20 million pregnant women to be vitamin A deficient, exhibiting a serum retinol concentration below 0.70 μmol/L, of whom nearly 9 million have gestational night blindness, an ocular manifestation of deficiency.¹ Night blindness during pregnancy is associated with increased risks of maternal anemia, morbidity, and mortality,^2- 3 suggesting that preventing vitamin A deficiency could improve maternal survival. Evidence of such an effect has been reported in rural Nepal, where a randomized controlled trial demonstrated a decrease of approximately 44% in mortality related to pregnancy following continuous, weekly receipt of vitamin A or beta carotene during the reproductive years at dosages approximating a recommended daily allowance.⁴ The reduction was most apparent in women with night blindness,³ and causes of death attenuated in risk were those involving infectious, obstetric, and miscellaneous causes as determined by interviews with family members.⁴ Supplementation also modestly decreased reported symptoms of morbidity.⁵ Plausible host defenses against severe illness, which likely rely on vitamin A–regulated cell proliferation and differentiation and could be compromised by vitamin A deficiency, include multiple mechanisms of epithelial innate and adaptive immunity,⁶ hematopoiesis,^7- 8 and coagulation,⁹ abnormalities of which could exacerbate infection, result in anemia, or impair wound recovery.⁹ Although in a trial conducted in Nepal there was no overall effect on infant mortality,¹⁰ there was evidence of a reduction in mortality among infants born to mothers with a positive history of gestational night blindness.¹¹

Given the importance of reducing maternal mortality, additional vitamin A intervention trials have been indicated. A recent trial conducted in Ghana, West Africa,¹² reported no effect of weekly supplementation on all-cause maternal mortality with a comparable dosage of vitamin A, suggesting that local nutritional, health, and vital risk contexts may influence effect of supplementation. In Bangladesh, we sought to extend the findings of the Nepal trial to another South Asian setting, delivering the same dosage of vitamin A or beta carotene to women from early pregnancy through 12 weeks postpartum to assess effects on pregnancy-related maternal, fetal, and early infant mortality.

We carried out a cluster randomized, double-masked, placebo-controlled trial in northwestern Bangladesh from August 2001 through January 2007. Its purpose was to determine the efficacy of providing an oral supplement containing the weekly equivalent of an RDA of vitamin A, either preformed or as beta carotene, from the first trimester of pregnancy through 12 weeks (84th day) after pregnancy termination, in reducing all-cause maternal mortality. We also assessed the efficacy of supplementation in reducing stillbirth and infant mortality through 12 weeks of age.

Our aim was to conduct the trial in a reasonably accessible, typical rural setting of the country that could be identified from reviews of available reports on population, health, nutrition, agriculture and infrastructure, site visits, and discussions with officials, community groups, and health and nutrition experts. The study site chosen, in the Division of Rangpur, comprised 18 contiguous, rural unions (subdistrict-level administrative divisions) in Gaibandha District and 1 adjacent union in Rangpur District, covering an area of approximately 435 km² with an estimated population of approximately 600 000. It was an area that appeared to have levels of undernourishment, health burdens—including vitamin A deficiency based on previous reports of maternal night blindness¹³—mortality risks, and societal characteristics typical of rural Bangladesh.¹⁴ Cadastral maps, drawn in the 1930s by the British Geological Survey and obtained from the Government of Bangladesh, provided unique, initial details of community boundaries, updated using handheld global positioning devices. We initially located and assigned unique addresses to 121 201 residences plus more than 100 000 other community landmarks (eg, public facilities, road intersections, tube wells) to enable field work and supervision.¹⁵

The site was divided into 596 study “sectors” of initially 109 to 377 households each to create community clusters of comparable population size and land area that could serve both as units of randomization and individual staff assignment. Changes in households and landmarks were regularly updated throughout the trial. Details of area characteristics and study design, organization, management, training, and quality control are described elsewhere.¹⁴

The primary outcome of interest in this trial was all-cause mortality related to pregnancy through 12 weeks after the end of pregnancy. We initially assumed a baseline mortality rate of at least 600 maternal deaths per 100 000 pregnancies, rounded up from an earlier estimate of 570 from previous surveys¹⁶ because of the area's remote, rural setting, where undernutrition, poor routine and emergency health care, and high maternal mortality often co-exist.^17- 18 Restricting the number of eligible pregnancies to 1 per woman, the sample size required to detect a 35% or greater reduction in all-cause mortality in either supplement group relative to control, conservatively estimated from the Nepal trial,⁴ was set at 54 000 (18 000 per group). This sample size estimation allowed for a type I error of 5%, power of 80%, equal supplement allocation, design effect for pregnancy-related mortality of 1.21,⁴ early pregnancy loss rate of 15% (12% miscarriage, 3% abortion), and loss to follow-up of 10%. The sample size was adequate to detect reductions of 15% or greater each in rates of stillbirth and infant mortality.

Meetings by the trial's data and safety monitoring board (DSMB) guided conduct decisions throughout the trial. During a DSMB meeting in November 2003, preliminary analyses began to reveal unexpectedly lower maternal mortality rates (≤500 vs 600 per 100 000 pregnancies) and higher abortion rates (12% vs 3%), although fewer losses to follow-up (2% vs 10%), leading us to revise upward the sample size to 67 740 pregnancies. In December 2006, with this recruitment goal attained, a DSMB analysis revealed no clinically or statistically significant differences, and an extremely low probability of finding any in primary (maternal) or secondary (fetal and infant) vital outcomes, across groups.

Given the futile projection, and aware of the large budget and more than 1-year period of field work that would be needed to complete follow-up for the approximately 7500 women remaining in the trial to 12 weeks after pregnancy, the DSMB recommended the trial be halted early. The closure date for pregnancy recruitment was set for January 4, 2007. Enrolled participants were notified by staff, and supplementation and pregnancy ascertainment activities of the trial ceased by February 1, 2007, leaving 60 294 pregnancies identified in our trial cohort (Figure 1).

Prior to the trial, sectors were contiguously listed geographically, assigned a number from 001 to 596, and randomized to 1 of 3 codes (1, 2, or 3), each representing 1 of the 3 batches of supplements confidentially labeled by the supplement packer and shipper (Amway Nutrilite Health Institute, Buena Park, California). Sectors were randomized, blocked on every ninth, to achieve geographic and numerical balance using a traditional method whereby 9 coins of identical size (3 per code) were drawn in a blinded fashion and assigned to each sequential sector. The blocked drawing exercise was repeated 67 times until all 596 sectors were assigned codes representing placebo (n = 198), vitamin A (n = 198), or beta carotene (n = 200). Sectors were geographically balanced by allocation as reported elsewhere.¹⁴

Three batches of encapsulated oily supplements of identical size and color were produced twice under contract for the trial in mid-2001 (Scherer Pharmaceuticals, Orlando, Florida) and late 2004 (Cardinal Health, Dublin, Ohio), containing 7000 μg of retinol equivalents as retinyl palmitate, 42 mg of all -trans beta carotene, or neither (placebo), all with 5 IU of vitamin E to extend shelf life. Supplements were blind-tested for potency, packed into opaque, plastic bottles labeled as batches 1, 2, or 3 (representing supplement codes), and air-freighted gratis to Bangladesh by the Amway Nutrilite Health Institute, where they were stored under protected conditions. Prior to dispensing to the field, bottles were relabeled with 3-digit sector numbers to ensure correct delivery and to further mask teams. Annually, capsules from storage were blind-tested for nutrient content (Medallion Laboratories, Minneapolis, Minnesota), yielding mean potencies of 6568 (SD, 1220) retinol equivalents as retinyl palmitate (94% of declared), 45.6 (SD, 4.3) mg of beta carotene (108% of declared), and undetectable amounts of either in placebos.

All data collectors were women. From early July to mid-August 2001, mapped and addressed households were visited by sector staff to enlist married women of reproductive age (13-45 y) into a surveillance system designed to recruit first trimester gravida. Pregnant women beyond their first trimester, assessed by eliciting a date of the last menstrual period, plus those breastfeeding an infant younger than 12 months and likely to be amenorrheic as a result of breastfeeding, were wait-listed until after their first year postpartum, when they became eligible for surveillance. Every 5 weeks thereafter, households were revisited and addresses updated, vital status of enlisted women was recorded, and newly married women were registered. Registrants were delisted if they had permanently moved, become menopausal or had been sterilized, died, or if their husbands had died. At each round, women were asked about menstruation in the previous month.

Women reporting amenorrhea were offered a urine test for detecting human chorionic gonadotropin at a concentration of 10 mIU/mL or greater (Clue; Orchid, Pune, India). Women were informed of the result. Within a week, women testing positive were informed by field leaders about the purpose, activities, types of supplements, potential risks and benefits, and voluntary nature of the trial, following standardized scripts. Women providing oral informed consent, as documented by staff, were enrolled into the trial. No incentives were provided for participation. Educational materials about care and diet during pregnancy were provided to all pregnant women, irrespective of participation.

Every week pregnant women were visited and given a coded supplement by local resident staff until 12 weeks postpartum. Records of weekly visits were kept in each home to facilitate oversight by supervisors and a trained quality control team.¹⁴ If a mother was temporarily absent, a capsule was left at the home with instructions to take the supplement. For participants gone longer (eg, to a parental home), families were provided 4 capsules for women to take while away and resupplied as needed. No other nutritional supplements were provided.

A 1-week history of morbidity, diet, and routine work; a lifetime pregnancy history; socioeconomic profile¹⁹; and mid–upper-arm circumference measurement were obtained shortly after providing consent, usually during the first trimester. Assessments were repeated 12 weeks after pregnancy, and infants were evaluated for growth, morbidity, and feeding patterns. At approximately 28 weeks' gestation, participants were visited at home to obtain a history of night blindness. Those with night blindness were treated with an oral regimen of 25 000 IU of vitamin A weekly²⁰ for 4 to 8 weeks, regardless of supplement group.

A report of pregnancy loss prompted home visits by trained interviewers to ascertain cause (ie, miscarriage, induced abortion, or stillbirth) and health status of the mother. Physicians trained in the conduct of verbal autopsy interviews met with family members of deceased women to elicit a history of illnesses, events, and treatments sought for the woman via closed-ended questions and open-ended histories. These were independently reviewed by 2 physicians, including an obstetrician-gynecologist, who met to assign consensus proximal and underlying or contributory causes of death. Trained interviewers conducted verbal autopsies with parents of deceased infants, using an instrument focusing on neonatal and postneonatal causes of death. Forms were independently reviewed by 2 or more physicians, who subsequently discussed and assigned consensus and underlying or contributory causes of death.

Clinical and biochemical studies were conducted in an area of the trial that comprised 32 contiguous sectors (5% of all, 10-12 per supplement group) involving additional maternal assessments in the first and third trimesters and at 12 weeks postpartum. Procedures included maternal anthropometry, clinical tests (eg, for bacterial vaginoses, proteinuria, blood pressure), phlebotomy, and breast milk sampling for nutritional analyses. Infant assessments included anthropometry, usually within 72 hours after birth and at 12 weeks of age, and heel-stick blood sampling at 12 weeks. Blood was processed to plasma and aliquoted for shipping under liquid nitrogen to the Johns Hopkins Micronutrient Analysis Laboratory (Baltimore, Maryland) for bioarchiving and analyses and to the Institute of Nutrition at Mahidol University (Salaya, Thailand), where plasma retinol and β-carotene concentrations were determined using high-performance liquid chromatography, following the methods of Yamini et al.²¹

Baseline characteristics of enrolled pregnant women were summarized and compared across groups. We evaluated adherence by dividing the number of supplements directly consumed or provided to women by the number of total supplements that could have been taken during each participant's time in the trial. We then plotted the percentage of possible supplements directly given or received by the cumulative percentage of women attaining each level of adherence. Supplemental nutritional effects were identified as differences in third-trimester and 12-weeks postpartum plasma retinol and β-carotene concentrations among substudy mothers.

Primary outcomes were compared on an intent-to-treat basis. To directly compare with the all-cause mortality findings from the Nepal trial,⁴ we divided the number of maternal deaths during gestation through the 12th week (84th day) postpartum by the number of enrolled pregnant women per 100 000. Distributions of causes of death were derived from verbal autopsy data, and rates of mortality attributed to direct and indirect proximal causes were calculated. Rates of stillbirth (born ≥28 weeks' gestation without moving or crying) were calculated per 1000 live births plus stillbirths. All-cause and consensus-cause infant mortality rates were calculated from numbers of deaths occurring from birth through the 84th postnatal day per 1000 live births.

In comparing effects of vitamin A or beta carotene vs placebo, the significance of differences in nutrient status was tested using generalized estimating equations linear regression analysis, with an identity link function and exchangeable correlation appropriate for normally distributed data. For maternal deaths, stillbirths, and infant deaths, relative risks with 95% confidence intervals (CIs) were estimated by generalized estimating equations binomial regression, with a log link function and exchangeable correlation appropriate for binary data. Both analyses account for the design effect associated with sector vs individual randomization.²² We set an a priori level of statistical significance at P < .05. All analyses were performed using SAS version 9.2 (SAS Institute Inc, Cary, North Carolina).

The protocol was approved by the Bangladesh Medical Research Council, Dhaka, and the institutional review board of the Johns Hopkins Bloomberg School of Public Health, Baltimore, Maryland. A DSMB of international membership was formed in the fall of 2001 and met 4 times thereafter through December 2006 to monitor recruitment, adherence, efficacy, and safety.

A total of 102 769 married women of reproductive age were enumerated for pregnancy surveillance at the outset across the 596 study sectors, with a total of 33 960 to 34 427 women, and a median of approximately 170 women per sector, in each of the 3 randomized groups (Figure 1). After excluding 5223 women, mostly because they had been sterilized or had become menopausal, and adding 27 711 newlyweds over the course of the trial, a total of 125 257 women were ever under pregnancy surveillance, with 32 180 to 32 719 in each group.

Because the study was halted early, we identified a cohort of women whose pregnancies started on or after August 17, 2001 (date of first pregnancy recruited), whose last menstrual period, signaling conception, occurred before January 5, 2006, and whose pregnancies ended, irrespective of outcome, on or before October 12, 2006. These criteria enabled as many participants and their live-born infants as possible an opportunity to complete 84 days of postnatal follow-up by the trial's close-out date of January 4, 2007. A total of 60 294 identified pregnancies met the inclusion criteria (20 060 in the placebo group, 20 012 in the vitamin A group, and 20 222 in the beta carotene group). Of these, 59 666 (99% in each group) consented, had a known pregnancy outcome, and had vital status recorded at 12 weeks after pregnancy (19 862 in the placebo group, 19 806 in the vitamin A group, and 19 998 in the beta carotene group) (Figure 1). These participants form the cohort of pregnancies in this analysis, among whom approximately 70% in each group gave birth to a live infant, leaving 13 972, 13 905, and 14 020 live-born infants in each respective group to constitute the cohort of infants in this analysis.

First trimester maternal characteristics were comparable across supplement groups, with 36% being younger than 20 years, 34% being nulliparous, 25% having a thin upper arm (mid–upper-arm circumference <21.5 cm), and 34% to 83% having eaten dietary sources of vitamin A (meat/liver, fish or crustaceans, dairy products) and carotenoids (yellow fruits and vegetables, dark green leaves), and 72% to 73% having chewed betel nut in the previous week. Eighty-five percent enrolled into the trial by 12 weeks' gestation (Table 1). Other supplements (iron and folic acid) were reported to have been taken during any time of pregnancy by 16.1% in the placebo group, 15.6% in the vitamin A group, and 15% in the beta carotene group.

Adherence to supplementation was comparable across groups, with approximately 80% of women having directly consumed (under staff supervision) at least 64% of eligible supplements. Inquiry about receipt of supplements left at home showed that 80% of participants received in total at least 94% of supplements (Figure 2). Within substudy sectors, plasma retinol distributions were comparable, withmean concentrations of 1.11 to 1.14 μmol/L (to convert to μg/dL, divide by 0.0349) and approximately 7% of mothers in each group initially classified as vitamin A–deficient, having a concentration less than 0.70 μmol/L (Table 2). By the third trimester, 14.2% in the placebo group vs 1.6% in the vitamin A group (P < .001) and 10.6% in the beta carotene group (P = .15) had deficient plasma retinol concentrations. These differences were sustained to 12 weeks postpartum. Weekly beta carotene supplementation had little effect on plasma retinol concentrations and modestly increased plasma β-carotene concentrations over placebo at both follow-up visits, by 0.01 μmol/L (P = .03) and 0.03 μmol/L (P < .001), respectively. Night blindness occurring anytime during pregnancy, assessed after pregnancy by history, was lower in the vitamin A (7.1%) than either the placebo (9.2%, P < .001) or beta carotene (8.9%, P < .001) groups. The relative risks of having night blindness in pregnancy were 0.77 (95% CI, 0.70-0.85) in the vitamin A group and 0.96 (95% CI, 0.88-1.05) in the beta carotene group.

There were 138 maternal deaths from all causes, with 41, 47, and 50 in the placebo, vitamin A, and beta carotene groups, respectively, yielding mortality rates of 206, 237, and 250 deaths per 100 000 pregnancies. Relative risks for maternal mortality were 1.15 (95% CI, 0.75-1.76) in the vitamin A group and 1.21 (95% CI, 0.81-1.81) in the beta carotene group. The combined relative risk for receiving either supplement was 1.18 (95% CI, 0.83-1.69). Compared with placebo, the risk ratio for direct obstetric causes, which combines deaths attributed to hemorrhage, eclampsia or preeclampsia, obstructed labor, puerperal sepsis, retained placenta, and shock based on verbal autopsy reviews, was 1.37 (95% CI, 0.75-2.52) in the vitamin A group and 1.57 (95% CI, 0.88-2.80) in the beta carotene group (Table 3).

There were 703 stillbirths in the placebo group, 665 in the vitamin A group, and 766 in the beta carotene group, yielding rates of 47.9 (95% CI, 44.3-51.5), 45.6 (95% CI, 42.1-49.2), and 51.8 (95% CI, 48.0-55.6) per 1000 births, respectively. Relative risks for infants being stillborn, compared with placebo, were 0.95 (95% CI, 0.85-1.06) in the vitamin A group and 1.08 (95% CI, 0.97-1.21) in the beta carotene group. Among 2834 infant deaths, 951 occurred in the placebo group, 904 in the vitamin A group, and 979 in the beta carotene group, yielding mortality rates of 68.1 (95% CI, 63.7-72.5), 65.0 (95% CI, 60.7-69.4), and 69.8 (95% CI, 65.4-72.3) deaths per 1000 live births, respectively, and relative risks of 0.95 (95% CI, 0.87-1.05) in the vitamin A group and 1.03 (0.94-1.12) in the beta carotene group (Table 4). We observed no discernable differences across groups in rates of infant death by consensus cause (per 1000 live births), which were dominated by prematurity (12.4-13.4), birth asphyxia (16.2-17.5), and sepsis (17.2-21).

In this rural setting of northern Bangladesh, weekly provision of a recommended daily allowance of vitamin A or equivalent as beta carotene failed to reduce mortality related to pregnancy or to decrease risks of stillbirth or infant mortality through 12 weeks of age. Throughout the trial masking was protected, safety monitored, pregnancy recruitment evenly accrued, baseline comparability attained, supplement potencies protected, and relatively high participant adherence maintained in the 3 groups. Following a DSMB recommendation, the trial was stopped for lack of effect and a cohort was defined to minimize outcome rate distortions. Although vitamin A deficiency was modest at baseline, the plasma retinol distribution appeared to normalize in women receiving vitamin A, while deteriorating in the placebo group. Beta carotene supplementation gradually increased plasma β-carotene concentrations, but only modestly affected plasma retinol concentrations, in relation to controls. These findings suggest that the lack of mortality effects appeared unrelated to aspects of study design, implementation, or supplement potency.

However, absence of a reduction in mortality among vitamin A and beta carotene recipients contrasts with the approximately 44% reduction in pregnancy-related mortality achieved by vitamin A and beta carotene supplementation in the trial conducted in Nepal, where dosages were comparable.⁴ It is possible that contextual differences between the Nepalese and Bangladeshi populations studied may have been responsible for the dramatic difference in outcomes, especially with respect to maternal mortality. In Bangladesh, mortality from early pregnancy through 3 months postpartum among placebo recipients was 206 per 100 000 pregnancies, a rate that was only about 30% of that observed among the Nepali placebo recipients,⁴ approximately 35% of the national estimate for Bangladesh prior to the trial,¹⁶ and just above the United Nation's lowest estimate of 170 for the country in recent years.²³ Among societal trends in Bangladesh that may have improved maternal survival are marked improvements in female secondary education.²⁴ In addition, another key difference was that Bangladeshi mothers were more likely to have a traditional or skilled health worker attend delivery than their Nepalese counterparts (42% vs 30%; odds ratio, 1.72 [95% CI, 1.65-1.78]) in the respective study areas.

The vitamin A status of Bangladeshi mothers was better than anticipated. In contrast to earlier prevalence reports of 25% night blindness,¹³ we observed that only 9% of women who received placebo reported the condition during pregnancy, a rate that was also below the range of 11% to 16% who reported the condition in Nepal.^3,25 Seven percent of Bangladeshi mothers had first trimester plasma retinol levels below 0.70 μmol/L, less than half of the prevalence (19%) among their Nepal counterparts.⁴ Better vitamin A status in the Bangladeshi women is consistent with their better diet; more mothers in Bangladesh reported consuming eggs and fish—foods rich in preformed vitamin A—in the previous week than in Nepal (K.P.W., et al, unpublished data, 2010). More generally, wasting malnutrition, reflected by upper-arm thinness, was also half as prevalent among Bangladeshi (26%) than Nepalese (>50%)²⁶ study participants.

While the mortality outcomes in Bangladesh were different from those seen in the Nepal trial, likely because of the contextual differences discussed above, they are similar to the outcomes of a large randomized controlled trial in Ghana, in which a comparable weekly supplement of vitamin A, delivered to enrolled women of reproductive age, failed to reduce all-cause pregnancy-related and all-female mortality.¹² The population studied in Ghana had a higher rate of all-cause maternal mortality (377 vs 206 deaths per 100 000 pregnancies), more biochemical vitamin A deficiency (15% vs 7%), and no night blindness, as compared with the population studied in Bangladesh. While other nutritional characteristics were not reported, Ghanaian women tend to be taller, heavier, and better nourished²⁷ than rural South Asian women, suggesting that other population factors may attenuate a maternal survival response to the same vitamin A intervention.

In this study, as in Nepal¹⁰ and Ghana,¹² neither maternal vitamin A nor beta carotene supplementation affected risks of stillbirth or early infant mortality. In Nepal, maternal vitamin A use appeared to lower mortality of infants born to mothers prone to night blindness,¹¹ an effect not evaluable in Bangladesh because in each group cases identified during the eighth month of pregnancy were treated with vitaminn A.

In conclusion, vitamin A deficiency is widespread in undernourished societies²⁸ and may pose health consequences such as night blindness^2,21,29- 31 and increased risks of infectious morbidity^5,32 and mortality^3- 4 to the mother. The effects of vitamin A on maternal survival may vary by severity of deficiency, mortality risk (by cause), general malnutrition, access to health services, and likely other factors. In this study, weekly supplementation of vitamin A and beta carotene in pregnant women in Bangladesh did not reduce all-cause maternal, fetal, or infant mortality. Irrespective of mortality effects, achieving maternal adequacy in vitamin A through diet, supplementation, or fortification is an important public health goal, especially in populations in which night blindness commonly occurs during pregnancy.