Bulletin of the World Health Organization

Neonatal vitamin A supplementation and infant mortality

Harshpal Singh Sachdev a, Betty R Kirkwood b & Christine Stabell Benn c

a. Sitaram Bhartia Institute of Science and Research, B-16 Qutab Institutional Area, New Delhi 110016, India.
b. Faculty of Epidemiology & Population Health, London School of Hygiene & Tropical Medicine, London, England.
c. Bandim Health Project, Statens Serum Institut, Copenhagen, Denmark.

Correspondence to Harshpal Singh Sachdev (e-mail: hpssachdev@gmail.com).

Bulletin of the World Health Organization 2010;88:875A-875B. doi: 10.2471/BLT.10.083030

The recent publication in the Bulletin on this subject1 presents a study whose findings lead to an erroneous conclusion that will misinform global health policy-makers regarding the benefits of neonatal vitamin A supplementation. The authors' stated objective was “to assess the relationship between the prevalence of vitamin A deficiency among pregnant women and the effect of neonatal vitamin A supplementation on infant mortality”. However, they excluded relevant studies, used inappropriate data on the prevalence of vitamin A deficiency in pregnant women and drew unsubstantiated conclusions.

To begin with, three relevant trials were not included in the analysis.24 One such trial2 was published in March 2010, before the authors submitted their paper. Another was excluded because it was conducted in infants born to HIV-negative mothers,3 but the authors do not explain why they viewed these participants as not belonging to the “general population”. A third trial4 was excluded because the vitamin A supplements were administered within the first month of life (i.e. in the neonatal period, as officially defined) rather than within the first 48 hours. However, the alleged differential effect of administering vitamin A in the first 48 hours of life as opposed to later in the neonatal period is not supported by any existing data.5

The authors have obtained prevalence figures for maternal vitamin A deficiency from a table that is no longer accessible at the site for which a link was provided.1 We fortunately had access to the data in paper format. The authors claim that “the included values represent the best estimates of the prevalence of vitamin A deficiency in pregnant women at the time individual subjects were enrolled in each study”. However, this is not correct. The prevalence figures for maternal vitamin A deficiency that were used are largely based on studies conducted in different time periods or on “guesstimates”, as witnessed by an unlikely prevalence of precisely 20.4% in several countries. Extrapolation of such national-level prevalence data to individual study subjects is prone to serious error. For example, in the Indonesian trial the imputed prevalence of vitamin A deficiency 6 was 34.2%, higher than in any of the other included trials. However, the women who participated in the trial had mean serum retinol levels very similar to those observed in a reference population of white women of reproductive age in the United States.6 Thus, the true prevalence of vitamin A deficiency among study subjects in the Indonesian trial, where the largest treatment effect was observed, was probably lower than in any of the other included trials.

Conclusions emanating from a meta-regression of the results of four trials are fragile. The Cochrane guideline states that “meta-regression should generally not be considered when there are fewer than 10 studies in a meta-analysis”. It further warns that even this will be too few when the covariates are unevenly distributed.7

Based on their analysis, the authors conclude that neonatal vitamin A supplementation may prove most beneficial for reducing infant mortality in settings where the prevalence of vitamin A deficiency among pregnant women is at least 22%. This cut-off is just below the prevalences of 22.5% and 22.8% used for two countries where trials showed a positive impact, yet barely above the 20.4% prevalence in Guinea-Bissau, where no benefit was found. Furthermore, the cut-off is not supported by Rotondi et al.’s figure, in which the relative risk (RR) of 1 (log(RR) of 0) corresponds to a prevalence of approximately 20.5% instead of 22%. Irrespective, it is biologically implausible that the effect of vitamin A supplementation should depend on such marginal differences in maternal vitamin A deficiency.

Finally, the title of the manuscript is inaccurate, since the outcome of interest in the included trials was infant mortality, not neonatal mortality, and three of the four included trials were individually randomized.

Although we consider the use of national-level vitamin A data of questionable relevance, we performed meta-regression using the “metareg” command in Stata version 9.2 software (StataCorp. LP, College Station, United States of America) with the restricted maximum likelihood option, and we included all relevant trials 24,6,810 to determine whether this analysis would have changed the authors’ conclusions. The national prevalence of maternal vitamin A deficiency turned out not to be a significant predictor of the effect of vitamin A supplementation on infant mortality, as expressed by the log(RR) (β: 0.003; 95% confidence interval: −0.022 to 0.029; P = 0.76) in this meta-regression (details available upon request).

The tendency to influence global policy on the basis of selective and inappropriate analysis of available data needs to be strongly discouraged. None of the evidence available at present provides grounds for determining where neonatal vitamin A supplementation is likely to reduce infant mortality based on the population prevalence of this micronutrient deficiency. Global policy formulation must await further input from the four ongoing trials evaluating this intervention1114 as well as documentation of consistent trends in these and previous trials.

Competing interests:

None declared.