Previous - Next
This is the old United Nations University website. Visit the new site at http://unu.edu
Vitamin A supplementation and child morbidity and mortality in developing countries
George H. Beaton, Reynaldo Martorell, Kristan A. Aronson, Barry Edmonston, George McCabe. A. Catharine Ross and Bart Harvey
A meta-analysis of eight mortality trials indicates that improving the vitamin A status of children aged six months to five years reduced mortality rates by about 23% in populations with at least low prevalence of clinical signs of vitamin A deficiency. The observed effect of supplementation, described in terms of relative risk (RR), was RR =0.77 (95% confidence interval 0.68-0.88; p < .001) and did not differ by sex or age. However, the number of lives saved was greater at younger ages because of higher mortality. A significant RR was shown for deaths attributed to diarrhoea and measles, but not for respiratory infection. Variability among the trials in effects was apparent, but attempts to explain it by descriptors of the population (baseline anthropometric status, prevalence of xerophthalmia, age profile, baseline mortality) were unsuccessful. Owing to the lack of data, firm conclusions could not be reached about effectiveness in children of less than six months and in settings where biochemical but not clinical evidence of vitamin A deficiency exists.
Information about morbidity outcomes from about two dozen studies was reviewed. No consistent effects on frequency or prevalence of diarrhoeal and respiratory infections were found. Improvement in vitamin A status did appear to reduce severe morbidity, particularly in children with measles.
Beginning in the 1950s, periodic dosing with high levels of vitamin A was tested as a method of controlling xerophthalmia and blindness in children, particularly in India and Indonesia. Much experience was gained through those trials and operational programmes [1]. At the same time, descriptive epidemiological studies documented the association between xerophthalmia and both morbidity and mortality. To test the causality of this association, a randomized (unblinded) field trial of the effect of vitamin A supplementation on young child mortality was conducted in Aceh province, Indonesia [2]. In 1986 the investigators reported a 34% reduction in mortality in vitamin A-supplemented preschool children age 12 to 71 months at entry in comparison with a control group. For the total study (all ages), the reduction was about 26%. These findings prompted the United Nations Subcommittee on Nutrition to issue a statement citing the study and noting that young child mortality reduction might be an additional reason for increasing efforts to control vitamin A deficiency.
In the five years that followed this report, ten field trials of vitamin A supplementation and mortality were initiated. Most involved high-potency periodic dosing, but one used weekly administration of a low-dose supplement and another was based on food fortification. At the time of the preparation of the report on which the present summary is based, the findings of all the studies had been published in the literature or were available to us in the form of draft manuscripts and other reports. For two of the projects we did not have sufficiently detailed data to include them in our formal analyses. The sources of data for the meta-analysis are given in the Appendix, with more details appearing in Beaton et al. [3].
While these mortality trials were under way, much interest was directed to morbidity as an outcome. Some trials included morbidity measures; most collected only prevalence data. A number of smaller and more intensive (in data collection) trials also were initiated. We examined the results of 23 of them. By the end of 1993, the results of several more studies will have been reported.
Although the morbidity and mortality results for some available studies are incomplete, a sufficient volume of information permits serious questions to be asked about the effectiveness of vitamin A supplementation. Estimates of the reduction in mortality associated with supplementation ranged from over 50% (Tamil Nadu, Bombay) to no significant effect (Hyderabad, Sudan, unpublished results from Haiti). Understandably, the apparent divergence in results has led to both confusion and concern among potential users of the results.
Most of the morbidity examinations reported in the literature failed to detect an effect of vitamin A supplementation on the frequency, duration, and prevalence of diarrhoeal and respiratory infections. A few described beneficial effects, and preliminary analyses of at least one suggest detrimental effects. Recently a large morbidity trial (Ghana VAST) reported a beneficial effect on severe illness with little or no effect on the frequency or duration of general illness. Also emerging are suggestions that vitamin A may act differently depending on the nature of the illness; only a few mortality studies reported results by attributed cause of death. Morbidity trials usually provide information classified by symptoms.
In 1992 the United Nations ACC Subcommittee on Nutrition, acting on behalf of the interested UN agencies, as well as reflecting the interest of attending bilateral donor agencies, urged that there be an independent, objective review of the experience to date regarding the effects of vitamin A interventions on morbidity and mortality. A similar recommendation was voiced by the International Vitamin A Consultative Group (IVACG) that year.
In response to these recommendations, the present study was commissioned by CIDA. The specific terms of reference were to review and assess the available experience with regard to the effect of vitamin A supplementation on young child morbidity and mortality, to advise CIDA on the apparent effectiveness of vitamin A supplementation in young children in developing countries, and to estimate, as far as possible, the range of effects for mortality and morbidity outcomes expected under various nutritional and ecological circumstances and for various subgroups of the population. Specifically, the mandate did not include the analysis of policy, nor did it call for the development of proposed policy. Rather, the intent was to gather background information that CIDA and others might use in formulating policy, which would then provide guidance in planning their own future programmes.
In preparing the report, we received extensive cooperation from a large number of the principal investigators of the original studies who responded to our requests for specific information about their study design and results. A preliminary report on the analysis of mortality experience was distributed to investigators for comment and criticism in the spring of 1 992.
Identification and review of controlled trials
We identified and examined 10 mortality and 23 morbidity trials (including morbidity results from the 10 mortality trials). The Nepalese study by West and colleagues (1991; see Appendix) was extended to assess the mortality effect of dosing under age six months. Preliminary results presented at the 1993 IVACG meeting indicated no effect under 10 months. The trials were both published and unpublished studies for which we obtained descriptions from the primary investigators. The Appendix lists the mortality trials; the sources for the morbidity trials are given in Beaton et al. [3].
For several published studies, we obtained supplementary information from the original investigators. We are aware of other morbidity trials still under way, and of plans for further analyses of existing trials. However, we know of no additional trials now being conducted or approved for implementation in the near future. Therefore, for the mortality outcome, we think we have captured the total experience.
Our only shortfall is with regard to two studies, one in Bombay, India, and one in Haiti, for which we could not obtain the level of detailed information necessary for inclusion in our formal analyses. In contrast, we expect that substantial information will be forthcoming in the next year or two, and therefore urge that our conclusions be seen as a valid interpretation of experience to date but subject to possible modification when those data are available.
Did vitamin A supplementation have an effect on young child morbidity and mortality?
Mortality effect
We established a definitive yes answer with regard to mortality. Vitamin A supplementation resulted in an average reduction of 23% in mortality of infants and children between six months and five years of age (see FIG. 1. Impact of vitamin A supplementation on mortality of infants and children six months to five years old. At left are the point estimates and 95% confidence intervals for the eight original studies reviewed in detail (see Appendix for references). Also shown are two summary estimates for the relative effect, taking into account all eight studies—a fixed and a random-effect model, with study variances adjusted for cluster sampling effects. These have the same point estimates, a 23% reduction in mortality, but differ in the estimated confidence intervals; the second takes into account the between-study variation that we believe exists and technically is derived from a random-effects model. At the extreme right is the prediction interval for a future programme or study; again the predicted average effect is 23% but now the interval is greatly expanded (see text for explanation).). The effect (0.77) was highly significant and identical (relative risk [RR] = 0.77) under either of two conceptual models examined, a fixed effect model or a random effect model, although the 95% confidence intervals were somewhat wider for the latter (95% Cl 0.71-0.84 and 0.68-0.88, respectively). Also shown in figure 1 is the prediction interval relating to the effect to be expected in a future programme or study in a new setting. The prediction interval is applicable to a very large population and/ or to one with very high baseline mortality rates; see Beaton et al. for additional details [3].
Over six months of age, the relative effect of vitamin A (per cent reduction in mortality) was not influenced by age or gender. That is, one would expect to see comparable reductions in males and females, and in infants over six months as well as in children up to five years.
The mortality effect is pronounced for diarrhoeal disease (RR = 0.71; Cl 0.57-0.88; p = .002), may be absent or very small for deaths attributed to respiratory disease (RR = 0.94; Cl 0.63-1.42; p = .777), and is demonstrable for deaths attributed to measles (RR = 0.46; Cl 0.22-0.98; p = .043), even though the number of cases is much smaller. Results of the Ghana VAST study indicate no effect of vitamin A on deaths attributed to malaria.
An important finding was that the effect on mortality was not dependent on very high-potency dosing. One trial was based on fortification of monosodium glutamate (MSG in Appendix) and another on weekly administration of physiologic doses (Tamil Nadu). This led us to infer that it was improvement of vitamin A status rather than the method of improving it that was the important determinant of effect.
Morbidity effect
In contrast to the very clear effect of vitamin A on mortality, we were forced to conclude that improvement of vitamin A status cannot be expected to affect the frequency, duration, or prevalence of diarrhoeal and respiratory infections. Conversely, we conclude that it is likely that improved vitamin A status does affect the progression of illness to its severe forms, and to its severest form, death. This important conclusion is largely based on the recent Ghana VAST morbidity trial in which such improvement had an impact on referrals and clinical admissions as well as on reported occurrence of severe morbidity per se. A Brazilian study reported that the frequency of severe diarrhoea is decreased by vitamin A treatment. This was also seen in studies of vitamin A administration in children with measles; both severity of the illness and case fatality rates were reduced. Since we know that hospital admission and clinical referral data were collected but not analysed by other projects, we expect that further information will he confirmatory.
The converse of these findings is that for controlling young child morbidity, vitamin A is not a panacea. The focus must be on the environment in which morbidity occurs. We can only suggest that vitamin A status appears to affect the child's ability to respond appropriately and adequately once infection has developed, and hence appears to affect the course of morbidity. As for mortality, there may well be differentials in the effect across different types of illness. Available evidence did not permit a conclusion on this matter.
One aspect of the morbidity analysis that has direct relevance to field programmes was the fact that vitamin A intervention after the onset of measles favourably affected the development of severe complications and reduced the case fatality rate. In the main mortality trials reviewed, it was not possible to ascertain when the vitamin A had been administered in relation to measles onset. We infer that it is vitamin A status during infection that is important, but also that this can be addressed before or after the onset of infection.
What can be expected in future programmes?
The third goal specified in the contract is perhaps the most important. It addresses the important planning question of what we should expect in a new programme in a new setting. The response can be divided into two sub-questions: in what population setting(s) can one expect vitamin A to be effective and what is the range of effect to be expected?
Where is improvement of vitamin A status most likely to affect morbidity and mortality?
The obvious answer to this question is: Where vitamin A deficiency is now a serious problem. For the mortality trials, all of which were conducted in settings where it was assumed vitamin A was a public health problem under the WHO definition, we attempted to ask about population-level predictors of the relative effect. Statistical power was very low for these analyses of the predictors of response (mortality of control group, xerophthalmia, stunting, wasting) since study was the unit of analysis (n = 8). Individual-level data might have uncovered more subtle effects had they been accessible.
We found no relationship between the baseline prevalence of xerophthalmia and the relative effect of vitamin A. Thus we have to conclude that whereas the existence of clinically apparent deficiency was a marker for all programmes, the actual prevalence added little additional information in predicting outcome.
One important question is unanswered. No studies were conducted in populations with biochemical evidence of vitamin A depletion but without associated evidence of clinical manifestations of deficiency (Ghana VAST came closest to this situation). Thus we can reach no firm conclusion about the impact of vitamin A to be expected in populations where evidence of depletion exists but not evidence that it is severe enough to produce clinical lesions in at least a small proportion of individuals. This unfortunately leaves as judgemental the potential effect of programmes in a substantial part of the developing world.
We found no impact of the prevalence of stunting or wasting on the prediction of the relative effect of vitamin A. We note, however, that all the population groups studied exhibited a high prevalence of stunting. They also shared the common feature of representing the poorer segments of the population with the stigmata of early deprivations. They undoubtedly also had a similar social-biological environment favouring high morbidity and mortality. Thus, stunting was seen more as a marker of the environment of early growth and development than as an index of current nutritional conditions.
We found no apparent association between the mortality rates of control groups and the relative effectiveness of vitamin A. The recorded mortality rates ranged between a low of about 5 per 1,000 to a high of 126 per 1,000.
As mentioned earlier, neither gender nor age appeared to influence relative effectiveness. The only factor we found that would serve to predict relative effectiveness of vitamin A was evidence of dependency on the attributed cause of mortality. From those analyses we conclude that the relative effect is most likely to be largest where mortality attributed to diarrhoeal diseases or measles is predominant, and diminished where deaths attributed to respiratory infection were increasingly prevalent. From these analyses we can add very little to the starting observations that in populations such as those studied, with evidence of poverty, general social and biological deprivation marked by stunting, and existing vitamin A deficiency marked by the presence of xerophthalmia, vitamin A can be expected to have an effect.
We can describe the apparent reason that two studies failed to show an effect of vitamin A supplementation: Hyderabad reported a 6% reduction in mortality and Sudan reported a 4% increase; neither was significant, and the confidence bounds for both included the estimated average effect for all studies combined. In each case the minimal difference in vitamin A status (marked by effect on xerophthalmia) was generated between treatment and control groups. In the case of Sudan, it appears that the vitamin A was not as biologically effective as expected (night blindness improved but not xerophthalmia) although its chemical stability was demonstrated. In the case of Hyderabad, the problem was an unexpected improvement in the vitamin A status of the control group.
Although these observations may explain why those trials failed to demonstrate effects, it is extremely important to recognize that in neither case could the outcomes have been predicted on the basis of information available to us for examination. We treat these two trials and their reported effects as a part of the collective experience and as contributors to our summary estimate of the effect of vitamin A supplementation. However, from the experience in these two studies, we conclude that it is essential that any future programmes monitor their impact on vitamin A status by repeated clinical surveys or by monitoring serum retinol levels, at least until it is established that the administered vitamin is biologically active in the particular setting.
We suggest, in keeping with conclusions reported by the Sudan study, that it may be timely to review with care the evidence supporting the existing guidelines for high-potency periodic dosing. It may be that the combination of dose (200,000 IU after one year) and frequency (six-month interval) were inadequate in the Sudan setting, although apparently adequate in other studies using a similar dosing schedule. Recent research in Indonesia showed that dose effects last at least one year, although protection is best two months after dosing [4].
What is the range of expected effects for future programmes?
Given that we were unable to explain the variation in reported results among the eight mortality trials, we must base any prediction on the total experience. Figure I included the prediction interval applicable to a new study but based on the review of past experience. This interval includes the possibility that a new study will have no effect on mortality (such was a part of the experience). It includes also the possibility that a new study might have an effect much greater than the average 23% reduction expected. In the main report we developed this concept further and actually developed probabilities that could be attached to various levels of effect (table 1 ). These can be interpreted as follows. If justification of a vitamin A-control programme requires a mortality reduction of at least 10%, we suggest that there are about 9 chances in 10 (p = .89) of an effect at least this large being present in a programme that does improve vitamin A status to a degree comparable to those reviewed. If a 20% reduction is required, the probability of achievement is 0.6 (3 chances in 5). However, if reductions of 30% and 40% are sought, the probabilities fall to 0.2 and 0.03 respectively. All of these may be contrasted to the probability of better than 97% that some effect will be produced.
TABLE 1. Probability that a vitamin A effect of specified magnitude will be present in a future study
| Mortality reduction (%) | Probability |
| Any effect | .98 |
| >1 0 | .09 |
| >20 | .62 |
| >30 | .23 |
| >40 | .03 |
In our main report we also cautioned that because of the predictable effects of sampling error, in a study of finite size, particularly in a population with low mortality rates, the investigator would not necessarily see an effect even if it were present. Table 2 presents this warning in the form of probability that an effect will not be seen as a function of intervention group size and baseline mortality rate. What this shows is that if one runs a pilot study in a relatively small population group (for mortality trials) and in the presence of a low mortality rate, there is a high chance that one will fail to see any effect even though the probability that there is an effect remains high. It is of interest that the Hyderabad trial would fall into this category. The opposite also holds: there is a greater chance of seeing an effect as large as that reported for Tamil Nadu (50% reduction) even though it is unlikely that the real effect is that large. Care must be taken in interpreting any pilot studies that are run in the future
TABLE 2. Probability of failing to see an effect of vitamin A, as a function of group size and baseline mortality rate
| Group size | Mortality rate per 1,00 | |||
| 5 | 15 | 25 | 45 | |
| 5,000 | .239 | .1 35 | .096 | .064 |
| 1 0,000 | .1 72 | .085 | .060 | .042 |
| 50,000 | .061 | .034 | .029 | .025 |
| 100,000 | .041 | .028 | .025 | .023 |
| 250,000 | .029 | .024 | .023 | .022 |
We caution also that our estimation of future effects rests on comparison of control and treated groups. However, the mortality rates observed in the control groups were often much lower than expected (than previously believed to exist as a baseline mortality rate). Several possible explanations exist for the discrepancy. These at least a possible non-specific effect of interventions, an effect operating in both control and treatment groups and unrelated to vitamin A; an effect secondary to treatment of high-risk xerophthalmic children with vitamin A in both groups; a phenomenon related to exclusion of high-risk children by design or by self-selection; the fact that the study population was actually different from the regional population for which mortality rates had been described, perhaps the result of selecting a study area that had somewhat better health services or other infrastructure; and inaccuracies in previously reported local mortality rates where these were not directly estimated by the research project.
We did not have an opportunity to test these hypotheses and warn only that we do not know whether vitamin A is equally effective in children who might have been excluded. Hence we do not know whether the predicted effect (23% reduction in mortality) is applicable to true baseline mortality rates. From studies in which the baseline and control group mortalities appeared comparable, the reported effect appeared comparable. Therefore we think the relative effect is applicable to true baseline mortality rates.
It was also reported in the Tamil Nadu study that inclusion or exclusion of children treated for xerophthalmia (and then left in their original treatment group assignments) did not change the estimated relative effect of vitamin A. Thus, that type of exclusion of a high-risk group might alter apparent mortality rates in troth control and treated groups without influencing the estimate of effect. What the planner must recognize is that in a programme setting, without a concurrent control group, reductions from baseline mortality attributable to any of these causes might appear to be results of the intervention. In this sense our estimates of the real effect could be smaller than the apparent effect seen in an operating programme. Offsetting this, of course, would be lower compliance rates expected in an operational programme as compared with a research study.
Distinction between relative and absolute effects of vitamin A on mortality
All of the results described refer to the relative effects of vitamin A, the proportional reduction in mortality. We have shown from those analyses that there was no apparent effect of gender, age, or mortality rate. However, it is to be recognized that if the relative effect is unchanged, the absolute effect (number of lives saved) must be directly proportional to the baseline mortality rate:
Lives saved per 1,000 treated =RR x baseline mortality rate per 1,000.
Since mortality rates in young children generally fall with age, and perhaps differ by sex, it follows that one would expect an impact of age and perhaps sex on the absolute effect of vitamin A (see FIG. 2. Absolute impact of vitamin A expressed as lives saved per 1,000 subjects covered. Estimates assume a 23% reduction in mortality and use the median mortality rates of the studies reviewed.). For the purpose of illustration, the median mortality rates of studies contributing age-specific data are used. Actual rates in a new programme might be quite different, but the phenomenon should be similar.
Some implications for programme targeting
Although the present analyses were not designed to address operational programmes, there are some apparent implications for targeting programmes. In terms of relative effects of vitamin A, the only targeting that we identified as potentially making a difference was with regard to cause-specific mortality. Populations in which deaths attributable to diarrhoeal disease or measles were much higher than deaths attributed to respiratory disease would be expected to show higher relative effects of vitamin A than would be seen under the reverse condition.
In keeping with earlier reviews, we demonstrated also that intervention after the onset of measles was effective in reducing severe morbidity and mortality. This has implications for the design of treatment protocols in primary and secondary health care. It also suggests the importance of determining whether a similar phenomenon holds for diarrhoeal disease and other types of infection. It might have implications for the design of population level control programmes, but this would imply the need for infrastructures capable of detecting and treating potentially severe illnesses.
When one thinks of programmes in terms of their impact expressed as lives saved per 1,000 infants and children covered, it seems clear that the following characteristics would increase their probable effect:
» high baseline mortality rates, particularly for diarrhoeal
disease or measles, the latter perhaps in conjunction with low
measles immunization rates;
» young ages; under one year, mortality rates are generally much
greater than those in children over age one.
Of course, all our analyses relate to populations determined in advance as likely to benefit from vitamin A, and thus our assessments apply to population groups characterized by: generalized poverty; high prevalence of stunting suggestive of disadvantageous social and biological environment and associated early growth failure; and the presence of clinical manifestations of vitamin A deficiency at least sufficiently prevalent to meet the WHO criterion of a public health problem.
An important unanswered question is whether such populations, lacking evidence of clinical manifestations of vitamin A deficiency but with biochemical evidence of major depletion, would also be responsive to improvement of vitamin A status.
This analysis of experience was not designed to compare programme approaches; nevertheless, some interesting observations relevant to the topic can be offered. First, it was demonstrated without doubt that daily (through fortification of monosodium glutamate, MSG) and weekly intakes of physiological levels of vitamin A (Tamil Nadu) were just as effective as periodic high-potency dosing. It follows in our judgement that any approach to improving vitamin A status that effectively controls xerophthalmia will have a beneficial impact on mortality comparable to that reported. In an Indonesian study one-time dosing of women shortly after giving birth was effective in raising breast-milk vitamin A levels and improving the vitamin A status of the infant for at least six months [5]. This might be a strategy worthy of exploration if the target group is young infants.
Finally, it must always be remembered that vitamin A is potentially toxic and may be teratogenic during pregnancy. In the studies reviewed there was some evidence of transient side-effects of high-potency dosing (e.g., reports from Ghana VAST) but no evidence of actual toxicity. Conversely, there was some suggestion in Sudan and perhaps also Hyderabad that 200,000 IU for six months for children over age one year may have been inadequate to evoke a beneficial response. That would be in keeping with an earlier review of oral dosing with vitamin A to control xerophthalmia. That review suggested that, although the suggested dosage appeared adequate to prevent xerophthalmia, it did not appear adequate to sustain blood and tissue levels over six months. It is suggested that continuing review of the norms for periodic high-potency vitamin A dosing is necessary if that approach is chosen. Such review might focus on the dose by frequency combination required to sustain blood levels (and presumably tissue stores) without necessarily having to document a mortality effect.
This project was carried out on the urging of the United Nations ACC-Subcommittee on Nutrition and was supported financially by the Canadian International Development Agency. We thank the authors of the investigations reviewed, many of whom generously provided additional information. The assistance of Dr. Barbara Underwood in helping us gather the literature is appreciated. Above all, we appreciate and respect the contribution made by the more than 175,000 children and their families who participated in the mortality and morbidity trials reviewed in this report. We trust that their contribution to improved knowledge of the role of vitamin A in the health and survival of children now alive and yet to be born will be judged to have been worth while.
Appendix: Sources of study data
Aceh
MSG
Tamil Nadu
Hyderabad
Sarlahi
Bombay
Jumla
Sudan
Ghana VAST
Haiti
