6. Discussion and Conclusions

While most of the studies reviewed involved the periodic administration of high potency doses, we feel that our conclusions are relevant to Improvement of vitamin A status by any effective means.

The major conclusions of this report are:

· Improvement of vitamin A status in young child populations exhibiting evidence of vitamin A deficiency (at the population level) does lead to a reduction in all-cause mortality rates. On average, this reduction is about 23% (RR = 0.77).

· There is a suggestion that improvement in vitamin A status can also be expected to reduce the chance of infectious diseases progressing to their severe forms.

· Conversely, there is very little evidence to suggest that vitamin A status impacts on the prevalence of general morbidity in young children. It would be unreasonable to expect such an effect in operational programs.

Of the eight mortality studies reviewed in detail, only one (SUDAN) saw no effect of vitamin A on mortality (RR=1.04). Another (HYDERABAD) did not detect any significant effect but the Relative Risk was slightly reduced (RR=0.96). A third study (HAITI), not examined in detail, advises that no mortality effect was seen. The other six studies reported statistically significant effects of vitamin A supplementation on total mortality. A seventh study (BOMBAY), again not examined in detail, reported a very major reduction in mortality. A new study of infants under 6 months failed to detect any beneficial effect of large doses of vitamin A.

The present mortality findings are comparable to the results of a meta-analysis originally reported at a meeting in Bellagio and more recently cited in print (Sommer, 1992; Tonascia, 1993). However, that report suggested an average 34% reduction in mortality in children 6 months to five years of age, while we report only a 23% reduction. The major distinction is that the earlier report analyzed data from only 6 mortality trials in S.E. Asia while we had access to data from 8, including the SUDAN and GHANA studies. If we examine only the S.E. Asia studies, the estimated reduction in mortality is 30%. A further distinction between the meta-analysis by Tonascia and the present results is that the earlier analyses selected for age over 6 months while we have analyzed the total data provided by the original studies; no comparison has been attempted to assess the impact of this selection. In 1993, two more meta-analyses were published (Fawzi et al., 1993; Glasziou and Mackeras, 1993). Since each analysis included a different set of projects, the derived summary RR estimates differ somewhat from analysis to analysis. Nevertheless, all meta-analyses conclude that vitamin A supplementation had a highly significant effect in the studies reviewed.

Four other conclusions can be drawn from the mortality study review:

· The effect of vitamin A is not dependent upon very high potency dosing (not a pharmacologic effect). Rather, it is reasonable to conclude that equivalent improvement in vitamin A status by any means would exert comparable effects.

This conclusion derives from the fact that one of the trials involved the use of fortified mono-sodium glutamate (MSG) resulting in a modest increase in daily intake and another (TAMIL NADU) administered a physiologic dose once per week. Both were demonstrably effective.

· The effects of vitamin A supplementation appear to be comparable in males and females and, at least from age 6 months to 5 years, appear to be comparable across ages (no gender or age effect detected).

An examination of the very limited experience reported in the original 8 studies for infants under six months suggested a reduction that averaged about 23% but did not achieve statistical significance. Subsequently West (1993) reported on the extension of the SARLAHI study to examine the impact of vitamin A administration between birth and 6 months of mortality under 10 months. No beneficial effect was seen. The effect of vitamin A under 6 months may not yet be clear but it should likely be presumed that the effect, if any, is small.

· In community-based programmes, it appears that there is a differential effect of vitamin A supplementation depending upon attributed cause of mortality. The effect is very pronounced for diarrhoeal diseases, may be absent in respiratory disease deaths and for deaths attributed to malaria, and was detectable in deaths attributed to measles.

This has important implications for planning since it implies that the effectiveness of vitamin A will be greatest in areas, and age groups where diarrhoeal disease is the major attributed cause of mortality. Given the uncertainties in the attribution of mortality, we do not feel this aspect of the analysis can be taken much further.

· It has also been shown that vitamin A administration after the onset of measles reduces severe complications and has a favourable effect on case mortality. Interestingly, in these interventions, in contrast to community-based interventions, pneumonia deaths were reduced.

These observations have great practical importance in considering treatment protocols.

Contrasting with these clear effects on mortality, in examining the available morbidity trials, and the morbidity results of studies designed primarily as mortality trials, we have come to the conclusions shown below. In offering these conclusions, we are cognizant of the fact that several morbidity studies have not yet published their final analyses and a few are still under way. Certain of our conclusions may be altered by further information that will become available within the next year or so.

· Vitamin A supplementation has no important effect on the incidence or duration of diarrhoeal and respiratory tract infections.

In our judgement, the above finding cannot be attributed to poor study design or methods. We are aware of other morbidity studies, using similar designs to ask about the impact of improvements in water supply and excreta disposal, that detected with statistical significance a reduction of 20-25% in morbidity rates. While small effects on morbidity prevalence or incidence might have gone undetected, we are confident that no major impact of vitamin A on general morbidity is to be expected. Some individual trials have reported beneficial effects of vitamin A on morbidity rates but our judgement and conclusions are based on a review of all trials, taking into account important design features, and giving emphasis to those studies which seem more convincing from the design standpoint. We are quite confident in this conclusion.

· While some studies have reported that vitamin A administration increases the risk of diarrhoeal diseases and respiratory infections, there does not appear to be consistent evidence for such an effect.

We do not place major credence in the few reports of a negative impact of vitamin A administration.

· Vitamin A supplementation appears to reduce the severity of infections.

Not all studies have assessed severity. A study in Ghana and another in Brazil, but not a study in Indonesia, suggest reduced severity as an outcome of vitamin A supplementation. The few studies that have assessed hospitalization rates have detected a decrease among treated children. Since many studies appear to have collected data that could be used to assess markers of severity, but have not yet reported analyses of those data, we expect that future reports will offer clarification of this important question. An effect of vitamin A on severity even without an effect on incidence or duration of morbidity, would be consistent with the results from the mortality trials. It would also be consistent with the results seen in hospital-based trials of intervention after measles.

· No reports of differential effects on morbidity by gender or by age (over six months) have appeared.

The pattern that seems to emerge from the review of morbidity and mortality trials is that vitamin A status impacts upon the response to infection rather than on resistance to becoming infected. The original expectation (when a number of the trials were being designed) was that general morbidity would show reductions in the same order of magnitude as the reported reductions in mortality. In hindsight, the pattern that has emerged is reasonably consistent with what is known about the biological roles of vitamin A (see chapter 3). There were two broad hypotheses about expected effects of vitamin A on morbidity and mortality. One focused upon the known role of vitamin A in epithelial tissues and postulated a barrier mechanism under which the vitamin A-deficient subject, would be more likely to become infected (seen as incidence). The other focused upon the roles of vitamin A in the immune system and hypothesized that the real effect of deficiency would be on the manner in which the organism responded to infection (seen as either or both of duration and severity). The morbidity and mortality results reviewed above would strongly favour the latter hypothesis - that vitamin A is influencing the child’s ability to respond appropriately and successfully to infections. There remains an anomaly - the apparent absence of an effect on respiratory disease-related mortality (except in the case of pneumonia after measles) vs a clear effect on diarrhoeal mortality. Neither theory of action of vitamin A would seem to explain this difference. Indeed, the very well documented role of vitamin A in the maintenance of epithelial tissue, linked to the barrier hypothesis, would also predict that respiratory disease would be more responsive to vitamin A status than would be diarrhoeal disease - the opposite of what has been seen.

There is ample evidence from animal studies that response to vitamin A can differ with infective agents and that may be what is involved here. The actual pathogens have not been identified in the reports available for review.

We conclude that the barrier hypothesis discussed in Chapter 3 is unlikely to be the most important path of effect. Instead we favour the “response hypothesis” suggesting that it is the body’s ability to generate the normal and appropriate responses to infection that play an important role.

It is tempting to speculate that the degree of deficiency is an important determinant of which type of mechanism is involved in the effect of vitamin A. Such speculation would hold that significant epithelial changes and associated weakening of the body’ s barrier system occur only in very severe deficiency while the immune system responses are affected by lesser degrees of depletion of tissue levels. It is emphasized that this is speculation. We do not have experimental data to test it. Interest arises because, in the studies we have examined, for ethical reasons, children who developed signs of xerophthalmia, severe vitamin A deficiency, were treated with vitamin A. This might have effectively removed, or at least reduced, very severe deficiency from the study groups (treatment and control) and hence diminished the chance of seeing effects that required a very severe state of depletion (impaired barrier function?) In turn that may help to explain why the epidemiologic experience suggested a linkage between xerophthalmia and incidence as well as outcome of infectious disease while the controlled trials failed to see the implied effect.

Having undertaken quantitative analyses of the mortality trials, we are able to offer some additional conclusions that are germane to the health planner. We did not attempt quantitative analyses of the morbidity trials because of substantive differences between projects in the way that morbidity data were collected, analyzed and reported. Nevertheless, some of our analyses of mortality data may be applicable also to severe morbidity (morbidity likely to lead to mortality).

Using all eight studies, including the two that failed to find significant effects (but omitting the recently reported extension of SARLAHI to examine dosing of infants under 6 months), we can provide estimates of the magnitude of effect that might be expected in a programme mounted in a new area.

· The average RR for the reported studies was 0.77. The 95% confidence interval attached to that estimate was only 0.71 to 0.84 and the p-value for the test of RR 1.0 (no effect) was 1.12 x 10”⁹. When this confidence interval is recomputed to allow for between study variation (i.e. accepting the Summary RR as an average value for the eight studies rather than as an estimate of a single true RR), the Confidence Interval increases to 0.68 to 0.87, but the effect remains highly significant. We are very confident that in this group of studies, vitamin A supplementation reduced mortality. We are confident also that in future programmes, conducted in populations like these (marked by poverty, evidence of widespread early growth failure (“stunting”), and exhibiting signs of vitamin A deficiency consistent with the international criteria of a public health problem), vitamin A is likely to have an effect. The expected effect, on average will be about a 23% reduction in mortality in pre-school children between 6 months and 5 years of age.

At the same time, we explicitly recognize that there were differences among the eight trials. We have to accept that the actual effect in a particular future programme may not be exactly a 23% reduction. Indeed, based on past experience it is possible, though unlikely, that no effect would exist in a particular program and a very large effect (e.g. 50% reduction) might be present in another. We have attempted to address this between project variation in two ways. First we attempted to identify population characteristics that would serve to predict a greater or smaller effect. In these analyses we only had an n of 8 (the 8 studies) so our power to detect subtle effects was very limited; major predictors should have been detectable. The results are presented below:

· Gender and age (over 6 months) profiles are unlikely to be predictors of effect since neither appears to influence the relative effectiveness of vitamin A (see above).

· The prevalence of wasting, prevalence of xerophthalmia and the interaction between these were not significant predictors. Since all study groups exhibited generally comparable degrees of stunting, it is not surprising that this was not an effective explanatory variable.

· No gross association between mortality rate (of the control group) and the relative effectiveness of vitamin A was seen. There is an association between mortality rate and absolute effect (the lives saved per 1000 treated is implicitly related to the basic mortality rate - see below).

· The observation that there appears to be a cause-specific differential in the impact of vitamin A on mortality would suggest that important differences mortality profiles would predict differences in the relative effect on total mortality. This was not formally tested.

From the above, we accept variation between studies but were unable to explain it. It follows that in offering predictions for the effect in future studies, we must allow for the between study variation that we have observed as well as the uncertainty of the estimate of the average RR for the 8 studies. In chapter 5 we presented, in graphic form, the prediction interval. It was in graphic form since a third variable to be considered in predicting the effect that would be seen is the size and mortality rates in the future programme or study. Below, in Figures 6.1 and 6.2 we present these intervals again but in a different form. Based on our analyses of eight studies we offer a portrayal of what a planner might reasonably expect. We do it as a probability statement - the probability of producing any effect, or of producing an effect exceeding a 10%, 20%, 30%, 40% or 50% reduction in young child mortality.

In Figure 6.1 we present probabilities that there would be a real effect of vitamin A in a new program. This display suggests that there is a 98% chance of there being some effect. The figure suggests also that there is an 89% chance that the real effect will be a reduction of at least 10%, 62% chance of a 20% reduction, and a 23% chance of reduction as great as 30%. We see the likelihood of a true reduction of 50% or more (reported in Tamil Nadu) as being effectively 0. These are predictions of the real effects to be expected. However a planner is more interested in knowing what effects s/he can expect to actually see. If s/he were working with a very large program with moderate to high mortality rates, the probabilities shown in Figure 6.1 might be expected to apply. However with smaller programmes or in programmes with very low mortality rates, the ‘sampling error’ is high. This means that the observed effect may not be the same as the real effect that would be seen with larger group sizes. We use the characteristics of the Hyderabad study (moderate sample sizes but extremely low mortality rates) to illustrate this situation in Figure 6.2. It will be noted that the probability of seeing any effect has fallen from 98% to 81%. The true effect is the same but there is less chance of seeing it. At the other end of the spectrum, the chance of seeing a 50% or greater reduction (an observed effect that would be greater than the real effect) has increased from 0 to about 6%. In the Technical Annex, programs (see Programmes J and K) are provided which compute these probabilities for given study characteristics. Information of this type may help the planner charged with choices in the allocation of resources.

Figure 6.1 Probability that Effect of Vitamin A Supplementation Will be Greater Than Specified Mortality Reductions in a Very Large Field Program

Note: See text for explanation.

Figure 6.2 Probability of Effects of Specified Magnitudes - in a Moderately Small Pilot Study

Note: Modelled after the variance characteristics of the Hyderabad trial.

To illustrate the operation of two of the key variables in the determination of expected variance of new programs, and to keep the presentation consistent with the ‘planning mode’, we present Figure 6.3 which shows the likelihood of failing to see any effect in a new programme as a function of the population group size and baseline mortality rate. The calculations assumed that the summary RR of 0.77 operates for this new population (that the population selected generally resembles those studied) and that compliance and coverage were at least as good as in the research studies.

Figure 6.3 Estimated Chances of Failing to See an Effect in a New Program

The important message from Figure 6.3, is that the results seen in “small” pilot studies could be quite misleading. A negative pilot study is not inconsistent with a true positive effect if the programme were applied to a much larger population. Similarly, a very promising (large effect) pilot study might be expected to be associated with smaller overall effects when scaled up.

The planner may face a problem in applying this approach. The mortality rates that s/he has available in background documents may be higher than the rates on which we have built our analyses and predictions. As we indicate in our report, the observed mortality rates in control groups were often much lower than rates anticipated from background information available for the district, region or country. There are many possible explanations for such a discrepancy. Possible explanations include, but are not limited to:

· possible limitations of vital statistics reporting in the country or region.

· possible effect of excluding high risk individuals and groups in the selection of study subjects (or refusal of high risk individuals to participate).

· reduction of mortality risk by treatment of active xerophthalmia.

· a non-specific beneficial effect of interventions and household visiting by study workers (e.g. unintentional encouragement to seek health care).

We could not test any of these implied hypotheses since none of the studies had appropriate controls for these types of effect. It is interesting that the exceptions to the pattern of lower than expected mortality rates appeared to be in studies in which there was minimal additional contact with households (MSG and JUMLA). That observation is consistent with the hypothesis that increased contact with households, as occurred in most of the trials, exerts a non-specific beneficial effect on young child mortality. As long as there is a blinded control group, this should not bias the results of the study (unless the ‘nonspecific’ effects swamped out any demonstrable effect of vitamin A). However, if an uncontrolled pilot study or operational programme were undertaken, the apparent effect of supplementation might be much greater than we have predicted (non-specific effect + specific effect of vitamin A).

Conversely, the planner will recognize that compliance is likely to be much greater in research programmes and pilot studies than in operational field programmes. As a modest warning, we have included whatever information was presented about compliance in the research studies, but have not attempted any analyses. The planner must expect that because of compliance, the vitamin A-specific effect s/he is likely to find may be less than we suggest. Clearly s/he will wish to examine compliance and other logistical aspects of operational programmes in any pilot study that is undertaken.

Above we noted that there was no detected gender or age effect on the estimated RR. We mentioned also that in fact, since mortality rates typically differed with age and perhaps also with gender, one should expect that programme effects, estimated as lives saved per 1000 children treated, or similar measures, would differ with age. The higher the mortality rate, the greater will be this index even though the relative effect (RR) does not change. This was exemplified in Chapter 5, Table 5.13.

It follows from this that a planner can consider targeting of vitamin A program to groups where the absolute effects per 1000 covered are greater. The indicator variables for such targeting would seem to be total mortality rates (group specific) and relative contribution of diarrhoeal disease and measles to the overall mortality (in contrast to respiratory disease and malaria mortality).

Targeting at the level of the individual could include ‘secondary prevention’, i.e. administration of direct supplements when a child becomes seriously ill. We offer no guidance on the logistical feasibility of such an approach or the coverage that might be expected. We do voice a note of scepticism that this could be seen as an effective approach to population control unless the primary health system were reasonably advanced. However, we do note that it could be a valuable attachment to other intervention strategies. We have reviewed studies that indicate that vitamin A supplementation in the face of serious illness can be efficacious in reducing the risk of more severe illness and mortality.

We note that the two studies that failed to find an effect of vitamin A supplementation on mortality shared one thing in common. They failed to generate the expected difference in vitamin A status between treated and control groups. The explanation for this is not clear and seems to differ between the two studies. Although potency of the distributed supplement was confirmed in both, the authors of the SUDAN study have postulated that the size of dose and interval between doses may have been inadequate to produce an effect in that setting.

This suggests that there may be need to carefully review the existing dosing guidelines for operational programs.

In our analysis of mortality trials we were unable to offer a clear answer to the question “should an effect of vitamin A be expected in situations where there is biochemical evidence of vitamin A depletion but no xerophthalmia?” We think it likely that a mortality effect would be present. This is based on three observations: i) a mortality effect was demonstrable in Ghana even though the prevalence of xerophthalmia was very low; ii) the relative effectiveness of vitamin A was not demonstrably related to the prevalence of xerophthalmia; and iii) the recent report from Brazil (Barreto et al, 1993) serves to demonstrate and effect of vitamin A on severe diarrhoea in a population with biochemical evidence of vitamin A depletion but no xerophthalmia. The report from Brazil takes on great importance since it is the only real link between the many vitamin A-deplete populations, in Latin America and elsewhere, and the mortality trial results.

The possibility of linking vitamin A supplementation to immunization programmes is currently a matter of high interest (WHO, 1993). This has logistical appeal, at least for very young infants (to the time of the last measles immunization at 14 months) and may carry some advantage in terms of improved response to immunization as well as the protective effect of vitamin A on mortality as discussed in the present report. However, there is at Least the possibility that deaths prevented by immunization and deaths prevented by vitamin A overlap - i.e. under these circumstances the effects of immunization and vitamin A might not be fully additive. We can offer some information pertinent to this with regard to measles. In Table 5.10 we presented an analysis by attributed cause of death. One can also compute, for the four studies reporting measles deaths (GHANA VAST, JUMLA, SARLAHI, and TAMIL NADU), the relative effects of vitamin A for all deaths, for measles deaths and for non-measles deaths. These are shown in Table 6.1. As can be seen, there is no detected effect of removing the deaths attributed to measles. This may be due to the very small number of cases identified (minimal impact on total mortality). One must interpret this with great caution - there is no way of ascertaining, from the data available, the number of deaths attributed to other causes that actually had measles as an underlying cause. One would expect that as measles immunization programmes take effect, there will be some reduction in the relative effectiveness of vitamin A even though it cannot be demonstrated in the present analyses. Precise figures are not available for the study projects but it was reported (personal communications) that measles immunization rates were very low in all four sites.

Table 6.1 Possible Effect of Specific Control of Measles on Relative Effect of Vitamin A (Ghana Vast, Jumla, Sarlahi and Tamil Nadu)

		95% C.I.
Attributed cause	RR	Lower	Upper	Z	Prob H0:RR=1
All	0.75	0.67	0.83	-5.294	< 0.000
Measles	0.74	0.53	1.03	-1.734	0.083
All, except measles	0.75	0.67	0.84	-4.945	< 0.000

In closing, the members of the Technical Advisory Group wish to be on record with the following statement:

We are very confident that vitamin A supplementation can effectively reduce mortality rates in young children, and probably also reduce the risk of severe morbidity. We believe that this is the result of improvement of vitamin A status. We expect that any other programme that effectively improved vitamin A status would have comparable effect.

Although the present review has been restricted to vitamin A supplementation programmes, usually involving the periodic administration of high potency doses, we do not wish to be interpreted as endorsing that as a preferred approach to the control of vitamin A deficiency.