Childhood malnutrition remains a common problem in much of the developing world. The United Nations agencies concerned with global nutrition isssues have compiled data indicating that approximately 25% of children less than five years of age in low-income countries have a height-for-age which is less than -2 SD with respect to international reference data (ACC/SCN, 1987; ACC/SCN, 1989; ACC/SCN, 1992; ACC/SCN, 1994). This situation has changed only little during the past 20 years since these figures have first been tracked. The causes of children's growth failure are not fully understood, although it has been assumed that inadequate food intake, poor nutritional quality of many traditional diets, and high rates of infections are primarily responsible. Recently, more consideration has also been given to the possible role of individual micronutrient deficiencies.

Approximately thirty years ago a clinical syndrome characterized by dwarfism, hypogonadism, and anemia was first attributed to zinc deficiency (Prasad, 1963). Since that time, zinc-responsive stunting has been reported from several parts of the world (Gibson, 19xx), including selected low-income populations in North America (Walravens, 1976; Walravens, 1983; Walravens, 1989). Several studies have also found that malnourished children gain weight more rapidly when supplemented with zinc (Golden, 1981; Castillo-Duran, 1987; Khanum, 1988). On the other hand, studies from some developing countries have failed to identify a growth resonse to zinc supplementation, so the importance of zinc deficiency as a public health problem remains controversial.

The global prevalence of zinc deficiency is unknown, largely because there are no simple indicators of zinc status (Golden, 1989). The diagnosis of zinc deficiency currently requires some evidence of a physiological response to a therapeutic trial of zinc; hence, a substantial investment of time and resources is necessary to establish the diagnosis, and large-scale population studies are still limited. Nevertheless, food composition and dietary intake data suggest that the majority of children in developing countries may be at risk of zinc deficiency beyond early infancy. The major sources of zinc are animal products, which are consumed in very low amounts by children in developing countries, and the embryo portion of grains, which are often removed during processing (Cousins, 1990). Moreover, phytic acid present in plant sources forms an insoluble zinc-chelate, which markedly reduces zinc availability from food (Solomons, 1984; Lonnerdal, 1988). Consequences of zinc deficiency other than growth failure include poor appetite, impaired dark adaptation, skin lesions and delayed wound healing, and immunosuppression and increased rates of infections (Prasad, 1990). Other recent studies, as briefly reviewed by Hambidge (1989), have found an association between low plasma or tissue zinc and complications of pregnancy, including impaired fetal development.

Because of the possibly widespread occurance of zinc deficiency in developing countries and the important functional consequences that have been reported, we have recently carried out a metaanlysis of all available zinc supplementation trials. The purposes of this analysis were to determine whether there is evidence of widespread zinc deficiency in developing countries and whether there are particular indicators that permit prediction of which populations are more likely to respond to zinc supplementation. This paper presents the preliminary results of the metaanlaysis, which was first presented at the Experimental Biology meetings in early 1995. A more comprehensive analyis with additional studies is now in progress. The current paper will describe briefly the methods we used for identifying potentially acceptable studies for the metanalysis, the criteria for inclusion of individual studies and the analytic methods that were used. The overall results of the acceptable studies will be summarized with regard to the effect of zinc supplementation on increments of linear and ponderal growth, changes in weight-for-length, and changes in plasma zinc concentrations.

Potentially acceptable studies were identified through an initial MEDLINE search, using key words zinc and growth. We also relied on several excellent published reviews of earlier zinc intervention trials, in particular the papers by Rosalind Gibson (ref) and Roger Shrimpton (ref). We then scanned the bibliographic citations from each of the studies we found to identify any other references that had not already been included in our database. Finally, we learned of several unpublished studies through personal communications with investigators working in this field.

Each study that was identified was evaluated according to pre-defined criteria for acceptability for the meta-analysis. We required the studies to be prospective intervention trials with a concurrent control group. We only considered those studies that included children less than 13 years of age to avoid possible confounding caused by differential pubertal growth spurts among adolescents in the trial cohorts. The studies selected had to include at least one of the major outcome measures of interest, specifically change in length or change in weight. Finally, the results had to be presented in sufficient detail to permit inclusion in the statistical analyses. In particular, we required that data were available for classifying initial anthropometric status of the study groups as well as the mean and standard deviations of changes in length or weight during the period of observation.

Subsequent to selection of acceptable studies, each paper was reviewed in detail, and data were independently abstracted by each of the co-investigators and then summarized in a common database. Because studies presented outcomes using a variety of different units — such as absolute change in length or weight, change in percent median length- or weight-for-age, or change in respective Z-scores — all results were standardized as "effect size" which was calculated as the mean change in nutritional outcome for the treatment group minus the mean change in the control group divided by the pooled standard deviation of change for both groups. Additionally, each study was weighted in the meta-analysis by considering the total number of subjects in the study and the effect size. The use of effect size as the response variable allows comparison of variables expressed in different units across studies. Ordinarily, a difference between means of 0.2 standard deviation units is considered a small effect size, a difference of 0.5 standard deviation units a medium effect size, and a difference of at least 0.8 standard deviation units a large effect size.

The mean and 95% confidence intervals for the weighted average effect size were calculated for each outcome variable. The homogeneity of effect size was then tested by chi-square analysis; and, when significant heterogeneity was identified, a regression analysis was performed to examine possible sources of heterogeneity.

We reviewed a total of 48 studies, of which 25 were felt to be acceptable for inclusion in our analysis. A total of 1,535 children were included in the full set of studies, which were published from 1974 through 1995. The remaining 23 studies were excluded from the meta-analysis for a variety of reasons, such as failure to include a control group, lack of experimental intervention, or lack of sufficiently detailed information. Nine of the acceptable studies were from Latin America and the Caribbean, seven from Asia and the Middle East, seven from North America or Europe, two from Africa and one from Australia. Eight of the studies were defined as community trials in developing countries, eight were either community or clinical trials in industrialized countries, six were clinical studies of hospitalized children recovering from severe malnutrition, and three could not be assigned to any of these categories.

Twenty of the twenty-five studies were double-masked, placebo-controlled, trials. In seventeen studies, the consumption of the supplement or placebo was directly observed by the study team; in seven studies compliance with the dosage schedule was monitored by examination of supplements remaining following different periods of monitoring; and in one study there was no reported monitoring of the dose. The studies varied from ten days to sixteen months in duration, with a mean of approximately seven months. The different studies included from 11 to 162 subjects, with a mean of 64.

A broad range of zinc compounds was used in the different trials. The doses of zinc supplements ranged from 1.5 to 50 milligrams per day, with a mean of 12 milligrams. In most cases the doses were given daily, six days per week or five days per week.

The study subjects ranged in age from birth to 13 years, with a mean of approximately 3½ years. The mean initial plasma zinc levels ranged from 42 to 139 micrograms per deciliter, with a mean of 82. Most of the studies were evenly balanced between boys and girls, although two studies included only boys. The initial mean height-for-age Z-scores ranged from -5.0 to +0.1 Z-scores, with a mean of -2.3 Z-scores. The initial mean weight-for-height Z-score ranged from -4.0 to +0.8 with a mean of -1.0.

Figure 1 shows the mean and 95% confidence limits for effect size for change in height from the 17 studies available to date that contained interpretable data. Three studies showed a negative mean effect size with zinc supplementation, although the confidence limits did not exclude zero in any of these studies. Of the remaining fourteen studies, all of which showed either no change with zinc supplementation or a positive effect of supplementation, the confidence limits from five studies excluded zero. Overall, there was a small weighted average effect size of +0.2 SD with zinc supplementation, which was highly statistically significant (p< 0.0001). Although these results are still preliminary, we estimate that there would have to be a total of 91 unpublished studies with no treatment effect to render the current positive results non-significant.

A chi-square test for homogeneity of the effect size indicated that there was, indeed, significant heterogeneity of results. We therefore explored whether specific aspects of study design or characteristics of study subjects may have explained the heterogeneity of results. There were negative correlations between the mean change in height in individual studies and two explanatory variables: namely, initial mean height-for-age, and initial mean plasma zinc. In other words, studies that included children with greater degrees of stunting and lower initial plasma zinc levels reported greater responses to supplementation. When all of these possible explanatory variables were included in a multiple regression analysis, the presence of stunting, defined as mean initial Z-score less than -2.0 was the only variable significantly associated with treatment effect size for change in height.

For those studies that included children with mean initial Z-score greater than -2, there was no overall effect of zinc supplementation. By contrast, for those studies that included children with a greater degree of stunting initially, there was a highly significant effect of treatment, with a medium effect size of 0.47 SD. There was no longer any heterogeneity within these two subgroups of studies.

We next examined the 19 studies that provided interpretable data on change in weight. As shown in Figure 2, none of the studies found a negative relationship between zinc supplementation and change in weight. Four of the studies showed positive effects of with confidence limits that excluded zero. Overall, there was a highly significant positive impact of zinc supplementation on change in weight, with a small weighted-average effect size of 0.26 SD.

Although there was no significant heterogeneity of the effect sizes for change in weight, there were some significant correlations between initial characteristics of the studies or study subjects and this outcome variable. For example, there were significant negative correlations between effect size and the following variables: initial mean weight-for-age, initial mean plasma zinc, and duration of study. There was no correlation between initial mean age, initial mean height-for-age, nor dose of zinc and change in weight for individual studies. When all of these variables were included in a multiple regression analysis with dependent variable effect size, initial mean plasma zinc concentration was the only explanatory variable that remained statistically significant. This result must be interpreted cautiously however, because three of these studies did not provide information for initial plasma zinc.

Six studies provided interpretable results on change in plasma zinc concentration. Each of these studies showed a positive impact of zinc supplementation, with an overall effect size of 0.67 SD, which was highly significant. There was significant heterogeneity of results, all of which could be explained by one study, which used an especially high dose of zinc supplements. Even after excluding this study from the database, there was still a significant effect size of 0.52 SD.

In summary, there was a small, but highly significant, impact of zinc supplementation on children's height increments, with an average effect size of 0.2 standard deviation units. This effect was present for the subgroup group of studies with mean initial height-for-age Z-score less than -2.0, but not for those with mean initial height-for-age Z-score greater than or equal to -2.0. The height response to zinc supplementation did not appear to be related to the dosage schedule employed, nor the duration of supplementation. Among studies with initially stunted children there was a medium average effect size of zinc supplementation, averaging 0.47 standard deviation units.

There was a small, but highly significant, impact of zinc supplementation on children's weight increments, with an average effect size of 0.26 standard deviation units. The effect of zinc supplementation on change in weight was negatively associated with mean initial plasma zinc levels, but not with mean initial age, presence of stunting, or dose of zinc provided. Among studies with initial plasma zinc concentrations less than 80 µg/dl, the average effect size was moderate, averaging 0.42 standard deviation units.

There was no impact of zinc supplementation on change in weight-for-height Z-score, but the number of available studies is still quite limited. There was a moderate, highly significant, impact of zinc supplementation on change in plasma zinc concentrations, but again the number of available studies is still very limited.