Contents - Previous - Next
This is the old United Nations University website. Visit the new site at http://unu.edu
Different nutritional deficiencies often coexist - both within populations and within individuals - and their interactive effects are likely to be qualitatively and quantitatively different from the summation of the independent effects of each deficiency in isolation. Thus, the generalizability of research findings on the functional consequences of specific nutrient deficiencies from one population to another is open to question if the general health and nutritional conditions of the target populations are not equivalent. However, population studies often lack information on other nutritional and health factors that have the potential of acting as effect modifiers. An analytic strategy to follow is to compare results from studies testing the same hypothesis in different population and ecological conditions. Consistency of findings using such a strategy provides insurance of external validity. This paper discusses the consistency of findings in four studies on the functional effects of iron deficiency on behavioural development and the discrepancy of the findings regarding physical growth.
Today's concern for dietary quality hits the critical problem of interaction among nutrients head-on and, implicitly, raises questions regarding the public health value of monofocal interventions and the validity of generalizations of findings from research on nutrients in isolation .
This basic issue of nutrition epidemiology is often observed in connection with research on nutritional deficiencies in populations in less developed countries, particularly when the research is concerned with one nutrient deficiency and substantive knowledge of factors that may modify the effect of the deficiency (e.g. dietary quality) is lacking. Nutritional deficiencies, within a population and within an individual, often coexist, and their interactive effects are likely to be qualitatively and quantitatively different from the summation of the independent effects of factors in isolation. A similar and related issue exists in connection with the remedial and preventive effects to be expected from the treatment of a particular deficiency if other correlated nutrient deficiencies remain unchanged.
This paper illustrates some of the problems of the generalization of findings on the basis of four studies of the functional consequences of iron deficiency. Although the issue of the public health value of monofocal interventions is not made explicit' it is also implicated.
The generalizability of research findings (e.g. on functional effects) from one population to another is open to question if the general health and nutritional conditions of the target populations are not equivalent. Illustrative are studies on the developmental consequences of iron deficiency in infants. On the one hand, most studies conducted in developed countries [2-4] have shown that iron-deficient infants treated with iron improve their performance on scales of mental and motor development after iron treatment. On the other hand, most studies conducted in developing countries such as Guatemala and Costa Rica fail to show such improvements [5-7]. Conceivably, these differences in developmental responses to treatment may be due to generally observable differences in the overall nutritional and health conditions of the children in developed and developing countries. Iron-deficient children in developed countries such as the United States are likely to be in otherwise good health, but this may not be true of children in underdeveloped countries.
TABLE 1. Number of subjects in four studies by iron-status group, and age ranges
|Study||Number||Age range (years)|
|West Java, Indonesia ||49||57||70||3-6|
|Athens, Greece ||21||-||20||3-4.6|
|Central Java, Indonesia ||78||-||41||8-13|
|Chon Buri, Thailand ||101||47||1,210||8-13|
At issue here is the principle that equivalence between the conditions of the reference population and the population to whom the data will be generalized is a core element of external validity. However, as is often the case in nutritional epidemiology, information is generally lacking on other nutritional and health factors that have the potential of acting as effect modifiers. Therefore, conclusive statements about equivalence and definitive generalizations to other populations are often precluded.
An analytical strategy to address the issue of external validity is to draw inferences after there is evidence of consistency in results from studies on the same question in populations living in different world regions, under different ecological conditions, and habituated to different diets. Consistency in the findings from well-grounded research (e.g. clinical trials) in different ecological contexts provides insurance for external validity even if the information on equivalence is not available. Conversely, discrepancy in the findings is likely to reflect differences in relevant variables (or non-equivalence) above and beyond the one particular deficiency under study. The comparative approach is the strategy used in this paper to address two specific questions regarding the growth and developmental effects of iron deficiency: (1) Do iron deficiency and anaemia affect cognitive function? And (2) do they slow physical growth?
Two of the studies were conducted in Indonesia (west [8; 9] and central  Java), one in Chon Buri province, Thailand , and one in Athens, Greece . The data analysis of the last study is still in progress (it is the doctoral dissertation of E. Metallino-Katsaras, a graduate student in the Department of Nutrition at the University of California at Davis). As these four studies followed closely the guidelines of a clinical trial, they provide a basis from which to draw robust inferences regarding the external validity of propositions on the functional effects of iron deficiency.
In particular, interest lies in determining consistency in the findings in connection with the two following research questions: Are iron-deficient and anaemic children smaller, lighter, and less competent in solving cognitive tasks than iron-replete children? And do children who change from an iron-deficient to an iron-replete state grow faster and improve more in their performance in cognitive tasks than children whose iron-deficient state remains unchanged?
The design of the four studies followed two variations of a clinical trial. In three of the studies the iron status of the subjects was defined prior to the experimental intervention. In the fourth the subjects were randomly assigned to one of the two experimental interventions without knowledge of their iron status. These differences in design are related more to strategies in collecting behavioural data than to the nature of the intergroup differences and therefore will not be discussed further. The location, the number of subjects per iron group, and the age range of the subjects in each study are identified in table 1.
Sample sizes varied widely among the studies. In Chon Buri, more than 1,350 children were included; in Athens, the sample included fewer than 50 subjects. This difference is partly explained by the kinds of questions asked by each study and the nature of the cognitive tasks that were administered. The study in Thailand included a representative sample of all children 8-11 years old in Chon Buri province. The tests were administered in groups and measured an amalgamation of intellectual skills and knowledge. The only concern regarding sample size in Athens was that of statistical power. The psychological tests were given individually and were purported to measure specific aspects of the way the brain processes information. The assessment of the time (measured in milliseconds) a child takes to press a key as a response to the display of a visual stimulus presented on a television screen is illustrative.
The differences in the questions asked were also related to the ages of the subjects. Fairly specific cognitive processes were studied in the children 3-6 years old, whereas more general intellectual performance and educational achievement were studied in the children 8 years old and older.
TABLE 2. Criteria for classification of iron status in each of the four studies
Serum ferritin (µg/L)
Transferrin saturation (%)
FEPa (µ/dl RBC)
|anaemic||< 110||-||< 15||-|
a. Free erythrocyte protoporphyrin.
b. An additional criterion was response to the iron treatment equivalent to 10 g per litre.
Causes of iron deficiency
The particular factors involved in the causation of iron deficiency were not the same in the four populations under study, but in most cases they included low consumption of available iron, particularly haem iron; low consumption of dietary factors such as ascorbic acid, meat, or fish that enhance iron absorption; and possible consumption of factors such as phytates which inhibit iron absorption. In addition, among the children in the populations from central and west Java and in Thailand, hookworm infection contributed to iron deficiency.
It is well recognized that the choice of diagnostic criteria used to define body iron status is a critical methodological problem in studies on the functional consequences of iron deficiency. Body iron stores, transport, and even the tissue level of essential haem-containing enzymes can be exhausted or markedly diminished before the circulating mass of red cells is affected. Reliance on a single criterion (or test) for subject selection can lead to sample heterogeneity . A way to increase diagnostic sensitivity is to use no fewer than three diagnostic tests and establish concordance in at least two of them to define iron status.
Except for the study in central Java, the criteria to establish iron status included haemoglobin (Hb) plus two or more biochemical indicators of iron (table 2). Because the diagnostic criteria in central Java included only haemoglobin and transferrin saturation, a decision was made to use two different cut-off points for each of these two indicators for iron-deficient anaemic (e.g. Hb<110 g/L) and iron-replete (Hb> 120 g/L) subjects respectively. In most cases the criteria were formulated on the basis of the distribution of the respective iron indicators from the HANES standards for the age group in question and on the clinical impressions of local paediatric haematologists.
The selection of a cut-off point for haemoglobin was particularly troublesome in Chon Buri because of a comparatively high prevalence of thalassaemia carriers. In such cases the mean haemoglobin of children with replete iron stores is about 5-10 g per litre less than that of non-carriers. It should be noted, however, that the inclusion or exclusion of such children in the analyses does not change the statistical findings on the outcomes.
The studies in central Java and Athens were restricted to iron-replete and iron-deficient anaemic children, while the studies in west Java and Chon Buri also included children who were iron-depleted without anaemia.
In the case of cross-cultural behavioural research, issues of construct and ecological validity are critically important. Requirements for assurance that the tests test the same constructs they are purported to test in different contexts are often difficult, if not impossible, to meet because of the large amount of information necessary for such a purpose.
Although a discussion of these issues falls outside the objectives of this paper, it is important to note that the problems of construct and ecological validity of the tests were significantly reduced in this comparative analysis because of the measures that were used. School achievement tests standardized in the respective populations were used in central Java and Chon Buri. Thus, in those two places there was no problem of transferring tests from one cultural context to another. The tests used in west Java and in Athens included visual stimuli such as simple geometric figures (e.g. circle, square) or pictures of objects (e.g. a table), plants, or animals from the respective immediate environments. Thus, here again, the possibility of measurement error due to poor cultural fit was removed.
TABLE 3. Daily dosage, form of delivery, and duration of iron treatments
|Elemental irona||Form||Duration (weeks)|
|West Java||50 mg||10 ml syrup||8|
|Athens||15 mg||chewable tablets||6-8|
|Central Java||10 mg||tablets||12|
|Chon Buri||4 mg/kg||tablets||12|
a. Ferrous fumarate was used in the study in Athens and ferrous sulphate in the other three studies.
Table 3 presents the dosages, forms of delivery, and duration of the iron treatments. The placebos were delivered in the same form as the iron, and to the extent possible had the same appearance and taste. The fieldworkers and teachers responsible for the delivery, as well as the children, were blind to the content in the tablets or syrup. The iron salts given were ferrous fumarate in Athens and ferrous sulphate in the other three studies. Differences in iron absorption with different iron salts are not an issue here, given the relatively long duration of the treatments.
An important difference among the studies was the duration of treatment, which lasted for only 8 weeks in west Java and Athens but for 12 weeks in central Java and Chon Buri.
The two studies with school-age children also included an anthelmintic treatment (albendazole; Smith, Kline and Beecham Laboratories, Philadelphia, Pa., USA). In central Java it was administered to all children and preceded iron treatment; subjects received ferrous sulphate after stool examinations showed that they were free of intestinal worms. In Chon Buri the approach was different: two doses of albendazole were given to all children during the distribution of iron and placebo, but no attempts were made to establish the success of the anthelmintic.
TABLE 4. Mean values for iron-status variables before experimental intervention
|Haemoglobin (g/L)||Serum ferritin (µg/L)|
Rx = subjects receiving iron treatment; Pl =
subjects receiving a placebo.
a. Subjects meeting only the criterion of response to treatment (N = 11) are not included in these means.
Table 4 presents the mean haemoglobin and serum ferritin levels for each iron group in the four studies. In west Java and Chon Buri there were three iron groups, while in the other two locations the groups were restricted to iron-replete and iron-deficient anaemic subjects. Superficial analysis of the mean haemoglobin and serum ferritin in the different groups suggests that the level of anaemia in schoolchildren in central Java and in Chon Buri was more advanced than that of preschool children in west Java and Athens. However, the sample in Chon Buri included about 20% of thalassaemia carriers; exclusion of these individuals raises the mean haemoglobin level to 103 g per litre. Thus, the level of anaemia in this group is comparable to that in west Java and Athens. A comparatively advanced state of anaemia was therefore observed solely in the sample from central Java. This conclusion is further supported by lower levels of transferrin saturation in this group than in the anaemic groups in the other three studies.
The post-treatment mean values for haemoglobin and serum ferritin are shown in table 5. In three of the four studies, the iron intervention elevated the mean haemoglobin values of the iron-deficient children to a level that was no longer significantly different from that of the iron-replete subjects. In west Java, the mean haemoglobin of the anaemic children after treatment was still significantly lower than that of the iron-replete subjects; however, the difference between the mean haemoglobin levels of the anaemic children with and without iron treatment was statistically significant.
TABLE 5. Mean values for iron-status variables after experimental intervention
|Haemoglobin (g/L)||Serum ferritin (µg/L)|
Rx = subjects receiving iron treatment; Pl = subjects receiving a placebo.
Table 6 presents the mean body weights by iron status before and after experimental intervention for the subjects in the studies from west and central Java and Chon Buri; the data from Athens are not yet available. (Note that the figures for the two groups of school-age children are Z scores using the NCHS reference standards, while those for the preschool children are weight in kilograms.)
The first question to be addressed is whether there is consistency among the three studies in the findings on the comparisons between groups (i.e. iron-deficient versus iron-replete) before the experimental interventions. Two between-group comparisons yielded no significant differences in mean weight values: in west Java and in Chon Buri there was no evidence that prior to the intervention the iron-replete children were heavier than the iron-depleted or irondeficient anaemic children. In central Java the negative Z-score weight means of the iron-deficient anaemic children were significantly larger than those of the iron-replete children. The differences, however, were relatively small: the means for the anaemic children who received iron and placebos were -1.39 and -1.41 respectively, and those for the iron-replete children were -1.34 and -1.31.
To assess the possibility of a differential effect of iron treatment and placebo, analyses of variance with repeated measures were calculated for each study. As in the case of the between-group comparison, the only statistically significant interaction (p<.05) was observed in central Java. Neither in west Java nor in Chon Buri was there any evidence that the children who received iron grew more than those who received a placebo.
TABLE 6. Mean body weights by iron status before and after experimental intervention: evidence of association and effects
West Java (kg)
Central Java (Z scores)
Chon Buri (Z scores)
Values in the same column with different superscript letters are significantly different (p<.05) from each other. Values printed in italic type show a statistically significant interaction (p<.05) between treatment and iron status.
In summary, the hypothesis that iron deficiency affects weight gain was supported by only one of the three studies reviewed. Two studies indicated that iron status and body weight were independent of each other.
Cognition and school achievement
As previously noted, the studies in west Java and Athens were targeted to the assessment of specific cognitive processes. In particular, the concern was with measures of attention and vigilance and with the formation of concepts to establish identities and differences between visual stimuli. The studies in central Java and Chon Buri focused on school achievement measures. The purpose was to assess whether the functional effects of iron deficiency limit learning in the classroom. Accordingly, comparisons of the results between the studies must be made with caution, as each test taps a distinct cognitive or learning construct.
Figure 1 is a summary of selective findings from the complete analyses of the information before treatment in the four locations. The data from Greece are restricted to one of the three cognitive tests used, and the results are not significant. The between-group analysis of the test results before treatment for west and central Java and Chon Buri, however, yielded statistically significant differences (p < .05) in the expected direction. In west Java the anaemic children performed more poorly than the iron-replete children in two of four odditylearning tasks. However, no difference was seen between the scores of the iron-replete and iron-depleted children without anaemia. In central Java a strong between-group difference was noted in the expected direction in school achievement scores that included four subject matters. Similarly, in Chon Buri the anaemic children did not perform up to the level of the iron-replete children on a language achievement test. Moreover, the level of performance of the iron-depleted children was significantly poorer than that of the iron-replete children.
FIG. 1. Scores of cognitive and school achievement tests before experimental intervention - evidence of statistical association. West Java: oddity learning. correct responses (p<.05). Athens: efficiency scores. Central Java: school achievement scores (p < .05). Chon Buri: Thai language achievement scores (p < .05).
In Athens the scores were derived from an attention task (i.e. a continuous performance test). By pressing a key on a computer keyboard, a subject discriminated between the display of a stimulus with and without the previous display of a probe. The time (milliseconds) it took to respond and the accuracy of the response determined the final efficiency scores. No statistically significant differences occurred between the scores of the anaemic and iron-replete children.
In summary, the findings of the three studies for which the analyses have been completed are consistent. Children who were anaemic did not learn concepts of identity and differences nor perform as well at school as children with replete iron stores. It is clear from these findings that the assumption that iron deficiency and comparatively poor cognitive performance covary is statistically valid.
Only one of the two studies that compared the performance of children without anaemia but with depleted iron stores and that of iron-replete children showed a difference in the expected direction. In Chon Buri, those subjects with depleted iron stores performed less well in both the Thai language and the math achievement tests.
Figure 2 presents the change scores and the results of repeated-measure analyses or analyses of the posttreatment differences using the pretreatment scores as covariates. In west and central Java the interactive term between treatment and iron status was statistically significant. In both places the difference (pre- to post-treatment) between the change scores of the anaemic children treated with iron and of those who received a placebo was statistically significant (p < .05) and in the expected direction, while the difference between the same scores for the iron-replete children was not significant.
In Athens the interactive term was not statistically significant. However, the difference between the change scores of the anaemic children who received a placebo and of those who were treated with iron was statistically significant (p < .05). Conversely, the same comparison among the iron-replete children did not yield a significant difference. Figure 2 shows that the performance of both subgroups of iron-replete children and that of the deficient children who were treated with iron improved, but no improvement was seen in the anaemic children who received a placebo.
FIG. 2. Change scores (post-treatment minus pretreatment) for cognitive and achievement scores - evidence of effects from change in iron status. West Java: oddity learning (p<.05). Athens: efficiency (positive scores indicate lower efficiency). Central Java: school achievement scores (p<.05)
The data from Chon Buri showed no significant effects of the treatment in any of the educational outcome variables.
In summary, iron-deficiency anaemia covaried at a statistically significant level with comparatively low performance in visual attention tasks and formal school achievement tests. Strong support exists for the assumption that among iron-deficient children iron treatment results in significant developmental changes in cognition and learning.
The data reported on the effects of iron treatment on physical growth and behavioural development among irondeficient and anaemic children illustrate the complexity of establishing external validity in research on nutritional deficiencies in less developed countries. The findings on the behavioural outcomes show a relatively high degree of internal consistency and point at the adverse effects of iron deficiency. Three of four clinical trials found differences in test performance in the expected direction between the iron-deficient and iron-replete children before treatment. Three studies found a beneficial effect of the iron treatment on the test performance of the iron-deficient children. Conversely, important discrepancies were noted between the findings of the studies regarding the association of iron status and anthropometry. In two of three studies the iron-deficient children were neither smaller nor lighter than the iron-replete children before treatment. The study in central Java  was the only one that found the expected anthropometric differences. Moreover, this study was the only one that found the expected beneficial effect of iron treatment on weight gain.
At issue here is how to reconcile findings that point in opposite directions. One set of findings (that from central Java) points in the direction of a causal effect of iron deficiency on growth retardation, whereas two other sets (from west Java and Chon Buri) suggest no statistical covariation between the two variables. In keeping with the argument presented in this paper, we have to consider the possibility of important differences between the health and overall nutritional status of the populations concerned that may be of direct relevance to physical growth. Such differences could explain the lack of more relevant findings on the association between iron status and growth. Before such a possibility is accepted, however, it is necessary to address some methodological issues within the studies reviewed that are relevant to internal validity.
The iron-deficient anaemic and iron-depleted children who received iron in west Java were not fully rehabilitated before termination of the treatment. Eight weeks of treatment was not enough time within this particular population for full rehabilitation. For example, the post-treatment mean haemoglobin levels of the anaemic and iron-replete children were 115 g and 127 g per litre respectively. It is conceivable that the expected growth differences could have become apparent if the anaemic children had been treated longer and reached a level of iron status not significantly different from that of the iron-replete group. By this same line of reasoning it may also be argued that 12 weeks of treatment was insufficient to result in an increase in growth. Thus it is not surprising that in Chon Buri we saw no significant between-group differences in body weight after treatment. This would be true particularly if changes in growth after iron treatment are related to changes in appetite; this course of events might require more than the 12 weeks spanned by the study. A counter argument is that the time of intervention in central Java also lasted for 12 weeks and was targeted to children whose ages were the same as those in Chon Buri. In central Java the effects of iron treatment on growth were indeed significant.
A second methodological issue arises because in Chon Buri both placebo- and iron-treated subjects received anthelmintic treatment. Because it reduced hookworm infection, this treatment resulted in a small but statistically significant (p < .05) rise in haemoglobin among the anaemic children who did not receive iron. Accordingly, it could be argued that in the absence of a truly placebo group, the comparison with the iron group is weakened. After pooling all subjects together, however, the regression of delta weight on delta haemoglobin is not statistically significant. Thus, no support exists for the assumption that the change in iron status among the iron-deficient children resulted in an acceleration of weight gain.
A third issue concerns the fact that in central Java the mean haemoglobin level of the anaemic children was about 5 g per litre lower than in the other studies, which suggests a more advanced state of anaemia. A differential response to treatment as a function of iron status is conceivable; however, this would not explain the growth differences between groups prior to treatment.
The three methodological constraints are therefore not sufficient to claim that the lack of consistency in the findings is more a problem of internal than of external validity. Thus, we conclude that the differences between the studies are due to non-equivalence among the populations.
In addition to the study in central Java, at least two studies in the literature have repotted improvement in growth after iron supplementation. One of these was conducted with infants in Great Britain , and the other focused on young schoolchildren in Kenya . In the latter study, children with similar anthropometry and morbidity history were randomly assigned to iron treatment or a placebo. After 32 weeks of treatment there were no significant changes in the prevalence of helminthic infections in the two groups. The differences in post-treatment measures of weight for age, weight for height, and triceps skinfold thickness were statistically significant, favouring the irontreated group.
To conclude, comparative analysis provides little support for the notion that changes in iron status among irondeficient preschool and school-age children produced by iron supplementation over 8 to 12 weeks result in comparable changes in body weight. A similar conclusion can be drawn for height. This analysis does not agree with results of at least two studies reported in the literature. It is inferred that the differences in the results may be closely related to differences in the nature of the nutritional deficiencies in the different populations. Accordingly, the research question must be placed in a broader nutritional context and must be addressed to nutritional factors that limit growth in conjunction with iron. Ideally the research paradigm for such a question should include variations in the overall quality of the diet to ascertain under what conditions iron interventions are effective.
Conversely, the data from all four studies reviewed support the contention that iron deficiency has adverse effects on cognitive function. These effects range from specific alterations in attention to complex forms of school learning.
This study was supported in part by the United States Department of Agriculture, State Agricultural Experimental Station Project CA-D-ABS-4625-H.
1. Brown K. The importance of dietary quality versus quantity for weanlings in less developed countries: a framework for discussion. Food Nutr Bull 1991; 13(2):86-94.
2. Aukett MA, Parks YA, Scotts PH, Wharton BA. Treatment with iron increases weight gain and psychomotor development. Arch Dis Child 1986;61 :849-57.
3. Oski FA, Honig AS. The effects of therapy on developmental scores of iron deficient infants. Pediatrics 1978;92:2125
4. Oski FA, Honig AS, Helu B, Howanitz P. Effect of iron therapy on behavior performance in nonanemic, iron deficient infants. Pediatrics 1983;71:877-80.
5. Lozoff B, Brittenham GM, Viteri FE, Wolf AW, Urrutia JJ. The effects of short-term oral iron therapy on developmental deficits in iron-deficient anemic infants. Pediatrics 1982;100:351-57.
6. Lozoff B. Methodologic issues in studying behavioral effects of infant iron-deficiency anemia. Am J Clin Nutr 1989;50:641 54,
7. Walter T. Infancy: mental and motor development. Am J Clin Nutr 1989;50:655-66.
8. Soemantri AG, Pollitt E, Kim I. Iron deficiency anemia and educational achievement. Am J Clin Nutr 1985; 42:1221-28.
9. Chwang L, Soemantri AG, Pollitt E. Iron supplementation and physical growth of rural Indonesian children. Am J Clin Nutr 1988;47:496-501.
10. Soewondo S, Husaini M, Pollitt E. Effects of iron deficiency on attention and learning processes in preschool children: Bandung, Indonesia. Am J Clin Nutr 1989: 50:667-74.
11. Pollitt E, Hathirat P, Kotchabhakdi NJ, Missell L, Valyasevi A. Iron deficiency and educational achievement in Thailand. Am J Clinic Nutr 1989;50:687-97.
12. Metallinos-Katsaras E, Adam E, Stathakopoulou N. Stamoulakatou A, Pollitt E. The effect of iron supplementation on pre-school children. Paper presented at the 14th International Congress of Nutrition, Seoul, Korea, Aug 1989.
13. Cook TD, Campbell DT. Quasi-experimentation: design and analysis issues for field settings. Boston, Mass, USA: Houghton Mifflin, 1979.
14. Pollitt E, Metallinos-Katsaras E. Iron deficiency and behavior: constructs, methods and validity of the findings. In: Wurtman RJ, Wurtman JJ, eds. Nutrition and the brain. Vol 8. New York: Raven Press, 1990:101-46.
15. Latham MC, Stephenson LS, Kinoti SN, Zaman MS, Jurz KM. Improvements in growth following supplementation in young Kenyan school children. Nutrition 1990;6: 159-65.
Contents - Previous - Next