QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Abstract Full Text (PDF) Purchase Article View Shopping Cart Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Grantham-McGregor, S. Articles by Ani, C. Search for Related Content PubMed PubMed Citation Articles by Grantham-McGregor, S. Articles by Ani, C.

---

Supplement

A Review of Studies on the Effect of Iron Deficiency on Cognitive Development in Children^{1 ,2}

Centre for International Child Health, Institute of Child Health, University College, London, UK

³To whom correspondence should be addressed. E-mail: s.mcgregor{at}ich.ucl.ac.uk.

	ABSTRACT

TOP ABSTRACT INTRODUCTION DEMONSTRATING CAUSAL... MECHANISMS REVIEW OF STUDIES OVERALL COMMENTS OVERALL CONCLUSION REFERENCES

 
Studies on the effect of iron deficiency on children’s cognition and behavior are selectively reviewed, looking for evidence of a causal relationship. Most correlational studies have found associations between iron-deficiency anemia and poor cognitive and motor development and behavioral problems. Longitudinal studies consistently indicate that children anemic in infancy continue to have poorer cognition, school achievement, and more behavior problems into middle childhood. However, the possible confounding effects of poor socioeconomic backgrounds prevent causal inferences from being made. In anemic children <2 y old, short-term trials of iron treatment have generally failed to benefit development. Most longer trials lacked randomized placebo groups and failed to produce benefits. Only one small randomized controlled trial (RCT) has shown clear benefits. It therefore remains uncertain whether the poor development of iron-deficient infants is due to poor social backgrounds or irreversible damage or is remediable with iron treatment. Similarly, the few preventive trials have had design problems or produced no or questionable benefits only. For children >2 y old, the evidence from RCT is reasonably convincing but not conclusive. RCT of iron treatment are warranted especially in younger children.

KEY WORDS: • iron deficiency • children • cognition • behavior • development

	INTRODUCTION

TOP ABSTRACT INTRODUCTION DEMONSTRATING CAUSAL... MECHANISMS REVIEW OF STUDIES OVERALL COMMENTS OVERALL CONCLUSION REFERENCES

 
Over the past three decades, there have been a considerable number of studies on the relationship between iron status and cognition, and behavior, but the topic remains controversial (Logan 1999 ). Many professionals are lobbying to promote fortification or supplementation programs, claiming conclusive evidence of a causal relationship between iron deficiency and poor cognitive development, whereas others consider that there is no clear evidence (Morley et al. 1999 ). In this review, we examine studies in humans (mostly children), looking for evidence of a causal relationship between iron status and cognition and behavior. Throughout the review, we refer to iron-deficiency anemia as anemia.

	DEMONSTRATING CAUSAL RELATIONSHIPS

TOP ABSTRACT INTRODUCTION DEMONSTRATING CAUSAL... MECHANISMS REVIEW OF STUDIES OVERALL COMMENTS OVERALL CONCLUSION REFERENCES

 
To demonstrate a causal effect, several conditions have to be fulfilled. Iron deficiency alone has to be shown to cause a change in development. There should also be a biologically plausible mechanism linking iron deficiency to development.

Demonstrating significant associations between anemia and poor development in correlational or case-control studies is helpful in identifying at-risk populations but cannot establish cause-and-effect relationships. They provide no information as to the timing of any relationship and it is possible that poor development precedes iron deficiency. In addition, there is considerable evidence that anemia is associated with a large number of socioeconomic and biomedical disadvantages that can themselves affect children’s development. Some of the factors found to be associated with both anemia and poor cognitive development are low socioeconomic status (Owen et al. 1971 ); poverty (Czajka-Narins et al. 1978 ); lack of stimulation in the home (de Andraca et al. 1990 ), including lack of maternal warmth; poor maternal education (de Andraca et al. 1990 , Idjradinata and Pollitt 1993 ) and intelligence quotient (IQ)⁴ (Lozoff et al. 1991 ); maternal depression (de Andraca et al. 1990 ); more absent fathers; low birth weight (<2.5 kg) and early weaning (Lozoff et al. 1991 ); parasitic infection (Ramdath et al. 1995 ); elevated blood lead levels; and undernutrition. It is highly unlikely that all of these factors are controlled for in one study, and there are probably many other confounding factors.

Longitudinal observational studies give additional useful information about the long-term prognosis of children with anemia and the types of deficits at different stages of development. However, they also cannot provide evidence of a causal relationship, but finding reasonably consistent associations between anemia and cognition—after controlling for the most obvious confounders—is a first step toward making causal inferences.

The most accurate way of pinpointing iron deficiency as a cause of poor development is to conduct a double-blind, randomized, controlled trial and demonstrate that producing or preventing anemia changes children’s development. Obviously, one has to use animal models for producing iron deficiency, but preventive trials beginning with nonanemic children are possible to conduct. Unfortunately, preventive trials are extremely difficult and expensive to run. They need large samples to have adequate statistical power, even in populations in which the prevalence of anemia is high. The samples must be followed for some time and it is essential that they remain intact.

Randomized controlled therapeutic trials in which iron is given to anemic children can demonstrate whether a developmental deficit is remediable with iron treatment. They are equally rigorous and are easier to conduct than preventive trials because much smaller numbers are needed. Unlike preventive trials in which a substantial proportion of the placebo children are not anemic and not all treated children are expected to benefit, in therapeutic trials, it is reasonable to expect all treated children to benefit from iron. However, failure of response to treatment does not necessarily negate the presence of a causal relationship because it is possible that the developmental deficit is irremediable at least in the short term.

	MECHANISMS

TOP ABSTRACT INTRODUCTION DEMONSTRATING CAUSAL... MECHANISMS REVIEW OF STUDIES OVERALL COMMENTS OVERALL CONCLUSION REFERENCES

 
Several mechanisms linking anemia to altered cognition are possible. The most direct one is that changes that affect development occur to the structure and function of the central nervous system (CNS). There is substantial evidence of such changes from animal research; these studies are being examined in other papers in this supplement and will not be discussed here.

Evidence from children of changes to the CNS is limited. However, investigators recently studied auditory brain stem responses in children with anemia (Roncagliolo et al. 1998 ). These responses provide a measure of the activation of the auditory pathway from the distal part of the acoustic nerve to the lateral lemniscus, and the central conduction time is an indicator of CNS development. The central conduction time was found to be prolonged in 6-mo-old children (n = 29) with anemia compared with nonanemic children (n = 26). Furthermore, they did not improve with correction of anemia and the difference was greater 6 and 12 mo later. The investigators speculated that the prolonged central conduction time was due to changes in myelination that have been reported in iron-deficient animals (Yu et al. 1986 ). Recent work has shown that formerly anemic children also have longer latencies in visual evoked potentials (B. Lozoff, personal communication). None of the above studies controlled for social background and it is possible that deprivation could affect brain development.

Another hypothesis linking anemia to poor development is functional isolation, which was originally conceived to explain poor development in children with protein-energy malnutrition (Levitsky and Strupp 1995 ). Anemic children explore and move around their environment less than nonanemic children, and they induce less stimulating behavior in their caretakers. These behaviors and the caretakers’ response are thought to delay the acquisition of new skills.

There are many reports of clinical impressions of anemic children being fearful. More systematic observations have been made comparing anemic with nonanemic children during testing with the Bayley Scales using the Bayley infant behavior ratings. These studies have found that anemic children tend to be more fearful (Lozoff et al. 1982a and 1996 ), withdrawn, tense, unreactive to usual stimuli (Lozoff et al. 1982a ), more solemn, less involved (Honig and Oski 1984 ) and more unhappy (Lozoff et al. 1996 , Walter et al. 1983 ).

Surprisingly, there have been few observations in nontest situations. In a study in the United States (Johnson and McGowan 1983 ), anemic children were observed in a standard situation that included two set tasks and free play. They were not different from nonanemic children in activity, reactivity, emotional tone, or attention span. However the observation period was extremely short (total of 16 min) and unlikely to provide a representative sample of behavior. Also, the groups showed no significant difference between their scores on the Bayley Scales, which is unusual. In another observation study of a very short free-play session (Lozoff et al. 1986 ), children stayed closer to their mothers and this was attributed to both the mothers’ and children’s behavior.

In a more extensive study, behavior observations were made in a 15-min free-play situation and throughout developmental testing. Children with anemia stayed closer to their caretakers, showed less pleasure, and were more wary, hesitant and easily tired. An average of 14 spot observations were also made during home visits, and anemic children were more likely to be asleep, irritable, doing nothing, being carried or in bed, and less likely to be on the patio or playing interactively with objects. During the Bayley test session, the anemic children made fewer attempts at test tasks, were less playful and had poorer attention than nonanemic children (Lozoff et al. 1998 ). These types of behaviors persisted after treatment. Most interestingly, mothers of anemic children were rated as being less affectionate, and even the testers behaved differently with the children, giving them fewer tasks and making fewer attempts to elicit responses.

It is of course possible that these behaviors could be due to deprived environments. In Jamaica, similar behavior was found in undernourished children and the behavior was changed with stimulation alone, without changing the children’s nutritional status (Grantham-McGregor et al. 1989 ).

One study linked children’s behavior during assessment on the Bayley Scales to their test scores (Lozoff et al. 1985 ). The children with abnormal ratings were more likely to have lower Bayley scores. The investigators hypothesized that the anemic children’s lower scores were mediated through behavior disturbances. There are therefore several biologically plausible ways, demonstrated in both animal and human research, in which iron deficiency could affect child development.

	REVIEW OF STUDIES

TOP ABSTRACT INTRODUCTION DEMONSTRATING CAUSAL... MECHANISMS REVIEW OF STUDIES OVERALL COMMENTS OVERALL CONCLUSION REFERENCES

 
Several recent comprehensive reviews exist (Lansdown and Wharton 1995 , Lozoff 1998 , Watkins and Pollitt 1998 ). In this review, we have chosen to consider critically selected important studies, focusing on evidence of causality. We discuss the studies grouped by study design and further divided by subjects’ age. The categories are correlational and case-control studies, longitudinal observation studies, therapeutic treatment trials and preventive treatment trials. Within these categories, the internal validity of the studies is examined. The definition of iron-deficiency anemia has been problematical in the past; this topic is discussed in another paper in this supplement and will not be dealt with here.

Correlational and case-control studies.

Beginning as early as 1919, many investigators found significant concurrent associations between hemoglobin concentrations and measures of cognitive development or school achievement (Agarwal et al. 1987 , Clarke et al. 1991 , Florencio 1988 , Grindulis et al. 1986 , Popkin and Lim-Ybanez 1982 , Waite and Neilson 1919 , Walker et al. 1998 , Webb and Oski 1973 ). In addition, baseline differences in developmental levels, cognition or school achievements were found between nonanemic and anemic groups in treatment trials. For example, in trials concerning children <2 y old, in six of seven studies with nonanemic and anemic children (Idjradinata and Pollitt 1993 , Lozoff et al. 1982b, 1987 and 1996 , Walter et al. 1983 and 1989 ), the anemic groups had significantly lower scores on the mental development index (MDI) of the Bayley Scales. There were only eight nonanemic children in the seventh study (Driva et al. 1985 ). Four of the studies also showed differences in the psychomotor development index (PDI) (Idjradinata and Pollitt 1993 , Lozoff et al. 1982b and 1987 , Walter et al. 1989 ). Most of these studies had some control for social background and biomedical conditions, but few had extensive controls for both socioeconomic and biomedical conditions. Although most studies found associations between anemia and a developmental outcome, a puzzling minority of the studies failed to find significant associations (Deinard et al. 1981 and 1986 , Huda et al. 1999 , Johnson and McGowan 1983 , Moock and Leslie 1986 ). Small sample sizes may explain some failures to find associations. Also, some of the studies did not have measures of iron status other than hemoglobin levels and it is possible, but not very likely, that iron deficiency was not the commonest cause of anemia.

Although correlational studies offer the opportunity to look for possible interactions between anemia and socioeconomic or biomedical conditions, few investigators have attempted to do so and most had sample sizes insufficient for doing so. Many studies of protein-energy malnutrition and low birth weight have shown that these conditions interact with social background and other biomedical conditions in their effect on child development (Grantham-McGregor et al. 1998 and 1999 , Pollitt et al. 1993 ). It is likely that analogous relationships exist with anemia. For example, anemia may have different effects in low-birth-weight infants than in normal-birth-weight infants. Most investigators have gone to great lengths to exclude high risk infants; thus, information on these types of questions is scarce.

Some investigators have examined the relationships between severity of anemia and developmental decline. In one study (Lozoff et al. 1987 ), a decline in concurrent motor development was found at hemoglobin values <105 g/L, whereas a decline in mental development appeared at values <100 g/L. In contrast, Walter and colleagues (1989) compared the development of children with hemoglobin concentrations <100 g/L with those with concentrations between 105 and 109 g/L and >109 g/L. The three groups were significantly different from each other in both their motor and mental developmental indices, which were in the same ranking order as their hemoglobin concentrations. Therefore it appears that the level of anemia associated with declining development varies in different populations

Longitudinal observation studies

We identified seven studies in which hemoglobin levels in early childhood were linked to cognitive development or school achievement in later childhood (Cantwell 1974 , de Andraca et al. 1990 , Dommergues et al. 1989 , Hurtado et al. 1999 , Lozoff et al. 1991 and 2000 , Palti et al. 1983 and 1985 , Wasserman et al. 1992 and 1994 ). The ages and size of the samples, outcome measures and findings are given in Table 1 .

Most importantly, all studies found that formerly anemic children continued to be at a developmental disadvantage at one or more of the follow-up assessments. All but one study (Cantwell 1974 ) reported controlling for some social background variables, gender and birth weight. Although the size and number of differences were reduced when scores were adjusted, some tests remained significant in all but one study. In that study (Wasserman et al. 1994 ), the final examination was at 4 y, and previous hemoglobin levels were only inconsistently related to IQ (negative at one age and positive at another). However, early hemoglobin levels had been related to development at 24 mo (Wasserman et al. 1992 ).

    Specific functions affected. All but one study (Hurtado et al. 1999 ) had a global measure of development, either an infant developmental assessment or an IQ test, and these were poorer in anemic children. Specific cognitive functions were assessed in only two studies. In Costa Rica and Chile (de Andraca et al. 1990 , Lozoff et al. 1991 and 2000 ), children were given a comprehensive battery of tests at 5 y of age. In both studies, the formerly anemic children had deficits, which were not identical, across a wide range of functions. Preschool skills, fine and gross motor skills, and visual-motor integration were affected in both studies, whereas language and global IQ were affected in the Chilean sample (de Andraca et al. 1990 ) and only performance IQ in Costa Rica (Lozoff et al. 1991 ). Children in Costa Rica were reassessed between ages 11 and 14 y for an even wider range of functions (Lozoff et al. 2000 ). The anemic children’s performance was worse in practically all tested functions, but they came from more deprived environments than did the nonanemic children. After many covariates were controlled for, the differences were reduced to writing, reading and arithmetic; motor skills; spatial memory; and, in the older children only, selective attention. The children’s behavior was also assessed by teacher and parent reports. The anemic children were reported to have higher scores in anxiety and depression, social and attentional problems, and total problems after covariates were controlled for. The only other study reporting behavior (Cantwell 1974 ) reported that anemic children were inattentive and hyperactive but gave no details.

In the four studies that assessed the children’s achievement in preschool or school subjects or placement in special classes, all found that formerly anemic children were poorer (de Andraca et al. 1990 , Hurtado et al. 1999 , Lozoff et al. 2000 , Palti et al. 1985 ). Two studies found that anemic children had minor neurological dysfunction at 5 (de Andraca et al. 1990 ) and 7 y of age (Cantwell 1974 ).

    Conclusions from longitudinal studies. In conclusion, longitudinal studies indicate consistently that children who were anemic in early childhood continue to have poor cognitive and motor development and school achievement into middle childhood. There is some evidence of behavior problems and minor neurological dysfunction, but evidence is not sufficient for identifying specific cognitive deficits.

The main question is whether the control for social background was adequate. Some environmental variables may not have been controlled for in the final analyses because they were not significantly related to the outcome variable. However, with small samples, this does not necessarily mean that they were not related. It remains possible that environmental variables, measured and unmeasured, could partly or completely explain these findings.

Therapeutic treatment trials

Numerous reviews concern the requirements that must be fulfilled in a treatment trial before it is possible to make causal inferences (Fairchild et al. 1989 ); thus, we will not discuss them in detail. Briefly, the following must be in place: the definition of initial iron-deficiency anemia should be clear and include at least three criteria; the samples must be large enough to provide adequate power; there should be a randomized control group that receives a placebo; treatment should be effective in improving the iron status; both tester and subjects should be blinded; and the outcome measures should have satisfactory construct validity, be sensitive to the range of changes expected and be reliable over time and between observers. If the data are required for determination of policy decisions, then the measurements must have good face validity for the policy makers. It is probably not necessary to point out to a group of fellow researchers that it is not always possible to meet all of these criteria for logistical and ethical reasons, but they remain the yardstick by which to evaluate studies.

In Tables 2 and 3 we have listed all of the treatment trials with iron-deficient children with and without anemia (children under and over age 2 y) that we located. We have attempted to focus on the most salient points that help determine the validity of the studies. We have identified the sample size and ages and type of exclusions, definition of initial iron deficiency, manner of group assignment, content and duration of treatment, and outcome variables. We have indicated whether a significant treatment effect was reported. Treatment effect was restricted to cases in which the change in development of the treated anemic or iron-deficient group was significantly different from the change in development of the placebo anemic or iron-deficient group. However, we have also indicated whether the placebo and treated groups were significantly different after treatment. For studies in which there was no randomized anemic placebo group or the analysis of differences in score change was not reported, we have not claimed a treatment effect. We have also noted whether a response in hemoglobin level was reported. We have divided the children <2 y old from those >2 y old because the findings tend to be different.

We identified nine studies of iron treatment in anemic children <2 y old (Aukett et al. 1986 , Driva et al. 1985 , Harahap et al. 2000 , Idjradinata and Pollitt 1993 , Lozoff et al. 1982b, 1987 and 1996 , Oski and Honig 1978 , Walter et al. 1983 and 1989 ) and one study of nonanemic iron-deficient children (Oski et al. 1983 ). Two of the studies conducted more than one trial with the same sample (Lozoff et al. 1987 , Walter et al. 1989 ), making a total of 12 trials (Table 2) .

The definition of anemia varied from hemoglobin <110 g/L in six studies (7 trials), <105 g/L in two studies (3 trials) and <100 g/L in two studies. It may be relevant that two of the most experienced investigators in this area, B. Lozoff and E. Pollitt, usually used the lower cut-off values (Idjradinata and Pollitt 1993 , Lozoff et al. 1982b, 1987 and 1996 ).

All studies except two (Aukett et al. 1986 , Driva et al. 1985 ) had criteria for other measures of iron deficiency, and all of the studies lasting longer than 2 wk provided evidence that the treatment given was satisfactorily delivered by demonstrating changes in hemoglobin levels.

All but one study (Aukett et al. 1986 ) used the Bayley Scales as an outcome variable, which facilitates comparison across studies. The Bayley is a global measure of development and gives both a PDI and MDI. Initial differences between nonanemic and anemic children were found in the MDI in six of seven studies reporting the comparison and in the PDI in five of the studies.

    Short-term trials. Seven early studies were short, lasting <15 d (Oski and Honig 1978 , Oski et al. 1983 , Lozoff et al. 1982b and 1987 , Walter et al. 1983 and 1989 ). Four of them were double-blind randomized controlled trials (DBRCT) (Lozoff et al. 1982b and 1987 , Oski and Honig 1978 , Walter et al. 1989 ) and one was a randomized controlled trial (RCT) but without a placebo group (Driva et al. 1985 ). None of these five reported a significant treatment benefit. Investigators in two of the studies claimed treatment benefits. Oski and Honig (1978) showed that more of the treated anemic group improved 10 MDI points than did the placebo group, but this was an exploratory post-hoc analysis. The other study claiming benefits (Driva et al. 1985 ) showed that in the 10 d immediately after an iron injection, children showed a significant benefit but not after that. However, the appropriate analysis of difference between the treated and placebo group in score changes was not reported. Further problems with this study are that hemoglobin was the only criterion for iron deficiency and no placebo was given.

The other two short-term trials had no randomized control groups. Treated anemic Chilean children (Walter et al. 1983 ) or nonanemic iron-deficient children from the United States (Oski et al. 1983 ) were compared with nonanemic iron-replete children. In Chile, the treated anemic children improved significantly more (10 MDI points) than did the nonanemic iron-replete children (-1 MDI point). In the American study (Oski et al. 1983 ), the nonanemic iron-deficient groups combined gained significantly more than did the nonanemic iron-replete and -depleted groups.

    Longer-term trials. In all, there were six studies in which anemic subjects were treated for 2–6 mo (Aukett et al. 1986 , Harahap et al. 2000 , Idjradinata and Pollitt 1993 , Lozoff et al. 1987 and 1996 , Walter et al. 1989 ). Four of these trials used nonanemic iron-replete subjects as controls (Harahap et al. 2000 , Lozoff et al. 1987 and 1998 , Walter et al. 1989 ). These studies were based on the idea that the anemic group would have an initial deficit and should catch up with iron treatment. In three of the studies (Lozoff et al. 1987 and 1998 , Walter et al. 1989 ), the anemic group failed to show an improvement significantly greater than the iron-replete group. However, in one study (Lozoff et al. 1987 ), the subset of children showing complete recovery in anemia and iron status caught up to the iron-replete group in MDI and PDI. In the fourth study (Harahap et al. 2000 ), anemic children initially had poorer motor development and caught up with the nonanemic children during the study.

Only two of the longer-term iron trials were DBRCT (Aukett et al. 1986 , Idjradinata and Pollitt 1993 ). One of the latter trials (Aukett et al. 1986 ) failed to find a significant treatment effect on scores of the Denver screening test; however, the authors reported a post-hoc analysis showing that significantly more treated anemic children gained the normal number of items than did children in the placebo group. The other study (Idjradinata and Pollitt 1993 ) showed a large significant treatment effect in both MDI and PDI.

    Discussion. There is no good evidence from RCT that short-term iron treatment benefits the development of anemic young children. However the anemic groups were very small in all five of the studies and one contained only 12 subjects. No investigators reported the study’s statistical power to show differences, but it must have been extremely low. In the four studies reporting the scores of placebo and iron-treated anemic groups, the iron-treated group improved more than the placebo group (Driva et al. 1985 , Lozoff et al. 1982b and 1987 , Oski and Honig 1978 ). Thus the hypothesis cannot be considered to have been tested rigorously. However, it may be that it takes longer than 2 wk for the type of skills measured by the Bayley Scales to develop. It is possible that other behaviors such as attention and motivation could change.

The Indonesian study (Idjradinata and Pollitt 1993 ) is unique in being a DBRCT with long treatment and using the Bayley test. It is also unique in showing a clear significant treatment effect in both MDI and PDI. The treated children showed an extremely large improvement in both indices. The size of the increase is surprisingly large for a 4-mo period. However, in one other study (Oski et al. 1983 ), children showed a similar increase in less time. In the only other DBRCT with this age group (Aukett et al. 1986 ), the Denver screening test was used as the outcome measure. This test is not sensitive to small differences and was intended to screen for children with abnormal development. Most studies indicate that the development of anemic children is within the normal distribution.

In four long-term trials (Harahap et al. 2000 , Lozoff et al. 1987 and 1996 , Walter et al. 1989 ), using only nonanemic iron-replete groups as controls, the anemic group did not catch up to the nonanemic group in three but did in one (Harahap et al. 2000 ). There are at least two reasons why it is not possible with this design to infer the presence or absence of a causal relationship. First, most children improve slightly with test practice and we cannot be sure that untreated anemic and nonanemic children respond in the same way. It is possible that anemic children have poor test behavior and do not learn as much as the nonanemic children from the test experience and thus improve less. In this case, improving the same amount as nonanemic children could actually represent an iron response that would not be detected. Another possibility is that anemic children’s fearful and unresponsive behavior causes them to perform badly at the first test and subsequently improve more that the nonanemic group.

When we examined the four studies with placebo anemic and nonanemic iron-replete groups there were no consistent differences between the groups in their changes in Bayley scores between tests. The change in scores in nonanemic and anemic placebo groups were +6 and +2 (Lozoff et al. 1987 ), +5.1 and +5.5 (Lozoff et al. 1982b ), +8.3 and +6.7 (Walter et al. 1989 ), +2.1 and +0.5 (Idjradinata and Pollitt 1993 ), respectively.

Nonanemic controls are helpful in assessing whether iron-deficient anemic children catch up to nonanemic children. However, anemic children usually come from poorer environments that should be controlled for before catch-up is examined. Most investigators restricted the range of social background and biomedical conditions found in the study children, but few controlled further for environmental factors when examining catch-up. Of the five longer-term studies that had nonanemic groups, the treated anemic group failed to catch up to the nonanemic group in scores on the Bayley test in three studies (Lozoff et al. 1987 and 1996 , Walter et al. 1989 ). Anemic children in the two other studies caught up to the nonanemic group (Harahap et al. 2000 , Idjradinata and Pollitt 1993 ). It is difficult to explain why children in two studies caught up and those in the others did not. Duration of treatment did not explain this because the Costa Rican study (Lozoff et al. 1996 ) had the longest treatment period. Severity of anemia did not explain this either because two of the studies had less severe criteria for anemia than did the Indonesian study (Idjradinata and Pollitt 1993 ). Disparities in social background between the anemic and nonanemic groups may explain some of the failure to catch up.

    Outcome measures. The Bayley test is a global measure and gives little indication as to any specific cognitive deficit. Its predictive ability in y 1 of life is extremely limited, but increases in y 2 (Colombo 1993 ). Other measures of infant cognitive development such as novelty preference and fixation time (Colombo 1993 , Goswami 1998 ) might be more sensitive and predictive in y 1. However, the Bayley test was sensitive to initial differences between anemic and nonanemic groups in nearly all studies. Furthermore, marked changes were found with iron treatment in the Indonesian study (Idjradinata and Pollitt 1993 ) in both MDI and PDI. The PDI of the Bayley test has also been sensitive to changes from nutritional supplementation in studies of protein-energy malnutrition (Husaini et al. 1991 , Joos et al. 1983 ).

    Summary. In general, these studies are difficult to interpret mainly because so few were RCT and the samples were often extremely small. There has been a hesitancy to use placebo groups in the field of iron deficiency on ethical grounds. This is the main reason we do not have clear answers to the important question of whether iron treatment can benefit the development of anemic children. There is no clear evidence that short-term iron treatment has such a benefit; however, the question has not been tested rigorously. Long-term treatment has clearly been shown to benefit the development of anemic children in only one relatively small study. We could find no other study that had rigorously evaluated (with an RCT) the effect with sensitive outcome measures. In several studies, but not all, anemic children have failed to catch up to nonanemic children with iron treatment. We located no study that looked at the effect of anemia in high-risk children (e.g., low birth weight)

Therapeutic treatment trials in children >2 y old

We identified 13 iron treatment trials with anemic children >2 y of age that included iron-deficient children and one trial in pregnant women. One reported no statistical analysis of the treatment and will not be discussed in detail (Soemantri 1989 ). The more important details of the others are given in Table 3 . (Bruner et al. 1996 , Deinard et al. 1986 , Groner et al. 1986 , Lynn and Harland 1998 , Pollitt et al. 1983, 1985, 1986 and 1989 , Seshadri and Gopaldes 1989 , Soemantri et al. 1985 , Soewondo et al. 1989 ).

    Comparison with nonanemic iron-replete children. Eight studies (Deinard et al. 1986 , Pollitt et al. 1983, 1985, 1986 and 1989 , Soemantri et al. 1985 , Soemantri 1989 , Soewondo et al. 1989 ) had nonanemic comparison groups and in one (Deinard et al. 1986 ), no initial difference was found on enrollment between the groups in developmental quotient and IQ. This study had a very wide age range (18–60 mo) and scores on the Bayley test were combined with scores on the Stanford Binet test. In the other seven studies, the anemic children had significantly poorer cognitive scores or school achievement than did the nonanemic iron-replete groups in at least one test. A variety of tests have been used. In the two studies that used a test of discrimination learning, the groups were different (Pollitt et al. 1983 and 1986 ); this test was reported to depend largely on attention. In two of three studies using an oddity learning test, which measures concept learning, the groups were different (Pollitt et al. 1986 , Soewondo et al. 1989 ). IQ tests did not show differences in three studies [the Stanford Binet test used by Pollitt et al. (1983) and the Ravens Progressive Matrices used by Soemantri (1989) and Soemantri et al. (1985) ] but did in one study [the Ravens Progressive Matrices used by Pollitt et al (1989) ]. Differences were not detected in short-term memory in two studies (Pollitt et al. 1983 and 1986 ). The Matching Familiar Figures test, which measures the efficiency of solving a visual perceptual problem, was different in another study (Pollitt et al. 1985 ), whereas scores on the Peabody Picture Vocabulary Test were not different in one study (Soewondo et al. 1989 ). Tests of school achievement were given in two studies and anemic children had lower scores in both (Pollitt et al. 1989 , Soemantri et al. 1985 ).

    Catch-up to nonanemic. In the six studies showing initial differences, the anemic children in three studies caught up to the level of nonanemic children in scores on cognition; the anemic children in one caught up in discrimination learning but not in oddity learning (Pollitt et al. 1986 ). In contrast, children in neither of the two studies measuring school achievement caught up (Pollitt et al. 1989 , Soemantri et al. 1985 ) although they improved on one of them (Soemantri et al. 1985 ). Catching up in school achievement would presumably take a long time because not only do children have to function well but they also need the opportunity to learn missed material. This may not occur in schools without facilities to give children personal attention.

    Study design. Three early studies (Deinard et al. 1986 , Pollitt et al. 1983 and 1986 ) had only nonanemic children as controls and suffer from the problems discussed previously; we will therefore not discuss these three further. One study had controls matched for age, gender and IQ but the method of assignment is not clear (Lynn and Harland 1998 ). Nine of the remaining studies were randomized controlled trials. One did not give a placebo (Seshadri and Gopaldes 1989 , study 1), but the others were all double blind. Several had sample sizes of <30 in each treatment group (Table 3) but four had reasonably adequate or good numbers (Bruner et al. 1996 , Pollitt et al. 1989 , Seshadri and Gopaldes 1989 , study 4; Soemantri et al. 1985 ). The study conducted by Lynn and Harland (1998) had large groups but most children were not anemic.

    School achievement. Two studies had robust designs and adequate sample sizes and examined school achievement, which has obvious policy implications. In Indonesia, anemic children showed a clear improvement with treatment (Soemantri et al. 1985 ), but it is puzzling that neither the placebo anemic group nor the two nonanemic groups made any improvement in 3 mo of schooling. This lack of progress suggests that the school was not teaching effectively; if this was the case, it may be that the treated anemic children’s ability to take tests in general improved rather than they had learned more. This would impute such factors as attention and motivation. An alternative explanation is that the test was insensitive over a certain level, causing a ceiling effect. In the other study (Pollitt et al. 1989 ), children in Thailand showed no improvement with treatment. The criterion for anemia in that study was high, hemoglobin <120 g/L, and the placebo group actually improved by 14 g/L, probably secondary to the deworming. It is possible that the placebo group responding to improved iron status may have threatened the integrity of the trial. In contrast, in the Indonesian study, the criterion for anemia was lower (<110 g/L) and the placebo group’s hemoglobin levels declined by 17 g/L during the study. A further consideration is that the children were tested in groups in the Thailand study, which is unlikely to be as accurate as individual testing; it is not clear how testing was conducted in the Indonesian study.

    Cognition. In another Indonesian study with only 24 treated anemic children (Soewondo et al. 1989 ), no effect was found on vocabulary scores (Peabody Picture Vocabulary Test), and many children were on the floor of the discrimination learning test, making its validity doubtful. In the oddity test, after initial test scores were controlled for, there was no significant difference between the iron-treated and placebo anemic groups, but an interaction occurred between treatment and group in two of four tasks. The treated anemic group improved more than the treated nonanemic group, indicating some catch-up. Only limited details were available from an Egyptian study (Pollitt et al. 1985 ). The Matching Familiar Figures test was used; the efficiency of the treated anemic children improved significantly, and they were significantly better than the placebo group at post-test. Group differences in change of scores were not reported. There were four studies from India (Seshadri and Gopaldes 1989 ) and in each the children were randomly assigned to treatment or control, regardless of hemoglobin level. The first study had no placebo; in the second study the children were given folic acid as well as iron, which may have had an independent benefit. However, the iron-treated group was significantly better than the control group at the end in the second study. In the third study, the treated group improved significantly in most of the cognitive tests, whereas the placebo group did not. At the end, the treated groups had higher scores than the nontreated groups. In the fourth study, iron-treated anemic children improved more than the placebo-treated anemic children in two of four tasks. There was a suggestion of a treatment effect in all four studies, but none reported the significance level of differences in change of scores by the randomized groups. Only the fourth study reported difference between the groups in change of scores, but they restricted the analysis to anemic children only, thus breaking the paired randomized design. It is probable that there was a treatment effect in the third and fourth studies.

In an English study (Lynn and Harland 1998 ), there was no overall treatment effect. When the children were divided into subgroups by iron status, the subgroup with low iron status (ferritin <12 µg/L) showed a significant treatment effect on a test of visual reasoning. However, the findings were inconsistent and the group with moderate iron status showed no benefit from treatment, whereas the group with high iron status showed significant benefits. Also, ascorbic acid was given in the iron treatment and may have had an independent effect.

In a trial with pregnant women, the treatment group improved significantly more than the placebo group in a test of short-term memory (digit span) and a test of attention (Consonant Trigrams). There was however, a large loss (n = 10) from the control group, leaving only 9 children.

The final study was with nonanemic iron-deficient older girls (Bruner et al. 1996 ). It was a well-conducted trial except that they did not have three measures of iron deficiency. Several cognitive functions were assessed, including auditory divided attention, speed of coding, visual search and attention, and verbal learning. There was no treatment effect on the first three tests. The learning test comprised three parts, and a significant effect was found in free call but not in delayed recall or recognition. Three other randomized trials included nonanemic iron-deficient or iron-depleted children in this age range (Pollitt et al. 1985 and 1989 , Soewondo et al. 1989 ). They all failed to find any treatment effect, but the samples were smaller.

    Summary. As with the younger children, anemic children usually had poorer cognition and school achievement than nonanemic children. They tended to catch up with repeated testing and treatment in cognition but not in school achievement.

There were eight DBRCT with anemic subjects or a mixture of anemic and nonanemic subjects and another in which the method of assignment is not given and ascorbic acid was given with the treatment (Lynn and Harland 1998 ). In two trials (Groner et al. 1986 , Soemantri et al. 1985 ), significant treatment effects were reported by randomized group. In another two, significant treatment effects were found in the subgroups of anemic or most iron-deficient children (Lynn and Harland 1998 , Seshadri and Gopaldes 1989 , study 4). Data suggestive of treatment benefits was reported from three other studies (Pollitt et al. 1985 , Seshadri and Gopaldes 1989 , studies 2 and 3). All three reported significant differences at the end but did not analyze differences in change in scores. These studies were reported as conference proceedings or a letter, and details were not available. In two studies, no significant treatment effect was reported (Pollitt et al. 1989 , Soewondo 1995 ).

Some of the studies were small and must have had limited power to show differences. None reported long-term follow-up of children to determine whether benefits arose later or whether benefits were sustained.

    Nonanemic iron-deficient or depleted children. We identified three randomized trials with nonanemic iron-deficient children >2 y old. Although one trial found a treatment effect in one of several tests (Bruner et al. 1996 ), three other studies with smaller samples did not (Pollitt et al. 1985 and 1989 , Soewondo et al. 1989 ). In addition, one study in the <2 y age range included iron-deficient children and found no benefit from treatment (Idjradinata and Pollitt 1993 ). Therefore the evidence for an effect of treatment is weak.

Preventive treatment trials

We identified six preventive trials (Heywood et al. 1989 , Lozoff 1997 , Moffatt et al. 1994 , Morley et al. 1999 , Walter et al. 1989 , Williams et al. 1999 ) (Table 4 ). Two trials are difficult to interpret. A large number of the children became infected with malaria and this confused the results in one study (Heywood et al. 1989 ). In the second study (Walter et al. 1989 ), the analysis was not reported for the original randomized groups; instead the groups were pooled and all anemic children were found to have significantly lower scores than the nonanemic groups in MDI and PDI. These two studies will not be discussed further.

In three of the studies, some beneficial response to iron treatment was found. In Canada (Moffatt et al. 1994 ), children were supplemented from 2 mo of age and were tested at 6, 9, 12 and 15 mo. The iron-fortified group had significantly higher PDI scores than did the placebo group at 9 mo (4 points) and 12 mo (6.3 points), but the benefit was only 2.9 PDI points at 15 mo and no longer significant. There was no significant effect on MDI. The difference between the groups in percentage anemic was greatest at 6 mo (19.9%) and had become small by 15 mo (8%). The loss from the study was approximately one third but the investigators controlled for any difference between lost and tested children. The findings suggest that the effect of iron deficiency is transient.

In England (Williams et al. 1999 ), children were supplemented from 7 to 18 mo of age and benefits were not found until 24 mo. The global developmental quotient fell 5.4 points more in the nonfortified group than in the fortified group (P < 0.05). The scores of all subscales fell less in the iron-fortified than nonfortified group but the personal social subscale was the only one to show a significant treatment effect. This was the only study to follow children for as long as 17 mo, including 6 mo after the cessation of treatment; thus, it is conceivable that other studies may have had undetected benefits. The main problem with the study is that fortified formula was given to one group and money to buy cow’s milk to the other group. It is possible that other constituents in the formula were responsible for the benefits or that cow’s milk reduced the absorption of other nutrients. Also, the study was not double blind. The loss was not large (15%), but unfortunately this included two children excluded from the nonfortified group because of anemia and two from the fortified group excluded because of failed protocol. Some children were already anemic on enrollment at 7 mo (fortified 13%, nonfortified 16%). The difference in hemoglobin levels between the groups was considerable at 12 mo (31%) and at 18 mo was 24%.

A Chilean study (Lozoff 1997 ) probably had the greatest statistical power. The children showed no benefit on the Bayley test, but at 12 mo the fortified group had shorter fixation times on the Fagan test (Lozoff, personal communication), which is thought to indicate better attention and ability to encode stimuli. However, within the groups, the anemic children did not have longer fixation times than nonanemic children. This study was the shortest and, furthermore, all anemic children were excluded at enrollment; therefore, any case of anemia would have been of short duration and not have been present for at least the first 6 mo of life. Both of these factors may have played a role in the lack of effect on the Bayley test scores. The Fagan test predicts later mental development and is probably more sensitive to small cognitive differences. However, the lack of consistency within the groups in the relation between fixation times and anemia makes the finding difficult to interpret. Details of the study are not yet fully reported; thus, in-depth evaluation is not possible.

In contrast to the above, another English study (Morley et al. 1999 ) found no benefit from 9 mo of iron treatment begun at age 9 mo. Fewer than one third of the children had hemoglobin levels available at the beginning and end and these were considered not valid because of technical problems (Lucas, personal communication). The other measures of iron status indicate very little iron deficiency; thus, the study would have had limited power to show benefits from treatment. The results are therefore not possible to evaluate except that there was no apparent harmful effect of giving iron to nonanemic children.

    Conclusions. Only three of the prophylactic studies can be assessed. All found at least a hint of some improvements, albeit in one study, treatment was confounded by formula (Williams et al. 1999 ). The other two studies were double-blind control trials. In the Canadian study (Moffatt et al. 1994 ), the benefits were small, transient and limited to motor development. In the other study (Lozoff 1997 ), benefits were not found on the Bayley test and inconsistent benefits were found on the Fagan test. These findings provide extremely limited evidence that preventing iron-deficiency anemia produces benefits to development. When benefits were found, they were transient or small. Longer-term follow-up may help to interpret the Chilean data.

	OVERALL COMMENTS

TOP ABSTRACT INTRODUCTION DEMONSTRATING CAUSAL... MECHANISMS REVIEW OF STUDIES OVERALL COMMENTS OVERALL CONCLUSION REFERENCES

 
We have discussed the problems of individual studies under the specific sections, but there are some comments common to several different types of studies. Child development is essentially longitudinal, changing over time. Events happening in early childhood may show immediate benefits or detrimental effects that disappear quickly; on the other hand, benefits or detrimental effects may appear at a later stage of development. Very few trials have followed the children after the treatment stopped, and this should be remedied.

Another problem is that many studies suffered from lack of statistical power; future studies not only should be randomized trials but also should have adequate sample sizes. A further problem is that when studies use a battery of tests and analyze several different scores from each test, there is a danger of spurious significant effects.

Last, there is an extreme lack of data on high risk children. In countries with large populations of high risk children, such as children with low birth weight, this may be important.

	OVERALL CONCLUSION

TOP ABSTRACT INTRODUCTION DEMONSTRATING CAUSAL... MECHANISMS REVIEW OF STUDIES OVERALL COMMENTS OVERALL CONCLUSION REFERENCES

 
It is clear that iron deficiency identifies children at concurrent and future risk of poor development. It is also clear that iron deficiency is usually associated with many pyschosocial, economic and biomedical disadvantages.

Anemic children <2 y old

There is no good evidence from RCT that short-term iron treatment benefits the development of anemic young children, but this has not been examined with rigor. The evidence of benefits from long-term trials is insufficient for drawing conclusions or extrapolating to other populations. There are insufficient RCT on the topic and larger studies are required in which both short- and long-term effects are assessed.

In several but not all studies, anemic children have failed to catch up to nonanemic children with iron treatment. This indicates that either their poor development is not caused by anemia or that the effect is irremediable by iron treatment alone, at least in the short term. This is surprising, considering the plasticity of child development; however, a vulnerable period exists for iodine deficiency and there is limited evidence that the first 2 y of life may be critical for protein-energy malnutrition (Grantham-McGregor and Ani 2001 ). It may be that improvements in the environment may be necessary for anemic children to catch up.

Anemic children >2 y old

Anemic older children also usually had poorer cognition and school achievement than did nonanemic children. They usually catch up in cognition with repeated testing and treatment but not in school achievement.

There are more RCT with this age group, and it was clearly shown that children benefited from iron treatment in four studies and a treatment benefit was highly likely in three others. However, two studies showed no effect. At present, the evidence for a beneficial effect of iron treatment on cognition in anemic older children is reasonably convincing, but it would be helpful to run one or two more rigorous RCT with detailed reporting of the results.

Preventive trials

Only limited data from preventive studies support a causal relationship. Preventing iron deficiency anemia can produce benefits to development but they are small and may be transient. The Chilean study should be followed up for longer to determine the implications of the benefits on the Fagan test.

It is difficult to come to unequivocal overall conclusions concerning the effects of iron deficiency in the first 2 y. There is some evidence of a causal relationship but this tends to be inconsistent. There are too few randomized trials of adequate size and appropriate analyses to make firm conclusions. More large randomized trials with anemic children are required before we can inform policy with confidence.

However, considering the stronger evidence of a causal relationship in school children it would be surprising if younger children were not also affected.

DISCUSSION

Participants: Lozoff, Pollitt, Grantham-McGregor, Beard, Haas, Habicht, Schultink, Sazawal, Allen, Lynch.

Dr. Lozoff: First, it is very sobering. I have been working on iron deficiency for 25 years now, and really trying to do good studies. After all this effort, we still cannot give definite answers.

Second, I was pleased to see that we did not buy in totally to the reasoning that if we could show that we could correct it, that would be the proof that iron was the cause. There was a period in the literature where, if you did not show that you corrected it with the iron treatment, people would conclude iron deficiency could not be the cause, because you correct the anemia, you correct the development. I was trying to think of an analogy for this. Take tuberculosis. A child comes into the hospital with tuberculosis meningitis. You give antituberculosis drugs. The child still has neurological sequelae. Well, the child might have had a neglectful or a misguided family, but the child nonetheless had tuberculosis meningitis. Tuberculosis has been cleared from the body. There was unequivocally a treatment response, but you still have a long-lasting effect. So, it is just fallacious thinking to hold as the only criterion that you reverse it with iron treatment and I was pleased that Dr. McGregor made that so clear. It would be great if we could do it, because that would settle the whole thing, but the reverse does not hold true.

Third, I want to highlight the unresolved issues as I see them. We have not really been looking at specific central nervous system functions. As we do that, we are going to have to revisit this iron deficiency without anemia. We have got to revisit these more sensitive measures. Similarly, we have got to revisit everything about what treatment affects or does not affect as you get more measures that make sense in terms of what iron is doing.

Finally, one comment about the magnitude of effect. The Bayley for infants and toddlers has been the only measure that we really have had. So, it has been used not just in iron deficiency but in a whole host of early biological risks. The magnitude of the differences that we have observed between anemic and nonanemic kids on the Bayley is of the same order of magnitude that is widely accepted as being clinically relevant and important, whether it is low birth weight or cocaine or alcohol or any of those things. I shared Dr. Pollitt’s concern about the Bayley Scales, but I wanted to be sure that we at least gave that information to everybody.

Dr. Pollitt: The data have shown differences in the Bayley Scales that were based on children who were 18 mo or older. They probably do have significant value. I am arguing that it is at the younger ages that the data are not good. Second, we also have to understand that a 3–5 point difference in the Bayley Mental Development Index may actually represent only about 1–2 wk of difference in development. In other words, a 3 or 5 point difference, as has been reported, does not correspond to a difference of 3 mo of development. It is equivalent to 15 d.

Dr. Lozoff: Six to 19 points difference in your study?

Dr. Pollitt: Nineteen points actually represents about a month and a half.

Dr. Grantham-McGregor: That is quite a lot in a young infant’s life. It is not like a month and a half in my life.

Dr. Beard: I would like all three of the pr