Improving the health of school children is emerging as a policy priority in international health. The recent World Bank Development Report, Investing in Health, recommended school-based health services as a top priority for public health action in developing countries after concluding that they are among the most cost-effective activities for reducing the burden of disease in developing countries. Proposed models for school-based health services typically include distribution of anthelminthic medications and micronutrient supplements, along with health education (Savioli et al. 1992, World Bank 1993).

Although it is likely that school-based interventions may prevent or cure micronutrient deficiencies in school children, it remains controversial whether these interventions could improve growth, particularly linear growth. Because linear growth retardation is believed to occur mainly in early childhood (Martorell et al. 1994, WHO Expert Committee 1995), the question has focused on whether stunted children can exhibit catch-up growth if their environment is improved in later childhood. Adopted children whose environment changes from one of poverty to one of adequacy can attain a normal adult height (Proos et al. 1991), and some authors have concluded that catch-up growth may occur even when children remain in their impoverished environment (Kulin et al. 1982). However, others maintain that catch-up growth does not occur, except in unusual populations in which maturation is markedly delayed (Martorell et al. 1979 and 1994).

Table 1. Zanzibari school children: study cohort at base line [View Table]

Schools are usually the most efficient channel for delivering health interventions to this age group, and school-based surveys are most often used to assess need for different interventions. Although a survey of height and weight of school children is relatively simple to implement, the data may easily be misinterpreted. Shortness-for-age (being stunted) is common among school children in developing countries. However, it is believed to reflect a process that almost always happened in early childhood (Martorell et al. 1994). If linear growth failure (stunting) is not happening during the school years, then it cannot be prevented by school-based interventions (although it might be overcome by catch-up growth). In addition, there are many pitfalls when using cross-sectional data from school children in a search for age effects. Nine-year olds may be more stunted than 7-y olds because they were more malnourished as young children, not because they are growing poorly as school children. Also, school enrollment and drop-out patterns may differ in stunted and nonstunted children, creating a spurious picture of stunting by age among children who attend school.

As part of an evaluation of a school-based deworming program, we monitored the longitudinal changes in height and weight of Zanzibari school children over a 1-y period. Data from the nonprogram schools enabled us to examine both the cross-sectional and longitudinal patterns of growth in primary school children in the absence of intervention. As we expected, the cross-sectional and longitudinal pictures differ in important ways. Contrary to our expectation, we found that these children are experiencing stunting of an important magnitude during the school years. These results are relevant for the planning and evaluation of school health programs in East Africa and perhaps other regions.

SUBJECTS AND METHODS

The present analyses were conducted using data collected in four schools that were assigned as a control group for an evaluation of a school-based deworming program. To form this control group, one school was randomly selected from the public primary schools in each of the four districts of the island and morning classes of grades 1-4 were selected. The base-line survey was conducted in March-May, 1994, and anthropometric measurements were conducted at base line, and 6 and 12 mo later. In the control schools, 1375 children were enrolled in the selected classes according to the school records, and 1261 (91.7%) of those children participated in the base-line survey. Children who were surveyed again 6 mo later (86.7% of those at base line) and who had sex and age data at base line were used in the present analyses; this sample numbered 1090 children, 86.4% of the base-line cohort. Of these children, 1072 (98.3%) had anthropometric data at base line and 6 mo, and 978 (89.8%) had complete data at base line, 6 and 12 mo.

The study protocol was reviewed and approved by the institutional review boards of The Johns Hopkins University, the Ministry of Health of Zanzibar and the World Health Organization.

Hemoglobin concentration was determined by use of the HemoCue (HemoCue AB, Angelhom, Sweden). Parasitologic methods are described in detail elsewhere (Stoltzfus et al. 1997). Briefly, thick and thin blood smears were fixed and stained with giemsa, and malaria parasites were counted against leukocytes, using standard methods. Malaria species was also identified. All infections were P. falciparum; in <5% of slides, P. malariae was also identified. Prevalence and intensity of helminth infections were determined using the Kato-Katz method (World Health Organization 1994). S. haematobium infections were determined indirectly by testing for microhematuria with Hemastix test strips (Ames Laboratories, Elkhart, IN); grade ++, +++ or the presence of visible hematuria was considered a positive test. This procedure screens for S. haematobium infection with 69% sensitivity and 86% specificity (Savioli et al. 1990). Any child with visual hematuria was treated with praziquantel.

Six-month height and weight increments were calculated as the difference in measurements between surveys, adjusted to 6-mo differences according to the actual number of days between surveys, which were approximately but not exactly 6 mo apart. These increments were compared with the incremental growth tables of Baumgartner et al. (1986), based on the growth of white children from Ohio (the Fels Longitudinal Study). For this purpose, an age variable in 6-mo increments was created, in which children aged 6.75-7.24 y at base line were categorized as age 7 y, children 7.25-7.74 y as 7.5 y, and so on. In the tables and figures that follow, our age groupings represent the age at the start of each growth interval. This is in contrast to the tables of Baumgartner et al. (1986), in which the tabulated age is at the end of the growth interval.

To illustrate the magnitude of the height and weight deficits at different ages in this population, we calculated the difference between each child's height or weight and the median height or weight of the WHO growth reference population (World Health Organization 1983). The difference in the height or weight deficit at base line and at 12 mo represents the height or weight deficit accrued by a child in a 1-y period relative to the growth reference. The average annual height or weight change relative to the reference was calculated for boys and girls, along with 95% confidence limits (Snedecor and Cochran, 1980).

To test potential predictors of variation in growth rate, we used multiple linear regression, with the base-line to 6-mo height or weight increment as the dependent variable. In both models, we first adjusted for the variation due to age, sex and school, then tested additional predictor variables and retained those with significance levels below 0.10. Interactions between predictor variables were also tested and considered significant at P < 0.20. Data analyses were conducted using Systat statistical software (SYSTAT, Evanston, IL).

RESULTS

In girls, the height increments were around the 5th percentile in 8-y olds, around the 25th percentile in 9- to 12-y olds, and above the 50th percentile in children 12.5-13.5 y. The pubertal growth spurt appears to occur in these girls 1.5-2 y later than in reference girls. The prevalence of height increments below the 5th reference percentile for girls was nearly 40% at age 8, ~20% at ages 9-10, and <5% after age 11 y.

For both girls and boys, the height increments for the two study intervals (roughly April to October and October to April) were very similar. Apparently, these rather wide seasonal intervals did not capture any seasonal patterns in linear growth.

The weight increments of boys and girls (Fig. 3) were low compared with the reference, lying between the 10th and 25th percentiles and closer to the 25th. There is no evidence of a growth spurt in this group of boys up to age 13.5 y. The weight increments of girls improved with age relative to the reference. Below age 11, girls' weight increments were between the 10th and 25th percentiles, but after age 11 gradually approached the reference median.

The magnitude of growth retardation in the Pembian children can be examined in Figures 4 and 5. The base-line and 12-mo follow-up data have been converted into mean differences from the National Center for Health Statistics (NCHS) reference so that the longitudinal and cross-sectional patterns of growth retardation can be compared in real terms. The cross-sectional picture in boys and girls confirms the patterns of prevalence of stunting by age seen in Figure 1.

In boys, the longitudinal picture of growth was similar to the cross-sectional picture, both for weight and for height. Apart from the group of boys 9.25-9.75 y at base line, whose growth tended to be better, boys' weight and height declined steadily away from the reference. Boys lost on average 1.99 cm/y (95% confidence limits: 2.16, 1.82) and 1.97 kg/y (95% confidence limits: 2.11, 1.83) relative to the reference population. In girls, the longitudinal fall away from the reference values was less steep than the cross-sectional trend suggested. Girls began to stop losing height and weight relative to the reference at ages 12-13 y. Still, girls lost on average 1.42 cm/y (95% confidence limits: 1.64, 1.19) and 1.63 kg/y (95% confidence limits: 1.81, 1.45) compared with the reference. Based on the average annual losses from age 8-14 y, the cumulative height deficit was 11.9 cm for boys and 8.5 cm for girls. The cumulative weight deficit in the same period was 11.8 kg for boys and 9.8 kg for girls.

Table 2. Health characteristics associated with poor growth in Zanzibari school children [View Table]

An association between hematuria and poor linear growth increment was only marginally significant in boys, but it is noteworthy because of its relationship to treatment. This association was found only with microhematuria, which was not treated. Visual hematuria, indicative of severe infection, was treated with a curative dose of praziquantel, and these treated boys exhibited growth similar to their uninfected peers. The linear growth of boys without hematuria was close to the 10th percentile, whereas that of boys with microhematuria was close to the 5th percentile.

DISCUSSION

Age is difficult to ascertain in African populations. In the context of the school survey, we could not discern with parents the child's age by use of a local events calendar; rather we relied on the school records. If our age data were very imprecise, the variability of children's Z-scores by age would be inflated. This was not the case. The standard deviation of children's Z-scores fell between 0.85 and 1.20 for all age groups. This is the expected variability based on many anthropometric surveys in developing countries (WHO Expert Committee 1995).

As mentioned previously, schools may provide a biased sample of children. In Zanzibar, school entrance rates are around 80%, but drop out rates are such that overall primary school enrollment is believed to be around 65% (H. M. Chwaya, unpublished data). Although patterns of school attendance have not been described for this particular population, we expect that better-off children enroll at earlier ages and remain in school longer. Because we sampled children only in the first three grades of school, the bias in age at enrollment is probably a greater factor in this sample than bias in drop-out rates. If better-off children enroll at earlier ages, then the youngest children in our sample tend to represent better-off-than-average children, whereas the older children in our sample tend to represent poorer-than-average children, because better-off 12-y olds would have moved beyond the 3rd grade and are no longer observed in our surveys. This bias probably explains why the cross-sectional trend of increased stunting by age is greater than the longitudinal trend of stunting by age. It would also explain why the prevalence of stunting in the 7-y olds is only 14%, far below the 40-50% rates typical of Africa (UN ACC/SCN 1992).

Our 1-y longitudinal follow-up period allowed us to avoid this cross-sectional bias when describing 6- or 12-mo growth increments. However, when we add up the yearly growth deficits observed in children of different ages, we are again subject to this bias. In this case, this bias would tend to underestimate the cumulative growth deficits experienced by typical Zanzibari children, because typical Zanzibari 8-y olds would tend to grow more poorly than the subgroup of 8-y olds who were privileged enough to be in our school-based sample. Secular trends in feeding practices and infections during infancy may also contribute to the cross-sectional pattern of stunting by age, but could not create the longitudinal deficits we observed.

Differences in the maturation rate of poorly nourished children compared with reference populations make the interpretation of growth data in this age range problematic. Sexual maturation was not assessed in this study. Modern-day studies of African children have found the median age at menarche to be 13.1-14.5 y, which represents a delay of around 1.5 y compared with U.S. blacks (Eveleth and Tanner 1990). A delay of 1-2 y in the start of the growth spurt of Zanzibari children is also suggested by the pattern of median growth increments we observed in girls. If the difference in growth increments between the study sample and the reference population was caused by the difference in their maturation, then the growth increments would be normalized by shifting the curve of growth increments by age of the Pembian children 1-2 y to the left in Figures 2 and 3. This does not normalize the growth increments. The median height increment for the boys in our sample was close to 2.0 cm from age 8 to 12 y. This does not resemble typical linear growth for reference children of any age from birth to 15 y. The lowest reference median height increment in this age range is 2.52 cm, observed between age 10.5 and 11.0 y (Baumgartner et al. 1986).

Perhaps the greatest evidence for the validity of our findings is their internal coherence. On the basis of the cross-sectional pattern of stunting by age, the incremental growth patterns or the longitudinal differences in height of age cohorts relative to the NCHS reference, the phenomenon of stunting is clear. Although the cross-sectional analysis at base line overestimates the magnitude of stunting, the pattern of stunting by age and sex is consistent.

Lawless et al. (1994) conducted an iron supplementation trial in 87 primary school children, ages 6-11 y, from the Kenyan coast (Kwale District), where endemic parasitic infections are very similar to Zanzibar. Although this study was too small to allow an examination of linear growth at different ages, the average height-for-age Z-scores of control children fell from 1.26 to 1.34 in the 14-wk study period, a small but significant decline. The average height gain of control children over the period was 1.1 cm. This would extrapolate to 1.9 cm/6 mo, a value very similar to the increments we observed. No comparison was made between boys and girls.

The growth of children from 2-18 y in rural Machakos District, Kenya, was studied by means of a cross-sectional survey (Stephenson et al. 1983 in Eveleth and Tanner 1990). The sample of preschool children was community-based, but the older children were from two village schools. The mean length of 2-y old children was already 7.3 cm shorter than the NCHS reference. On average, children from the ages of 2-5 y "lost" 1.0 cm/y (total: 3.0 cm), and those from the ages of 6-11 y "lost" 1.5 cm/y (total: 7.5 cm). This pattern of height by age in the school children may be exaggerated because of school enrollment biases, although the authors claimed that below the age of 14 y, the school children were representative of the community. By age 17, the height deficit was 18.2 cm in boys, compared with 12.7 cm in girls. These older children were not representative of the community, but rather tended to come from more privileged families.

Thus it appears that the Pembian children are not unique. These studies provide evidence that African children over the age of 3 y do not grow parallel to reference populations and that boys are more vulnerable to stunting than girls.

Iron status is very low in these children, but low hemoglobin was the only base-line indicator of iron deficiency that was associated with growth, and ponderal growth only. However, iron supplementation caused a significant increase in linear growth of children from coastal Kenya, based on the randomized controlled trial mentioned previously (Lawless et al. 1994). Iron-supplemented children gained on average an additional 0.3 cm/14 wk, equivalent to 0.6 cm/6 mo. This amount of additional height gain would bring the rate of height gain close to reference values.

The NCRSP study of Embu, Kenyan children also suggests that dietary etiologies of stunting are operating in this age group. In multivariate models that included maternal height and socioeconomic status, dietary energy intake was significantly associated with attained height at age 8 y, and animal protein intake was significantly associated with height increment (Neumann and Harrison 1993). Animal protein intake might indicate better overall protein intakes, better dietary iron availability or more bioavailable zinc. A higher phytate:zinc molar ratio was also associated with poor linear growth in these children. Inadequate dietary zinc has also been documented in Ghana and Malawi (Ferguson et al. 1993).

Our data suggest that parasitic infections may also contribute to stunting, although the relationships we observed were not statistically very strong, and they explained only a small fraction of the variability in growth. Two trials of anthelminthic treatment with albendazole (effective against the geohelminths: hookworms, T. trichuria, and Ascaris lumbricoides) (Stephenson et al. 1989 and 1993) and one trial of treatment with metrifonate and praziquantel (effective against the geohelminths and schistosomiasis) (Latham et al. 1990) found positive effects on height gain in primary school children on the Kenyan coast, whereas another publication from one of the same albendazole trials reported no effect on height gain.

A remarkable aspect of helminth infection and malaria is their chronicity. In Zanzibar and many parts of Africa, it is rare to find a child who is not infected with at least one of these parasites throughout their entire childhood and adolescence. Even small decrements in growth may become large when accumulated over 10-12 y.

Slow rates of linear growth in the primary school years might have either of two implications for health. First, the process of stunting might be associated with concurrent risks to the health and development of school-age children, as is the case with young children who experience stunting (WHO Expert Committee 1995). Second, school-age stunting might result in shorter adult height, which decreases work capacity and increases reproductive risks for women (WHO Expert Committee 1995). We cannot describe the effect of the growth retardation we observed on final adult heights, because we did not measure growth during and after puberty. A portion of the height deficit we observed will be compensated by a longer duration of prepubertal growth. Between the ages of 8 and 14 y, Pembian boys are gaining around 4 cm/y, while "losing" 2 cm/y relative to the reference. A delay in puberty of 2 y relative to the reference population would allow them to recover 8 of the 12-cm deficit accrued in this period. A pubertal delay of 3 y (i.e., mean age at menarche of 16-16.5 y) would be needed to compensate fully the deficit we observed. This seems unlikely, based on African values in the literature (Eveleth and Tanner 1990).

It has been suggested from cross-sectional data that rural Kenyan stunted children catch up fully during late adolescence (Kulin et al. 1982). However, this evidence is based on observations of seven 18-y-old boys and eighteen 17-y-old girls compared with reference data collected by other investigators in 1961. Compared with the NCHS reference data, these Kenyan boys had an average height-for-age Z-score of 1.55 at 10 y and 2.74 at 17 y, whereas the girls improved from 2.57 to 1.14.

Our findings raise several new questions: 1) How much of the height deficit accrued during the primary school years is made up through longer prepubertal growth or during the pubertal growth spurt? 2) What are the functional implications of slower linear growth during the primary school years; for example, do these slower-growing children also have slower social or cognitive development? 3) Is school-age stunting preventable through school-based or other public health interventions, and will these interventions correct the functional decrements that might be associated with stunting in this age group? Further research is required to describe the growth of African children based on community-based samples and to evaluate potential interventions to prevent stunting in African school-age children.

ACKNOWLEDGMENTS