点击显示 收起
1 From the Institut de Recherche pour le Développement (IRD, formerly ORSTOM), Nutrition Unit and Infectious Diseases Research Unit, Montpellier, France, and the Niakhar Population and Health Project, Dakar, Senegal.
2 From 1989 to 1997, the Niakhar Population and Health Project was supported mainly by Pasteur-Mérieux Sérums et Vaccins, Paris. The present study was financed by the IRD. 3 Address reprint requests to KB Simondon, UR Epidémiologie et Prévention, Centre IRD de Montpellier, BP 5045, 34032 Montpellier Cedex 1, France. E-mail: kirsten.simondon{at}mpl.ird.fr.
ABSTRACT
Background: Prolonged breast-feeding is frequently associated with malnutrition in less-developed countries, even after adjustment for socioeconomic confounders. However, in rural Senegal, breast-feeding is prolonged when the child is stunted.
Objective: We aimed to test whether the lower height-for-age of children weaned late is explained by their height before weaning or whether prolonged breast-feeding is associated with impaired growth.
Design: A cohort of 443 Senegalese children recruited from dispensaries at 2 mo of age were visited in their homes at 6-mo intervals when they were 1.5 to 3 y of age. Weight, length, arm circumference, and triceps skinfold thickness were measured. Six-month increments were analyzed in relation to breast-feeding (breast-fed compared with weaned children or breast-feeding duration), season, and maternal housing with use of multiple linear regression.
Results: The mean duration of breast-feeding was 24.1 mo (quartiles 1 and 3: 21.9 and 26.4). Height-for-age at the age of 3 y was negatively associated with age at weaning (P < 0.01), but this association disappeared after adjustment for height-for-age in infancy. Length increments were significantly greater in both the second and third years of life in children breast-fed for longer durations (P < 0.05) and tended to be greater in breast-fed than in weaned children in the second year of life (P = 0.05). In the third year of life, breast-fed children had greater length increments than did weaned children in the subgroup with poor housing (P for interaction < 0.05). Growth in weight did not differ significantly according to breast-feeding.
Conclusion: Prolonged breast-feeding improved linear growth, and the negative relation between height-for-age and duration of breast-feeding was due to reverse causality.
Key Words: Breast-feeding • weaning • linear growth • weight • reverse causality • stunting • seasonality • socioeconomic groups • morbidity • Africa • Senegal • infants • children • height-for-age
INTRODUCTION
In many developing countries, breast-fed children have a lower nutritional status than do weaned children in terms of both height-for-age and weight-for-height from the age of 12 mo (1–10). This negative association cannot be explained merely by confounding (6, 8, 9); therefore, it was suggested that breast-feeding beyond 12 mo depresses growth (1, 3, 6) and even that malnourished children should be weaned by 12 mo (3).
Another hypothesis is that the observed relation is due to reverse causality, ie, later weaning of initially malnourished children (11–13). Evidence for reverse causality was reported in studies from several countries in Africa (7, 14–16) and in South America (17). In a rural area of Senegal, children who were weaned late (after the age of 24 or 30 mo) had a significantly lower mean height-for-age and a greater prevalence of stunting than did children who were weaned early (before 18 or 24 mo) as early as the age of 9–10 mo, at which time all of the children studied were still breast-fed (16).
This hypothesis does not exclude the possibility that prolonged breast-feeding impairs growth, perhaps by depressing a toddler's appetite for non-breast-milk foods (1, 3). Existing data on the relation between breast-feeding and growth in toddlers are conflicting. In periurban Peru, breast-feeding was associated with faster growth in length from age 12 to 15 mo in a subgroup of children but not in the entire sample (18). In periurban Guinea-Bissau, weaning was associated with a relative decrease in weight, but not length, after adjustment for age, sex, season, and the prevalence of diarrhea (19). In an urban-rural survey in Sudan, breast-fed children grew significantly slower in weight and length than did weaned children up to the age of 2 y, even after stratification for economic level and maternal literacy (15). In Kenya, breast-feeding was associated with faster growth in both length and weight over a 6-mo period in a cohort of 9–18-mo-old children (20).
To investigate the relation between nutritional status, duration of breast-feeding, and growth within the same area of Senegal that we studied previously (16), we followed up a cohort of 443 children from 2 mo to 3 y of age. Our objectives were to compare height-for-age at the age of 3 by duration of breast-feeding, while taking into account height-for-age during infancy, and to analyze growth in toddlers by duration of breast-feeding, season, and economic level.
SUBJECTS AND METHODS
Subjects
The infants studied were born from January to September 1995. Oral informed consent was given by the infants' mothers at the time of inclusion in the study. During their second and third years of life, all children who had complete measurements at the time of inclusion and who were still living in the study area were visited at home. For the analyses, children lost to follow-up before the last home visit in April 1998 were excluded, as were children with missing data on nutritional status at the age of 3 y, maternal housing, or duration of breast-feeding. Children who were absent at any of the first 3 home visits were not excluded from the analysis. The study was conducted in accordance with the Declaration of Helsinki of 1975, as revised in 1983.
Study area
The study was conducted in the Niakhar area in central Senegal, West Africa, which is 150 km east of the capital, Dakar. This area is inhabited by nearly 30000 persons belonging to the Sereer ethnic group who live in 30 villages. The climate is Sahelian, with a long dry season from November to June and a short rainy season from July to October. Annual rainfall averaged 433 mm (range: 227–632 mm) between 1984 and 1996. More than 90% of the population is composed of farmers who grow millet, mainly for their own consumption, and groundnuts, mainly for sale; 98% of the compounds have livestock.
The area has been under continuous demographic surveillance by The Niakhar Population and Health Project, Institut de Recherche pour le Développement (IRD, formerly ORSTOM), since 1983, when a central demographic and epidemiologic database was set up. From 1987 to February 1997, all births, deaths, migrations into and out of the study area, and weaning events were monitored through weekly home visits; bimonthly visits have been made since March 1997. Mortality in infants and in children aged <5 y has decreased sharply over the past 30 y: from 223 and 485, respectively, per 1000 live births in the 1960s (21) to 77 and 182, respectively, in 1994–1996 (22). The total fertility rate was 7.1 live-born children per woman in 1994–1996 (22). Monthly vaccination sessions for infants, organized by the IRD from 1987 to 1997 according to the national extended vaccination program in the 3 dispensaries of the area, were well attended (by 80% of eligible infants).
Study design
The study was prospective and observational. The infants were enrolled at dispensaries from April to December 1995 at the time of their first vaccination at the age of 2–3 mo. Length and weight measurements, taken routinely for all infants attending the vaccination sessions, were obtained from the database together with precise birth dates and sociodemographic information. During the children's second and third years of life, dietary, morbidity, and anthropometric data were collected in the children's homes during 4 rounds of visits at 6-mo intervals: November 1996, May and October 1997, and April 1998. The children's age ranges at the times of these visits were 13–23, 18–28, 23–33, and 29–39 mo, respectively.
Dietary data
Semiquantitative dietary data were collected during the home visits with use of 24-h maternal recalls. Any consumption of breast milk or snacks was noted, as was the number of meals of family food and gruel. During the last round of visits, the frequency of consumption of specific foods (meat, fish, and milk products) during the preceding week was also recorded. For the analysis, dates of weaning were obtained from the demographic database, which was last updated in September 1999.
Anthropometry
Length measurements taken at the time of inclusion were taken from the central database for both children included in the analysis and those lost to follow-up. Length was measured to the nearest mm by trained staff using a measuring board.
During the 4 home visits in the children's second and third years of life, several anthropometric measurements were taken according to standard procedures (23). Weight was measured to the nearest 10 g with a Seca baby scale (Hamburg, Germany), left midupper arm circumference was measured to the nearest mm, and recumbent length was measured to the nearest mm with a measuring board. From 24 mo of age, standing height was measured in addition to length. Triceps skinfold thickness was measured only during the last 3 rounds of visits with a pair of Holtain calipers (Zurich, Switzerland) precise up to 0.2 mm; the mean of 2 consecutive measurements was used. Two teams of 3 persons were in charge of data collection. The teams were well trained and measurements were standardized before each round. Each team was in charge of one-half of the children, so that a child was usually measured by the same person at each round. A standardized procedure for counseling the mother was used when a malnourished child was detected (arm circumference 12.5 cm).
Socioeconomic status and morbidity
Information on the age, religion, education, and occupation of the parents was obtained from the central database. Information on the precise dates of birth, death, or migration out of the study area of the study subjects or their parents was also obtained from the database when relevant. More than 90% of the parents had never attended school. Information about latrines, possessions, and the composition of each mother's hut were collected at the last home visit. The main possessions of the compounds were petrol lamps (98%) and radios (75%); the roofs of the huts were made of straw (68%) or corrugated iron. The type of wall—cement (20%) or mud bricks (80%)—was used as an indicator of economic status. During each home visit, the mothers reported episodes of fever, diarrhea, and coughing during the preceding week using their own definitions.
Statistical analysis
Increments in length, weight, arm circumference, and triceps skinfold thickness were computed as the difference between measurements taken at 2 consecutive rounds and were standardized to exact 6-mo intervals by dividing by the exact duration of the interval and multiplying by 6. Height-for-age, weight-for-age, and weight-for-height z scores were computed by using recumbent length before 24 mo, standing height thereafter, and the ANTHRO software (version 1.01; CDC/WHO, Atlanta).
The relation between the duration of breast-feeding and nutritional status was investigated first by two-factor analysis of variance (ANOVA) by comparing mean height-for-age and weight-for-height at the last round among 4 groups of children categorized by duration of breast-feeding. Prevalences of stunting [height-for-age <-2 z scores of the National Center for Health Statistics–World Health Organization reference (24)] and mild or moderate wasting (weight-for-height <-1 z score) were also compared by using chi-square tests. Second, age at weaning was tested for association with height-for-age at the last round, with and without adjustment for height-for-age in early infancy, by multiple linear regression analysis with adjustment for age at the last round, sex, maternal age, education, and economic status (quality of housing).
The relation between growth and breast-feeding was investigated by two-factor ANOVA and multiple linear regression. First, 6-mo increments were compared with two-factor ANOVA, by season, among 3 groups of children: those breast-fed throughout the interval, those weaned before the interval, and those weaned during the interval. Tukey's studentized range method was used for multiple comparisons among the 3 feeding-mode groups by one-factor ANOVA.
Thereafter, multiple linear regression analyses were performed with 6-mo increments as the dependent variable. First, the growth of children breast-fed throughout the interval was compared with the growth of those weaned before the interval with use of breast-feeding as a dichotomous variable. Second, another set of regression analyses used breast-feeding as a continuous variable, including only children weaned during the interval. For both sets of analyses, other independent variables were season and the quality of maternal housing, which might modify the relation between growth and breast-feeding, and age and anthropometric status at the beginning of the interval, which might act as confounders. Indeed, the last 2 variables were associated with the risk of weaning and potentially also with growth in length or weight. To avoid problems of correlated errors between initial length and length increments, initial weight was used to adjust for anthropometric status in analyses of growth in length, whereas initial length was used for the analyses of growth in weight. Weight was also used for the analyses of growth in arm circumference and in triceps skinfold thickness. Age and anthropometric variables were centered at their mean values for each analysis.
The significance of interactions (of first and second order) between breast-feeding, season, and maternal housing was tested, and interactions were included in the final model when P values were <0.10. The second order interaction (breast-feeding x season x maternal housing) was not significant for any analysis.
All growth analyses considered 6-mo increments in length, weight, arm circumference, and triceps skinfold thickness for 2 independent age groups: children in the second year of life and children in the third year of life. These groups were defined such that the former was close to the 18–24-mo interval and the latter was close to the 24–30-mo interval. Conditions were that the age at the beginning of the intervals could not overlap (ie, no increment could be included in both age groups) and that only one increment per child could be included in each age group. For the second year, children aged 16.0–20.9 mo at the beginning of an interval were included; for the third year, those aged 21.0–25.9 mo at the beginning of the interval were included.
The Mann-Whitney rank-sum test was used for comparisons between groups of variables with significant deviations from normality. For the analysis of differences in nutritional status and morbidity across the 4 home visits, global tests and tests for linear trends used ANOVA for repeated measures and the Mantel-Haenszel procedure applied to repeated measures (25) for continuous and dichotomous variables, respectively. Significance for two-tailed tests was P < 0.05. All analyses were performed with BMDP (version 7.0; SPSS Inc, Chicago) and SAS (version 6.12; SAS Institute Inc, Cary, NC).
RESULTS
Of 915 infants born in the study area from January to October 1995 and still residing in the area at the age of 2 mo, 176 did not attend the vaccination sessions and length measurements at 2 mo of age were missing for 128 children vaccinated during a cholera epidemic. Fifty-five of the included children died and 42 migrated out of the study area before the first home visit; an additional 27 died and 10 migrated out of the study area before the last home visit. Thirty-one children had left the study area temporarily at the time of the last home visit and thus had no height measurement taken at the age of 3 y. Age at weaning was not precisely known for 3 children weaned outside the study area. Data for the remaining 443 children were retained in the analysis.
The 472 children not included in the analysis differed from those included: their prevalence of a birth order <4 was greater (43.9% compared with 37.2%; P < 0.05), their mothers were younger (28.0 compared with 29.2 y; P < 0.05), and a higher proportion of their mothers had an occupation (10.3% compared with 7.5%; P < 0.05). There were no significant differences in the sex ratio and duration of breast-feeding of the children or the religion and education of the parents between the children not included and those included. The children who died during infancy or early childhood had a lower mean height-for-age than did those retained in the analysis (z score at 2–3 mo of age: -0.9 compared with -0.3; P < 0.01), whereas those lost to follow-up because of migration out of the study area or travel at the time of the last home visit were not significantly different from the children retained in the analysis.
All 443 children had complete data for the analysis of factors associated with nutritional status at the age of 3 y. For the growth analyses, 312 children were included for the second year of life and 402 for the third year. Children not included in the analyses of the second year of life were too old at the first home visit (n = 106), were too young at the first visit and too old at the second visit (n = 10), or had been absent during one round (n = 15). Similarly, children not included in the analyses of the third year of life had an inadequate age at measurement (n = 27) or missing data because of being absent at one round (n = 14). However, 436 children were included in 1 of the 2 series of analyses. There were no significant differences in socioeconomic or nutritional status between the 312 children included in the growth analyses for the second year of life and the remaining 131 children, or between the 402 children included in the growth analyses for the third year of life and the remaining 41 children (results not shown).
The cohort was characterized by high prevalences of malnutrition during the second year of life, especially during the rainy season (Table 1). The prevalence of stunting decreased slowly (P for trend < 0.01), mainly from the first to the second home visit, whereas the prevalence of wasting decreased dramatically across the 4 home visits (P for trend < 0.0001). The occurrence of fever during the week preceding the home visits was reported often at the end of the rainy season (for about one-half of the cohort) and less often at the end of the dry season (for one-fifth of the cohort).
View this table:
TABLE 1.. Children's characteristics at the 4 home visits1
The children's mean (±SD) age at weaning was 24.1 ± 3.7 mo. One-half of the children were weaned between 21.9 and 26.4 mo; the vast majority were weaned between 18 and 30 mo (92.1%). The mean number of daily meals was 3.3 for the weaned children compared with 2.6 for the breast-fed children (P < 0.001, Wilcoxon's rank-sum test), with no significant difference according to age. The staple food was cereal-based (millet and rice); fish was consumed at least once during the preceding week by 91.7% of the 3-y-old children, animal milk by 20.0%, and meat by 13.3%. Consumption of snacks, mainly groundnuts, during the preceding day was common (80.2% of the 12–23 mo-olds and 94.9% of the 24–36-mo-olds), whereas specially prepared gruels or other special foods were seldom consumed (3.5% and 2.8% during the second and third years of life, respectively).
Malnutrition and duration of breast-feeding
Mean height-for-age at the last round of visits, when the children were 3 y old (29–39 mo), was negatively related to age at weaning (z scores:-1.25, -1.05, -1.42, and -1.52, respectively, for ages at weaning of <22, 22–23.9, 24–25.9, and 26 mo; P < 0.01). The prevalence of stunting at 3 y of age also differed according to age at weaning (25.4%, 17.9%, 32.5%, and 35.3% in the 4 groups respectively; P < 0.05). The negative association between height-for-age and age at weaning, as a continuous variable, remained significant in a multiple linear regression analysis adjusted for sex and maternal age, housing, and education (P < 0.05).
Height-for-age at the age of 2–3 mo was also significantly related to age at weaning (z scores: -0.16, -0.09, -0.26, and -0.52 in the 4 groups, respectively; P < 0.01). When the analysis was adjusted for height-for-age during early infancy in addition to the variables mentioned above, height-for-age at 3 y became independent of age at weaning (P = 0.58). Weight-for-height at 3 y of age was not significantly associated with the duration of breast-feeding, nor was the prevalence of a weight-for-height <-1 z score.
Growth in relation to breast-feeding
Of the 312 children included in the analysis of the second year of life, 137 were breast-fed throughout the 6-mo interval, 27 were weaned before the interval, and 148 were weaned during the interval. Of the 402 children included in the analysis of the third year of life, 40 were breast-fed throughout the 6-mo interval, 184 were weaned before the interval, and 178 were weaned during the interval. Because triceps skinfold-thickness measurements were not taken during the first home visit, sample sizes in the 3 subgroups were smaller: 63, 13, and 56, respectively, for the second year of life and 30, 162, and 111, respectively, for the third year of life. For the analysis of the second year of life, the children's mean (±SD) age at the beginning of the interval was 19.0 ± 1.3 mo and the mean duration of the interval was 5.4 ± 0.2 mo. For the third year, the children's mean initial age was 23.6 ± 1.5 mo and the mean duration of the interval was 5.5 ± 0.2 mo.
The variable most closely associated with growth was season. Increments in length, weight, arm circumference, and triceps skinfold thickness were all significantly lower during the rainy season than during the dry season (Table 2). There was no interaction between season and breast-feeding for any variable in any age group, perhaps because the statistical power of the tests was insufficient. Length increments tended to be greatest in children breast-fed during the entire interval, and differences among the 3 groups were significant for the second year of life. For the second year of life, children weaned before the interval had significantly lower length increments than did those breast-fed throughout the interval (P < 0.05, Tukey's studentized range test). No significant differences were noted for weight increments, whereas the main effect of breast-feeding was significant for increments in both arm circumference and triceps skinfold thickness during the third year of life (Table 2). When increments were compared two-by-two according to feeding mode by using one-factor ANOVA, children weaned from the beginning of the interval had significantly greater increments in triceps skinfold thickness than did those breast-fed throughout the interval (P < 0.05), whereas no significant differences were found for growth in arm circumference.
View this table:
TABLE 2.. Six-month increments in length, weight, midupper arm circumference (MUAC), and triceps skinfold thickness for children breast-fed during the entire interval (BF), breast-fed during part of the interval (PT), or weaned before the interval (WE), by season and age1
Growth of breast-fed compared with weaned children
During the second year of life, children breast-fed throughout the interval tended to grow faster in length than did children weaned before the interval (Table 3). During the third year of life, breast-feeding and maternal housing interacted negatively with growth in length, whereas the main effect of breast-feeding was positive. Breast-feeding was associated with faster growth in length among children with poor housing, whereas the opposite was true in the subgroup with good housing (Figure 1). When the interaction term between breast-feeding and housing was not included in the model, no significant association was found between breast-feeding and growth in length.
View this table:
TABLE 3.. Multiple linear regression analysis of 6-mo increments in length, weight, and arm circumference, comparing children breast-fed throughout the interval (BF) with children weaned before the interval (WE), with adjustment for season, quality of housing, and initial age and weight1
FIGURE 1. . Mean (+1 SEE) estimated 6-mo length increments for breast-fed () and weaned () children, by season and maternal housing, during the second (n = 164) and third (n = 224) years of life. The results of multiple linear regression analysis with adjustment for age and weight at the beginning of the interval were as follows: P < 0.05 for the differences between breast-fed and weaned children in the third year of life; P < 0.01 for the differences between seasons for both the second and third years of life; P < 0.01 and P < 0.05 for the differences between children with good and poor housing for the second and third years of life, respectively; and P < 0.05 for the interaction between breast-feeding and housing in the third year of life (see Table 3).
Breast-feeding was not associated with growth in weight, neither as a main effect nor in interaction with another variable (Table 3). The rainy season was associated with slower growth in weight and good maternal housing was associated with faster growth in weight. However, in both age groups, a negative interaction was found between maternal housing and season. Children with good housing had greater weight increments than did those with poor housing during the dry season, but this advantage of good housing was not found during the rainy season (Figure 2).
FIGURE 2. . Mean (+1 SEE) estimated 6-mo weight increments for breast-fed () and weaned () children, by season and maternal housing, during the second (n = 164) and third (n = 224) years of life. The results of multiple linear regression analysis with adjustment for age and length at the beginning of the interval were as follows: P < 0.01 and P < 0.001 for the differences between seasons for the second and third years of life, respectively; P < 0.001 and P < 0.05 for the differences between children with good and poor housing for the second and third years of life, respectively; and P < 0.05 and P < 0.01 for the interactions between season and housing in the second and third years of life, respectively (see Table 3).
No main effect of breast-feeding was found for growth in arm circumference in any age group; however, during the third year, breast-feeding and maternal housing interacted negatively (Table 3). Among those with good housing, weaned children had greater arm circumference increments than did breast-fed children in both seasons. No significant differences in growth in arm circumference were found between breast-fed and weaned children in the subgroup with poor housing.
Triceps skinfold-thickness increments were not available for the dry season during the second year of life. During the rainy season, growth in triceps skinfold thickness did not differ significantly between weaned and breast-fed children (results not shown). During the third year of life, breast-fed children tended to have slower growth in triceps skinfold thickness than did weaned children, but not significantly so (Table 4).
View this table:
TABLE 4.. Multiple linear regression analysis of the 6-mo increments in triceps skinfold thickness during the third year of life, by breast-feeding, season, quality of housing, and initial age and weight1
Growth in relation to duration of breast-feeding
The association between growth and the duration of breast-feeding was also analyzed. Duration of breast-feeding was positively associated with growth in length: the longer the children were breast-fed, the faster they grew during both the second and third years of life (Table 5). No association with duration of breast-feeding was found for growth in weight and arm circumference (Table 5) or for growth in triceps skinfold thickness (Table 4).
View this table:
TABLE 5.. Multiple linear regression analyses of 6-month increments in length, weight, and arm circumference, by duration of breast-feeding (BF) within the interval, season, quality of housing, and initial age and weight, including only children weaned during the interval1
No significant interactions were found between duration of breast-feeding and the quality of maternal housing for growth in either length, weight, arm circumference, or triceps skinfold thickness. Season and maternal housing interacted negatively with growth in weight (in both the second and third years of life) and with growth in arm circumference (in the third year of life only, Table 5). Children with good housing grew faster than did those with poor housing during the dry season, but, as in the analysis between breast-fed and weaned children, this advantage of housing was not found during the rainy season. As noted in the previous analyses, the main effect of season was highly significant: growth in length, weight, and arm circumference (Table 5) and in triceps skinfold thickness (Table 4) was faster during the dry season than during the rainy season.
DISCUSSION
The results of our study suggest that a long duration of breast-feeding has a positive effect on linear growth in the second and third years of life in rural African toddlers. Our results also show how both a greater prevalence of stunting and enhanced linear growth can be associated with a long duration of breast-feeding within the same population. This apparent contradiction arises from a lower initial height-for-age of children breast-fed for long durations (16), because the difference disappears when height-for-age is adjusted for length during infancy.
Maternal reasons for early and late weaning were investigated in this cohort through interviews with the mothers. As reported elsewhere (26), about one-third of the mothers who breast-fed for >2 y said that they delayed weaning because their children were "little and weak." Children weaned late because of low nutritional status were shorter than weaned children of a similar age, whereas those weaned late for other reasons were not (26).
Given the scarce knowledge of maternal motivations for weaning in most parts of Africa, reverse causality is a plausible explanation for the greater prevalence of stunting among breast-fed children in many countries (9), although residual confounding by socioeconomic factors is possible in some settings. In a population studied in rural Kenya, mothers did not seem to take their children's nutritional status into account when deciding to wean (20, 27), but in that setting there was no significant difference in height-for-age or weight-for-height between breast-fed and weaned children when age, income, and maternal education were adjusted for (27). In southern Brazil, breast-fed infants had a lower nutritional status than did weaned infants, without any evidence of reverse causality (10), but these Brazilian mothers lived in a very different socioeconomic context with access to formulas and processed infant food. In such environments, breast-feeding may not provide any benefit for growth. Breast-fed infants from a poor population of Mexico City were reported to grow faster than formula-fed infants (28).
Another reason for differences in results among previously published studies could be differences in the definition of breast-feeding. Victora et al (29) suggested that mothers may modify infant feeding, ie, they may introduce complementary food or stop breast-feeding, in response to slow growth. Thus, these authors argued that for analyses of the association between growth and breast-feeding, children should be separated into 2 groups according to feeding mode at the beginning of the interval considered: those breast-fed throughout the interval or weaned during the interval compared with those weaned from the beginning of the interval, a so-called intent-to-feed analysis. However, we consider that if weaning depresses growth, children weaned during the interval should not be pooled with those breast-fed throughout the interval, because this approach would underestimate the growth of breast-fed children. The only study that reported slower growth in weight and length for children breast-fed during their second year of life used precisely this approach (15).
In the present study, we found no evidence that slow growth leads to weaning. However, we did not pool children weaned during the interval with children in either of the 2 other groups. Rather, we used one set of analyses to compare children breast-fed throughout the interval with those weaned before the interval and a second, independent set of analyses to compare all children weaned during the interval in terms of duration of breast-feeding. A positive association of breast-feeding with growth in length was found with use of both approaches.
Breast-feeding was associated with improved linear growth without any significant relation to growth in weight. A negative association with growth in arm circumference was found in the subgroup of children with good housing in 1 of 4 analyses, together with a nonsignificant tendency toward slower growth in triceps skinfold thickness for breast-fed children in 1 of 4 analyses. Onyango et al (20) found a positive association between breast-feeding and growth in both length and weight, which was stronger for length. Marquis et al (18) reported on growth in length only. As far as we know, no previous studies considered arm circumference and skinfold thickness.
One explanation for the lack of association between breast-feeding and growth in weight may be that prolonged consumption of breast milk improves the children's intake of particular nutrients and micronutrients more than their overall energy intake. Another explanation may be that growth in weight is more strongly affected by illness than is growth in length, so that reverse causality (a higher probability of weaning among healthy, steadily growing children) obscures the effect of breast-feeding on growth in weight. In a Peruvian study, the positive association of breast-feeding with linear growth existed only after exclusion of a subgroup of children with a high prevalence of diarrhea and low weight-for-age before 12 mo, who were less likely to be weaned from 12 to 15 mo (18). Evidence of reverse causality related to illness was found in Senegal as well: mothers reported delaying weaning not only when the child was stunted, but also in cases of frequent or current illnesses (26). Thus, the strength of the association between linear growth and prolonged breast-feeding in this Senegalese setting was probably underestimated.
Even after exclusion of children with evidence of reverse causality in the Peruvian cohort, the main effect of breast-feeding on linear growth was not significant: breast-feeding was positively associated with linear growth only in children with low intakes of animal products (18). In Senegal, a long duration of breast-feeding was significantly associated with a higher length velocity in the entire cohort during both the second and third years of life, probably because this rural West African setting is poorer, with an overall low intake of animal products and higher prevalences of morbidity, especially fever. However, when breast-fed and weaned children were compared, not all children benefited equally from breast milk during the third year of life: breast-feeding appeared to be associated with enhanced growth in length only for children from the poorest families, ie, 80% of the cohort.
During the rainy season, growth was significantly reduced, as was consistently reported in West Africa (14, 30, 31) and other parts of the developing world (32, 33). Malaria, which is transmitted mainly at the end of the rainy season in this area (34); other illnesses; low food availability; and heavy maternal workloads are potential explanations for the children's slow growth during the rainy season.
No interaction was found between season and breast-feeding, although a greater positive effect of breast-feeding on growth in length was expected during the rainy season because of food shortages and higher prevalences of fever during this time. Breast-fed infants were reported to maintain their food intake during illness better than do other infants because their appetite for breast milk is not affected (35). However, some mothers of children in this cohort stated that they delayed weaning during the rainy season either because of frequent illness or because of a lack of food and money before the harvest (26). Therefore, as for growth in weight, reverse causality may have masked any particular benefit of breast-feeding during the rainy season. Conversely, the greater ability of breast-fed children to maintain their food intake during illnesses may not exist beyond infancy, when breast milk contributes less to the diet.
In conclusion, the higher prevalence of stunting in children breast-fed for long durations in this community can be explained entirely by reverse causality, because the higher prevalence disappeared after adjustment for height-for-age in infancy. A positive effect of prolonged breast-feeding on linear growth probably exists in many parts of rural Africa, where the prevalence of morbidity and the quality of complementary food are equivalent, but not necessarily in more-developed settings.
ACKNOWLEDGMENTS
We thank Adama Marra for programming assistance, Laurence Chabirand and Agnès Gartner for logistic support, and Sybil Pinchinat and Yves Martin-Prevel for help with the SAS analyses.
REFERENCES