The Journal of Nutrition Vol. 128 No. 8 August 1998, pp. 1328-1335

Infant Feeding Practices in Barbados Predict Later Growth^1,2,3

Janina R. Galler⁴, Frank C. Ramsey^*, Robert H. Harrison, Rechele Brooks, and Sally Weiskopf-Bock

Center for Behavioral Development and Mental Retardation, Boston University School of Medicine, Boston, MA 02118 and ^* Rus-in-Urbe Clinic, Bridgetown, Barbados

	ABSTRACT

Abstract Introduction Methods Results Discussion References

This longitudinal study was designed to examine reciprocal relationships between feeding practices and infant growth over the first 6 mo of life. The following three hypotheses were tested: 1) early feeding practices predict later infant growth; 2) early infant growth predicts later feeding practices; and 3) these relationships occur after controlling for related background variables. The sample included 226 healthy, well-nourished infants born at the Queen Elizabeth Hospital, Bridgetown, Barbados. Assessments were made at birth, 7 wk, and 3 and 6 mo of age. Factor analysis of a feeding practices questionnaire for those mothers who attended all three postnatal visits yielded five uncorrelated factors. Three of these factors, preference for breast-feeding, feeding intensity and feeding difficulty, declined with infant age. Two of these factors, father helps and relatives help, increased with infant age. Several background variables, including maternal age, anthropometry and reproductive history, and reliance on outside sources of information were correlated with infant growth. Multivariate analyses confirmed all three hypotheses. The group of feeding practices (particularly the preference for breast-feeding) at 7 wk predicted increases in infant lengths at subsequent ages. Conversely, infant weights at 3 and 6 mo predicted subsequent feeding practices, especially feeding intensity. These reciprocal relationships remained even after statistically controlling for the influence of the background variables. Implications for public policy include the need for comprehensive programs advocating breast-feeding and supporting the general health of mothers and infants.

	INTRODUCTION

Abstract Introduction Methods Results Discussion References

Published research has shown that breast-feeding improves infant physical growth and motor development at least in the first 6 mo of life (Castillo et al l996, Rao and Kanade 1991, Seward and Serdula 1984). There are fewer studies relating breast-feeding and cognitive outcome (Rogan and Gladen 1993, Horwood and Ferguson l998), but these have also demonstrated lasting benefits. The potential role of breast-feeding in promoting both physical and cognitive development is thought to be greatest for disadvantaged populations. However, in these settings, it often remains difficult to dissociate the effects of feeding practices on infant outcome from the consequences of malnutrition, poor environmental conditions, overdilution of formula and infectious diseases (Van Derslice et al. 1994). This study, which was conducted in Barbados, was uniquely designed to address these issues. Barbados is one of the few developing countries that is not characterized by extreme poverty or malnutrition; it has high standards of health care and education (Ramsey et al. 1984, UNICEF 1993). Thus, confounding effects of poverty, malnutrition and infection were not present.

Several reports have now been published showing the complexity of the relationships between feeding practices and infant outcomes. We suggested the reciprocal relationship of mother and infant during feeding in our earlier review of the role of mother-infant interaction in nutritional disorders (Galler et al. 1984). Two recent longitudinal studies in the Philippines (Adair et al. 1993) and in Peru (Piwoz et al. 1994) showed that infant characteristics, especially infant weight gain, affected later feeding decisions in low income populations in which malnutrition was present. In both studies, heavier babies were more often breast-fed at later ages. Moreover, low infant weight gain in Peru was associated with more changes in maternal feeding practices over time. The Peruvian study also emphasized the bidirectional nature of the relationship between feeding practices and infant outcomes.

We addressed these issues in a series of reports on the relationships between feeding practices and infant growth in Barbados (Galler et al. 1998). In the 1980 UNICEF report on the State of the World's Children (UNICEF 1980), Barbados was highlighted because of its low prevalence of breast-feeding, raising concern about possible implications for children growing up in this island nation. We undertook this study as a means of documenting those factors associated with the preference to breast- or bottle-feeding, and we also documented the growth and development of children exposed to different feeding practices in Barbados. We followed the growth of 226 healthy infants born at the Queen Elizabeth Hospital (QEH), Bridgetown for 6 mo after birth. Assessments were made at 7 wk, and at 3 and 6 mo of age. We documented feeding practices using a range of questions, which were then factor analyzed to create independent categories of interrelated items. We were thus able to avoid the use of categorical descriptions of feeding, such as predominant or partial breast-feeding, which may be misleading (Rao and Kanade 1991, Winikoff 1981). Conclusions based on such categories often ignore other aspects of feeding practices and their variations over time that are present in most families (Piwoz et al. 1994). The longitudinal design of this study permitted us to test the following three hypotheses: 1) that early feeding practices predict infant growth at later ages; 2) conversely, that infant size at early ages predicts feeding practices at later ages; and 3) that hypotheses (1) and (2) are independent of background variables, including environmental conditions and maternal characteristics.

	SUBJECTS AND METHODS

Abstract Introduction Methods Results Discussion References

Site.  We conducted this study in Barbados, a small island in the Eastern Caribbean with a population of 260,000 inhabitants. The population is homogeneous, generally of low middle class. There is little migration into or out of the island. With a literacy rate of 99%, socioeconomic standards in Barbados are higher than in other Caribbean islands. We based our study at the QEH where 99% of all births occur. The hospital has ~4000 births per year. Health care is of a high standard; all women receive routine prenatal care at the hospital or at local polyclinics. We have previously followed women and children from this population with little attrition after many years and have had maximal cooperation from all participants in our surveys. Conditions in Barbados were optimal for ensuring accurate and reliable information. The study was approved by the Boston University Medical Center Institutional Review Board for Human Subjects (Protocol E1962).

Sample.  The infants recruited for this investigation included every other healthy baby born at the QEH during a 2-mo period from October 17 to December 17, 1986. We identified a total of 257 mothers and infants as meeting the following criteria: 1) adequate maternal nutritional status; 2) birth weight 2.5 kg; 3) vaginal delivery; 4) no prenatal, perinatal or postnatal complications, and Apgar score 5; and 5) agreement by the mother to participate in the survey. Children admitted to the Premature or Intensive Care Units were not included. Twenty-eight women, mothers of 20 boys and 8 girls (² = 5.11; P < 0.05), chose not to participate in the study for a variety of reasons. We excluded three infants from the analysis after the study began: one infant died after her 7-wk follow-up appointment; the second child was hospitalized for malnutrition secondary to a misshapen palate interfering with routine feeding; the third was found to have incomplete birth records. Of the final 226 infants, 115 were girls and 111 were boys. The full sample included 217 healthy infants with a gestational age (GA) of 37 wk or more and nine babies with a GA between 35 and 37 wk. The latter were included because they met the inclusion criteria at birth. Mothers (n = 226) included in the study were typical of Barbadian women of child-bearing age (Ramsey et al. 1984). Their mean age was 25.3 ± 5.3 y. They had a mean of 2.3 ± 1.5 births (including the study child) and a total of 2.6 ± 1.6 prior pregnancies. Nonparticipants were comparable to participants with respect to age, parity and gravidity, and their infants had comparable birth weights and lengths.

Time line.  We undertook evaluation of mothers and infants at four time points: birth, 7 wk, and 3 and 6 mo of age. Birth and obstetric records were reviewed within 24 h of birth. Mothers were interviewed concerning their feeding practices on each of the subsequent visits, and anthropometric measurements were taken for infants at each time point. The number of mothers attending the visits was as follows: 158 (69%) at 7 wk, 168 (74%) at 3 mo and 209 (92%) at 6 mo. A total of 112 women and infants attended all three postnatal visits.

Measures

Anthropometry.  Infant birth weights and lengths were derived from the hospital records. Anthropometric measurements at all three follow-up visits were done by the project pediatrician using the CMS scale (CMS Weighing, UK). The project scale was calibrated against standard weights at regular intervals and no less than weekly. Infants were weighed in their diapers to the nearest 0.1 kg. Infant lengths were measured to the nearest 0.5 cm using an infantometer (Graham-Field Surgreal, New York, NY) as the child lay in a supine position. In the absence of weight-for-age or length-for-age norms for Barbados or the Caribbean, we used a computer program (Sullivan and Gorstein 1990) to convert raw measurements into Z-score values, based on the National Center for Health Statistics (NCHS 1977) reference data. Maternal anthropometry was assessed only at the 6-mo follow-up visit. Weights were measured to the nearest 0.1 kg, and standing heights were measured to the nearest 0.5 cm. Arm circumference was taken at the midpoint between shoulder and elbow on the left arm.

Maternal reproductive history.  Measures of maternal age, parity, gravidity and length of gestation were derived from the clinical record. In the case of gestational age, maternal reports of last menstrual periods and results of the clinical examination were both reviewed, but the latter were used for the analyses described below.

Feeding practices interview.  We developed and piloted a feeding practices interview in Barbados. This interview was administered by local community health sisters who were familiar with the population and with field research. Each interview protocol included 40 quantitative and descriptive questions, which were potentially codable into 149 items. For this report, 24 items derived from the interview were used in the overall factor analysis described below because they were asked at all three follow-up visits. Thus, we excluded 58 items because they depended on prior yes or no responses or did not apply to all women at each visit. Four items were redundant, and were asked primarily as a measure of reliability, and 31 items were open-ended questions and not easily coded. Eighteen other items had insufficient variability. Finally, 18 items were summarized and recoded as 4 variables. Answers either were yes/no or ranked by the mother on a three-point scale.

Because there were no standard scoring procedures for the feeding practices interview, exploratory factor analysis was used to develop appropriate scales. The 24 variables were subjected to principal component analyses, (see Data analysis), one for each of the three follow-up visits at 7 wk, and at 3 and 6 mo after birth. Factor analyses of the data produced interpretable factor structures, but they were not directly comparable from one age to the other. Therefore, we combined the data from all three ages into one overall factor analysis, using only mothers and babies who had complete data at all three follow-up ages (n = 112). For this analysis, we treated each mother as three independent subjects only for the purposes of grouping the interview items. Thereafter, each mother was assigned three sets of factor scores, one set for each follow-up visit, using the same factor score coefficients. This procedure increased the number of observations from 112 to 336 and permitted the comparison of changes in feeding practices over time in the same mother-infant pair. The factor structure of the overall analysis was similar to those of the cross-sectional analyses at each age separately, confirming the validity of the comprehensive analysis. Five principal components, accounting for 45% of the total variance, were rotated by the normal varimax criterion, yielding an easily interpretable solution. The Scree test (Cureton and D'Agostino 1983), which focuses on a significant drop in eigenvalues, was also used as a criterion in selecting five factors. For each factor, variable loadings (>0.40) and Armor thetas (1973), measuring internal consistency, are presented in Table 1.

To supplement the factor analysis and make the data comparable to other published studies, we also coded mothers' reports of daily breast-feeding practices and use of breast-milk substitutes into a five-point feeding scale. Feeding categories were similar to those developed by Piwoz et al. (1994) in a longitudinal study of Peruvian infants. Because most of the mothers offered liquids, all feeding categories included mothers who fed water and other nonnutritive liquids or fruit juice to their infants. The five scale categories were as follow: 1= exclusive bottle-feeding; 2 = mostly bottle-feeding, when mothers reported at least three more bottle-feedings than breast-feedings during the previous 24 h; 3 = mix of breast- and bottle feeding; 4 = mostly breast-feeding, when mothers reported at least three more breast-feedings than bottle-feedings during the previous 24 h; and 5 = breast only, when mothers reported no formula use.

Environment questionnaire.  The questionnaire was derived from earlier questionnaires we have administered in Barbados (Galler and Ramsey 1985, Galler 1987). The questionnaire consisted of 57 scorable items covering three main areas, i.e., socioeconomic conditions, maternal medical, social and educational histories, and current living situations. These items were considered the most representative in describing the background histories of Barbadian families. We interviewed mothers at the 7-wk visit. Items derived from the questionnaire were grouped into nine categories (factors) of related items using a principal components analysis. The factors were not correlated with one another. The first factor concerned household composition (Factor 1), which included the number and classification of adults in the family and sources of support. Two factors reflected the family's standard of living, home conveniences (Factor 2) and family income (Factor 3). Father's income (Factor 4) referred to the nature of the father's contribution to the household income. The remaining factors focused specifically on the mother as follows: maternal reproductive age (Factor 5), child-rearing experience (Factor 6), childhood closeness to parents (Factor 7), maternal health (Factor 8) and information seeking (Factor 9). The latter refers to the number of books, magazines and newspapers read by the mother and whether the mother routinely sought advice from professionals. Armor's thetas confirmed good internal consistencies of the individual factors. They ranged from 0.86 and 0.84 for Factors 1 and 2 to 0.43 and 0.39 for Factors 8 and 9.

Data analysis.  Data reduction was undertaken using principal components analysis with rotation to the orthogonal simple structure using the normal varimax method (Cureton and D'Agostino 1983). Factor scores from the questionnaires were computed by the SAS factor scoring program (SAS Institute 1990). Each factor score was entered into a two-way ANOVA with infant age and sex as independent variables and infant age as a repeated measure to compare changes in feeding practices over time. Similarly, two-way (infant age by sex) ANOVA with age as a repeated measure was applied to analyze changes in infants' weight-for-age and length-for-age Z-scores over time. In analyzing changes of Z-scores since birth, we used residual Z-scores adjusted for birth scores (subsequent scores adjusted for birth scores through linear regression). We selected residual scores rather than delta scores for these analyses, because the former have lower standard deviations and produce lower error variance (Keppel 1973). Residual scores are not adversely affected when outcomes are corrected by less reliable predictors. Although most predictors in the study were reliable, birth weights and lengths were taken from hospital records and not measured directly by the research team. When indicated, Newman-Keuls tests were used to analyze pairwise differences (Miller 1981).

Before combining data for boys and girls, we computed correlations separately. Bartlett's test for differences between covariance matrices was used to evaluate sex differences in correlation coefficients (Kendall et al. 1983). Canonical correlations (Rc) were used to describe overall relationships between sets of variables (Harris 1974). These were supplemented by multiple correlations (R) to describe the relationship between one set of variables and a single outcome, and by Pearson correlations (r). When it was necessary to control for confounding factors (e.g., maternal anthropometry), we used partial canonical correlations (Rcp), partial multiple correlations (Rp) and simple partial correlations (rp). Following Cohen and Cohen (1983) we report significant multiple correlations only when a significant (P < 0.05) canonical correlation "protected" their significance. Similarly, simple correlations were "protected" by significant multiple correlations. This was done to reduce the likelihood of chance findings.

Following Lawley (l959), Wilks' lambda and its accompanying F-ratios were used to test the significance of entire sets of canonical roots (SAS 1990). Roy's Greatest Root Test was used to test the significance of infant weight and length outcomes because sets of follow-up weights were considered to represent variations on a single weight construct. Follow-up lengths were considered to represent variations on a single length construct (Harris 1974).

	RESULTS

Abstract Introduction Methods Results Discussion References

For each analysis reported below, we used all available mother-infant pairs who had complete data for that analysis. As more measures were added for each successive analysis, sample sizes diminished. For the final analysis relating feeding practices and growth over time, we included only mother-infant dyads who attended all three follow-ups visits at 7 wk, and at 3 and 6 mo. To exclude the possibility of sampling bias, we compared those mothers and infants who attended all three follow-up visits (n = 112) with those who did not (n = 114). The excluded mothers did not differ (P > 0.05) from the included mothers on any demographic or anthropometric measurement. They also did not differ in their preferences for breast- or bottle-feeding on any of the three follow-up visits.

Infant ages at the three interview points varied more than we had intended. The mean infant age was 7.01 (±1.37) wk at the 7-wk visit, 13.72 (±1.06) wk at the 3-mo visit and 26.56 (± 1.70) wk at the 6-mo visit. To evaluate the significance of these variations, we first compared infants whose ages were within 25% of the mean follow-up age with those whose ages fell outside these limits. Although the numbers of within-range and out-of-range infants varied with each measure and with each follow-up visit, we found no significant differences (P < 0.20) between the two groups on any of the maternal or infant measures considered. Therefore, we did not exclude these mother-infant pairs in the data analyses described below.

Consistency and change in feeding practices as a function of age.  All women in the sample initiated breast-feeding. At 7 wk, 69% of women were exclusively or mostly breast-feeding (feeding scale category 4 or 5) and 22% of the sample were breast- and bottle-feeding approximately half the time (category 3). At 3 mo, these percentages were 44% (categories 4 and 5) and 34.5% (category 3). At 6 mo, the percentages were 17 and 31%. Thus, approximately half of the women employed some form of breast-feeding at 6 mo. Table 2 shows the mean scores derived from the feeding scale and from the five interview factors (from an overall mean = 0, SD = 1). Repeated measures ANOVA (age by sex) yielded highly significant F-values for age for both the feeding scale and for all five factors, but there were no sex effects or interactions of sex by age. The F-value for preference for breast feeding (Factor 1) (P < 0.0001) showed that formula feedings were increasingly used over the first 6 mo of life. The linear correlation of infant age with preference for breast feeding (Factor 1) was 0.50 (P < 0.0001; df = 334). ANOVA results for the feeding scale were similar to those for the preference for breast feeding (Factor 1), and scores from the two measures were highly correlated (r = 0.78). Due to collinearity, we excluded the feeding scale from the multivariate analyses described below.

Table 2 also shows that feeding intensity (Factor 3) declined steadily over time (r = 0.44; P < 0.0001), as did feeding difficulty (Factor 5), (r = 0.30; P < 0.001). Father helps (Factor 2) increased with infant age (r = 0.12; P < 0.05), particularly at the 6-mo follow-up. The relationship between relatives help (Factor 4) and infant age (P < 0.01) was curvilinear. Scores increased markedly from 7 wk to 3 mo, and then declined slightly after that. In general, the relative position of individual mother-infant dyads remained stable over time. Thus, the longitudinal correlations for each factor at adjacent ages averaged about 0.55 except for the feeding difficulty factor whose consistency over time was lower (r = 0.35).

Infant anthropometry as a function of age.  In all analyses of infant anthropometry, we report infant weights and lengths as Z-scores. Among the boys, rank values for mean weight at all four ages varied from the 46th to the 67th percentile of the NCHS (1977) norms. Rank values for mean length varied from the 45th to the 52nd percentile. Values for boys and girls were similar. No infants were found to be undernourished. ANOVA showed that mean Z-scores for weight were above the NCHS mean [F(1,107) = 33.41; P < 0.001]. ANOVA also showed that infant age was associated with the weight-for-age Z-scores [F(3,107) = 35.22; P < 0.0001], but there was no significant effect of sex or any interaction of age and sex. A post-hoc Newman-Keuls test, showed that mean Z-scores at 7 wk (0.59) and 3 mo (0.61) were higher than NCHS means and both were higher than the mean Z-score at birth (0.06; P < 0.05) or at 6 mo (0.28; P < 0.05). In contrast, length did not differ from the NCHS average at any age tested. Because there were no significant sex effects on weight- and length-for-age Z-scores at any of the four ages, we combined data for boys and girls in all subsequent analyses.

The role of background factors.  Before testing the three main hypotheses in this study, we considered the possibility that both feeding practices and infant outcomes might be modified by other background factors, including gestational age, maternal reproductive history, anthropometry and the set of environmental factors.

Relationships with feeding practices. It was possible that maternal characteristics, including reproductive history, anthropometry and environmental conditions could alter the pattern of feeding practices and thus affect infant outcomes. This was tested by applying separate multivariate correlations at each follow-up age between the group of feeding practice factors and each set of background factors. Reproductive history (maternal age, length of gestation, and parity) was significantly correlated with the group of feeding practices measured at 7 wk (Rc = 0.42; P < 0.05) and 3 mo (Rc = 0.38; P < 0.05) but not at 6 mo. Maternal age and parity were responsible for these significant relationships. Older mothers and those with more children were more likely to report feeding difficulties and to have the help of the infant's father. Environmental conditions (nine factors reported in Subjects and Methods section) were significantly correlated with feeding practices at 7 wk (Rc = 0.67; P < 0.0001), 3 mo (Rc = 0.55; P < 0.0001) and 6 mo (Rc = 0.65; P < 0.0001). In contrast, maternal anthropometry (weight, height and arm circumference) was not associated with feeding practices.

Relationships with infant anthropometry. In a parallel series of analyses, we also examined the relationships between the background variables and infant weight and length over the first 6 mo of life. As noted above, it was possible that the background variables could also play a role in determining infant size. We therefore examined the extent to which potential confounding may have occurred.

First, reproductive history was examined in relationship to infant weights and lengths at birth and the three follow-up ages. An overall canonical correlation between reproductive history and infant length was significant (Rc = 0.38; P < 0.01), and there was also a significant parallel correlation with weight (Rc = 0.36; P < 0.01). The relationship between reproductive history and length was most striking at 3 mo (R = 0.28; P < 0.05). At this age, simple correlations between gestational age, parity and maternal age and infant length showed that only the first-mentioned was a significant factor (r = 0.21; P < 0.05). Thus, babies with a longer gestational age were longer at 3 mo. Reproductive history and weight were significantly associated only at birth (R = 0.34; P < 0.01), and this relationship was accounted for by gestational age and parity (r = 0.24 and 22, respectively; both P < 0.05). Next, these analyses were repeated controlling for birth weight and length. When the birth measure was corrected, the relationships between reproductive history and length persisted (Rc = 0.38; P < 0.05). The correlations approached significance only in the case of weight (P = 0.0503). The step-down multiple and simple correlations relating reproductive history and infant anthropometry (weight and length) at individual ages were no longer significant. However, the multiple correlation between maternal age and infant weights at the three postnatal ages tended to be significant (P = 0.0512).

Second, we assessed the role of maternal anthropometry in predicting infant growth. Overall canonical correlations between the group of maternal anthropometric measures and infant growth at the four ages were significant for both infant lengths (Rc = 0.32; P < 0.05) and infant weights (Rc = 0.40; P < 0.01). In the case of infant length, there were no significant multiple correlations with maternal measures when the follow-up ages were considered one at a time. In contrast, associations between maternal measures and infant weights were present at both 7 wk (R = 0.37; P < 0.01) and 6 mo (R = 0.29; P < 0.05). Maternal height, weight and arm circumference were all correlated with infant weight at these two ages. In the parallel series of analyses in which birth measures were controlled, a significant relationship with weight was still present (Rcp = 0.30; P < 0.05), whereas the relationship with length was no longer significant. Corresponding multiple correlations showed significant associations between maternal anthropometry and infant weight at 7 wk (R = 0.28; P < 0 .05), and a significant simple correlation between maternal height and infant weight (r = 0.28; P < 0 .01) at this age. When examined separately, both maternal height and weight were correlated with the group of postnatal weights at three ages (both r = 0.28; P < 0.05). Thus, the relationship between maternal anthropometry and infant weights persisted even after controlling for the corresponding birth measures.

Third, we examined the relationships between the environmental conditions (nine factors) and infant anthropometry. An overall canonical correlation between the group of environmental factors and the four infant lengths was significant (Rc = 0.45; P < 0.05) when birth length was not controlled. In contrast, there were no significant associations with infant weight. The association between environment and infant length was still significant even after birth length was controlled (Rc = 0.41; P < 0.05). When the environment factors were examined individually, information seeking (Factor 9) was found to be significantly correlated with the group of three infant lengths (R = 0.29; P < 0 .05). However, multiple correlations between the group of environment factors and length at the individual infant ages were not significant.

In summary, these results show that all three background factors were significantly associated with infant growth. After controlling for birth measures, maternal anthropometry was significantly associated with the group of infant weights. Similarly, reproductive history and environmental factors were found to be significantly associated with infant lengths, even after correcting for birth size. We therefore controlled for maternal weight, height and age and also information seeking (Factor 9) in the subsequent analyses examining the relationships between feeding practices and infant outcomes.

Correlations between feeding practices and infant anthropometry.  For these analyses, 101 mother-infant pairs had complete data sets containing all maternal and infant measures for all visits. The maternal characteristics of this latter group did not differ significantly from the group of women for whom complete data sets were not available. In all analyses, birth weight was partialled out from subsequent weight measurements and birth length from subsequent length measurements to eliminate possible differences among infants at birth. Thus, we could interpret significant correlations between feeding practices and infant anthropometry as referring to weight and length gains after birth. The analyses were also performed with and without controlling for the background variables addressed in the previous section. Because the results of the two sets of analyses were similar, we report only those analyses that controlled for these variables.

We first compared the correlations between feeding practices and infant outcomes to determine whether these were different for boys compared with girls. Bartlett's test for homogeneity of covariance matrices (Kendall 1983) was applied separately for each gender to the relationships between the 15 feeding factor scores (5 factors at three infant ages) and infant weights at four ages (including birth). The resulting chi-square for infant weight was not significant (² = 172; df = 190). The parallel test for the four infant lengths was also not significant (² = 206; df = 190). These results indicated that gender did not affect the predictive relationships between feeding practice factors and infant anthropometry. We therefore combined the data for boys and girls for the analyses reported below.

Next, we computed correlations between the group of feeding practices at each age and the group of infant outcomes at all three postnatal ages. The overall (canonical) correlations were supplemented with multiple-partial correlations between individual feeding practices factors and infant outcomes. These included predictive, proximate or postdictive relationships. Predictive relationships, which tested hypothesis (1), refer to the correlations between feeding practices taken at previous visits and subsequent infant anthropometry. Proximate relationships are correlations between feeding practices (over the period immediately preceding the visit) and infant anthropometry assessed at that same visit, and also test hypothesis (1). Postdictive relationships refer to the correlations in which feeding practices were assessed following the anthropometric measurement, testing hypothesis (2).

Feeding practices at 7 wk and 6 mo were associated with infant size. Data analysis confirmed these observations, especially in the case of length. Thus, significant associations were present between the group of feeding practices reported at 7 wk and the group of infant lengths across the three ages (Rcp = 0.42; P < 0.01). When the feeding practices were examined separately, we found that this result was in fact attributable to the significant multiple correlation between the preference for breast-feeding (Factor 1) at 7 wks and the set of three birth-adjusted lengths (Rp = 0.34; P < 0.05). When each age was examined separately, we found a significant proximate relationship between the preference for breast-feeding at 7 wk and infant length at the same age (rp = 0.23; P < 0.05). There was also a predictive relationship between this feeding factor and infant length at 6 mo (rp = 0.26; P < 0.05). The predictive relationship between Factor 1 and infant outcomes is illustrated in Figure 1. The figure shows that, when factor scores were dichotomized at the mean, mothers with a higher preference for breast-feeding at 7 wks had longer infants at most ages tested. A similar, though less striking relationship was present for infant weights. The five-point feeding scale was not included in the multivariate analysis, but was analyzed for the purpose of comparing our results with those of other studies. We found that the feeding scale was not correlated with the group of infant lengths, although there was a significant correlation between the 7-wk score and infant length at 6 mo (rp = 0.22; P < 0.05). Thus the feeding scale was less likely to predict outcomes than the factors derived from the feeding practices questionnaire.

View larger version (21K): [in this window] [in a new window]   Fig 1. Infant weight- and length-for-age Z-scores for high and low breast-feeding. Mothers' feeding factor scores were dichotomized into those indicating a higher (greater than the mean) preference for breast-feeding and those indicating a lower (less than the mean) preference for breast-feeding. Infant weight and lengths for those two groups are plotted as a function of age. At birth, maternal weight, height, and parity, and infant gestational age were all held constant. At 7 wk and at 3 and 6 mo, birth weight or birth length, maternal weight, height, and parity, and infant gestational age were all held constant. Values are means ± SEM, n = 101.

For weight, a significant association was present between feeding practices recorded at 6 mo and the group of birth-adjusted infant weights (Rcp = 0.47; P < 0.001). The result was attributable mainly to the significant multiple correlation between feeding intensity (Factor 3) and the set of infant weights (rp = 0.44; P < 0.001). Correlations at the individual infant ages showed that feeding intensity at 6 mo was proximately related to infant weight at 6 mo (rp = 0.40; P < 0.001). It was postdictively associated with infant weight at 3 mo (rp = 0.34; P < 0.01). These correlations indicated a bidirectional relationship between suckling intensity and weight gain. Of interest, scores derived from the feeding scale were not correlated with infant weights at age tested.

These results confirmed all three hypotheses tested. Thus, early breast-feeding predicted infant length at later ages. Moreover, infant size predicted later feeding practices, as shown by our finding that infant weight predicted greater feeding intensity at later ages. These findings were significant even after controlling for environmental factors and other background variables, confirming the third hypothesis.

	DISCUSSION

Abstract Introduction Methods Results Discussion References

In this study, we followed the growth and development of healthy Barbadian infants over the first 6 mo of life. Regardless of the type of feeding, the infants did not show any evidence of malnutrition or growth faltering at any age. Thus, average weights of the infants corresponded to or exceeded the 50th percentile for U.S. children at 7 wk and at 3 and 6 mo. The apparent advantage of this population at 3 and 6 mo may be an artifact of the modeling used to create the standardized growth charts. Thus, the growth pattern of infants in this study was similar to that of infants in other countries (WHO 1995). Average lengths were also comparable to U.S. reference lengths at all ages studied (NCHS 1977). There are currently no norms available for weights and lengths of Caribbean children; clearly, these are needed.

Several background variables, including maternal anthropometry, reproductive history and environmental conditions, were significantly associated with infant outcomes. Maternal anthropometry (heights and weights) was positively correlated with infant weights after birth, when birth weight was corrected. These relationships were similar to those reported in other studies of infant growth (Adair et al. 1993, Diaz et al. 1995, Pérez-Escamilla et al. 1995). In contrast, maternal age was positively correlated with changes in infant length over the first 6 mo of life, after correcting for birth size. Older mothers generally had longer babies at all ages studied. The association between maternal reproductive history and infant lengths has not been extensively documented (Delgado et al. 1982).

Among the various background factors, one of the home environmental conditions was significantly related to changes in infant lengths after birth. Information seeking (Factor 9) was the most striking in this regard, referring to the important role of maternal reading habits and exposure to a variety of sources of information, including reliance upon the advice of professionals. Of interest, we did not find any association between family income and socioeconomic status (macroenvironmental variables) and infant outcomes in this population, consistent with our previous studies in Barbados on the growth of school-aged children (Galler and Ramsey 1985, Galler 1987). However, these macroenvironmental variables were significantly correlated with feeding practices in this study, and may thereby have affected infant outcomes indirectly. Our findings were consistent with those reported in a prospective study of healthy, full-term infants in Montreal (Kramer et al. 1985), in which socioeconomic conditions per se were found to have little direct effect on infant growth. Thus, the lack of association between macroenvironmental conditions and infant growth in our study may also be attributed to the selection of healthy, lower middle class children. An additional possibility was that the quality and uniformity of health care in Barbados may have mitigated any adverse effects of the home environment. The health care system in Barbados is comprehensive, especially for children <5 y of age, who are required to have regular medical visits to qualify for school entry. Consequently, adequate growth may have occurred despite variations in macroenvironmental conditions. The important role of health care as a path between environmental conditions and growth has been discussed by Martorell and Habicht (1986).

Despite the independent relationships between the background variables and infant size, our study provided evidence supporting predictive associations between early feeding practices and later infant growth, and vice versa, even after controlling for environmental conditions, maternal anthropometry and maternal age. Thus, all three hypotheses proposed in designing this study were confirmed. First, we showed that the early preference for breast-feeding and associated variables in Factor 1 were found to predict infant lengths at later ages. Conversely, earlier infant weights predicted later maternal reports of increased feeding intensity. The positive association between breast-feeding and infant growth parallels other reports in the literature. Predictive contributions of infant characteristics to later feeding practices were previously reported by Adair et al. (1993). These authors demonstrated that greater infant weight (ponderal index) predicted a longer duration of breast-feeding in their longitudinal study of Filipino infants. However, in our study, we found that infant weights primarily predicted feeding intensity but not necessarily the preference to breast-feed at later ages. The Adair study followed infants to 24 mo and was specifically designed to assess factors contributing to the duration of breast-feeding. As we postulated in our earlier review article (Galler et al. 1984), the current study confirmed that the relationship between feeding practices and infant outcomes was reciprocal. Our findings were therefore consistent with those from the study of low income Peruvian mothers and their babies by Piwoz et al. (1994) who demonstrated the presence of a bidirectional relationship between feeding practices and infant growth. They also reported that poor infant weight gain predicted an increased number of changes in feeding practices. Unlike the aforementioned studies, mothers and babies in our study were not from low income homes and did not show any evidence of malnutrition. Thus, the predictive relationships found by us and the reciprocity between the feeding practices and infant outcomes were not restricted to a specific setting or income group.

Although the frequency of breast-feeding among the mothers was the major factor predicting infant growth, other variables also played a significant role. In this study, the application of factor analysis to a range of feeding practices allowed us to group related items into factors that were statistically reliable, independent and theoretically meaningful. It also allowed us to compare these groups of items at three ages. Factor 1, the preference for breast-feeding, produced more striking results than those obtained by use of the feeding scale, which included only a unidimensional measure of number of breast- vs. bottle-feedings per day. The factor included the quantitative measure of breast-feeding vs. bottle-feeding; it also measured the level of satisfaction shown by the baby in association with breast-feeding, the maternal perception that her milk supply was adequate for breast-feeding and the limitation of extenders and other foods. The improved infant outcomes predicted by Factor 1 were also associated with these conditions, which may have important policy implications. For example, the maternal perception of having an inadequate milk supply to sustain breast-feeding has been pointed to repeatedly as a reason for the premature termination of breast-feeding and should be an important subject for future intervention and education.

The significant role of feeding intensity (Factor 3) in relationship to infant growth may be a reflection of the health and well-being of the infants. However, it may also represent differences in maternal perceptions of their infants' behaviors, which may be viewed as more advantageous in the case of heavier babies. These relationships remain to be studied further. In summary, this study suggests that comprehensive programs, which also take into account the health and behavior of both mothers and infants pre- and postnatally, and their interrelationships over time, will be most likely to succeed.

There were several limitations to this research. First, the study was not designed as an intervention. Therefore, relationships reported here are predictive rather than causal. However, the longitudinal design of the study permitted testing of possible causal pathways, including the relationship between early breast-feeding and later infant outcomes, which could not be explored with more conventional cross-sectional methods. Second, we cannot eliminate the possibility of selection bias, although missing data did not exceed that reported in other longitudinal studies of feeding practices (Fleming et al. 1988, Hundt and Forman 1993). Our study differed from some of these in that mothers and babies were evaluated at more frequent intervals, and more measures were applied, thus increasing the risk of missing data. Thus, the final number of women and infants used to test all three hypotheses was approximately half of the original sample. Even though there were no differences in the characteristics of mother-infant pairs with and without complete data sets, it was possible that they differed in ways we did not measure. Because 92% of the sample was tested at the 6-mo visit, we were able to confirm that cross-sectional analyses based on the larger sample did not differ from analyses based on the restricted sample. The high percentage of returns for the final visit confirms that women who missed earlier visits continued to be cooperative. A replication study in another setting may be warranted, nevertheless, to confirm the validity of the findings and their applicability to other populations.

Third, a multitude of correlations were generated used overall (canonical) correlations, supplemented by protected multiple and simple correlations. To avoid reporting spurious findings, we considered simple relationships as significant only when the overall relationships were found to be significant. The significance of the overall correlations between the sets of variables also indicated that these findings were not random. A fourth concern was that the reliability of the birth measures, which were derived directly from the hospital record, could not be easily tested. To adjust subsequent measurements by the birth measures, we therefore elected to use residual scores, rather than delta scores, specifically because the former were not adversely altered by the less reliable birth measures.

These limitations are not unique to this study, but are characteristic of most longitudinal field investigations. Nevertheless, this type of research is best capable of addressing the complex relationships between feeding and growth and their consequences for future public policy.

	FOOTNOTES

Manuscript received 9 June 1997. Initial reviews completed 22 July 1997. Revision accepted 14 April 1998.

	ACKNOWLEDGMENTS

Esther Archer, MRCP, who helped to design and conduct this study, passed away in May 1991. Her enthusiasm and dedication will always be remembered. The authors also thank Jean Ramsey, Marjorie Bowen and Victor Forde for their participation in the field study and Samantha Butler and Antonia Biggs for their assistance with data analysis.

	LITERATURE CITED

Abstract Introduction Methods Results Discussion References

• Adair L., Popkin B. M., VanDerslice J., Akin J., Guilkey D., Black R., Briscoe J., Flieger W. Growth dynamics during the first two years of life: a prospective study in the Philippines. Eur. J. Clin. Nutr. 1993; 47:42-51[Medline]
• Armor, D. (1973) Theta reliability and factor scaling. In: Sociological Methodology (Castner, H., ed.), pp. 17-50. Jossey-Bass, San Francisco, CA.
• Castillo C., Atalah E., Riumallo J., Castro R. Breastfeeding and the nutritional status of nursing children in Chile. Bull. PAHO 1996; 30:125-133
• Cohen, J. & Cohen, P. (1983) Applied Multiple Regression/Correlation Analysis for the Behavioral Sciences, 2nd ed. Erlbaum Associates, Hillsdale, NJ.
• Cureton, E. E. & D'Agostino, R. B. (1983) Factor Analysis: An Applied Approach. Erlbaum Associates, Hillsdale, NJ.
• Delgado H. L., Valverde V. E., Martorell R., Klein R. E. Relationship of maternal and infant nutrition to infant growth. Early Hum. Dev. 1982; 6:273-286[Medline]
• Díaz S., Herreros C., Aravena R., Casado M. E., Reyes M. V., Schiappacasse V. Breast feeding duration and growth of fully breast-fed infants in a poor urban Chilean population. Am. J. Clin. Nutr. 1995; 62:371-376[Abstract/Free Full Text]
• Fleming A. S., Ruble D. N., Flett G. L., Shaul D. L. Postpartum adjustment in first time mothers: relations between mood, maternal attitudes, and mother infant interactions. Dev. Psych. 1988; 24:71-81
• Galler, J. R. (1987) The interaction of nutrition and environment in behavioral development. In: Early Nutrition and Later Achievement (Dobbing, J., ed.), pp. 175-207. Academic Press, New York, NY.
• Galler, J. R., Harrison, R. H., Ramsey, F. C., Biggs, M. A. & Forde, V. (1998) Maternal moods predict breast-feeding in Barbados. J. Dev. Behav. Pediatr. (in press).
• Galler J. R., Ramsey F. The influence of early malnutrition on subsequent behavioral development: VI. The role of the microenvironment of the household. Nutr. Behav. 1985; 2:161-173
• Galler, J. R., Ricciuti, H. N., Crawford, M. A. & Kucharski, L. T. (1984) The role of the mother-infant interaction in nutritional disorders. In: Nutrition and Behavior (Galler, J. R., ed.), pp. 269-304. Plenum Press, New York, NY.
• Harris, R. A. (1974) A Primer of Multivariate Statistics. Academic Press, New York, NY.
• Horwood, L. J. and Ferguson, D. M. (1998) Breastfeeding and later cognitive and academic outcomes. Pediatrics 101: url:http://www. Pediatrics. Org/cgi/content/full/101/1/e9.
• Hundt G. A., Forman M. R. Interfacing anthropology and epidemiology: The Bedouin Arab infant feeding study. Soc. Sci. Med. 1993; 36:957-964
• Kendall, M. G., Stuart, A. & Ord, J. K. (1983) The Advanced Theory of Statistics, Vol. 3, 4th ed. MacMillan Publishing Company, New York, NY.
• Keppel, G. (1973) Design and Analysis: A Researcher's Handbook, p. 513. Prentice Hall, Englewood Cliffs, NJ.
• Kramer M. S., Barr. R. G., Leduc D. G., Boigjoly C., McVey-White L., Pless I. B. Determinants of weight and adiposity in the first year of life. J. Pediatr. 1985; 106:10-14[Medline]
• Lawley D. N. Tests of significance in canonical analysis. J. Biometrika 1959; 46:59-66
• Martorell, R. & Habicht, J. P. (1986) Growth in early childhood in developing countries. In: Human Growth: A Comprehensive Treatise (Faulkner, F. & Tanner, J. M., eds.), pp. 241-262. Plenum Press, New York, NY.
• Miller, R. G., Jr. (1981) Simultaneous Statistical Inference. Springer-Verlag, New York, NY.
• National Center for Health Statistics (1977) NCHS Growth Curves for Children, Birth to 18 Years, United States. National Health Survey, No. 165, DHEW Publication No. PS 78-165, Rockville, MD.
• Pérez-Escamilla R., Cohen R. J., Brown K. H., Rivera L. L., Canahuati J., Dewey K. G. Maternal anthropometric status and lactation performance in a low-income Honduran population: evidence for the role of infants. Am. J. Clin. Nutr. 1995; 61:528-534[Abstract/Free Full Text]
• Piwoz E. G., Black R. E., Lopez de Romaña G., Creed de Kanashiro H., Brown K. The relationship between infants' preceding appetite, illness, and growth performance and mothers' subsequent feeding practice decisions. Soc. Sci. Med. 1994; 39:851-860
• Ramsey, F., Demak & Trotter, P. (1984) National Health and Nutrition Survey of Barbados, 1981. Pan American Health Organization, Washington, DC.
• Rao S., Kanade A. N. Prolonged breast-feeding and malnutrition among rural Indian children below 3 years of age. Eur. J. Clin. Nutr. 1991; 46:187-195
• Rogan W. J., Gladen B. C. Breast feeding and cognitive development. Early Hum. Dev. 1993; 31:181-193[Medline]
• SAS Institute Inc. (1990) SAS/STAT User's Guide, Version 6, 4th ed, vols. 1 & 2. SAS Institute, Cary, NC.
• Seward, J. F. & Serdula, M. K. (1984) Infant feeding and infant growth. Pediatrics 74 (suppl.): 728-762.
• Sullivan, K. V. & Gorstein, J. (1990) Anthropometry Software for Calculating Pediatric Anthropometry (Version 1.01), U.S. Department of Health Services, Centers for Disease Control, Atlanta, GA and Nutrition Unit, World Health Organization, Geneva.
• UNICEF (1980) The State of the World's Children. United Nations, New York, NY.
• UNICEF (1993) The State of the World's Children. United Nations, New York, NY.
• VanDerslice J., Popkin B., Briscoe J. Drinking-water quality, sanitation, and breast-feeding: their interactive effects on infant health. Bull. WHO 1994; 72:589-601[Medline]
• WHO Working Group on Infant Growth An evaluation on infant growth: the use and interpretation of anthropometry in infants. Bull. WHO 1995; 73:165-174[Medline]
• Winikoff B. Issues in the design of breast-feeding research. Stud. Fam. Plann. 1981; 14:177-184

0022-3166/98 $3.00 ©1998 American Society for Nutritional Sciences