Community and International nutrition

Low Birth Weight Reduces the Likelihood of Breast-Feeding among Filipino Infants1'2
LINDA S. ADAIR3 AND BARRY M. POPKIH
Department of Nutrition, University of North Carolina at Chapel Hill, Chapel Hill, NC 27516 environment. How they are fed represents an impor tant determinant of their survival. An extensive and often controversial literature ex ists concerning the feeding of LBW infants in devel oped countries, but it deals almost exclusively with preterm infants or those weighing < 1500 g (American Academy of Pediatrics 1985, Churella et al. 1985, Steichen et al. 1987, Verronen 1985). In developing countries, very little is known about how LBW infants are fed. Furthermore, the relevant issues are very dif ferent (Narayanan 1986). First, in lower income coun tries, early preterm infants have little chance of sur vival in the absence of special care facilities. Second, in contrast to developed countries where most LBW is accounted for by prematurity, the majority of LBW infants in developing countries are mature, but smallfor-gestational-age (Villar and Belizan 1982). Many, particularly those who weigh >2000 g, require no spe cial care (Narayanan 1986). The question of how LBW infants in developing countries are fed is an important one. Some research has suggested that LBW infants are less likely to be breast-fed at all or are breast-fed for shorter periods of time (Barros et al. 1986, Butz and DaVanzo 1981, World Health Organization 1981). Given its proven beneficial effects in reducing morbidity and mortality, breast-feeding is of special importance to infants al ready at risk because of LBW (Popkin et al. 1986). The small amount of available data on feeding of LBW infants in developing countries is based primarily
' Funding for parts of the program design, data collection, and computerization was provided by the National Institutes of Health (NIH) (grants ROI-HD 19983A, ROl-HD 18880 and ROI-HO23137). Data analysis was supported by a grant from Wellstart International. 1 The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 USC section 1734 solely to indicate this fact. 3 To whom correspondence should be addressed at Carolina Pop

ABSTRACT We studied the relationship of low birth weight (LBW) to concordance of mother's feeding in tentions during pregnancy with actual feeding practices; initiation of breast-feeding; and patterns of feeding in the first 6 mo. Data came from the Ceb Longitudinal Health and Nutrition Survey, which followed 3080 mother-infant pairs from urban and rural communities of Metro Ceb,Philippines. We used logistic regression to estimate the effects of LBW on feeding practices, controlling for place of delivery (home, public or private health facility), receipt of free infant formula samples; infant sex, urban residence; primiparity, education level and age of the mother; and family income and assets. Birth of a LBW infant significantly decreased the like lihood that women would initiate breast-feeding. Of particular note is the finding of this decreased likelihood among women who during pregnancy had stated an intention to breast-feed. In a comparison of 6-rno feeding patterns, we also found that LBW increased the likeli hood of not breast-feeding or of weaning before 6 mo. Among breast-feeding mothers, LBW increased the likelihood of full breast-feeding for 6 mo compared with patterns characterized by earlier supplementation with other foods and liquids. The negative relationship of LBW to breast-feeding was strongest when births took place in private or public health facilities. Given the known health risks of LBW and the proven benefits of breast-feeding, these results emphasize the need for special efforts to promote breast-feeding of LBW infants born in clinical settings. J. Nutr. 126: 103-112,1996. INDEXING KEY WORDS:

Downloaded from jn.nutrition.org by guest on August 1, 2013

breast-feeding infant feeding low birth, weight human milk

Low birth weight (LBW)4 continues

to be a problem

of particular concern for developing countries, where the estimated overall prevalence of LBW is 19% (World Health Organization 1992), and in some countries, up to 50% of infants are born weighing <2500 g (World Health Organization 1984). As a group, LBW infants are less able to cope with the stresses of the postnatal
0022-3166/96 $3.00 1996 American Institute of Nutrition. Manuscript received 30 March 1995. Initial review completed

ulation Center, University of North Carolina, University Square, CB #8120, 123 W. Franklin St., Chapel Hill, NC 27516-3997. 4 Abbreviations: CI = confidence interval; LBW = low birth weight; RRR = relative risk ratio.

16 June 1995. Revision accepted 25 August 1995.

103

104

ADAIR AND POPKIN

on retrospective data and is mostly descriptive, usually presenting only the prevalence of breast-feeding among infants of different weights at birth. Further more, there is no consistency in results from different populations. For example, the WHO Collaborative Study on Breast-Feeding (World Health Organization 1981) found no relationship between birth weight and the prevalence of breast-feeding in Ethiopia, Nigeria, Zaire, Guatemala and India, nor was a relationship found in Machakos, Kenya (Renquist et al. 1985). In contrast, among urban poor and rural subjects in Chile, a lower percentage of LBW infants were breast-fed at 3 mo of age compared with infants whose weights ex ceeded 2500 g at birth (World Health Organization 1981). In an urban Brazil population, infants weighing <2000 g at birth were less likely to have been breast fed in the first week of life and throughout the entire first year (Barros et al. 1986). None of these studies differentiated immature LBW infants from those with intrauterine growth retardation. To date, there have been no systematic studies of how birth weight interacts with other factors known to influence breast-feeding. Physical conditions that influence the infant's ability to suck are likely to pro hibit or complicate the initiation of breast-feeding or affect its duration. Before 33 wk of gestation, the in fant's sucking reflexes are ineffective, and between 34 and 36 wk may be inefficient or poorly coordinated. Even in mature infants, there is a relationship between sucking behavior and birth weight (Pollitt et al. 1978). Mothers may also differ in their responses to an infant who does not suck well. Their perceptions of the size and capabilities of their infants and their reactions to the stimuli they receive from their infants may influ ence their decisions about breast-feeding. Thus, factors such as LBW may alter feeding choices, even among mothers who were strongly inclined to breast-feed their infants. LBW infants born in hospitals may be subject to policies that single them out for special care (Popkin et al. 1984). When this occurs, mothers and infants may be separated for prolonged periods of time. This delayed contact may impair or prevent the establish ment of breast-feeding (de Chateau and Wiberg 1977, Salariya et al. 1978). Thus, different patterns of feeding LBW infants may be established on the basis of the place of delivery. Finally, numerous studies have shown that birth weight and breast-feeding are both associated with socioeconomic status and education of the mother (Adair et al. 1993, Forman 1984, Popkin et al. 1983). The confounding effects of variables such as these must be sorted out to determine the relationship between birth weight and patterns of breast-feeding. This study examined feeding of LBW infants in the Philippines. We used data from a community-based survey that included infants born at home as well as in public or private health facilities. Our major objec

tives included the following: 1} to compare feeding patterns among LBWvs. normal weight and term LBW vs. preterm LBW infants, with special attention to the concordance of feeding intentions and feeding prac tice, initiation of breast-feeding and patterns of feed ing in the first 6 mo of life; 2) to determine the extent to which place of delivery (home or hospital) influ enced the feeding of LBW and preterm infants and 3) to assess, in a multivariate context, the relative im portance of LBW compared with other biological, sociodemographic and economic factors as a determinant of feeding patterns.

MATERIALS AND METHODS Sample. Data were collected during the CebLon gitudinal Health and Nutrition Survey. This is a pro spective, community-based survey that followed women from midpregnancy to 24 mo postpartum. The initial sample included all pregnant women in 33 ran domly selected barangays (smallest administrative units) of Metro Ceb.Metro Cebis one of the most rapidly growing and developing areas of the Philip pines, with a current population of ~ 1.9 million. The Cebsample is drawn from an ecologically and socioeconomically diverse population. About one-quarter of participants lived in rural communities, some of which were in isolated mountainous areas. The re mainder were residents of urban or peri-urban com munities with varying levels of modernization but fairly high exposure to modern media. All data were collected using procedures approved by the University of North Carolina School of Public Health Institu tional Review Board for research involving human subjects. Our target analysis sample consisted of 3,080 single live births occurring in a 1-y time period (1983-1984). Acceptable birth weight data were available for 3022 infants, and gestational age was known in 2891 of these infants. Among infants with birth weight and gestational age data, 334 (11.55%) were LBW, and of LBW infants, 93 (27.8%) were preterm (completed gestational age of <37 wk, range was 25-36 weeks). Infants were assessed at birth and at bimonthly inter vals until they reached 24 months of age. Data. During the pregnancy baseline survey, ex tensive background socioeconomic and demographic data were collected. Within 3-6 d of birth, a birth in formation survey was conducted to identify the place of delivery, birth attendant, type of delivery, compli cations of labor and delivery and timing of first mother-infant contact. Infant weight was initially measured by birth attendants who were provided with project scales and instructed in their use. Project in terviewers subsequently weighed infants during the birth information survey. All scales were regularly

Downloaded from jn.nutrition.org by guest on August 1, 2013

FEEDING

OF LOW BIRTH WEIGHT

INFANTS

105

checked and calibrated. The Ballard method was used to assess gestational age clinically of all LBW infants, all infants whose mothers had complications such as bleeding during pregnancy and all infants whose mothers were uncertain about the date of their last menstrual period (Ballard et al. 1979). When both Bal lard and last menstrual period estimates of gestational age were available, we use the Ballard estimates as the most accurate representation. Previous analyses have shown a high level of agreement between these two measures in the Cebsample. Infant-feeding data were collected during the birth information survey and subsequent bimonthly inter views with mothers. At each interview, mothers were asked whether they breast-fed and to recall all foods and liquids fed to the infant in the past 24 h. For the same period, detailed information was collected on the types, quantities and method of preparation of all foods and liquids given to infants. Mothers were also asked to recall general feeding patterns (exclusive breast-feeding, feeding of nonnutritive liquids, nutri tive liquids, solids and semisolid foods) 7 d before each interview. Using a health facilities survey administered to hospital and clinic personnel, we gathered information on facility practices related to infant feeding. Person nel were asked questions about receipt and distribu tion of infant formula, rooming-in policies, breast feeding education, policies regarding feeding of water and formula, etc. Analysis methods. Although Ceb data were available for the first 24 mo of life, we focused our work on the first 6 mo because this is the time period during which LBW is likely to influence infant-feeding patterns, and full breast-feeding is rare in this popu lation after 6 mo. Furthermore, in preliminary anal yses, we found no consistent effects of LBW on feeding patterns after 6 mo. We defined three feeding outcomes for analysis. The analysis samples for the three outcomes differ in size because the feeding pattern analysis requires that in fants have complete data for the first 6 mo and because of missing data critical to the definition of the out come. 1. Concordance of breast-feeding intentions and practices (n = 2955). One major way to understand how biological and other factors in the perinatal period affect feeding behaviors is to explore the role of breast feeding intentions (see Stewart et al. 1991) and their concordance with actual feeding practices. During the baseline survey, mothers were asked how they planned to feed their infant. We compared their intentions with actual feeding practices and defined the following four groups: 1}intended to breast-feed and did breast-fed, n = 2498 (84.5%); 2) intended to breast-feed but did not breast-feed, n = 85 (2.9%); 3) did not intend to breast-feed but did breast-feed, n = 260 (8.8%) and 4)

did not intend to breast-feed and did not breast-feed, n = 112(3.8%). Of particular interest is the second group, because condition of the infant at birth or other barriers to breast-feeding encountered at delivery may cause the mother to change her mind about feeding. 2. Initiation of breast-feeding (n = 2991). A dichotomous variable indicates whether or not the mother ever initiated breast-feeding, either in the first several days after birth or subsequently. Studies of prelacteal feeding patterns in the Cebsample (Fer nandez and Popkin 1988) showed that women may discard colostrum and not initiate breast-feeding for several days. Only 6.75% of mothers in this sample never initiated breast-feeding. 3. Patterns of feeding in the first 6 mo of life (n = 2757). Infants were classified in groups on the basis of overall feeding patterns during the first 6 mo. We first analyzed results from six groups in which infants who never breast-fed were differentiated from those who were weaned early, and those who fully breast fed for only 2 mo were in a different group than those fully breast-fed for 4 mo. We found no differences in the determinants of never breast-feeding vs. early weaning, nor did we find differences in predictors of 2 vs. 4 mo of full breast-feeding. Thus although there may be important policy reasons to identify these groups separately for studies of infant growth and morbidity, none of the variables of interest in the present analysis significantly differentiated the groups. These results provided the rationale for a simplifica tion to only four groups as follows: ) fully breast-fed for at least 6 mo, n = 388 (14.1%); 2) fully breast-fed for more than 2 but fewer than 6 mo and then mixed fed, n = 1321 (47.9%); 3) mixed fed throughout, n = 307 (11.1%) and 4) not breast-fed or weaned early, n = 741 (26.9%). Infants were considered fully breast-fed if, on the basis of the 24-h food recall, they received <83 kj (20 kcal) from supplemental foods. In most cases, such infants were receiving tastes of weaning foods or sweetened liquids such as herbal teas. For this analysis, we focused on the dominant feeding pattern and ig nored the frequent switching of feeding methods typ ical in this sample (Zohoori et al. 1993). For example, infants reported to be fully breast-fed at 2 and 6 mo, but who received >83 kj/d from supplemental foods or liquids at 4 mo, were still considered to be fully breast-fed for 6 mo. Infants were considered to have been weaned early if they initiated breast-feeding but were totally weaned from the breast before 6 mo of age. Infants were considered not breast-fed if they never initiated breast-feeding. We used multinomial logistic regression models to predict membership in the different groups (categories of each outcome variable). A review of the estimated coefficients allowed an assessment of how each inde pendent variable in the model affected the likelihood

Downloaded from jn.nutrition.org by guest on August 1, 2013

106

ADAIR AND POPKIN

of membership in each group. Coefficients were ex ponentiated to calculate relative risk ratios (RRR) for a 1-unit change in each independent variable. Risk was measured for a comparison of each category to a ref erence category. All possible intergroup comparisons were made and used to calculate the predicted prob ability of membership in each feeding category. For each model, our primary independent variable of in terest was LEW. We also compared LEW term with LEW preterm births. Because infant birth weight and mothers' decisions about where to deliver and how to feed the infant are highly interrelated, we had to take several potentially important statistical problems into consideration. We controlled for the effects of confounding by including place of delivery (home, public or private facility); re ceipt of a free sample of infant formula at delivery; sex of the infant, place of residence (urban or rural); primiparity, education level and age of the mother; and household income and assets. However, inclusion of these variables does not eliminate potential biases associated with endogeneity. Variables are considered endogenous to a model when they are jointly deter mined by a set of common, underlying, unobservable variables. Consider the case where a mother's innate healthiness affects both the likelihood that she will breast-feed her infant and whether the infant is LBW. In a multiple regression model, error terms associated with estimates of feeding and of LBW will be corre lated. If we ignore these relationships, we may mis takenly attribute part of the variation in feeding prac tices to LBW, when in fact, it should be attributed to the unmeasured maternal health status. (Seeadditional examples of endogeneity bias in the article by Briscoe et al. 1990.) Note that if the variables are observable and measurable, then they are potential confounders and can be included in the model as control variables. We hypothesized that LBW and place of delivery vari ables would be endogenous to our feeding models. Economists have developed approaches to correct for endogeneity of explanatory variables, which in volve the use of instrumental variables (see Briscoe et al. 1990, CebStudy Team 1991 and Maddala 1988 for further information and another example of use of the instrumental variables approach). An instru ment or variable predicted from a set of strictly ex ogenous factors is substituted in the model for the actual value of the variable. We tested the assumption of endogeneity of the LBW and place of delivery vari ables using the Hausman test (Hausman 1978), which requires that both the instrument and actual values of the variable be included in the model. When the in strument has a statistically significant coefficient, it is considered endogenous to the model. Using this method, we found place of delivery but not LBW to be highly endogenous in all of our models. This in dicates that unmeasured factors such as the innate healthiness of the mother affect the delivery decision

and the likelihood that the infant is LBW. Accordingly, all models were specified with place of delivery in struments, but actual values for all other variables. RRR and 95% confidence intervals (CI) were cal culated for each independent variable in the models. Then, to assist in the interpretation of results, we pre dicted the likelihood of membership in each of the groups based on mean values of the independent vari ables. Finally, we simulated the effects of specified conditions, principally LBW vs. normal birth weight and place of delivery.

RESULTS Descriptive results. Table 1 presents character istics of sample women in strata defined by place of delivery (at home, in public or private hospitals or clinics). The decision of where to give birth is an im portant maternal choice, conditioned by a wide range of sociodemographic factors. Women who gave birth at home were significantly more likely than mothers delivering in health facilities to be from rural com munities, have less education and lower household in come and assets. In addition, higher parity women were more likely to deliver at home. Exposure to for mula industry practices varied considerably by place of delivery. Mothers delivering in private hospitals were most likely to receive free formula samples and had significantly longer hospital stays. In other work (Adair et al. 1993), we showed that such practices de creased the duration of breast-feeding. Finally, delayed first contact between mother and infant after birth oc curs more frequently in health facilities, especially private facilities that are more likely to have newborn nurseries. Table 1 also shows differences in the prev alence of LBW among sample infants born in different settings. A higher percentage of LBW infants were born in public facilities, which tended to serve pop ulations at the highest risk for LBW (e.g., primiparae, younger women, women with complications of preg nancy and women from communities with higher dis ease prevalence). Table 2 shows comparisons of feeding practices in LBW vs. normal weight infants by place of delivery. Within each facility, we tested for significant differ ences in the distribution of feeding practice by birth weight status. Within a facility category, there was a consistent trend of lower percentages of LBW com pared with normal weight infants breast-feeding at each point in time. These differences, although similar across the three delivery settings, were most often sta tistically significant in public facilities. Within each place of delivery, there was a trend toward increased prevalence of full breast-feeding in LBW infants at all ages. For infants born in public facilities, there was a highly significant difference in the prevalence of full

Downloaded from jn.nutrition.org by guest on August 1, 2013

FEEDING

OF LOW BIRTH WEIGHT

INFANTS

107

TABLE 1 Characteristics of women delivering infants in different settings1'*

Place of delivery characteristicsAge, Mother's y3Education, 1843)26.1 (n = 6.16.0 2.865.42. facility 545)25.3 (n = 5.77.7 2.893.21.96 facility 634)26.6 (n = 5.710.0 3.394.21.75 1.9828.214.668.53.04 +

completed^Urban, y
%cParity0Primiparous, %cHeld %bReceived baby in first 4 h, %Duration a sample of infant formula, d~LBW, of hospital stay, %aPrterai, % |<37 wk)dHome

52 2.3217.274.54.2na311.013.3Public 1.9323.832.226.92.52

1.7215.315.7Private

0.969.912.2

1 Values are means so. 2 Statistical comparisons based on ANOVA for continuous variables or chi-squared for categorical variables, P < 0.05: a public differs significantly from home and private; " all intergroup differences are significant; c home differs significantly from public and private; no significant differences. LBW = low birth weight; na = not applicable.

Downloaded from jn.nutrition.org by guest on August 1, 2013

breast-feeding of LBW vs. normal weight infants at 6 mo. The lowest overall prevalence of breast-feeding occurred in private facilities, where mothers are typ ically more educated and come from higher income households. The basis for these associations can be best understood after multivariate analyses that ac count for other factors known to influence feeding. Concordance of feeding intentions and prac tices. The reference group for the multinomial logis tic regression was women who intended to breast-feed and did breast-feed. LBW dramatically increased the likelihood of not breast-feeding among women who intended to do so (RRR = 3.13, P < 1CT4,95% CI 1.755.60). This effect of LBW was slightly stronger for LBW preterm (RRR = 3.37, P = 0.002, 95% CI 1.23-9.28) compared with LBW term infants (RRR = 2.86, P = 0.019, 95% CI 1.45-5.64). Other factors signifi cantly associated with not breast-feeding in those mothers who had intended to do so included delivery in a private hospital, receipt of a sample of formula after delivery and primiparity. For a clearer interpretation of results, we predicted the probability of membership in each of the groups under specified conditions. Results based on multi variate analyses are presented in Table 3. The overall probability of intending to, but not breast-feeding a normal birth weight infant was 2.2%, whereas for a LBW infant, this probability was increased to 6.3% (7.1% for LBW preterm infants and 6.0% for LBW term infants). The effect of LBW appeared most dra matic in private facilities, where the predicted prob ability of not breast-feeding when mothers intended to do so was 16.7% for LBW vs. 6.1% for normal birth weight infants.

Initiation of breast-feeding. The initiation of breast-feeding model had the same specification as the intention-practice model. The full set of results is pre sented in Table 4. LBW increased by 67% the chances of never breast-feeding. Comparison of term and pre term LBW infants showed a significant effect of LBW term delivery, but not of LBW preterm delivery. Other significant predictors of never breast-feeding included delivery in a private facility, maternal education, ma ternal age and household income and assets. Predicted probability of never breast-feeding based on the mul tivariate models is presented in Table 5. Note that within each facility category, LBW decreased the probability that a mother would initiate breast-feed ing. Furthermore, LBW term infants were less likely to be breast-fed than LBW preterm infants. Infant feeding patterns in the first 6 mo. Multinomial logistic regression models allowed for the identification of significant predictors of mem bership in each of the four feeding pattern groups. LBW significantly incrreased the following: 1} the likelihood of not breast-feeding or of early weaning compared with mixed feeding for 6 mo (RRR = 2.22, P = 0.003, 95% CI = 1.30-3.79); 2} the likelihood of not breast-feeding or early weaning compared with full breast-feeding for at least 2 but <6 mo RRR = 1.49, P = 0.012, 95% CI= 1.09-2.03); 3) the likeli hood of full breast-feeding for 6 mo compared with mixed feeding for 6 mo (RRR = 2.02, P = 0.016, 95% CI = 1.09-2.03). Predicted probability of each feeding pattern by place of delivery for normal vs. LBW infants, based on the multivariate model, is presented in Table 6. The effects of LBW can be seen by comparing the LBW

108

ADAIR AND POPKIN

TABLE 2 Comparison of feeding patterns among low birth weight {LBWJand normal weight infants Place of delivery Home NBW (n= 1641)LEW (n = 202)P1NBW Feeding patterns Ever breast-fed, % Breast-fed at 2 mo, % Breast-fed at 4 mo, % Breast-fed at 6 mo, % Breast-fed at 12 mo, % Fully breast-fed at 2 mo, % Fully breast-fed at 4 mo, % Fully breast-fed at 6 mo, % Public facility |n = 462)LEW(n = 83)P1NBW Private facility (NBW) in different delivery settings

[n-571]LEW |n = 63)P1
85.4 66.2 58.1 51.7 37.1 34.2 23.3 4.7 79.3 64.2 61.4 50.0 33.3 37.7 22.5 4.6 Private facility ns ns ns ns ns ns ns
Downloaded from jn.nutrition.org by guest on August 1, 2013

96.6 90.5 87.4 85.0 70.8 71.8 58.3 18.1

93.0 90.6 84.9 82.8 68.3 74.6 60.3 19.0 Home NBW

0.01 ns ns ns ns ns ns ns

93.2 83.9 77.4 73.1 60.0 60.8 43.5 10.6

90.0 75.7 67.6 58.2 46.0 59.5 45.6 22.4 Public facility

ns2 0.08 0.07 0.01 0.03 ns ns 0.01

ns

LEW

NBW

LEW

NEW

LEW

Distribution of feeding patterns3 No. in group Fully breast-fed for 6 mo, % No. in group Fully breast-fed for > 2 but < 6 mo and then mixed-fed, % No. in group Mixed-fed for 6 mo, % No. in group Never breast-fed or weaned early, % 1 P values for comparison

270 17.7 834

33 18.2 99

45 10.8 205

15 21.4 22

23 4.5 147

3.9 14
27.9 5 9.6 31 59.6

54.7 161 10.6 260 17.1

54.7 11 6.1 38 21.0

49.2 48 11.5 119 28.5

31.4 2
2.9 31 44.3

28.7 80 15.6 262 51.2

of prevalence of age-specific feeding method in LBW versus NEW infants, within each place of delivery are

based on ANOVA. 2 ns = not significant, P > 0.05. 3 Results of chi-squared tests for significant differences in distribution of feeding patterns in LBW vs. NBW infants, by place of delivery are as follows: Home, chi-squared = 4.69, P = 0.196; public facility, chi-squared = 18.75, P < 0.000; private facility chi-squared = 1.88, P = 0.598.

and normal birth weight columns within each place of delivery. The overall pattern was the same across all sites. The predicted probability of never breast feeding or weaning early was higher in LBW compared with normal weight infants. Consistent with the frequencies of the different feeding patterns presented in Table 2, the multivariate analyses also showed that if women elected to breast feed, LBW increased the probability of full breast feeding for 6 mo. This was true for home and public facility deliveries. Among private deliveries; LBW had no impact on the likelihood of full breast-feeding for 6 mo, probably because full breast-feeding is a rela tively rare practice among women delivering in this setting.

The effects of term vs. preterm delivery were ex amined in general because a further breakdown by place of delivery resulted in a very small number of preterm LBW infants in each place. The RRRs for term and preterm LBW deliveries were similar for most intergroup comparisons, but since the standard errors were larger for the estimates of the LBWpreterm effect because of the small number of infants in this category, the RRR for LBW preterm generally did not reach sta tistical significance. These predictions were based on the use of instru mental variables for place of delivery. Using instru ments, we obtained unbiased estimates of the effects but noted that some of coefficients for place of delivery were very large. The resulting predicted probabilities

FEEDING

OF LOW BIRTH WEIGHT

INFANTS

109

TABLE 3

Concordance of breast-feeding intentions and practices among CLHNS women: predicted probabilities of membership in groups are affected by place of delivery and low birth weight1
NBWZ LEW LEW prterai LEW term

GroupIntended, breast-fedIntended, breast-feedDid did not breast-fedDid not intend, breast-feedBy not intend, did not facility: tobreast-feed predicted probability of intending breast-feedingHomePublicPrivate85.02.28.93.80.61.36.182.96.38.31.63.04.016.782.57.18.81.682.86.08.33.1 but not

1 Numbers represent predicted probability (%) of membership 2 NBW = normal birth weight; LEW = low birth weight.

in specified group.

Downloaded from jn.nutrition.org by guest on August 1, 2013

of full breast-feeding in the public facility group were higher than would be expected on the basis of obser vation of actual feeding practices in the sample. To try to understand the basis of this result, we estimated the models using actual values for place of delivery instead of instruments. For home deliveries, the results were very similar. For public facility deliveries, our predicted probability of full breast-feeding for 6 mo was quite close to the expected value. The most im portant result of this comparison relates to the LBW findings. The magnitude and direction of the effects of LBW were the same. That is, regardless of whether we used instrumental variables, a consistent finding was that LBW significantly increased the probability that an infant would never be breast-fed or would be weaned early. Mothers' perceptions of the size of the in fant. Mothers were asked whether they thought their infant was of normal size, big or small at birth. We used this information in an effort to sort out the effects of mothers' perceptions of the infant from biological factors that influence feeding decisions. About 18% of mothers thought their infant was small, and of these, 40% were actually LBW. We looked at the ef fects of mothers' perceptions by substituting small for LBW in the models and by adding small to the model with LBW, to measure the independent effect of mothers' perceptions. We found no significant effects of mothers' perceptions on initiation of breast-feeding. In the breast-feeding intention model, both small and LBW significantly predicted membership in the group who intended to breast-feed but did not breast-feed, suggesting independent effects of infant biology and mothers' perceptions. In the feeding pattern analysis, small was statistically significant, but LBW was not significant when both terms were included in that model. Hospital policies. Within categories of facilities, there was little variation in practices related to infant

feeding. Between facility types there was considerable variation. Private hospitals were most likely to receive and distribute formula samples and to offer all infants other liquids in the first several days of life. Roomingin was more prevalent in public hospitals. All facilities claimed to have personnel instructing mothers on the initiation and maintenance of breast-feeding. How ever, within facilities, we had no information on spe cific treatment of LBW infants. Thus, although we found important facility effects on breast-feeding practices in general, we could not point to specific practices affecting LBW infants in particular. We initially hypothesized that delayed contact with LBW infants would reduce the likelihood of establish ing breast-feeding. We found a significant association of delayed first contact with infants and reduced like lihood of breast-feeding in general but no significant interaction with LBW. Furthermore, there were no significant differences within facilities in the propor tion of LBW vs. normal weight infants who had early contact with their mothers.

DISCUSSION
The set of analyses presented here shows a strong and consistent effect of LBW on infant-feeding prac tices. Giving birth to a LBW infant was a deterrent to breast-feeding, but in cases where mothers still elected to breast-feed, the infant's LBW status increased the likelihood that mothers would fully breast-feed their infant. We also found important effects of place of delivery, showing a larger impact of LBW on feeding decisions of mothers who deliver away from home in private or public health facilities. Our analyses were unable to identify clearly the un derlying reasons for the LBW effect on feeding. How ever, we have some important clues from our com-

110
TABLE 4

ADAIR AND POPKIN

duration of breast-feeding is increased when the infant has a high ponderal index or relative fatness. We in Determinants of the likelihood of never initiating breast terpreted this as an effect of positive feedback to the feeding: resalta from multinomial logistic regression mother. If she perceived that her infant was doing well, she was more likely to continue to breast-feed (Adair RRRLEW |z|0.02 CI1 et al. 1993). This is also relevant to our finding that model1'2 mothers who chose to breast-feed a LBW infant were more likely to breast-feed that infant fully. We know LEW .71-24.09 Public delivery 4.15 72 1. 59 0.11 02..07-2.62 that LBW infants tend to undergo a period of catch .50-11.830.97-10000000-01002000000000,61-173-291-130-014-087-228-051-105-260-3.9212.74909280617508 deliveryFormula Private 5.4410110010.36.83.42.32.52.34.50.460.7511.75464.856.101011.37.84,32300.55010035.44.48.77SE0.3 162413482515154212140.450.664.382.460.0.24130.450.0.00.00251615411215z2. 200000000038 4.1.-1.1.1-2.-2.1,-3.-12,01.25 288022034623474608531584744.501.101.-2.-2.1.-2.-1.781783390341267833P> 0.000.070.220.300.140.020.010.140.000.120.030.400.080.000.070.270.410.170.040.020.210.000.1995% up growth in the first 2 mo of life, particularly if they sampleMale are fully breast-fed and thus well nourished and more infantUrbanPrimiparousLow protected from infectious diseases (Adair 1989). In other analyses of growth of infants with different educationYoung feeding patterns, we found that fully breast-fed infants y)Old (<20 weighed more in the first 4 mo of life. The good growth y)Lowest (>35 performance of fully breast-fed infants may serve to SESMiddle reinforce the mother's breast-feeding behavior. SESLEW LEWpreterm term vs. The second set of clues relates to hospital policies. model^LEW Although we do not have specific information on pol termLEW icies related to possible differential treatment of LBW pretermPublic infants, we can show dramatic differences in general deliveryPrivate policies that relate to feeding by place of delivery, and deliveryFormula sampleMale in turn, significant effects of place of delivery on feed infantUrbanPrimiparousLow ing. Early mother-infant contact is one important as pect of postnatal care that affects the initiation and success of breast-feeding. We found dramatic differ educationYoung ences in the percentage of women who had early con yrs)Old (<20 yrs)Lowest (>35 tact with infants across delivery settings, with private SESMiddle facilities having the lowest prevalence of early contact. SES1.67 We did not find differences in the percentage of LBW infants with early contact, nor did we find a significant 1 LEW = low birth weight; RRR = relative risk ratio (for a 1interaction of LBW with early contact in our multi unit change in the corresponding variable, expressed relative to the initiated breast-feeding category; CI = confidence interval; SES = variate models. Thus lack of early contact was not socioeconomic status based on household income and asset tertiles. shown to be a factor which can explain the decreased 2 LEW model = number of observations = 2976; chi-squared likelihood of breast-feeding among LBW infants in = 152.89, P < IO"4; log likelihood = -627.419; pseudoR2 = 0.1086. particular. ^ LEW term vs. LEW preterm model. Number of observations = 2847; chi-squared = 146.90, P < IO"4; log likelihood = -613.798; A second policy that varied substantially across de pseudoR2 = 0.1069. livery setting is the provision of free samples of infant formula to mothers. Only 4% of mothers who deliv ered at home received a free sample, whereas 27% who delivered in public hospitals and 68% who delivered parison of characteristics of facilities, as well as some of the multivariate findings. First, we found a strong effect of LBW among women who had intended to breast-feed their infant. This strongly suggests that the mother's plans were TABLE 5 altered either by the biological condition of the infant Predicted probability of never initiating breast-feeding among CLHPiS women1 at birth (such as poor infant health or poor sucking ability in preterm infants), by her perceptions of the preterm4.112.918.98.7LEW term4.714.721.39.9 PlacedeliveryHomePublicPrivateAllNEW22.98.312.76.0LEW4.813.019.49.5LBW of ability of the infant to breast-feed or the appropriate ness of breast-feeding for a small infant or by barriers to breast-feeding encountered in the postnatal envi ronment. We have no additional information on the infant's health status at birth, but we explored the possible role of the mother's perceptions and showed that controlling for actual LBW, a mother's perception that her infant was small tended to increase the like 1 Numbers represent predicted probability (%) of membership lihood that she would not breast-feed, even when she in specified group. 2 NEW = normal birth weight; LEW = low birth weight. intended to do so. In earlier work, we showed that the
Downloaded from jn.nutrition.org by guest on August 1, 2013

FEEDING

OF LOW BIRTH WEIGHT

INFANTS

111

TABLE 6 Predicted probability of breast-feeding patterns among CLHNS women1 Place of delivery Public facility LBW NEW LEW NEW Private facility LEW

Home NEW

Feeding pattern Fully breast-fed Fully breast-fed Mixed fed for 6 Never breast-fed

6 mo for >2 but <6 mo and then mixed-fed mo or weaned early

16.3 56.1 13.0 14.6

2,0.3 51.3 8.1 20.3

30.2 47.6 5.0 17.1

38.4 40.3 2.9 21.9

1.5 19.5 12.7 66.3

1.6 15.1 6.6 76.7

NEW

LBW preterm

LBW term
Downloaded from jn.nutrition.org by guest on August 1, 2013

Overall deliveriesFully effects of LEW term vs. LEW preterm moFully breast-fed 6 mixed-fedMixed breast-fed for >2 but <6 mo and then moNever fed for 6 breast-fed or weaned early13.849.111.525.618.242.35.134.315.144.77.332.8 Numbers represent predicted probability (%) of membership 2 NEW = normal birth weight; LEW = low birth weight. in specified group.

in private hospitals received samples. In the present analysis, receipt of a sample significantly increased the likelihood of not initiating breast-feeding or weaning early compared with the feeding patterns involving full breast-feeding. In our previous work (Adair et al. 1993) we also showed that receipt of a sample de creased the duration of breast-feeding. However, we found no significant effect of an interaction of LBW and receipt of formula, suggesting that this policy does not single out LBW infants. An important strength of the Cebstudy is that it is community based and allows for a comparison of home births and those that occur in a wide range of public and private facilities. Furthermore, by control ling for a number of factors that we know influence choice of place of delivery as well as feeding practices, we can be more assured that we have true estimates of the effects of LBW status. The research findings should have important health implications for infants born in less developed coun tries. A lower incidence of breast-feeding among LBW infants is an unfortunate consequence because of the greater health risks they face. The health risks are re lated in part to biological characteristics of the LBW infant but also to poor environmental conditions that contribute to LBW and poor postnatal outcomes. Given the proven nutritional and immunological su periority of breast milk, breast-feeding is of special importance to the LBW infant, or as Narayanan (1986)

has stated, a "passport to life." Efforts must be made to single out and eliminate policies that hinder the establishment of breast-feeding in clinical settings. Such policies would benefit all infants, regardless of birth weight status. Furthermore, special policies are needed to promote breast-feeding of LBW infants. En hanced education and encouragement of mothers of LBW infants should address their concerns about the infant's size and capabilities, as well as their special needs.

ACKNOWLEDGMENT We thank Judith Borja from the Office of Population Studies in Cebfor assistance with data analysis.

LITERATURE CITED
Adair, L. S. (1989) Growth of Filipino infants who differed in body composition at birth. Am. J. Hum. Biol. 1: 673-682. Adair, L. S., Popkin, B. M. & Guilkey, D. K. (1993) The duration of breast-feeding: how is it affected by biological, sociodemographic, health sector and food industry factors? Demography 30: 63-80. American Academy of Pediatrics: Committee on Nutrition (1985) Nutritional needs of low-birth-weight infants. Pediatrics 75: 976986.

112

ADAIR AND POPKIN

Popkin, B., Lasky, T., Litvin, J., Spicer, D. & Yamamoto, M. E. (1986) The Infant Feeding Triad: Mother, Infant and Household. Breach Science Publishers, New York, NY. Popkin, B. M., Yamamoto, M. & Griffin, C. (1984) Traditional and modern health professionals and breast-feeding in the Phil ippines. J. Pediatr. Gastroenterol. Nutr. 3: 765-776. Renquist, U. H., Kusin, J. A. & Jansen, A. A. J. (1985) Feeding pattern, supplementary food and growth of low birth weight and normal birth weight infants in Machakos, Kenya. East Afr. Med. J. 62: 299-308. Salariya, E. M., Easton, P. M. & Cater, J. I. (1978) Duration of breast-feeding after early initiation and frequent feeding. Lancet 2: 1141-1143. Steichen, J. J., Krug-Wispe, S. K. & Tsang, R. C. (1987) Breast feeding the low birth weight preterm infant. Clin. Perinatol. 14: 131-171. Stewart, J. F., Popkin, B. M., Guilkey, D. K., Akin, J. S., Adair, L. S. & Flieger, W. (1991) Influences on the extent of breast feeding: A prospective study in the Philippines. Demography 28: 181-199. Verronen, P. (1985) Breast-feeding of low birth weight infants. Acta Paediatr. Scand. 74: 495-499. Villar, J. & Belizan, J. (1982) The relative contribution of pre maturity and fetal growth retardation to low birth weight in developing countries. Am. J. Obstet. Gynecol. 143: 793-798. World Health Organization (1981) Contemporary Patterns of Breast-Feeding: Report on the WHO Collaborative Study on Breast-Feeding. World Health Organization, Geneva, Switzer land. World Health Organization (1984) The incidence of low birth weight: an update. Weekly Epidemiol. Record 59: 205-212. World Health Organization (1992) Global Health Situation and Projections: Estimates. Division of Epidemiological Surveillance, World Health Organization, Geneva, Switzerland. Zohoori, N., Popkin, B. M. &.Fernandez, M. (1993) Breast-feeding patterns in the Philippines: a prospective analysis. J. Biosoc. Sci. 25: 127-138.

Ballard,J. L, Novak, K.K..& Driver, M. (1979) A simplified score for assessment of fetal maturation in newly born infants. J. Pediatr. 95: 769-774. Barros, F. C., Victora, C. G., Vaughn, J. P. & Smith, P. G. (1986| Birth weight and the duration of breast-feeding: are the beneficial effects of human milk being overestimated? Pediatrics 78: 656661. Briscoe, ]., Akin, J. S. & Guilkey, D. K. (1990) People are not passive acceptors of threats to health: endogeneity and its con sequences. Int. J. Epidemiol. 19: 147-153. Butz, W. P. & DaVanzo, J. (1981) Determinants of Breastfeeding and Weaning Patterns in Malaysia. Rand Corporation, Santa Monica, CA. CebStudy Team (1991) Underlying and proximate determinants of child health: the Ceblongitudinal health and nutrition study. Am. J. Epidemiol. 133: 185-201. Churella, H. R., Bachhuber, W. L. &MacLean,W. C. (1985) Sur vey: methods of feeding low birth weight infants. Pediatrics 76: 243-249. de Chateau, P. & Wiberg, B. (1977) Long term effect on motherinfant behavior of extra contact during the first hours post-partum. Acta Paediatr. Scand. 66: 137-151. Fernandez, Ma. E. &.Popkin, B. M. (1988) Prelacteal feeding pat terns in the Philippines. Ecol. Food Nutr. 21: 303-314. Forman, M. R. (l984) Review of research on the factors associated with choice and duration of infant feeding in less-developed countries. Pediatrics 74 (suppl part 2): 667-694. Hausman, J. A. (1978) Specification tests in econometrics. Econ ometrica 46: 1251-1271. Madalla, G. S. (1988) Introduction to Econometrics. Macmillan, New York, NY. Narayanan, I. (1986) Care of the low birth weight infant in de veloping countries. Ann. Trop. Pediatr. 6: 11-15. Pollitt, E., Gilmore, M. &. Valcarcel, M. (1978) The stability of sucking behavior and its relationship to intake during the first month of life. Infant Behav. Devel. 1: 347-357. Popkin, B., Akin, J. S., Billsborrow, R. E. a Yamomoto, M. E. (1983) Breast-feeding practices in low income countries. Med. Anthropol. 7: 1-31.

Downloaded from jn.nutrition.org by guest on August 1, 2013