In affluent populations, most infants with low birth weight (LBW)⁴ are the consequence of preterm delivery and are adequately grown for their gestational age. In contrast, in developing countries the majority of low-weight births are full-term infants who have experienced growth retardation in utero, with the weight deficit reflecting stunted linear growth and/or reduced deposition of soft tissue, notably fat. Since 70% of linear growth normally occurs by 28 wk of gestation, and fetal fat deposition occurs at the end of gestation (Falkner et al. 1994), the differing anthropometric characteristics of LBW newborns permit inferences to be made about the timing and chronicity of fetal insults. Thus stunted newborns with a normal ponderal index [body weight, g × 100/(length, cm)³] are inferred to have suffered an insult early in gestation, whereas nonstunted, wasted newborns with a low ponderal index are considered to have experienced a late insult.

Villar and colleagues in Guatemala have done much to focus attention on the differential postnatal outcomes of stunted vs. wasted LBW newborns, reporting poorer growth and development of those born stunted (Villar et al. 1984). Similar findings have been reported elsewhere (Adair 1989). However, not all stunted LBW newborns appear destined to remain small. For example, in Sweden excellent catch-up of stunted newborns has been reported (Albertsson-Wikland et al. 1993). This could be interpreted in two ways: 1 ) good postnatal socioeconomic conditions overcome fetal insults or 2 ) good prenatal socioeconomic conditions protect the fetus from the types of insult that have permanent adverse effects.

Being small at birth and remaining small during childhood is of concern because stunting is associated with reduced cognitive outcomes, school achievement and adult work capacity (Haas et al. 1996, Powell et al. 1995). Low birth weight combined with poor compensatory growth by one year of age has also been reported to be causally linked to cardiovascular disease in later life (Barker 1994). This is part of a larger concept in which it is hypothesized that the timing of the fetal insult has profound effects on metabolism, and that persistence of these metabolic changes, together with changes in the structure of organs, may predispose growth-retarded infants to hypertension, diabetes, coronary heart disease and hemorrhagic stroke (Barker 1994). Opinion is divided, however (Elford et al. 1991 and 1992, Paneth and Susser 1995), and some hold the view that the observed associations between early growth and adult disease may be the result of confounding. For example, indices used to control for socioeconomic differences may have been too crude (Ben-Shlomo and Davey Smith 1991).

In view of the importance attached to postnatal catch-up growth in LBW infants, a fuller understanding of the enabling/inhibiting factors would be both advantageous and of public health interest. In 1993 we intensively monitored a cohort of 133 LBW infants living in poor socioeconomic conditions in northeast Brazil. Despite restricting participation to poor families from a relatively small catchment area, infant growth proved to be highly heterogeneous. In this paper we report the relative contributions of a number of variables to this heterogeneity.

METHODS

Study infants were recruited over a one-year period beginning January 1993 from the maternity wards of two state hospitals (in Palmares and Ribeirão), one private hospital (Palmares), and three smaller government health centers (in Catende, Água Preta and Joaquim Nabuco). There are no other maternity centers in the area, and more than 90% of deliveries occur in these facilities. The incidence of LBW is 9%. The infants and their mothers have been described previously (Lira et al. 1996).

All infants in this study, both LBW and ABW, were randomly assigned to receive either 1 mg zinc daily for eight weeks from birth, or a placebo preparation similar in taste and appearance. This intervention had no impact on any of the outcome measures considered, and identical conclusions are reached controlling for treatment allocation.

At 4, 8, 17, 26 and 52 wk, a team of three field workers specially trained in anthropometry visited the infants in their homes, and 98.4% of measurements were made within 3 d of the scheduled date. Weights were measured with a regularly calibrated portable scale (Model MP10 and MP25, CMS, London, UK), and lengths were measured twice using a Harpenden Infantometer. Standard techniques were again used. A difference of 0.5 cm was considered adequate. Infant weights and lengths were converted into weight-for-age z-scores (WAZ) and length-for-age z-scores (LAZ) using median values from the National Center for Health Statistics (NCHS) (1978) as the reference. Exact age in days at the time of measurement was used to calculate these age-standardized measures. It has previously been shown that the growth of elite infants in this region of Brazil is comparable to the NCHS reference population. Information on the infants' family backgrounds, housing conditions, and maternal smoking, alcohol consumption and work during pregnancy was collected by questionnaire applied in the maternity center at the time of enrollment.

For estimating maximum gain in length-for-age z-score, individual growth curves were fitted without a cubic term in age, as this did not improve the fit of the model. The models were fitted using the software package ML3 (Institute of Education, London, UK, Prosser et al. 1991). Goodness of fit was assessed by critical examination of the model residuals, both at level one and at level two. Infants whose z-score remained constant or deteriorated from birth were considered to have achieved a maximal z-score gain of 0. The few infants who were still improving in length-for-age z-score at 12 mo of age were considered to have reached maximal gain at exactly 12 mo.

The estimated maximum z-score gains were related to individual-level covariates using multiple linear regression. Possible socioeconomic confounders were introduced into the model first, with an entry criterion of P  0.2. Other variables of interest were then entered into the model in a stage-by-stage process, so that the underlying hierarchies in the causal pathway would be respected (Victora et al. 1997). In practice, this meant that maternal factors were controlled only for socioeconomic status, whereas anthropometric characteristics of the infant at birth were controlled for both socioeconomic status and maternal variables, while neonatal morbidity and feeding patterns were controlled for socioeconomic status, maternal variables and birth anthropometry. Because peak gains occurred very early, postneonatal morbidity and postneonatal feeding mode were not included in the models to minimize the risk of identifying associations resulting from reverse causality.

RESULTS

Table 1. Mean weight-for-age z-scores (WAZ) and the variation in WAZ at intervals during the first year of life for infants of low birth weight (1500-2499 g) and appropriate birth weight (3000-3499 g) [View Table]

Associations were sought between maximal WAZ gain in the LBW infants and a number of different variables relating to the infants' prenatal and neonatal environment. As a first step, the infants' socioeconomic background was examined. Seventeen variables relating to housing construction and facilities, income, father's presence and literacy, and crowding were entered in a stepwise multiple linear regression, and those significant at the 20% level were retained. Three variables remained in the model: presence/literacy of the father (coded 1 = absent, 2 = present but illiterate, and 3 = present and literate), type of toilet facility, and ownership of a television. Together these three variables explained 21% of the variation in maximal WAZ gain, and were used to control other associations for the effects of socioeconomic confounding.

Table 2 shows the associations between maximal WAZ gain and three sets of possible early enabling/inhibiting factors: mother's size, selected insults during pregnancy, and birth anthropometry. Effects are shown both before and after controlling for socioeconomic status, which is likely to confound all these associations. Associations between infants' anthropometry at birth and subsequent WAZ gain are additionally controlled for mother's weight. Mother's weight and height were both associated with maximal WAZ gain after adjusting for socioeconomic status, with mother's weight showing the stronger association. A 10 kg difference in mother's weight was positively associated with a 0.2 z-score difference in maximal WAZ gain. Neither smoking nor work during pregnancy was associated with maximal WAZ gain, but both infant length and ponderal index at birth were, with a 1 z-score difference in birth length positively associated with a 0.3 z-score difference in maximal WAZ gain.

Table 2. Associations between prenatal (maternal) and natal risk factors and maximal gain in weight-for-age z-score achieved at any time during the first year of life for infants of low birth weight1 [View Table]

Table 3 presents the associations between maximal WAZ gain and 1 ) the average number of breast feedings/nonbreast feedings per day during the first four weeks of life, and 2 ) the proportion of all neonatal days that the infant was reported by his/her mother/caregiver to be suffering from diarrhea, cough, vomiting or fever. Neither the number of breast feedings nor the number of nonbreast feedings was associated with WAZ gain, although marked confounding was seen in the case of breast feeding (indicated by the marked change in the coefficient on controlling for the measured socioeconomic variables), and the potential for residual confounding due to other unmeasured socioeconomic variables should perhaps be considered. Proportion of time sick was strongly associated with maximal WAZ gain, with a 20% absolute difference in the proportion of neonatal days spent sick negatively associated with a 0.2 z-score difference in maximal WAZ gain.

Table 3. Associations between neonatal risk factors (feeding mode and morbidity during weeks 0-4) and maximal gain in weight-for-age z-score achieved at any time during the first year of life for infants of low birth weight1 [View Table]

In a final multivariate model, socioeconomic variables explained 21.4% of the variation in maximal WAZ gain, mother's weight explained a further 4.4%, infant birth length another 4.7%, and illness in the neonatal period a further 5.4%. Thus 35.9% of the total variance of this outcome was explained by the six variables considered (adjusted R² = 0.328).

Table 4. Mean length-for-age z-scores (LAZ) and the variation in LAZ at intervals during the first year of life for infants of low birth weight (1500-2499 g) and appropriate birth weight (3000-3499 g) [View Table]

Associations were sought with a variety of potential enabling/inhibiting factors, following the procedures for weight gain that were shown in Tables 2 and 3. The results of the final model for length gain are shown in Table 5. Socioeconomic factors accounted for 24.4% of the total variance in maximal LAZ gain. After controlling for socioeconomic variables, mother's height (not weight) accounted for a further 4.9% of the variance. A 10 cm difference in mother's height was positively associated with a 0.2 z-score difference in maximal LAZ gain. Smoking during pregnancy was associated with a 0.3 z-score shortfall in maximal LAZ gain, and a 20% absolute difference in the proportion of days spent sick during the first 4 wk of life was negatively associated with a 0.1 z-score difference in maximal LAZ gain. Infant anthropometry at birth was not associated with this outcome (P > 0.1 in all cases, after adjusting for socioeconomic variables and mother's height and smoking habit). A total of 35.7% of the total variability in maximal LAZ gain was explained by the variables considered in the model (adjusted R² = 0.321).

Table 5. Associations between socioeconomic indicators, maternal risk factors and neonatal morbidity with maximal gain in length-for-age z-score for infants of low birth weight: multi-stage, multivariable model1 [View Table]

DISCUSSION

In reviewing the results the following may be noted: 1 ) WAZ gain differentials were most apparent in the first 8 wk, with some LBW infants having a dramatic initial increase, or take-off, 2 ) LAZ gain differentials emerged more gradually, 3 ) 36% of the variability in maximum WAZ and LAZ gains was explained by just six variables (three socioeconomic and three other). The rapidity of onset of differential growth patterns so soon after birth, and the asynchrony of the timing of the peak WAZ and LAZ gains, suggest that the differential response is due, at least in part, to rapid soft tissue gain. This is borne out by comparing the stunted-only group with the group that was both wasted and stunted, as shown in Figure 1. Infants in the latter group experienced accelerated weight gain initially, but then were indistinguishable from the stunted-only infants.

Despite the fact that only poor families were eligible for the study, socioeconomic status explained 21 and 24% of the variance in maximal WAZ and LAZ gains, respectively. How socioeconomic status, if acting only postnatally, could so dramatically differentiate the patterns of take-off is not clear. Whatever the ultimate controlling mechanism, differences in postnatal growth among individual infants can only involve differences in 1 ) dietary intake, 2 ) prenatal nutrient stores, 3 ) efficiencies of absorption and utilization, 4 ) basal metabolism, 5 ) physical activity and/or 6 ) morbidity. Socioeconomic status could conceivably affect dietary intake, either by influencing a ) the number of breast or supplementary feedings given, b ) the volume of breast milk or supplement consumed per feeding and/or c ) the quality of the breast milk or supplement. Regarding a ), neither the number of breast feedings nor supplementary feedings in the first month were associated with WAZ or LAZ gains, and it was during this period that differential growth patterns were most pronounced. We have no data pertaining to b ) or c ), but if these were mediating factors one might expect them to affect postnatal growth rather later, when nutrient demands are greater. Socioeconomic status could also influence the fetal store of micronutrients, but again, if this were the mediating factor one might expect a later effect, after existing stores have been depleted. Psychosocial care mediators are unlikely to operate so soon after birth, except through the measured feeding and morbidity variables.

Small maternal stature and low maternal weight, as well as maternal smoking and neonatal illness, were all more common in the poorest families in this population. It is therefore clear that part of the large effect of socioeconomic status was mediated by these variables, which we have shown were associated with infant growth. However, when these associations were examined by path analysis (data not shown), we found that neither the associations between these proximal determinants and the socioeconomic measures, nor the associations between these determinants and the growth outcome, were sufficiently strong to explain more than a small proportion of the overall socioeconomic effect.

Taken together, we surmise that socioeconomic status is acting prenatally, and that the early differential growth patterns are mediated through maternal factors that determine fetal linear and ponderal growth. If the outcome is a wasted LBW newborn, it is reasonable to expect that such a baby would catch-up in weight-for-length in a comparable way to postnatally wasted infants. Stunted infants might be down-regulated with altered gene expression, reduced cell numbers, altered hormone release and receptor sensitivity, and altered organ structure (Barker 1994, Dahri et al. 1995, Owens et al. 1994, Widdowson et al. 1972). Interestingly, in Guatemala stunted fetal growth was associated with several socioeconomic indicators, but this was not the case for wasted newborns (Neel and Alvarez 1991). Down-regulation could explain why LBW Indian orphans born in the poorest communities but adopted soon after birth by affluent Swedish families attained an adult height comparable to Indians who had experienced lifelong disadvantage (Gopalan 1994).

In New York City, Naeye et al. (1971) compared organ weights and body size from consecutive autopsies on stillborn and newborn infants according to urban poverty indices of the US Social Security Administration. Those from families below the poverty line were shorter, lighter, and had smaller liver, spleen, thymus and adrenal glands compared with infants from families above the poverty line. These changes together with morphological differences were attributed to maternal undernutrition. Altered organ structure has been found to have marked effects on postnatal function in infants growth-retarded in utero. For example, altered intestinal and pancreatic structure adversely affects nutrient absorption (Boehm et al. 1991 and 1992, Kolacek et al. 1990) and reduced numbers of  cells may limit insulin production (Barker 1994). Recent data suggest that maternal undernutrition may increase the fetal production of untoward cytokines including interferon (IFN-) and tumor necrosis factor (TFN-) (Hanson et al. 1996, Heyborne et al. 1992). Insulin-like growth factor I (IGF-I) plays a central role in regulating fetal growth, and fetal IGF-I can be modulated by maternal dietary intake (Gluckman 1993). Postnatally, IGF-I concentrations remain low in infants who have been growth-retarded in utero (Theiriot-Prevost et al. 1988).

In conclusion, the public health implications are that the differential rates of growth in these LBW infants are associated with socioeconomic status, and that marked differentials exist even within a population that is already labeled "poor." Five other determinants of maximal WAZ and LAZ gains were maternal weight and infant length (WAZ alone), maternal height and maternal smoking (LAZ alone) and neonatal illness (both WAZ and LAZ). These variables explained an additional 15% of the variance in WAZ gain and 11% of the LAZ gain. This analysis serves to reinforce the need to break the vicious cycle of intergenerational disadvantage through prenatal intervention programs in the short-term and poverty alleviation in the long-term.