Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Abstract
Background: Interactions between nutrition and infections in developing countries are complex, and analyses of the
interrelations require data from longitudinal studies that span several years and a comprehensive framework facilitating
the formulation of food and health policies.
Objective: The effects of dietary intakes in the Philippines for birth outcomes, heights, weights, and morbidity during ages
2–24 mo; heights and weights during ages 8–19 y; and completed adult heights were analyzed.
Methods: Data on >3000 children from the Cebu Longitudinal Health and Nutrition Survey were modeled in 4 sets of
analyses. Cross-sectional and multilevel models were estimated for length and weight at birth. Dynamic random-effects
models were estimated for childrenÕs heights, weights, and morbidity levels during ages 2–24 mo. ChildrenÕs heights and
weights during ages 8–19 y were modeled by using random-effects models. ChildrenÕs heights at age 22 y were modeled
for investigating the effects of nutritional intakes in early childhood and in adolescence.
Results: Maternal anthropometric indicators, energy intakes, and sociodemographic variables were significantly
associated with childrenÕs length and weight at birth. Dynamic models for childrenÕs heights and weights during ages
2–24 mo showed significant effects of calcium and protein intakes; child morbidity levels were significantly associated
with height and weight. Higher b-carotene intakes were significantly associated with lower morbidity levels. Analyses of
data on children during ages 8–19 y showed significant effects of protein and calcium intakes and of morbidity levels on
heights and weights. Models for childrenÕs heights at age 22 y indicated significant effects of protein and calcium intakes
during early childhood and in adolescence.
Conclusions: The results underscored the need for formulating long-term food and health policies for the Philippines that
enhance childrenÕs physical development and ultimately their adult stature, which is important for physical work capacity
and labor productivity. J Nutr 2016;146:133–41.
Keywords: anthropometric indicators, diet quality, econometric modeling, food policies, longitudinal data,
long-term evaluation, maximum likelihood estimation, morbidity
Introduction
to analyze the factors underlying childrenÕs physical growth and
The importance of interactions between infections and individ-
morbidity that ultimately determine their adult physical work
ualsÕ nutritional status has long been recognized, especially for
capacity (10) and labor productivity (11).
developing countries (1–6). Inadequate protein and micro-
For the design of efficacious food policies, it is important to
nutrient intakes lower the capacity of the body to fight infections, adopt multidisciplinary approaches that incorporate interactions
which, in turn, adversely affects nutritional status by prolonging between nutritional status and infections. For example, weaning
sickness spells. Moreover, absorption rates for nutrients such as programs for breastfed infants may not be effective if they
iron and calcium are often low due to excessive phytate intake inadvertently increase morbidity levels because the mothers failed
from grains (7, 8). In addition, food contamination due to poor to maintain hygiene in preparing the supplements (12). Moreover,
hygiene exacerbates nutrient loss (9). It is therefore important it is important for food policies to match childrenÕs nutritional
requirements with their intakes at various ages. This is an intricate
1
task because childrenÕs requirements depend on their ages and
Supported in part by the Harvest Plus program of the International Food Policy
Research Institute.
growth pattern history (13, 14). It is therefore important to adopt
2
Author disclosures: A Bhargava, no conflicts of interest. a long-term framework for analyzing the effects of childrenÕs
*To whom correspondence should be addressed. E-mail: bhargava@umd.edu. nutritional intakes and morbidity on their physical development.
ã 2016 American Society for Nutrition.
Manuscript received August 20, 2015. Initial review completed October 1, 2015. Revision accepted November 2, 2015. 133
First published online December 9, 2015; doi:10.3945/jn.115.222869.
Furthermore, there has been emphasis in the literature on early May 1983 and April 1984 were invited to enroll in the study, and the refusal
nutritional interventions, especially during the first 2 y of life (15). rate was 4%. The 3080 women in the sample with single births were
Although such interventions may be necessary for undernourished interviewed during pregnancy, at birth, and then bimonthly from ages 2
children growing up in poverty, it is important to adopt a broader to 24 mo (i.e., there were 13 observations on infants in the initial surveys).
Subsequently, mothers and children were interviewed in follow-up surveys
perspective on ‘‘critical windows’’ that may be available for catch-
in 1991, 1994, 1998, 2002, and 2005 when the children were 8, 10, 15,
up growth (16). Thus, for example, heights in the Netherlands 19, and 22 y old, respectively.
surpassed all countries after the Second World War presumably The first survey round conducted during pregnancy measured
due to the school milk programs, which rewarded children for mothersÕ heights and weights around the 30th week of gestation.
daily drinking up to 5 glasses of milk to become ‘‘M-brigadiers’’ MothersÕ food intakes were measured by using the 24-h recall method
(17). The programs were gradually discontinued in the 1980s, and were converted into energy and nutrient intakes with the use of
and some reversion to the mean in Dutch heights is apparent tables for the Philippines (28). MothersÕ food intakes were again
(18). Because childrenÕs requirements for protein and micronu- measured when the infants were born and when they were ;6 and 12 mo
134 Bhargava
addition, mothersÕ energy and nutrient intakes are important for during the 2- to 24-mo period and was included as an explanatory
intrauterine development (31). The correlation between mothersÕ highest variable in the models. In addition, data on dietary intakes were compiled at
grade and socioeconomic index was 0.41, and it did not present any 3 time points, and protein and calcium intakes were averaged to produce
difficulties for the estimation of model parameters. figures for 10–19 y; averaging was helpful in reducing the impact of
Second, the dynamic random-effects model for childrenÕs heights within-child variations in intakes (36). In alternative versions of the
with the use of bimonthly observations in the period from 2 to 24 mo is model, 2 indicator variables for time periods were included as explan-
given in Equation 2: atory variables to account for the trends underlying the dietary intakes.
Dynamic and static random-effects models were also estimated for
lnðHeightÞit ¼ b0 þ b1 ðGenderÞi þ b2 ðBirth orderÞi
childrenÕs weights. Last, 4 specifications were estimated for investi-
þb3 ðNumber of persons in householdÞi gating the effects of childrenÕs heights at age 2 y on their completed
þb4 ðSocioeconomic indexÞi þ b5 ðMothers highest gradeÞi heights.
þb6 ðMothers morbidity indexÞit
þb7 lnðMothers energy intake=dÞit Statistical methods and test statistics. The statistical estimation
Age
Variable Birth 1y 2y 8y 10 y 15 y 19 y
Girls, % 47 — — — — — —
Birth order 3.14 6 2.0 — — — — — —
Number of persons in household 5.57 6 2.8 — — — — — —
Socioeconomic status index 7.25 6 4.0 — — — — — —
MothersÕ — — — — — — —
Highest grade 7.56 6 3.7 — — — — — —
Results for infant lengths and weights at birth. Table 2 test indicated that the variance of random effects was statisti-
presents the results for natural logarithms of lengths and weights cally greater than zero. The main findings from these models
of infants at birth with the use of 2 specifications. The model was were, first, that childrenÕs birth order and number of persons in
first estimated by using ordinary least squares; ‘‘barangay’’-level the household were associated in a nonlinear fashion with their
random effects were included in the second specification and the lengths and weights at birth. The estimated coefficients showed
variables were estimated by maximum likelihood (38, 39). The that infant lengths and weights significantly (P < 0.05) increased
results from the 2 models were very close, although a statistical with birth order, although at a declining rate; lengths and
TABLE 2 Ordinary least-squares and maximum likelihood estimates of models for childrenÕs lengths and weights at birth in Cebu,
Philippines1
Dependent variable
ln (Length), cm ln (Weight), kg
Explanatory variables Ordinary least squares Maximum likelihood2 Ordinary least squares Maximum likelihood2
136 Bhargava
weights declined with respect to household size at an increasing the previous time period was also negative and significant.
rate. Girls were shorter and lighter than boys at birth. Although childrenÕs intake of protein expressed as a ratio to
Second, the coefficients of mothersÕ highest grade and socio- energy intake was not significantly associated, the coefficient of
economic index were positive and significant. This was also calcium intake was positive and significantly associated with
true for mothersÕ heights and weights. The ‘‘elasticity’’ of infant childrenÕs heights. The coefficient of the lagged dependent
length with respect to mothersÕ height (percentage change in variable was 0.88, implying large long-run effects of explanatory
infant length resulting from a 1% change in mothersÕ height) was variables on childrenÕs heights. For example, whereas the short-
0.13; the elasticity with respect to mothersÕ weight was 0.05. The run elasticity of childrenÕs heights with respect to the calcium
elasticities of infant birth weight with respect to mothersÕ heights to energy intake ratio was 0.001, the long-run elasticity was
and weights were 0.30 and 0.25, respectively. Third, mothersÕ ;0.01.
energy intakes were significantly associated with infant lengths; The results for childrenÕs weights in Table 3 are presented for
the coefficients were positive but not significant in the model for 2 specifications—namely, where childrenÕs height was excluded
TABLE 3 Maximum likelihood estimates of dynamic random-effects models for heights and weights of
children in Cebu, Philippines, ages 2–24 mo1
Dependent variable
Explanatory variables ln (Height), cm ln (Weight), kg ln (Weight),2 kg
Dependent variable
Explanatory variables ChildrenÕs morbidity index 1 ChildrenÕs morbidity index 2 ChildrenÕs diarrhea index
but was not significant in the model for diarrhea index. Whereas corresponding estimate, 0.35, from the static model. In contrast,
mothersÕ highest grade was significantly negatively associated, the estimated long-run elasticity of childrenÕs weights with
mothersÕ morbidity index was positively associated with all 3 respect to mothersÕ heights was 0.47, which was very close to the
indexes. Moreover, greater hygiene in the house was associated corresponding estimate from the static model.
with lower child morbidity levels. This was also true for number Third, the number of immunizations received by children
of immunizations that the children received and for the ratio of during the ages of 2–24 mo was significantly associated with
b-carotene to energy intakes. height. Moreover, childrenÕs morbidity levels during the ages of
Finally, the lagged dependent variables were estimated with 2–24 mo were negatively associated with height and weight in
small coefficients of 0.05 and 0.04 in the models for morbid- later years; the coefficients were significant in the static versions
ity indexes 1 and 2, respectively, and were significant. The co- of the models. Last, coefficients of childrenÕs average intakes of
efficient of the lagged dependent variable in the model for protein and calcium during the ages of 10–19 y were significant
diarrhea index was larger (0.58), indicating longer persistence in the model for height; coefficients of protein intakes were
in the effects of explanatory variables on diarrhea index. positive and significant in the model for weight. Coefficients of
Coefficients of previous heights were negative, indicating that lagged dependent variables were lower in Table 5 than those in
taller children were sick less often and with lower intensity. The Table 3 because the models were estimated by using observa-
exogeneity hypothesis for previous heights could not be tions separated by several years.
rejected in the models for morbidity indexes 1 and 2; the
exogeneity hypothesis was rejected in the model for diarrhea Results from models for childrenÕs heights at 22 y
index. explained by current and early variables. Table 6 reports
the results from 4 specifications for childrenÕs completed
Results for childrenÕs heights and weights during ages heights at 22 y with the use of data on heights at age 2 y. The
8–19 y. Table 5 reports the estimates from dynamic and static results from specification 1 showed that the coefficients of
random-effects models for childrenÕs heights and weights with mothersÕ height and childrenÕs calcium intakes in adoles-
the use of data at 8, 10, 15, and 19 y. First, the coefficient of the cence were positive and significantly associated with completed
socioeconomic index was significant only in the dynamic model heights at age 22 y. However, the coefficient of calcium intakes
for childrenÕs heights; this coefficient was positive but was not at 2 y was estimated with an unexpected minus sign, which was
significant in the static version of the model. The coefficient of significant. The coefficient of height at age 2 y was 0.43, which
mothersÕ highest grade was significant in all models except the was large and significant. In specification 2, estimated by using
dynamic model for weights. instrumental variables, this coefficient decreased to 0.14 and
Second, the coefficient of mothersÕ height was estimated to be was no longer significant. The instrumental variables used for
large and significant in the models for childrenÕs heights and heights at age 2 y were mothersÕ highest grade and weight. A
weights. Whereas the short-run elasticity of childrenÕs heights test for exogeneity (41) showed that there was likely to be a
with respect to mothersÕ heights was 0.30 in the dynamic model, correlation between errors affecting this model and childrenÕs
the long-run elasticity was 0.76, which was larger than the heights at age 2 y.
138 Bhargava
TABLE 5 Efficient estimates of dynamic and static random-effects models for heights and weights of
children in Cebu, Philippines, by using data at ages 8, 10, 15, and 19 y1
Dependent variable
ln (Height), cm ln (Weight), kg
Explanatory variables Dynamic model2 Static model2 Dynamic model Static model
In specification 3, the dynamic model was estimated by using when height was treated as an endogenous variable and remained
data at ages 2 and 22 y and by treating height at 2 y as an significant.
exogenous variable. However, coefficients of childrenÕs protein
and calcium intakes were not statistically different from zero. In
Discussion
specification 4, which treated previous height as an endogenous
variable, the estimated coefficients of protein and calcium This article presented 4 sets of comprehensive analyses for
intakes were 0.006 and 0.003, respectively, and were significant. growth and morbidity patterns of children in Cebu, Philippines,
The likelihood ratio test statistic for the exogeneity hypothesis and tackled the conceptual and methodologic aspects. First, the
for heights at 2 y was 13.6 and indicated that child-specific results for infant lengths and weights at birth underscored the
random effects were correlated with previous heights. Moreover, importance of mothersÕ nutritional status reflected in height,
the coefficient of childrenÕs heights at 2 y decreased to 0.025 weight, and energy intakes. Moreover, in alternative models for
TABLE 6 Efficient estimates of models for completed heights of children in Cebu, Philippines, at age 22 y
explained by variables including heights at age 2 y1
140 Bhargava
22. Adair LS, Popkin B, Akin J, Guilkey D, Gultiano S, Borja J, Perez L, 34. Bhargava A, Reeds PJ. Requirements for what? Is the measurement of
Kuzawa C, McDade T, Hindin M. Cohort profile: the Cebu Longitu- energy expenditure a sufficient estimate of energy needs? J Nutr
dinal Health and Nutrition Survey. Int J Epidemiol 2011;40:619–25. 1995;125:1358–62.
23. Bhargava A, Sargan JD. Estimating dynamic random effects models 35. Bhargava A. Identification and panel data models with endogenous
from panel data covering short time periods. Econometrica 1983;51: regressors. Rev Econ Stud 1991;58:129–40.
1635–59. 36. Nelson M, Black AE, Morris JA, Cole TJ. Between-and-within subject
24. Allen LH, Wood RJ. Calcium and phosphorus. In: Shils ME, Olson JA, variation in nutrient intake from infancy to old age: estimating the
Shike M, editors. Modern nutrition in health and disease. 8th ed. number of days to rank dietary intakes with desired precision. Am J Clin
Philadelphia: Lea and Febiger; 1994. Nutr 1989;50:155–67.
25. Rand Corporation. Measurement of physiologic health for children. 37. Numerical Algorithm Group. Numerical Algorithm Group mark 13.
Vol. 1–5. Santa Monica (CA): Rand Corporation; 1983. Oxford (United Kingdom): Oxford University; 1991.
26. Bhargava A. Modelling the health of Filipino children. J R Stat Soc Ser 38. Goldstein H. Multilevel statistical models. New York: Oxford Univer-
A Stat Soc 1994;157:417–32. sity Press; 1995.