TRANSACTIONSOFTHEROYAL SOCIETYOFTROPICALMEDICINEANDHYGIENE(1996)90,442-445

Seasonal change in nutritional status among young children in an urban shanty

town in Peru

Carmen M. Ma&, JosC Luis Segural, Caryn Bern 2, David S. Freedman2, A. Guillermo Lescanol, Luis E.
Benaventel, Luis G. Corderol, Laura Clavijo’ and Josephine B. Gilman 1 ‘Asociacibn Bent;fica PRISMA, Lima, Peru;
2Division of Nutrition and Physical Activity, National Center for Chronic Disease Prevention and Health Promotion, Centresfor
DiseaseControl and Prevention, Atlanta, Georgia, USA

Abstract
Seasonalvariation in nutritional status among young children has been described in rural populations, but

Downloaded from https://academic.oup.com/trstmh/article/90/4/442/1866793 by University of Basel user on 21 January 2021

in few urban settings. We examined seasonality in 7 years of nutritional surveillance data from an urban
shanty town near Lima, Peru, where childen O-35 months old were measured at intervals of 45 months.
We compared nutritional status by month, using generalized estimating equations to account for the inter-
correlations among measurements of the same person at different times. The periodicity of the seasonal
variation was found to fit a model in which the month of the year was sine-transformed, and this sine-
transformed model was used to examine possible interactions with age?sex and year of examination. A total
of 38 626 measurementswas available from 11 333 children. In late winter, mean weight-for-height was an
estimated 0.38 Z score higher than in late summer. The seasonaleffect occurred at all ages,in both sexes,
and in each year of surveillance. The amplitude was greatest for children 6-23 months old. The summer
trough in weight-for-height was lower in 1989 than in other years; children who experienced this summer
low had lower mean height-for-age in subsequent years. The seasonalvariation in nutritional status may be
related to differences in dietary intake! or to the higher prevalence of bacterial diarrhoea in summer than in
winter. The more marked drop in weight-for-height in 1989 and subsequent trough in height-for-age may
be related to political and economic changesthan adversely affected food accessin Peru.
Keywords:nutritional status,seasonality,urban children, Peru
Introduction months. We assessednutritional status using the weight
Seasonal changes in nutritional status have been re- and height curves of the NCHS/CDC growth reference.
ported predominantly among rural populations depend- Three anthropometric indicators-weight-for-height,
ent on an anricultural cvcle (MCGREGOR et aZ.. 1968: weight-for-age, and height-for-age-were expressed as
ROWLAND et& 1977; BR&N et al., 1982).This skasonai standard deviation or Z scores. For example, a weight-
variation may be due to several factors, including differ- for-height Z score of +l corresponds to a value one
encesin food availability before and after the harvesting standard deviation above the median of the reference,
season,increased workload and decreasedtime for child while a Z score of zero corresponds to the median.
care during planting and harvest seasons,and seasonal To determine the association between nutritional
peaks in infectious diseaseprevalence (CHEN et aZ., 1979; status and season, we first examined monthly mean Z
TROWBRIDGE & NEWTON, 1979; BAIRAGI, 1980). In the scoresfor the period between January 1987 and October
few urban areas where seasonal variation has been re- 1993. A locally weighted scatterplot smoother (lowess)
ported, the effect was less consistent or more modest curve (CLEVELAND, 1979) which relied entirely on the
(TROWBRIDGE & STETLER, 1982; TOMKINS et al., 1986). data to specify the form of the model, was used to sum-
We used 7 years of surveillance data to examine the sea- marize the mean weight-for-height Z score over time.
sonality of nutritional status of young children in an ur- This technique, which is somewhat analogous to a mov-
ban shanty town in Peru. ing average,uses weighted least-squaresregression on a
subset of the data (5 values for weight-for-height) to fit
Materials and Methods each value of the smoothed curve, with the assigned
We examined data from thepueblo joven (shanty town) weights decreasing as the distance from the point of in-
of Pampas de San Juan de Miraflores (population terest increases.
35 000), located 20 km south of the centre of Lima, Peru These descriptive analyses suggested that several of
(FERNANDEZ-CONCHA et al., 1991).In this marginal peri- the Z scoresfollowed a cyclic pattern with a periodicity
urban community, about half the dwellings are con- of one year, and we then fitted regression models incor-
structed of brick or cement, and the remainder have porating 11 indicator variables (for the 12 months), to-
walls made from straw mats, cardboard, or a combina- gether with sex, age.,and year of examination. We also
tion of such materials. The major sources of drinking found that a model m which the month of the year was
water are either public standpipes or tanker trucks; few sine-transformed fitted the data nearly as well as those
houses have formal sewagedisposal. Most people hving containing indicator variables for the months, and this
in Pampasmigrated to Lima from the countryside in the simpler, sine-transformed model (containing only one
1980s and now work as day labourers, as either small term for the months of the year) was used to examine
vendors or participants in cottage industries such as possible interactions with age, sex, and year of examina-
hand manufacture of wool and alpaca sweaters.There is tion. In this model, for example, the value for February,
no major variation in price or availability of food staples the second month, was 0.866 (sine [(2+12)x360’]),
by season.However, many fruits and vegetables are sea- while that for July was -0.5 (sine [(7+12)x360’]). The
sonal. estimated regression coefficient for the sine-transformed
Since 1987, Asociacion Ben&a PRISMA has con- month would then represent the amplitude, or one-half
ducted a comprehensive primary health care project in of the difference between the peak and trough of the
Pampas.As part of this project, a complete census of the curve. More complicated models, incorporating addi-
community was conducted in 1986-1987, and has been tional terms to examine a periodicity other than one year
updated regularly. Trained female field workers made or a shift along the x axis, did not improve the fit.
home visits to weigh and measure all community chil-
dren from birth to 3 years at regular intervals of 4-5 Many children had more than one measurement made
during the time period and, because serial measure-
Address for correspondence: Caryn Bern, MD, MPH, Division ments from an individual are not independent, standard
of Nutrition and Physical Activity, Mailstop K-25, Centers for regression techniques may produce biased results. We
Disease Control and Prevention, Atlanta, GA30341, USA. therefore used generalized estimating equations to ac-
SEASONAL CHANGE IN NUTRITIONAL STATUS 443

count for within-person correlations (DIGGLE et al.,
1994). These iterative procedures, which independently
estimate regression coefficients and their standard er-
rors, were implemented in SAS using the GEE 1 macro
of Karim (Technical report no. 674, Department of
Biostatistics, The Johns Hopkins University, Baltimore,
Maryland, USA, 1989). Although the presented results
are based on the assumption that all correlations be-
tween responses at different times are identical (i.e., an
exchangeable correlation structure), other possible cor-
relation structures, such as auto-regressive or m-depend-
ence, yielded very similar results. The GEE regression
to 0.7 standard deviations above the reference. The mean
height-for-age increased bv 0.15 Z scores from 1987 to
1993. but the lowest mean height-for-age occurred in
1990:
Mean weight-for-height and weight-for-age varied sig-
nificantlv with the season. while mean height-for-age
showed no seasonal variation. Additional analyses in&
cated that the seasonal variability of weight-for-age was
largely accounted for by the variation in weight-for-
height; subsequent analyses focused on weight-for-
height. Examination of the data, both in aggregate and
for each year of the 7 years period, indicated that mean

Downloaded from https://academic.oup.com/trstmh/article/90/4/442/1866793 by University of Basel user on 21 January 2021

coefficients (which account for within-person correla- weight-for-height was highest in August-October (late
tions) typically differed by less than 10% from those of winter) and lowest in March-May (late summer) of each
standard regression analyses. year (Fig. 1). Furthermore, the amplitude and peri-
odicity of the seasonality effect in the aggregate data
Results closely fitted that predicted by a sine transformation
A total of 38 626 measurements was available from (Fig. 2). This simpler model, which modelled the peri-
11333 children over the 7 years. The mean number of odicity with a single term, was therefore used for further
measurements available per child was 3.4 (range l-9), analysis of the seasonality effect.
and the mean number of measurements per year was For all age groups combined, the estimated difference
5525 (range 4023-7946). At least 100 measurements were in mean weight-for-height between September and
available for each month of the 7 years period with the March was 0.38 Z scores (twice the amplitude of 0.19)
exception of March-April 1990, when surveillance was (P<O.OOOl). The results were comparable for girls and
interrupted due to lack of funding. In the first 6 months boys, and in each year of the study, both in terms of the
of life, the mean weight-for-age and height-for-age of amplitude and the months in which the maximum and
Pampas shanty town children were close to those of the minimum Z scores were observed (Table 2). Although
reference population, but after 6 months of age, Pampas the effect was largest among children 6-11 and 12-23
children were both shorter and lighter than reference months of age, for whom the mean weight-for-height
children (Table 1). The mean weight-for-height was 0.3 was estimated to be nearly one-half of a standard devia-
Table 1. Anthropometric status by age, sex and year of measurement for 11333 children measured as part of nutri-
tional surveillance in a peri-urban shanty town in Lima, Peru
No. of Mean Z score
Measurements Height-for-age Weight-for-age Weight-for-height
Aged_;ll$nths)
6124 -0.18 0.50 0.72
6-l 1.9 6507 -0.75 -0.13 0.58
12-23.9
24-35.9 13091 -1.31 -0.52 0.34
12904 -1.16 -0.43 0.37
Sex
Male 19681 -0.99 -0.27 0.42
Female 18945 -0.99 -0.25 0.48
Year
1987 5012 -1.05 -0.21 0.54
1988 7928 -0.98 -0.22 0.51
1989 4594 -0.98 -0.33 0.36
1990
1991 4019 -1.11 -0.26 0.53
1992 6592 -0.99 -0.26 0.46
1993 6147 -0.92 -0.28 0.39
4334 -0.90 -0.31 0.33

0.87 ~0.8 0.7,

0.7

0.6:

0.51

0.41
sine-transformed model
0.3:

0.2:

o-o 1987 1988 1989 1990 1991 1992 1993

1
Year 0.2 1
1 2 3 4 5 6 7 a 9 IO 11 12
Fig. 1. Mean weight-for-height Z score by calendar month from Month of year
January 1987 to October 1993 for children O-35 months of age.
Curves were smoothed by the lowess method (see text). Asterisk Fig. 2. Seasonal variation in weight-for-height Z score predicted
(*) signifies 2 months of insufficient data (March-April 1990). by sine-transformed model and observed in surveillance data.
444 CARMENM.MARIN ETAL.

Table 2. Seasonal variation in weight-for-height by months old; we examined in more detail the growth of
age, sex, and calendar year children during this period. A similar trough in height-
for-age was seen one year later: for children who were
Amplitude 12-23 months old in 1989, the mean height-for-age Z
(Z scores) Minimum Maximum score at 24-35 months was -1.33, compared with -1.12
All children 0.19 March September to -1.14 for children born either 2 years earlier or 2 years
later (Table 3). Children who were younger than 12
AT-(;lonths) months in March 1989 were also shorter at 24-35
0.16 April August months of age than earlier or later cohorts of children,
6-l 1
12-23 0.24 March August although this difference was less pronounced. The eco-
24-35 0.23 March September nomic changes that occurred in the late 1980s in Peru
0.13 March September are reflected-in the yearly change in the per caput gross
national oroduct (GNP): after slow but consistent

Downloaded from https://academic.oup.com/trstmh/article/90/4/442/1866793 by University of Basel user on 21 January 2021

Sex
Male 0.19 March September growth in-1986 and 1987,yhe per caput GNP fell by 9.9%
Female 0.19 April September in 1988and by 14.1%in 1989 (Table 4). The GNP recov-
Year
ered to above US $1000per caput only in 1994.
1987 0.18 March August Discussion
1988
1989 0.19 March August The young children of this Peruvian shanty town had
1990 0.22 May August low height-for-age, but their weight-for-height was
1991 0.16 Februarya October slightly above that of the reference population; this pat-
1992 0.18 April September tern of anthropometry was found previously among chil-
1993 0.20 March October dren in another periurban Lima shanty town, as well as
0.16 April September among Bolivian and Mexican-American children
aOnly one child was measured in March, and 8 in April, 1990. (NATIONALINSTITUTEOFFOODANDNUTRITION, 1981;
LOPEZDE RoMA~~A~~ aZ.,1987; MARTORELL etal., 1989).

Table 3. Weight-for-height at 12-23 months and height-for-age at 24-35 months of age for children born in differ-
ent calendar periods
Mean weight-for-height Z Mean height-for-age
score at 12-23 months Z score at 24-35 months
Born during O-l 1 months 12-23 months
perioda old in March old in March Summerb Winterb
4185-3186 1986 1987 0.38 0.64 -1.15
4186-3187 1987 1988 0.33 0.65 -1.05
4187-3188 1988 1989 0.09 0.54 -1.33
4188-3189 1989 1990 0.22 0.45 -1.19
4189-3190
1990 1991 0.17 0.47 -1.12
“Month/last 2 digits of year.
bSummer defined asJanuary to June; winter asJuly to December.

Table 4. Per caput gross national product LGNP) and An investigation of body fat indicators such as skin fold
percentage change by year, Peru, 1986-1993 thicknesss and total body water in a similar population
indicated that Peruvian children were leaner than their
Per caput GNP Change in per caput North American counterparts, despite having weight-
Yearb (1986 US dollars) GNP (%) for-height above that of the reference population; thus
the same level of weight-for-height in different nonula-
1986 1249 +8.6 tions may not reflect Fhesamedggreeof leannessbr-obe-
1987 1352 +8.2
1988 sity (TROWBRIDGE et al., 1987). Nonetheless, variation in
1218 -9.9 weight-for-height did correlate with variation in lean-
1989 1046 -14.1
1990 ness within the population of Peruvian children (TROW-
1991 976 -6.7 BRIDGE et al., 1987).
1992 986 +41
In PRISMA’S 7 years of nutritional surveillance data,
1993 943 -4.3 children were significantly thinner in the summer of
1994 986 +4.5 each year; this effect was most marked among children
1100 +11.5 between 6 and 23 months of age. Despite the fact that
%ource: Instituto National de Estadistica e Informatica, Direc- the mean weight-for-height Z score remained near or
cidn National de Cuentas Nacionales, Lima, Peru. above the reference median, the findings of TROW-
bData for 1990-1994 are preliminary. BRIDGE et al. (1987) suggestthat this variation represents
a real difference in thinness. In fact, the difference in
tion higher in August-September than in March, it was weight-for-height by season (0.3-0.5 Z score) is of the
also seen in infants younger than 6 months and children same magnitude as that observed in an urban study in
aged 24-35 months. As assessedin multiple regression The Gambia (TOMKINS et al., 1986), and is about
models, in addition to the seasonalvariation, weight-for- 60%-70% of the magnitude of the variation seenin rural
height was estimated to decreaseby 0.01 units with each Bangladesh (BROWN et al., 1982).
month increase in age; it was 0.05 Z scores higher in Although we have no direct data to explain why nutri-
airls than in bovs and decreasedbv 0.02 Z scores with tional status varies by seasonin urban Peru, it is likely
each calendar year. However, even-after controlling for that a combination of factors is involved. Dietary intake
age, sex, and calendar year, there was still a strong sea- varies somewhat by season,with more fruits and vegeta-
sonal variation in mean weight-for-height with an over- bles but fewer energy-dense animal products in the-diet
all amplitude of 0.19Z scores. in the summer than in the winter (PRISMA, 1995). How-
The summer minimum in weight-for-height tended to ever, seasonal dietary variation in the periurban cash
be lower in 1989 than in othe; years, faliing below a economy is likely to be much less than that in agricul-
mean Z score of zero for children 12-23 and 24-35 tural rural areas; and morbidity, in particular diarrhoeal
SEASONALCHANGEINNUTRITIONALSTATUS
disease, probably plays an important role. Although no
marked seasonalitg was noted in overall diarrhoeal dis-
ease incidence iti a similar periurban area of Lima
(BLACK et al.. 1989). the incidence of diarrhoea associ-
ated with specific pathogens such as Shigella species and
Campylobacter jt?juni varies seasonally, with peaks in the
summer month; (BLACK et al., 1989; PRISMA, unpub-
lished data). The effects on growth differ according to
pathogen @LACK et al., 1984% a predominance of b&e-
rial pathogens in the summer may be associated with a
more marked effect on growth than, for example, rota-
virus, which tends to be less seasonal in Peru (PRISMA,
unpublished data). In addition, the age range at which
the most pronounced seasonal effect is seen (6-23
months) is the same as the age at which the incidence of
diarrhoea is at its height (LOPEZ DE ROMAI~A et al., 1987;
BERN et al., 1992).
The deeper summer trough in weight-for-height in
March 1989 occurred at a time when the economic situ-
ation of Peru’s poor communities was at a low ebb. In
late 1988, the combination of falling GNP, marked infla-
tion, and the withdrawal of subsidies for a series of sta-
ple food!, including wheat, rice, meat, sugar, and pow-
dered milk, caused the food purchasing power of the
poor to drop precipitously (PRISMA, 1989). After May
1989, some of these subsidies were reinstated, but the
econbmy has recovered substantially only since about
1993. During the oeriod Seotember 1988 to Mav 1989,
food access was suhstantiall; lower than usual. ‘ihe co:
hort of children in their second year of life during this
period were shorter in their third year of life than co-
horts of children either older or younger than they.
Those who were younger than 12 months in March 1989
were also shorter at 24-35 months than earlier or later
groups, but the difference was smaller; infants may have
been relatively protected by breastfeeding. Although our
data represent only an ecological association, they sug-
gest that decreased food access due to the economic diff-
culties of 1988-1989 may have caused a more marked fall
in weight-for-height in the summer of 1989, and that the
subsequent increased stunting may reflect a more de-
layed effect. The timing of this effect is consistent with
thk delay between weight gain and height gain seen in
someother studies(B~~wN etal.. 1982; WATERLOW, 1994).
Our results dociment a sig&ficar& seasonal effect c&
the nutritional status of urban children in Peru, consis-
tent with that found in other developing countries. We
cannot definitively identify the cause of the seasonal ef-
fect from available data. However, closer investigation of
the major diarrhoeal pathogens among young children,
and their effects on linear growth and weight gain, may
help to explain this phenomenon. These data suggest
that nutritional evaluations and interventions should be
designed bearing in mind the possibility that nutritional
status may vary significantly by season. In addition, the
changes in growth documented among children who
were younger than 2 years during the economic difficul-
ties of 1989-1990 suggest that preventive interventions
should be targeted to this age group during times of eco-
nomic stress.
References
Bairagi, R. (1980). Is income the only constraint on child nutri-
tion in rural Bangladesh? Bulletin of the World Health Organi-
zation, 58,767-772.
Bern, C., Martines, J., de Zoysa, I. & Glass, R. I. (1992). The
445
magnitude of the global problem of diarrhoeal disease:a ten-
year update. Bulletin of the World Health Organization, 70,
705-714.
Black, R. E., Brown, K. H. & Becker, S. (1984). Effects of diar-
rhea associated with specific enteropathogens on the growth
of children in rural Bangladesh. Pediatrics, 73,799-805.
Black, R. E., Lopez de Romafia, G., Brown, K. H., Bravo, N.,
Grados Bazalar. 0. & Kanashiro. H. C. (1989). Incidence and
etiology of infantile diarrhea and major rouies of transmis-
sion in Huascar, Peru. American Journal of Epidemiology, -. 129,
785-799.
Brown, K. H., Black, R. E. & Becker, S. (1982). Seasonal
changes in nutritional status and rhe prevalence of malnutri-
tion in a longitudinal study of young children in rural Bang-
Downloaded from https://academic.oup.com/trstmh/article/90/4/442/1866793 by University of Basel user on 21 January 2021
ladesh. AmericanJournal of Clinical Nutrition, 36,303-313.
Chen, L. C., Chowdhury, A. K. M. A. & Huffman, S. L. (1979).
Seasonal dimensions of energy protein malnutrition in rural
Bangladesh: the role of agriculture, dietary practices, and in-
fection. Ecology of Food and Nutrition, 8,175-187.
Cleveland, W. S. (1979). Robust locally weighted regression and
smoothing scatterplots.Journal of the American Statistical Asso-
ciation, 74,829-836.
Diggle, P. J., Liang, K. Y. & Zeger, S. L. (1994). Analysis of Lon-
gitudinal Data. New York: Oxford University Press.
Fernandez-Concha, D., Gilman, R. H. & Gilman, J. B. (1991). A
home nutritional rehabilitation programme in a Peruvian
peri-urban shanty town (pueblo joven). Transactions of the
Roval Societv of TroDical Medicine and HYeiene. 85809-813.
Lopez de Ro&&a, I??.,Brown, K. H. 8;-Blaik, k. E. (1987).
Health and growth of infants and young children in Huascar,
Peru. Ecology of Food and Nutrition, 19,213-229.
Martorell, R., Mendoza, F. S. & Castillo, R. 0. (1989). Genetic
and environmental determinants of growth in Mexican-
Americans. Pediatrics, 84,864-871.
McGregor, I. A., Rahman, A. K., Thompson, B., Billewicz, W.
Z. & Thomson, A. M. (1968). The growth of young children
in a Gambian village. Transactions of the Royal Society of Tropi-
cal Medicine and Hygiene, 62,341-352.
National Institute of Food and Nutrition [Ministrv of Plannine
and Coordination, Government of Bdlivia] (1981). Balk&
National Nutritional Status Suruey. Washington, DC: USAID.
PRISMA (1989). The probable effects-of the present economic crisis on
the status of young children in a peri-urban population. Lima,
Peru: PRISMA, Report to UNICEF, September 1989.
PRISMA (1995). Encuesta de consumo de alimentos en Pampas de
&n~UUn. LiIIIa,hXII:PRISMA.
Rowland, M. G. M., Cole, T. J. & Whitehead, R. G. (1977). A
quantitative study into the role of infection in determining
nutritional status in Gambian village children. BritishJournal
of Nutrition, 37,441-450.
Tomkins, A. M., Dunn, D. T., Hayes, R. J. & Bradley, A. K.
(19861. Seasonal variations in the nutritional status of urban
ham&an children. BritishJournal of Nutrition, 56,533-543.
Trowbridge, F. L. & Newton, L. H. (1979). Seasonal changes in
malnutrition and diarrhoeal disease among preschool chil-
dren in El Salvador. American 7ournal of Troaical Medicine and
Hygiene,28,135-141. I ’ z
Trowbridge, F. L. & Stetler, H. C. (1982). Results of nutritional
status surveillance in El Salvador. 1975-77. Bulletin of the
World Health Organization, 60,433440.
Trowbridge, F. L., Marks, J. S., Lopez de Romaba, G., Madrid,
S., Boutton, T. W. & Klein, P. D. (1987). Body composition
of Peruvian children with short stature and high weight-for-
height. II. Implications for weight-for-height as an indicator
of nutritional status. American Yawnal of Clinical Nutrition, 46,
411-418.
Waterlow, J. C. (1994). Relationship of gain in height to gain in
weight. European Journal of Clinical Nutrition, 48, supplement
1, S72-S74.
Received IS January 1996; revised 4 March 1996; accepted
publication 27 March I996
for