How did the collaboration come about?
In 2005, plans were made for a series of papers on maternal and child undernutrition
for publication in The Lancet. The second paper of the series aimed to describe the
long-term consequences of maternal and child undernutrition for health and human capital.
The focal person for this article—Cesar Victora from the Federal University in Pelotas,
Brazil—decided to bring together available long-term data from low- and middle-income
countries. He identified the five largest prospective birth cohort studies from these
regions, all of which had at least 15 years of follow-up and an initial sample size
of 2000 or more newborns. The principal investigators were approached, and all agreed
to join the writing team.
The five studies included the 1982 Pelotas (Brazil) Birth Cohort Study,
1
the Institute of Nutrition of Central America and Panama Nutrition Trial Cohort (INTC;
Guatemala),
2
the New Delhi Birth Cohort (India),
3
the Cebu Longitudinal Health and Nutrition Survey cohort(CLHNS; Philippines)
4
and the Birth-to-Twenty (Bt20; Soweto-Johannesburg, South Africa) cohort.
5
All the studies were population-based and started recruitment during gestation or
at delivery; they have long follow-up periods; their study populations experienced
high rates of maternal and/or child undernutrition, and all are currently undergoing
rapid demographic, nutritional and epidemiological transitions.
The experience of working together on the original paper, which was published in 2008,
6
was highly positive. This motivated Cesar Victora, on behalf of the principal investigators,
to apply for a research grant from the Wellcome Trust aimed at establishing a long-lasting
collaborative network among the five cohorts. With funding, the group named the Consortium
of Health-Orientated Research in Transitioning Societies (COHORTS), was formed and
a logo created (Figure 1).
Figure 1
COHORTS logo
What does the collaboration cover?
The overall objectives of the consortium include: (i) to strengthen the collaboration
among five of the largest and longest running birth cohort studies in low- and middle-income
countries, including capacity building among young scientists, epidemiologists and
statisticians in the cohort teams; (ii) to jointly produce high-quality scientific
evidence on the early origins of chronic diseases and human capital by analysing data
from five cohort studies from low- and middle-income countries; and (iii) to disseminate
the findings of the collaboration through scientific meetings and journal articles.
The collaboration has developed through three main phases:
Phase 1 (2006–07) included the preparation and publication of the original Lancet
article on the long-term consequences of maternal and child undernutrition.
6
Phase 2 (2007–09) was aimed principally at answering two broad research questions:
(i) What is the effect of weight gain during infancy and early childhood on adult
health and human capital? In particular, during which period does weight gain have
a stronger effect on each outcome? (ii) How does infant feeding (duration of breastfeeding
and age of introduction of complementary foods) relate to later health outcomes?
7–9
Phase 3 (2009–11) is directed at answering three main research questions: (i) What
is the association between the physical and social environment in early childhood
and adult indicators of health and capacity? (ii) What is the influence of the physical
and social environment in early childhood on growth in the first 2 years of life?
(iii) How does the physical and social environment in early childhood moderate the
relationships between growth in the first 2 years of life and adult indicators of
health and capacity?
All three phases have been supported by grants from the Wellcome Trust. Since the
beginning of the collaboration, a schedule of annual meetings has been held at the
cohort study sites—Brazil (2006), South Africa (2008) and India (2010)—or coincident
with the World Congress on Developmental Origins of Health and Disease—DOHaD (Australia
2007, Chile 2009). In addition to these whole-team meetings, smaller subgroup data
analysis and writing meetings also take place as needed, sometimes several times a
year.
From 2010, COHORTS is also being supported by a grant from the Bill and Melinda Gates
Foundation for secondary analyses of the data to address the following issues: preterm
birth, subsequent growth and human capital and health; healthy post-natal growth,
in particular by attempting to disentangle (i) the effects of rapid weight gain vs
rapid linear growth during early life on adult human capital and health, as well as
consequences of rapid subsequent weight gain in relation to stunting at 2 years and
(ii) the relationship between early growth failure, delayed menarche and attained
final stature; inter-generational influences on child growth and its consequences,
specifically the differential impact of poverty and low parental schooling on growth
before and after the age of 2 years, and estimating the inter-generational constraints
on birth size and infant growth; and lastly, breastfeeding duration and achieved schooling.
Who is in the sample?
The 1982 Pelotas (Brazil) Birth Cohort study originally included all 5914 children
born in the city’s hospitals during the 1982 calendar year. At that time, >99% of
all deliveries in the city took place at hospitals. Because the sample was representative
of the city’s population, it included participants from all social classes.
1
The INTC study in Guatemala was originally an intervention trial in four villages.
All children <7 years in 1969, and all children born between 1969 and 1977, were enrolled
(N = 2392) and followed up until 7 years of age or until the end of the study in 1977.
2
For the COHORTS analyses, only children enrolled at birth were included in the analyses,
irrespective of whether they were born in the intervention or control villages (a
dummy variable for treatment arm was included in all analyses).
The New Delhi Birth Cohort in India recruited all married women living in a defined
area of the city from 1969 to 1972. Pregnancies were identified and the newborns were
enrolled (N = 8181) and followed up. The cohort was primarily middle class.
3
The CLHNS in the Philippines recruited pregnant women living in 33 randomly selected
neighbourhoods of the metropolitan Cebu area in 1983 and 1984. Data collection started
at the 30th week of gestation. Cohort members (N = 3080) were from all social classes.
4
Finally, the Bt20 study in South Africa identified pregnant women with gestational
age of 26–40 weeks living in the largest urban area (Soweto-Johannesburg) in 1990.
The sample enrolled (N = 3273) was predominantly Black and generally poor.
5
How often have cohort members been followed up?
Table 1 shows all waves of follow-up for each site included in the COHORTS collaboration.
All sites have some data available at birth, 1 and 2 years of age, mid-childhood and
late adolescence or adulthood. In the last follow-up visits (2009), participants had
an average age of 23 years in Brazil, 32.7 years in Guatemala, 36 years in India,
21.5 years in the Philippines and 18.5 years in South Africa.
Table 1
Data collection waves in the individual studies in the COHORTS
Country
Enrolment
Follow-up assessments
Brazil
Birth (1982)
2, 4 and 23 years; subsamples were seen at 1, 15, 18, 19 and 26 years of age
Guatemala
Birth (1969–77)
11–19, 25–30, 25–35, 28–38 years
India
Conception (1969–72)
Birth to 1 year—every 3 months
1–20 years—every 6 months
29–32 years
37–40 years
The Philippines
Pregnancy (1983–84)
Birth to 2 years—every 2 months
8, 11, 15, 19 and 22 years
South Africa
Pregnancy (1990)
Birth and 6 months
1, 2, 3, 4, 5, 7.5, 11.5, 13, 14, 15, 16, 17.5, 18.5 and 19.5 years
What has been measured?
Detailed lists of what has been measured in each cohort are available elsewhere.
1
,
2
,
5
,
10
,
11
We provide here key variables that are common in most if not all cohorts.
Growth-related variables
Information on birthweight is available for all sites. Birth length is available for
three of five sites (Delhi, Guatemala and Cebu). Data on weight and length/height
are available for all cohorts throughout infancy and childhood. In the late adolescence
or adult visits, all cohorts collected data on weight and height and some indicator
of body composition.
Early-life socio-economic, infrastructure and environment variables
All cohorts except Delhi allow calculation of household asset scores in early life
(e.g. TV, fridge, car, radio and telephone). Information on maternal and paternal
education is available for all sites, as are a range of variables related to family
income and paternal employment. All studies have common data on basic services (e.g.
water, electricity and toilets), and all sites have information on the size of the
household and the quality of building materials used.
Human capital
Socio-economic indicators continued to be collected in all sites, from initiation
and throughout adolescence and adulthood. Information on schooling attainment is available
for all sites, and individual income at follow-up is available in three of the five
sites (Pelotas, Guatemala and Delhi).
Behavioural variables
Feeding patterns in infancy were collected in all sites. In three of the sites, data
are available on the age of first alcohol intake. All sites have data on smoking and
physical activity, either in adolescence or adulthood. In addition to variables on
the behaviour of cohort members, some information is available about their parents
(e.g. maternal smoking during pregnancy).
Chronic disease-related variables
In late adolescence or adulthood, all sites have data on blood pressure, plasma/blood
glucose concentrations, lipids and body composition.
What was attrition like?
The current COHORT analyses are based on close to 11 000 individuals. The number of
people not traced into adulthood varies from 1% in the Philippines to 17% in Pelotas.
Other reasons for the loss of participants are known deaths and established movements
out of the study area. Table 2 presents attrition information by site.
Table 2
Attrition information by site
Attrition information
Brazil
Guatemala
India
The Philippines
South Africa
Original cohort size
5914
2392
8181
3080
3273
Known to have died
282
274
590
237
47
Moved out of study area
334
162
a
3918
724
292
b
Not traced
1001
101
1311
24
441
c
63
d
Sample size available for main COHORTS analyses
4297
1571
e
1526
f
(2002)
2032
1869
g
1149 (2009)
aOne hundred and sixty-two left Guatemala.
bAn additional 498 moved to an unknown address, either within or outside the study
area.
cThe cumulative attrition >20 years for all reasons was 961, of which 441 were deemed
as ‘not traced’, having only been seen once in the first year of the study and never
contacted again.
12
dRefused, institutionalized, physically/mentally ill, pregnant in 2002 or 2005, invalid
baseline/birth data.
eOne thousand five hundred and seventy-one completed at least one interview in the
2002–04 study.
fTwo thousand three hundred and sixty-two available; numbers analysed based on consent
and eligibility.
gAt the age of 18/19 years, 1869 had data collected, 2312 had true contact.
What has COHORTS collaboration found to date?
Phase 1
The group carried out a comprehensive literature review on the long-term consequences
of 12 exposure variables on 14 adult outcomes, and conducted meta-regression analyses
from the five cohorts on the effects of six-exposure variables on seven adult outcomes.
6
Our main findings were (i) intra-uterine growth retardation and stunting in the first
years of life is associated with shorter adult height, lower attained schooling, reduced
adult income and lower offspring birthweight; (ii) there were no associations between
anthropometric status at 2 years of age and the risk of chronic disease, even in children
with intrauterine growth retardation; (iii) children who are undernourished in early
life and then gain weight rapidly after infancy are at high risk of chronic diseases.
The findings suggest that the prevention of maternal and child undernutrition is a
long-term investment that will benefit both the current generation and their children.
Phase 2
The second set of analyses investigated associations between weight gain and linear
growth (first year of life vs second year vs 2 years in early to mid-childhood) and
two sets of outcomes measured among young adults. The first set included risk factors
for complex chronic diseases, including blood pressure, body composition and glucose.
The second set included human capital outcomes, namely achieved schooling and adult
height. All analyses used conditional growth modelling in order to remove the correlation
between growth measured in different age ranges.
13
Our main findings so far are the following:
Schooling
8
: growth during the first 2 years of life was found to be the most critical factor
for schooling outcomes, followed by status at birth; but growth from 2 to 4 years
has little relationship to schooling outcomes. Stunting was associated with a reduction
in attained schooling of 1.8 and 0.9 years before and after controlling for confounding,
respectively. In fully adjusted models, one z-score of birthweight, weight gain between
0 and 24 months and 24–48 months were each associated with 0.21, 0.43 and 0.07 additional
years of schooling, respectively.
Height
9
: the five cohorts differed in their patterns of child growth in length/height, with
Guatemala and Cebu showing marked growth failure in early childhood, Pelotas showing
growth quite similar to international standards and South Africa and India showing
intermediate patterns. Nonetheless, across the five cohorts, linear growth failure
was associated with reduced adult height. In the cohorts with measured length at birth
(Guatemala, India and Cebu), linear growth failure during the period from 0 to 12
months was most strongly associated with attained height, whereas growth failure in
the periods 12–24 months or from 24 months through to mid-childhood was not associated
with attained adult height.
Blood pressure
7
: weight gain in childhood was associated with increased blood pressure and the risk
of pre-hypertension and hypertension, and this association was mediated through adult
body mass index (BMI). After adjustment for adult height and BMI, systolic blood pressure
and risk of hypertension and pre-hypertension were inversely associated with birth
weight, but not with any of the conditional weight gain measures. Relationships were
similar in adults who had been born small for gestational age and those with normal
birth weight. These analyses suggest that rapid weight gain in infancy does not pose
a higher risk for elevated blood pressure than gains at other ages.
Long-term consequences of breastfeeding: there were no differences in outcomes related
to chronic disease between children whose mothers initiated breastfeeding compared
with children who were never breastfed.
14
Associations with duration of breastfeeding were most U-shaped and of small effects.
Children who started complementary foods later in infancy were less adipose in adult
life. These associations remained significant after adjusting for the confounding
factors, indicating that delayed introduction of complementary foods until child age
of 6 months may protect against later overweight.
Analyses are underway on the association between early growth and later glucose and
body composition. A paper on the methods used in the DOHaD field is also under preparation.
Taken together, the series of analyses on the timing of early weight and/or height
gains show that children from low- and middle-income countries with greater gains
in the first and possibly in the second year of life show improved human capital indicators
as adults, and no evidence of increased risk of chronic diseases in later life.
What are COHORTS' main strengths and weaknesses?
The main strengths of the collaboration include: (i) the possibility of confirming
individual-site findings by verifying whether they are also observed in other low-
and middle-income contexts; (ii) the enhanced statistical power obtained by pooling
five data sets; (iii) the high prevalence of maternal and child undernutrition in
our cohorts as compared with studies from high-income countries, which allows studying
their long-term consequences; (iv) for selected exposures and outcomes, the different
patterns of confounding as compared with high-income country cohorts (e.g. breastfeeding
being more common among the poor in most of the five cohorts, or overweight/obesity
being more common among the better off); and (v) the avoidance of publication bias
by agreeing a priori on which analyses are to be undertaken and jointly publishing
the results for the five cohorts.
Some of the weaknesses include: (i) differences in variable definitions, or in measurement
techniques across sites (this affects exposures, outcomes and confounding variables),
which means that major effort has gone into producing a common data set; (ii) the
different ages of individuals across the five cohorts and the different time periods
they reflect; (iii) the different ages for which data are available throughout infancy
and childhood; (iv) heterogeneity in the results for some of the analyses, for example,
those on body composition. In order to overcome these limitations, we have restricted
potential analyses to those including variables collected consistently across the
cohorts; likewise, we have limited the analyses to ages with data available for all
(or most) cohorts. In some analyses, different outcome variables have been used (e.g.
pre-hypertension for adolescents and hypertension for adults)
7
because of the different ages of the individuals across cohorts.
Can I get hold of the data? Where can I find more?
Individual sites have extensive collaborations with groups outside COHORTS. Further
information is available on the websites of the individual birth cohort projects (Table
3). We are currently developing a website for the collaboration and anticipate this
being accessible by January 2011. The contact details of the principal investigators
and individual study websites are shown in Table 3. For further information, please
e-mail LRichter@hsrc.ac.za (principal investigator of the current Wellcome Trust grant),
prchallal@gmail.com (data manager) or aryeh.stein@emory.edu (coordinator of the Gates
Foundation funded analyses).
Table 3
Contact information for each study in the COHORTS collaboration
Site
Principal investigator for the COHORTS collaboration
Email
Website
Pelotas (Brazil)
Cesar G Victora
cvictora@gmail.com
http://www.epidemio-ufpel.org.br/_projetos_de_pesquisas/coorte1982/
NTCS (Guatemala)
Reynaldo Martorell
rmart77@sph.emory.edu
http://www.sph.emory.edu/humancapital/index.php
New Delhi (India)
Santosh Bhargava
dskbhargava@yahoo.com
No website
Cebu (The Philippines)
Linda Adair
linda_adair@unc.edu
http://www.icpsr.umich.edu/cocoon/DSDR/STUDY/00178.xml
Bt20 (South Africa)
Linda Richter
LRichter@hsrc.ac.za
http://web.wits.ac.za/Academic/Health/Research/BirthTo20/
Funding
The Wellcome Trust for funding the COHORTS collaboration; Bill and Melinda Gates Foundation
for additional COHORTS analyses; The Wellcome Trust (Pelotas and Soweto); US National
Institutes of Health and the US National Science Foundation (Guatemala); British Heart
Foundation, the Medical Research Council UK, and the Indian Council of Medical Research
(India); Human Sciences Research Council, South African Medical Research Council,
the Mellon Foundation, the South-African Netherlands Programme on Alternative Development
and the Anglo American Chairman’s Fund (Soweto/Johannesburg); Ford Foundation, USAID,
The World Bank, Nestle Coordinating Center for Nutrition Research (Cebu).