Published by Oxford University Press on behalf of the International Epidemiological Association
ß The Author 2013; all rights reserved. Advance Access publication 7 August 2013
International Journal of Epidemiology 2013;42:1390–1398
doi:10.1093/ije/dyt116
The causal effect of malaria on stunting:
a Mendelian randomization and matching
approach
Hyunseung Kang,1y Benno Kreuels,2,3y Ohene Adjei,4 Ralf Krumkamp,3 Jürgen May3 and
Dylan S Small1*
1
Department of Statistics, The Wharton School, University of Pennsylvania, Philadelphia, PA, USA, 2Section for Tropical Medicine,
I. Department of Internal Medicine, University Medical Center Eppendorf, Hamburg, Germany, 3Research Group Infectious Disease
Epidemiology, Bernhard Nocht Institute for Tropical Medicine, Hamburg, Germany and 4Kumasi Center for Collaborative Research
in Tropical Medicine, Kumasi, Ghana
*Corresponding author. Department of Statistics, The Wharton School, University of Pennsylvania, 3730 Walnut Street,
Philadelphia, PA, USA. E-mail: [email protected]
y
These authors contributed equally to this work.
Accepted
21 May 2013
Background Previous studies on the association of malaria and stunted growth
delivered inconsistent results. These conflicting results may be due to
different levels of confounding and to considerable difficulties in
elucidating a causal relationship. Randomized experiments are impractical and previous observational studies have not fully controlled
for potential confounding including nutritional deficiencies, breastfeeding habits, other infectious diseases and socioeconomic status.
Methods
This study aims to estimate the causal effect between malaria episodes and stunted growth by applying a combination of Mendelian
randomization, using the sickle cell trait, and matching. We demonstrate the method on a cohort of children in the Ashanti Region,
Ghana.
Results
We found that the risk of stunting increases by 0.32 (P-value:
0.004, 95% CI: 0.09, 1.0) for every malaria episode. The risk estimate based on Mendelian randomization substantially differs from
the multiple regression estimate of 0.02 (P-value: 0.02, 95% CI:
0.003, 0.03). In addition, based on the sensitivity analysis, our results were reasonably insensitive to unmeasured confounders.
Conclusions The method applied in this study indicates a causal relationship
between malaria and stunting in young children in an area of
high endemicity and demonstrates the usefulness of the sickle
cell trait as an instrument for the analysis of conditions that
might be causally related to malaria.
Keywords
Malaria, stunting, children, Mendelian randomization, matching
Introduction
It is estimated that there were 174 million cases of
malaria and 596 000 malaria-related deaths in subSaharan Africa in 2010.1 In addition to being one of
the major causes of death in early childhood, repeated
malaria episodes are a major cause of chronic
anaemia and may impair child growth and development.2 Consequently, it is important to study the
impact of malaria on child development to prioritize
public health resources.
Previous epidemiological studies on the association
between malaria and child growth have produced
1390
THE CAUSAL EFFECT OF MALARIA ON STUNTING
inconsistent results; this is partly rooted in different
methodological approaches. Several studies assessed
growth using the mean height-for-age Z-score,
whereas other studies used the prevalence of stunting
(height-for-age Z-score < 2) as an indicator of insufficient growth. Stunting is a common condition in
African children and is often used to assess chronic
malnutrition, which is one of the main determinants
of childhood morbidity and mortality.3 In 1956 a study
from Gambia first showed a tendency to higher mean
Z-scores in infants who received malaria prophylaxis
compared with children who did not.4 Later an association between malaria and growth or risk of stunting
was also seen in Nigeria,5 Kenya,6 Gambia,7 Ghana8
and Uganda.9 Other studies, however, found no association10–13 or even a higher risk of malaria in children
with better Z-scores.14 Finally, one study demonstrated
that the association between stunting and malaria
might be strongest in young children.15
A major limitation common to all previous studies is
the inability to fully adjust for confounding.
Specifically, nutritional deficiencies are important potential confounders because they are an important
determinant of stunting and also compromise
immune function, which could result in a higher
risk of infection.11 Further potential confounders are
socioeconomic status, living conditions and other infections. In addition, reverse causality in the association of stunting and malaria seems possible.
Randomized trials recruiting children at birth could
account for potential confounders and reverse causality but are impractical in this context.
In this paper, we seek to control for confounders in
estimating the causal effect of malaria on stunting by
using a combination of Mendelian randomization
(MR) and matching.16 The basic idea of MR is to extract variation in an exposure (i.e. malaria) that is
due to a gene, which is independent of confounders,
and use this confounder-free variation to estimate the
effect of the exposure on the outcome (i.e. stunting).17–22 The haemoglobin variant HbS, which is
caused by a point-mutation at the sixth position of
the b-Globin gene (b6Glu ! Val), serves as the paradigm for balanced polymorphisms; whereas people
homozygous for HbS (HbSS) have sickle cell disease
with an increased mortality, heterozygote carriers
(HbAS, sickle cell trait) are asymptomatic and protected from malaria.23,24 A previous analysis of the
current data showed a negative association between
the HbAS genotype and stunting in an area of
high malaria endemicity and computed the magnitude of the association.25 However, the study did
not analyse the effect of malaria on stunting or the
magnitude of such an effect. In this analysis we use
HbAS as a Mendelian gene to expand on this finding
and estimate the effect of malaria on stunting. To
control for measured confounders (e.g. birthweight,
ethnic group, mosquito protection) we will use
matching.
1391
Methods
Study population and design
The study was conducted in the Ashanti region in
Ghana. A cohort of 1070 infants was recruited as
part of a clinical trial on intermittent preventive
treatment with sulphadoxine-pyrimethamine (SP).26
Infants were recruited at 3 months of age and followed up monthly until age 2 years with comprehensive examinations including a standardized medical
history, a measurement of body temperature and a
thick-and-thin smear for microscopic malaria diagnostics. Passive case detection was performed between scheduled visits. A child was diagnosed with
malaria if he or she had a parasite density of more
than 500 parasites/ml and a body temperature greater
than 388C or the mother reported a fever within the
last 48 h. In 3-monthly intervals, standardized anthropometric measurements, including height and
weight, were performed. A child was deemed stunted
if his or her length/height-for-age Z-score was less
than 2 (i.e. moderate or severe stunting).27 Further
details of the study population are published in a previous paper.26
Definition of instrument, exposure and
outcome
For this analysis only infants with heterozygote HbAS
or wildtype HbAA were considered. Children with
homozygote mutation (HbSS) or a different mutation
on the same gene leading to haemoglobin C (HbAC,
HbCC, HbSC) were excluded. The instrument was a
binary variable indicating the HbAS or HbAA genotype. The exposure of interest was the malarial history
defined as the total number of malaria episodes
during the study. The outcome of interest was
whether the child was stunted at the last recorded
visit, which took place when the child was approximately 2 years old.
Assumptions for Mendelian randomization
A combination of prior biological and clinical evidence
along with empirical methods was used to assess the
sickle cell trait as a valid instrument for MR. In particular, the sickle cell trait must satisfy the following
assumptions (Figure 1) to be a valid instrument and
to provide an unbiased estimate of the causal effect
between malaria and stunting28: (1) the sickle cell
trait is associated with malaria episodes; (2) there
are no unmeasured confounders that are associated
with the sickle cell trait and stunted growth; and
(3) all directed pathways from the sickle cell trait to
stunting pass through malaria episodes (i.e. there is
no pathway that goes directly from the sickle cell
genotype to stunted growth).
Assumption (1) states that the sickle cell trait must
be associated with malaria episodes. A strong association is preferred between these two variables because
it leads to lower-variance estimates of the causal
1392
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
Confounders
(2)
Sickle
cell trait
(1)
Malaria
episodes
Stunting
African American children,39,40 as well as two studies
among children from Dominica and Jamaica,41,42 for
whom the sickle cell trait is not uncommon but who
live in a non-malaria-endemic area, found no evidence of differences in physical and cognitive development between children with HbAS and HbAA. This
provides support for Assumptions (2) and (3).
(3)
Figure 1 Causal diagram for the MR study. Arrows represent associations between variables. Absence of arrows between two variables indicates no relationship between them.
Numbers (1, 2, 3) represent MR assumptions and dashed
arrows indicate possible violations of these assumptions as
outlined in the methods section
effect and makes estimates less sensitive to bias.29,30
A review of previous literature found numerous
clinical studies, which suggested that HbAS was associated with 90% protection against severe and 20–30%
protection against mild malaria,23,31–34 with some attempts to find biological pathways.35–38 Based on this
clinical and biological evidence, we expect that HbAS
is negatively associated with malaria episodes and
that the association is not small in size. We performed
a two-sided t-test to check whether the time of the
last visit differed among those with different genotypes and used a standard Poisson regression, with
the number of malaria episodes as an outcome, to
analyse the association between HbAS and malaria.
Assumption (2) could be violated if, for example,
population stratification or linkage disequilibrium is
present, the two most common biological mechanisms
for violation of this assumption. Population stratification is a condition where some subpopulations are
more likely to have HbAS, and some are more likely
to be stunted through mechanisms other than malaria. An example for this is the ethnicity of study
participants. The genetic background may differ by
ethnic group (i.e. the chances of inheriting HbAS
may differ) and at the same time the socioeconomic
background may also differ leading to a different risk
of stunting. For linkage disequilibrium, there might
be a gene that causes stunting and that is in linkage
disequilibrium with the sickle cell trait. We will control for all measured covariates through matching to
mitigate this concern. To measure the potential effect
of unmeasured confounders, we conducted a sensitivity analysis. Assumption (3) could be violated if the
sickle cell trait had effects on stunting other than
through causing malaria (i.e. if the sickle cell trait
was pleiotropic).
A combined test of Assumptions (2) and (3) is provided by looking at whether the sickle cell trait is
associated with stunting among people living in
non-malaria-endemic areas. In non-malaria-endemic
areas, the arrow between sickle cell trait and stunted
growth in Figure 1 is removed, so that under
Assumptions (2) and (3) the sickle cell trait should
have no effect on stunting. Two studies among
Matching
To control for potential confounders of the sickle cell
trait-stunting relationship, one child with HbAS was
matched with five children with HbAA, based on
characteristics that were previously found to be associated with malaria risk and could therefore act as
potential confounders (Table 1). Specifically, we
matched for all measured covariates, which are birthweight, sex, birth season, ethnic group, presence of
alpha-thalassaemia, village of birth, mother’s occupation, mother’s education, family’s financial status,
mosquito protection and treatment arm of the original
trial (SP vs placebo). We used propensity score caliper
matching with rank-based Mahalanobis distance to
measure covariate similarity between children.43,44
Propensity scores were calculated by logistic regression. The propensity score here was on an instrumental propensity score, which is the probability of having
the sickle cell trait given the measured confounders
(J. Cheng. Using the instrumental propensity score in
observational studies for causal effects. Unpublished
presentation. Joint Statistical Meeting, American
Statistical Association, 3 August 2011). Children
with missing values in the covariates were matched
to other children with similar patterns of missing
data.43 Once the measurement of covariate similarity
was calculated, the matching algorithm matched each
child with HbAS to five children with HbAA in such a
way that their covariates are similar.
Estimation of effect ratio
We estimated the effect ratio, which measures the
effect of a change in malaria episodes on the risk of
stunting. For example, if the sickle cell trait were to
reduce malaria episodes by 0.5 per child and this resulted in a reduction of stunting by 0.05, the effect
ratio would be 0.05/0.5 ¼ 0.1. Under the assumption
that the effect of changing the rate of malaria episodes per child by A is proportional to the effect of
HbAS that reduces malaria episodes by B, i.e. the
effect of changing the rate of malaria episodes per
child by A is C(A/B), where C is the effect of the
sickle cell trait on stunting, we can interpret such
an effect ratio of 0.1 as meaning that for every 10
malarial infections, one of them would lead to
stunted growth. For a discussion of when such a proportionality assumption is justified, refer to these
studies.45,46
The effect ratio was estimated with a modified signscore statistic. The distribution of this statistic is
approximated by its asymptotic distribution (see
THE CAUSAL EFFECT OF MALARIA ON STUNTING
1393
Table 1 Characteristics of study participants at recruitment
HbAS (n ¼ 110)
HbAA before
matching (n ¼ 774)
HbAA after
matching (n ¼ 550)
Birth
Birthweight, mean (SD)
Sex (male/female)
3112.44(381.9)
2978.7(467.9)***
46.4% Male
51.0% Male
3065.2(390.1)
51.1% Male
Birth season (dry/rainy)
56.4% Dry
55.3% Dry
55.6% Dry
Ethnic group (Akan/Northerner)
86.4% Akan
88.8% Akan
90.9% Akan
a-globin genotype (norm/hetero/homo)
75.7%/21.5%/2.8%
74.4%/23.1%/2.6%
76.7%/20.9%/2.4%
Village composition
Village of birth:
Afamanso
4.5%
4.8%
4.4%
Agona
10.0%
13.6%
11.8%
Asamang
13.6%
11.1%
12.72%
Bedomase
5.5%
4.5%
5.1%
Bipoa
14.5%
10.7%
13.3%
Jamasi
15.5%
13.8%
14.0%
Kona
16.4%
12.8%
15.1%
Tano-Odumasi
Wiamoase
4.5%
12.3%**
15.5%
16.4%
5.6%
18.0%
Mother and household
Mother’s occupation (non-farmer/farmer)
79.0% Non-farmer
77.7% Non-farmer
81.1% Non-farmer
Mother’s education (literate/illiterate)
91.7% Literate
90.5% Literate
95.1% Literate
Family’s financial status (good/poor)
69.1% Good
70.1% Good
70.9% Good
Mosquito protection (none/net/screen)
55.7%/32.0%/12.4%
45.4%*/35.1%/19.5%
50.8%/35.2%/14.0%
Other
Treatment group
(sulphadoxine-pyrimethamine/Placebo)
49.1% Placebo
50.1% Placebo
48.4% Placebo
We conducted hypothesis tests to examine the differences between children with HbAS and HbAA for each covariate. P-values were
obtained by doing a Pearson’s chi-square test for categorical covariates and two-sample t-tests for numerical covariates. For the
covariate a-globin genotype, norm indicates wildtype, hetero indicates a+-thalassaemia, and homo indicates homozygote a+thalassaemia
***P < 0.01; **P < 0.05; *P < 0.1. Before matching, some characteristics between HbAS and HbAA show P-values < 0.10.
After matching, all the characteristics between HbAS and HbAA are not statistically different from each other.
Supplementary data Section 1.3, available as
Supplementary data at IJE online). We used this
asymptotic distribution to obtain inference about the
effect ratio, compute its power under the current
sample size and derive confidence intervals. For details on the power calculation see Supplementary Data
Section 1.5 (available as Supplementary data at IJE
online). For comparison, we computed the multiple
regression estimate of the effect ratio, an estimate
that only adjusts for measured confounding, not for
unmeasured confounding. This estimate is derived
from a multiple linear regression with stunting as
the dependent variable and all measured confounders
and the number of malaria episodes as independent
variables. From the regression, we take the estimated
slope coefficient for malaria episodes, which is the
reduction in the risk of stunting per malaria episode.
Sensitivity analysis
To quantify the effect of unmeasured confounders on
the obtained inference, a sensitivity analysis was performed. We used a standard sensitivity analysis for
multiple controls with a sign-score statistic and its
interpretation in higher dimensions.44,47 Specifically,
we consider a binary unmeasured confounder that
1394
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
has a specified effect on the odds of inheriting HbAS
over HbAA and specified effect on the odds of stunting (conditional on measured confounders), and
evaluate the effect such an unmeasured confounder
would have on the inference we make (see Supplementary data Section 1.4, available as Supplementary
data at IJE online).
Results
Basic data
The analysis was conducted on 884 children with
HbAA or HbAS genotype: 774 children were HbAA
homozygotes; 110 children were HbAS heterozygotes;
35 children (4.0%) were already stunted at the beginning of the trial and, by the end, 168 children (19.0%)
were stunted. The t-statistic to test the difference in
the time of the last recorded visit amongt HbAA and
HbAS children did not indicate any variation
[P ¼ 0.21, 95% CI: (-3.70, 16.68)].
Table 1 shows characteristics of the HbAS and HbAA
subjects. The first two columns of numbers compare
the HbAS and HbAA groups before matching. Before
matching, most characteristics at recruitment were
similarly distributed between children with HbAS
and HbAA. The notable exception is birthweight.
There was evidence that birthweight of children
with HbAA was lower than that of children with
HbAS [P ¼ 0.006, 95% CI: (-228.27, -39.14)].
We characterize the effect of the sickle cell trait on
malaria as based on a Poisson regression. As expected,
the sickle cell trait was protective against malaria. The
HbAS group had fewer malaria episodes than the
HbAA group before matching (RR ¼ 0.82, P ¼ 0.02,
95% CI: 0.70, 0.97) and after matching (RR ¼ 0.82,
P ¼ 0.03, 95% CI: 0.70, 0.97). This is in alignment
with previous literature on the relationship between
sickle cell genotype and malaria for these data.24
Matching
The first and third columns of numbers in Table 1
compare the HbAS and HbAA groups after matching.
After matching, all confounders between children
with HbAA and HbAS were similarly distributed.
Specifically, after running a simple hypothesis test
for each covariate, with the null of no difference between HbAA and HbAS, all the P-values were well
above 0.10.
Effect ratio
Using a multiple regression analysis that does not
control for unmeasured confounders, the effect ratio
was estimated at 0.02 (P ¼ 0.02, 95% CI: 0.003, 0.03),
indicating only a weak association between malaria
and stunting (Table 2).
When the sickle cell trait is used to control for unmeasured confounding, the effect ratio was estimated
at 0.32 (P ¼ 0.004, 95% CI: 0.09, 1). That is, for every
Table 2 Estimates of the causal effect using MR compared
with multiple regression. The strong difference between
estimates from MR and multiple regression analysis (adjusted for measured confounders) for an association between malaria and stunting indicates a high level of
unmeasured confounding in the multiple regression analysis
Methods
MR with matching
Estimate
0.32
P-value
0.004
95%
Confidence
Interval
(0.09, 1)
Multiple Regression
0.02
0.02
(0.003, 0.03)
unit increase in malaria episodes, there is an average
0.32 increase in the risk of stunting.
Sensitivity analysis
Our analysis indicated that malaria causes stunting
under the assumptions described in the methods
section. The results of the sensitivity analysis, which
we performed to assess the sensitivity of our results
against a violation of Assumption (2) that HbAS must
be independent of unmeasured confounders, is presented in Figure 2.
The plot shows the potential effect of a binary unmeasured confounder on the odds of stunting (x-axis)
and the odds of inheriting HbAS (y-axis). Any points
between the curves represent an unmeasured confounder whose effect on stunting (x-axis) and the
sickle cell trait (y-axis) does not change the inference
that malaria causes stunting. The bold curves represent the boundaries for which the inference that malaria causes stunting would no longer hold due to
unmeasured confounders. For example, the point
(2.0, 1.75) represents an unmeasured confounder
that increases the odds of stunting by a factor of 2
and increases the odds of inheriting HbAS over HbAA
by a factor of 1.75; this point is between the two bold
curves and its associated P-value is between 0.025 and
0.05, indicating strong evidence that malaria causes
stunting even if such an unmeasured confounder
existed. On the other hand, an unmeasured confounder associated with increasing the odds of stunting by
a factor of at least 1.75 and increasing the odds of
inheriting HbAS over HbAA by a factor of at least 2.5
lies outside the bolded curves, indicating at most
weak evidence for an effect of malaria on stunting.
Discussion
By using MR with the sickle cell trait as the instrument and matching techniques to account for potential confounders, we found evidence of a causal effect
of malarial episodes on stunting. Each increase by one
malaria episode increased the risk of stunting by 0.32
(95% CI: 0.09, 1), indicating that the effect of malaria
on stunting is substantial in our cohort of infants
under 2 years of age.
THE CAUSAL EFFECT OF MALARIA ON STUNTING
1395
Figure 2 Sensitivity analysis. Each point between two curves corresponds to effects by unmeasured confounders that result
in P-values specified by the legend. Points within the two bold curves correspond to effects by unmeasured confounders
where our inference will give us P-values < 0.05, meaning there is strong evidence that malaria causes stunting even in the
presence of the confounders. Any points outside the two bold curves correspond to effects by unmeasured confounders that
result in a P-value 40.05
Our results confirm findings about an association
between malaria and stunting from previous studies7–9 as well as findings from earlier studies on an
association between mean height-for-age Z-scores and
malaria,4–6 Previous studies were unable to fully
adjust for confounding; a large number of personal
characteristics, such as nutritional deficiencies, low
socioeconomic status and poor living conditions, are
likely to be predictors for both malaria and stunting.
Differing levels of confounding in previous studies
may have led to findings of no association between
malaria and stunting or mean Z-scores10,12,13 or a
negative correlation14 or false conclusions about
associations.
In our study, the large difference between the estimate for the effect ratio from the multiple regression
and the estimate derived after matching (0.02 vs.
0.32) indicates a substantial level of confounding in
the multiple regression. MR takes into account unmeasured confounders that are frequently present in
observational studies and are not controlled for in
standard regression. Under the assumptions stated
in the methods section, MR will control for both unmeasured and measured confounding and provide an
unbiased estimate.19,20,28 The necessity of these assumptions is a potential limitation that is inherent
in our approach. However, we are convinced that
the assumptions of an association between HbAS
and malaria23,31–34 and no association between
HbAS and stunting other than through malaria39-42
are valid.
Ghanasah et al. have described the HbAS haplotype
in a Ghanaian population as an extended haplotype
of 1.5 Mb containing 25 additional genes.48 Their analysis shows that this genomic region has a considerable degree of linkage disequilibrium, which
potentially could violate our assumption that HbAS
is independent of unmeasured confounders. To identify a potential violation, we searched PubMed for
reports on associations between stunting or
1396
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
malnutrition and any of the other 25 genes on the
extended haplotype, including possible alternative
gene names, allelic variants and resulting phenotypes,
based on searches in the National Center for
Biotechnology Information (NCBI) gene database
and the Online Mendelian Inheritance in Man database (OMIM) (see Supplementary data Section 2,
available at IJE online). These searches did not
reveal any reports of an association between genes
or genetic variants on the haplotype and stunting.
In addition, the sensitivity analysis showed that the
estimate of the effect ratio was robust with respect to
potential effects by unmeasured confounders. For example, a gene that was in linkage disequilibrium with
the sickle cell trait and caused a doubling in odds of
inheriting the sickle cell trait would have to at least
increase the odds of stunting by a factor of 1.75 to
overturn our conclusion. Such an effect of a gene on
stunting would be large and surprising. In fact, for
any unmeasured confounder to have effects specified
in the region above the bolded curve in Figure 2 is
very unlikely and, hence, we believe our result is insensitive to unmeasured confounders.
A further limitation to previous studies is potential
reverse causality in the association of stunting and
malaria. As discussed by Arinaitwe et al, it is difficult
to distinguish whether stunting increases the risk of
malaria or whether malaria increases the risk of
stunting.9 The MR design of this study solves part
of this limitation. It enables us to see whether an
increased frequency of malaria causes stunting.
Specifically, any association between the sickle cell
trait and stunting must come from an effect of malaria on stunting rather than the reverse. The sickle
cell trait, which is determined at conception, only affects stunting through its effect on malaria. If malaria
did not affect stunting, there would be no association
between the sickle cell trait and stunting.
However, there are several additional factors that we
cannot analyse or adjust for in our analysis that may
have contributed to the differing findings between
studies. For example, several studies were of crosssectional design8,12,13 and looked at a potential
association between current malaria and stunting
prevalence. Malaria at the time point of the study
may or may not correlate to previous exposure. This
correlation is likely to differ by transmission intensity
of malaria and this varied from low-seasonal to highperennial transmission. Although the assessment of
malaria incidence in the longitudinal studies was
probably a more accurate measure of exposure, it
seems plausible that the effect of malaria on growth
is modulated by immunity and thereby may vary with
age.4–7,9–11,14 In fact, a study from Tanzania found an
effect modification by age with the strongest effect of
malaria on stunting in children <1 year of age.15
A further potential limitation of our model is the
measurement of exposure. We have assumed that
the simple sum of malaria episodes over a child’s
life is what affects the child being stunted at age
2 years. It may be that a more complex function of
a child’s malaria history affects stunting; we plan to
investigate this in future work. In addition, the population in this study was enrolled in a clinical trial and
seen by medical personnel at close intervals. Prompt
medical treatment and nutritional interventions were
available free of charge during follow-up. It is possible
that the effect of malaria on stunting in this population may differ from that in the general population
and especially from that in populations where nutritional deficiencies are more common.
This interpretation of the effect ratio assumes that
the effect HbAS has on stunting is solely mediated by
a reduction of the number of malaria episodes.
However, HbAS also reduces the severity of every
malaria episode and the effect on stunting may
partly be due to this.24 This would lead to an overestimation of the effect that is attributable to each
malaria episode. However, the causality conclusion
would not change and even the lower boundary of
the 95% confidence interval for the effect (0.09) still
indicates a substantial effect of malaria on stunting.
Our analysis demonstrates the applicability of HbAS
as an instrumental variable for the analysis of conditions related to malaria. As in all observational studies, research on the association of malaria with other
medical conditions is often difficult due to the strong
influence of confounders and randomized trials are
almost always impractical. The method we propose
can be applied to reanalyse previous studies in this
area, specifically those where the genotyping of the
sickle cell gene has already been performed.6,15 We
hope that our findings will encourage the application
of MR to such analyses in the future. A potential
further application of MR using HbAS is the elucidation of associations between malaria and other infections. One such analysis was performed by Scott et al,
who used MR to analyse an association between malaria and bacteraemia caused by Salmonella spp.49
Our analysis provides evidence of a substantial
causal effect of malaria episodes on stunting, at
least in children <2 years of age in an area of high
endemicity. Our findings will hopefully spur further
research on this important epidemiological concern in
sub-Saharan Africa and increase the application of the
sickle cell trait as an instrumental variable in malaria
research.
Supplementary Data
Supplementary data is available at IJE online.
Funding
This work was supported by the Bundesministerium
für Bildung und Forschung (grant 01KA0202) that
funded the original trial, La Roche (Basel,
THE CAUSAL EFFECT OF MALARIA ON STUNTING
Switzerland) manufactured study drugs free of charge
and Sanofi-Aventis donated Artesunate tablets for
treatment of uncomplicated malaria episodes. The
German Centre for Infection Research (DZIF) supported B.K. with a clinical leave stipend. The funders
had no influence on design or implementation of the
study or analysis and interpretation of the data.
Acknowledgements
The authors would like to thank the children that
participated in the study and the fieldworkers who
1397
contributed to the data collection. We would also
like to thank the reviewers for their helpful suggestions to improve the manuscript.
H.K., B.K. and D.S. analysed the results and wrote the
manuscript. H.K. and D.S. conducted the statistical
analysis. B.K. was involved in the collection of the
data. R.K. contributed to the writing of the manuscript and cross-checked the analysis. J.M. and O.A.
conducted the original trial and contributed to the
writing of the manuscript. All authors read and
approved the final version of the manuscript.
Conflict of interest: None declared.
KEY MESSAGES
Previous association studies between malaria and stunted growth were inconsistent, mainly because
of their inability to fully adjust for confounding factors
In this paper, we use Mendelian randomization and matching to control for both unmeasured and
measured confounders where the sickle cell gene was used as the instrument.
We find a causal relationship, where the risk estimate of stunting increases by 0.32 for every malaria
episode.
A sensitivity analysis shows that our result is reasonably insensitive to violations of Mendelian randomization assumptions.
References
1
2
3
4
5
6
7
8
World Health Organization. World Malaria Report 2011.
Geneva: World Health Organization, 2011.
Korenromp
EL,
Armstrong-Schellenberg
JRM,
Williams BG, Nahlen BL, Snow RW. Impact of malaria
control on childhood anaemia in Africa – a quantitative
review. Trop Med Int Health 2004;9:1050–65.
Rice AL, Sacco L, Hyder A, Black RE. Malnutrition as an
underlying cause of childhood deaths associated with infectious diseases in developing countries. Bull World
Health Organ 2000;78:1207–21.
McGregor IA, Gilles HM, Walters JH, Davies AH,
Pearson FA. Effects of heavy and repeated malarial
infections on Gambian infants and children; effects of
erythrocytic parasitization. Br Med J 1956;2:686–92.
Bradley-Moore AM, Greenwood BM, Bradley AK,
Kirkwood BR, Gilles HM. Malaria chemoprophylaxis
with chloroquine in young Nigerian children. III. Its
effect on nutrition. Ann Trop Med Parasitol 1985;79:575–84.
ter Kuile FO, Terlouw DJ, Kariuki SK et al. Impact of
permethrin-treated bed nets on malaria, anemia, and
growth in infants in an area of intense perennial malaria
transmission in western Kenya. Am J Trop Med Hyg 2003;
68:68–77.
Deen JL, Walraven GEL, von Seidlein L. Increased risk for
malaria in chronically malnourished children under 5
years of age in rural Gambia. J Trop Pediatr 2002;48:78–83.
Ehrhardt S, Burchard GD, Mantel C et al. Malaria,
anemia, and malnutrition in african children–defining
intervention priorities. J Infect Dis 2006;194:108–14.
9
10
11
12
13
14
15
16
Arinaitwe E, Gasasira A, Verret W et al. The association
between malnutrition and the incidence of malaria
among young HIV-infected and -uninfected Ugandan
children: a prospective study. Malar J 2012;11:90.
Snow RW, Byass P, Shenton FC, Greenwood BM. The
relationship between anthropometric measurements and
measurements of iron status and susceptibility to malaria
in Gambian children. Trans R Soc Trop Med Hyg 1991;85:
584–89.
Fillol F, Sarr J, Boulanger D et al. Impact of child malnutrition on the specific anti-Plasmodium falciparum antibody response. Malar J 2009;8:116.
Deribew A, Alemseged F, Tessema F et al. Malaria and
under-nutrition: a community based study among
under-five children at risk of malaria, south-west
Ethiopia. PLoS One 2010;5:e10775.
Crookston BT, Alder SC, Boakye I et al. Exploring the
relationship between chronic undernutrition and asymptomatic malaria in Ghanaian children. Malar J 2010;9:39.
Genton B, Al-Yaman F, Ginny M, Taraika J, Alpers MP.
Relation of anthropometry to malaria morbidity and immunity in Papua New Guinean children. Am J Clin Nutr
1998;68:734–41.
Nyakeriga AM, Troye-Blomberg M, Chemtai AK,
Marsh K, Williams TN. Malaria and nutritional status
in children living on the coast of Kenya. Am J Clin Nutr
2004;80:1604–10.
Baiocchi M, Small DS, Lorch S, Rosenbaum PR. Building
a Stronger Instrument in an Observational Study of
Perinatal Care for Premature Infants. J Am Stat Assoc
2010;105:1285–96.
1398
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
Davey Smith G, Ebrahim S. ‘Mendelian randomization’:
can genetic epidemiology contribute to understanding environmental determinants of disease? Int J Epidemiol
2003;32:1–22.
Davey Smith G. What can mendelian randomisation tell
us about modifiable behavioural and environmental exposures? BMJ 2005;330:1076–79.
Lawlor DA, Harbord RM, Sterne JAC, Timpson N, Davey
Smith G. Mendelian randomization: using genes as instruments for making causal inferences in epidemiology.
Stat Med 2008;27:1133–63.
Sheehan NA, Didelez V, Burton PR, Tobin MD. Mendelian
Randomisation and Causal Inference in Observational
Epidemiology. PLoS Med 2008;5:e177.
Ebrahim S, Davey Smith G. Mendelian randomization:
can genetic epidemiology help redress the failures of observational epidemiology? Hum Genet 2008;123:15–33.
Davey Smith G, Ebrahim S. Mendelian randomization:
prospects, potentials, and limitations. Int J Epidemiol
2004;33:30–42.
May J, Evans JA, Timmann C et al. Hemoglobin variants
and disease manifestations in severe falciparum malaria.
JAMA 2007;297:2220–26.
Kreuels B, Kreuzberg C, Kobbe R et al. Differing effects of
HbS and HbC traits on uncomplicated falciparum malaria, anemia, and child growth. Blood 2010;115:4551–58.
Kreuels B, Ehrhardt S, Kreuzberg C et al. Sickle cell
trait (HbAS) and stunting in children below two years
of age in an area of high malaria transmission. Malar J
2009;8:16.
Kobbe R, Kreuzberg C, Adjei S et al. A randomized controlled trial of extended intermittent preventive antimalarial treatment in infants. Clin Infect Dis 2007;45:16–25.
World Health Organization. Child Growth Standards
based on length/height, weight and age. Acta Paediatr
Suppl 2006;450:76–85.
Glymour MM, Tchetgen Tchetgen EJ, Robins JM.
Credible mendelian randomization studies: approaches
for evaluating the instrumental variable assumptions.
Am J Epidemiol 2012;175:332–39.
Small DS, Rosenbaum PR. War and Wages. J Am Stat
Assoc 2008;103:924–33.
Wooldridge J. Econometric Analysis of Cross Section and Panel
Data. 1st edn. Cambridge, MA, USA: MIT Press, 2001.
Allison AC. Polymorphism and Natural Selection in
Human Populations. Cold Spring Harb Symp Quant Biol
1964;29:137–49.
Willcox M, Björkman A, Brohult J, Pehrson PO, Rombo L,
Bengtsson E. A case-control study in northern Liberia of
Plasmodium falciparum malaria in haemoglobin S and
beta-thalassaemia traits. Ann Trop Med Parasitol 1983;77:
239–46.
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
Hill AV, Allsopp CE, Kwiatkowski D et al. Common West
African HLA antigens are associated with protection from
severe malaria. Nature 1991;352:595–600.
Aidoo M, Terlouw DJ, Kolczak MS et al. Protective effects
of the sickle cell gene against malaria morbidity and mortality. Lancet 2002;359:1311–12.
Friedman MJ. Erythrocytic Mechanism of Sickle Cell
Resistance to Malaria. Proc Natl Acad Sci U S A 1978;75:
1994–97.
Friedman MJ, Trager W. The biochemistry of resistance to
malaria. Sci Am 1981;244:154–55, 158–64.
Williams TN, Mwangi TW, Roberts DJ et al. An Immune
Basis for Malaria Protection by the Sickle Cell Trait. PLoS
Med 2005;2:e128.
Cholera R, Brittain NJ, Gillrie MR et al. Impaired cytoadherence of Plasmodium falciparum-infected erythrocytes
containing sickle hemoglobin. Proc Natl Acad Sci U S A
2008;105:991–96.
Kramer MS, Rooks Y, Pearson HA. Growth and development in children with sickle-cell trait. A prospective
study of matched pairs. N Engl J Med 1978;299:686–89.
Rehan N. Growth status of children with and without
sickle cell trait. Clin Pediatr (Phila) 1981;20:705–09.
Ashcroft MT, Desai P, Richardson SA. Growth, behavior,
and educational achievement of Jamaican children with
sickle-cell trait. Br Med J 1976;6022:1371–73.
Ashcroft MT, Desai P, Grell GA, Serjeant BE. Heights and
weights of West Indian children with the sickle cell trait.
Arch Dis Child 1978;53:596–98.
Rosenbaum PR. Design of Observational Studies. New York:
Springer, 2012.
Rosenbaum PR. Observational Studies. New York: Springer,
2002.
Angrist JD, Imbens GW, Rubin DB. Identification of
causal effects using instrumental variables. J Am Stat
Assoc 1996;91:444–55.
Angrist JD, Graddy K, Imbens GW. The interpretation of
instrumental variables estimators in simultaneous equations model with an application to the demand for fish.
Rev Econ Stud 2000;67:499–527.
Rosenbaum PR, Silber JH. Amplification of Sensitivity
Analysis in Matched Observational Studies. J Am Stat
Assoc 2009;104:1398–405.
Ghansah A, Rockett KA, Clark TG et al. Haplotype analyses of haemoglobin C and haemoglobin S and the dynamics of the evolutionary reponse to malaria in
Kassena-Nankana District of Ghana. PLoS One 2012;7:
e34565.
Scott JAG, Berkley JA, Mwangi I et al. Relation between
falciparum malaria and bacteraemia in Kenyan children:
a population-based, case-control study and a longitudinal
study. Lancet 2011;378:1316–23.