17

THE ASSOCIATION BETWEEN PROLONGED

BREASTFEEDING AND POOR GROWTH-
What are the implications?

Jean-Pierre Habicht, M.D.,Ph.D.,MPH

Division of Nutritional Sciences, Cornell University, Ithaca. NY 14853

Key words: Breast feeding, growth, reverse causality, confounding, complementary

feeding:

Abstract: The smaller size of breast fed children in infancy and thereafter in
malnourished and well-nourished populations has resulted in rushes to
judgement that have been shown to be ill-advised. The reasons for the smaller
size in malnourished populations is due to retaining the small and sickly child
at the breast (reverse causality) and the consequent continuing sickliness of
this breast fed child (negative confounding). Once the reverse causality and
negative confounding have been taken into account breast feeding improves
growth, at least through the second year of life. Thus prolonged breastfeeding
should always be fostered, especially in malnourished populations. An
exception remains when breast milk may transmit disease to the suckling
child. In well-nourished populations the magnitude of the difference between
breast fed and weaned children is much less than in malnourished populations,
is observed to increase over the first year of life, but to have disappeared by
the end of the second year. One may never-the-less be concerned that
complimentary feeding practices are not adequate for these children.

1. INTRODUCTION
The practical program and policy implication of the previous reports is
that continued breastfeeding is good. The three studies from poor
malnourished populations in Kenya, Senegal and Peru found (published here
and in other reports) that the benefits of breastfeeding and suckling extend
even into the third year of life in spite of the finding that sometimes babies

Short and Long Term Effects of Breast Feeding on Child Health

Edited by Berthold Koletzko et al., Kluwer Academic/Plenum Publishers, 2000 193
194 J. Habicht

who are partially breastfed beyond the age of six months are not as big as
non-breast babies. When this occurred it was not that breastfeeding stunted
growth but that small baby size and poor health determine continued
breastfeeding.

While the studies from developing countries are clear, it is less clear
why there is a negative association between infant size and continued
breastfeeding in the latter half of infancy in the study from Denmark.
However this study does not affect practical recommendations because there
is no evidence that this decreased rate of growth is detrimental.

2. EVOLUTION OF ISSUE
A review of the evolution in our thinking on the relationship between
breastfeeding and size is instructive. The contribution by Simondon in this
volume cites the various authors who have reported a negative association
between breastfeeding and babys size. Initially investigators who were
known for their scientific rigor reported this negative association but were
careful in their assessment of the implications for recommendations,
awaiting further research to clarify the situation. However, a highly
publicized report in 1988 (Brakohiapa et al. 1988 cited in Simondon) stated
in their summary that , These results indicate that prolonged
breastfeeding can reduce total food intake and thus predispose to
malnutrition. The possibility ofreverse causality was not considered.
Reverse causality refers to the situation when the putative cause and
effect relationship are reversed. This was, for instance, the reason for the
extremely high association between not breastfeeding (putative cause) and
infant mortality (effect) originally reported in the literature. The association
was mostly due to the fact that children who died shortly after birth never
breastfed and not because a lack of breastfeeding caused these deaths
(Habicht et.al. 1986). In this case an important remaining component of the
association was never-the-less causal, in that breast feeding did save lives.
The example of the breastfeeding-mortality association shows that after
discarding the effect of reverse causality one must continue to search for and
demonstrate causality. In other words, it is not enough to show reverse
causality to negate causality. Analyses to determine causality and reverse
causality require longitudinal data to establish a time sequence. The first
study to do this in the examination of breast feeding and growth was in
Africa but was reported in French in a source that was so obscure that it was
unknown (Garrenne et al. 1987 cited in Simondon). This report was followed
up shortly by a publication from Bangladesh, whose senior author was also
The Association between Prolonged Breastfeeding and Poor Growth 195

French (Briend and Bari, 1989 cited in Simondon), but published in the peer
reviewed English literature. Since at least 1993 (review by Grummer-
Strawn cited in Simondon) the possibility of reverse causality was well
known, although it took some further time before it was established as the
cause of the negative association between breastfeeding and growth, as
reported in the three studies in this section. It is interesting that quantitative
proof of reality comes for this issue, as for so many others, decades after
anecdotal evidence is reported (e.g., Morley 1968 cited in Simondon) based
upon facts that are well known to those working in the field.
In contrast to silence on the possibility of reverse causality, the
methodological problem of confounding was considered as of the first
reports about the negative association between breast feeding and child size.
Confounding is the term used to refer to a situation in which the relationship
between a putative determinant (e.g., breastfeeding) and an outcome (e.g.,
growth) is caused by a third factor (e.g., illness). The association of poor
growth with breastfeeding would then be due to illness, or to poor
complementary feeding, or to biased measurements and reporting, but not to
breastfeeding. Longitudinal data do not help much in dealing with
confounding. This issue must be addressed with imagination and persuasion
to make a plausibility argument. Alternatively specially designed and
difficult intervention trials can be used to make a statistical probability
statement, which is stronger than a plausibility argument (Habicht et
a1.1999).
The position taken by a number of authors who have urged the
abandonment of prolonged partial breastfeeding because of some biological
anorexic properties is surprising, even without an understanding of the issue
of reverse causality, Some of the authors who had looked for confounding
and did not find evidence of it apparently were then led to the conclusion
that it must be some unknown factor in breastmilk itself. But our
understanding at the time of the advantages of breastfeeding in the kinds of
populations in which it most occurs was solid. In particular the advantage of
continuing breastfeeding when the complementary diet was poor in protein
was often mentioned, and has since been shown to be the case in just these ,
kinds ofpopulations (Marquis et al 1997a).

The papers questioning the benefits of prolonged breastfeeding and

those discussing these reports were not chary in inventing mechanisms to
explain why breastfeeding stunted growth. Overwhelmingly they were
biological and child- oriented in the sense that breastmilk was supposed to
reduce appetite. None of these mechanisms were ever tested. The reverse
causality mechanism, which is behavioral and under the control of the
mother (mother-oriented), is easily testable. Why was it not invoked first?
196 J. Habicht

We believe it is because quantitative research in breastfeeding is

insufficiently informed about breastfeeding decisions and behaviors (such as
those described Simondons paper, in Dettwyller et al. (1987) and in
Marquis et al (1997a) both cited in Simondon). This expertise and similar
expertise about breast feeding in general (Pelto 198 1) would be useful to our
ISHM&L. It will require some recruiting to bring this scientific expertise to
our society.

While inferring causality from statistical associations without other

plausible evidence is a perennial problem, the scientific mill ultimately
grinds finely enough to correct the mistake. However, the harm is not so
quickly erased. In December, 1999 a graduate student approached me about
investigating the anorectic properties of breast milk, in spite of having read
all the pertinent literature! Apparently the idea was too beguiling. More
serious than the effect on graduate students is the effect on programs and
policy recommendations where such mistakes are not so benign, and
continue to bedevil policy discussions long after the scientific mistake is
corrected.

3. SUMMARY OF THREE STUDIES IN

DEVELOPING COUNTRIES
The reports at this meeting from rural Kenya, rural Senegal and a slum in
Peru had a number of similarities and some differences. The similarities
were that they were longitudinal, were of children who were more than a
year old, and who were already markedly stunted. One can infer from the
data presented that the mean stunting was in the range of -1.6to -2.0 Z-
score for children in their second year of life. This corresponds to the level
of malnutrition usually seen in rural communities and slums around the
world. Some countries, such as Guatemala and India, and the Sahel during
famine periods have more severe levels (around -3.0 Z-scores). All the
studies used linear OLS statistics with interactions. All three studies
presented analyses that take into account confounding, either in this
meetings reports or in previous publications.

The major dissimilarities relate to the amount of information that was

presented about each study in this book. The Senegal report reported a
generally beneficial effect of breast feeding on growth up to 28 months of
age. The Kenyan report mentioned that there was a positive association
between the duration of breastfeeding and growth.. The Peruvian report
followed the evolution of the negative association between breast feeding
The Association between Prolonged Breastfeeding and Poor Growth 197

and growth through the second year of life - it did not present evidence of a
beneficial effect of breast feeding as had the other two. However, that
evidence had been presented in a previous publication ( Marquis et al. 1997
b).
The Peruvian studies had ethnographic information, which showed the
dynamics of decision making that led to reverse causality. Mothers are
concerned that prolonged breast feeding weakens them so they stop when
their child no longer needs it. In Peru the information presented at this
meeting showed that the strength of this preferential prolongation of breast-
feeding for sickly children changes markedly over the second year of life. It
was very strong at the beginning and had ceased by the end.

Reverse causality can be reinforced by confounding in that the children

who continue to be breastfed are more sickly after their selection. This
negative confounding has been described in the Peruvian study (Marquis et
a1 1997b). Of course, reverse causality and negative confounding biases are
related in that children who are sickly and retained at the breast tend to
remain sickly and continue to grow poorly.

Once reverse causality and negative confounding have been taken into
account all three studies showed a beneficial effect of breast feeding on
growth. At the meeting analyses of all three studies were discussed but not
presented, Apparently the usual confounders that might explain this positive
effect were investigated and found not to be reponsible. This increases the
plausibility that the benefits on growth associated with breast feeding are
causal and not artifactual, Plausibility is further increased if the conditions
that potentiate this beneficial effect are those that would be expected given
the biological benefits that are ascribed to breastfeeding. This was the case in
these studies: the benefits were greater among children in environments with
poor housing or hygienic conditions (verbally reported for Kenya and
Senegal), and who have poor complementary diets (Peru in Marquis 1997a).
Another finding in Peru confirms the mothers perceptions about the special
benefits of breast feeding for children who are not growing well (not
thriving). These finding underscore the importance of recognizing and
factoring in heterogeneity of response to nutritional activities when planning
interventions and in analyzing study results. Testing for statistical
interactions (synergisms and antagonisms) is important, but needs to be
supported by thoughtful comparative analyses of subgroups to fully
understand the relationships.

The practical implication of these studies is that in poor populations

breast feeding improves growth and is most beneficial when other household
198 J. Habicht

circumstances are least propitious. This heterogeneity is not yet being taken
into account adequately in the recommendations relative to breast feeding by
HIV mothers. Neither are the heterogeneities in the likelihood of HIV
transmission through breast milk being adequately considered. For instance
the recent finding of much reduced transmission during exclusive breast
feeding (Coutsoudis et al. 1999) may change the recommendations
substantially for certain poor populations.

4. THE COPENHAGEN STUDY

This study also found a negative association between breast feeding and
growth. The longer the baby suckled the shorter it was at one year of age.
This, and the above studies, were also comparable in that they were
longitudinal, and they sought to take into account confounding and reverse
causality. However, they were not comparable in age or in the level of
poverty. In contrast to the other studies, this study collected data in infancy.
The babies in the Copenhagen study lived in one of the most healthy and
well-grown populations of the world, and were not stunted at the onset ofthe
study, in contrast to the poverty-stricken, unhealthy and stunted children in
the other studies. These differences are so profound that one might think that
there is nothing to be learned from the other studies as relates to this study.

In fact the Copenhagen study is an important representative of many

other studies that have shown that breastfeeding duration is associated with
decreased growth in infancy in well off populations (WHO 95). This effect
is presumed to be due to a biological effect of breast milk on child.appetite.
It is a child-oriented mechanism in contrast to a behavioral mother-
oriented mechanism. The Copenhagen study took into account biases due to
self-selection and confounding. Once these were considered there was no
difference in growth until after four months of age. Thereafter children who
were breastfed were almost a centimeter shorter by nine months of age, and
another centimeter for those who breast fed longer. One might conclude
from this evidence that nothing more is needed to assert that this decreased
growth is natural, and is a healthy consequence of prolonged breast feeding.
This assertion is the basis of a world wide effort to establish growth
standards for breast fed infants (De Onis et al. 1997). It is also the basis for
the UNICEFrecommendation to breastfeed exclusively to six months of age.

Aspects of the Copenhagen study, as well as studies in other countries,

should give one some pause before concluding that no further research is
needed on this subject. The magnitude of the growth effect associated with
The Association between Prolonged Breastfeeding and Poor Growth 199

duration of breastfeeding is substantially different across all these studies.

For instance the Copenhagen study found differences of two centimeters at
one year of age in comparison to a half centimeter difference reported in
WHO (1999). This variability across studies is not likely to be due to factors
mediated only through breastfeeding.

In the Copenhagen study the proportion who breastfed to five months

was 20%, and only 3% were breastfed beyond nine months. This very small
proportion indicates that longer breastfeeding was not culturally normative
in Copenhagen, and the small minority of mothers who did continue were
likely to be different from the great majority who did not in multiple ways
that affect growth. It is likely that child care and dietary patterns were
different in this subgroup, and these differences need to be investigated
before one can conclude that observed differences are due to breastfeeding.
In the case of Denmark Dr. Michaelsen has pointed out that this seems to be
particularly the case for those who breastfeed longest. Is this the reason for
the much larger difference in length between the longest breast feeding
group and the others? The possibility that other factors are responsible
underscores the importance of further investigation, particularly because of
concern about the adequacy of complementary feeding, even in wealthy
families in developed countries.

On the other hand, scientific uncertainty about feeding in later infancy

should not affect breastfeeding recommendations in wealthy countries,
because the differences in growth between those who breast feed longer and
those who receive breastmilk for a shorter period of time is usually small at
one year of age and disappears thereafter.

However, it may well turn out that the effects on growth, especially
among those who breast feed for longer periods of time, is due to
inadequacies In complementary feeding. Our knowledge on this score is
woefully lacking (Brown et al 1998), in contrast to our knowledge about the
adequacy and benefits of human milk and breast-feeding. It is time that we
turned our attention to this essential complementary element to breast-
feeding.

ACKNOWLEDGMENT

I would like to thank Gretel Pelto for substantive discussions and critical
comments.
200 J. Habicht

REFERENCES
Brown,K., Dewey,K., and Allen,L., 1998, Complementary Feeding of Young Children in
Developing Countries, World Health Organization, Geneva

Coutsoudis, A., Kubendran, P., Spooner, E., Kuhn, L., and Coovadia, H.M.,, 1999, Influence
of infant-feeding patterns on early mother-to-child transmission of HIV- 1 in Durban,
South Africa: a prospective cohort study. Lancet 354:471-76.

de Onis, M., Gam, C., and Habicht, J-P., 1997,Time for a new growth reference. Pediatrics,
100: E8

Habicht, J-P., DaVanzo,J., and Butz,W.P., 1986, Does breastfeeding really save lives, or are
apparent benefits due to biases? American Journal of Epidemiology 123: 279-290.

J. Habicht, J-P., Victora, C.G., and Vaughan, J.P., 1999, Evaluation designs for adequacy,
plausibility and probability of public health programme performance and impact.
International Journal of Epidemiology. 28: 10- 18.

Marquis, G.S., Habicht, J-P., Lanata, C.F., Black, R.E., and Rasmussen, K.M., 1997a, Breast
milk or animal-product foods improve linear growth of Peruvian toddlers consuming
marginal diets. American Journal of Clinical Nutrition 66: 11 02-1 109.

Marquis, G.S., Habicht, J-P., Lanata, C.F., Black, R. E., Rasmussen. K.M., 1997b,
Association of breastfeeding and stunting in Peruvian toddlers: An example of reverse
causality. International Journal of Epidemiology 26: 349-356.

Pelto, G.H., 198 1, Perspectives on infant feeding: decision-making and ecology. Food and
Nutr. Bull. 3:16-29.

WHO, 1995,-Physical Status:The Use and Interpretation ofAnthropometry. Technical Report

of a WHO Expert Committee, Geneva, World Health Organization, Geneva