Contents - Previous - Next


(KM. Rasmussen, Division of Nutritional Sciences, Savage Hall, Cornell University, Ithaca, NY 14853, USA)

I would like to complement Dr. Leon on his useful, thoughtful paper (Leon, 1997). He rightly put the work by Barker and members of his group clearly in the context of 'hypothesis generation', at which they have been most effective. What is needed now is much more attention to 'hypothesis testing'. This is because of problems inherent to the retrospective approach when it is used for an issue such as this one, which doubtlessly has multiple causes - most of which are not documented in available records.

My approach will be to use Leon's paper as a point of departure and to focus on areas that I think deserve further scrutiny. I will offer a critique of some of Barker's ideas from the perspective of someone who has focused on the earlier, rather than the later, periods of life. In doing this, I will include some observations from the recent conference on this subject, which was entitled "Frontiers in Maternal, Fetal and Neonatal Health: Programming for a Life-Time of Good Health", organized by Peter Nathanielsz and held at Cornell University Ithaca in August 1996.

The 'Barker Hypothesis' (Barker, 1994) would not have generated such interest if it were not based on sound biological principles and if it didn't make some sense. As someone who has devoted her scientific career to the concept that maternal nutritional status matters, I welcome the expansion of this idea. We can all agree that the concept of developmental programming is an important and valid one; many examples of it were presented in detail at the August meeting in Ithaca. The novelty of Barker's hypothesis is the extension of this concept to much later in life than had previously been the case. It is also more specific about the insult, inadequate maternal nutritional status, that negatively affects fetal growth. However, Barker's hypothesis also would not be as controversial as it has been if it did not have some elements that seem implausible.

There are a number of issues that are worthy of further discussion. The first of these is Leon's observation (Leon, 1997) that most of the effect that has been observed is related to variation in fetal growth rather than prematurity. However, it must be remembered that early in this century, most premature infants died and also that not all premature infants weighed less than 2,500 g (Lang et al, 1996). Thus, this generalization may not actually have been tested adequately. It remains for future researchers to explore the actual importance of prematurity as a predictor of adult onset disease.

A second issue worthy of further discussion is Barker's concept that the association between birth weight (BOO) and cardiovascular disease (CVD) may be 'amplified' or 'modified' by factors that act at later periods of life (Barker, 1994). Often, it is difficult to measure and control for such factors. In the data sets that have been explored to date, information on most such factors has been absent. How obesity might participate in the association between size at birth and later CVD is an example of this difficulty (Figure 1). Obesity may be in the causal pathway between size at birth and CVD - in which case is it not an 'amplifier'. Obesity, on the other hand, could act as effect modifier-in which case it would be an 'amplifier'. The statistical techniques that one would use to explore this differ depending on how these variables are conceptualized. This concept also has implications for public health because it affects the educational message one might want to deliver to the public.

A third issue that merits further consideration is Leon's statement that "If a common genetic factor does underlie part of these associations, this requires size at birth to have an appreciable genetic component." I believe there is evidence confirming this: The work of Klebanoff and coworkers (Klebanoff et al, 1984; Klebanoff and Yip, 1987) establishes that BW tends to repeat from one generation to the next; it also repeats from one birth to the next within the same woman as described by Bakketeig earlier (Bakketeig et al, 1979) and also in this meeting (Bakketeig, 1997).

Obesity is in causal pathway; no amplification

IUGR (r) (r) Obesity (r) CVD Mortality

Figure 1. A schematic diagram showing obesity in the causal pathway between size at birth and later adult disease contrasted with obesity acting as an "amplifier" of this relationship.

IUGR, intrauterine growth retardation, CVD, cardiovascular disease.

There are several possible ways in which genetic and environmental factors may contribute to the association between size at birth and disease in adulthood. Some of these cannot easily be disentangled:

(a) What is being seen is purely genetic (i.e. it is expressing what is in the fetus's chromosomes).

(b) What is being seen is genetic-but it is in the mother's genome (e.g. her height). However, it is difficult to prove that mother's height is not environmental because her height may also reflect her environment (Figure 2).

(c) What is being seen is purely environmental. A good example of this is first parity; the environment here is that of the maternal system in responding to pregnancy. This is a significant source of growth retardation (100 g on average, nearly as large as the size of the nutritional effect in most supplementation trials, and half the size of the smoking effect) that has not been evaluated for its relationship to CVD.

(d) What is being seen is purely environmental where the environment here is more external to the mother than in the previous example (e.g. high altitude or maternal dietary intake).

(e) A final possibility is confounding by socioeconomic status, which is possible with most of the associations that have been observed. It is important to remember that when Barker's subjects were born, there was a clear social class gradient in England, and when they developed CVD, there was a clear gradient of social class for this too. Unfortunately, it is virtually impossible to disentangle this degree of confounding by post hoc measures, particularly when they are as limited as those available in the data sets that have been examined to date.

Exemple of intergeneration repetition of birth weight

Maternal birth weight (r) Child’s birth weight

Figure 2. A schematic diagram showing different ways in which maternal height may relate to birth weight and mortality from cardiovascular disease (CVD).

A fourth issue of concern is that BW is probably mainly a proxy for some aspect of the intrauterine environment, but what this is has not been specified and remains unknown. BW played an important role in Barker's earlier publications (Barker, 1994), but was no longer part of the conceptual framework he presented at the Ithaca meeting. He and his coworkers have expanded this concept to include weight at one year but, in my opinion, this just confuses the issue. One of the strongest predictors of size at one year is BW, so this does not distance this new predictor much from BW. In addition, it adds method of infant feeding, which may itself be a predictor of CVD (Hamosh, 1988; McGill et al, 1996; Wong et al, 1993; Salmenperð et al, 1988), as a complication; thus any association may be with this instead of with intrauterine environment (Figure 3).

From a biological perspective, if one assumes that BW is a proxy for some other factor then one might reasonable expect that a different aspect of the intrauterine environment might be crucial for each of the chronic diseases with which BW has been associated. In fact, this is what others have found. In addition, as Leon has made clear, low BW due to prematurity is not associated with a generally higher susceptibility to chronic disease in adulthood (Leon et al, 1995). Therefore, future studies should distinguish among the various forms of growth retardation among term infants and between full-term and premature infants. I reiterate what we have already discussed at this meeting: it is tricky to distinguish infants who have intrauterine growth retardation (IUGR) from those who do not because their distributions of BW overlap. As a result, not all small babies have IUGR and not all larger babies are not IUGR (Lang et al, 1996). Furthermore, some of the non-IUGR babies are 'overgrown' (large-for-gestational age for lack of a more precise term) (Figure 4).

It is reasonable to think that being 'overgrown' would itself have adverse consequences. This is suggested by the upturn of the reverse, J-shaped relationship that Leon discusses. The paper just presented by Martorell and his colleagues (Martorell et al, 1997) is a positive step in this direction, but more such research is needed.

From my point of view, the issue that merits the greatest attention is the observation that the association between BW and, for example, mortality from premature CVD is continuous 2 (i.e. a 'dose-response' relationship) over the entire range of 'normal' BW. There are two key words here: 'dose-response' and 'normal'; I will discuss each in turn.

Figure 3. A schematic diagram showing how birth weight and breastfeeding may each be related to cardiovascular disease (CVD).

Figure 4. A schematic diagram showing that the range of "normal" birth weight includes some infants who are growth-retarded as well as some who are "overgrown".

This finding is provocative because it calls into question our definition of 'normal' BW (i.e. between 2,500 and 4,000 g), which is based on survival during the first month and year of life. It suggests, if one is to believe Barker's data, that any BW under 4,000 g carries an excess risk of death from CVD for males. This is a difficult proposition to accept given that this BW is not achievable or desirable (even if a cesarean delivery were available worldwide) for the majority of the world's women. We also know that many of the heavier of these infants often have a rocky course in the early days and weeks of life (they may have been excluded from Barker's studies because they did not survive infancy).

This finding also suggests that we should not be comparing IUGR infants with normally grown and/or overgrown babies, or even asking "what is it about the intrauterine environment of a small baby that leaves a lasting imprint?" Perhaps we should instead phrase the question as "why should progressively bigger babies have progressively less CVD?" The only theory that I can offer is a reiteration of what we have already discussed: bigger babies come from better nourished women and the predictors of being born to a better nourished mother are also the predictors of a healthier, lower-stress life, which would lead to lower CVD. In this case good fetal growth is a proxy for good socioeconomic circumstances.

The intrauterine mechanism that one would postulate here is one of increased substrate availability with accompanying hormonal changes resulting in larger births. How this is related to CVD remains unexplained. Interestingly, the reverse example (poor fetal growth is a marker for poor socioeconomic conditions leading to a less successful, more stressful life and, thus, higher CVD rates) also is reasonable and is supported by, for example, the work of Ben-Schlomo and Smith (1991). This line of reasoning reinforces the importance of what many others have already contended: control for such factors is notoriously difficult but, without adequate measures of and control for socioeconomic conditions before and after birth, Barker's results and similar findings of others are difficult to explain.

In summary, Leon has carefully and thoughtfully summarized the growing epidemiologic evidence for/against the 'Barker Hypothesis'. I agree with him that there is something important and worthy of study here. However, it is certainly more complex than Barker has suggested in his recent publications and presentations.

Testable, refutable hypotheses are desperately needed. I agree with Leon's call for testing hypotheses related to the basic biology of this association between size at birth and later chronic disease. Such hypotheses should permit the evaluation of causality, not just association; thus experimental work, not just more observational studies, is required.

Further population-based studies should distinguish among the various forms of growth retardation that may occur among term infants as well as between full-term and premature infants.

In addition, it would be helpful if we could design studies in which it was possible to apportion the risk of CVD that related, for example, to current nutrition/lifestyle, genetics, and history of prior disease as well as intrauterine experience. Available data could be used to make a preliminary calculation of etiologic fraction. Given the strength of other predictors of CVD, BW itself (or whatever it is a proxy for) may actually play a relatively minor role.

Acknowledgements - The thoughtful comments of Drs. Jean-Pierre Habicht and Cutberto Garza on an earlier version of this commentary are gratefully acknowledged.


Bakketeig LS (1997): Current growth standards, definitions, diagnosis and classification of fetal growth retardation. This Volume.

Bakketeig LS, Hoffman HJ & Harley EE (1979): The tendency to repeat gestational age and birth weight in successive births. Am. J. Obstet. Gynecol. 135, 1086-1103.

Barker DJP (1994): Mothers, Babies, and Disease in Later Life. BMJ Publishing Group: London.

Ben-Schlomo Y & Smith GD (1991): Deprivation in infancy or in adult life: which is more important for mortality risk? Lancet 337, 530-534.

Hamosh M (1988): Does infant nutrition affect adiposity and cholesterol levels in the adult? J. Pediatr. Gastroenterol. Nutr. 7, 10-16.

Klebanoff MA & Yip R (1987): Influence of maternal birth weight on rate of fetal growth and duration of gestation. J. Pediatr. 111, 287-292.

Klebanoff MA, Graubard BI, Kessel SS & Berendes HW (1984): Low birth weight across generations. JAMA 252, 2423-2427.

Lang JM, Lieberman E & Cohen AP (1996): A comparison of risk factors for preterm labor and term small-for-gestational-age birth. Epidemiology 7, 369-376.

Leon DA (1997): Fetal growth and adult disease. This Volume.

Leon DA, Smith GD, Shipley M & Strachan D (1995): Adult height and mortality in London. Early life, socioeconomic confounding, or shrinkage? J. Epidemiol. Comm. Health 49, 5-9.

Martorell R. Ramakrishnan U, Schroeder D, Melgar P & Neufeld L (1997): Intrauterine growth retardation, body size, composition and physical performance in adolescence. This Volume.

McGill HC, Mott GE, Lewis DS, McMahan CA & Jackson EM (1996): Early determinants of adult metabolic regulation: effects of infant nutrition on adult lipid and lipoprotein metabolism. Nutr. Rev. 54, S31-S40.

Salmenperä L, Perheentupa J, Slimes MA, Adrian TE, Bloom SR & Aynsley-Green A (1988): Effects of feeding regimen on blood glucose levels and plasma concentrations of pancreatic hormones and gut regulatory peptides at 9 months of age: Comparison between infants fed with milk formula and infants exclusively breast-fed from birth. J. Pediatr. Gastroenterol. Nutr. 7, 651-656.

Wong WW, Hachey DL, Insull W, Opekun AR & Klein PD (1993): Effect of dietary cholesterol on cholesterol synthesis in breast-fed and formula-fed infants. J. Lipid. Res. 34, 1403-1411


There is a considerable amount of literature on experiments with animals supporting the concept of developmental programming, i.e. the idea that factors impinging on fetal growth and development determine to some extent what is happening to the organism after birth. Interpretations of the animal literature beyond this become more complicated because of the many different paradigms used in animal models, the wide range of outcomes, and doubts as to the extent to which generalizations are possible across species.

Most of the information available on humans today comes from medical records of people born in high-income countries during the first half of the century on the one hand and information from death certificates, medical records or results of medical examinations of some of the same people in adulthood. Epidemiologists depended on whatever data were available in records and used them opportunistically. In many instances only a small subgroup of the original cohort could be traced later in life, and it is difficult to argue convincingly that the subsamples that could be followed-up were unbiased.

It is also unlikely that the etiologic factors that led to IUGR in Europe in the first half of the century and the associations with non-communicable diseases later are the same that predominate in developing or industrialized countries today. Extrapolations from results of these studies therefore have to be made with caution, and experimental studies are needed to investigate the main etiologic factors leading to variations in fetal growth, their consequences in adulthood and mechanisms that lead from the former to the latter.

The hypothesis that maternal nutrition during pregnancy and low birthweight are important determinants of chronic, non-communicable diseases provides this issue with a great deal of public health legitimacy and importance, particularly if the relationships can be shown to be causal. Causality will have to be established by results of experimental studies. The question therefore arises whether the time has come to advocate randomized controlled trials in pregnancy designed solely to study long-term effects of fetal growth. The prevailing opinion was that, prior to that, one should identify appropriate intervention trials in pregnancy, set up to study shorter-term outcomes (e.g. birthweight), and extend them by adding follow-up studies in adulthood. It is important to do this both in developed and developing countries since differences in the determinants of fetal growth can be expected to lead to differences in long-term outcomes. A problem is that even the most successful intervention trials carried out in pregnancy so far have only had moderate effects on fetal growth, so that the variability introduced by them may be too small to expect long-term effects.

Early growth restriction may create a disposition which later turns into disease only if certain conditions are met at a later age. A study of arteriosclerosis in the US (New Orleans) and South America showed that in both regions coronary arteriosclerosis began very early developing slowly and continuously up to about age 20. In South America this development continued to be slow and never reached a level resulting in a high incidence of coronary heart disease, whereas in New Orleans the development of arteriosclerosis accelerated and the incidence of CHD increased markedly with age. IUGR is much more prevalent in developing than in developed countries, whereas the prevalence of chronic non-communicable diseases has generally been lower in developing than in developed countries. This apparent contradiction is generally explained by the argument that, while the predisposing factors were present, the conditions needed to express the disease at a later age were not widely prevalent in developing countries. The prevalence of these conditions, like high energy intake in the form of fat and sedentary lifestyle, is rapidly increasing in certain segments of developing country populations. In many countries of Latin America, like Chile, Argentina, Costa Rica, Panama and parts of Mexico, overweight is already a greater problem in adults than underweight.

If this hypothesis is correct, one should find strong interactions between variation in fetal growth and dietary and other adult risk factors. This seems actually to be the case. Evidence is accumulating that deprivation followed by plentiful supply is particularly harmful. This is illustrated by data from South India (Mysore) from Ethiopian Jews who migrated to Israel and now suffer from very high rates of diabetes. In Guatemala, the association between LBW and obesity in adulthood is greater in migrants to Guatemala City than in adults who remain in rural areas. Hispanics in the US, with similar genetic background, have the highest rates of obesity.

Some doubts are expressed about low birthweight or IUGR being sufficient and necessary conditions on the causal chain from intrauterine conditions and events to chonic non-communicable diseases in adulthood. There are unmodifiable predictors of low birth weight like first parity and female sex of the newborn that do not appear to entail a higher risk. An example of programming independent of birthweight is provided by a Ca supplementation study in Argentina. Pregnant women were randomly assigned to Ca supplementation or placebo treatment. The risk of having high systolic and diastolic blood pressure was greater in the offspring of mothers who had received the placebo treatment. This effect was independent of birthweight but stronger in children who were overweight at the time blood pressure was measured. In many instances low birthweight may not be important per se, but a proxy of a growth inhibiting factor or another consequence of it. A low number of nephrons could, for instance, be associated with low birthweight and, at the same time, be responsible for later hypertension which in turn could be responsible for a higher incidence of CVD.

It could also be that a family history of cardiovascular disease, associated with an increased risk of being SGA, is the more immediate reason for an increased risk of CHD also in the offspring. One of the best predictors of size at birth is the mother's ability to expand her vascular volume during pregnancy. If this is not possible to the full extent, she will produce an IUGR baby and could, at the same time, but independently, transmit to the baby factors entailing a higher risk of hypertension and CVD in adulthood.

The most important single factor affecting blood pressure is body weight at the time of measurement the relationship between weight at birth and blood pressure at a later age is always heavily confounded by body weight at the time of measurement.

The need, from a public health point of view, to identify groups at high risk and to target interventions at them, encourages the definition of cut-off points and binary divides of populations into those at greater and those at lesser risk. Even though this is necessary for an efficient targeting of interventions, it is artificial from an epidemiological point of view, when we are dealing with continuous variables like birthweight and blood pressure and associations between them that extend over their whole range. This dilemma is not easily resolved.

Contents - Previous - Next