Contents - Previous - Next

Fetal growth and adult disease

1. Evidence for fetal origins of adult disease
2. Discussion
References
Commentary
References
Discussion

DA Leon

Correspondence: Dr DA Leon

Epidemiology Unit, Department of Epidemiology and Population Sciences, London School of Hygiene & Tropical Medicine, Keppel Street, London WC1E 7HT, UK

Current knowledge on associations between variations in fetal growth on the one hand and blood pressure, non-insulin dependent diabetes, coronary heart disease and cancer in adulthood on the other is reviewed and related to more conventional preoccupations of perinatal epidemiology. Commonly used definitions and indicators of impaired fetal growth, possible explanations and mechanisms of the association between fetal growth impairment and later disease, and the concept and operational definitions of programming are discussed. Implications and research priorities that can be derived from this information are presented.

1. Evidence for fetal origins of adult disease

The notion that some adult diseases may have their origins in utero has recently captured attention across a wide area of big-medical science. Editorial comment on this idea has ranged from the enthusiastic, referring to 'paradigm shifts' (Robinson, 1992), to the critical and sceptical (Paneth and Susser, 1995). Perinatal epidemiologists have tended to welcome it as a boost and extension of scope while those working with experimental systems have taken up the challenge to develop and refine informative animal models. Among researchers in cardiovascular disease, this idea has been more controversial, as it has sometimes been regarded as undermining efforts to establish and promote a clear public health message focussed on modifications of adult behaviours and circumstances.

The idea that adult health may reflect circumstances in childhood or even earlier in life is one that dates back many years (Kuh and Davey Smith, 1993). However, in the field of cardiovascular disease the work of David Barker's group in the UK, has been central to the development and study of the fetal origins hypothesis in its current form (Barker 1992, 1994). As mentioned below, others have been developing a parallel interest in the fetal origins of various cancers.

The early work suggesting that cardiovascular disease and cancer may have origins in utero was no more than suggestive. For instance, geographical correlation studies showed that contemporary mortality rates from coronary heart disease were related to infant mortality rates 50 or more years ago by area within England and Wales (Williams et al, 1979; Barker and Osmond, 1986). Although intriguing, these studies could only be treated as hypothesis generators. In particular, there was widespread concern (Ben-Shlomo and Davey Smith, 1991) that the apparent association between early life circumstances and later disease may be simply a consequence of socioeconomic confounding. It was argued that infant mortality could simply be an indicator of relative socioeconomic conditions of each area at the time of birth and hence in adult life given the persistence of the geography of socioeconomic deprivation in Britain this century.

Since these early studies, much more work has been done, particularly with respect to cardiovascular disease. This paper sets out to summarise what is now known about the association between in utero circumstances/fetal growth and risk of disease in adult life. Emphasis will be placed upon reviewing the most robust and direct evidence. A more general, demographically orientated review has been published elsewhere (Elo and Preston, 1992), as has another review focusing on methodological problems with the evidence (Joseph and Kramer, 1996). The paper also aims to relate this work to the more conventional preoccupations of perinatal epidemiology, and to develop a discussion of possible mechanisms and directions for future research.

1.1 Blood pressure

The effect of birth weight on blood pressure is the most studied association between fetal growth and later physiological characteristics. This association has recently been reviewed by Law and Shiell (1996) who examined the results of 32 papers identified in a systematic search of the literature. This review shows clearly that at all ages, with the possible exception of the pubertal period, there is a tendency for blood pressure to increase as birth weight goes down. This association is seen in different countries and in males and females.

The possibility of socio-economic confounding has been addressed in a number of studies, and it is apparent that this does not provide an explanation for the association between birth weight and blood pressure. For instance, in the Uppsala study, adjustment for maternal marital status and socio-economic status at birth, and social class, drinking habits and education at age 50 had little substantive effect upon the association between birth weight and blood pressure (Koupilová et al, 1997).

In general it has been found that adjustment for weight or body mass index (BMI) at the time at which blood pressure is measured increases the strength of the association. This is because

a) birth weight is positively correlated with later body size and BMI
b) blood pressure goes up with weight and BMI.

It has been suggested (Paneth and Susser, 1995) that it is misleading to adjust for current body size, as from a public health perspective it is the net effect of birth weight that is important. It may be, however, that the effect of birth weight on blood pressure is potentiated by body size later in life. This is clearly illustrated in Table 1, taken from the Uppsala study of 50 year old men (Leon et al, 1996), where it is apparent that men in the top third of the BMI distribution showed a stronger relation of birth weight to systolic and diastolic blood pressure compared to those in the bottom third of the BMI distribution.

Table 1 also illustrates that the association between birth weight and blood pressure is continuous over the entire range of birth weight. This is a very consistent and thought provoking finding that does not naturally fit with the tendency within peri-natal epidemiology to think in terms of binary divides between 'small' and 'normal'. The long-term effects of birth weight are not simply driven by recognisable peri-natal pathology. The Uppsala data also suggest that the blood pressure-birth weight association is strongest among those men who were born at term (38-41 weeks).

One of the striking findings of the Uppsala study has been that the strength of the birth weight blood pressure association is much steeper among men who were of above median height at age 50, compared to those who were at or below median height (Table 2). This finding has been replicated in an analysis of the Hertfordshire data (Law et al, 1996). The question is how do we interpret this result? In pert-natal epidemiology much thought has been given to the question of how to assess whether a new born has suffered an impairment of growth relative to its genetic potential, as already discussed by Bakketeig in his contribution to this meeting. Looking at size at birth relative to that of previous births has been an interesting approach (Skjaerven and Bakketeig, 1989). By analogy, it could be argued that the group composed of those who are relatively tall as adults but small at birth contain a greater proportion of people who failed to achieve their growth potential in utero than the group of similar size at birth but who were shorter as adults. Here we are using adult height as an index of differences in genetic growth potential between individuals. Thus it may be this sort of failure to achieve ones growth potential in utero that is particularly significant for later blood pressure.

Barker and colleagues, more than other researchers in this field, have examined the association between blood pressure and measures of size and shape at birth other than weight. For example, the ratio of birth weight to placental weight was suggested to be an important predictor of subsequent blood pressure (Barker et al, 1990), with men who were light at birth but had heavy placentas having the highest mean blood pressure. However, the Uppsala study has failed to replicate this observation, finding in fact that blood pressure tended to go down as placental weight increases. Other studies have also failed to find any consistent association with placental weight or the placental to birth weight ratio.

Finally with respect to blood pressure, two analyses suggest that blood pressure in pregnancy is related to the birth weight of the mother. A recent report from a Belgian case-control study of 1,822 women delivering with hypertension at a university hospital between 1988 and 1993 plus matched 'normal' delivery controls, found the risk of preeclampsia or pregnancy-induced hypertension to be inversely related to the pregnant women's own birth weights (Hanssens et al, 1996). Relative to women weighing 3000-3999 g at birth, those who weighed < 2500 g, 2500-2999 g and > 4000 g had a risk of preeclampsia that was respectively 2.16 (95% CI 1.07-4.38), 2.35 (95% CI 1.52-3.62) and 0.81 (95% CI 0.44-1.48). Consistent with this, preliminary analyses (Leon, personal communication) of data from the Scandinavian SGA study (Bakketeig et al, 1993) suggest that mean blood pressure at each stage of pregnancy is inversely related to the women's own birth weights.

1.2 Non-insulin dependent diabetes

The evidence for an association between size at birth and non-insulin dependent diabetes and insulin action is summarised in a forthcoming review (McKeigue, 1997). Birth weight and/or ponderal index (birth weight/birth length³) have been found to be inversely related to impaired glucose tolerance/non-insulin dependent diabetes in adults in Hertfordshire (Hales et al, 1991) and Preston (Phipps et al, 1993) in the UK, among Pima-Americans (McCance et al, 1994), US male health professionals (Curhan et al, 1996) and in Uppsala men (Lithell et al, 1996). Several studies have also suggested that size at birth is related to glucose levels in children (Yajnik et al, 1995; Law et al, 1995). In general, these associations are strengthened by adjustment for current body mass index. Size at birth has also been found to be inversely related to measures of insulin resistance or insulin levels in these and other populations.

In 60 year old Uppsala men (Lithell et al, 1996), the prevalence of diabetes was more strongly associated with ponderal index than birth weight. Men in the bottom quintile of ponderal index had a three-fold greater prevalence of diabetes than men in the other four quintiles. This effect was independent of body mass index. As measures of beta-cell function and insulin action were available in these men at age 50, it was possible to investigate whether either of these factors mediated the association between size at birth and diabetes. It was found that adjustment for the acute phase insulin response to an intravenous glucose tolerance test (IVGTT) was unrelated to size at birth. This suggested that beta-cell function was not the principal mediating factor. However, the 1-hour insulin level from the IVGTT at 50 was strongly related to ponderal index. Moreover, adjustment for this reduced the strength of association between ponderal index and diabetes at age 60, although it did not entirely eliminate it. This has been interpreted as suggesting that impairments in fetal growth, best measured by ponderal index rather than birth weight, result in a permanent reduction in insulin sensitivity. These conclusions are in line with a range of results from David Barker's group. The focus of current work in this area is to try and uncover the specific defect in insulin action which seems to be affected by fetal growth.

Table 1. Mean systolic and diastolic blood pressure (mm Hg) at age 50 by birth weight and body mass index at age 50 among men resident in Uppsala, Sweden in 1970 (from Leon et al, 1996)

Figures in parentheses are numbers of men.

Table 2. Mean systolic and diastolic blood pressure (mm Hg) at age 50 by birth weight and height at examination among men resident in Uppsala, Sweden in 1970 (from Leon et al, 1996)

Figures in parentheses are numbers of men.
*In the full data set of 2322 men median height is 176 cm.

1.3 Coronary heart disease

The classic epidemiological studies of cardiovascular disease (and cancer) have tended to start with cohorts of people already in adult life, many being restricted to people in middle age or older. They provide little if any information about the influence of early life factors on the incidence of cardiovascular events. The evidence that does exist for pre-adult influences on cardiovascular disease, extending back from adolescence through childhood, into infancy and ultimately fetal life, has been recently reviewed elsewhere (Leon and Ben-Shlomo, 1997). As already mentioned, geographic analyses, and other designs including mobility studies (Strachan et al, 1995; Elford et al, 1989; Notkola et al, 1985; Vågerö and Leon, 1994), have played a useful role in generating hypotheses. Their central weakness is that it is difficult to convincingly exclude the possibility that the measures of early life circumstances are only associated with disease risk in so far as they are correlates of adult circumstances and behaviour.

The British Hertfordshire cohort study has been an important source of information about the relationship between birth weight and mortality from coronary heart disease (Barker et al, 1989; Fall et al, 1992; Osmond et al, 1993; Fall et al, 1995; Fall et al, 1995; Vijayakumar et al, 1995; Martyn et al, 1996). This area of Southern Britain was fortunate to have an exceptionally well organised midwifery service in the first part of this century. From 1911 births were required to be notified by the attending midwife within 36 hours of delivery. Information on birth weight as well as details of the mother were reported. In addition, local health visitors recorded information on a special form about the infants progress throughout the first 12 months of life, including their weight at one year. Barker and colleagues extracted this information, and set about tracing the fate of the 22 thousand singleton births born 1911-30 who survived childhood and for whom birth weight and weight at one year was known. Limitations of the British system used to determine date and cause of death of such cohorts meant that it was only possible to look at mortality from 1951. Moreover, only those members of the cohort who could be definitely identified in the national register in 1951 were included. Because of difficulties introduced by change of name for women on marriage, the trace rate for women (60%) was lower than for men (79%). Thus, of the 37 thousand live singleton births originally identified, the mortality analyses (1951-1992) are based on only 16 thousand (43%). This low rate of inclusion has been the focus of criticism. However, although the mortality cohort may not be entirely representative of all the births 1911-30, it is difficult to see how the analysis of mortality in relation to size at birth or at 1 year within the 16 thousand could be seriously biased.

What then has the Hertfordshire cohort revealed? As shown in Table 3, there is a decline in the rate ratio for coronary heart disease (CHD) as birth weight increases, although this is only statistically significant for men. An even more consistent pattern of decline in the rate ratios for CHD was found among men by weight at one year. Weight at one year has also been found to be inversely related to the prevalence of CHD (Fall et al, 1995) and left ventricular mass (Vijayakumar et al, 1995) in special studies of several hundred men from the original Hertfordshire cohort. Associations of birth weight with these outcomes were not as smooth.

The Hertfordshire findings with respect to CHD and birth weight have now been confirmed in studies of men in Sheffield, England (Barker et al, 1993), Caerphilly, Wales (Franker et al, 1996) and Uppsala, Sweden (Køupilová and Leon, 1996). Data for women are less common, but the same picture has emerged from an analysis of CHD from the US Nurses Health Study (Rich-Edwards et al, 1997). The only published data that have failed to confirm this general pattern is from an analysis of mortality among a relatively small number of men born in Gothenburg, Sweden in 1913 (Eriksson et al, 1994). Information on size at birth in Sheffield, Uppsala and Gothenburg was obtained from hospital or midwives' records, while in the case of Caerphilly and the US Nurses Health Study, birth weight was obtained from recall by the subjects or their female relatives. The main shortcoming of the Hertfordshire mortality study is that information on socioeconomic circumstances, health and lifestyle in adult life is not available for the cohort. As a consequence, it has not been possible to look at the confounding and mediating influences of these factors on the association between birth weight and CHD. However, the Uppsala and Caerphilly cohorts and the US Nurses Health Study are based on prospectively followed study populations recruited in adult life for whom information on adult circumstances and cardiovascular risk factors is available. In these studies adjustment for socio-economic factors in adult life and at birth had little if any impact on the strength of the birth weight CHD associations. Furthermore, adjustment for a range of cardiovascular risk factors including blood pressure and serum cholesterol had no substantive impact on the strength of the associations.

Table 3. Rate ratios for coronary heart disease mortality among singleton men (ages 20-81) and women (ages 20-69) born in Hertfordshire in relation to birth weight (based on Osmond et al, 1993)

Figures in parentheses are numbers of deaths.

On closer inspection, the pattern shown in Table 3 from Hertfordshire, is not one of a consistent linear decline in mortality from the lowest to the highest birth weight categories. There is a slight up turn in the CHD rate in the highest birth weight group. It is striking that this reverse-J shaped association is also apparent in the results from Uppsala and the US Nurses Health Study. The published results from the Sheffield study do not show separate estimates for birth weights over 9 lbs. It is probable that this upturn in risk seen in the highest birth weight category is because this group contains an appreciable proportion of macrosomic births to mothers who had impaired glycaemic control in pregnancy. This itself may lead to an increased risk of non-insulin dependent diabetes in the offspring either for genetic reasons, or because of long-term down-regulation of insulin receptors or some other similar phenomenon in response to the high levels of fetal exposure to glucose.

In focussing upon the associations between fetal growth and CHD we should not lose sight of the evidence that postnatal growth may be important. There is now substantial evidence that adult height is inversley associated with CHD risk. This association has been found in a range of different settings, and appears to be independent of other known risk factors (Yarnell et al, 1992; Palmer et al, 1990; Leon et al, 1995). The few studies that have been able to look at the height association adjusting for birth weight (Yarnell et al, 1992; Rich Edwards et al, 1995) suggest that these factors have an independent effect on CHD risk. Recent analyses of a cohort of children whose height was measured in the late 1930s, has found that height under the age of 14 years is inversley correlated with mortality from coronary heart disease, with leg length being a particularly strong predictor (Gunnell et al, in press). Leg length, more than trunk length, is believed to be sensitive to environmental circumstances.

In summary, a consistent picture is emerging of a decline in risk of CHD with increasing size at birth, with a suggestion of a small upturn in risk among those weighing over 9.5 Ibs. This pattern has been observed in a number of different populations. It does not appear to be accounted for by socio-economic confounding. More strikingly, the association does not seem to be mediated by blood pressure or other established adult cardiovascular risk factors.

1.4 Cancer

It has been suggested that in utero circumstances may affect later risk of a range of hormonally related cancers, i.e., cancers of the breast (Le Marchand et al, 1988; Ekbom et al; 1992; Sanderson et al, 1996; Ekbom et al, 1995; Michels et al, 1996), testis (Depue et al, 1983; Brown et al, 1986; Akre et al, 1996), ovary (Walker et al, 1988) and prostate (Ekbom et al, 1996; Tibblin et al, 1995). At present, however, there is little consistency in the findings of the few studies that have investigated these associations (Leon and Ben-Shlomo, 1997). Despite this there has been an appreciable amount of hypothesising on the possible mechanisms that could lead to in utero influences on risk of these malignancies (Trichopoulos, 1990a/b; Trichopoulos and Lipworth, 1995; Lipworth et al, 1995). Further work needs to be done on this area.

Contents - Previous - Next