Baker, Jennifer L.; Olsen, Lina W.; Sørensen, Thorkild I. A.
It has been proposed that suboptimal growth in utero and during the early postnatal period has negative long-term effects on health.1,2 Weight and other measures of size at birth are used as indicators of intrauterine growth and the prenatal environment. Numerous studies have established that lower size at birth is associated with heart disease3,4 and diabetes5,6 whereas larger size at birth is associated with the risk of obesity and various types of cancer.7–9 The net effect of these associations on total mortality is less clear.
Research on birth weight and adult all-cause mortality has been conducted in a variety of populations that were born in different years and followed for various lengths of time. Smaller size at birth has been associated with increased all-cause mortality in men, although there is considerable variation in both the strength of the association and the ages at which the association was observed.4,10–13 Other studies have not detected an association.14,15 Among women, a Finnish study11 showed an association between smaller size at birth and increased all-cause mortality, whereas other studies have not.4,10,13,14 Collectively, these studies cover birth years from 1911 to 1976, although no single study spans more than 29 birth years. The ages at follow-up range from 28 to 88 years. Given that birth weight has opposing associations with cardiovascular disease and cancer mortality, and that causes of death change with age, it is possible that the shape and strength of the association with total mortality changes with age. Although birth weight itself may not play a causal role, it is potentially an important indicator of risks for adult mortality that are established in the prenatal period.
We used a large population-based register, including subjects born during a 44-year period, to investigate the association between birth weight and mortality in adulthood.
This cohort study is based on data from the Copenhagen School Health Records Register. This registry contains information on all schoolchildren born from 1930 to 1983, who underwent annual health examinations in the municipality of Copenhagen. On 1 April 1968, the Danish Civil Registration System (Central Person Register) of vital statistics assigned a unique identification number to every Danish citizen.16 Linkage using this number provided information on vital statistics for the subjects in the school registry.
To be eligible for this study, subjects had to have been born between 1936 and 1979. Birth weight was recorded starting in 1936, and 1979 was the last birth year in which subjects could reach 25 years of age (the youngest age at entry into the study). Further, subjects had to have a unique identification number, and be alive and living in Denmark at age 25. Of 254,525 subjects fulfilling these criteria, 220,644 (87%) had a recorded birth weight, which was reported at the first school examination by the parents and thus could be included in the study. Information on birth length and gestational age was unavailable. We excluded persons with birth weights less than 2000 g (n = 3379) or greater than 5500 g (n = 801) to avoid potential recording errors. A total of 110,860 men and 105,604 women were included in this study.
Follow up of subjects began at the age of 25 years or at their age in 1968, when the unique identification number was assigned (whichever came later). Age 25 was chosen as the earliest entry point to restrict our analyses to adult mortality. Follow-up of subjects ended at the date of death, loss to follow-up, emigration, or 14 June 2004.
Birth weight was divided into 5 categories: 2000–2750, 2751–3250, 3251–3750, 3751–4250, and 4251–5500 g. These categories were chosen to minimize the effects of digit preference. The highest category was widest, reflecting the relatively smaller number of subjects with birth weights above 4250 g. To assess a possible birth cohort effect, we created a categorical variable that divided the year of birth into 4 intervals: 1936–1939, 1940–1945, 1946–1952, and 1953–1979. These categories correspond to the late interwar period, World War II, reconstruction, and a period characterized by social and political change (known as the “social expansion and new departure” time period).17,18
Research into fetal origins of adult disease has focused extensively on associations of birth weight with cancer and circulatory diseases. We analyzed mortality from these causes to understand how they may contribute to the pattern of all-cause mortality. Causes of death were obtained from the Danish National Cause of Death Register.19 Although this registry was established in 1942, it could be linked with our cohort only from 1970 onwards due to administrative reasons. Follow-up ended on 31 December 2001. Causes of death were defined according to the International Classification of Diseases, eighth revision (ICD-8) until 1994, and according to the tenth revision (ICD-10) thereafter. We investigated the association of categories of birth weight with cancer (ICD-8: 140–209; ICD-10: C00-C99), diseases of the circulatory system (ICD-8: 390–458; ICD-10: I00-I99), and death from all other causes.
We used Cox proportional hazards regression to examine the association among categories of birth weight and all-cause mortality. The birth weight category of 3251 to 3750 g was chosen as the reference as it contained the most subjects. Age was chosen as the underlying time scale.20
To investigate whether the hazard ratio changed across age, we created graphs of the cumulative hazard (Λ) from one birth weight category versus another.20 Plots were generated of Λi(age) versus Λj(age), which is the cumulative hazard in category i compared with the cumulative hazard in category j. If the resulting plot shows a straight line that intersects zero, then the cumulative hazards are proportional [Λi(age) = kΛj(age)] and the hazards are proportional as well [λi(age) = kλj(age)]. The interpretation of the line is that the risk is constant across age and that the assumptions of the Cox model are fulfilled. If the graph does not show a straight line, the inflection point will illustrate at which age this occurs.
Graphs were generated separately by sex for the 10 unique combinations of birth weight categories (20 graphs), and for each of the 4 birth cohort groups (80 graphs). The proportionality assumption was also assessed using the Schoenfeld test.21 Analyses were performed stratified by birth cohort and sex. We examined potential interactions between category of birth weight and birth cohort and between category of birth weight and sex on all-cause mortality. The shape of the association was tested for linearity.
The association of categories of birth weight with cancer, circulatory disease, and death from all other causes were examined in the same way as the all-cause analyses. In our cohort, even subjects without recorded birth weight values were followed up. Therefore, we conducted analyses to test if the hazard ratios differed between subjects with birth weight values and those without. We also conducted a series of sensitivity analyses. Subjects without a recorded birth weight were sequentially added to each birth weight category, the analyses were repeated, and the effects on the hazard ratios were examined. Statistical analyses were performed using SAS (version 9.2, SAS Institute Cary, NC) and STATA (version 9.2, StataCorp LP College Station, TX).
During the 44-year time period covered by our study, the mean and median birth weight changed very little among men and women (Fig. 1). The 5th percentile remained relatively stable during this period, whereas the 95th percentile became stable after 1955.
There were 11,149 deaths among the 110,860 men, and 6609 deaths among the 105,604 women (Table 1). Few subjects emigrated (n = 6209; 3%), very few were lost to follow-up (n = 270; <0.2%), and all were censored in the analyses at their last known date of residence in Denmark. In total, there were 5,205,477 person-years of follow-up.
The cumulative hazards associated with the various birth weight categories formed a straight line that intersected zero for both men and women (Fig. 2, selected graphs only). A slight deviation from the predicted line was observed for subjects with the oldest ages (>65 years) where the data were limited. No inflection points (at which the slope of the line changed) were identified in any graphs for either men or women. The lack of inflection points indicates that the risk of mortality by category of birth weight is constant across age in men and women, and shows that the proportionality assumptions were fulfilled. Schoenfeld tests in each birth weight category and each birth cohort stratum also confirmed that the proportionality assumptions were fulfilled (not shown).
We did not identify an interaction between birth weight categories and birth cohorts (P = 0.24), or between birth weight categories and sex (P = 0.59) on the risk of all-cause mortality. That is, the association of birth weight with all-cause mortality did not differ between men and women. Subsequently, the analyses were performed for men and women combined, and stratified by sex.
Compared with subjects in the reference category (3251–3750 g), the hazard ratio for all-cause mortality was 1.17 for those in the lowest birth weight category (95% confidence interval [CI] = 1.11–1.22). Those in the second lowest category had a hazard ratio of 1.01 (0.98–1.05), those in the second highest category had a hazard ratio of 0.96 (0.92–1.00), and those in the highest weight category had a hazard ratio of 1.07 (1.01–1.15) (Fig. 3). The association was U-shaped, and the test for linearity was rejected (P < 0.0001).
A diagnosis of cancer was recorded on the death certificates for 2110 men and 2335 women, a diagnosis of circulatory disease for 2747 men and 1213 women, and another diagnosis for 4501 men and 2076 women. Men and women were analyzed together, with the models stratified by sex. As the birth weight increased, the risk of cancer increased (Fig. 4, Panel A) in a linear manner and the test for linearity was not rejected (P = 0.62). For circulatory disease, there was a nonlinear trend, with the test for linearity rejected (P = 0.01). Compared with the reference category, elevated risks were identified for the 2 lower and the highest birth weight categories (Fig. 4, Panel B). For all other causes of death, the trend was also nonlinear (P = 0.0001) (Fig. 4, Panel C).
Although subjects without recorded birth weights were excluded from our study, they were also followed up. We tested whether their risk of all-cause mortality differed from subjects with a recorded birth weight. Overall, birth weight was not recorded on 13% of the health records. The proportion of missing birth weights was highest (31%) for the 1936–1939 cohort, and declined in the subsequent 3 cohorts (11%, 8%, and 12%, respectively). Compared with subjects who had a recorded birth weight, subjects without birth weights had a higher mortality (hazard ratio of 1.16 [95% CI = 1.11–1.21] for all-cause mortality). Sensitivity analyses showed that irrespective of the birth weight assigned to subjects with missing values, the hazard ratios remained virtually unchanged. For example, when subjects with missing birth weights were added to the smallest birth weight category, the hazard ratio for all-cause mortality associated with this category changed very little (1.16 compared with 1.17).
Our study shows a U-shaped association between birth weight and all-cause mortality as both low and high birth weights are associated with a significantly elevated risk. This association was identical for men and women, and was stable from 25 to 68 years of age.
Having a birth weight in the lowest category carried a 17% increased risk of death, whereas having a birth weight in the highest category carried a 7% increased risk, compared with the reference group. The excess risk with the smallest birth weight is consistent with results from other studies.4,10–13 Many low birth weight babies are also preterm, and it is possible that the association between low birth weight and a higher risk of mortality reflects this. The excess risk for the heaviest birth weight has apparently not been previously reported. It can be speculated that excessive rates of fetal growth are problematic for adult survival. Other studies have identified linear or J-shaped associations, but only lower birth weights have been associated with a convincingly elevated risk of all-cause mortality. Tests for nonlinearity have been conducted in many of these studies.4,11,13,14 Among Swedish men at ages 65 years and above, there was a suggestion of a reverse J-shaped association between birth weight and all-cause mortality, but the estimates in the highest birth weight category had very broad confidence limits.4 Our results, based on a much larger number of cases, suggest that linear modeling of the association may be too simplistic.
The elevated risks associated with lower and higher birth weights were stable across age in our study. This is unexpected in that causes of death change with age, and furthermore, the distance from the exposure of birth weight to the outcome of death increases as people age. Our results are consistent with those from a study of British men and women,13 but not with a study of Finnish men and women11 where an inflection point was identified at age 55 years for men; low birth weight was associated with all-cause mortality only at ages ≤ 55 years. Unlike subjects in the Finnish study, who were born in hospitals that served mostly families of low socioeconomic status, our study was relatively free of selection bias and had greater power to detect an inflection point had one been present.
Birth weight is an indicator of prenatal growth. There is a strong, reverse J-shaped association between birth weight and infant mortality,22 similar to (although stronger than) what we observe with adult mortality. Mechanisms underlying the association between birth weight and infant mortality are not well understood. It has been proposed that the association is not biologic but rather is due to the effects of an unmeasured confounder.22 We cannot exclude the possibility that confounding explains the observed association with adult mortality as well. Regardless, birth weight serves as an indictor of risk even if its associations with adult mortality are not entirely due to birth weight itself.
All-cause mortality, circulatory disease mortality and noncancer, noncirculatory disease mortality all shared a similar U-shaped association with birth weight. These results contrast with negative and linear trends reported by studies on all-cause mortality10,11,13 and circulatory disease mortality.4,11,13 As expected, the trend for the association between categories of birth weight and cancer was positive and linear.7 The consistency of the associations across time in our study, which spanned 44 years of birth—and the numerous social, environmental, and changes in exposure for both mothers and babies—makes it plausible that biologic mechanisms contribute to the association.
Evidence from neonatal autopsy studies23 and animal models24 links low birth weight with a reduced number of nephrons in the kidney, thus providing a biologic link to hypertension and circulatory disease. In rodent models,25 larger size at birth is associated with a greater number of cells, and it has been speculated this could create a greater potential for induced cellular transformations leading to cancer. These opposing mechanisms may contribute to the overall pattern we observe for all-cause mortality, but cannot account for what we observe with the noncancer, noncirculatory disease mortality. Low birth weight infants suffer from hypo- and hyperglycemia26,27 whereas high birth weight infants are likely to suffer from hypoglycemia.27 Sustained exposure to either state can permanently impair brain development.26,27 We speculate that this could lead to cognitive deficits and impairments in the regulation of body processes, which might contribute to the elevated risks of mortality observed at both ends of the birth weight spectrum.
Unique aspects of this study are the relatively minimal exclusion of subjects due to missing information and the completeness of follow-up for the members. A common criticism of the “developmental origins” hypothesis is that studies have been conducted on selected populations, combined with extreme losses to follow-up.28,29 Our study is based on a register that includes every child who attended school in the Copenhagen municipality and was born from 1936 through 1979. School attendance was compulsory and school health exams were mandatory throughout this time period. Exams were performed at all public and private schools, thus eliminating selection by socioeconomic status. Complete follow-up was available for every subject who was eligible for our study. Therefore, the cohort can be considered nearly complete with regard to inclusion and follow-up of Copenhagen’s school children.
Our investigation used subjects from a cohort based on children who survived until age 25 years (the age of entry into our study). This criterion raises the possibility of a healthy-cohort effect. Low birth weight is associated with infant and childhood mortality.30 We cannot assess this in our cohort because it is based on children who survived long enough to enter school. It is also possible that low birth weight babies were the least likely to survive until the age of 25 years. As a result, our analyses would be based on a sample of relatively healthy babies; however, this would only lead to an underestimation of the total impact of low birth weight.
At the first in-schooling health examination (when children were approximately 7 years old), parents were asked to recall their child’s birth weight. The method has been shown to be valid31,32; any imprecision in the recall would introduce random error and attenuate the results. Information on gestational age and birth length was not available. Studies on the association between birth length or ponderal index (kg/m3) and all-cause mortality in adulthood report stronger associations than between birth weight and all-cause mortality.11,12 Thus, had an examination of these factors been possible, stronger associations might have been identified.
The birth weight distribution in our cohort changed very little over 43 years. World War II affected Denmark less than other countries, as demonstrated in a comparison of the anthropometric changes among schoolchildren in the Nordic countries during this time period.33 Birth weight was not available for all cohort members. The recording of birth weight was phased in, and from the 1940s onwards it was consistently available. Unlike many studies in this area, we also traced subjects without a recorded birth weight and investigated their risk of mortality. Subjects without a recorded birth weight had a slightly higher risk of mortality, presumably reflecting differences in social class and education. Missing birth weight values are unlikely to have biased our results, as shown by the sensitivity analyses.
In these analyses, it was not possible to account for the effects of socioeconomic status (SES). We cannot exclude the possibility that residual confounding by SES contributes to the observed associations between low birth weight and increased mortality.34 Results from a smaller study on selected Danish boys in Copenhagen showed that adjustments for social factors only marginally attenuated an association between low birth weight and all-cause mortality.12 The question remains, however, how SES may have influenced our results. Subjects in our study were born across a long time period, from 1936 through 1979. During these years, there were dramatic societal, nutritional, medical, and environmental changes. The prevalence of smoking changed among women during these years.35 By stratifying the analyses by birth cohort, we adjusted indirectly for these and other, often immeasurable, factors. The consistency of the association across the different birth cohorts, given their very different exposures, suggests that adjusting for SES in these analyses would only marginally attenuate the results.
In Denmark, the prevalence of low birth weight infants has remained stable over the last 30 years, whereas there have been large increases in the numbers of high birth weight infants.36 Currently, more than 15% of girls and 22% of boys are born with a birth weight >4000 g.36 The increases in numbers of heavy infants in recent years, many of which are a result of increasing maternal body size,37–39 suggests that the already sizeable impact of high birth weight on all-cause mortality in adulthood may be elevated in the future. More generally, these results suggest that interventions aimed at increasing birth weight as a method of improving health are too simplistic and may actually cause harm.28 Future research on the fetal origins hypothesis must expand the definition of suboptimal growth to include excess growth, which is associated with an excess risk of all-cause mortality in adulthood.
We are grateful to Kathleen M. Rasmussen for her scientific contributions and analytic suggestions on a previous version of the manuscript.
1. Barker DJP. Mothers, Babies, and Disease in Later Life
. London: BMJ Publishing Group; 1994.
2. Lucas A. Programming by early nutrition in man. In: Bock GR, Whelan J, eds. The Childhood Environment and Adult Disease
. CIBA Foundation Symposium 156. Chichester: Wiley; 1991:243
3. Barker DJ, Winter PD, Osmond C, et al. Weight in infancy and death from ischaemic heart disease. Lancet
4. Leon DA, Lithell HO, Vagero D, et al. Reduced fetal growth rate and increased risk of death from ischaemic heart disease: cohort study of 15,000 Swedish men and women born 1915–29. BMJ
5. Hales CN, Barker DJ, Clark PM, et al. Fetal and infant growth and impaired glucose tolerance at age 64. BMJ
6. Forsén T, Eriksson J, Tuomilehto J, et al. The fetal and childhood growth of persons who develop type 2 diabetes. Ann Intern Med
7. Ahlgren M, Wohlfahrt J, Olsen LW, et al. Birth weight and risk of cancer. Cancer
8. Sørensen HT, Sabroe S, Rothman KJ, et al. Relation between weight and length at birth and body mass index in young adulthood: cohort study. BMJ
9. Rugholm S, Baker JL, Olsen LW, et al. Stability of the association between birth weight and childhood overweight during the development of the obesity epidemic. Obes Res
10. Friedlander Y, Paltiel O, Deutsch L, et al. Birthweight and relationship with infant, child and adult mortality in the Jerusalem perinatal study. Paediatr Perinat Epidemiol
11. Kajantie E, Osmond C, Barker DJP, et al. Size at birth as a predictor of mortality in adulthood: a follow-up of 350 000 person-years. Int J Epidemiol
12. Nybo Andersen AM, Osler M. Birth dimensions, parental mortality, and mortality in early adult age: a cohort study of Danish men born in 1953. Int J Epidemiol
13. Syddall HE, Sayer AA, Simmonds SJ, et al. Birth weight, infant weight gain, and cause-specific mortality: the Hertfordshire cohort study. Am J Epidemiol
14. Power C, Li L. Cohort study of birthweight, mortality, and disability. BMJ
15. Frankel S, Elwood P, Sweetnam P, et al. Birthweight, adult risk factors and incident coronary heart disease: the Caerphilly Study. Public Health
16. Pedersen CB, Gotzsche H, Moller JO, et al. The Danish Civil Registration System. A cohort of eight million persons. Dan Med Bull
17. Hansen SA, Henriksen I. Sociale brydninger 1914–39 [Social strife 1914–39]. Dansk socialhistorie [Danish social history]. Vol. 6. Copenhagen: Glydenal; 1980.
18. Hansen SA, Henriksen I. Velfærdsstaten 1940–78 [The welfare state 1940–78]. Dansk socialhistorie [Danish social history]. Vol. 7. Copenhagen: Gyldenal; 1980.
19. Juel K, Helweg-Larsen K. The Danish registers of causes of death. Dan Med Bull
20. Andersen PK, Borgan Ø, Gill RD, et al. Statistical Models Based on Counting Processes. Springer Series in Statistics
. New York: Springer-Verlag; 1992.
21. Schoenfeld D. Chi-squared goodness of fit tests for the proportional hazards regression model. Biometrika
22. Basso O, Wilcox AJ, Weinberg CR. Birth weight and mortality: causality or confounding? Am J Epidemiol
23. Hughson M, Farris A, Douglas-Denton R, et al. Glomerular number and size in autopsy kidneys: the relationship to birth weight. Kidney Int
24. Bagby SP. Maternal nutrition, low nephron number, and hypertension in later life: pathways of nutritional programming. J Nutr
25. de Assis S, Khan G, Hilakivi-Clarke L. High birth weight increases mammary tumorigenesis in rats. Int J Cancer
26. Hemachandra AH, Cowett RM. Neonatal hyperglycemia. Pediatr Rev
27. McGowan JE. Neonatal hypoglycemia. Pediatr Rev
28. Rasmussen KM. The. “fetal origins” hypothesis: challenges and opportunities for maternal and child nutrition. Annu Rev Nutr
29. Kramer MS, Joseph KS. Enigma of fetal/infant-origins hypothesis. Lancet
30. Li CI, Daling JR, Emanuel I. Birthweight and risk of overall and cause-specific childhood mortality. Paediatr Perinat Epidemiol
31. O’Sullivan JJ, Pearce MS, Parker L. Parental recall of birth weight: how accurate is it? Arch Dis Child
32. Walton KA, Murray LJ, Gallagher AM, et al. Parental recall of birthweight: a good proxy for recorded birthweight? Eur J Epidemiol
33. Angell-Andersen E, Tretli S, Bjerknes R, et al. The association between nutritional conditions during World War II and childhood anthropometric variables in the Nordic countries. Ann Hum Biol
34. Naess O, Claussen B, Thelle DS, Smith GD. Four indicators of socioeconomic position: relative ranking across causes of death. Scand J Public Health
35. Juel K. Increased mortality among Danish women: population based register study. BMJ
36. Schack-Nielsen L, Molgaard C, Sørensen TIA, et al. Secular change in size at birth from 1973 to 2003: national data from Denmark. Obesity (Silver Spring)
37. Kramer MS, Morin I, Yang H, et al. Why are babies getting bigger? Temporal trends in fetal growth and its determinants. J Pediatr
38. Surkan PJ, Hsieh CC, Johansson AL, et al. Reasons for increasing trends in large for gestational age births. Obstet Gynecol
39. Dietl J. Maternal obesity and complications during pregnancy. J Perinat Med
© 2008 Lippincott Williams & Wilkins, Inc.