Birth weight is strongly associated with the risk of perinatal death, probably because it reflects fetal maturation and pathology. Although this association is undisputed at the individual level, it is less clear whether differences in average birth weight between populations also lead to differences in perinatal mortality rate. 1
Average birth weight differs substantially between national populations and racial groups, as was shown, for example, by the large International Collaborative Effort in the 1980s. 2,3 In this study, which covered 11 different populations, median birth weight for singletons ranged between 3186 gm among black babies in the United States and 3544 gm in Norway. These differences were attributable to shifts towards lower or higher birth weights of the entire birth weight distributions of these populations. It has been suggested that these shifts are accompanied by similar shifts in the birth-weight–specific perinatal mortality curves, so that the optimal birth weight is lower in populations with lower average birth weights. 4–6 This shift has indeed been found in several one-by-one comparisons:eg, between black and white babies in the United States, 7 between Norway and the United States, 8 between Norway and Hungary, 9 and between Belgium and the United States. 10
Within the framework of the EuroNatal study, 11,12 a large European collaborative project, we were able to use a large dataset covering seven countries to make similar comparisons. The main aim of the analysis reported in this paper was to test the hypotheses that (1) differences between countries in average birth weight are attributable to shifts of the entire birth weight distributions towards higher or lower birth weights, and (2) these shifts are accompanied by similar shifts of the birth-weight–specific perinatal mortality curve, in such a way that countries with lower average birth weights also have lower optimal birth weights. We expected that confirmation of this hypothesis would considerably strengthen the case for the development and use of population-specific standards for birth weight.
We selected those countries participating in the EuroNatal study for which numbers of births and perinatal deaths could be classified simultaneously by gestational age (completed weeks) and birth weight (grams). These were (data sources in parentheses): Finland (Finnish Medical Birth Registry and cause-of-death register), Sweden (Medical Birth Registry), Norway (Medical Birth Registry), Denmark (Danish Birth Register), Scotland (United Kingdom; National Health Service, ISD [SMR2]), The Netherlands (National Obstetrics and Neonatology Registries), and Flanders (Belgium; Gezondheidsindicatoren, Ministerie Vlaamse Gemeenschap).
We used an “extended” definition of perinatal mortality, combining stillbirths with neonatal deaths (up to 28 days after birth), because neonatal care is increasingly able to delay death beyond the end of the first week. The perinatal mortality rate was calculated as the number of perinatal deaths per 1000 live births plus stillbirths. Only singleton births were included. Because some countries have a lower limit of 28 completed weeks of gestational age for the registration of stillbirths, we excluded births with a gestational age of less than 28 completed weeks. To increase the statistical power of our analysis, we obtained data for a 3-year period, 1993–1995. This was impossible for both Denmark, for which we used data for 1992–1994, and the Netherlands, for which we used data for 1995 only. This produced 1,372,092 singleton births, among which 7900 extended perinatal deaths occurred.
Data from the Dutch National Obstetrics and Neonatology Registries were incomplete. No births were recorded in the Registries by general practitioners (who attend fewer than 5% of all births in the Netherlands), and births were recorded by 89% of midwife practices, by 84% of level I hospitals, and by 100% of level II hospitals. Because the completeness depended on the level of care and because birth weight distributions and perinatal mortality rates differed between levels of care, we reweighted the data by level of care, assuming that births attended by general practitioners have the same birth weight distributions and perinatal mortality rates as births attended by midwives (details available from authors). Gestational age and birth weight were missing for fewer than 5% of births in all countries, except Norway, where gestational age was missing for 9.8% of births.
We analyzed between-country differences in the position of the birth weight distribution by making frequency plots of birth weights, and by determining each country's most frequent birth weight, ie, the mode or maximum of its birth weight distribution. Because we wanted to obtain a 95% confidence interval (CI) for the mode, we also estimated the mean value and the standard error of the mean (SEM). As others have shown before us, however, birth weight distributions are skewed to the left because of the presence of small preterm births in the left tail of the distribution;4 therefore, the mean does not correspond exactly with the mode. After exclusion of births of gestational ages of 36 weeks or less, mean birth weight corresponded exactly with modal birth weight and we decided to use this mean value and its SEM to quantitatively characterize the mode.
The rate of perinatal mortality decreases steeply from very low to normal birth weights, reaches its lowest value within the range of the most common birth weights, and increases again at higher birth weights. Various statistical models, including logistic parabolas and combinations of two linear logistic risks, 5 have been used to characterize this distribution and to determine its minimum. We decided to use a logistic parabola model. This model was a logistic quadratic regression model, in which perinatal mortality is a function of birth weight and birth weight squared. We preferred a logistic parabola to Wilcox and Russell's combination of two linear logistic risks, 5 because a logistic parabola would permit us to calculate 95% CIs of the minimum of the birth-weight–specific perinatal mortality curve. To improve its fit, however, we limited the application of this model to the symmetric part of the curve around the minimum. This part was identified by fitting a segmented linear model to the left hand part of the birth-weight–specific perinatal mortality curve, with a smooth joining point for the linear and quadrate models (“smooth” is the same predicted values for the two models at the joining point, and the same slope of the two models at both sides of the joining point). The position of the joining point was determined by fitting the two models with different joining points, and by choosing the joining point at which the models fit best. The minimum of the quadratic part of the curve (“optimal birth weight”) was determined as the birth weight for which the first derivative of the fitted parabola equalled 0, and the 95% CI of the minimum was estimated using the point estimators and the covariances of the logistic model. 13 We checked the regression-based minima by comparing them with the lowest point of a smoothing spline, fitted using a generalized additive model, 14 which showed that quadratic regression provided unbiased estimates of the minimum of the birth-weight–specific perinatal mortality curve.
Subsequently, the relation between the mode of the birth weight distribution and the minimum of the birth-weight–specific perinatal mortality curve was examined, and a linear regression coefficient was calculated.
Gestational age is strongly associated with birth weight and perinatal mortality. 15 Gestational age itself may be affected by the frequency of obstetric interventions such as induced labor or elective delivery. International differences in the distribution of gestational age may therefore bias the relation between the mode of the birth weight distribution and the minimum of the perinatal mortality curve. To eliminate this possible bias, we restricted additional analysis of the mode of the birth weight distribution, the minimum of the mortality curve, and their relation to term deliveries (37–41 weeks’ gestation). These represent 89% of births but only 47% of perinatal deaths; because of small numbers, it was not possible to restrict the analysis to even smaller strata of gestational age.
Figure 1 illustrates the international differences in the position of the birth weight distribution (panel A) and in the position of the birth-weight–specific perinatal mortality curve (panel B); Scotland, the country with the second lowest modal birth weight, is compared with Norway, which has the second highest modal birth weight. Differences among other countries in the position of the birth weight distribution and the position of the perinatal mortality curve show similar patterns. Generally, countries with lower average birth weights (eg, Scotland) show lower perinatal mortality rates (compared with other countries) at lower birth weights. In contrast, countries with higher average birth weights (eg, Norway) show lower mortality rates at higher birth weights. Because of this pattern, the mortality curves for countries with lower and higher average birth weights will usually cross.
The mode of the birth weight distribution and the minimum of the birth-weight–specific perinatal mortality curve are shown for each country in Table 1. The highest modal birth weights were found in Finland, followed by Norway, Sweden, Denmark, the Netherlands, and Scotland, whereas the lowest modal birth weight was found in Flanders (Belgium). The same ranking was found for the minimum of the birth-weight–specific perinatal mortality curve, except that Norway and Finland changed places to highest and second highest position, respectively. Optimum birth weight is consistently higher than modal birth weight, with a tendency for the difference to be larger in countries with higher average birth weight. Figure 2 shows the close association between modal weight and optimum weight: for every 100 gm higher modal birth weight, optimal birth weight was 170 gm higher (95% CI = 104–236 gm).
This close association may be attributable to a third variable, gestational age. If countries differed in average gestational age, this could potentially affect both the mode of their birth weight distributions and the minimum of their birth-weight–specific perinatal mortality curves. This does not explain our findings, however, because the countries included in our study do not differ substantially in their gestational age distribution. The median gestational age was 40 weeks for all countries, and the mean ranged between 39.2 and 39.7 weeks (standard deviation = 1.5–1.9 weeks). Exclusion of preterm births did not change our results; among deliveries at 37–41 weeks of gestation, there was a strong positive association between the mode of the birth weight distribution and the minimum of the birth-weight–specific perinatal mortality curve. For every 100 gm higher birth weight (mode), the minimum of the birth-weight–specific mortality curve was 152 gm higher (95% CI = 84–219 gm).
This European collaboration provided an opportunity to investigate international differences in birth weight distributions and their relation with perinatal mortality rates. Our study showed a strong positive correlation between the mode of the birth weight distribution and the minimum of the birth-weight–specific perinatal mortality rate, suggesting that shifts in the birth weight distribution are accompanied by similar shifts in the birth-weight–specific perinatal mortality curve. The differences in the minimum of the birth-weight–specific perinatal mortality rate are considerable: more than 400 gm between the “extremes” of Flanders and Finland.
We used routinely available data, which may differ across countries in the accuracy of birth weight data or in the completeness of registration of perinatal deaths. 16 However, it is unlikely that any inaccuracies would affect the mode of the birth weight distribution or the minimum of the birth-weight–specific perinatal mortality curve. The measurement of gestational age, which we used as an inclusion criterion and as a stratification variable, could also differ among countries as a result of being calculated from last menstrual period, ultrasound examination, or a combination of the two, in varying proportions. It has been shown, however, that for babies who are born at or near term, last menstrual period and ultrasound give similar results. 17
We stratified on gestational age in a number of broad categories (data not shown). Despite the large size of our dataset, the numbers of perinatal deaths were low, particularly among term births, and this prevented us from stratifying into narrower categories of gestational age. Among-country differences in gestational age distribution may still exist within each of the broad categories, for example, because of differences among countries in proportions of induced labors or elective deliveries. 18 We therefore looked at among-country differences in gestational age distribution, particularly within the 37–41 weeks category, but found that they were small (data not shown).
What could explain these among-country differences in position of the birth weight distribution? It is striking that various national populations living close to each other within a relatively homogeneous part of the world still differ so much in their birth weight distribution. Additional information on specific factors associated with birth weight would have helped the interpretation of our results, but such information was not available in an internationally comparable form. Among-country differences in the position of the birth weight distribution could be attributable to a wide range of genetic and environmental influences, such as mother's birth weight, height, weight, weight gain, age, parity, disease status, smoking, nutrition, socioeconomic status, and ethnic origin. 19–28
Because populations with a lower average birth weight also have a lower optimal birth weight, we should probably focus on possible reasons for lower birth weight that are not associated with a higher risk of perinatal death. If this line of reasoning is correct, adverse environmental conditions such as maternal disease or smoking are less likely explanations, and genetic or other intergenerational factors seem more likely. The heritability of birth weight has been estimated to be about 30% in studies on term infants 29 and between 25% and 40% in studies on monozygotic and dizygotic female twins. 30 The intergenerational transmission of low or high birth weight, however, is not restricted to genetic factors. 31 A study of cases of ovum donation revealed that birth weight was not substantially related to characteristics of the donor, but was related to the recipient's weight. 32 This relation may be a matter of intrauterine physical constraints, 33,34 or more generally a matter of smaller organs in poorly grown mothers. 31 To the extent that lower birth weight reflects smaller mothers, environmental conditions must also be considered. Favorable environmental conditions are thought to be responsible for a trend toward increases in average height attributable to a better use of the full genetic potential. 35–37 Higher living standards, like those prevailing in the Nordic countries, may have gradually produced larger mothers and larger babies over several generations.
Optimal birth weight for perinatal mortality differs among countries, depending on shifts in their birth weight distribution. This variation has two implications: one for research and one for clinical practice. The results of our study show that researchers trying to explain among-country differences in perinatal mortality should interpret differences in average birth weight very cautiously. For comparisons among populations, it has been suggested that relative measures of birth weight, such as z-scores or percentile values derived from the birth weight distribution within each population, should be used instead of absolute birth weights. 6,38,39
The results of this study also have clinical implications relating to the measurement of intrauterine growth retardation and the definition of low birth weight. Differences in optimal birth weight among countries suggest that prenatal growth curves used for the detection of growth retardation and the definition of low birth weight need to be differentiated among populations. Despite the fact that the standard birth weight criterion of 2500 gm to identify small babies at high risk of perinatal death has been questioned many times, suggestions for differentiation have been ignored in practice. 7,40 In our dataset, the percentage of births below 2500 gm ranges between 2.8% in Finland and 5.5% in Scotland, and the proportion of perinatal deaths below 2500 gm ranges between 42% and 59%. This range clearly shows the variation in predictive power of such a uniform criterion, and together with our findings strengthens the case for population-specific standards for birth weight.
We acknowledge Luis F. Lopes de Oliveira, Coimbra (Portugal), who kindly hosted one of the meetings of the EuroNatal Working Group. We also acknowledge the contribution of James Chalmers (Edinburgh, UK) to the collection of Scottish data. We are indebted to Lya den Ouden and Sabine Anthony, who prepared the Dutch data. We also thank the statistical offices in the participating countries for providing data.
1. Bakketeig LS, Hoffman HJ, Titmuss Oakley AR. Perinatal mortality. In: Bracken MB. Perinatal Epidemiology. New York: Oxford University Press, 1984.
2. Alberman E, Bergsjø; P, Cole S, et al
. International collaborative effort (ICE) on birth weight; plurality; and perinatal and infant mortality. I: Methods of data collection and analysis. Acta Obstet Gynecol Scand 1989; 68: 5–10.
3. Evans S, Alberman E, Pashley J, Hampton B. International collaborative effort (ICE) on birth weight; plurality; and perinatal and infant mortality. II: Comparisons between birth weight distributions of births in member countries from 1970 to 1984. Acta Obstet Gynecol Scand 1989; 68: 11–17.
4. Wilcox AJ, Russell IT. Birth weight and perinatal mortality I. On the frequency distribution of birth weight. Int J Epidemiol 1983; 12: 314–318.
5. Wilcox AJ, Russell IT. Birth weight and perinatal mortality II. On weight-specific mortality. Int J Epidemiol 1983; 12: 319–325.
6. Wilcox AJ, Russell IT. Birth weight and perinatal mortality III. Towards a new method of analysis. Int J Epidemiol 1986; 15: 188–196.
7. Wilcox A, Russell I. Why small black infants have a lower mortality rate than small white infants: the case for population-specific standards for birth weight. J Pediatr 1990; 116: 7–10.
8. Wilcox A, Skjaerven R, Buekens P, Kiely J. Birth weight and perinatal mortality. A comparison of the United States and Norway. JAMA 1995; 273: 709–711.
9. Bjerkedal T, Czeizel A, Hosmer DW. Birth weight of single livebirths and weight specific early neonatal mortality in Hungary and Norway. Paediatr Perinat Epidemiol 1989; 3: 29–40.
10. Buekens P, Wilcox AJ, Kiely J, Masuy-Stroobant G. Birth weight, preterm births and neonatal mortality in Belgium and the United States. Paediatr Perinat Epidemiol 1995; 9: 273–280.
11. Richardus JH, Graafmans WC, Pal-de Bruin van der KM, Amelink-Verburg MP, Verloove-Vanhorick SP, Mackenbach JP. An European concerted action investigating the validity of perinatal mortality as an outcome indicator for the quality of antenatal and perinatal care. J Perinat Med 1997; 25: 313–324.
12. Richardus JH, Graafmans WC, Verloove-Vanhorick SP, Mackenbach JP. The perinatal mortality rate as an indicator of quality of care in international comparison. Med Care 1998; 36: 54–66.
13. Braak ter CJF, Looman CWN. Regression. In: Jongman RHG, Braak ter CJF, Tongeren OFR, eds. Data Analysis in Community and Landscape Ecology. Cambridge: Cambridge University Press, 1995; 65–66.
14. Hastie TJ, Tibshirani RJ. Generalized Additive Models. London: Chapman and Hall, 1990.
15. Wilcox AJ, Skjaerven R. Birth weight and perinatal mortality: the effect of gestational age. Am J Public Health 1992; 82: 378–382.
16. Gourbin G, Masuy-Stroobant G. Registration of vital data: are live births and stillbirths comparable all over Europe? Bull World Health Organ 1995; 73: 449–460.
17. Kramer MS, McLean FH, Boyd ME, Usher RH. The validity of gestational age estimation by menstrual dating in term, preterm and postterm gestations. JAMA 1988; 260: 3306–3308.
18. Macfarlane A, Chamberlain G. What is happening to caesarian section rates? (commentary). Lancet 1993; 342: 1005–1006.
19. Kramer MS. Determinants of low birth weight: methodological assessment and meta-analysis. Bull World Health Organ 1987; 65: 663–737.
20. Cnattingius S, Axelsson O, Eklund G, Lindmark G. Smoking, maternal age, and fetal growth. Obstet Gynecol 1985; 66: 449–452.
21. Rossner S, Ohlin A. Maternal body weight and relation to birth weight. Acta Obstet Gynecol Scand 1990; 69: 475–478.
22. Kariniemi V, Rosti J. Maternal smoking and alcohol consumption as determinants of birth weight in an unselected study population. J Perinat Med 1988; 16: 249–252.
23. Lumley J, Correy JF, Newman NM, Curran JT. Cigarette smoking, alcohol consumption and fetal outcome in Tasmania 1981–1982. Aust N Z J Obstet Gynaecol 1985; 25: 33–40.
24. Dougherty CR, Jones AD. The determinants of birth weight. Am J Obstet Gynecol 1982; 144: 190–200.
25. Lekea-Karanika V, Tzoumaka-Bakoula C, Matsaniotis NS. Sociodemographic determinants of low birth weight in Greece: a population study. Paediatr Perinat Epidemiol 1999; 13: 65–77.
26. Lumey LH, Stein AD. Offspring birth weights after maternal intrauterine undernutrition: a comparison within sibships. Am J Epidemiol 1997; 146: 810–819.
27. Skjærven R, Wilcox AJ, Øyen N, Magnus P. Mothers’ birth weight and survival of their offspring: population based study. BMJ 1997; 314: 1376–1380.
28. Guendelman S, Buekens P, Blondel B, Kaminski M, Notzon FC, Masuy-Stroobant G. Birth outcomes of immigrant women in the United States, France and Belgium. Matern Child Health J 1999; 3: 177–187.
29. Langhoff-Roos J, Lindmark G, Gustavson K-H, Gebre-Medhin M, Meirik O. Relative effect of parental birth weight on infant birth weight at term. Clin Genet 1987; 32: 1–9.
30. Clausson B, Lichtenstein P, Cnattingius S. Genetic influence on birth weight and gestational length determined by studies in offspring of twins. Br J Obstet Gynaecol 2000; 107: 375–381.
31. Emanuel I. An assessment of maternal intergenerational factors in pregnancy outcome (invited commentary). Am J Epidemiol 1997; 146: 820–825.
32. Brooks AA, Johnson MR, Steer PJ, Pawson ME, Abdalla HI. Birth weight: nature or nurture? Earl Human Develop 1995; 42: 29–35.
33. Carr-Hill R, Campbell DM, Hall MH, Meredith A. Is birth weight determined genetically? Br Med J 1987; 295: 687–689.
34. Chike-Obi U, David RJ, Coutinho R, Wu S. Birth weight has increased over a generation. Am J Epidemiol 1996; 144: 563–569.
35. Eveleth PB, Tanner JM. Worldwide Variation in Human Growth
. 2nd ed. Cambridge: Cambridge University Press, 1990.
36. Lumley J, Bakoula C. Perinatal mortality in Greece and Greek-born women in Victoria. What does a ’natural experiment’ suggest? Eur J Obstet Gynecol Reprod Biol 1993; 50: 65–70.
37. Dubrova YE, Kurbatova OL, Kholod ON, Prokhorovskaya VD. Secular growth trend in two generations of the Russian population. Human Biol 1995; 67: 755–767.
38. Wilcox AJ, Russell IT. Perinatal mortality: standardizing for birth weight is biased. Am J Epidemiol 1983; 118: 857–864.
39. Hertz-Picciotto I, Din-Dzietham R. Comparison of infant mortality using a percentile-based method of standardization for birth weight or gestational age. Epidemiology 1998; 9: 61–67.
40. Rooth G. Low birth weight revised. Lancet 1980; 1: 639–641.
Keywords:© 2002 Lippincott Williams & Wilkins, Inc.
perinatal mortality; birth weight; international comparison