Comparability of published perinatal mortality rates in Western Europe: the quantitative impact of differences in gestational age and birthweight criteria

Authors


  • Participants in the EuroNatal Working Group are listed on page 1244

*Professor J. P. Mackenbach, Erasmus University Rotterdam, Department of Public Health, PO Box 1738, 3000 DR, Rotterdam, The Netherlands.

Abstract

Objective To quantify the impact of publication criteria on differences in published national perinatal mortality rates among Western European countries.

Design Descriptive study of perinatal mortality rates in Western European countries with adjustments for international differences in publication data.

Population All live births and perinatal deaths in 1994 in Western European countries.

Methods The 1994 perinatal mortality rates were obtained from national and Eurostat publications for Belgium, Denmark, Finland, France, Germany, Greece, The Netherlands, Norway, Portugal, Spain, Sweden, United Kingdom (England, Wales, Scotland, Northern Ireland). Two methods, one direct and one indirect, were used for adjusting these officially published rates for differences in registration laws or publication practices. For the indirect method adjustment factors were derived from an analysis of a large Finnish database using different cutoff points for gestational age and birthweight. For the direct method a common cutoff point was imposed for birthweight (1000g) and gestational age (28 completed weeks) on national perinatal mortality data, obtained from civil registration or hospital/obstetrics databases in each country.

Results The published perinatal mortality rates ranged from 5.4 per 1000 total births in Sweden and Finland to 9.7 in Greece and Northern Ireland. The indirect adjustment method showed that some countries apply cutoff points for registration or publication of perinatal mortality which may raise the perinatal mortality rate by up to 17% above the most commonly used threshold for including live and stillbirths. The direct adjustment method showed that a common lower limit of 1000g for birthweight or 28 weeks for gestational age would reduce the perinatal mortality rate, but by a differing extent ranging from 14% to 40%. Both adjustment methods reduced the contrast between the countries’ perinatal mortality rates, and changed their rank order.

Conclusion These quantitative results confirm that international differences in countries’ published perinatal mortality rates partly reflect differences between countries’ criteria for registration and publication of perinatal deaths.

Introduction

The term ‘perinatal mortality’ came into use in the 1940s and 1950s, when it had become clear that, despite a huge fall in the total infant mortality rate, infants’ risk of dying shortly after birth were still considerable1. The measurement of perinatal mortality proved difficult, however, because the definitions of its components (stillbirths and first week deaths among live births) are problematic2. Despite attempts of the World Health Organisation to introduce common definitions of perinatal mortality, over the years many studies have shown that there is still substantial international variation in the measurement of perinatal mortality2–5.

This variation makes it difficult to compare perinatal mortality rates for different countries. Still, perinatal mortality rates frequently have been used to monitor the achievements of a country, in terms of the quality of its obstetric and neonatal care, in comparison with other countries1. In the latter half of the twentieth century the perinatal mortality rate declined rapidly in Western Europe due both to the improvement of social conditions and to advances in health care, and the proportion of perinatal deaths which are preterm and/or have a very low birthweight has increased substantially1,2,6. The perinatal mortality rate may therefore now be much more sensitive to subtle differences between countries in criteria for registering and publishing perinatal deaths, particularly when they relate to the viability of the fetus or infant.

A study in the early 1990s showed that differences between countries in legal criteria for registration were considerable2. All countries use some threshold, in terms of gestational age or birthweight, for registering stillbirths, and some use a threshold for registering live births, and hence deaths among them, as well. Despite the fact that in the 1980s and 1990s these criteria shifted to lower gestational age and birthweight cutoff points in several countries, they still differ substantially between countries7,8. Although many studies have now pointed to this problem, its quantitative impact on comparisons of published perinatal mortality rates is largely unknown9, except for a study from the late 1980s focusing on the effect of introducing a common lower threshold for birthweight10. It should be noted, however, that the legal criteria for registering perinatal deaths within a country are not necessarily identical to the criteria used for including perinatal deaths in published tables. Sometimes, registration criteria include a wider range of gestational ages and/or birthweights than are actually included in the official national statistics, in order to promote completeness of registration within the range included in published data. Also, some countries publish their perinatal mortality rates in two forms, one for national statistics and one for international comparisons. For example, the Nordic countries (Denmark, Norway, Sweden and Finland) publish harmonised figures for comparison between them, and such harmonised figures are also supplied to international statistical agencies like Eurostat. Consequently, to evaluate the international comparability of officially published perinatal mortality rates it is important to look at the end result of applying criteria for both registration and (international) publication in each country.

In international comparisons of officially published perinatal mortality rates there is another potential source of bias which has to be mentioned briefly. There may also be differences between countries in registration practices (i.e. the extent to which people comply with the legal registration requirements). Compliance with the legal registration requirements is the responsibility of the birth attendant and/or the parents, and is dependent on the extent of their knowledge of the criteria2,11,12, and on social, cultural and economic factors such as whether they wish to bury the infant or to obtain certain benefits2,13,14. As a consequence, births which should be declared may stay undeclared (and vice versa), and births which are declared may be declared incorrectly, so stillbirths may be registered as live births followed by a neonatal death (and vice versa). The latter may affect the perinatal mortality rate if the fetus's gestational age is below the threshold for registration of a stillbirth. Indications of a lack of compliance have been found in many countries15–18. Although there may well be differences between countries in the extent of the problem11, such differences in registration practice are outside the scope of this paper, because in-depth studies are needed to investigate them.

The purpose of the study reported here was to investigate the quantitative impact on officially published perinatal mortality rates of differences between countries in criteria for registering perinatal deaths and including them in published data. This paper focuses on lower limits for birthweight and gestational age, because these were identified as the main sources of variation between countries2,9.

Methods

The EuroNatal Study

This study was performed within the framework of the EuroNatal study19, a collaborative European effort aiming to establish the extent to which the perinatal mortality rate can be used as an outcome indicator for perinatal care within Europe. The study was co-ordinated in The Netherlands. Participating countries were: Belgium, Denmark, Finland, France, Germany, Greece, Norway, Portugal, Spain, Sweden, The Netherlands, and the UK (England, Wales, Scotland, and Northern Ireland).

Officially Published Perinatal Mortality Rates

As a starting point for our analyses of differences between countries in the officially published perinatal mortality rate, we chose data sources that are commonly used for purposes of international comparison such as: national statistical publications (e.g. statistical yearbooks) and Eurostat publications20. Some countries publish separate rates for international comparison, and in that case these harmonised rates were chosen as the starting point for our analyses. The perinatal mortality rate was defined as the number of stillbirths and first-week deaths per 1000 total births. Confidence intervals for the perinatal mortality rates were calculated using the binomial distribution. Data on perinatal mortality and information on the criteria used for these published rates were collected for 1994 or if this was not available, a year as close to 1994 as possible. Two important areas of international variation were identified: the lower gestational age or birthweight limit for inclusion of live births, and the lower gestational age or birthweight limit for inclusion of stillbirths. We found some other differences as well, but these are unlikely to have a major impact on the (comparability of the) perinatal mortality rates. For example, the definition of the early neonatal period also differed between countries, and varied from six to seven days. However, an analysis of unpublished Dutch data suggests that the number of deaths between day six and day seven is very small, and that this difference will not affect the perinatal mortality rate by more than 1%.

Indirect Adjustment Method

The various criteria for inclusion of live births and stillbirths used in the participating countries were applied to a Finnish database. This contained all 194,904 births in the years 1993–1995, and was compiled from the Finnish Medical Birth Register (STAKES) and the cause of death register (Statistics, Finland). This data set was chosen for its low gestational age and birthweight limits for registering births, and for its low proportion of missing gestational ages and birthweights (<2%). We started by calculating the perinatal mortality rates in this data set for live births without a threshold and stillbirths at 28 or more weeks of gestation. We then calculated the proportional changes in these perinatal mortality rates after application of different cutoff points for birthweight and gestational age. These cutoff points were chosen to reflect the criteria for registration and publication of perinatal mortality in the other countries participating in the study. Finally, the proportional changes observed in the Finnish data set were used as adjustment factors for an upward or downward adjustment of the officially published, national perinatal mortality rates, depending on the specific cutoff points applying in the publications for that country. The main advantage of this indirect approach is that it can be applied without having access to detailed national data on the distribution of births by gestational age and birthweight. The disadvantage is that it assumes common distributions of birthweight and gestational age.

Direct Adjustment Method

The assumption of a similar birthweight and gestational age distribution in all countries may not be valid. Therefore, if information was available, perinatal mortality rates were also adjusted by directly applying common cutoff points for gestational age and birthweight. Participants in all countries were asked to send anonymous individual level records on live births, stillbirths, and early neonatal deaths, including data on gestational age and birthweight. Data were obtained for the year 1994 or, if this was not available, for a year as close to 1994 as possible.

To calculate an adjusted perinatal mortality rate, we used two relatively high cutoff points which could be applied to all the countries in the analysis: a lower gestational age limit of 28 weeks and a lower birthweight limit of 1000g for all births. These are also cutoff points which have been recommended for international comparisons by the World Health Organization (WHO)21,22. The necessary information was available for Denmark, Finland, Belgium, The Netherlands, Northern Ireland (UK), Norway, Scotland (UK), and Sweden. In England and Wales information was available on birthweight only. In Spain, data were available but gestational age and birthweight were missing in about 20% of cases, and therefore Spanish data were not included in this part of the analysis. The databases with information on gestational age and birthweight were obtained from civil registration, hospital databases or national professional databases. It was assumed that missing data on gestational age and birthweight were random. Fewer than 2.5% of gestational ages and birthweights were missing from the databases included in the analysis, except in Norway where 9.8% of gestational ages were missing. Gestational age and birthweight data from the Dutch National Obstetrics and Neonatology Registries were incomplete. Births were not recorded by general practitioners (who, however, attend less than 5% of all births), by 11% of midwife practices and by 16% of level I hospitals. Because the completeness depended on the level of care, and birthweight distributions and perinatal mortality rates differed between levels of care, we reweighted the Dutch data by level of care, assuming that births with GPs have gestational age and birthweight distributions, as well as perinatal mortality risks, equal to those with midwives (details available from the authors).

Results

The gestational age and birthweight cutoff points used in published perinatal mortality rates are shown in the first two columns of Table 1. The criteria for including stillbirths differ substantially between countries: 28 weeks of gestational age in most countries, but 26 weeks in Spain, 24 weeks in the UK and 500g birthweight in Germany. The perinatal mortality rates in published national statistics (‘Nordic’ statistics in the case of the Nordic countries) are shown in the third column, and range between 5.4 (Finland and Sweden) and 9.7 (Greece and Northern Ireland) per 1000 total births. The perinatal mortality rates published by Eurostat are shown in the final column of Table 1. As they are derived directly from the nationally published figures, they might be expected to be identical but that was not the case for The Netherlands and France. For The Netherlands the difference is due to the fact that Eurostat used an incomplete count of stillbirths. For France, stillbirths at 26 or 27 weeks after conception (28 or 29 weeks after last menstrual period) are included in national statistics but not in Eurostat statistics.

Table 1.  Lower gestational age and birthweight limits, and perinatal mortality rates in European countries in 1994
CountryLower GA (wks) and/or BW (g) limit for published PMRsaPMR per 1000 total births (95% CI)
Live birthsStillbirthsNational and ‘Nordic’ statisticsbEurostatc
  1. a GA = gestational age, BW=birthweight, PMR = perinatal mortality rate

  2. b The following statistical publications were used: Belgium: Institut National de Statistiques 1999; Denmark, Finland, Norway, and Sweden: Yearbook of Nordic Statistics 1996; France: Annuaire Statistique de la France 1998; Germany: Statistisches Jahrbuch 1997, Bundesrepublik Deutschland; Greece: Greek National Statistical Office 1999; The Netherlands: CBS, Statistisch jaarboek 1999; Portugal: Instituto Nacional de Estatistica 1998; Spain: Instituto Nacional de Estadistica (INE) 1994; United Kingdom: Annual Abstract of Statistics 1998.

  3. c Eurostat Demographic Statistics 1997.

  4. d 1993

  5. e 1995

BelgiumNone≥28wks7.3 (6.8–7.8)Not available
DenmarkdNone≥28 wks7.4 (6.8–8.0)7.4 (6.8-8.0)
FinlandNone≥28 wks5.4 (5.0–5.9)5.4 (5.0-5.9)
France≥22 wks or ≥500 g≥287.4 (7.2–7.6)7.0 (6.8-7.2)
GermanyNone≥500 g6.4 (6.2–6.6)6.4 (6.2-6.6)
GreeceNone≥289.7 (9.1–10.4)9.7 (9.1-10.4)
The NetherlandseNone≥288.1 (7.7–8.5)7.6 (7.2-8.0)
Norway≥16 wks≥287.5 (6.8–8.2)7.5 (6.8-8.2)
PortugalNone≥289.2 (8.6–9.8)9.3 (8.7-9.8)
SpainNone≥266.6 (6.3–6.8)6.5 (6.2-6.7)
SwedenNone≥285.4 (5.0–5.8)5.4 (5.0-5.8)
UKNone≥248.9 (8.7–9.1)8.9 (8.7-9.1)
England/WalesNone≥248.9 (8.7–9.2)Not available
ScotlandNone≥249.0 (8.4–9.6)Not available
N. IrelandNone≥249.7 (8.7–10.7)Not available

The different gestational age and birthweight cutoff points presented in Table 1 were applied to a Finnish database, and the results of this ‘indirect’ method are shown in Table 2. The most commonly applied cutoff points for live births (none) and stillbirths (28 or more weeks of gestational age) form the reference point for the analysis. Differences between countries in cutoff points for stillbirths have a substantial impact on the officially published perinatal mortality rate. The most restrictive cutoff point, 28 weeks of gestation, is used in publications for most of the participating countries, and the least restrictive cutoff point, 500g birthweight, is used in publications for Germany. The analysis of Finnish data suggests that a cutoff point of 500g raises the perinatal mortality rate in this country by 17%, as compared with countries where the cutoff point is 28 weeks. Even a lower threshold of 24 instead of 28 weeks of gestation for stillbirths raises the perinatal mortality rate by 15%, according to this analysis of Finnish data.

Table 2.  The effect of applying different gestational age (GA) and different birthweight (BW) cut-off points for the inclusion of live births and stillbirths on the perinatal mortality rate (PMR) in a Finnish database, 1993-1995a.
Lower thresholds for inclusion of live and stillbirths    
Live birthsStillbirthsNo. of perinatal deathsPMR per 1000 births(95% CI)Relative PMR
  1. a No. of births: 194904. Source: STAKES and cause of death register

None≥28 weeks GA9815.15(4.84 – 5.47)1.00 (ref.)
None≥26 weeks GA10425.47(5.14 – 5.79)1.06
None≥24 weeks GA11295.91(5.57 – 6.26)1.15
None≥500 g BW1148 (5.70 – 6.39)1.17
≥16 weeks GA≥28 weeks GA9815.15(4.84 – 5.47)1.00
≥22 weeks GA≥28 weeks GA9815.15(4.84 – 5.47)1.00
≥1000g BW≥1000g BW7724.07(3.78 – 4.35)0.79
≥28 weeks GA≥28 weeks GA7944.16(3.87 – 4.45)0.81

Differences between countries in gestational age cutoff points for live births (16 weeks or more in Norway, 22 weeks or more in France) do not appear to have a substantial impact on the officially published perinatal mortality rate, as is shown by the 0% change in perinatal mortality rate in Table 2.

The final two rows of Table 2 show the effect of applying the WHO recommendations for international comparisons (≥1000g for birthweight and ≥28 weeks for gestational age for all births).

The results of the direct method are shown in Table 3. Here, common thresholds of ≥28 weeks for gestational age and ≥1000g for birthweight were applied directly to all births in the national datasets of seven countries. The changes in the number of live births, stillbirths, early neonatal deaths and in the perinatal mortality rate, as compared with the published figures, are shown in the Table. The extent of the changes depends on, first, the national criteria used for registering and publishing perinatal deaths, and, second, the country-specific distribution of births by gestational age and birthweight. Applying common cutoff points for gestational age or birthweight to live births did not have a clearly noticeable effect in any of the seven countries, due to the fact that such a small proportion of live births was below one of the two thresholds in these countries. There are important changes in the number of stillbirths, however, depending on the nationally used criteria for registering and publishing perinatal deaths. In many countries, stillbirths are registered and included in published tables from 28 weeks of gestation onwards, so applying the common gestational age criterion does not have an effect, except in Belgium where the official statistics do include stillbirths below the official threshold for gestational age. However, in Scotland and Northern Ireland it has a major impact. The results of applying the common cutoff points on the number of first-week deaths are even more striking. For example, when we restrict births to gestational ages of ≥28 weeks of gestation, the absolute number of early neonatal deaths decreases from 176 to 83 in Scotland (detailed results not shown), leaving only 47% of the officially published first-week deaths. Smaller, but still substantial, effects are seen for many other countries. As a result of all these changes for each of the components of perinatal mortality, applying the common cutoff points had important impacts on the officially published perinatal mortality rates. Overall, these rates were reduced substantially, by between 10% and almost 40% (last columns of Table 3). A stronger reduction was found after exclusion of all births with birthweight <1000g than after exclusion of all births before 28 weeks of gestation. More importantly, the extent of reduction differs importantly between countries, and is largest in Scotland, smallest in Denmark.

Table 3.  The effect of applying common gestational age (≥28) and birthweight (≥1000) cut-off points on the perinatal mortality rate (PMR) in national datasets.a,e
 Proportional change of numbers published in national statisticsb  
 Live birthsStillbirths Early neonatal deathsPMR
Country≥28 wks≥1000 g≥28 wks≥1000 g≥28 wks≥1000 gPMR ≥28w /PMR Nt.st.PMR ≥1000 g /PMR Nt.st
  1. a NA= not available

  2. b Nt.st. according to statistical publications

  3. c Information in 1993

  4. d Information in 1995

  5. The following registers and databases were used for calculations: Belgium: Institut National de Statistiques; Denmark: Danish birth Register; Finland: Finnish Medical Birth Register (STAKES) and Cause of death register (Statistics Finland); Norway and Sweden: Medical Birth Register; Germany: Statistischen Bundesamt, Wiesbaden; The Netherlands: SIG (LVR1/LVR2/LNR); England/Wales: Office for National Statistics; Scotland: National Health Service, ISD (SMR2); N-Ireland: Northern Ireland's Child Health System (CHS)

Belgium0.991.000.930.820.690.670.830.76
Denmarkc1.001.001.000.930.730.750.900.86
Finland1.001.001.000.950.590.590.790.76
FranceNA.NA.NA.NA.NA.NA.NA.NA.
GermanyNA.NA.NA.NA.NA.NA.NA.NA.
GreeceNA.NA.NA.NA.NA.NA.NA.NA.
The Netherlandsd1.001.001.000.860.510.490.810.72
Norway1.001.001.000.870.580.500.840.73
PortugalNA.NA.NA.NA.NA.NA.NA.NA.
SpainNA.NA.NA.NA.NA.NA.NA.NA.
Sweden1.001.001.000.930.700.680.870.82
UKNA.NA.NA.NA.NA.NA.NA.NA.
England/WalesNA.1.00NA.0.68NA.0.52NA.0.63
Scotland1.001.000.740.690.470.420.660.61
N-Ireland1.001.000.790.720.550.570.710.67

In order to compare the two methods, Table 4 shows the final results of both. Both adjustment methods reduce the variation between countries in perinatal mortality rates, and change the rank order of countries, as compared with the variation and rank order observed in official publications. With the indirect method we found that compared with most countries, Spain includes about 6% more perinatal deaths in its published perinatal mortality rate, the UK includes 15% more, and Germany includes 17% more (see Table 2). Adjustment for these differences produces the perinatal mortality rates in the first two columns of Table 4. This results in some changes in the ranking of countries, although the highest (Greece) and the lowest (Sweden and Finland) positions remain unchanged. The direct method, which could only be applied to a limited number of countries, produces largely similar changes, as shown by the third and fourth column of Table 4. However, there are also some important differences between both methods. For example, when applying the birthweight criteria to The Netherlands the direct adjustment method has a stronger (downward) effect on the perinatal mortality rate than the indirect method. This must be due to differences between the Netherlands and Finland in the distribution of births and deaths by birthweight, and suggests that the indirect method does not always perform well.

Table 4.  Perinatal mortality rates (PMR) after indirect and direct adjustment (≥28 gestation, or ≥1000 g birthweight)a
 PMR per 1000 births
 After indirect adjustment, based on Finnish dataAfter direct adjustment, based on country-specific data
Country≥28≥1000 g≥28≥1000 g
  1. a NA.: not available

  2. b Information in 1993

  3. c Information in 1995.

Belgium5.95.86.15.6
Denmarkb6.05.86.76.4
Finland4.44.34.24.1
France6.05.8NA.NA.
Germany4.44.3NA.NA.
Greece7.87.7NA.NA.
The Netherlandsc6.66.46.65.8
Norway6.15.96.35.5
Portugal9.29.2NA.NA.
Spain5.04.9NA.NA.
Sweden4.44.34.74.5
UK6.26.1NA.NA.
England/Wales6.26.1NA.5.6
Scotland6.36.25.95.5
N. Ireland6.86.76.96.5

In order to visualise the effect of the direct adjustment, Fig. 1 shows both the officially published perinatal mortality rates, and the rates adjusted according to the direct method, on the basis of a common gestational age cutoff point of ≥28 weeks. Fig. 1 illustrates the slight reduction from 43% to 39% in the difference between the lowest and the highest rate as well as the change in rank order. After adjustment, Finland no longer shares its leading position with Sweden, while Denmark and the Netherlands now have almost the same perinatal mortality rate as Northern Ireland.

Figure 1.

Perinatal mortality rates and 95% confidence intervals based on national publications (open bars) and revised perinatal mortality rates based on a common criterion, using births of 28 weeks gestation onwards (dark bars).

Discussion

Published perinatal mortality rates differ substantially between Western European countries. Our analysis shows that different cutoff points for inclusion of perinatal deaths strongly affect the comparability of these perinatal mortality rates. Both the indirect and the direct method of adjustment suggest that application of common criteria for inclusion of perinatal deaths would reduce the range of variation between countries, and would slightly change the rank order of countries with regard to their perinatal mortality rate.

Our conclusions apply not only to the perinatal mortality rates as published in national publications of vital statistics, but also to Eurostat and other international publications. Figs. in Eurostat publications20are often directly derived from national publications, but sometimes differ from nationally published figures, without specified or obvious reason. In The Netherlands, for example, the nationally published rate (8.1/1000) is different from the figure published by Eurostat (7.6/1000) (Table 1). This difference is due to the fact that Eurostat used an incomplete count of stillbirths. Since Eurostat figures are intended to be used for comparison of countries, we recommend that a stronger effort is made to base Eurostat figures on common thresholds for including perinatal deaths, and if this is absolutely impossible, that the differences in thresholds are more clearly indicated in Eurostat publications.

The results of the indirect and direct adjustment methods are not identical. Because the indirect method assumes that the distribution of births and deaths by gestational age or birthweight of Finland applies to other countries as well, the difference between both methods is likely to be due to violations of this assumption. This is confirmed by a detailed look at some of the changes arising from using the direct method. Differences between countries in the extent of change in the number of first-week deaths after application of a common cutoff point of ≥28 weeks for gestational age and of ≥1000g for birthweight, as observed in Table 3, are due to differences between countries in the distribution of births and deaths by gestational age and birthweight. First-week deaths among babies born before 28 weeks of gestation, as a proportion of the number included in the officially published perinatal mortality rate, ranged between 53% in Scotland and 27% in Denmark. Similar results were found for the 1000g birthweight threshold. These differences between countries in the distribution of births and deaths by gestational age and birthweight had important effects on the outcome of the direct method of adjustment. For example, Finland and Sweden reported the lowest perinatal mortality rate in their national publications, using the same criteria. After introducing the common, relatively high gestational age criterion of ≥28 weeks, however, the Finnish perinatal mortality rate was lowered to 79% of the nationally published rate, and the Swedish rate was reduced to 87%. This is due to the fact that Sweden had a lower proportion of first-week deaths before 28 weeks of gestation (30%) as compared with Finland (41%), and resulted in a change in ranking: after the direct adjustment Sweden ranked second lowest after Finland. These differences in distribution of births and deaths by gestational age may partly result from different strategies in obstetric care. Obviously, the indirect adjustment method, based on the distribution observed in one particular country, in this case Finland, may be misleading. Nevertheless, it does give an indication of the direction and order of magnitude of the bias present in officially published perinatal mortality rates, and for many countries is the only way of evaluating this bias.

This study focused on the bias introduced by differences in cutoff points for gestational age and birthweight. Other differences in official criteria were ignored, mainly because we expected these differences to have a smaller impact. Previous studies have pointed to differences in, among other things, the definition of signs of life (important for the distinction between stillbirths and live births), the definition of the first week (see our comments above), and the rules for registration of liveborn children who die before their birth is registered2,23. It might be worthwhile investigating the quantitative impact of these differences in future studies.

Another limitation of our study is perhaps more important. We did not investigate differences in registration practice between countries, for example differences in under-reporting of perinatal deaths because of non-compliance with the legal registration requirements15–18. Perinatal mortality rates are probably subject to under-reporting, and the extent of under-reporting may differ between countries11. For that and other reasons, the adjusted perinatal mortality rates as reported in this paper should not be seen as the ultimate truth, but as a correction of the nationally published rates to values which are more closely, but still not perfectly, comparable between countries.

Another limitation is that we accepted measures of gestational age and birthweight, as registered in different national databases, at face value. In particular, gestational age may be measured by ultrasound, last menstrual period, or a combination of both. Differences can occur between countries in the way gestational age is measured. Although both methods correspond reasonably well in term births, discrepancies may occur in preterm births24. Differences between countries in the measurement of gestational age may therefore have influenced our results, but due to lack of information it is difficult to say how and to what extent.

Our results show that differences within Europe in the perinatal mortality rate do exist, but are in reality slightly smaller and of a different rank order than the official publications suggest. For the purpose of international comparisons, perinatal mortality figures should be adjusted for differences in cutoff points for gestational age and birthweight. Within each country, registration of stillbirths and live births should be based on a low threshold (e.g. ≥22 completed weeks of gestation, or ≥500g birthweight), which does not need to be the same in all countries. However, before perinatal mortality rates are tabulated for international comparison, common thresholds should be imposed on the original birth and death data in all countries, for example following the recommendations of the World Health Organisation (≥28 completed weeks of gestational age or ≥1000g birthweight)21,22. These recommendations, however, are already quite old and imply that a substantial part of perinatal mortality is left out of consideration. Perhaps it is time for a downward revision of these recommendations. In addition, our own experience with collecting such data shows that applying common lower thresholds may be unrealistic, because many countries do not routinely include birthweight and gestational age in data compiled at birth registration. In any case, this will never eliminate all comparability problems, because there may be differences between countries in the frequency of terminations of pregnancy before these registration limits are reached. For example, screening for congenital malformations and consequent medical abortion before 28 weeks of gestation may have a substantial effect on the perinatal mortality rate calculated from 28 weeks of gestation onwards25. At the other end of the ‘time-window’ for registering perinatal mortality, there may be some other remaining problems, because in some countries there is an active policy of resuscitation at birth followed by a high proportion of withdrawal of life-sustaining treatment at the end of the first month or even later26. This will also influence the international comparability of perinatal mortality rates, but could at least partly be remedied by extending perinatal mortality to include all first month deaths instead of first week deaths only, or by compiling stillbirths and neonatal deaths separately27.

If imposing common cutoff points is not feasible, indirect methods may permit a crude adjustment of the nationally published perinatal mortality rates. Adjustment factors will then have to be derived from an external and more detailed dataset, as we did with data from Finland. Such adjustment will be sensitive to assumptions about the shape of the distributions.

We conclude that differences between Western European countries in perinatal mortality rates are biased by differences in criteria for registering and publishing perinatal mortality, and that more should be done to harmonise perinatal mortality rates before they are included in international overviews.

The EuroNatal Working Group Willem Aelvoet (Brussels, Belgium); Fiona Alderdice (Belfast, UK); Marianne Amelink (Leiden, The Netherlands); Leiv Bakketeig (Odense, Denmark); Chryssa Bakoula (Athens, Greece); Birgit Bødker (Copenhagen, Denmark); Francisco Bolumar (Alicante, Spain); Leslie Davidson (Oxford, UK); Mika Gissler (Helsinki, Finland); Anna-Liisa Hartikainen (Oulu, Finland); Jan Holt (Bodø, Norway); Monique Kaminski (Villejuif, France); Jens Langhoff-Roos (Copenhagen, Denmark); Johannes Leidinger, (München, Germany); Vasso Lekea (Athens, Greece); Gunilla Lindmark (Uppsala, Sweden); Luis F. Lopes de Oliveira (Coimbra Portugal); Godelieve Masuy-Stroobant (Louvain-la-Neuve, Belgium); Karin van der Pal (Leiden, The Netherlands); Gillian Penney (Aberdeen, UK); Maria da Purificação Araújo (Lisbon, Portugal); Paula Rantakallio (Oulu, Finland).

Acknowledgements

The EuroNatal study was funded by the BIOMED 2 programme of the European Commission (BMH4-CT96-0746). The authors would like to thank Ms L. den Ouden and Ms S. Anthony who prepared the Dutch data, the statistical offices in the participating countries for providing data, and Mr J. Chalmers for the collection of Scottish data.

Ancillary