Clarity and consistency in stillbirth reporting in Europe: why is it so hard to get this right?

Abstract Background Stillbirth is a major public health problem, but measurement remains a challenge even in high-income countries. We compared routine stillbirth statistics in Europe reported by Eurostat with data from the Euro-Peristat research network. Methods We used data on stillbirths in 2015 from both sources for 31 European countries. Stillbirth rates per 1000 total births were analyzed by gestational age (GA) and birthweight groups. Information on termination of pregnancy at ≥22 weeks’ GA was analyzed separately. Results Routinely collected stillbirth rates were higher than those reported by the research network. For stillbirths with a birthweight ≥500 g, the difference between the mean rates of the countries for Eurostat and Euro-Peristat data was 22% [4.4/1000, versus 3.5/1000, mean difference 0.9 with 95% confidence interval (CI) 0.8–1.0]. When using a birthweight threshold of 1000 g, this difference was smaller, 12% (2.9/1000, versus 2.5/1000, mean difference 0.4 with 95% CI 0.3–0.5), but substantial differences remained for individual countries. In Euro-Peristat, missing data on birthweight ranged from 0% to 29% (average 5.0%) and were higher than missing data for GA (0–23%, average 1.8%). Conclusions Routine stillbirth data for European countries in international databases are not comparable and should not be used for benchmarking or surveillance without careful verification with other sources. Recommendations for improvement include using a cut-off based on GA, excluding late terminations of pregnancy and linking multiple sources to improve the quality of national databases.


Introduction
R ecent work by international collaborations has called attention to the health burden associated with stillbirth and to the absence of, or very slight declines, in stillbirth rates in high-income countries. [1][2][3][4] These reports have also revealed high heterogeneity in stillbirth rates across countries with comparable standards of living and health systems. Rates of stillbirth are over twice as high in those countries with the highest rates compared with those with the lowest. 3 Since the causes of up to one-half of stillbirths remain unknown and many are associated with sub-optimal health care, 5,6 it is essential to improve the availability and quality of data on these deaths to tackle this important health problem and identify the levers for achieving continuous improvement.
Despite longstanding registration of births and deaths in vital statistics and birth registries and recommendations by the World Health Organization (WHO) to collect data on stillbirths from 22 weeks' gestational age (GA), many countries lack complete and reliable data on stillbirths, especially early stillbirths. [7][8][9] Globally, differences between countries in criteria and practices for recording stillbirths render international benchmarks unreliable for fetal deaths before the third trimester of pregnancy. This limitation is acknowledged by the WHO which has recommended international comparisons of stillbirth rates only for births with a birthweight of !1000 g or, in more recent work, !28birthweeks' GA. 4 The comparability of stillbirth rates in Europe has been investigated by the Euro-Peristat network, a European research network that aims to improve monitoring and reporting of perinatal health indicators and that periodically compiles data using a common protocol in 31 European countries. Data on births and fetal deaths are collected by GA and birthweight groups which allow the application of denominators based on both these criteria. This also makes it possible to harmonize the population studied, despite the differences between countries in Europe in thresholds for recording stillbirth. 10,11 Euro-Peristat has also collected information about terminations of pregnancy (TOPs), which are inconsistently reported and can have a strong impact on rates and trends, especially at early gestations. 12,13 Analyses of these data have quantified the extent to which these differences influence stillbirth rates and thus their comparability across countries. 9 Euro-Peristat also illustrated the importance of improving data on early stillbirths by showing that up to 27% of all stillbirths with a birthweight of 500 g or more occur between 500 and 999 g and 33% occur between 22 and 27 weeks' GA. 10 So, while the exclusion of early stillbirths may improve comparability, it substantially underestimates the health burden and creates discrepancies with neonatal mortality which is measured for live births of any gestation. In recent analyses, Euro-Peristat concluded that stillbirth rates could be reported for births at 24 weeks' GA and over with good reliability 8 although this could not currently be extended to stillbirths at 22 and 23 weeks' GA. 14 Euro-Peristat only collects new data periodically, 3 and therefore data to assess annual trends in stillbirth in Europe must be taken from Eurostat, the official statistical system for demographic and health data in Europe based on data reported by national statistics offices. In 2011, Eurostat published implementing regulations for stillbirth reporting as part of regulation governing reporting of causes of death (EU 328/2011) 15 under the regulation on community statistics on public health and health and safety at work (EU 1338/2008 with amendment EU 2019/1700). 16 This regulation specifies that stillbirths be reported principally by birthweight criteria and that data be provided separately for stillbirths weighing 500-999 g and !1000 g. Providing data on stillbirths by year is compulsory, but data by birthweight group is voluntary. A separate department at Eurostat collects demographic statistics and includes information on stillbirths !28 weeks.
Given the importance of good quality data on stillbirths for health monitoring, we sought to assess the concordance between Eurostat routine statistics and data collected within the Euro-Peristat research network where quality checks based on GA and birthweight limits are applied.

Methods
Data on stillbirths and live births were abstracted from Eurostat databases and compared with birth data collected for the Euro-Peristat project for the year 2015.
Thirty-one countries participated in the Euro-Peristat data collection (28 EU Member States at the time, Iceland, Norway and Switzerland). In most countries, data on stillbirths were taken from medical birth registers or demographic statistics, with largely mandatory provision and good coverage. 10,17 When there were several birth data sources, the country team decided which was the most reliable. Euro-Peristat collects data on births !22 weeks' GA or when GA is missing, with a birthweight !500 g. When countries cannot provide data using this definition, they use their national definition and specify this (see Supplementary Appendix for sources and criteria). Data on live births and stillbirths are collected for each gestational week and also by birthweight in 500-g intervals with options for specifying the number of births with missing data on GA or birthweight. Data are collected for stillbirths and TOPs separately when countries are able to distinguish between these outcomes.
As part of its Causes of Death Statistics, Eurostat compiles annual data on stillbirths for the 31 countries included in Euro-Peristat. 18 Data are based on each country's national definition. As specified in regulations, 15 stillbirths are also collected in two groups: (i) stillbirths with birthweight from 500 to 999 g or (when birthweight does not apply) GA from 22 to 27 weeks, or (when neither of the two applies) crown heel length from 25 to 34 cm and (ii) stillbirths with birthweight !1000 g or (when birthweight does not apply) GA !28 completed weeks, or (when neither of the two applies) crown heel length !35 cm. National definitions for stillbirth were abstracted from meta-data files 19 to create Supplementary table S1. Verification of the accuracy of these definitions was not undertaken.
Stillbirth rates are also reported in Eurostat's European Demographic Statistics defined as late fetal deaths at 28 weeks or over. 20 However, unlike the Cause of Death statistics, there is no legislated data collection guidelines for stillbirth and the instructions provided in the most recent 2015 Demographic Statistics manual do not specify GA reporting criteria. 21 Therefore, although we included these data, we conducted our primary analysis on Cause of Death data. Data on live births were taken from demographic statistics, as live births are not collected for Cause of Death statistics.
Stillbirth rates were calculated per 1000 live and stillbirths. Since we did not have data on live births by birthweight or GA for Eurostat, we used all live births for the calculation of total births for both data sources regardless of the cut-off used for stillbirths. As very early live births <500 g, <1000 g or <28 weeks are infrequent (<1% of births), this imprecision in the denominator will have a minor effect on stillbirth rate estimates, although variation may be higher in smaller countries. We separately calculated stillbirth rates including and excluding late TOPs (!22 weeks) when data were available. Using Euro-Peristat data, we also calculated the proportion of stillbirths with missing data on birthweight and on GA. For each country, we computed absolute and relative differences in stillbirth rates from both sources for all reported stillbirths and using 500-g, 1000-g and 28-week thresholds. For these analyses, Euro-Peristat was considered as the reference. Table 1 provides data on live births and stillbirths for the two data sources. The number of total births ranged from a little over 4000 in Iceland and Malta to over 700 000 in Germany, France and the UK. Estimates were similar between Eurostat and Euro-Peristat, with most discrepancies totaling less than 1%. One exception was Luxembourg (11%). Stillbirths are presented as all reported stillbirths and using thresholds of 500 and 1000 g. For Euro-Peristat, most countries used the Euro-Peristat definition of !22 weeks for all stillbirths, unless this was not possible, as detailed in the Supplementary Appendix. For Eurostat, definitions, abstracted from meta-data files, are presented in Supplementary table S2. Some countries were missing Eurostat stillbirth data for birthweight groups in 2015, even though data were available for adjacent years (for instance, Italy had data for 2016 and 2017, Sweden for 2017 and Iceland for 2013 and 2014). Several countries provided data only by GA to Euro-Peristat (Denmark and Portugal) so we used 22 weeks as a proxy for 500 g and 28 weeks for 1000 g.

Results
The extent of differences between the two sources in the number of stillbirths varied widely between countries. The number of stillbirths was exactly or almost (65 cases) the same for 12 countries. In contrast, Ireland reported over 50% more cases to Euro-Peristat compared with Eurostat, while the reverse was true for Luxembourg, France and Slovenia which reported >40% cases to Eurostat compared with Euro-Peristat. Information on birthweight was missing for 8.7% of all stillbirths (1817/20 774). The country average was 5.0% and the highest proportions of missing values were observed for Cyprus (29%), Italy (26%), Hungary and France (both 22%) and Spain (16%). Twelve countries reported no missing cases. Table 1 also presents Euro-Peristat data on the number of TOPs. Several countries cannot distinguish terminations in overall stillbirth data (Belgium, Cyprus and the Netherlands), whereas other countries do not record terminations (Germany). Some countries do not Clarity and consistency in stillbirth reporting in Europe 201 have terminations because they are not authorized or are rare at or after 22 weeks (Croatia, Estonia, Iceland, Ireland, Lithuania, Malta, Poland, Slovakia, Spain and Sweden). Figure 1 illustrates substantial differences in overall stillbirth rates from the two sources. Countries are sorted based on the Euro-Peristat stillbirth rate, without including terminations, although how the addition of terminations affects rates is indicated by a separate bar. This illustrates that the inclusion of terminations in Eurostat stillbirth statistics is one reason for the observed differences in rates, particularly affecting France, Luxembourg and Slovenia. Of note, Finland reports terminations to Euro-Peristat, but does not include them in stillbirth statistics provided to Eurostat. Other differences which are not due to terminations can also be observed, for instance in Hungary, Ireland and Latvia. Tables 2 and 3 provide stillbirth rates using a lower limit of 500 and 1000 g, respectively. For stillbirths with a birthweight of 500 g or more, the difference between the averages for Eurostat data (4.4/ 1000, Standard Deviation (SD) 1.7) and Euro-Peristat data (3.5/ 1000, SD 0.9) was pronounced, due in part to terminations, as seen previously. Differences in stillbirth rates for birthweight of 1000 g or more were smaller between Eurostat data (2.9/1000, SD 0.9) and Euro-Peristat data (2.5/1000, SD 0.8) (table 3), but substantial differences existed for individual countries.
Eurostat demography statistics showed larger disparities when compared with Euro-Peristat (Supplementary table S3). While data are reported as stillbirths at !28 weeks, this threshold does not seem to be applied in many countries, leading to substantial overestimation of rates. Comparisons with Euro-Peristat revealed that some countries provided data on all stillbirths instead of late stillbirths (Estonia, Germany, Latvia, Lithuania, Sweden, Switzerland and UK), while elsewhere other definitions were used (Czech Republic, Greece, Hungary, Italy and Norway). This table also presents Euro-Peristat data on stillbirths with missing GA. GA was missing less often than birthweight, with only Cyprus and Spain reporting proportions >5%.

Discussion
Our comparison of stillbirth rates in Europe based on data reported by national statistical offices to the official European statistical office, Eurostat, and by the Euro-Peristat research network revealed substantial discrepancies between the two sources. These discrepancies remained even after adopting a common inclusion limit of 1000 g and ranged from rates that were 52% lower to 36% higher in Euro-Peristat compared with Eurostat; this range was even wider, À63% to þ51% when a lower limit of 500 g was used. Differences of this magnitude affect benchmarking because they disrupt the order of country rankings. As a research network, Figure 1 Stillbirth rate per 1000 total births in Eurostat cause of death statistics and Euro-Peristat by country in 2015, distinguishing between stillbirths and TOP and sorted by rates of stillbirth using Euro-Peristat data (solid red bar). which collects data using a standardized protocol with data quality checks, Euro-Peristat statistics are more comparable. Therefore, these differences are a cause for concern as Eurostat data represent official European statistics and are the only source of stillbirth data compiled annually for Europe. Eurostat also provides data to other international organizations, such as the OECD, which does not report stillbirth data, but uses these data to compute perinatal mortality statistics (stillbirths and early neonatal deaths) and to WHO which reports stillbirth rates from 1000 g. 22 There are multiple reasons for inconsistency between these two data sources. Data provided to Eurostat come predominantly from demographic statistics collected by national statistical agencies, which may not collect information on birthweight or GA, whereas Euro-Peristat collects data from sources which have these data, such as medical birth registers and perinatal databases. Registration rules and criteria can also differ between data sources. In Ireland, for instance, Eurostat uses data from the Central Statistics Office which only includes registered births and perinatal deaths. In Ireland, it is not a legal requirement for parents/guardians to register stillbirths. Euro-Peristat uses data from the National Perinatal Reporting System, validated using hospital data by the Healthcare Pricing Office, which can provide data using Euro-Peristat definitions. For the UK, we used national birth registrations of live births from England and Wales, Scotland and Northern Ireland combined with stillbirth data for the UK from national perinatal mortality surveillance by MBRRACE-UK as this includes stillbirths starting at 22 weeks, whereas civil registration data only records stillbirths starting at 24 weeks. Some countries combine data from several sources to improve completeness for Euro-Peristat, such as in Italy where data come from the spontaneous abortion register and civil registration data. A similar situation contributes to inconsistency between sources in the Netherlands, where stillbirths are also only registered from 24 weeks in civil registers, but data provided to Euro-Peristat come from linked clinical registers. Likewise, TOPs may not be identifiable in demographic data and these were major contributors to the discrepancies between Euro-Peristat and Eurostat. In the Euro-Peristat project, the country's Scientific Committee member selects the highest quality population-based data with national coverage for describing perinatal indicators and these are usually birth registers or other medical data sources. 17 In contrast, national statistical offices are responsible for reporting to Eurostat, so they do not have a choice and are bound by the limitations of legislation governing national statistics. In addition, stillbirths are usually reported to Eurostat as part of a much larger process of reporting all deaths or a wide range of demographic data.
Discrepancies can also result from population inclusion and exclusion criteria. Demographic statistics most often exclude cases where the mother is not a citizen or permanent resident, while medical birth registers and perinatal database include births in the country without applying any restrictions to citizenship or residents. For instance, in Luxembourg, this was the explanation for large discrepancies between births included in Euro-Peristat (de facto births: any birth occurring in Luxembourg) and those in Eurostat (de jure births: those to residents of Luxembourg only, and wherever the place of occurrence). 23 Third, countries may not consistently follow requested definitions when providing data to Eurostat or the discrepancies could be due to errors. Euro-Peristat collects data by 500 g groups and by GA so that cross-checks are possible, but Eurostat only collects data already grouped together as provided by the statistical offices, so verification is difficult. Furthermore, Eurostat accepts data by GA if countries do not report stillbirth data by birthweight, but this is not explicitly noted in most cases. For instance, in France, stillbirth data by birthweight are not routinely produced because these rely on linkage between maternal and birth hospitalizations in hospital discharge data and incomplete linkage leads to missing data, as seen in table 1. Malta also provides data to Eurostat by GA. A final difference relates to the denominator for rates; Euro-Peristat collects data only on births at 22 weeks and over. In contrast, Eurostat has no GA threshold, so reporting is defined by countries' own thresholds.
The 2011 implementation regulation on causes of death (EU 328/ 2011), following the EU regulation on the Community statistics on public health and health and safety at work (1338/2008) was an important improvement to stillbirth reporting as previously no information on stillbirths by birthweight groups was collected. However, problems remain. First, the sum of the two subgroups (500-999 and !1000 g) is not necessarily equal to the total number of stillbirths because stillbirths with birthweight below 500 g or GA < 22 weeks or crown heel length <25 cm might be recorded in the total number of stillbirths. Second, the implementation regulation is mandatory for the total number of stillbirths, but the more detailed information is voluntarily. Finally, there is no agreement across European countries on whether and how TOPs !22 weeks should be reported. It is preferable to present stillbirth data without terminations because of differences in screening and late termination policies between countries 24 and the high impact that they can have on stillbirth rates. 13,25,26 A more general issue to consider in evaluating current Eurostat rules is the use of birthweight over GA. Euro-Peristat recommends using GA to establish cut-off thresholds for stillbirth reporting since country regulations governing registration of stillbirths principally use GA and not birthweight. 27 Furthermore, as growth restriction is a major cause of stillbirth, using birthweight underestimates thirdtrimester stillbirths. 9 WHO has also recently changed its recommendations to use of GA as opposed to birthweight. 4 Our study also addressed the issue of missing birthweight and GA data. Higher proportions of missing data for birthweight than GA among stillbirths may reflect practices for weighing stillbirths or for recording this information when there is a stillbirth. While birthweight is more straightforward to measure than GA and is usually more complete in most countries, GA data are well recorded in Europe, likely because of high uptake of early antenatal care and dating ultrasounds. Higher proportions of missing birthweight are another reason to prefer GA for reporting of stillbirth indicators. Our results also suggest that internationally agreed upon methods for imputing missing data are needed when reporting thresholds are used, as simply excluding cases with missing data artificially lowers rates.
One way to improve data and to ensure compatibility between data collection in countries with multiple data sources is to encourage linkage, in particular between vital statistics and medical birth registers. Several European countries link these data on a routine basis or for research, showing its technical feasibility, but these practices are far from universal. 28 Birth registers tend to have higher quality data on the clinical conditions affecting stillbirths enabling evaluation of perinatal policies and studies show that linking data between medical and vital statistics registries improves the quality of information. 29 Use of medical data sources can also make it possible to identify terminations. Consolidating multiple sources of data would also avoid problems of inconsistency in international reporting related to use of one source over the other. Finally, a general recommendation is that any perinatal data including information on GA and birthweight should be cross-checked to identify discrepancies caused by coding or reporting error. Improving the quality of national birth data is the best way to ensure high-quality international statistics.
The strength of our study is the careful compilation of validated population-based data from a large number of countries. The main limitation is use of aggregated data which meant that we could not cross-check cases across the sources. We also did not have information on the characteristics of the stillbirths which limited our ability to describe the clinical or social characteristics of discrepant cases. Furthermore, due to small annual births in some countries, the number of stillbirths was small, making it difficult to measure rates with high precision.

Conclusion
The stillbirth rate is a key indicator of population health and offers vital signals about the quality of maternity care and the health of mothers and newborns. International comparisons are a powerful tool to encourage political and societal debate and motivate countries to improve their perinatal health and healthcare systems. However, basing such inferences on comparisons that are not valid or robust can lead to inappropriate conclusions regarding healthcare provision with potentially significant financial and social implications. Our study suggests that the current Eurostat regulations and procedures for stillbirth reporting should be updated to ensure that the burden of stillbirth is accurately captured in European statistics and to enable robust, valid and effective comparisons. This should be done in tandem with more frequent collection of data, including micro-data, through research networks, such as Euro-Peristat, that make it possible to carry out comprehensive analyses of stillbirth and validate routine data. Improvement of national data on stillbirths, including amending legislation to be compatible with WHO and combining information sources to optimize reporting, is needed for the full success of these European initiatives and to guide effective policy to prevent stillbirth.

Supplementary data
Supplementary data are available at EURPUB online.

Funding
The Euro-Peristat project received funding from the European Commission through the Health Programme (grant numbers 20101301 and 664691). The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript.
Conflict of interest: None declared.