International variations in the gestational age distribution of births: an ecological study in 34 high-income countries

Abstract

Background

Few studies have investigated international variations in the gestational age (GA) distribution of births. While preterm births (22–36 weeks GA) and early term births (37–38 weeks) are at greater risk of adverse health outcomes compared to full term births (39–40 weeks), it is not known if countries with high preterm birth rates also have high early term birth rates. We examined rate associations between preterm and early term births and mean term GA by mode of delivery onset.

Methods

We used routine aggregate data on the GA distribution of singleton live births from up to 34 high-income countries/regions in 1996, 2000, 2004, 2008 and 2010 to study preterm and early term births overall and by spontaneous or indicated onset. Pearson correlation coefficients were adjusted for clustering in time trend analyses.

Results

Preterm and early term births ranged from 4.1% to 8.2% (median 5.5%) and 15.6% to 30.8% (median 22.2%) of live births in 2010, respectively. Countries with higher preterm birth rates in 2004–2010 had higher early term birth rates (r > 0.50, P < 0.01) and changes over time were strongly correlated overall (adjusted-r = 0.55, P < 0.01) and by mode of onset.

Conclusion

Positive associations between preterm and early term birth rates suggest that common risk factors could underpin shifts in the GA distribution. Targeting modifiable population risk factors for delivery before 39 weeks GA may provide a useful preterm birth prevention paradigm.

Introduction

The typical length of pregnancy is 39–40 weeks, but annually 15 million infants are born preterm, defined as birth before 37 completed weeks of gestation.¹ Preterm birth is associated with adverse child health and long-term neurodevelopmental outcomes.¹ While medical advances have reduced mortality and morbidity risks among preterm infants over past decades, little progress has been made in preventing the incidence of preterm birth.² There are few effective interventions for preventing preterm delivery and, with the exception of programs promoting smoking cessation, they tend to target selected groups of high risk pregnancies, such as women with previous preterm deliveries in the case of progesterone or cervical cerclage.^2,3 However, differences in rates across high-income countries from 5% to 10%, as well as heterogeneous time trends,^4–6 suggest that there are modifiable population factors that affect preterm birth risk.⁷

Identifying population-wide exposures and designing policies to mitigate them could be facilitated by a broader focus on early delivery. Whereas preterm birth is associated with the greatest risks, recent research on early term births (37 and 38 completed weeks) also highlighted their elevated risks of adverse health outcomes compared to those born full term, at 39 or 40 weeks.⁸ Given their larger numbers, early term births may provide studies with greater power to detect risk factors which influence changes in preterm birth rates but also impact on the population gestational age (GA) distribution.

Few studies have investigated international variations in the GA distribution,⁹ and it is not known whether countries with higher preterm birth rates also have higher rates of early term births. If common risk factors affect earlier delivery across the GA spectrum, we would expect associations between preterm and early term birth rates across countries and across time. Furthermore, policies that successfully shift the GA distribution towards later delivery could reduce preterm births as well as early term births. We thus aimed to investigate variations in early term births and their association with preterm birth rates, and mean term GA in high-income countries.

Methods

Data sources

We used data from the Euro-Peristat project, which monitors a set of 30 perinatal health indicators in European countries using data available in national health information systems.^10,11 Data were also obtained for the United States (US), Canada and Japan as part of the PREBIC Epidemiology Working Group (www.prebic.org). Data from Australia were ascertained from the New South Wales Perinatal Data Collection.

The Euro-Peristat project collected aggregate data on all births starting at 22 weeks of GA in 2004 and 2010.⁴ One of Euro-Peristat’s Core indicators is the distribution of GA in completed weeks by vital status (stillbirths and live births) and multiplicity. The project also conducted a separate study on preterm birth for all live births by multiplicity and mode of onset and delivery for the years 1996, 2000, 2004 and 2008 in 19 countries.⁵ The data collection sheets for the Euro-Peristat core indicators and the study on preterm birth were used to acquire data from the non-European countries in the PREBIC project (US, Canada, Japan) and Australia.

Data came from vital statistics, civil registers and medical birth registers in most countries and from a nationally representative survey of births in France⁴ (Supplementary Appendix A). If countries could not provide national data, population-based data from geographically defined regions were accepted. Data for Belgium came from Brussels, Wallonia and Flanders; data from the United Kingdom were provided by England and Wales combined, Northern Ireland and Scotland; data from Germany came from the regions of Hesse, Lower Saxony and Bavaria in 2000, 2004 and 2008, but national-level data were available in 2004 and 2010. Data from Canada do not include births in the Province of Québec, and data from Australia were limited to the region of New South Wales, which represents one-third of annual births in Australia. In the US, births from California were excluded due to non-reporting of clinical estimates of GA before 2007. In France, data were from 1995, 1998, 2003 and 2010; in Canada and UK: England and Wales, from 2005 instead of 2004; and in Sweden from 2009 instead of 2008.

Study population

Our study population was all singleton live births with a GA of 22 weeks or over. We focused on singletons, because preterm birth rates are much higher for multiple pregnancies and multiple pregnancy rates differ widely among countries.¹² Stillbirths were excluded, as stillbirth data were not available in 1996, 2000, or 2008. GA data were available from 16 countries/regions in 1996, 20 in 2000, 29 in 2004, 22 in 2008 and 34 in 2010. Fourteen countries had data for all five years. Data on mode of onset of labor (i.e. spontaneous or provider-initiated delivery) were available for 12 countries in 1996, 14 in 2000, 15 in 2004 and 16 in 2008, but not in 2010 (participating years of data are available in Supplementary Appendix A).

Definitions

GA was requested based on the best available obstetrical estimate. In Europe, Canada, Australia and Japan, GA is derived from ultrasound and other prenatal assessments of gestational length. In the United States, birth certificates historically relied on a clinical estimate of GA, which included postnatal assessments. In 2003, a revised version of the birth certificate using only antenatal assessments (the obstetric estimate) was devised and by 2010, 35 states had adopted the 2003 revision;¹³ for the analyses, we used either the clinical or the obstetrical estimate.

We assessed countries’ GA distributions based on the following outcomes: rates of preterm (22–36 weeks) and early term births (37–38 weeks) and mean GA at term, excluding preterm births. We recoded births 42 weeks GA and over to 41 weeks, because countries differed in their policies for management of post-term pregnancies, and our focus was whether the pregnancy progressed to term.¹⁴ Subgroup analyses were done for very preterm births (24–31 weeks) and moderate and late preterm births combined (32–36 weeks). Births at 22–23 weeks GA were not included in very preterm birth analyses due to the impact of differing registration practices on very preterm birth rates.¹⁵ We also computed rates by mode of delivery. We identified indicated deliveries where the mode of onset was provider-initiated: i.e. induction of labour, or prelabour or elective caesarean delivery, based on national classifications.

Missing data

Overall, less than 2% of GA data were missing, except in Germany (missing 3% in 2000), Norway (10% in 1996) and Spain (11–20% from 1996 to 2010).¹⁶ Similarly, less than 2% of mode of delivery onset data were missing. Observations with missing data were excluded from the analyses.

Analysis strategy

We summarized countries’ GA distributions using descriptive statistics of preterm and early term birth rates including median, interquartile range (IQR), and mean GA at term calculated for each country/region, year of birth, and mode of delivery onset.

We used Pearson correlation coefficients to assess the magnitude of the associations between rates of preterm and early term birth in each study period and in time trends over the study periods.To investigate time trends, we calculated compound annual growth rates of preterm and early term births between data points; where compound rates allowed us to take into account differences in time periods for which GA data were available across countries.

The Pearson correlation coefficient was used because it can be adjusted for clustering in time series analyses based on Lorentz’ formula.¹⁷ The adjusted Pearson correlation coefficient is a marginal association measure derived from generalized estimating equations which remain valid under the clustered framework, and take account of informative cluster size. We examined the associations overall and by mode of onset of delivery. We also carried our sensitivity analyses using Spearman non-parametric tests, which do not rely on assumptions of normality. All data used for GA subgroup rates in 1996, 2000, 2004, 2008, 2010, and growth rates between periods are provided in Supplementary Appendix B.

Data were analysed using STATA 13.0 software (StataCorp LP, College Station, TX, USA), while adjusted analyses were conducted using R Statistical Software (Foundation for Statistical Computing, Vienna, Austria).

Results

Table 1 shows live singleton preterm (22–36 weeks) and early term (37–38 weeks) birth rates in 2010 in 34 countries/regions. In 2010 across 34 high-income countries and regions, preterm birth rates varied between 4.1% and 8.2%, with a median of 5.5%. The median rate of early term birth was 22.2% with a range between 15.6% and 30.8%. Mean GA at term ranged between 39.0 and 39.7 weeks.

Table 1.

Live singleton preterm (22–36 weeks) and early term (37–38 weeks) birth rates in 2010

Country: region	Code	N	GA in completed weeks
			22–36%	37–38%	Mean GA at term
Austria	au	75 950	6.3	25.5	39.3
Australia: New South Wales	nsw	92 974	5.5	23.4	39.3
Belgium: Brussels	be_bu	23 731	6.2	23.6	39.3
Belgium: Flanders	be_fl	67 029	6.0	24.3	39.2
Belgium: Wallonia	be_wa	36 965	6.5	29.1	39.1
Canada (without Québec)	ca	270 401	6.3	25.3	39.2
Czech Republic	cz	111 616	6.1	21.9	39.3
Denmark	dk	60 667	4.9	18.1	39.6
Estonia	es	15 357	4.6	17.1	39.6
Finland	fi	59 318	4.3	16.1	39.6
France	fr	14 326	5.5	22.5	39.4
Germany	ge_ntl	611 864	6.5	27.3	39.3
Iceland	ice	4 751	4.1	15.7	39.6
Ireland	ir	72 707	4.2	15.6	39.7
Italy	it	527 845	5.7	28.3	39.2
Japan	ja	1 080 089	4.7	30.8	39.0
Latvia	lv	18 662	4.9	17.5	39.4
Lithuania	li	30 035	4.3	15.7	39.5
Luxembourg	lu	6 285	6.3	29.7	39.1
Malta	mt	3 856	5.4	30.7	39.0
Netherlands	ne	171 781	5.9	21.8	39.4
Norway	no	60 623	4.9	16.4	39.6
Poland	po	402 171	5.3	19.9	39.5
Portugal	pt	98 386	5.9	26.5	39.0
Romania	ro	208 325	7.6	23.6	39.1
Slovakia	sa	54 041	5.7	19.9	39.4
Slovenia	se	21 482	5.5	19.1	39.4
Spain	sp	382 136	5.9	22.4	39.4
Sweden	sw	111 474	4.7	18.3	39.6
Switzerland	ch	77 016	5.2	26.5	39.2
UK: England and Wales	uk_ew	696 087	5.6	18.1	39.6
UK: Northern Ireland	uk_ni	24 804	5.6	16.6	39.6
UK: Scotland	uk_scot	55 367	5.5	16.3	39.6
USA (without California)	usa	3 363 032	8.2	27.2	39.0

Variation was seen in both spontaneous and provider-initiated births. In 2008, between 2.8% and 5.1% of live births were spontaneous preterm births, whereas from 1.1% to 4.4% were provider-initiated (cf. Supplementary table S1). Even greater variation was seen in early-term births: between 9.8% and 16.6% for spontaneous births, and between 4.3% and 15.5% for provider-initiated births.

Figure 1 displays the associations between preterm birth rates and early term birth rates in 1996, 2000, 2004, 2008, 2010. The strength of the correlation ranged between r = 0.55 and r = 0.58 in 2004–2010 (P < 0.01) whereas in 1996 and 2000 the magnitude of the association was lower and not statistically significant; results using Spearman’s rank test were similar. Japan was an outlier in all years, with low preterm birth rates and high early term birth rates. Without Japan, correlations were stronger (r=.57 to .75, P ≤ 0.01 in 2000–2010).We also studied countries with data available in all five study years (Supplementary table S2) and found similar results. Finally, we looked at associations for spontaneous and indicated deliveries separately and observed similar trends, although results were significant only for provider-initiated deliveries (cf. Supplementary figures S1A and S1B).

Figure 1.

Associations between preterm birth rates (<37 weeks) and early term birth rates (37–38 weeks) in 1996, 2000, 2004, 2008, 2010. Note: For country codes see table 1. ‘ge’ refers to German data from the regions of Hesse, Lower Saxony and Bavaria in 2000, 2004 and 2008; ‘ge_ntl’ refers to national data available in 2004 and 2010

Associations between preterm birth rates and mean GA at term are shown in figure 2. Preterm birth rates were negatively correlated with mean GA at term in all years, with significant correlation coefficients of -0.51 in 2004, -0.58 in 2008 and -0.68 in 2010; in 1996 and 2000 however, the correlations were lower and were non-significant. Results using Spearman’s rank test were similar.

Figure 2.

Associations between preterm birth rates (<37 weeks) and mean GA at term (37–41 weeks GA) in 1996–2010. Note: For country codes see table 1. ‘ge’ refers to German data from the regions of Hesse, Lower Saxony and Bavaria in 2000, 2004 and 2008; ge_ntl refers to national data available in 2004 and 2010

Figure 3 illustrates the correlations between compound annual preterm and early term growth rates across the periods in our study, representing 83 time points from 29 countries/regions with data in at least two years. Temporal changes in preterm and early term birth rates were strongly correlated (adjusted Pearson’ r = 0.55, P < 0.01). Although annual changes were more heterogeneous for provider-initiated births than for spontaneous births, correlations by delivery mode of onset for spontaneous and indicated births were similar (cf. Supplementary figure S2A and S2B). Changes in indicated preterm deliveries were not significantly correlated with changes in spontaneous early term deliveries (adj-Pearson’s r = 0.11, P > 0.05, N = 42), nor were changes in spontaneous preterm deliveries significantly correlated with changes in indicated early term deliveries (adj-Pearsons’ r= -0.32, P > 0.05, N = 42).

Figure 3.

Associations between annual growth rates for preterm and early term births between 1996 and 2010. Note: Live births: adjusted Pearson’s r = 0.55, P < 0.01; N = 83

In preterm subgroup analyses, very preterm birth rates (24–31 weeks) were not correlated with early term births or mean GA at term in 1996–2010. Moderate and late preterm births (32–36 weeks) were positively correlated with early term births however (adj-Pearson’s r = 0.56, P < 0.01, and negatively correlated with mean GA at term (r ranging from -0.6 to -0.7, P < 0.01 in 2004–2010 (cf. Supplementary table S3).

Discussion

Main findings

We found that early term birth rates varied by a factor of 2, comparable to the relative variation in preterm birth rates although higher in absolute terms: up to 15%. Countries with high early term rates and lower mean GA were more likely to have high preterm rates. These associations increased over time, especially from 2004 onwards. Time series results were similar for spontaneous and provider-initiated births, but cross-sectional results were significant only for provider-initiated births. Finally, these associations were observed for moderate and late preterm births, but not for the sub-group of very preterm infants born before 32 weeks of GA.

Strengths and limitations

Our study’s strengths include the use of population-based data on births at each completed week of gestation compiled using a common protocol for a large number of high-income countries from North America, Europe and Asia-Oceania. Data available from several years also allowed us to study time trends, and we adjusted for informative clustering of rates within countries in our time series analyses. Compound annual growth rates took into account relative changes in risks over differing time periods, and smoothed year-to-year volatility in preterm and early term birth rates.

One limitation was that although we requested data using the best obstetric estimate of GA, we had no further information on how that estimate was derived. Ultrasound dating was the norm, but other methods of GA assignment were likely used and could impact estimates of both the preterm and early term birth rates.¹⁸ Also, we had data on mode of delivery onset from fewer countries, as such data are not always collected in routine data systems. In Japan, for example, the absence of data on delivery onset limited further exploration of the high relative rates of early term vs. preterm births. Moreover, differences in definitions may affect the comparability of rates of spontaneous and indicated deliveries across countries.^4,19 Finally, since our data were aggregated, we were unable to stratify by other factors that may affect the preterm birth rate, including maternal age, parity, smoking and socioeconomic status.

Interpretation

Our data showing an association between early term and preterm births suggest that variations in preterm birth rates reflect, a more general shift in the GA distribution. Overall, we observed robust correlations between rates of preterm and both early term births and mean GA at term in time series analyses, which are less sensitive than cross-sectional analyses to varying definitions among countries. Cross-sectional correlations were not significant in 1996 and 2000, and for spontaneous-onset births in all years. The absence of associations in those years could be due to fewer countries with data, or to more recent changes. The absence of an association between very preterm and early term births may reflect differences in both etiology and practices for this sub-group.

GA subgroup rate associations across countries suggest that variation in risk factors for preterm birth may influence early delivery risk across the GA continuum. Maternal characteristics such as maternal age, smoking during pregnancy and BMI, are known to differ among high-income countries²⁰ and have been found to affect preterm birth trends within countries,^21–23 as well as socio-demographics factors such as maternal educational level and migrant status.²⁴ Environmental factors may also partially explain our findings. Policies to reduce exposure to secondary smoke have been found to correlate with reductions in the preterm birth rate in Belgium,²⁵ and in preterm and early term deliveries in Switzerland.²⁶ The impact of chemical exposures and air pollution on duration of gestation is of increasing interest although more research is needed on the underlying physiological mechanisms.^27,28 In a recent population-based study from Canada, associations of ambient air pollution with preterm birth were stronger among women with pre-existing diabetes, asthma and preeclampsia, suggesting that environmental factors interact with other population characteristics.²⁹

Clinical practices related to indicated deliveries also likely contribute to our findings. Studies have shown wide variations in rates of obstetric intervention for subgroups at higher risk of intervention,¹⁹ and throughout the GA continuum.³⁰ In New South Wales, a decreasing GA from 1994 to 2009 was associated with decreases in spontaneous birth and increases in early term birth and provider-initiated deliveries.³¹ In the US, changes in the use of obstetric interventions have been studied as drivers of variation in the preterm birth rate,³² and recommendations to decrease provider-initiated deliveries before 39 weeks have been linked to decreases in late preterm and early term birth.^9,33

Guidelines related to screening, antenatal care and the management of pregnancy complications are different across countries and evolve over time; these may contribute to rate differences, in particular for provider-initiated births. For example, gestational diabetes will increase the risk of indicated preterm or early term delivery, but not all countries offer routine screening. Policies related to the timing delivery aim to maintain low rates of perinatal and maternal morbidity and mortality overall, but these also change over time which could contribute to variation in GA subgroup rates.

Finally, methods of GA estimation and the more frequent use of ultrasound for pregnancy dating could impact on GA subgroup trends. Some studies find that US dating increases preterm birth rates (because LMP estimates assume all women have a 28 day cycle, whereas the average is slightly longer),³⁴ while others have documented decreases in preterm birth (due to the elimination of erroneous GA).³⁵ The determination of GA is an important area for further research into cross-national variation in preterm and early term rates.⁶

Proposals for research and practice

GA at delivery is a strong determinant of perinatal and child health. Our findings show that variations in preterm and early term birth rates and trends tend to co-occur in most high-income countries, suggesting a common aetiology for early delivery.^34,35 These results warrant the evaluation of risk factors affecting both preterm and early term birth as opposed to targeting the highest-risk group of preterm births only. Based on the premises of Rose’s population approach to the prevention of disease: changes in mean level of exposures and clinical practices may explain the observed heterogeneity in preterm and early term birth rates over time within, and among countries.³⁶ This carries implications for research and programme evaluation, in particular for the choice of outcome variables.

A population-based approach to early delivery prevention is related to mitigating demographic, behavioral and environmental risks in the general population, as well as evaluating the impact of clinical practices. Moreover, by focusing on shifting determinants of earlier birth among the low-risk majority it may be possible to achieve a similar impact on higher-risk groups as well, in line with a stewardship model of public health that is both ethical and efficient.³⁷ A ‘population vision’ of preterm birth prevention could also potentially link more global initiatives to reduce unnecessary obstetric interventions³⁸ to those intended to reduce preterm birth.

In conclusion, we observed wide variation in early term birth rates across high-income countries which were associated with preterm birth rates cross-sectionally and over time, with the exception of very preterm births. Our results suggest that a more general focus on identifying, designing and implementing interventions to target modifiable population-level risk factors for preterm as well as early term deliveries may provide a useful prevention paradigm.