Study design and data sources

We conducted a national cohort study using routine data provided by the Office for National Statistics (ONS), based on National Health Service birth notifications linked with civil birth and death registrations. Details are reported elsewhere.16 The full cohort comprised all 4.7 million singleton babies born alive at 22+ completed weeks of gestation in England and Wales from 1 January 2006 to 31 December 2012. We sequentially excluded births with implausible gestational age (43+ weeks), missing birth weight, or implausible birth weight for gestational age (more than twice the IQR above or below the median within sex–gestation–ethnicity strata of the dataset). We analysed the subset born at 24 to 36 completed weeks of gestation, applying the lower limit because of known variation in completeness of registration of births before 24 weeks.17

Variables

Using the ONS hierarchical classification based on The Tenth Revision of the International Statistical Classification of Diseases and Related Health Problems (ICD-10),18 we grouped the causes of infant death as congenital anomalies, immaturity-related conditions, and all other causes combined (’other'). Following WHO guidelines,19 we categorised gestational age at birth as 24–27 completed weeks (extremely preterm), 28–31 (very preterm) and 32–36 (moderate/late preterm). We grouped the baby’s ethnicity (stated by the mother) as White British, White Non-British, Pakistani, Indian, Bangladeshi, Black African, Black Caribbean, other stated and unstated16; the ‘other stated’ group included categories that were mixed (eg, mixed White and Black Caribbean), rare (eg, Chinese) or non-specific (eg, ‘any other Asian background’). We assessed socioeconomic disadvantage using quintile categories of the respective national Index of Multiple Deprivation (IMD) for England 201520 and for Wales 2014,21 based on the mother’s address at birth registration. Gestational age was obtained from routine maternity records and is likely to have been based on the results of ultrasound examinations.22 No data were available for other aspects of clinical care, such as the use of antenatal corticosteroids.

Statistical analysis

We conducted a separate analysis for each combination of gestation category and cause group, using Stata V.13 (StataCorp, 2013). Within gestation groups, we calculated cause-specific infant mortality rates per 1000 live singleton preterm births during 2006–2012, by ethnicity and by IMD quintile. To estimate the association of infant mortality with ethnicity, we fitted a Poisson regression model with robust standard errors, adjusting for potential confounding by birth year and gestation week (within gestation category) as continuous variables and for IMD as a categorical variable. To estimate the association with socioeconomic circumstances, we used IMD instead of ethnicity as the main exposure variable and included ethnicity instead of IMD among the adjustment factors. Causal diagrams for these models are presented in online supplementary figure S1.23 Variation in adjusted risk was measured by the appropriate Wald statistic, using a Bonferroni significance level of 0.0021 to compensate for multiple testing.24

Sensitivity analysis

We used IMD in our main analysis because this area-based measure of disadvantage was available for all members of the cohort. ONS applies the occupation-based National Statistics Socioeconomic Classification25 (NS-SEC) to a 10% random sample of births. We refitted our socioeconomic model in this subgroup of the dataset to assess sensitivity to the socioeconomic measure used, replacing IMD with household NS-SEC (higher managerial, administrative and professional occupations; intermediate occupations; routine and manual occupations; other, including unstated/unspecific occupations, students, long-term unemployed and never worked).

Descriptive analysis

After exclusions for implausible gestational age (0.4%), missing birth weight (0.4%) and implausible birth weight (1.5%), there were 4 634 932 live singleton births,16 of which 256 142 were at 24–36 completed weeks of gestation (table 1). The proportion of births in the ‘unstated’ category of ethnicity was 6%, decreasing from 10% in 2006 to 4% in 2012. There were no other missing data.

The proportion of babies in the most disadvantaged IMD quintile (category 5) varied with ethnicity, being 26% to 28% for the Indian, White Non-British and White British groups but 50% to 58% for the Black African, Black Caribbean, Pakistani and Bangladeshi groups (table 1). The proportion of ethnic minority babies was higher for births at earlier gestational ages; in particular, the combined proportion of Black African and Black Caribbean babies increased from 5% of births at 32–36 weeks to 10% of births at 24–27 weeks.

Overall, 6480 babies died in infancy (table 2), a rate of 25.3 per 1000 live singleton births at 24–36 weeks; 52% of infant deaths were attributed to immaturity-related conditions, 26% to congenital anomalies and 22% to other causes (postpartum infections 5%, intrapartum asphyxia/anoxia/trauma 4%, sudden infant death 3%, antepartum infections 2% and unclassified 5%). Births at 24–27 weeks constituted just 5% of babies in the study population but 47% of those who died in infancy, while births at 32–36 weeks constituted 85% of babies but 32% of those who died in infancy. Hence, crude all-cause infant mortality rates decreased from 261 per 1000 for births at 24–27 weeks to 10 per 1000 for births at 32–36 weeks (table 3).

Ethnicity

In babies born at 24–27 weeks, risk of immaturity-related death was low for the Black Caribbean (adjusted rate ratio 0.63), Black African (0.74), Indian (0.75), Pakistani (0.79) and White Non-British (0.83) groups compared with White British (p<0.0001 for ethnic variation) (table 4).

In babies born at 28–31 weeks, risk of death from congenital anomalies was high for the Pakistani (2.00), Bangladeshi (1.91) and White Non-British (1.53) groups, but low for the Black Caribbean group (0.14) (p=0.0008).

In babies born at 32–36 weeks, risk of death from congenital anomalies was high for the Pakistani (4.50), Bangladeshi (2.89), Black African (2.06), Indian (1.79) and White Non-British (1.29) groups (p<0.0001).

There was no statistically significant ethnic variation in the other cause-gestation strata.

Area deprivation

In babies born at 32–36 weeks, risk of death from congenital anomalies was highest in the most disadvantaged IMD quintile (adjusted rate ratio 1.54 relative to the most advantaged) (p=0.0009 for variation between quintiles) (table 5). Risk of death from ‘other’ causes was highest in the most disadvantaged quintile (2.05) (p<0.0001) with an increasing trend.

There was no statistically significant variation between quintiles for any cause of death in babies born before 32 weeks or for immaturity-related death in any gestation category.

Sensitivity analysis

Although the NS-SEC analysis found no strong evidence of socioeconomic differences, the results were consistent with a socioeconomic trend in risk of death from ‘other’ causes in babies born at 32–36 weeks (online supplementary table S1), as for the IMD analysis.

Statement of principal findings

Among live singletons born at 24–36 completed weeks’ gestation, extremely preterm babies (24–27 weeks, 5% of the births) accounted for nearly half of infant deaths, very preterm babies (28–31 weeks, 10%) a fifth and moderate/late preterm babies (32–36 weeks, 85%) a third. Four-fifths of deaths were attributed to immaturity-related conditions or congenital anomalies.

Patterns of ethnic variation differed by gestational age at birth. Among extremely preterm babies, all minority ethnic groups had lower risk of immaturity-related death than White British, with adjusted rate ratios of 0.63–0.83 for Black Caribbean, Black African, Indian, Pakistani and White Non-British. Among moderate/late preterm babies, all minority groups had higher risk of death from congenital anomalies than White British, with adjusted rate ratios of 4.5 for Pakistani, 2.9 for Bangladeshi and 2.1–1.2 for Black African, Indian and White Non-British. Patterns of ethnic variation for very preterm babies were intermediate.

There was evidence of socioeconomic variation only for moderate/late preterm babies and for causes other than immaturity-related conditions: risks of death from congenital anomalies and combined rarer causes (infections, intrapartum conditions, sudden infant death and unclassified) increased with area deprivation, adjusted rate ratios being 1.54 and 2.1, respectively, comparing the most with the least disadvantaged IMD quintile.

Strengths and limitations

To our knowledge, this is the first study to describe ethnic and socioeconomic variation in gestation-specific preterm infant mortality by cause of death. Some of our findings may be generalisable to similar ethnic and social groups resident in other countries. The study was based on linked routine data from England and Wales for a 7-year national birth cohort, with virtually complete follow-up through the death registration system. Gestational age was derived from maternity records, and the proportion of exclusions for missing or implausible information was very small (2%).

Where infant deaths were attributed to congenital anomalies, no information on the specific type of congenital anomaly was available. The study period (2006–2012) is now several years old, but the observed patterns of ethnic and socioeconomic variation seem likely to be reasonably stable over time. The proportion of births with unstated ethnicity decreased over time, from 10% in 2010 to 4% in 2016, but all regression analyses were adjusted for potential confounding by birth year. Although birth notifications nominally recorded the baby’s ethnicity as stated by the mother, in practice they may have included reports by healthcare professionals of the baby’s or mother’s ethnicity26; any resulting misclassification might have caused underestimation of differences between ethnic groups, but the potential bias is probably small. Area deprivation scores do not necessarily reflect individual socioeconomic circumstances; the sensitivity analysis using individual-level NS-SEC was underpowered, being based on just a 10% sample of births, but gave compatible results. Using separate IMD quintiles for England and for Wales is unlikely to have materially affected the results, as deprivation-specific mortality rates are similar,27 and only 5% of the cohort were born to mothers resident in Wales.

Relation to other findings

Our results appear broadly consistent with international evidence on ethnic3–11 or socioeconomic12–15 variation in gestation-specific all-cause mortality of preterm babies. Comparisons require caution, however, as designs, settings, time periods, outcome definitions, gestation categories and ethnic groupings varied; births before 24 weeks and multiples were included; and maternal (not infant) ethnicity was generally examined. Some studies did not provide CIs or hypothesis tests, and none reported cause-specific risks.

In the UK in the 1990s, a relatively small hospital-based study found that neonatal mortality rates were similar for Black, South Asian and White mothers before 28 weeks, but lower for Black mothers than White among later preterm babies3; a larger hospital-based study found that perinatal mortality rates were lower for Black mothers than White before 32 weeks, but higher for Black than White thereafter and highest for babies of South Asian mothers at all preterm gestations.4 In the Netherlands during 1999–2007, national neonatal mortality rates for babies born at 26–31 weeks were higher for White mothers than African, South Asian, Mediterranean or East Asian.5 In the USA, national infant mortality rates until the early 2000s were lower for Black mothers than White among extremely/very preterm births and higher among later preterm births,6 7 but more recently were higher for Black mothers at all preterm gestations8; other findings were similar.9–11

For England and Wales in 2005, national infant mortality rates within each preterm category were higher in households with routine/manual occupations than managerial/professional.12 Studies from France, Sweden and the Trent region of England found no evidence of socioeconomic variation in predischarge or infant mortality among births before 32 or 33 weeks,13–15 but a socioeconomic gradient in infant mortality for later preterm births.15

Meaning of the study

A population-based study in the Trent region of England found a socioeconomic gradient in incidence of birth at 22–32 weeks’ gestation but not in subsequent survival, suggesting that deprivation was not a barrier to accessing and receiving effective neonatal care.13 Our results for immaturity-related deaths confirm this finding in the national population and extend it to moderate/late preterm babies. Despite similar antenatal detection rates, women living in more deprived areas of the Trent region had lower rates of termination of pregnancy for major congenital anomalies than women in more affluent areas; this differential contributed to higher rates of birth with such anomalies and hence to higher rates of neonatal mortality.28 Similarly, for moderate/late preterm babies in the national population, we found that infant mortality from congenital anomalies was highest in the most disadvantaged group. We also found a socioeconomic trend in death from ‘other’ causes (including infections, intrapartum conditions and sudden infant death), but this is difficult to interpret, as the component causes will have different mechanisms and were too rare to analyse separately in our dataset.

Among moderate/late preterm babies, we found relatively high rates of death from congenital anomalies in the Pakistani and Bangladeshi groups (nearly five and three times the corresponding White British rate, respectively). Pakistani babies are known to have a relatively high incidence of congenital anomalies, particularly autosomal recessive conditions.29 Diabetes may play a role,30 but it is generally accepted that the most important single cause is the high prevalence of endogamy and consanguineous marriage in some Pakistani communities29: for example, in a recent study in Bradford UK, 59% of 5127 Pakistani mothers were married to a blood relative.31 Accessibility or use of fetal anomaly screening may also contribute: limited evidence suggests that rates of prenatal diagnosis might be relatively low for South Asian women in the UK32 and attitudes to termination of pregnancy may vary.

We found that the risk of immaturity-related death among extremely preterm babies was relatively low for Black and South Asian groups compared with White, despite their generally higher risks of preterm birth.16 We speculate that this apparently paradoxical observation might reflect variation between ethnic groups in the proportions of the many different biological causes of extremely preterm birth operating within those groups. For example, maternal sickle cell disease (SCD)33 34 incurs a risk of small fetal size and preterm delivery35 36 and will therefore contribute to the incidence of extremely preterm birth in the Black ethnic groups. The baby of a mother with SCD usually does not have SCD and may have no special problems after birth, other than immaturity. The spectrum of causes of extremely preterm delivery in babies of mothers without SCD will be slightly different, and may include a higher proportion of fetal conditions with serious postpartum implications. Maternal thalassaemia might provide a similar mechanism for South Asian babies.33 34 37

Ethnic variation in immaturity-related infant death is relevant to ‘customisation’ of the standards that support clinical decisions to induce labour for inappropriate fetal growth.38 Previous evidence that extremely preterm babies had better survival rates if their mothers were Black and that preterm births were more frequent in Black and South Asian families generated a hypothesis that the optimal period of gestation might be genetically shorter for Black and South Asian than White babies.39 The findings of our study are consistent with that hypothesis, but other explanations are possible (as discussed above, for example). A recent large international study on pregnancies of women with ‘optimum health, nutrition, education and socioeconomic status’ concluded that optimal fetal growth is similar in diverse geographical settings.40 This seems to imply that ethnic differences might be mediated either by disadvantage or by maternal ill-health. Our results suggest the latter should be considered.