Study population

We derived nationally representative cohorts of singleton live births in 2003–2012 to resident mothers using linked administrative databases from England and Sweden. The English birth cohort was based on birth admissions identified in Hospital Episode Statistics (HES).13 For each birth, two new records are created in HES—a birth admission for a baby and a delivery admission for the mother, both covering information about key risk factors at birth. The two records were probabilistically linked to enhance completeness of birth characteristics and socioeconomic factors in the birth cohort.14 Birth records were then linked to the baby’s subsequent hospital admissions and Office for National Statistics death registration records using a study-specific identifier generated by the data provider.13 We excluded 29% of births in hospitals with low completeness of recorded variables or poor linkage to mortality records. Detailed derivation and validation of the cohort is described elsewhere.2 The Swedish birth cohort was based on births identified in the Medical Birth Register.15 Births were linked to hospital records16 and death records17 using the child’s pseudonymised personal identity number generated by the data provider.

Children were followed up from the 31st day of life until their fifth birthday, death or 31 December 2013, whichever occurred first. We excluded deaths at ≤27 days due to differences in death certification practices in England and Sweden.18 19 Deaths at 28–30 days were excluded as 75% of these deaths in the English birth cohort were missing all causes of death, which likely reflects a data extraction error by the data provider.

Outcomes

RTI-related deaths were defined as non-injury, non-SUDI with an appropriate International Statistical Classification of Diseases and Related Health Problems 10th Revision (ICD-10) code recoded as any cause of death or any diagnosis in a hospital admission that started <31 days before death (listed in online supplementary appendix table 1).9 We excluded injury-related deaths as they are considered accidental and not a direct result of RTI. Deaths were classified as injury-related if the underlying cause of death included any code from chapter 20 of ICD-10.20

SUDI was indicated if the underlying cause of death included any ICD-10 code from a code list recommended for international comparisons to minimise bias from differences in coding and investigative practices (listed in online supplementary appendix table 2).12 The code list included all unexpected deaths: SIDS, sudden deaths from unspecified causes and sudden deaths that were subsequently explained (eg, accidental suffocation).

Risk factors

We categorised birth weight as 500–1499 g, 1500–2499 g, 2500–3499 g and ≥3500 g and gestational age as 24–34, 35–36, 37–38 and ≥39 weeks. We developed an indicator of congenital anomalies using a subgroup of codes from chapter 17 of ICD-1020 included in a code list of chronic conditions in children.9 Congenital anomalies were identified if any diagnosis in hospital admissions within the first 2 years of life or any cause of death under the age of 5 included a relevant ICD-10 code.

Measures of socioeconomic status (SES) were not directly comparable in the two countries. In England, quintiles of SES were calculated using the Index of Multiple Deprivation score, an area-level indicator measured per 200–1400 households, allocated according to a child’s postcode.13 In Sweden, we used quintiles of family’s disposable household income in the year before childbirth.21 We also included maternal age as an individual-level variable indicating SES, categorised as <20, 20–24, 25–29, 30–34 and ≥35 years.

We excluded births with missing information on either of the risk factors and births weighing <500 g or at <24 weeks’ gestation.

Statistical analyses

For each country we derived numbers and percentages of all children, total deaths, RTI-related deaths and SUDIs by each risk factor category. We also calculated cause-specific mortality rates per 100 000 child-years. We fitted Cox proportional hazards (PH) regression models to estimate HRs for cause-specific mortality in England relative to Sweden (the baseline), which we used as a measure of relative risk of death in England versus Sweden. First, we fitted unadjusted models including only a covariate for country of birth. Then, we added birth characteristics and socioeconomic factors to determine the contribution of these risk factors to excess mortality in England.

We used Schoenfeld residual plots for each covariate to test the underlying assumption of the Cox PH models that HRs for all covariates remain constant over time.22 We fitted additional Cox PH regression models with an effect modification term between each covariate for which PH assumption was violated and survival time.

Since 90% of RTI-related child deaths in England occur among children with at least one chronic condition, we repeated the analyses for RTI-related mortality at 1–4 years also including an indicator of chronic conditions (identified if any diagnosis recorded during infancy included ICD-10 codes from the code list of chronic conditions, excluding codes for congenital anomalies).9 We did not repeat the analyses for RTI-related mortality at 31–364 days as any chronic illness at age 0–30 days was likely to be associated with birth characteristics already included in the model. We used Stata MP V.14.2 for all analyses.

Study population

The cohort comprised 3 928 483 births in England and 1 012 682 births in Sweden, with follow-up of 15 167 185 and 4 132 324 child-years, respectively (figure 1). The prevalence of adverse birth characteristics such as low birth weight (<2500 g), preterm birth (at <37 weeks’ gestation) and congenital anomalies was higher in England than in Sweden (table 1). Young motherhood (aged <20 years) was also more common in England.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Flow diagram showing steps taken to develop comparable, nationally representative cohorts of children who were born in 2003–2012 in England and Sweden. The numbers of live births (n) and deaths (d) are presented. For each exclusion criterion, the percentage of all live births and all deaths is shown in brackets. Crude mortality rates at age 31 days–4 years per 100 000 child-years are presented for each country before and after applying all exclusion criteria (95% CI).

RTI-related deaths contributed 16% (768) of all deaths at 31–364 days in England and 17% (131) in Sweden. In both countries, a third of children who died in the first year of life with RTI had low birth weight, 27% were born prematurely, half had a congenital anomaly (table 1), and over 70% had at least one of these risk factors (559 in England and 92 in Sweden). At 1–4 years, RTI-related deaths contributed 31% (691) of all deaths in England and 25% (118) in Sweden, of which approximately 20% were in children with low birth weight or preterm birth, 50% and 36%, respectively, were in children with congenital anomalies, and 69% and 57%, respectively, were in children with any of these risk factors (475 in England and 67 in Sweden). In both countries SUDI accounted for a further 24% of all deaths at 31–364 days (1166 in England and 189 in Sweden). One in five of these deaths occurred in children with low birth weight or preterm birth, 4% in England and 13% in Sweden in children with congenital anomalies, and half in children with any of these characteristics (565 in England and 93 in Sweden). Unadjusted mortality rates were 50%–60% higher in England than in Sweden (online supplementary appendix table 3).

Comparison of RTI-related and SUDI mortality

At 31–364 days, children born in England had 52% higher risk of RTI-related death than children born in Sweden (HR=1.52, 95% CI 1.26 to 1.82; table 2). After adjusting for birth characteristics, the relative risk of RTI-related death in England versus Sweden reduced by approximately 70%, to 16% higher relative risk (HR=1.16, 95% CI 0.96 to 1.40). Further adjustment for socioeconomic factors reduced the relative risk to 11% (HR=1.11, 95% CI 0.92 to 1.34). The risk of RTI-related death was highest for children with congenital anomalies (HR=25.55, 95% CI 22.20 to 29.41, compared with children with no anomalies) and low birth weight (HR=5.30, 95% CI 3.54 to 7.94 for 500–1499 g and HR=5.38, 95% CI 4.13 to 7.02 for 1500–2499 g vs birth weight ≥3500 g).

At 1–4 years, the risk of RTI-related death was 58% higher in England relative to Sweden (HR=1.58, 95% CI 1.30 to 1.92; table 3). After adjusting for birth characteristics, the relative risk of RTI-related death in England versus Sweden nearly halved to 32% higher risk (HR=1.32, 95% CI 1.09 to 1.61) and did not change appreciably after further adjustment for socioeconomic factors. Having a congenital anomaly increased the risk of RTI-related death 28 times relative to having no anomalies (HR=28.39, 95% CI 24.55 to  32.82).

Children born in England had 59% higher risk of SUDI than children born in Sweden (HR=1.59, 95% CI 1.36 to 1.85; table 4). Adjustment for birth characteristics and socioeconomic factors each reduced the relative risk of SUDI in England versus Sweden by a third, to 40% (HR=1.40, 95% CI 1.20 to 1.63) and 19% (HR=1.19, 95% CI 1.02 to 1.39), respectively. The risk of SUDI was highest for children with low birth weight (HR=6.83, 95% CI 4.62 to 10.09 for 500–1499 g and HR=3.33, 95% CI 2.66 to 4.18 for 1500–2499 g vs birth weight ≥3500 g) and for children of young mothers (HR=4.45, 95% CI 3.86 to 5.37 for mothers aged <20 years vs 30–34 years).

Sensitivity analyses

The PH assumption was not met only for the indicator of congenital anomaly for RTI-related deaths at 31–364 days. We repeated the analyses allowing for different HRs for congenital anomaly at 1–2 months, 2–3 months and 3–12 months. The HRs did not change appreciably in sensitivity analyses (online supplementary appendix table 4). Cox PH models for RTI-related mortality at 1–4 years and SUDI met the PH assumption.

The relative risk of an RTI-related death at 1–4 years declined from 58% to 32% after adjustment for birth characteristics in the main analyses, and further to 3% after including an indicator of chronic conditions in sensitivity analyses (online supplementary appendix table 5). The relative risk did not change appreciably after further adjustment for socioeconomic factors. Children with chronic conditions had 20 times higher risk of RTI-related death compared with children with no chronic conditions; children with congenital anomalies had nearly 9 times higher risk (compared with 28 times higher risk in the main analyses without adjusting for chronic conditions).

Use of ICD-10 codes included in the definition of SUDI varied between the two countries (online supplementary appendix table 6). ICD-10 code ‘R95’ denoting SIDS accounted for 63% of SUDIs in England and 75% in Sweden; ‘R99’ for unknown cause accounted for a further 33% and 22%, respectively.

Interpretation

Previous comparisons of cause-specific child mortality hypothesised that increased infection-related mortality in England relative to Sweden reflected differences in provision and organisation of healthcare in the two countries.23 24 Our results suggest that risk factors operating before and during pregnancy are more important: excess RTI-related mortality in England relative to Sweden was largely explained by higher prevalence of adverse birth characteristics in England, with congenital anomalies and low birth weight being associated with the highest relative risks of RTI-related deaths. Unfavourable distribution of birth characteristics also contributed to excess SUDI mortality in England. Adverse birth characteristics are strongly associated with maternal risk factors such as smoking,28 29 obesity,30 31 and young and old maternal age,32–34 many of which are more prevalent in England than in Sweden. These maternal risk factors are, in turn, strongly associated with socioeconomic disadvantage.35 36 These findings suggest that reducing adverse maternal health and socioeconomic circumstances before and during pregnancy to improve health outcomes at birth could reduce preventable child mortality in England relative to Sweden. Further work is needed to identify specific interventions which would be most effective at improving birth characteristics of children in England.

Socioeconomic factors contributed a third of the excess risk of SUDI and 10% of the excess risk of RTI-related infant death in England relative to Sweden, over and above their effect on birth characteristics. This likely reflects wider income inequality in England relative to Sweden (in 2009, 5.5% of children in the UK lived in deprived households, compared with 1.3% in Sweden).37 Linkage to external data sources is needed to identify specific risk factors through which socioeconomic circumstances after birth contribute to the risk of RTI-related death or SUDI. Detailed characteristics of children who died and their familial and social environment could be described using the National Child Mortality Database (NCMD), a national data set collated from child death overview panels, available from April 2019.38 Additional linkage of NCMD to data on exposures in children who did not die, however, would be required to identify risk factors for child deaths.39 Linkage to census records in England could provide insights into the importance of factors such as the number of household members and rooms as indicators of overcrowding and housing conditions; socioeconomic inequalities in access to and uptake of preventive services could be explored through linkage to primary care records. Further analyses using causal mediation methods are also needed to quantify the total effect of socioeconomic factors on the excess cause-specific mortality in England relative to Sweden, including the effect mediated by adverse birth characteristics.

Some of the differences in RTI-related and SUDI mortality which remained unexplained could reflect differences in provision of preventive healthcare. Support for families with young children differs between the two countries; for example, in Sweden parents have longer paid parental leave and subsidised day care.5 Parents also build a trusting relationship with their nurse from the child health centre, who sees the family on average 14 times in the first 18 months of life.40 Such provision appears to improve uptake of vaccines and other preventive services.3 In England services are more fragmented: health visitors have only five mandated contacts with families in the first 2 years of a child’s life and some of the care is provided by general practitioners in primary care.41 Linkage to primary care records is needed to explore the contribution of differences in healthcare organisation and use to higher preventable child mortality in England compared with Sweden. Such data, however, are currently not collected in Sweden.

Our sensitivity analyses showed that chronic conditions contributed to an increased risk of RTI-related mortality beyond infancy in England, which remained after accounting for birth characteristics. Further work is needed to determine whether this is due to intercountry differences in the prevalence of chronic conditions, in the provision of healthcare for children with chronic conditions or in the recording of these conditions in hospitalisation records. Further comparison should also include information on what proportion of these deaths were expected (ie, whether RTI was part of terminal decline in child’s health, not possible to prevent) and what proportion were unexpected42.

Policy implications

Our results suggest that policies focusing on improving the health of women and socioeconomic circumstances of their families before and during pregnancy to reduce the prevalence of maternal risk factors and in turn of adverse birth characteristics could lead to substantial reductions in RTI-related and SUDI mortality in England. After accounting for birth characteristics, socioeconomic factors and chronic illness, we found weak evidence for an excess of RTI-related and SUDI mortality in England relative to Sweden. This finding suggests that improving the socioeconomic conditions of disadvantaged families and investment in health care before, during and after pregnancy for women and their children could reduce the gap in preventable child mortality in England and Sweden.