Data Resource Profile: A national linked mother-baby cohort of health, education and social care data in England (ECHILD-MB)
International Journal of Epidemiology · 2024
DOI: 10.1093/ije/dyae065 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Epidemiology, 2024. Verfasst von Qi Feng und 3 weiteren Autor:innen. Beteiligte Einrichtung: Great Ormond Street Institute of Child Health, University College London, London, UK. Hauptthemen laut MEDLINE: Mothers. Die Arbeit wurde bisher 15-mal zitiert (Crossref) und nennt 38 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0). Gefördert von: Economic and Social Research Council.
Abstract (Originalsprache)
The Education and Child Health Insights from Linked Data Mother-Baby (ECHILD-MB) cohort was created by linking data from National Health Service (NHS) Hospital Episode Statistics (HES) with the National Pupil Database (NPD), to examine intergenerational effects of maternal exposures on child outcomes in education, health and social care in England. We extracted 14.5 million baby records from HES for births between 1 April 1997 and 31 January 2022 and, using a validated mother-baby data linkage algorithm, linked 13.6 million (94.1%) of these to the delivery records of 8.0 million mothers. The linked cohort captures 87.3% of all births and 87.7% of all live births in England, and as of 2023, includes mothers aged 12–37 and their children aged 0–24 years. The cohort is representative of national birth statistics. All individuals in the cohort are linked to HES and the NPD, which include routinely collected data on sociodemographics, education, health and use of children’s social care. The ECHILD-MB cohort data is accessible to accredited researchers as part of the ECHILD project [echild.org.uk]. Maternal physical, psychological and social risk factors extend beyond affecting mothers' individual wellbeing to significantly influence their children. Research consistently shows that maternal exposure to poor nutrition and psychological and social stressors prior to and during pregnancy is associated with an increased risk of adverse birth outcomes, developmental disorders and chronic health conditions of children in both childhood and later life.1–10 These maternal exposures may even date back to the mother’s childhood.8 Recognizing the intricate interplay of these factors within families and across generations is pivotal for designing targeted interventions to enhance the health and wellbeing of current and future generations. Longitudinal mother-baby cohort studies are useful in addressing research questions about the influence of maternal exposures on outcomes in their child. There are generally two types of mother-baby cohorts: recruited and consented cohorts, and administrative data-derived cohorts. A key advantage of recruited and consented cohorts is the capture of assessment data from mothers and their children throughout various stages of pregnancy and beyond. In the past decades, many such cohorts have been established.11 Examples in England include Avon Longitudinal Study of Parents and Children study,12 1970 British Cohort Study,13 Born in Bradford cohort,14 Southampton Women’s Survey15 and the Gateshead Millennium Study.16 Some of these cohorts are linked to administrative data for follow up.12,17 These cohorts collect data on maternal exposures during pregnancy, but often have limited data on maternal exposures prior to pregnancy, such as the mother’s childhood, adolescence, and early adulthood. Furthermore, these cohorts can be constrained by relatively small sample sizes, limited follow-up periods or geographical restrictions.11 By contrast, administrative data-derived cohorts have distinct advantages, including comprehensive coverage, being less prone to the selection biases, the ability to examine cohorts from past decades and, very recently, the power to examine rare conditions. Several countries have derived national longitudinal administrative cohorts.18,19 We derived a national mother-baby cohort nested within the Education and Child Health Insights from Linked Data (ECHILD) Database.20,21 ECHILD is a linked collection of the National Health Services (NHS) Hospital Episode Statistics (HES) and the Department for Education (DfE) National Pupil Database (NPD) for a population-based cohort of children and young people born in England. In this data resource profile, we describe how we derived the mother-baby cohort, and we present the results of the linkage and its validation, as well as the basic characteristics of the cohort. HES captures separate records of births for babies and deliveries for mothers in NHS hospitals. However, routine identification of mother-baby pairs is lacking. This section outlines the statistical process used to perform mother-baby linkage using de-identified data, aiming to identify and link baby birth records to maternal delivery records. A concise overview of the method is provided here, with detailed information available in the Supplementary Methods (available as Supplementary data at IJE online). We first identified 14 494 782 birth records and 14 611 863 delivery records between 1 April 1997 and 31 January 2022 from HES data using validated codelists. In HES, delivery episodes for mothers and birth episodes for babies include additional fields on delivery procedures and outcomes, which are called the baby/maternity tail. The baby/maternity tail ideally contains the same information in delivery and birth records, but is sometimes incomplete. Second, we used a previously validated algorithm for linking delivery and birth records.22,23 We performed deterministic linkage using seven linking variables (including location, delivery and birth characteristics), and probabilistic linkage using 23 linking variables for the unlinked records. Record pairs with implausible dates were not considered; for example, babies discharged prior to the mother’s admission, or mothers discharged prior to the baby’s admission. For record pairs surpassing selected match weight cutoff values, we prioritized the highest match weight for each baby and then for each mother. Mother-baby linkage was conducted individually for each financial year (1 April to the next 31 March) due to varying data quality over time in HES. In probabilistic linkage, matches across financial years were permitted by incorporating maternal delivery data from March of the preceding financial year to April of the subsequent financial year. In total, 13 635 655 out of 14 494 782 birth records (94.1%) were linked to delivery records (Figure 1). The linkage rate varied by year, with the lowest linkage rate in 1997 (89.5%) and 1998 (89.4%), and the highest linkage rate in 2010 (96.2%); the linkage rate for singleton births was higher than for multiple births (94.4% vs 80.7%) (Supplementary Figure S1, available as Supplementary data at IJE online). Flow diagram of participant recruitment in ECHILD-MB cohort. For records where the same TokenID (unique individual identifier in HES data) was assigned to multiple individuals, records were considered as different individuals when they had discrepant values in birthweight, maternal age or gestational age. For the records where the same individual was assigned multiple TokenIDs, records were considered as the same individual when they had consistent values in all other variables. aIn HES, one birth event may take multiple episodes, in which case we included the episode that had the highest data completeness (lowest missing level). HES, Hospital Episode Statistics; ECHILD-HES, Education and Child Health Insights from Linked Data—Hospital Episode Statistics; ECHILD-MB: national linked mother-baby cohort of health, education and social care data in England Linkage rates also varied by hospital. Among the 341 NHS hospital groups (identified by three-digit hospital codes), 43 groups (birth n = 9148) had a 0% linkage rate, 13 groups (birth n = 55 846) 0.1–10.0%, nine groups (birth n = 14 024) 10.1– 50.0% and 93 groups (birth n = 4 062 049) 50.1–94.1%. The remaining 183 groups (birth n = 10 353 715, 71.4% of all births) had linkage rates of >94.1%; 22 groups (birth n = 135 349) had linkage rates >99.0%. Compared with linked babies, the 5.9% of unlinked babies were more likely to have Black or Mixed ethnic background (12.5% vs 10%), to have an older mother (31.0 vs 29.2 years), to be born before or at 37 weeks (24.4% vs 14.3%) and to have lower birthweight (3120 vs 3340 g) (Table 1). Unlinked babies were also more likely to be admitted to special care (17.0% vs 10.3%) or intensive care (6.8% vs 2.7%), or be recorded as being a stillbirth (1.4% vs 0.2%). Similar patterns were observed in singleton births and multiple births (Supplementary Table S1, available as Supplementary data at IJE online). Characteristics of the linked and unlinked baby birth records When calculating percentage for the Unknown category, the denominator was the total number of babies; when calculating percentage for other categories, the denominator was the total number of babies excluding the unknown category. IQR, interquartile range; SD, standard deviation. Characteristics of the linked and unlinked baby birth records When calculating percentage for the Unknown category, the denominator was the total number of babies; when calculating percentage for other categories, the denominator was the total number of babies excluding the unknown category. IQR, interquartile range; SD, standard deviation. We used the 2021 Community Services Data Set (CSDS) 24 as a gold standard to evaluate the linkage results. Despite representing only a subset of all births, CSDS facilitates valuable linkage validation, as the mother’s unique ID is captured on the child’s record, which is the same as the IDs used in HES. CSDS comprised records for 2 581 393 babies from 313 local authorities in England. We estimated missed match rates and false match rates.23,25 Overall, 2 541 772 (98.5%) were linked in this study, and the remaining 39 621 were missed (1.5%). Of the linked records, 2 523 476 were true matches, giving a positive predictive value of 99.3% (2 523 476 true matches/2 541 772 linked records), a false match rate of 0.72% (18 296 false matches/2 541 772 linked records) and a sensitivity of 97.8% (2 523 476 true matches/2 581 393 total mother-baby dyads in CSDS; Supplementary Table S2, available as Supplementary data at IJE online). We evaluated the coverage of mother-baby linkage among all births and live births, relative to the Office for National Statistics (ONS) statistics (1998–2021). The identified birth records covered 92.7% of all births and 92.9% of live births, and the linked babies covered 87.3% of all births and 87.6% of live births (Supplementary Table S3, available as Supplementary data at IJE online). We compared the distributions of gestational age, birthweight and maternal age with national birth statistics published by ONS, and observed high levels of agreement (Figure 2, Supplementary Figure S2, available as Supplementary data at IJE online). Comparison of the mother-baby cohort with data from the Office of National Statistics (ONS) on maternal age, gestational age and birthweight in the 2021 financial year. Red: ECHILD-MB (national linked mother-baby cohort of health, education and social care data in England) data. Green: Office for National Statistics. Each financial year starts on 1 April and ends on 31 March next calendar year The mother-baby cohort, nested within the ECHILD database,20,21 shares data sources (NPD and HES) with ECHILD and is de-identified. Both NPD and HES regularly collect and compile information, undergoing quality assurance checks upon submission to DfE and NHS England. NHS England performed the linkage between HES and NPD, with details outlined elsewhere.26 Briefly, DfE securely transferred identifiers from NPD to NHS England, where deterministic linkage was used to create an anonymized linkage spine connecting the NPD unique identifier (aPMR) to the HES unique identifier (TokenID). HES contains records for all hospital activity provided or paid for by NHS, including births, inpatient admissions, outpatient appointments, accident and emergency attendances, mortality, demographics and standardized codes for diagnosis and procedures.27 Data on socioeconomic status are collected at area level using Index of Multiple Deprivation. NPD contains records related to state-funded education and the use of social care services. NPD data are collected by local authority, but the specification of data that are collected is centralized and determined by the DfE. It includes information from different educational settings about pupils’ characteristics, including age, gender, ethnicity, special educational needs and free school meals. Educational outcomes are also documented, including national assessments and examinations, absences, exclusions and participation in post-16 education.26,28 NPD's social care modules, Children in Need and Children Looked-after Return, collect information on children using social care services and those in care (looked-after children in the UK) (Table 2). Data on socioeconomic status are collected at area level using the Index of Multiple Deprivation and at individual level using eligibility for free school meals as a surrogate measurement. Key data sources included in a national linked mother-baby cohort of health, education and social care data in England (ECHILD-MB) Information on diagnoses, treatments, and procedures for each episode of care is recorded by clinical coders based on patient care records and/or discharge summaries using standardized codes. In the Admitted Patient Care, Critical Care and Outpatient modules, diagnoses are recorded using the International Classification of Diseases (ICD) version 10, and treatments and procedures are recorded using the Office of Population Censuses and Surveys (OPCS) version 4. In the Accident and Emergency module, bespoke codes are used to record diagnoses and treatments14; however, these are much more limited than ICD-10 and OPCS-4 codes. The School Census Pupil Level module is collected on a termly basis in October (Autumn census), January (Spring census), and May (Summer census). The other education census modules are collected in January only. HES, Hospital Episode Statistics; ONS, Office for National Statistics; NPD, National Pupil Database; PRU, pupil referral unit; SEN, special educational needs; FSM, free school meals; GCSE, General Certificate of Secondary Education (national examinations taken by students at the end of compulsory education); KS, Key Stage. Key data sources included in a national linked mother-baby cohort of health, education and social care data in England (ECHILD-MB) Information on diagnoses, treatments, and procedures for each episode of care is recorded by clinical coders based on patient care records and/or discharge summaries using standardized codes. In the Admitted Patient Care, Critical Care and Outpatient modules, diagnoses are recorded using the International Classification of Diseases (ICD) version 10, and treatments and procedures are recorded using the Office of Population Censuses and Surveys (OPCS) version 4. In the Accident and Emergency module, bespoke codes are used to record diagnoses and treatments14; however, these are much more limited than ICD-10 and OPCS-4 codes. The School Census Pupil Level module is collected on a termly basis in October (Autumn census), January (Spring census), and May (Summer census). The other education census modules are collected in January only. HES, Hospital Episode Statistics; ONS, Office for National Statistics; NPD, National Pupil Database; PRU, pupil referral unit; SEN, special educational needs; FSM, free school meals; GCSE, General Certificate of Secondary Education (national examinations taken by students at the end of compulsory education); KS, Key Stage. Due to variations in age coverage and collection periods across NPD datasets, the availability of specific datasets for mothers and babies is contingent on their respective birth years. Figure 3 demonstrates the availability for a mother born in 1988 and her child born in 2012. To facilitate visualization of data availability for any mother-baby pair, we created an online tool [https://ermadake.shinyapps.io/echild_mb_data/]. Example of data availability of different National Pupil Database (NPD) datasets for a mother born in 1988 and her child born in 2012. Shaded: data available. Empty: data not available. Green: education datasets. Blue: social care datasets. Orange: health data. EYFSP, Early Year Foundation Stage Profile; EYC, early year census; NCCI, National Client Caseload Information system. We developed an online tool to visualize the data availability for all possible mother-baby pairs [https://ermadake.shinyapps.io/echild_mb_data/]. Data resource profile: a national linked mother-baby cohort of health, education and social care data in England (ECHILD-MB) We imputed missing values in birth records using the linked delivery record of the mother. We removed implausible values of gestational age and birthweight that fell more than three standard deviations from the average.29 In case of multiple births, we also imputed missing values in gestational age and maternal age by copying data across multiples. After imputing missing values in baby records using values in mother records, data completeness increased (Supplementary Table S4, available as Supplementary data at IJE online). The ECHILD-MB cohort serves as a valuable resource for investigating intergenerational effects of maternal exposure before and during pregnancy on children's health, education and social care outcomes. For example, previous research has examined the associations between pre-pregnancy psychosocial risk factors and infant outcomes, finding that exposure to psychosocial risk factors, including teenage motherhood, previous teenage motherhood, a history of adversity, mental health or behavioural conditions, or living in the most deprived quintile during the 2 years before pregnancy, significantly increased children's risk of low birthweight, injury admission and mortality during infancy.7 The ECHILD-MB will allow this area of research to be further developed due to the inclusion of information on educational and social care contacts. As the ECHILD-MB cohort matures, the extended follow-up will enable further comprehensive analyses. The cohort extends its utility beyond examining health outcomes at birth and infancy by facilitating investigations into long-term health trends during childhood, adolescence, early adulthood and later life stages. Additionally, the cohort uniquely allows for the exploration of educational outcomes and use of social care services. These findings are instrumental in identifying vulnerable children at heightened risk of adverse outcomes, informing the design of early intervention strategies for targeted and effective support. The ECHILD-MB cohort is characterized with a substantial sample size, seamless integration with national administrative databases, and extensive data encompassing education, health and social care. Further strengths include nationwide coverage and representativeness of mother-baby pairs, good linkage accuracy and the unique capability to explore intergenerational effects of maternal exposures. The mother-baby cohort encompassed 87.3% of all births and 87.6% of live births in England from 1998 to 2021. The under-ascertainment can be explained by several factors. First, a 5.4% opt-out rate (as of November 2023) among English residents limited data sharing for research or policy purposes.30 Second, the ONS Birth Registration dataset, mandated to report all births in England, includes records from various settings such as NHS hospitals, private hospitals and homes. on HES birth records in NHS hospitals, in the of births not recorded in HES. has that of Birth Registration records can be linked to The linkage rate in mother-baby linkage was with a prior using a HES for linking singleton Linkage rates varied across hospitals and clinical for NHS data The linkage rate was with data higher rates in later consistent with the data completeness in HES birth records over with of births in hospitals linkage rates linkage rate and lower data of linked and unlinked babies identified risk factors for including Black or Mixed and older maternal age, all associated with adverse birth as observed in this individuals and adverse clinical data in a lower linkage linkage in imputing missing values, dat
Schlagwörter
MeSH-Schlagwörter (MEDLINE)
Stichwörter der Autor:innen
Autor:innen und Institutionen
- Great Ormond Street Institute of Child Health, University College London, London, UK.
ORCID-Profile: Qi Feng, Katie Harron
Förderung
- Economic and Social Research Council
Zitieren
APA
Feng, Q., Ireland, G., Gilbert, R., & Harron, K. (2024). Data Resource Profile: A national linked mother-baby cohort of health, education and social care data in England (ECHILD-MB). International Journal of Epidemiology. https://doi.org/10.1093/ije/dyae065
Vancouver
Feng Q, Ireland G, Gilbert R, Harron K. Data Resource Profile: A national linked mother-baby cohort of health, education and social care data in England (ECHILD-MB). International Journal of Epidemiology. 2024. doi:10.1093/ije/dyae065
BibTeX
@article{Feng2024,
author = {Feng, Qi and Ireland, Georgina and Gilbert, Ruth and Harron, Katie},
title = {Data Resource Profile: A national linked mother-baby cohort of health, education and social care data in England (ECHILD-MB)},
journal = {International Journal of Epidemiology},
year = {2024},
doi = {10.1093/ije/dyae065}
}
EUVIMED