Childhood Health and Educational outcomes afteR perinatal Brain injury (CHERuB): protocol for a population-matched cohort study

Aim

The CHERuB (Childhood Health and Educational outcomes after perinatal Brain injury) study is a population birth cohort study that aims to investigate longitudinal childhood health and educational outcomes after perinatal brain injury. We propose using a national database capturing all children with perinatal brain injuries admitted to a neonatal unit and two matched control cohorts, linked to administrative health, education and social care data to ascertain long-term outcomes.

Objectives

• Work with affected families to inform the study design, focus, interpretation of results and their dissemination.

• Create a population-based national matched cohort of children with and without perinatal brain injury from existing national data sets.

• Examine neurodevelopmental, school and survival outcomes after perinatal brain injury up to 13 years of age compared with unaffected peers.

• Examine the prevalence and nature of chronic health conditions in children with perinatal brain injury compared with unaffected peers.

• Examine mental health outcomes after perinatal brain injury compared with unaffected peers.

Study population

The CHERuB study will include approximately 170 996 children: all infants born in England between 1 January 2008 and 31 December 2019 who were admitted to a neonatal unit with a perinatal brain injury (cohort 1; n=54 176) and two matched comparator groups of infants admitted to the neonatal unit and born at <34 (cohort 2) and infants born at ≥34 weeks’ gestation (cohort 3).

Cohort 1 will include infants with perinatal brain injury as per the DHSC definition in addition to infants with intraventricular haemorrhage grades 1–2 and mild HIE (online supplemental file 1).1 Infants born outside of England or with congenital infections, encephalopathies or brain abnormalities will not be included.1

Identifying comparators

Infants with perinatal brain injury born at <34 weeks’ gestation will be matched 1:1 with unaffected infants admitted to a neonatal unit using propensity scores accounting for the week of gestation, birth weight Z-score, sex, mode of delivery, multiplicity, maternal smoking status, receipt of antenatal steroids, receipt of antenatal magnesium sulfate and year of birth (cohort 2). This will be done using Mahalanobis matching (in the psmatch2 function in Stata) to identify the 20 nearest neighbour matches for each preterm infant with brain injury based on the prespecified covariates.11 The nearest unique match (of the 20) will then be selected to minimise duplicate comparators. Additional confounders affecting the preterm population necessitate more complex matching to create a balanced comparable cohort for meaningful analysis.

Infants with perinatal brain injuries born at 34 weeks’ gestation or greater will be matched to unaffected comparator infants indicated from a database of all births in England in a 1:3 ratio on sex, year of birth, birth weight (within 100 g), week of gestation (from 2015) and multiplicity (cohort 3).

Follow-up

Included children will be followed-up to 13 years of age, 31 August 2023, or their date of death through linked hospital, death, school and social care records. Follow-up time will vary by birth year.

Data sources

Cohorts 1 and 2 will be derived from the National Neonatal Research Database (NNRD). The NNRD contains demographic, clinical and organisational data derived from electronic patient records for all neonates admitted to a neonatal unit across Great Britain.12 13 The NNRD also captures information on short-term developmental outcomes at 2 years of corrected age for certain groups of neonates.13 Cohort 3 will be derived from Office for National Statistics birth registration data which pertains to all births registered in England. These data sets are described in brief in table 1.

Follow-up data will be obtained from the NNRD and linkage to administrative health and education records. All data sets are described in table 1. In brief, we will use:

• Diagnostic data recorded during hospital presentations, admissions and appointments in health records to measure indicators of neurodevelopmental impairments, chronic conditions and mental health conditions.

• ONS-Hospital Episode Statistics (HES) linked mortality records to obtain information about causes and timing of death.

• Education records to obtain information about educational attainment, level of provision and recorded type of special educational needs.

Data linkage

The suggested data flows for study data are outlined in figure 1. The main steps are:

Figure 1

CHERuB study data flows. MHSDS, Mental Health Services Data Set; NDAU, Neonatal Data Analysis Unit; NHS, National Health Service; NNRD, National Neonatal Research Database; ONS, Office for National Statistics; PDS, Personal Demographic Service.

1. Extraction and transfer of NNRD data to the ONS Secure Research Service (SRS): Infants with perinatal brain injury will be identified within the NNRD (cohort 1; n=54 176). The pseudonymised neonatal care data for this cohort will be transferred to the ONS SRS. Preterm infants (<34 weeks’ gestation) will be propensity score matched to a comparator group of infants in the NNRD (cohort 2; n=22 854). The pseudonymised neonatal care data for cohort two will also be transferred to the ONS SRS.

2. Linkage of the NNRD to the Personal Demographic Service (PDS): The NNRD reliably captures data items such as date of birth, postcode and infant NHS number. The NNRD cohorts will be linked to the PDS, a database that contains demographic data for all NHS patients in England with an NHS number and serves as a linkage spine for health records. Linkage will be done using NHS number, date of birth, sex and postcode at birth: to identify registered forename, surname and postcodes changes (which are required for linkage to education records).

3. Deriving cohort 3: The remaining unmatched infants (≥34 weeks’ gestation) with perinatal brain injury will be matched in a 1:3 ratio to a comparator group of infants, identified from ONS birth notifications by NHS England (cohort 3; n=93 966).

4. Linkage to health data sets via the PDS: Using NHS number, infant surname, forename, sex and date of birth, all three cohorts will be identified in the PDS by NHS England using their Master Person Service bespoke deterministic person-matching algorithm.14 The PDS serves as a spine for linkage to routinely collected health records: ONS mortality records, HES and the Mental Health Services Data Set (MHSDS). The data will be linked to the most recent available health records for the included children up to 13 years of age.

5. Linking to education and social care data: One file containing a list of personal identifiers for the linkage of the three cohorts to the education and social care data sets will be transferred to the Department for Education. The following identifiers will be used by the Department for Education to link to education and social care data: forename, surname, date of birth, sex and postcodes.15 Another file, containing the three pseudonymised cohorts of infants and their health outcomes, will be transferred to the ONS SRS.

Outcomes of interest

1. Survival and neurodevelopmental impairment.

The primary outcome of interest, survival without neurosensory impairment, will be determined up to 2, 5, 7 and 11 years of age using ONS mortality, HES, NNRD and National Pupil Database (NPD) records as binary outcomes at four time points. Two-year outcomes will be presented corrected and uncorrected for prematurity where possible.

The specific definitions for neurosensory impairment are denoted in online supplemental file 2 and include cognitive, motor, speech and language, visual and hearing impairments. The potential of extending this definition to include data from the MHSDS will be explored on data receipt as the utility of the MHSDS (available from 2016) for paediatric studies is still uncertain.

Survival without neurosensory impairment is a priority outcome for parents (as highlighted by previous core outcome studies in neonatology and HIE specifically).16 17 This definition of survival without neurosensory impairment has been developed by scoping the existing literature and through expert consensus with patient and public involvement (PPI) representatives, community paediatricians, general paediatricians, neonatologists and academics with expertise in NNRD, HES and NPD data.

We will consider the individual components of the composite primary outcome in secondary analyses as well as other important neurodevelopmental outcomes (online supplemental files 2, 3). These include:

• Neurosensory impairment (cognitive, motor, speech and language, hearing or visual).

• Any neurodevelopmental impairment.

• Cognitive impairment.

• Not meeting expected level of academic attainment.

• Learning difficulty components of special educational needs.

• Motor/physical impairment.

• Speech, language and communication difficulty.

• Visual impairment.

• Hearing impairment.

• Additional educational needs (any recorded special educational needs, Educational Health and Care Plan and attending a specialist school).

• Developmental delay (although we recognise this is not an impairment).

• Epilepsy.

• Behavioural, emotional and social disorders (hyperkinetic, autism spectrum disorder, tic disorder).

• Survival.

• Cause-specific mortality.

2. Chronic health conditions (* outcomes are not mutually exclusive; some conditions are classed as both a type of neurodevelopmental impairment and a chronic health condition).

This is defined as the occurrence of the chronic health conditions included in the Hardelid classification up to age 2, 5 and 11 years.18

• Infections.

• Cancer and chronic blood conditions.

• Cardiovascular conditions.

• Respiratory conditions.

• Musculoskeletal/dermatological conditions.

• Neurological conditions.

• Metabolic/endocrine/digestive conditions.

• Renal/genitourinary conditions.

3. Mental health conditions (* outcomes are not mutually exclusive; some conditions are classed as both an NDI and a mental health condition).

This is defined as the occurrence of behavioural, emotional, pervasive, mood and/or anxiety disorders up to age 5 and 11 years of age in HES or NPD data (online supplemental file 4). The potential of extending this definition to the MHSDS will be explored on data receipt.

Potential confounding or effect modifiers

Directed acyclic graphs (DAGs) will be created to map the relationships of the variables in table 2, types of exposures (ie, brain injuries) and outcome types.19 Selection of key covariates for each analysis will be guided ‘a priori’ by the DAGs. Potential effect modifiers will also be identified using DAGs and explored further within each analysis through stratification.

Statistical analysis

Analyses will be reported as per the Reporting of studies conducted using observational routinely-collected health data (RECORD) guidelines.20

The probability of survival without neurosensory impairment among the included population up to 2, 5, 7 and 11 years of age will be determined using multilevel logistic regression methods adjusting for the key covariates identified from DAGs and fitting hospital of birth as a random effect (to account for heterogeneity at the hospital level). The crude and adjusted ORs will also be calculated and presented for each of the secondary outcomes (ie, type of neurodevelopment diagnosis or additional needs) and by brain injury type accounting for the competing risk of death. These methods will also be used to study the risk of chronic health conditions and mental health conditions after perinatal brain injury. We will use Stata V.18 (or later) software for all analyses.

Further analysis of academic attainment will be undertaken using a linear regression model (adjusting for the key covariates) to determine the differences in mean academic attainment Z-scores on both teacher and national assessments at Phonics, Early Years Foundation Stage (EYFS), Key Stage 1 (KS1) and Key Stage 2 (KS2) after perinatal brain injury and by type of brain injury. This will also be undertaken for the specific domains that make up the overall teacher and/or national scores at each of the educational assessment stages (table 3).

Data permitting, a group trajectory analysis may be undertaken to study the differing academic trajectories at EYFP, KS1 and KS2 of infants with perinatal brain injury and to determine if and how these differ by brain injury type.21 22 This could provide insights into whether impairments as a result of specific perinatal brain injuries are fixed, deteriorate or even improve throughout childhood, in addition to identifying characteristics associated with these different trajectories.

To determine the difference in absolute mortality rates (and 95% CI) per 1000 person-years between those with and without perinatal brain injury (and by brain injury type) a survival analysis with Cox proportional hazards modelling will be undertaken. This will include censoring follow-up time, fitting the admission hospital or neonatal operational delivery network (data permitting) as a random effect using the frailty function, adjusting for covariates identified from DAGs and examining the data for breaches in the proportionality assumption.

Subgroup analyses

The primary and secondary outcomes will be reported for those with and without brain injury but also by type of brain injury. Additionally, outcomes for these specific populations will be explored geographically, temporally and in relation to specific treatments. Additional subgroup analyses of neurodevelopmental and school-based outcomes will include exploring how these outcomes differ based on eligibility for free school meals, being a ‘looked after child’, first language spoken at home, exclusions from schools and school type.

Missing outcome data

Infants from cohorts 1 and 2 (identified from the NNRD) with and without a 2-year follow-up record in the NNRD will be compared. Sensitivity analyses will be undertaken to assess the potential impact of missing data. For cohorts 1 and 2 the standardised difference for those with and without 2-year follow-up data in the NNRD will be calculated for the NNRD covariates specified in table 1. The standardised difference for those with and without 2-year follow-up data will also be calculated by the type of perinatal brain injury for cohort 1. The predominant mechanism of missing 2-year NNRD follow-up data, is likely to be missing at random (MAR), that is, predictable missingness related to covariates such as ethnicity, neonatal network responsible for follow-up and index of median deprivation.23 Some neonatal follow-up studies have shown an element of missing not at random (MNAR) in their missing follow-up data, where children with NDI are less likely to attend research study follow-up (due to a myriad of factors such as the burden of care appointments independent of ‘research’ appointments).23–25 Other studies have demonstrated that children without NDI might be less likely to attend follow-up.26 However, these were typically follow-up studies conducted alongside, rather than as part of, routine care. The data in the CHERuB study are collected as part of routine care, we would therefore expect this MNAR missingness mechanism to play less of a role. To explore the potential impact of MNAR, sensitivity analyses will be undertaken to simulate violations of the MAR assumption in either direction. The use of inverse probability weighting or multiple imputation methods will be considered to adjust for sources of potential bias resulting from missing outcome data.

The incidence of specific NDIs, chronic health conditions and mental health conditions in cohort 3 will be compared with estimates of the prevalence of these conditions within the general UK population from the literature and existing registries to assess the validity of the CHERuB study estimates.

Examining linkage quality

The linkage rate for cohort 1 and 2 from the NNRD to a HES records, and for cohort 3 from the ONS birth notifications to a HES records will be calculated. The characteristics of infants in cohorts 1–3 who did and did not link to a HES record will be compared by calculating standardised differences for the year of birth, sex, maternal ethnicity and index of multiple deprivation decile for each group, to evaluate for potential linkage bias as a result of missed matches.15 Furthermore a multivariate logistic regression analysis will be undertaken to determine whether these variables independently increase the risk of linkage failure. If necessary, multiple imputation methods will be considered to adjust for potential sources of bias resulting from linkage error.

The linkage rates of cohorts 1–3 from HES to the NPD will be calculated; and the characteristics of infants from the NNRD who did and did not match to an NPD record in addition to those from cohort 3 in HES who did and did not match to an NPD record, will be compared using the same methods as for the NNRD-HES linkage evaluation.

Patient and public involvement

The study and in particular the PPI components of the study are overseen by a PPI advisory committee which consists of parent representatives and representatives from Bliss, Peeps HIE and the meningitis research foundation. Families of infants with perinatal brain injury have also been recruited via charity partners (Bliss and Peeps HIE) to participate in PPI workshops. Early family workshops have been undertaken to inform the study design and focus. Further workshops are planned later in the study to explore the study results with families: particularly what specific results might mean to families and to seek their input on how best to disseminate and use these findings going forward.