Linkage of administrative family court care proceedings and hospital records for mothers in England: linkage accuracy and cumulative incidence of family court care proceedings after a first live birth

Abstract

Introduction
Linkage of public law family court care proceedings (CP) data to all women giving birth in NHS hospitals in England allows calculation of the cumulative incidence of CP involvement for mothers with first children born.

Objectives
To assess linkage accuracy and determine the 10-year cumulative incidence of CP after a first live birth (FLB) for population subgroups.

Method
NHS England linked records for mothers in Cafcass (Children and Family Court Advisory and Support Service) involved in CP (2007-2021) to all mothers with a delivery in England using Hospital Episode Statistics (HES: 1997-21). We calculated match rates and assessed indirect evidence of potential false positive and missed links. We used survival analyses to estimate cumulative incidence of CP within 10 years overall and for five-year maternal age groups at first live birth.

Results
Of 120,937 mothers involved in CP, 6.6% (n = 8,010) were excluded due to missing postcode or date of birth, or age < 15 or > 50. Of the remaining 112,927 mothers, 92,891 (82.8%) were linked to a HES delivery record. Match rates were lowest for mothers with an ethnic minority background, older at first case, or residing in Greater London, but improved over time.

Of 3,572,737 mothers with a FLB, 38,462 had CP involvement. The cumulative incidence of CP at 10 years from FLB was 1.31% (95% Confidence Interval [CI]; 1.29-1.32) overall and highest in mothers aged 15--19 years (6.79%, 95% CI: 6.69-6.89) and those living in the most deprived areas (2.47%, 95% CI: 2.43-2.51).

Conclusion
One in 77 of all mothers and one in 15 aged less than 20 at first live birth were involved in CP within 10 years. Linkage error may underestimate the incidence of CP for mothers in London or with an ethnic minority background.

Key points

• Overall, 82.8% of women recorded as a mother in Cafcass care proceedings were linked to a hospital delivery record.
• Match rates were lowest for mothers with an ethnic minority background, older age at first child, or residing in Greater London.
• 1.3% of all mothers (1 in 77) with a first birth were involved in care proceedings within 10 years and 6.8% (1 in 15) of mothers aged <20 at first live birth.

Introduction

Involvement in public law family court care proceedings reflects serious breakdown in the care or safety of a child. Care proceedings are brought by local authority (LA) social care departments under section 31 of the Children Act 1989 due to concerns about significant harm, or risk of significant harm to the child attributable to care given by the parents or the child being beyond parental control. The court decides on whether or not to place the child under LA care or supervision. Of the 10,000 mothers involved with care proceedings each year in England, over 80% have their child(ren) placed in state care or in kinship care under state supervision [1, 2].

Population-based retrospective and prospective studies using linked administrative data in Wales and Canada report that young maternal age, living in the most deprived neighbourhoods, and problems such as poor mental health, substance use problems, and domestic violence are risk factors for care proceedings or children being taken into care, and often precede childbearing [3–7]. Qualitative studies in England reveal that around 2 in 5 mothers involved in recurrent care proceedings have themselves experienced child maltreatment [8]. Prospective, population-based administrative data studies in Canada and Sweeden report that mothers of children taken into care have elevated mortality compared with peers [9–11]. These findings suggest that early health interventions and intensive support for parenting might avoid court involvement entirely or mitigate the resulting health harms on parents and children.

National, whole population administrative data linking healthcare utilisation to records of family court care proceedings has not before been available for the whole of England. Pearson et al. (2021) linked administrative data from family court care proceedings to de-identified records from mental health services in 4 local authorities in South London, involving 3,200 mothers followed for up to 12 years [12]. Two-thirds of mothers had contact with mental health services- 80% of whom received secondary or tertiary mental health care before onset of family court care proceedings [13].

We developed an administrative database to research the upstream health determinants and outcomes of care proceedings for mothers and their children in England. We linked administrative data from Children and Family Court Advisory and Support Service (Cafcass) care proceedings and hospital records from Hospital Episode Statistics (HES) for women giving birth in an NHS hospital in England. In this study, we describe how the linked cohorts were derived and assess the match rate and indirect evidence of potential false or missed matches. We also estimate the cumulative incidence of care proceedings within 10 years of a first live birth, according to demographic subgroups, and discuss potential biases relating to linkage error. These findings are relevant to policy makers in health and family justice and will underpin applications for permissions to enable wider access to these linked data for wider research.

Methods

Data sources, flows and linkage

We used routinely recorded administrative data collated from NHS hospitals in England (HES) and data from public law family court care proceedings, collated by Cafcass [14, 15]. The data flows are summarised in Figure 1 and briefly described below. For more detail see Supplementary Material Sections 1–3.

Figure 1: Data flow and linkage process between administrative datasets from Cafcass, the Personal Demographic Service and Hospital Episode Statistics. Notes: Dark shading indicated de-identified and light shading identified data. Cafcass linkage data refers only to identities associated with mothers aged 15-50 years at start of a section 31 public family court case in England between 1^st April 2007 and 31^st December 2021. PDS = Personal Demographic Service, HES = Hospital Episode Statistics, UCL = University College London, DSH = Data Safe Haven.

Cafcass: Public law family court care proceedings

Cafcass holds data on all public family law cases involving section 31 applications in England (Supplementary Table 1). Children in public care proceedings are automatically party to proceedings and are represented by a Cafcass guardian and their own solicitor (see definitions relevant to Cafcass in Supplementary Material Section 1). Cafcass was established on 1^st April 2001 and de-identified data recording cases have been usable from 1^st April 2007 [14]. Data associated with each case includes dates of hearings, number and ages of children involved in the application and age of adults considered as carers, and thereby, party to the case.

Linkage was conducted following the separation principle whereby identifiers were transferred from Cafcass to NHS England (NHSE) and only attribute data were transferred to the UCL Data Safe Haven [16]. Figure 1 Step A shows the creation of a Cafcass extract within the Cafcass secure case management system for mothers aged 15-50 at the start of their first case (1^st April 2007 to 31^st December 2021) and classified as a parent in section 31 proceedings. Cases were deduplicated probabilistically within Cafcass, using name, date of birth and sex before transfer to NHSE. Step B shows the transfer of the Cafcass linkage file from Cafcass to NHSE, which contained first name, middle name (where available), surname, date of birth, sex (all female) and address (≤3 most recent addresses). Identifiers available in Cafcass, the personal demographic service (PDS) and HES are listed in Supplementary Material Section 2.1, Table 2. Separately, a de-identified Cafcass attribute file containing attribute data on family court care proceedings were transferred from Cafcass to the UCL Data Safe Haven. This process is known as the separation principle. A project-specific anonymised Cafcass-ID was attached to each record in the attribute and linkage files.

NHS England (NHSE): Personal Demographic Service (PDS)

NHSE linked the Cafcass linkage file to PDS. PDS contains a full chronology of identifiers from 2004 onwards for all people who are registered with an NHS GP or referred to or attend NHS care in England [17].

During linkage stage 1 (Step C), NHSE linked the Cafcass linkage file to PDS using two automated processes: the alphanumeric trace and algorithmic trace (Supplementary Material Section 2.2) [18]. The alphanumeric trace is a deterministic link using name, date of birth, sex and address. If Cafcass records were not linked to PDS during the alphanumeric trace, the probabilistic algorithmic trace was used. The algorithmic trace block linked the remaining records using combinations of name, date of birth, sex and postcode, and match options were scored. Each PDS candidate was scored based on the similarity of features from the query record. The score is calculated from the average of similarity scores of given and family names, date of birth (including partial date of birth), postcode (including partial and ≤3 historical postcodes) and sex. The highest scoring candidate PDS record was chosen as the matching record. If no candidates were found, or >1 PDS record had the highest score, then no record was returned. Manual review of links within NHSE was not possible. The NHS number was obtained from PDS for Cafcass-PDS linked records.

During linkage stage 2 (Step D), Cafcass-PDS linked identifiers were matched to HES identifiers using the NHS number from stage 1 and the NHS number was replaced with the project specific HES-ID for UCL. NHSE then securely transferred a linkage bridging file, containing only the Cafcass-ID and the UCL-HES-ID to the UCL Data Safe Haven (Step E). The Cafcass identifier linkage file was destroyed six months after NHSE first received it.

UCL Hospital Episode Statistics (HES): Extract for mothers with a delivery admission

We used an extract of HES, supplied by NHSE that contained all Admitted Patient Care (APC) and any death registration records for mothers with a record of a live or still birth in a NHS unit in England between 1^st April 1997 and 31^st December 2021 [15, 19]. Derivation of the delivery cohort (Figure 2 boxes a to d) is described in detail in Supplementary Materials Section 3). We excluded mothers who were not recorded as resident in England and those aged less than 15 years or older than 50 years at first delivery. All hospital admissions for the same woman were linked over time using the anonymised UCL-HES-ID provided by NHSE, specifically encrypted for our data sharing agreement. Data were analysed in the UCL data safe haven.

Figure 2: Results of data extraction from Cafcass and HES delivery records and linkage of Cafcass to HES, the Delivery Cohort and the First Live Birth Cohort. Notes Orange box represents exclusion from Delivery or Cafcass data at each stage. Light blue line represents number of mothers within Cafcass who linked to the Delivery and First Live Birth Cohorts. 1-Includes delivery episodes for women 12-14 years old at delivery. Women under 15 years old at first delivery were excluded following processing of delivery episodes (steps b->c). 2-see Supplementary Material Section 3.4 3-female, party to public family law case and parent of child involved. 4-Deduplicated using HES ID on accepted matches during step iv. Deduplicated when 1 HES ID matched to multiple mothers in Cafcass, following which the Cafcass records of these mothers were merged.

Linkage of the deidentified Cafcass attribute file to the HES records was done in the UCL Data Safe Haven using the linkage bridging file (Step F). NHSE did not provide any information on which records linked at stage 1 but not at stage 2 (i.e., they only provided information on those who linked at both stages).

Study populations

We used the HES extract to create two study populations. First, we derived a delivery cohort of all mothers residing in England who had a delivery record for a live or still born baby recorded in HES between 1^st April 1997 and 31^st December 2021 linked to Cafcass data (Figure 1 step F, Figure 2-a to d). The delivery cohort was used to determine the match rate and evidence of false positive matches. Second, we created a subset of mothers in the delivery cohort with a first live birth (FLB), referred to as the FLB cohort, between 1^st April 2007 and 31 December 2021 (Figure 2-e). The FLB cohort included mothers involved in care proceedings and their children born from 1^st April 2007 in England (Figure 2-vii). We used the FLB cohort to estimate the cumulative incidence of Cafcass proceedings after a first birth (Figure 2-e).

The process of creating a delivery cohort of women who gave birth in the NHS, including sequential delivery episodes, in England between 1^st April 1997 and 31^st December 2021 was conducted in four stages (Supplementary Material Section 3). In stage 1, we extracted delivery episodes from HES APC based on ICD 10 diagnosis and OPCS procedure codes and maternity tail (MT) variables [19], identified the first delivery and estimated delivery date for each woman.

In stage 2, we identified subsequent deliveries that were at least 24 weeks after the first delivery. We repeated the process to identify deliveries and estimated date of delivery until no further delivery episodes were available for each woman.

In stage 3, we assigned birth status (live or stillborn) and identified multiple births using diagnostic and maternity tail information (Supplementary Material Section 3.3). We adjusted the sequence of first, second and subsequent births by comparing the sequence of births in HES APC and the maternity tail variable indicating the number of registrable pregnancies prior to the delivery. Finally, exclusion criteria were applied to the cohort to maintain data quality, removing 217,835 mothers (Figure 2-b to c). A full list of exclusion criteria are provided in Supplementary Material Section 3.4 but include under 15 years at first delivery, resident outside of England, poor data quality from specific providers and years, delivery episodes <24 weeks apart and maternal deaths recorded prior to or between deliveries.

We identified first-time mothers to derive the FLB cohort from 1^st April 2007 until 31^st December 2021 (Figure 2-e; Supplementary Material Section 3.5).

To assess representativeness of the delivery cohort to all births in England, we compared annual aggregate numbers and prevalences from the Delivery Cohort with birth registrations published by the Office for National Statistics (ONS) for England for: annual number of deliveries (to provide delivery cohort coverage estimates); maternal age-group and deprivation quintile at delivery at 5 year intervals between 2000 and 2020; annual proportion of deliveries with multiple births and stillbirths; and the annual proportion of all deliveries born to primiparous women (first-time mothers (Supplementary Material Section 4) [20].

Evaluation of linkage quality

Linkage between de-identified Cafcass attribute file and the delivery cohort occurred at Step G (Figure 1) using the linkage bridging file. Figure 2 shows the number of unique mothers with care proceedings in Cafcass attributed records transferred to UCL DSH at Step B (Figure 2-i), the number of mothers sent to NHSE for linkage (Figure 2-ii) and the number of linked cafcass ids included in the returned linkage bridging file (Figure 2-iii).

Cafcass-HES matches in the returned bridging file were assessed. Results for mothers who linked to PDS but not to a HES record were not provided by NHSE.

We determined the overall match rate by dividing the numerator of mothers recorded in Cafcass who linked to the HES delivery cohort (1997–2021, Figure 2-vi) by the number of all mothers recorded in Cafcass and eligible for linkage (Figure 2-iii).

To evaluate potential bias resulting from missed links between Cafcass and the HES delivery cohort, we described demographic characteristics of mothers recorded in Cafcass by whether they linked to the delivery cohort or not.

Univariate and multivariable logistic regression, adjusted for ethnicity, age, number of children, year and region at first case, number of cases between 2007 and 2021 and postcode availability was used to assess predictors of linkage to a delivery record among mothers eligible to link to HES (had date of birth and between 15–50 at first case). Unlinked records could be interpreted as missed links, but could also include duplicate records of mothers in Cafcass (with one record having poorer identifiers) or mothers who linked to more than one HES ID. We could not distinguish duplicates as we did not have access to identifiable records held by Cafcass. Unlinked mothers could also have given birth outside England or prior to 1997.

To provide indirect evidence of potential false links, or erroneous recordings, we explored disagreement between characteristics that were recorded in both HES and Cafcass among mothers with a first live birth from 1^st April 2007 who linked to Cafcass. Common variables in HES and Cafcass were ethnicity, region, number of children, age at first delivery (HES)/oldest child(Cafcass) and year of birth (estimated in Cafcass using year of first case and age at case start and allowing for ±1 year difference). For each variable, HES and Cafcass could either agree, disagree or the information was missing from one or both datasets. Frequency of combinations were reviewed and overall disagreement scores calculated (for each variable: 0 if agree or missing, 1 if disagree). Disagreement provided indirect evidence of a potential false match (although we could not distinguish between errors in data recording and linkage errors).

Estimation of the cumulative incidence of care proceedings at 10 years

We use Kaplan-Meier analyses to estimate the cumulative incidence, and 95% confidence interval (CI), of a first care proceedings at 5 and 10 years after first live birth (using Stata/MP Version 18). Follow-up started at first live birth and ended at whichever came first; start of first care proceeding, the date 10 years after first live birth, 31^st December 2021, or death (using linked death registration or death recorded as discharge method in HES). Cumulative incidence at 10 years was estimated for the following population subgroups: maternal age-group, ethnic group, quintile of area-based index of multiple deprivation (IMD) a measure of the concentration of deprived households within a small area (lower super output area, which contains an average of 1,500 individuals), year of first live birth and health region of first delivery [21].

Sensitivity analysis to addressing underestimation of cumulative incidence estimates

Match rates demonstrate that underestimation of the cumulative incidence of care proceedings is likely. The extent to which incidence of care proceedings within 10 years of a FLB was underestimated due to missed matches was also assessed quantitatively using the records of mothers who did not link to HES. Among unlinked mothers recorded only in Cafcass, year of FLB was estimated using maternal age at oldest child and first case start (where available). We excluded mothers with an estimated FLB before 2007. Under scenario 1, the least conservative estimate, all remaining mothers were added to the crude numerator of mothers who experienced care proceedings within 10 years of a FLB. Under scenario 2, only mothers with a child less than 2 years at first case were added to the numerator, as we assumed that births close in time to care proceedings were more likely to have been in England. The percent increase in crude incidence from scenario 1 and 2 was applied to Kaplan-Meier estimates of cumulative incidence. These scenarios assume there are no duplicate mothers in the unlinked pool, that all children associated with the case were delivered within the NHS in England, all mothers without a date of birth were 15-50 years old at case start and that unlinked mothers would contribute similar follow-up time to linked mothers.

Results

Data flows and linkage

Between 1^st April 2007 and 31^st December 2021, 120,937 mothers were party to care proceedings involving one or more children (Figure 2-i). The number of unique mothers eligible for linkage was reduced to 112,927 (Figure 2-ii) because mother’s date of birth was not recorded in 4.9% (n = 5,935) and maternal age was outside age 15-50 years (1.7%; n = 2,075). Recording of date of birth varied by ethnic background, number of care proceedings and region- with London (9.2%) and the East of England (6.8%) having the highest proportions of mothers missing date of birth (Supplementary Table 1). Completeness of date of birth improved over time, from 90.2% in 2007-10 to 97.6% in 2019–21. An additional 1.7% (n = 2,075) were excluded as they were not aged between 15 and 50 years at start of first case. These 112,927 mothers (93.4%) were included in the Cafcass Linkage File and were sent to NHSE.

Study populations

Between 1^st April 1997 and 31^st December 2021, 14,881,615 delivery episodes in 8,488,511 women aged 12–50 years old were extracted from HES (Figure 2-a, see Supplementary Material Section 3). Processing steps identified 14,581,400 separate deliveries in these women. Overall, 168,608 women (equivalent to 282,654 deliveries) were excluded from the cohort (see Supplementary Material Section 3.4), leaving a total of 8,319,907 women aged 15 to 50 years with 14,298,746 deliveries.

For the 1998-2021 period, the Delivery Cohort was estimated to contain 92.9% of all births registered in England by the Office for National Statistics, increasing from 87.0% in 1998 to 95.2% in 2013, then falling to 91.7% in 2021. Further comparison of the delivery cohort to ONS data on birth registrations for England is reported in Supplementary Material Section 4.

The FLB cohort comprised a subset of 3,572,737 mothers in the delivery cohort with a first recorded live birth from 1^st April 2007 (Figure 2-e). There were a further 2,303,476 deliveries (total 5,876,213) for these mothers up to end-2021. In the FLB cohort, 1.4% of first time and 1.3% of subsequent deliveries included more than 1 infant and 0.3% of subsequent deliveries included at least 1 still born infant.

Evaluation of linkage quality

The Linkage Bridging File transferred from NHSE to UCL for analysis (Figure 1 Step E, Figure 2-iii) contained 100,125 unique Cafcass-HES ID combinations, with 88.5% of Cafcass records (99,971/112,927) matching to at least one HES record (99,340 HESIDs). Where mothers linked to 2 or more HES IDs, the most common link was accepted (≤3 possible), otherwise the matches were discarded (77 combinations accepted, 227 rejected).

Amongst the 99,898 accepted matches (Figure 2-iv), where 2 or more mothers linked to the same HES ID (n = 1,708) the care proceedings records were merged into a single ID (n = 929), thus removing duplicates missed prior to linkage and resulting in 99,118 mothers who matched to HES, and a final HES match rate of 88.4% (99,118/112,147).

Figure 2 shows that 5,743 mothers recorded in Cafcass were linked to a HES record but not to a delivery record and were excluded. A further 455 mothers were removed from the delivery cohort as their first care proceeding started more than 39 weeks before their first liveborn delivery recorded in HES, and 29 due to date of death recorded as before to case start. This left 92,891 (82.8% of 112,147) mothers with care proceedings linked to a delivery record between 1997 and 2021 (Figure 2-vi). Of these, 38,462 mothers had a first live birth between 1^st April 2007 and 31^st December 2021 (34.3% of 112,147) and were followed up in the FLB cohort of 3,572,737 mothers (Figure 2-vii).

Characteristics of mothers in Cafcass according to linkage status

Of the 112,147 mothers recorded in Cafcass with identifiers for linkage, 13,029 (11.6%) did not link to any HES record, 6,227 (5.6%) linked to a HES record but the final Delivery Cohort, and 92,891 (82.8%) linked to the Delivery Cohort (Table 1 and Figure 2). Mothers from an ethnic background classified as black or other, aged over 40 years at first case, with a first case between 2007–2010 and whose first case was in London or the East of England, were least likely to link to the Delivery Cohort.

Table 1: Characteristics of mothers recorded in Cafcass care proceedings between 1st April 2007 and 31st December 2021 in England according to linkage status: linked to a UCL-HES extract record and linked to a delivery cohort record. N.B. Characteristics are based on information as recorded within Cafcass, and thus may differ to characteristics recorded in HES and presented in Table 2. 1-Of those available for linkage (max 112,147), calculated using row total as a denominator. 2-MR = Match rate. 3-due to no information on date of birth or if outside age range (15-50 years) included for linkage. 4-As recorded in Cafcass. 5-Rounded to nearest 5 in accordance with NHS England disclosure control methodology for HES. 6-Yorkshire and the Humber.

Characteristics recorded in Cafcass	Mothers	% 1	Linked to UCL HES				Linked to Delivery Cohort
			n	%	MR 1,2		n	%	MR 1,2
Total	120,157		9,118				92,891
Not available for linkage3	8,010
Available for linkage	112,147	100	99,118	100	88.4		92,891	100	82.8
Ethnicity4
White	70,470	62.8	63,385	63.9	89.9		60,251	64.9	85.5
Black	4,487	4.0	3,573	3.6	79.6		3,127	3.4	69.7
Asian	3,116	2.8	2,603	2.6	83.5		2,340	2.5	75.1
Mixed	3,067	2.7	2,642	2.7	86.1		2,470	2.7	80.5
Other	1,000	0.9	797	0.8	79.7		647	0.7	64.7
Missing	30,007	25.0	26,118	26.4	87.0		24,056	25.9	80.2
Number of children4
1	48,091	42.9	41,863	42.2	87.0		38,242	41.2	79.5
2	29,446	26.3	26,076	26.3	88.6		24,706	26.6	83.9
3	17,769	15.8	15,876	16.0	89.3		15,264	16.4	85.9
4+	16,841	15.0	15,303	15.4	90.9		14,679	15.8	87.2
Age at first s31 case
15–19	11,902	10.6	10,539	10.6	88.5		9,978	10.7	83.8
20–24	22,858	20.4	20,284	20.5	88.7		19,551	21.0	85.5
25–29	22,659	20.2	20,112	20.3	88.8		19,341	20.8	85.4
30–34	22,169	19.8	19,680	19.9	88.8		18,635	20.1	84.1
35–39	17,149	15.3	15,145	15.3	88.3		13,931	15.0	81.2
40–44	10,093	9.0	8,822	8.9	87.4		7,703	8.3	76.3
45–50	5,317	4.7	4,536	4.6	85.3		3,752	4.0	70.6
Year of first s31 case
2007–2010	20,968	18.7	18,574	18.7	88.6		16,958	18.3	80.9
2011–2014	29,837	26.6	26,293	26.5	88.1		24,638	26.5	82.6
2015–2018	36,976	33.0	32,526	32.8	88.0		30,710	33.1	83.1
2019–2022	24,366	21.7	21,725	21.9	89.2		20,585	22.2	84.5
Region of first case5
North East	9,225	8.2	8,385	8.5	90.9		8,015	8.6	86.9
North West	18,865	16.8	16,675	16.8	88.4		15,755	17.0	83.5
York. & Humber6	13,085	11.7	11,915	12.0	91.1		11,250	12.1	86.0
East Midlands	8,875	7.9	7,985	8.1	90.0		7,535	8.1	84.9
West Midlands	12,050	10.7	10,730	10.8	89.0		10,110	10.9	83.9
East of England	9,635	8.6	8,545	8.6	88.7		7,970	8.6	82.7
London	15,435	13.8	12,785	12.9	82.8		11,455	12.3	74.2
South East	14,700	13.1	13,055	13.2	88.8		12,240	13.2	83.3
South West	9,775	8.7	8,610	8.7	88.1		8,155	8.8	83.4
Not available	505	0.5	430	0.4	85.1		405	0.4	80.2
Number of s31 cases
1	85,606	76.3	75,119	75.8	87.7		69,885	75.2	81.6
2	19,328	17.2	17,459	17.6	90.3		16,714	18.0	86.5
3+	7,213	6.4	6,540	6.6	90.7		6,292	6.8	87.2

Additionally, an adjusted logistic regression analysis to summarise associations between characteristics of mothers recorded in Cafcass and linkage to the Delivery Cohort showed mothers with a postcode recorded (92.0%; 8,939/112,147) were 1.77 times (95% Confidence Interval [CI]: 1.69-1.87) more likely than mothers without a postcode in Cafcass to match to a HES delivery record (Supplementary Material Section 2.2).

Disagreement between common variables in HES and Cafcass

Missingness was low (≤0.5%) for all variables other than ethnicity (where 22.6% of match combinations had ethnicity missing from either data source). Of the 38,462 mothers recorded in Cafcass linked to the FLB Cohort, 74.6% disagreed on no variables (Supplementary Table 8a). Disagreement was highest for the number of children at case start (17.2%, n = 6,608), followed by age at first child (10.7%, n = 4,126), ethnicity (4.2%, n = 1,634), region (2.7%, n = 1,051) and maternal year of birth (1.0%, n = 383) and Supplementary Table 8b shows the 5 most common patterns of disagreement.

Cumulative Incidence of care proceedings

The FLB cohort contained 3,572,737 mothers, with 5,876,213 deliveries, and 1.1% of mothers (38,462/3,572,737) had a Cafcass care proceeding between 1^st April 2007 and 31^st December 2021 (Table 2). Overall, the majority of deliveries were singletons (98.6%; n = 5,794,598), with 1.4% (n = 81,615) recorded as a multiple, and 0.15% (n = 8,701) were recorded as having at least 1 still born infant and in <0.03% (n = 1,372) the outcome was uncertain. 6,655 (0.2%) of mothers died within 10 years of their first live birth.

Compared to mothers not in care proceedings, mothers subject to care proceedings were younger at first delivery (78.4% vs 28.8% under 25 years) and more deprived (72.0% vs 47.0% residing in the two most deprived quintiles).

The cumulative incidence of a care proceeding by 10 years after a first live birth was 1.31% (95% CI: 1.29–1.32%) (Table 2; Figure 3). Cumulative incidence was highest in mothers who were aged 15–19 years (6.79%, 95% CI: 6.69–6.89%) or 20–24 years (1.89%, 95% CI: 1.85–1.92%) at first delivery, most deprived at first delivery (2.47%, 95% CI: 2.49–2.51%) and of mixed ethnicity (1.93%, 95% CI: 1.80–2.07%). Sensitivity analysis (Supplementary Material Section 6), using data from unlinked mothers, increased estimated cumulative incidence up to 1.7%.

Table 2:Characteristics of mothers, aged 15-50 years, who had a live birth in an NHS service, and were resident in England between 1st April 2007 and 31st December 2021 (the First Live Birth Cohort), number and proportion with care proceedings and the cumulative incidence, and 95% Confidence Interval (CI), of care proceedings within 10 years of a first live birth. N.B. Characteristics are based on information as recorded within HES and the delivery cohort, and thus may differ to characteristics recorded in Cafcass and presented in Table 1. 1-based on first delivery between 1^st April 2007 and 31^stDecember 2021. 2-Rounded to nearest 5 in accordance with NHS England disclosure control methodology for HES. 3-Yorkshire and the Humber. 4-from 1^st April 2007.

	Mothers			Not experienced care proceedings			Experienced care proceedings
	n	%		n	% of total		n	% of total	10-year cumulative incidence	95% Confidence interval
Total	3,572,737	100		3,534,275	100.0		38,462	100.0	1.31	1.29–1.32
Age at first delivery
<20	292,316	8.2		274,201	7.8		18,115	47.1	6.79	6.69–6.89
20–24	754,819	21.1		742,766	21.0		12,053	31.3	1.89	1.85–1.92
25–29	1,044,397	29.2		1,039,939	29.4		4,458	11.6	0.53	0.51–0.54
30–34	978,389	27.4		976,077	27.6		2,312	6.0	0.30	0.28–0.31
35–39	412,709	11.6		411,505	11.6		1,204	3.1	0.36	0.34–0.38
40–44	84,146	2.4		83,836	2.4		310	0.8
45–50	5,961	0.2		5,951	0.2		10	0.0
Ethnicity
White	2,621,788	73.4		2,588,348	73.2		33,440	86.9	1.51	1.49–1.53
Black	137,028	3.8		135,554	3.8		1,474	3.8	1.25	1.18–1.32
Asian	349,866	9.8		348,688	9.9		1,178	3.1	0.40	0.38–0.43
Mixed	57,181	1.6		56,275	1.6		906	2.4	1.93	1.80–2.07
Other	121,865	3.4		121,199	3.4		666	1.7	0.67	0.61–0.72
Not reported	285,009	8.0		284,211	8.0		798	2.1	0.43	0.40–0.46
IMD Quintile1
1-most deprived	872,836	24.4		854,949	24.2		17,887	46.5	2.47	2.43–2.51
2	812,742	22.7		803,046	22.7		9,696	25.2	1.43	1.40–1.46
3	701,728	19.6		696,311	19.7		5,417	14.1	0.95	0.92–0.98
4	611,178	17.1		607,919	17.2		3,259	8.5	0.65	0.63–0.67
5-least deprived	566,446	15.9		564,380	16.0		2,066	5.4	0.45	0.42–0.47
Not reported	7,807	0.2		7,670	0.2		137	0.4	1.57	1.32–1.88
Region1,2
North East	154,310	4.3		150,936	4.3		3,374	8.8	2.62	2.52–2.71
North West	439,805	12.3		433,483	12.3		6,324	16.4	1.71	1.67–1.76
York. & Humber3	324,220	9.1		319,604	9.0		4,614	12.0	1.63	1.58–1.68
East Midlands	250,685	7.0		247,693	7.0		2,994	7.8	1.38	1.33–1.44
West Midlands	342,505	9.6		338,421	9.6		4,085	10.6	1.41	1.37–1.46
East of England	363,315	10.2		359,996	10.2		3,318	8.6	1.09	1.05–1.13
London	669,390	18.7		665,132	18.8		4,258	11.1	0.75	0.72–0.77
South East	530,995	14.9		525,953	14.9		5,041	13.1	1.11	1.08–1.15
South West	318,155	8.9		314,467	8.9		3,689	9.6	1.35	1.30–1.40
No Fixed Abode	565	0.0		536	0.0		30	0.1	5.60	3.94–7.95
Not reported	178,790	5.0		178,054	5.0		735	1.9	1.36	1.16–1.60
Year of first delivery
20074–2010	1,010,671	28.3		996,179	28.2		14,492	37.7	1.20	1.18–1.22
2011–2014	993,652	27.8		981,664	27.8		11,988	31.2	1.28	1.26–1.31
2015–2018	928,000	26.0		919,436	26.0		8,564	22.3
2019–2022	640,414	17.9		636,996	18.0		3,418	8.9
Multiples1
Singleton	3,521,210	98.6		3,483,141	98.6		38,069	99.0	1.32	1.30–1.33
Multiples	51,527	1.4		51,134	1.4		393	1.0	0.88	0.79–0.97
Birth Status1
All live born	3,571,397	100.0		3,532,946	100.0		38,451	100.0
Some live born, some still born	1,146	0.0		1,137	0.0		c.	0.0
Uncertain	194	0.0		192	0.0		c.	0.0

Figure 3: Kaplan-Meier estimates of cumulative incidence of care proceedings at 5 and 10 years after a first live birth, with 95% confidence intervals, between 1st April 2007 and 31st December 2021.

Discussion

We linked 88.4% of mothers recorded in Cafcass care proceedings to a HES admission, and 82.4% to a HES delivery record. Mothers residing in London, East of England, older age and reported as black or other ethnic background were least likely to link to a HES delivery record. Amongst mothers with care proceedings who linked to the first live birth cohort, 74% had Cafcass and HES records agree on five compared variables, but disagreement on number or age of children was present for 26%. The cumulative incidence of Cafcass care proceedings within 10 years of a first live birth was 1.3% for all mothers, and higher for mothers aged less than 20 years at their first live birth (6.9%), most deprived (2.50%) and of mixed ethnicity (1.96%).

Results in context

For the first time, national court and hospital records for England were linked and our match rate is comparable to that reported for Welsh data [22]. Johnson et al. linked Welsh family court and electronic healthcare records, including general practice records, via the Secure Anonymised Information Linkage Databank [SAIL] for all people associated with cases, and achieved a match rate of 74% [22]. In contrast, in Sweden and Canada it is possible to link across multiple databases deterministically using unique national identification numbers [9, 23]. We found match rates vary by region, maternal age and ethnicity, but similar figures are not available in Welsh-linked data. These differences will impact cumulative incidence estimates. A key driver of differences in match rates is the variability in availability of date of birth and postcode, both of which were lowest in London and the East of England. This may be because London is more ethnically diverse than other regions, a higher proportion of mothers in London may have delivered outside of England, thus not contained in the delivery cohort, or the population of London is more mobile, meaning postcode data in Cafcass may not be up to date. The higher match rate to HES via PDS of 88% highlights the potential for public law family court care proceedings data to be linked to a range of NHSE owned healthcare datasets. Of particular interest are the mental health, maternity and emergency care data services. As data linkage of Cafcass to administrative datasets continue, the data quality, particularly of linkage identifiers remains important and may require improvements.

We found deprived, younger mothers were more likely to experience care proceedings within the first 10 years of parenthood, corresponding with previous, descriptive analyses from England and Wales care proceedings data [24]. To date, English estimates have used unlinked data or Children Looked After returns. Using these data, it has been estimated that between 1.2% to 1.8% of children enter out-of-home care by 10 years, increasing to 3.3% by 18 years old [25–27]. Our estimate of 1.3% of mothers experiencing care proceedings within 10 years of a FLB, and an upper estimate of 1.7% under sensitivity scenario 1, aligns with these previous estimates. Whilst strategies to reduce teenage pregnancy rates have resulted in declining rates, teenage mothers remain at greatest risk of care proceedings [28]. Young mothers require early, sustained, co-ordinated, multiagency support to improve the outcomes of both mother and child(ren) [29]. Following assessment, support may include help to continue with education, return to work, or to develop adequate networks and avoid social isolation, as well as addressing health needs [30, 31]. However, investment in early intervention services, such as Sure Start children’s centres, has fallen by 46% since 2010–11 [32]. Our results show the utility of data linkage of health and court data, and in the future, exploration of up-stream health profiles will provide further insights into predictors of care proceedings and direction of preventive measures.

Strengths and limitations

Strengths include use of whole of England data, with up to 24 years of follow up. This allowed us to derive a FLB cohort so that all mothers can be followed from the same point in their reproductive life course. A limitation is that we could not exclude left censoring, by which we mean births that occurred outside the NHS in England, prior to 1^st April 1997, or were not linked to the woman’s records from 2007 onwards. As longer look-back periods become available and the data quality of the maternity tail improves, the impact of this censoring will be reduced. Knight et al. (2013), referenced unpublished data stating that 90% of mothers have their second infant within 7 years of the first, which is similar to the 86.5-90.2% we estimate based on early deliveries in our cohort [33]. This means that from 2005, less than 10% of mothers will be wrongly recorded as primiparous from cohort processing alone, a figure further reduced as 66.1% of mothers records had data on the number of previous registrable pregnancies. In 2021, 29% of live births were to non-UK born women, some of whom will be highly mobile [34]. The Delivery Cohort is also influenced by right censoring, particulary if women leave England.

The drivers of differences in match rates are discussed earlier, but the proportion of mothers who should have matched to the FLB cohort is difficult to ascertain through Cafcass data alone. We estimate, through sensitivity scenario 1, cumulative incidence could be as high as 1.7%. However, we have assumed all children associated with these cases were born in NHS maternity units in England and are not aware of data to verify this. Additionally, the records of mothers without date of birth are more likely to be unmatched, duplicate records.

Due to the data linkage process of NHE, we were unable to review match quality with identifiable data or know whether unlinked mothers did not match to any PDS records or matched to multiple PDS records with the same match quality, and so not provided in the linkage bridging file. Where mothers did match and FLB cohort and Cafcass data was compared, <0.4% of records disagreed on 4 or more variables. The most common disagreement was for age at first child and number of children, which may occur if care proceedings do not include all children and/or some children do not live in the family home (ie live with another family member).

Conclusion

We have demonstrated successful linkage of court, health and delivery records in England, although improved identifier quality would reduce linkage biases for vulnerable groups and between regions. Using the resulting data, we estimate one in 15 mothers younger than 20 years at their first birth were involved in care proceedings within 10 years. In the future, this data will provide opportunities to better understand the health needs of mothers prior to, and following, care proceedings. The deployment of appropriate healthcare services could reduce the need for children to be removed from the family home and improve the health and wellbeing of mothers before and following child removal.

Acknowledgments

We are grateful to the Children and Family Court Advisory and Support Service (Cafcass), for providing extracts of their case management data to establish this linkage, and the patients, their families, and NHS staff for their ongoing contribution to research. This work uses data provided by patients and collected by the National Health Service as part of their care and support. Permission to use de-identified data from Hospital Episode Statistics was granted by NHS Digital (DARS-NIC-196263).

This project was funded by the Nuffield Foundation (grant number: JUS/FR-000020122). RG, LW and MJ were (in part) supported by the National Institute for Health and Care Research (NIHR) Children and Families Policy Research Unit (PR-PRU-1217-21301); RG by a NIHR Senior Investigator award and by Health Data Research UK (HDRUK2023.0029); GI, MJ, QF, KH, LW and RG by the NIHR GOSH Biomedical Research Centre, and KH, RG and QF by ADR UK (Administrative Data Research UK), an Economic and Social Research Council (part of UK Research and Innovation) programme (ES/V000977/1, ES/X000427/1 and ES/X003663/1). CG is supported by the Economic and Social Research Council (ESRC) UBEL Doctoral Training Programme (ES/P000592/1).

The views expressed are those of the author(s) and not necessarily those of the Nuffield Foundation, NIHR, the Department of Health and Social Care, ADR UK or ESRC.

Data availability

Hospital Episode Statistics data and was provided within the terms of a data-sharing agreement (DARS-NIC-196263-J9Q7Z-v1.4) to the researchers by NHS England. The data do not belong to the authors and may not be shared by the authors, except in aggregate form for publication. The data is provided by patients and collected by the NHS as part of their care and support. Data can be obtained by submitting a data request through the NHS England Data Access Request Service and Cafcass.

Statement on conflicts of interest

Nothing to declare

Ethics statement

Ethical approval was obtained from the UCL Research Ethics Committee (reference: 19/LO/0103), and the Cafcass Research Governance Committee to work with the Cafcass data. The Health Research Authority Confidentiality Advisory Group granted exemption from s251 to enable linkage between identifiers supplied by Cafcass to NHS identifiers by NHS England. Data sharing agreements are in place to use de-identified attribute data from Cafcass and NHS England (DARS-NIC-196263-J9Q7Z-v1.4) in the UCL Data Safe Haven (privacy notice).

Funding

This work was supported by the Nuffield Foundation [grant number JUS/FR-000020122]. The views expressed are those of the author(s) and not necessarily those of Cafcass or the Nuffield Foundation. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Abbreviations

Cafcass	Children and Family Court Advisory and Support Service
CI	Confidence Interval
CP	care proceedings
FLB	First live birth
HES	Hospital Episode Statistics
HES APC	Hospital Episode Statistics Admitted Patient Care
ID	Identifiers
IMD	Index of multiple deprivation
LA	Local Authority
MT	Maternity Tail
NHSE	NHS England
ONS	Office for National Statistics
PDS	Personal Demographic Service
UCL	University College London
UCL DSH	UCL Data Safe Haven

References

1. Broadhurst K, Alrouh B, Yeend E, Harwin J, Shaw M, Pilling M, et al. Connecting Events in Time to Identify a Hidden Population: Birth Mothers and Their Children in Recurrent Care Proceedings in England. Brit J Soc Work. 2015;45(8):2241–60. 10.1093/bjsw/bcv130

   10.1093/bjsw/bcv130

2. Cafcass. Cafcass Official Demand Statistics. 2023. https://www.cafcass.gov.uk/about-us/our-data.

3. Wall-Wieler E, Roos LL, Brownell M, Nickel NC, Chateau D. Predictors of having a first child taken into care at birth: A population-based retrospective cohort study. Child Abuse Neglect. 2018;76:1–9. 10.1016/j.chiabu.2017.09.033

   10.1016/j.chiabu.2017.09.033

4. Griffiths LJ, Johnson RD, Broadhurst K, Bedston S, Cusworth L, Alrouh B, et al. Maternal health, pregnancy and birth outcomes for women involved in care proceedings in Wales: a linked data study. Bmc Pregnancy Childb. 2020;20(1). 10.1186/s12884-020-03370-4

   10.1186/s12884-020-03370-4

5. Johnson RD, North L, Alrouh B, John A, Jones K, Akbari A, et al. A population level study into health vulnerabilities of mothers and fathers involved in public law care proceedings in Wales, UK between 2011 and 2019. International Journal of Population Data Science (Ijpds). 2022;7(1). 10.23889/ijpds.v7i1.1723

   10.23889/ijpds.v7i1.1723

6. Johnson R, Alrouh B, Broadhurst K, Ford D, John A, Jones K, et al. Health vulnerabilities of parents in care proceedings in Wales. London: Nuffield Family Justice Observatory; 2021. https://www.nuffieldfjo.org.uk/resource/health-vulnerabilities-parents-care-proceedings.

7. Johnson R, Griffiths LJ, Cowley LE, Broadhurst K, Bailey R. Risk Factors Associated With Primary Care-Reported Domestic Violence for Women Involved in Family Law Care Proceedings: Data Linkage Observational Study. J Med Internet Res. 2023;25. 10.2196/42375

   10.2196/42375

8. Broadhurst K, Mason C, Bedston S, Alrouh B, Morriss L, McQuarrie T, et al. Vulnerable Birth Mothers and Recurrent Care Proceeding. Lancaster University; 2017. https://www.nuffieldfoundation.org/sites/default/files/files/rc-final-summary-report-v1_a6.pdf.

9. Wall-Wieler E, Vinnerljung B, Liu C, Roos LL, Hjern A. Avoidable mortality among parents whose children were placed in care in Sweden: a population-based study. J Epidemiol Commun H. 2018;72(12):1091–8. 10.1136/jech-2018-210672

   10.1136/jech-2018-210672

10. Wall-Wieler E, Roos LL, Nickel NC, Chateau D, Brownell M. Mortality Among Mothers Whose Children Were Taken Into Care by Child Protection Services: A Discordant Sibling Analysis. Am J Epidemiol. 2018;187(6):1182-8. 10.1093/aje/kwy062

    10.1093/aje/kwy062

11. Wall-Wieler E, Roos LL, Brownell M, Nickel N, Chateau D, Singal D. Suicide Attempts and Completions among Mothers Whose Children Were Taken into Care by Child Protection Services: A Cohort Study Using Linkable Administrative Data. Can J Psychiat. 2018;63(3):170–7. 10.1177/0706743717741058

    10.1177/0706743717741058

12. Pearson RJ, Grant C, Wijlaars L, Finch E, Bedston S, Broadhurst K, Gilbert R. Mental health service use among mothers involved in public family law proceedings: linked data cohort study in South London 2007- 2019. Social Psychiatry and Psychiatric Epidemiology. 2022;57(10):2097–108. 10.1007/s00127-022-02221-1

    10.1007/s00127-022-02221-1

13. Pearson RJ. Using longitudinal administrative data to characterise mental health problems and substance misuse among women whose children enter care in England: University College London; 2021. https://discovery.ucl.ac.uk/id/eprint/10139871/.

14. Bedston SJ, Pearson RJ, Jay MA, Broadhurst K, Gilbert R, Wijlaars L. Data Resource: Children and Family Court Advisory and Support Service (Cafcass) public family law administrative records in England. International Journal of Population Data Science. 2020;5(1). 10.23889/ijpds.v5i1.1159

    10.23889/ijpds.v5i1.1159

15. Herbert A, Wijlaars L, Zylbersztejn A, Cromwell D, Hardelid P. Data Resource Profile: Hospital Episode Statistics Admitted Patient Care (HES APC). Int J Epidemiol. 2017;46(4):1093-i. 10.1093/ije/dyx015

    10.1093/ije/dyx015

16. Gilbert R, Lafferty R, Hagger-Johnson G, Harron K, Zhang LC, Smith P, et al. GUILD: GUidance for Information about Linking Data sets. J Public Health-Uk. 2018;40(1):191–8. 10.1093/ pubmed/fdx037

    10.1093/ pubmed/fdx037

17. NHS England. Personal Demographics Service 2023 [Available from: https://digital.nhs.uk/services/personal-demographics-service.

18. NHS Digital. The Person_ID handbook for HES users. 2021. https://digital.nhs.uk/services/personal-demographics-service/master-person-service#user-guidance.

19. Harron K, Gilbert R, Cromwell D, van der Meulen J. Linking Data for Mothers and Babies in De- Identified Electronic Health Data. PLoS One. 2016;11(10):e0164667. 10.1371/journal.pone.0164667

    10.1371/journal.pone.0164667

20. Office for National Statistics. Number of maternities by Index of Multiple Deprivation (IMD), age of mother, type of outcome and previous live born children 2023 [Available from: https://www.ons.gov.uk/peoplepopulationandcommunity/birthsdeathsandmarriages/maternities/adhocs/1637num berofmaternitiesbyindexofmultipledeprivationimdageof mothertypeofoutcomeandpreviouslivebornchildren.

21. Office for National Statistics. English indices of deprivation 2019 [Available from: https://www.gov.uk/government/collections/english-indices-of-deprivation.

22. Johnson RD, Ford DV, Broadhurst K, Cusworth L, Jones KH, Akbari A, et al. Data Resource: population level family justice administrative data with opportunities for data linkage. Int J Popul Data Sci. 2020;5(1):1339. 10.23889/ijpds.v5i1.1339

    10.23889/ijpds.v5i1.1339

23. Wall-Wieler E, Roos LL, Bolton J, Brownell M, Nickel N, Chateau D. Maternal Mental Health after Custody Loss and Death of a Child: A Retrospective Cohort Study Using Linkable Administrative Data. Can J Psychiat. 2018;63(5):322–8. 10.1177/0706743717738494

    10.1177/0706743717738494

24. Griffiths LJ, Johnson RD, Broadhurst K, John A. Born into care: One thousand mothers in care proceedings in Wales. London: Nuffield Family Justice Observatory; 2021. https://www.nuffieldfjo.org.uk/resource/1000-mothers-care-proceedings-wales.

25. Jay M, Mc Grath-Lone L, Pearson R, Gilbert R. The cumulative proportion of children receiving social care services in England: a whole population administrative data cohort study. International Journal of Population Data Science. 2022;7(3). 10.23889/ijpds.v7i3.1895

    10.23889/ijpds.v7i3.1895

26. Mc Grath-Lone L, Dearden L, Nasim B, Harron K, Gilbert R. Changes in first entry to out-of-home care from 1992 to 2012 among children in England. Child Abuse Negl. 2016;51:163–71. 10.1016/j.chiabu.2015.10.020

    10.1016/j.chiabu.2015.10.020

27. Ubbesen MB, Gilbert R, Thoburn J. Cumulative incidence of entry into out-of-home care: changes over time in Denmark and England. Child Abuse Negl. 2015;42:63–71. 10.1016/j.chiabu.2014.10.006

    10.1016/j.chiabu.2014.10.006

28. Office for National Statistics. Conceptions in England and Wales: 2020. 2022. https://www.ons.gov.uk/peoplepopulationandcommunity/birthsdeathsandmarriages/conceptionandfertilityrates/bulletins/conceptionstatistics/2020#teenage-conceptions.

29. Public Health England. Teenage mothers and young fathers: support framework. 2016. https://www.gov.uk/government/publications/teenage-mothers-and-young-fathers-support-framework.

30. Public Health England. A framework for supporting teenage mothers and young fathers 2019. https://assets.publishing.service.gov.uk/media/5cb85bc640f0b649e47f2983/PHE_Young_Parents_Support_Framework_April2019.pdf.

31. Barnardo’s. Not the end of the story: Supporting teenage mothers back into education. 2010. https://www.barnardos.org.uk/sites/default/files/2020-11/not_the_end_of_the_story_-_march_2010_pdf.pdf.

32. Franklin J, Larkham J, Mansoor M. The well-worn path. 2023. https://www.child renssociety.org.uk/information/professionals/resources/the-well-worn-path#facts-about-childrens-services-funding.

33. Knight HE, Gurol-Urganci I, van der Meulen JH, Mahmood TA, Richmond DH, Dougall A, Cromwell DA. Vaginal birth after caesarean section: a cohort study investigating factors associated with its uptake and success. BJOG. 2014;121(2):183–92. 10.1111/1471-0528.12508

    10.1111/1471-0528.12508

34. Office for National Statistics. Births by parents’ country of birth, England and Wales: 2021. 2022. https://www.ons.gov.uk/peoplepopulationandcommunity/birthsdeathsandmarriages/livebirths/bulletins/parentscountryofbirthenglandandwales/2021#:..:text=Main%20points,decrease%20from%2029.3%25%20in%202020.