Results of the First Steps study: a randomised controlled trial and economic evaluation of the Group Family Nurse Partnership (gFNP) programme compared with usual care in improving outcomes for high-risk mothers and their children and preventing abuse

Policy context

A range of cross-party policy documents has now explicitly highlighted the importance of promoting children’s well-being during pregnancy and first 2 years of life,12–14 and recent key documents include Conception to Age 2: The Age of Opportunity15 and The 1001 Critical Days: The Importance of the Conception to Age Two Period. 16

Fair Society, Healthy Lives. Independent Review Into Health Inequalities in England Post 201017 focused on the importance of pregnancy and the first 2 years of life in terms of equalising the life chances of children, and Healthy Lives, Healthy People18 similarly points to the importance of ‘starting well’, focusing in particular on the health of mothers during pregnancy and parenting during the early years. Recent research has identified that this period is key because of the ‘biological embedding of social adversity’ that takes place during sensitive developmental periods. 19,20 This research showed that toxic stress caused by high levels of anxiety and depression during sensitive developmental periods (e.g. pregnancy and the postnatal period) can disrupt the developing brain architecture and other organ systems and regulatory functions, impacting the fetal/infant physiology in terms of hyper-responsive/chronically activated stress response; their resulting behavioural adaption; and the long-term cognitive, linguistic and socioemotional development. The long-term impact occurred in terms of increased stress-related chronic disease, unhealthy lifestyles and widening health disparities.

Evidence context

There is limited evidence available about ‘what works’ to support vulnerable parents during pregnancy and infancy. Although evidence concerning the effectiveness of home-visiting programmes in general in reducing child maltreatment is inconclusive,21 the US-developed Nurse Family Partnership (NFP) was one of nine home-visiting programmes identified as effective by the US Department of Health and Human Services as part of their Home Visiting Evidence of Effectiveness Review. 22 It is commonly named when examples of programmes with high-quality evidence for success are sought. For instance, the US coalition for evidence-based policy, responding to a Congressional directive that funds be directed to programmes with top-tier evidence of effectiveness identified only two programmes for children aged 0–6 years and their families that could be thus categorised, one of which was the NFP. 23 The Blueprints mission of the ‘Center for the Study and Prevention of Violence’ was charged with identifying outstanding violence and drug prevention programmes that meet a high scientific standard of effectiveness and, out of 800 with published research, found 12, one of which was NFP. 24 A similar conclusion was reached by academics seeking evidence-based home-visiting programmes likely to reduce child abuse and neglect. 25 The NFP was found to be effective in both decreasing child maltreatment and improving parenting practices. 22 Long-term follow-up of the NFP in the USA suggests a 48% reduction in cases of child abuse and neglect by the age of 15 years. 26

The NFP curriculum has strong theoretical underpinnings, in terms of both risk and protective factors, and the mechanisms through which change may be produced,27 drawing on ecological,28 self-efficacy29 and attachment30 theories. Ecological theory emphasises the importance of interactions between the characteristics of individuals and their contexts; self-efficacy theory focuses on an individual’s beliefs that they can successfully carry out behaviour required for good outcomes; and attachment theory highlights the importance of the early interactions with the primary caregiver in terms of the child’s later capacity for affect regulation. The cornerstone of the NFP model is the therapeutic nurse–client relationship. Beneficial outcomes found in the US trials included improved prenatal health, fewer childhood injuries, fewer subsequent pregnancies, increased intervals between births, increased maternal employment and improved school readiness;23,26,31–33 it has also been shown to have the potential to be cost-effective. 34 Results from the US trials of NFP found that it was particularly beneficial for women with ‘low psychological resources’, namely a combination of lower intelligence, mental health problems and low self-efficacy. 35

The NFP programme was introduced into England in 2007, renamed the Family Nurse Partnership (FNP),36 and has been offered to first-time teen mothers in more than 70 locations in England, Scotland and Northern Ireland,37 although recent RCT evidence has failed to support it as a way to reduce child abuse and neglect in the UK. 38 An implementation evaluation in the first 10 areas to provide FNP found that the programme was perceived in a positive light by potential clients and the nurses responsible for its delivery and uptake was high, with delivery close to the stated US objectives. 39 Nevertheless, potential sustainability issues were identified and in particular local concerns about its cost set against long-term rather than immediate gains. 40,41 Issues of eligibility were also examined, with the conclusion that, over time, the criteria might have to be changed to include additional risk factors beyond young age, although this could cause difficulties in identifying women early in their pregnancy. 42

In addition to being trained according to the US requirements, UK nurses are trained in ‘motivational interviewing’43 so that they can develop in-depth engagement with families to achieve change. As is the case in the USA, fathers are encouraged to be present for home visits and they have reported positively about the programme, in particular that the nurses invested time in developing relationships with them and identified their strengths in addition to areas that needed support, and that the programme was holistic in its approach. 44

Participants

The participants were young (aged < 25 years) pregnant women.

Eligibility criteria

The requirement of the UK FNP NU was that gFNP should be offered to women who were not eligible for FNP, but who would be likely to benefit from the content of programme, based on research in the USA. 23,26 Women eligible for the trial, based on criteria defined by the FNP NU, were expectant mothers with expected delivery dates (EDD) within approximately 10 weeks of each other, for each group in each site. The range of EDDs was specified in relation to the expected date of the first meeting per site so that the majority would be 16–20 weeks pregnant when programme delivery commenced in that site. Specific criteria, beyond similar EDDs and gestation, were that participants should be either:

• aged < 20 years at their last menstrual period (LMP) with one or more previous live births; or

• aged 20–24 years at their LMP with no previous live births and low educational qualifications [defined as not having both mathematics and English language General Certificate of Secondary Education (GCSE) at grade C or higher or, if they had both, no more than four GCSEs at grade C or higher].

Exclusions were:

• expectant mothers aged < 20 years who had previously received home-based FNP

• mothers in either age group with psychotic mental illness (defined as bipolar disorder or schizophrenia)

• mothers who were not able to communicate orally in English.

Study setting

Family Nurse Partnership teams are located around England, but with various dates of starting ranging from 2007 to the time that the study was being planned (October 2012). FNP teams were eligible to be part of the trial if:

• the team had delivered the home-based FNP programme in its entirety (from birth to age 24 months) to a cohort of women

• the team included at least one family nurse (FN) practitioner who had notified their intention to practise as a midwife.

Invitations were sent by the FNP NU to eligible teams, noting that they could take part if, in addition:

• they could demonstrate from birth records that sufficient women of the relevant age and parity in the local area had given birth in the previous year for recruitment of two groups of 16–20 women (8–10 intervention, 8–10 control), each recruited within approximately 6 weeks, assuming that at least three times that number would need to be identified to gain sufficient agreement

• they could confirm good links with community midwifery such that they also signed the expression of interest.

Seventeen teams expressed initial interest and eight sent in formal expressions of interest. Following site visits to discuss the likelihood of sufficient birth data and good midwifery collaboration, seven teams agreed to take part in the trial, located across England in Barnsley, Dewsbury, Lewisham (London), Nottingham, Sandwell (Birmingham), South Tyne and Wear and Waltham Forest (London). The eighth site found that their birth rate would not support the numbers needed for the trial.

The selection of FNs within sites to be involved in the trial was the responsibility of FNP teams. FNs, all of whom had substantial experience of delivering FNP, in general volunteered, and the majority had previous experience of running other types of group. At least one FN at each site had to have an intention to treat (ITT) as a midwife. The FNs received several days of training specific to delivering gFNP, which focused on group dynamics and the different aspects of the curriculum designed to generate interactions between group members. The training, from FNs who had developed the programme materials and been involved in feasibility research, covered topics such as using communication and motivational interviewing skills within a group context. 70 Although in theory FNs could have withdrawn from involvement, any FN withdrawing during the study did so owing to illness. Most sites were not able to send to training more than the two FNs needed for the programme. In cases of short-term absence, the supervisor or another FN from the team usually deputised.

Study intervention

Group Family Nurse Partnership is designed to run from the first trimester of pregnancy until infants are aged 12 months, with 44 group meetings in the curriculum (14 covering pregnancy and 30 covering infancy). 57 It was delivered to a group of women living in relatively close proximity, with similar EDDs (within 8–10 weeks). 46 Meetings lasted around 2 hours and were held in children’s centres, health centres or other suitable community facilities in the local areas served by the FNP teams. Sessions were facilitated by two experienced FNP FNs, one of whom had notified their intention to practise as a midwife. The two FNs exchanged the roles of active leader (facilitating a topic and activity) and active observer, noticing behaviours and body language of members and stepping in to support the leader and maintain a positive and inclusive group environment.

The gFNP programme includes content to improve maternal health and pregnancy outcomes; to improve child health and development by helping parents provide more sensitive and competent care; and to improve parental life course by helping parents develop effective support networks, plan future pregnancies, complete their education and find employment. 23 The curriculum domains were mother’s personal health, the maternal role, maternal life course, family and friends, environmental health and related health and human services, with referrals made when necessary. The gFNP curriculum materials and activities were modified from those used to deliver FNP to reflect group administration. They were designed to avoid a lecture context and to facilitate interaction between group members and between group members and the nurses, providing a range of engaging, often ‘hands-on’ activities. In particular, gFNP had a particular focus on enhancing social support and social networks through dialogue between group members, which is not a specific focus of home-based FNP. 46,57

Specific to the gFNP programme and following NICE guidelines,62 the family nurse midwife (FNMW) provided routine antenatal care during the meeting, taking an approach based on the Centring Pregnancy Programme,47,59,61 which encourages women to monitor their own health (e.g. by testing their own urine or listening to the fetal heartbeat). The Centring Pregnancy approach was perceived to correspond well with the gFNP aims in that both focus on developing self-efficacy and encouraging women to be more self-aware. 46 Once infants were born, both FNs were involved in routine infant checks, conducted in accordance with the UK NHS HCP. 54

An appreciation of the diversity of group members is central to thinking about how the content is delivered, especially for some emotive topics such as ‘safe relationships for our children’. 46 Although there was a curriculum for each meeting, the nurses were sensitive to the need for ‘agenda matching’ related to particular issues raised; this requires the practitioners to listen to the issues that are uppermost for the group members and agree how these can be met, while at the same time ensuring that the session agenda is realised and behaviour adaptation is progressed for everyone. In addition to modelling of infant care, they model respectful relationships and turn-taking,71 which are expected to be of benefit to any group members with poor social skills, especially if they are experiencing difficult interpersonal relationships. 46 Study participants allocated to gFNP could also access any aspect of the HCP usual care that they wished, independently or with the guidance of the gFNP nurses.

Control: usual care

Complete details of the care offered through the NHS to pregnant women and those with infants up to the age of 1 year at the time that the research was conducted can be found in the Healthy Child Programme. Pregnancy and the First Five Years. 54 The HCP, led by health visitors, is delivered through integrated services that bring together Sure Start Children’s Centre staff, general practitioners (GPs), midwives, community nurses and others. In summary, it offers every family a programme of screening tests, immunisations, developmental reviews, and information and guidance to support parenting and healthy choices. There are core universal elements provided for all families with additional progressive, preventative elements for those with medium or high risk. The universal programme includes a neonatal examination, a new-baby review at about 14 days, a 6- to 8-week baby examination, and a review by the time the child is 1 year old and at 2–2.5 years.

It aims to develop strong parent–child attachment and positive parenting, resulting in better social and emotional well-being among children; care that helps to keep children healthy and safe; healthy eating and increased activity, leading to a reduction in obesity; prevention of some serious and communicable diseases; increased rates of initiation and continuation of breastfeeding; readiness for school and improved learning; early recognition of growth disorders and risk factors for obesity; early detection of – and action to address – developmental delay, abnormalities and ill health, and concerns about safety; identification of factors that could influence health and well-being in families; and better short- and long-term outcomes for children who are at risk of social exclusion.

There is a focus on supporting mothers and fathers to provide sensitive and attuned parenting, in particular during the first months and years of life. From the 12th week of pregnancy, women are encouraged to see a midwife or maternity health-care professional for a health and social care assessment of their needs, risks and choices.

Primary outcome measures

Two primary outcome measures of parenting were used because of the difficulties associated with the detection of low frequency events such as child abuse. One is a self-report measure of parenting opinions and the other is an objective measure of maternal behaviour during a parent–infant interaction. Both are known to be able to identify mothers at risk for abusive parenting.

1. The Adult Adolescent Parenting Inventory Version 2 (AAPI-2)72 is a 40-item self-report measure able to discriminate between abusive and non-abusive parents. The total raw score is converted to a standard 10 score, with low scores indicating a higher risk for practising abusive parenting practices. Subscales are also available: ‘inappropriate’ expectations of children (seven items); inability to demonstrate empathy to children’s needs (10 items); strong belief in the use of corporal punishment (11 items); reversing parent–child family roles (seven items); and oppressing children’s power and independence (five items). Responses are on a five-point Likert-type scale ranging from strongly agree to strongly disagree. Internal reliability of the subscales ranges from 0.83 to 0.93, Cronbach’s alphas range from 0.80 to 0.92. The scales were constructed based on factor analysis to demonstrate construct validity and the inventory has discriminant validity comparing abusive and non-abusive parents.

2. The observational CARE-Index73,74 is based on a video recording of 3–5 minutes of mother–child play, and measures three aspects of maternal behaviour (sensitivity; covert and overt hostility; unresponsiveness) and four aspects of infant behaviour (co-operativeness; compulsive compliance; difficultness; and passivity). For this study only maternal sensitivity has been used as the co-primary outcome and has been shown to differentiate between abusing, neglecting, abusing and neglecting, marginally maltreating and adequate dyads. 75 Scores can range from 0 to 14, higher scores indicating better maternal sensitivity and/or infant co-operation. Scoring was conducted blind to allocation. Reliability scoring was completed on a random 10% sample of the recordings.

Secondary outcome measures

Eight secondary outcomes assessed socioemotional aspects of parenting and family life and service use.

1. The observational CARE-Index infant co-operativeness.

2. Maternal depression was assessed (at baseline and at 2, 6 and 12 months post partum) using the Edinburgh Postnatal Depression Scale (EPDS),76 a well-validated 12-item measure of postnatal depression with high reliability (0.88) and internal consistency (0.87), 86% sensitivity and 78% specificity. This questionnaire was scored within 24 hours of its administration so that any woman with a total score above the recommended cut-off indicating a risk of depression, or who responds affirmatively to the question asking about self-harm, could be identified and a health-care professional contacted to give appropriate support.

3. Maternal stress was assessed (at 2 and 12 months post partum) using the Abidin Parenting Stress Index (PSI), Short Form,77 a well-validated 36-item measure of perceived stress in the parenting role with sound test–retest reliability (r = 0.84) and internal consistency (α = 0.91). High scores on the PSI have been associated with abusive parenting,78,79 with some evidence that parenting stress is higher in women with five or more risk factors for child abuse. 80

4. Parenting sense of competence was assessed with the Parenting Sense of Competence (PSOC) scale81 at 2 and 12 months. This 17-item measure has three factors – satisfaction, efficacy and interest – established by factor analysis in a normative non-clinical sample, each with acceptable internal consistency (from 0.62 to 0.72). 82

5. The extent of social support available to the mothers was assessed (baseline and 12 months) using the Medical Outcomes Study (MOS), Social Support Survey. 83 The 20-item scale measures four dimensions of support, established using confirmatory factor analysis (emotional support, tangible support, positive interaction and affection), each with internal consistency of 0.91 or higher, and also provided a total support score (Cronbach’s alpha 0.97); stability over time is also high for each scale (ranging from 0.72 to 0.78). 83

6. Brief questions designed for the study and based on those developed for use when delivering FNP40 asked about maternal smoking and alcohol and drug use.

7. Brief questions designed for the study and based on those developed for use when delivering FNP40 asked about relationship violence.

8. Brief questions designed for the study asked about infant feeding.

Information other than for the primary and secondary outcome at different time points was collected and is shown but was not formally tested (e.g. baby demographics; immunisations; maternal smoking, alcohol and drug use).

Data collection

The trial commenced in February 2013, recruitment and baseline data collection commenced in July 2013, continuing to July 2014, and data collection was completed in March 2016. Data collection was conducted by researchers making four visits to participants’ homes (baseline in early pregnancy, when infants were 2, 6 and 12 months of age), when they administered structured questionnaires, and (at 12 months) made a 3- to 5-minute video recording of the mother and infant together, presented with a standardised set of toys.

At a Project Management Committee meeting (31 October 2014), it was agreed that the target windows for data collection were 2–3.5 months (60–105 days) for the 2-month outcomes, 6–7.5 months (180–225 days) for the 6-month outcomes and 12–14 months (365–425 days) for the 12-month outcomes, although data would still be collected outside those windows if the participant was available. It was also agreed that interviews with mothers whose babies were premature would be timed as much as possible according to their chronological age. The participants were given ‘High Street’ vouchers for £20 at each home visit data collection point to acknowledge their time for participation. All reasonable attempts were made to contact any participants lost to follow-up during the course of the trial to complete the assessments.

Data management

Each participant was allocated a unique identifier (UID) prior to the baseline interview and this UID was recorded on each questionnaire completed for that participant. All questionnaires were anonymous. The researchers sent completed questionnaires by post directly to the London School of Hygiene and Tropical Medicine (LSHTM) Clinical Trials Unit (CTU) and checks were made for receipt. The questionnaires received at the LSHTM CTU were reviewed for errors and omissions and, when possible, these were resolved via communication with the researchers. The questionnaires were stored in a locked cabinet. The data were double entered into a database by trained data personnel. All electronic trial data from questionnaires and electronic management data with personal participant content stored at the LSHTM CTU were password protected and held on secure servers at the LSHTM.

Video-taped play interactions were transferred by the fieldworkers from the camera to encrypted USB flash drives with Advanced Encryption Standard 256-bit military-level security, sent by recorded delivery to the principal investigator (PI), with files deleted from the camera by the fieldworkers. Recordings were decrypted by the PI and saved with full anonymisation of filenames on a dedicated drive separate from any other study information. Copies of recordings were sent on DVDs to the coder by special delivery and codings returned on a password-protected Microsoft Excel (2010; Microsoft Corporation, Redmond, WA, USA) file to the study PI via e-mail. These were converted to a Statistical Product and Service Solutions (SPSS version 22; SPSS Inc., Chicago, IL, USA) data file once all codings had been received and sent by the PI as a password-protected file by e-mail to the trial statistician at the LSHTM CTU.

Sample size calculation

The sample size was calculated for the two primary outcomes: the AAPI-272 and maternal sensitivity from the observational CARE-Index. 73–75 The standard deviation (SD) of the AAPI-2 based on a total sum of the raw scores of 40 items (range 40–200 items) is 10, with differences of 6.7 identified in the normative sample between abusive and non-abusive adult females. 72 The SD for the CARE-Index sensitivity scale (range 0–14) was expected to be around 2.3. 73

For this individually randomised trial, we initially proposed to recruit sufficient mothers and babies (families) to allow the trial to detect a difference between groups of 0.5 SD, with 90% power at a significance level of 0.05 (two-tailed), considered to represent a moderate size of effect. 84 Basing calculation on the AAPI-2, very conservatively assuming a correlation of 0.4 between pre- and post-intervention scores, at least 71 families were needed in each arm of the trial to detect this difference. Allowing for an expected 30% dropout rate (based on the first two applications of the programme in England), we planned to recruit a minimum of 84 families per arm of the trial. We therefore proposed, conservatively, to recruit a minimum of 100 families per arm (n = 200). The proposed sample size would similarly allow us to detect a change of approximately 0.5 SD in the CARE-Index maternal sensitivity score. 73–75 If this was achieved, we expected to be able to detect a difference at follow-up between arms of the trial of approximately 1.2 with 90% power and a 5% level of significance.

However, owing to ongoing slow recruitment, and because two of the Phase I groups with very low numbers were discontinued prematurely, the allocation ratio was changed during the trial from 1 : 1 to 2 : 1 in favour of the intervention arm. As a result of this and the actual recruitment rate, sample sizes were revised to 100 families in the intervention arm and 65 families in the control arm. With the expected dropout rate of 30%, we would still have 82% power to detect the planned differences in the primary outcomes.

Recruitment and consent

Community midwives were initially involved in identifying potentially eligible women based on their age, parity and gestation,85 giving them a study leaflet describing the study (see Report Supplementary Material 1) and asking for written agreement to give their names and contact details to the local researcher as part of a staged consent process, using an ‘agreement to contact’ form (see Report Supplementary Material 2). Owing to a range of factors,86 the identification of potentially eligible participants subsequently involved both Clinical Local Research Network (CLRN) midwives and FNP FNs who generally gained oral agreement for research contact, as approved by the Research Ethics Committee (REC) (amendment #1).

The first research contact was by telephone to confirm eligibility. Women who were not eligible were thanked for their time. Those eligible were given an information sheet about the trial (see Report Supplementary Material 2), and time to think about participation. After at least 24 hours, researchers arranged a home visit, so that written consent could be obtained (see Report Supplementary Material 2) and baseline data could be collected.

Randomisation procedure

The process was overseen by the LSHTM CTU. The UID (which included a site identifier) and age at the LMP of eligible consenting mothers-to-be were passed by the researchers to the central randomisation service at the Health Services Research Unit, Aberdeen, using an automated telephone procedure. Minimisation criteria [site and age group (< 20 years and 20–24 years)] were used to ensure a balance of key prognostic factors. Allocation to one of two arms was securely computer generated and delivered by e-mail to the LSHTM, which conveyed the information to study participants by post and conveyed to each gFNP team the names and contact details of women allocated to the intervention arm by fax or password-protected e-mail, receiving confirmation of receipt by e-mail.

Blinding

The research team collecting the data and the psychologists scoring the videos were blind to treatment allocation.

Statistical analyses

The primary analyses were by ITT and included adjustment for baseline measure of the outcomes when possible (analysis of covariance; ANCOVA). When outcomes were collected at multiple time points to gain power, random-effects models, using a likelihood-based approach, were fitted to the outcomes at all the time points they were measured at simultaneously (Table 1 and see Appendix 1). This has the additional advantage that the data from all participants contribute to the analysis, even if there are missing data at some follow-up time points.

Reflecting the discussion at the Project Management Group (31 October 2014), about appropriate time windows for data collected at 2, 6 and 12 months, the statistical analysis plan as agreed with the Data Monitoring Committee in December 2014 was for the primary analysis to exclude all data outside the windows (i.e. after 12 months + 60 days; 6 months + 45 days; and 2 months + 30 days). A sensitivity analysis was then conducted including all data, even those outside the windows.

For the primary outcome of the AAPI-2,72 a linear regression model was used to estimate a mean difference in AAPI-2 score between the two arms of the trial. For the primary outcome maternal sensitivity score, a mixed-effects model was used with a random effect at the mother level (to allow for multiple births), to estimate a mean difference in maternal sensitivity score between the two arms of the trial. However, only one set of twins was available for this analysis and their responses were identical. Therefore, it was not possible to include a random effect and the analysis was carried out at the mother level using a linear regression model.

For the secondary outcomes, appropriate generalised linear models were used to examine the effect of the intervention. Odds ratios (ORs) and mean differences are reported with 95% confidence intervals (CIs). When continuous measures were available at baseline, they were adjusted for in the analysis.

When there was evidence of non-normality in the continuous outcome measures, the non-parametric bootstrapping, with 1000 samples, was used to estimate the effect of the intervention and bias-corrected CIs are reported. 87 When this was done, p-values were estimated using permutation tests.

An adjusted analysis, adjusting for site and maternal age group, was also carried out. A pre-specified subgroup analysis was planned based on ‘looked-after’ history, but, as there was only one participant in the intervention arm (see Chapter 6), this was not carried out.

It was planned that the impact of being a twin would be explored by including a covariate in all models, but owing to the low number of twins this was not carried out. However, exploratory analyses were carried out to examine the impact of premature birth on all outcomes. Further exploratory secondary analyses were also carried out, in which the small group in which the intervention was delivered was fitted as a random effect to allow for any potential clustering by group.

A complier average causal effect (CACE) analysis88 was also carried out. The CACE analysis estimates a measure of the effect of the intervention on those participants who received it as intended by the original allocation.

A reliability analysis was carried out for the CARE-Index. Ten randomly selected videos (stratified by site) were scored by a second scorer and Lin’s concordance correlation coefficient was calculated and Bland–Altman plots were produced to assess reliability.

Health economic study

A prospective economic evaluation, conducted from a NHS and Personal Social Services (PSS) perspective, was integrated into the trial. The economic assessment method adhered as closely as possible to the recommendations of the NICE reference case. 89 Primary research methods estimated the costs of the delivering gFNP, including development and training of accredited providers, the cost of delivering the group sessions, participant monitoring activities and any follow-up/management. Broader resource utilisation was captured through participant questionnaires administered at baseline, 2, 6 and 12 months post partum. Maternal health-related quality of life (HRQoL) was assessed using the EuroQol-5 Dimensions, five-level version (EQ-5D-5L) measure90 at baseline and at 2, 6 and 12 months post partum. This contains a visual analogue scale (VAS) asking patients to rate their current HRQoL on a scale from 0 to 100, and a five-dimension health status classification system, which can then be converted to a multiattribute utility score by applying a UK tariff. 91

In addition, information was collected about service use that could indicate a risk factor for abuse or neglect, namely contact with a social worker and the child’s attendance at hospital accident and emergency (A&E) departments (all based on maternal reports at 2, 6 and 12 months). Confirmation was to be from Hospital Episode Statistics (HES) records, but these could not be obtained. The unit costs for health and social care resources were largely derived from local and national sources and were estimated in line with best practice. For further details, see Chapter 4.

Process study

The uptake rate of women who agreed to the intervention involved an assessment of the ratio of women randomised to receive the intervention who then attended at least one session relative to those who either refused after meeting with the FN or agreed but never attended any sessions based on standardised data forms completed by FNs.

The study attrition rate was estimated in terms of the proportion of women who dropped out relative to those who continued in either arm of the trial and also those who may or may not have taken part in research visits, but ceased to receive the intervention, based on information provided by the nurses delivering the programme. This included both women who stopped attending and women in areas where the programme delivery ended prematurely.

The extent to which the programme was delivered with integrity was assessed through an analysis of data from the programme’s standardised data forms, documenting attendance and the content domains covered in sessions.

A parallel qualitative appraisal was concerned with understanding ‘how’ the gFNP service:

• was implemented based on data collated by the FNP NU on sessions delivered and attendance of clients, to develop evidence for future roll-out and potential fidelity measures

• was experienced by families and practitioners, to develop recommendations for improvement

• impacted on established roles to understand how barriers to and drivers of change manifest in distinct professional knowledge, practice and cultural domains.

The appraisal was informed by both quantitative data and qualitative interviews, which are further detailed with the results in Chapter 5.

Focus on mothers with a ‘looked-after’ history

After the programme concluded, delivery interviews were sought with participants who had identified at 6 months post partum that they had spent time away from their parent(s) during childhood, in the care of social services. Interviews were also conducted with FNs involved in delivering gFNP in sites, which included the self-identified ‘looked-after’ participants, and with other professionals involved in providing support to young parents who had in their childhood or adolescence been ‘looked after’. For further details, see Chapter 6.

Study harms/adverse events

Information was collected on any hospitalisation of mother or infant other than for delivery; congenital anomaly or birth defect; persistent or significant disability; and death identified by information from participants at data collection points or using pre-paid change of circumstances cards. All events were reported to the REC, which gave a favourable opinion within 15 days of the PI becoming aware of the event.

Ethics

Ethics approval for the main study was granted in May 2014 by the National Research Ethics Service (NRES) Committee South West – Frenchay (REC reference 13/SW/0086). Six substantial amendments to the study protocol were also approved, as follows, most of which were changes that were designed to boost the poor recruitment.

October 2013: approval of –

1. FNs with access to midwifery records and Comprehensive Clinical Research Network midwives (when available), having access to midwifery booking lists to identify potentially eligible participants

2. contact with potentially eligible participants to be by telephone to gain ‘agreement to research contact’

3. a study poster to highlight the study in GP clinics and midwifery waiting rooms

4. extending the recruitment period by 2 months

5. adding one additional exclusion criterion – any woman already enrolled in the trial who experiences fetal death and becomes pregnant again within the recruitment period

6. a letter to be sent to any participant experiencing fetal death

7. a change in the original analysis plan, with a CACE analysis to be carried out after the ITT analysis to determine the effect of the intervention on those who received gFNP as intended.

November 2013: approval of –

1. including in the groups a small number of women who are not part of the research study (called in subsequent sections ‘buffer clients’). They were women not eligible for the research because they were aged 20–24 years but had more educational qualifications than were allowed for eligibility. This was to facilitate the groups being of the minimum size (set at eight), which had become a concern with slow recruitment. The presence of buffer clients has been taken into account in the analyses.

December 2013: approval that –

1. owing to ongoing slow recruitment and two of the Phase I groups with very low numbers being discontinued prematurely, the allocation ratio be changed from 1 : 1 to 2 : 1 in favour of the intervention arm (this was predicted to lead to a reduction in the power of the study from 90% to 80%)

2. additional questions added to the process qualitative interviews so that the experience of a group being discontinued could be examined.

April 2014: approval of –

1. a simplification of the eligibility criteria for 20- to 24-year-olds for the final (third) phase of recruitment, removing the requirement for low/no educational qualifications

2. a slightly revised study leaflet removing mention of the educational requirement.

June 2014: approval that –

1. contrary to the original proposal, the 6-month data collection would be by a home visit rather than a telephone call, a change based primarily on feedback from clients when visited at 2 months, that they did not want to talk extensively on the telephone when coping with a baby, and also as a strategy to maximise study retention

2. participants would be provided with a £20 voucher at 6 months rather than the planned £10, as it was a home visit, rather than the original plan of a telephone call and a voucher to be sent in the post

3. one final question be added to the 6-month interview so that participants could identify whether or not they had any history of being ‘looked after’ by the local authority.

November 2014: approval of –

1. all of the study materials (consent form, information sheets, interview guides) to conduct the qualitative interviews with study participants who had been allocated to receive gFNP and with FNs who had delivered gFNP; interviews to begin once gFNP delivery was complete in the area.

Attendance at Group Family Nurse Partnership groups

Programme delivery and attendance is covered in detail in Chapter 5 so it is only summarised here. In total, the 97 trial participants were allocated to 16 planned groups; five sites planned to offer two groups (A and B) and two sites planned to offer three groups (A, B and C) (Table 6), although in some cases no sessions were delivered for a planned group. In addition, one participant attended sessions offered in groups A and B as the first group was terminated prematurely.

The mean number of gFNP sessions attended by the 117 clients allocated to groups (see Table 6), comprising 99 trial participants (97 allocated to gFNP and two control group participants mistakenly offered gFNP as buffers by FNP teams) and the 18 buffer clients who were not eligible for the trial owing to educational qualifications but were offered gFNP to boost group sizes to a viable number, was 11.8 (SD 13.8).

Overall, the 97 trial participants in the intervention arm attended a mean of 10.3 sessions (SD 13.4), but a substantial proportion (39, 40%) did not attend any sessions (Table 7). Of the 97 randomised to the intervention, 17 were never allocated a gFNP UID number by the relevant gFNP team and did not attend any sessions. The reasons for this are given in Chapter 5 (see Uptake of the programme). Twenty-two of the remaining 80 participants registered for gFNP did not attend any sessions, 10 of whom were allocated to groups that did not offer any sessions. Five of those were offered one-to-one FNP, but no information was available about how much of that service was received and others were referred back to existing services. Thus, of the 97 study participants allocated to the intervention arm, 58 took part in at least one gFNP session. A summary of attendance overall and by group is given in Table 6 (trial participants and buffer clients), Table 7 (only intervention arm trial participants), Table 8 (only intervention arm trial participants, pregnancy sessions) and Table 9 (only intervention arm trial participants, infancy sessions).

Baseline demographics for all intervention arm trial participants and for those who attended at least one group session are given in Table 10. There are no apparent differences between the demographic characteristics of women who attended at least one group session and those of the intervention arm trial participants as a whole.

Reliability of the CARE-Index

Ten videos were analysed for the CARE-Index by two coders. Lin’s concordance coefficient for mother’s sensitivity based on these 10 videos was 0.56 (95% CI 0.27 to 0.85), suggesting poor agreement. Examination of the Bland–Altman plot for this outcome (see Appendix 3) suggests systematic bias for this component of the CARE-Index, with one scorer consistently scoring higher than the other.

Demographics and outcomes at 2, 6 and 12 months

See Appendix 1 for demographic updates. Baby demographics (Table 14), infant feeding (Tables 15 and 16), immunisations (Tables 17 and 18), smoking, alcohol and drug use (Tables 19 and 20), questionnaire scores (Tables 21–23) and all CARE-Index scores (Table 24) follow.

Costing of the Group Family Nurse Partnership programme

A particular focus of the economic evaluation was the assessment of the cost of delivering the gFNP programme in community settings, including the costs of programme development, training of accredited gFNP practitioners, cost of delivering the group sessions, participant monitoring activities and any follow-up/management. This primarily involved asking each of the gFNP practitioners in each site to prospectively complete detailed weekly activity logs outlining the cost of delivering each gFNP session, including costs associated with preparation time, programme delivery time, indirect administrative activities, home visits and telephone contacts, as well as gFNP-related training and supervision activities. The weekly activity logs also recorded the mode, distance and time spent travelling by each practitioner as a result of gFNP-related activities. They also recorded additional expenditures associated with refreshments, materials, cards or gifts, participant travel, partner travel, child care costs and miscellaneous expenditures associated with weekly gFNP-related activities. The costs of venue hire were estimated separately at each site. A copy of the weekly FN activity log is provided online (see Report Supplementary Material 4). The total costs of delivering the gFNP programme across each group and site were subsequently converted into group- and site-specific estimates of average cost per session per attending woman using separately collected attendance data for each group within each site.

Collection of broader resource use data

Data were also collected about all significant health and personal social service and broader societal resource inputs over the period between randomisation and 12 months post partum. Trial participants were asked to complete detailed resource use questionnaires via researcher-administered face-to-face interviews at baseline and at 2, 6 and 12 months post partum to minimise loss of information due to recall difficulties. The data collected from the trial participants at each time point covered their (and, in the case of the postpartum questionnaires, their babies’) use of hospital care services, community-based health care, community-based social care, and medicines and drugs. Information was also collected regarding the use of legal services and costs borne by the trial participants or their family members or friends as a result of the trial participants’ (and, in the case of the postpartum questionnaires, their babies’) health status, over the relevant time horizons. Medication use was categorised by chemical entity, mode of administration, dosage frequency and duration of use. The service use questionnaires were piloted to assess their acceptability and women’s comprehension levels of the questions. Copies of the service use questionnaires administered at each time point are provided online (see Report Supplementary Material 3).

As part of our strategy to measure broader resource use, an application was also submitted to the Health and Social Care Information Centre (HSCIC) during the course of the study for a new data sharing agreement relating to bespoke data linkage of HES admitted inpatient care, A&E and outpatient data to complement data collected within our battery of research instruments. We wanted to use the HES data for our trial participants to validate the self-reported hospital service utilisation data. Unfortunately, despite following the stipulated guidance from the HSCIC at the time of the study design, our application to the HSCIC was rejected on the following grounds: (1) the original consent material (consent form/patient information sheet) was from Birkbeck University of London and did not mention the University of Warwick CTU (the unit co-ordinating the economic evaluation) on either document; and (2) neither document mentioned the HSCIC or mentioned the flow of identifiable data to the HSCIC. Nevertheless, we proceeded to measure hospital service utilisation on the basis of participant self-reports and are confident that the integrity of the study design, conduct and analysis was not compromised.

Valuation of resource use

Resource inputs were valued using a combination of primary research, based on established accounting methods and data collated from secondary national tariff sets. Direct and indirect staff time associated with the delivery of the gFNP programme was valued using national unit costs per working hour for each Agenda for Change band of staff. 93 These unit cost estimates were inclusive of components for staff salaries, employer salary on-costs, qualifications, and revenue and capital overheads. The travel costs for gFNP practitioners were based on standing and running costs per mile provided by the Automobile Association (AA) for travel by car,94 and values published in the Department for Transport Public Service Vehicle Survey for travel by public transport. 95 Inpatient admissions over the study time horizon were delineated by type and duration and valued using per diem costs extracted from the NHS Reference Costs trusts schedule. 96 The use of other hospital-based care was valued by applying unit costs extracted from national tariffs. 97 Costs for the community-based services were calculated by applying unit costs from national tariffs93,97 to resource volumes. NHS net prices per milligram for the medications were obtained from the British National Formulary98 or the British National Formulary for Children. 99 The costs for individual participants or their children were estimated based on their reported doses and frequencies where these were available, or otherwise on an assumed daily dose based on British National Formulary or British National Formulary for Children recommendations. The costs of time taken off work were estimated by applying sex-specific median earnings data to occupational classifications100 derived from self-reported work status information. Other family-borne costs were valued using data reported by the participants as part of the follow-up resource use questionnaires. Unit costs were inflated where necessary to 2014–15 prices (pounds sterling) using the NHS Hospital and Community Health Services Pay and Prices Index. Costs accrued by each trial participant beyond the first 12 months of follow-up were discounted at 3.5%, as recommended by NICE. 89

Cost-effectiveness analyses

The main CEAs were conducted following multiple imputations of all missing cost and outcomes data. The cost-effectiveness results were primarily expressed in terms of an incremental cost-effectiveness ratio (ICER). This was calculated as the difference in mean costs divided by the difference in mean outcomes (QALYs or maltreatment outcome measure) between the trial comparators. The primary analyses adopted the perspective of the NHS and PSS. The non-parametric bootstrapping approach was used to determine the level of sampling uncertainty surrounding the mean ICER by generating 10,000 estimates of incremental costs and benefits. These were represented graphically on four-quadrant cost-effectiveness planes. Cost-effectiveness acceptability curves (CEACs) showing the probability that the gFNP programme is cost-effective relative to standard care across a range of cost-effectiveness thresholds were also generated based on the proportion of bootstrap replicates with positive incremental net benefits.

Unless otherwise stated, all statements about cost-effectiveness were based on a £20,000 per QALY gained threshold. 89 The probability that the gFNP programme was less costly or more effective than standard care was based on the proportion of bootstrap replicates that have negative incremental costs or positive incremental health benefits, respectively. Published estimates of willingness to pay for unit changes in the maltreatment outcome measures are not available in the public domain. Consequently, statements about cost-effectiveness estimated using either the AAPI-272 or the CARE-Index (maternal sensitivity domain)73,74 are based on a hypothetical range of values for the cost-effectiveness threshold (£0–50,000).

Sensitivity and subgroup analyses

Several sensitivity analyses were undertaken to assess the impact of areas of uncertainty surrounding components of the economic evaluation. These involved re-estimating the main cost-effectiveness outcomes under the following scenarios: (1) adopting a wider societal perspective that includes costs incurred by all sectors of the economy and by families and informal carers; (2) restricting the analyses to complete cases (i.e. those with complete cost and outcome data); (3) recalculating the average cost per gFNP session per attending woman by varying the mean number of gFNP sessions attended to the highest and lowest mean number of sessions observed across all groups across all sites; and (4) recalculating the average cost per gFNP session per attending woman by varying the number of gFNP group participants to the highest and lowest number of participants observed across all groups across all sites.

Subgroup analyses were also conducted for the main cost-effectiveness results to explore heterogeneity in the trial population. These were conducted by (1) programme completers (no, yes), where women who participated in a pre-specified number of group sessions of the gFNP programme (set at ≥ 17 sessions to ensure consistency with the main clinical analyses reported in Chapter 3) were regarded as ‘programme completers’ (i.e. as having complied with the protocol sufficiently); and (2) programme phase (I, II, III) to test whether or not organisational learning may have influenced the cost-effectiveness of the gFNP programme.

Long-term cost-effectiveness model

The trial-based economic evaluation focused on the short- and medium-term costs and consequences of the gFNP programme in expectant mothers aged < 20 years with one or more previous live births or expectant mothers aged 20–24 years with low/no educational qualifications and no previous live births. The study protocol allowed for extrapolation of costs and consequences over a longer time horizon if the trial demonstrated statistically significant differences in medium-term outcomes. This would have required the development of a de novo decision-analytic model. Accepted guidelines for good practice in decision-analytic modelling and the general principles outlined in the NICE ‘reference case’ were to be followed. 89,107 Long-term extrapolation of outcomes were to be expressed in terms of QALYs in the event of differences in medium-term outcomes. Both costs and outcomes accruing beyond the first year post partum were to be discounted using a 3.5% annual discount rate in line with current guidance. 89

Objective

It was felt that presentation of the results of the economic evaluation in terms of incremental cost per maternal QALY gained had the potential to miss effects of the gFNP programme on the child (or the broader family), whereas presentation of the results of the economic evaluation in terms of incremental cost per unit change (or unit difference) in each of the primary maltreatment outcomes [(1) AAPI-2 or (2) CARE-Index (maternal sensitivity domain)] was likely to miss relevant consequences of the gFNP programme for the mother and be less amenable to overall judgments of cost-effectiveness by decision-makers.

A discrete choice experiment (DCE) was therefore conducted that aimed to quantify people’s preferences for the disparate outcome measures collected in evaluating the gFNP programme. This would allow decision-makers to look explicitly at the trade-offs between different possible outcomes, and help to assess the net benefit of the gFNP programme in a manner that values the plethora of costs and outcomes across several domains. Ethics approval for the DCE was provided by the University of Warwick’s Biomedical and Scientific REC: REGO-2016-1769.

Background to discrete choice experiments

Discrete choice experiments are increasingly being used in health economics to address a wide range of health policy-related concerns. 108,109 The approach draws its microeconomic foundations from the characteristics theory of demand110 and random utility theory (RUT). 111 The characteristics theory of demand assumes that goods, services or types of health-care provision can be valued in terms of their constituent characteristics (otherwise known as attributes). DCEs involve respondents making a number of stated preference choices in response to DCE questions. According to RUT, respondents are assumed to act in a utility maximising manner and make choices contingent on the levels of attributes in DCE scenarios. Therefore, choice data obtained from respondents’ stated preferences can be analysed using econometric methods compatible with RUT. If the specified attributes are significantly related to respondent choices, findings from data analysis should provide information relating to how the average respondent’s utility (or willingness to pay) is affected by changes in the levels of attributes.

There are five identifiable stages in the design and analysis of stated preference DCEs: (1) identifying the attributes to include in the study; (2) assigning levels to these attributes; (3) designing the orthogonal matrix of attributes and levels using design theory; (4) eliciting preferences for these scenarios; and (5) analysing the responses.

Selection of attributes and levels

A number of approaches have been suggested to identify potential attributes for DCEs, including literature reviews, other evidence on the impact of disease or health technology being assessed, expert opinion, qualitative research and other preliminary studies. 112,113 In this DCE, the attributes were framed by the primary and secondary outcomes of the gFNP trial rather than developed de novo. The attributes were chosen to cover a wide range of potential outcomes that could impact on both the mother and the child, with attributes based on questions from the following trial instruments:

• child abuse potential based on the AAPI-272

• maternal stress based on the Abidin PSI, Short Form77

• parenting sense of competence based on the PSOC scale81

• maternal HRQoL based on the EQ-5D-5L101

• social support based on the MOS, Social Support Survey. 83

In addition, we aimed to estimate marginal rates of substitution between changes in the EQ-5D-5L attribute and the remaining attributes. The intention was to estimate changes in all the attributes on an overall ‘utility’ scale. To enable accurate quantification on the utility scale, it was decided to include two attributes derived from the EQ-5D-5L.

A key consideration was the levels with which each attribute was delineated for the purposes of the DCE. In selecting the levels for each attribute, a balance had to be struck between keeping the task simple and manageable for the respondents while obtaining the necessary information for assessing the relative importance of these attributes and linking to the gFNP trial outcomes. In the DCE conducted as part of the ‘Building Blocks’ trial,38 each attribute, with the exception of EQ-5D health states, was described in a binary format. The attributes selected for the gFNP DCE are potentially describable in terms of two or more levels. We analysed the outcomes data for the selected attributes within the gFNP data set, blinded to trial allocation, with a view to assessing the distribution of scores for each attribute. We initially selected levels for each attribute based on these distributions. Further refinements of the levels chosen was informed by the opinion of the Trial Management Group (TMG), to ensure that the levels chosen reflected plausible states for individuals eligible for the programme to be in. The final attributes and levels chosen, and the instrument from which they were derived, are shown in Table 26.

Discrete choice experiment pilot study

Given the preliminary evidence on the likely number of attributes and levels, we anticipated that using a full factorial design for the DCE would prove impractical. Therefore, the range of final choices to be specified in the pilot DCE questionnaire was defined using an orthogonal fractional factorial design within the SAS package (version 9.4; SAS Institute Inc., Cary, NC, USA). Three rounds of piloting of the DCE questionnaire were conducted. First, we conducted eight cognitive debrief interviews among women representative of the gFNP study population who were identified through local health visitors. This initial phase of the piloting aimed to assess respondents’ ability to complete the task. The specific objectives of this part of work were (1) to determine if respondents understood the DCE task in the way that it was intended; (2) to identify any problems with any individual attributes; and (3) to identify any problems with the questions and DCE structure. In the second part of the pilot study, the revised DCE questionnaire was piloted among 10 women representative of the study population who were identified through Sheffield-based health visitors. In the third part of the pilot study, conducted in a sample of 50 representative respondents identified by the online survey company Ipsos MORI Fieldwork International, the DCE was translated into a web-based format and simply catered for ‘main effects’. The issue of which attributes (and their levels) might interact with each other, and how they interact, was explored using data from these 50 pilot DCE questionnaires.

Development of the final discrete choice experiment questionnaire

The design of the final DCE questionnaire was informed by the pilot exercise and followed best practice in DCE design. 114 The final DCE questionnaire adopted an orthogonal fractional factorial design, which was developed using the SAS software package. We also used evidence from the pilot exercise to explore whether or not the adoption of clearly defined labels for models of social care during the antenatal and postnatal periods might be appropriate within the final DCE design. 115 Based on the pilot research, it was decided not to adopt a ‘labelled choice’ DCE design for the final DCE. The final design contained 16 questions, to which a 17th was added, which was a duplicate of a previous question to check for consistency in respondents’ answers. This repeated question was not included in the final analysis of the data set. A copy of the final DCE questionnaire is provided online (see Report Supplementary Material 5).

Participant recruitment and data collection

The final DCE survey was conducted by an online survey company (Ipsos MORI), which was responsible for the design of survey web pages, translation of the paper-based questionnaire to a web-based one, recruitment of study participants, data collection and data cleaning. The participants were invited to complete the online survey by Ipsos MORI and were reimbursed for their participation (£1 per participation). A copy of the DCE participant information leaflet is provided online (see Report Supplementary Material 6). Previous research has indicated that estimation precision in the design of DCEs (which accounts for the potential competing concerns of statistical efficiency and response efficiency) flattens out at around 300 observations. 116 A total of 600 respondents therefore allowed for two samples to be included in the full survey: one composed of women whose characteristics broadly match the trial eligibility criteria, and the other composed of a representative sample of the general population whose values can be considered relevant for social decision-making purposes. For the sample that aimed to match the general population, a stratified probabilistic sampling approach was adopted to ensure that a representative sample was achieved; the strata were defined by age, sex and region. For both populations, questions were presented in a random order to each individual participant to remove potential biases from the order in which they were asked. The quality of retrieved data, both during a ‘soft launch’ and the main study was assessed. Descriptive statistics for responses to each question were estimated. In addition, responders who were classified as ‘speeders’ (i.e. completed the survey much faster than anticipated) were removed by Ipsos MORI from the data set.

Analysis of final discrete choice experiment data

The analysis followed standard practice in the DCE literature, and involved the estimation of a conditional logit model clustered on patient UID (to allow for multiple responses from each respondent). The model also contained a term for whether the option chosen was the first or second one presented in each pairwise choice, to adjust for any potential for individuals to preferentially favour one alternative based on the ordering of options.

Study population

A total of 166 women were randomised into the ‘First Steps’ trial, 99 to the gFNP intervention and 67 to usual care (control). Two women in the intervention arm were ineligible, and baseline information is provided for 97 women in the intervention arm and 67 in the control arm. Consequently, the baseline study population for the bulk of the health economic analyses was 164 women. There were four sets of twins: three in the intervention arm and one in the control arm. A complete profile of resource use was collected for 141 women and their infants at 2 months post partum (representing 86.0% of the baseline study population). A complete profile of resource use was collected for 136 (82.9%) women and their infants and 138 (84.1%) women and their infants for the 2–6 months post-partum period and the 6–12 months post-partum period, respectively. Overall, a complete profile of resource use over the entire follow-up period was available for 129 (78.7%) women and their infants. A complete QALY profile was available for 103 (62.8%) women, whereas QALY calculations based on baseline and 12 months post-partum EQ-5D-5L data were possible for 131 (79.9%) women.

Resource use and costs

Health-related quality-of-life outcomes

There were no statisticaly significant differences between the intervention and control groups in suboptimal levels of function in HRQoL, as measured by five dimensions of the EQ-5D-5L,90 at each of the follow-up time points (Table 32). Similarly, there were no statistically significant differences in the overall EQ-5D-5L utility score or EQ-5D VAS score between the intervention and control groups, at each of the follow-up time points (see Table 32).

Analyses of incremental costs and incremental health outcomes

A bivariate regression, in the form of a seemingly unrelated regression, was carried out with a view to estimating the incremental costs and incremental health outcomes associated with the gFNP programme (Table 33). The adjusted incremental cost associated with gFNP over the entire follow-up period was £1776 (95% CI –£42 to £3593) when a NHS and PSS perspective was adopted and the analyses were restricted to participants with complete cost and QALY data. The respective values were £1593 (95% CI –£264 to £3451) and £2200 (95% CI £97 to £4304) when the analyses were restricted to participants with complete cost and AAPI-2 data and complete cost and CARE-Index (maternal sensitivity domain) data, respectively. The other pre-specified prognostic factors of maternal age, sex of infant, looked-after status and the presence of twins within the trial population did not have independent significant effects on either costs or health outcomes.

Cost-effectiveness results

Baseline analysis

The incremental cost-effectiveness of gFNP is shown in Table 34 for the women with costs and health outcomes data subject to multiple imputation, by outcome measure. Adopting a study perspective of the NHS and PSS (i.e. that adopted for the baseline analysis) and measuring health outcomes in terms of QALYs, the average total cost was £8179 in the gFNP intervention group, compared with £6107 in the usual-care group, generating a mean incremental cost of £2072. The mean incremental cost-effectiveness of gFNP was estimated at –£247,485 per QALY gained (i.e. on average the intervention was associated with a net positive cost and a net negative effect). The bootstrapped mean ICERs largely fell in the north-west quadrants of the cost-effectiveness plane (Figure 2). The CEAC shown in Figure 2 indicates that, regardless of the value of the cost-effectiveness threshold, the probability that gFNP was cost-effective does not exceed 3%. If decision-makers are willing to pay £20,000 for an additional QALY, the probability that gFNP was cost-effective is approximately 2.3% (see Table 34). This pattern of results was broadly replicated when outcomes were measured using the AAPI-2 score (Figure 3) or the CARE-Index (maternal sensitivity domain) (Figure 4). It is notable, however, that when outcomes were measured in terms of change in AAPI-2 score between baseline and 12 months post partum, the gFNP intervention was associated with a positive health effect (mean incremental gain in AAPI-2 score 0.02). For this outcome measure, the probability that gFNP was cost-effective reached 25.1% at a notional £20,000 cost-effectiveness threshold.

TABLE 34 - Baseline cost-effectiveness results based on the QALY and primary trial outcomes: imputed data, NHS and PSS perspective (£, 2014–15 prices)

NE, north-east quadrant of the cost-effectiveness plane; NW, north-west quadrant of cost-effectiveness plane.

Based on 10,000 bootstrap replicates of the data set.

The gFNP intervention was considered to be ‘cost-effective’ if it had positive net benefit at a £15,000 cost-effectiveness threshold.

The gFNP intervention was considered to be ‘cost-effective’ if it had positive net benefit at a £20,000 cost-effectiveness threshold.

The gFNP intervention was considered to be ‘cost-effective’ if it had positive net benefit at a £30,000 cost-effectiveness threshold.

Table reproduced from Barnes et al. 92 This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.

Outcome measure	Mean costs (95% CI)			Mean effects (95% CI)			ICER (£)	Probability gFNP intervention is
	Intervention (£)	Control (£)	Difference (£)	Intervention	Control	Difference		More effectivea (%)	Less costlya (%)	Cost-effectivea (%)b	Cost-effectivea (%)c	Cost-effectivea (%)d
QALY	n = 97	n = 67		n = 97	n = 67
	8179 (5397 to 10,961)	6107 (5029 to 7184)	2072 (–843 to 4988)	0.92 (0.84 to 1.00)	0.93 (0.85 to 1.00)	–0.01 (–0.05 to 0.03)	–247,485 (NW)	19.2	2.8	2.0	2.3	3.0
AAPI-2	n = 97	n = 67		n = 97	n = 67
	8179 (5903 to 10,455)	6107 (5160 to 7054)	2072 (–392 to 4537)	0.27 (0.14 to 0.40)	0.25 (0.12 to 0.38)	0.02 (–0.17 to 0.21)	111,334 (NE)	58.4	1.9	19.1	25.1	32.9
CARE-Index (maternal sensitivity)	n = 97	n = 67		n = 97	n = 67
	8179 (5903 to 10,455)	6107 (5160 to 7054)	2072 (–392 to 4537)	3.97 (3.54 to 4.39)	4.84 (4.30 to 5.38)	–0.87 (–1.55 to –0.19)	–2382 (NW)	1.2	1.4	< 1	< 1	< 1

FIGURE 2.

(a) Cost-effectiveness plane; and (b) CEAC based on the QALY outcome: imputed data, NHS and PSS perspective (£, 2014–15 prices). Figure reproduced from Barnes et al. 92 This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.

FIGURE 3.

(a) Cost-effectiveness plane; and (b) CEAC based on the AAPI-2 outcome: imputed data, NHS and PSS perspective (£, 2014–15 prices).

FIGURE 4.

(a) Cost-effectiveness plane; and (b) CEAC based on the CARE-Index outcome: imputed data, NHS and PSS perspective (£, 2014–15 prices).

Discrete choice experiment