Evaluation of the Healthy Start voucher scheme on maternal vitamin use and child breastfeeding: a natural experiment using data linkage

Article history

The research reported in this issue of the journal was funded by the PHR programme as project number 13/164/10. The contractual start date was in June 2015. The final report began editorial review in June 2021 and was accepted for publication in May 2022. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The PHR editors and production house have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the final report document. However, they do not accept liability for damages or losses arising from material published in this report.

Copyright statement

Copyright © 2023 Dundas et al. This work was produced by Dundas et al. under the terms of a commissioning contract issued by the Secretary of State for Health and Social Care. This is an Open Access publication distributed under the terms of the Creative Commons Attribution CC BY 4.0 licence, which permits unrestricted use, distribution, reproduction and adaptation in any medium and for any purpose provided that it is properly attributed. See: https://creativecommons.org/licenses/by/4.0/. For attribution the title, original author(s), the publication source – NIHR Journals Library, and the DOI of the publication must be cited.

2023 Dundas et al.

Background

Intervention in the early years is one of the most effective ways to prevent poor health and reduce inequalities. Health improvements during pregnancy, infancy and early childhood, and social patterning of any improvements, can track through life into adulthood and old age. Any health improvements that are made during pregnancy and the early years can reduce the prevalence of poor health and disease, and therefore increase life chances. For example, poor nutrition, tobacco exposure and pollution in pregnancy and infancy affect growth and lung development, in turn leading to respiratory illness in childhood and eventually lung disease in adulthood. 1 Poor nutrition in pregnancy and the early years of life have impacts on health. The developmental model of the origins of chronic disease proposes the causal influence of undernutrition in utero on coronary heart disease and stroke in adult life. 2 Maternal nutrition strongly influences child nutrition, which in turn tracks to adult eating patterns. 3–5 As poor nutrition is associated with lower socioeconomic status,6 these impacts may also perpetuate health inequalities. An intervention that modifies both maternal and infant diets thus has the potential to improve the health of the infant, and for these health gains to track into childhood and adulthood, thus reducing health inequalities.

The First 1000 Days7 initiative started at an international conference on global child undernutrition and has now influenced health policy in some high-income countries, including the United Kingdom (UK). 8 The premise is that the first 1000 days from conception to age 2 years is a period of rapid growth, and therefore what occurs during this time lays the foundations for future development. In 2010, the Fair Society, Healthy Lives report9 recommended giving every child the best start in life as a key policy intervention for a healthier society.

The principle of early life intervention has influenced government policy across the UK, ranging from high-level strategy documents such as Getting it Right for Every Child,10 Every Child Matters11 and Healthy Child Wales,12 through universal policies and interventions such as the Health in Pregnancy Grant13 and universal maternal and child health services to targeted interventions including the Family Nurse Partnership14 and Flying Start Wales. 15 Although these are in place across the UK countries, implementation and structure may differ. Therefore, there is a need to understand what strategies and policies improve health that so policy-makers can make better decisions about policies in the early years.

There have been some evaluations of early years policies. The Healthy Baby Program in Manitoba, Canada, was a programme for low-income pregnant women and then for their children. 16 This programme was associated with an increase in breastfeeding initiation (BI). In Sweden, the Salut Programme is a suite of interventions that combines health promotion and universal prevention interventions with the aim of improving health and well-being in children and families. 17 A positive impact on child and maternal health outcomes was found. However, there are other policies that have been introduced that have not been evaluated in terms of outcomes. In the UK these include enhanced provision of early years childcare,12,18,19 Healthy Start vouchers20 (HSVs) and the Baby Box in Scotland. 21 There have been process evaluations of some of these policies to understand acceptability and how they work in practice, but there is a need to evaluate interventions in pregnancy and childhood to determine which have an impact on health outcomes22 and which are cost-effective.

Intervention: Healthy Start vouchers

Healthy Start vouchers have the potential to improve the health of low-income families, who are likely to have poorer diets than more affluent families, and thereby reduce health inequalities. It is estimated that around 33,745 children and 10,551 women per annum across Scotland23 are entitled to HSVs and over 423,000 households in the UK claim HSVs. 24

Healthy Start vouchers, a means-tested voucher scheme, was introduced UK wide in 2006 as a replacement to the Welfare Food Scheme (WFS). The WFS started in 1940 as a universal benefit to ensure expectant and new mothers had a healthy diet during rationing. It subsequently became targeted to lower-income mothers and families. It provided tokens that could be exchanged for liquid and formula milk and vitamins to expectant and nursing mothers, and to children under the age of 4 years. The HSV scheme expands what the vouchers can be used for. The HSVs can be spent on milk, infant formula milk, fruit and vegetables. Between 2006 and 2021, vouchers worth £3.10 per week were given to eligible women; mothers with a child aged under 1 year received £6.20. The amounts were reviewed annually and have not changed since 2006. There are four main aims of the HSV scheme: to improve the nutrition of pregnant women, to increase fruit and vegetable intake, to initiate and maintain breastfeeding, and to introduce foods in addition to milk as part of a progressively varied diet when infants are 6 months old. 20

In 2002 it was estimated that it would cost the Exchequer £147M per annum to fund the HSV scheme. 25 In 2013/14, £93M was paid to retailers as reimbursement for vouchers. 24 This represented a substantial investment at a time when the UK government was trying to reduce the welfare bill.

Rationale

The effectiveness and cost-effectiveness of the HSV scheme on a range of maternal and child health outcomes have not yet been demonstrated. A previous small (n = 336) comparison of the HSV scheme to the WFS found that mothers eligible for HSV had higher daily intakes of iron, calcium, folate and vitamin C than mothers eligible for WFS. 26 Three months postpartum, the HSV scheme mothers again reported higher iron, calcium, folate, and vitamin C intakes, as well as a higher consumption of fruit and vegetables than mothers eligible for the WFS. 27 A mixed-methods study28 of practitioners and low-income mothers found that recipients valued the vouchers but that there were substantial barriers to access, including low levels of awareness of the HSV scheme among both mothers and practitioners and uncertainty about the eligibility criteria among health professionals. A repeated cross-sectional analysis of the Health Survey for England29 found that fruit and vegetable intake in households eligible for the HSV scheme was similar to that of households in control groups.

A report on the operational aspects of the HSV scheme30 concluded that a comparative study is needed to assess the effectiveness and cost-effectiveness of the scheme. To determine the effectiveness and cost-effectiveness of population-based feeding programmes including the HSV, such interventions need to be evaluated in larger studies using a wider range of outcomes with longer-term follow-up and methods appropriate for deriving causal inferences from observational data.

The evaluation of the effect of a means-tested voucher scheme can inform policies aiming to change behaviour, improve health outcomes and reduce health inequalities. It can inform public health decision-makers as to the benefits of vouchers for milk, fruit and vegetable intake during pregnancy and the postnatal/infancy period. Information on the characteristics of participants and non-participants and the reasons why women do and do not participate can also inform policy, so that if the vouchers are shown to be effective in improving vitamin use and breastfeeding rates, in addition to other health outcomes, take-up can be improved.

One criticism of the WFS was the potential to disincentivise breastfeeding because it offered tokens for formula milk. 31 The HSV scheme widened the scope of the vouchers to include fruit and vegetables and offered free vitamins to pregnant women. Therefore, the hypotheses for this study were that the HSV scheme would increase the use of vitamins in pregnancy, but that through the provision of vouchers for formula milk, it has the potential to decrease BI and duration (BD).

Aim and objectives

The overall aim is to evaluate the HSV scheme in relation to the extent to which HSV improves the nutrition of pregnant women and the health outcomes of their infants.

There are five objectives to investigate:

1. the effectiveness of the HSV scheme in relation to vitamin use in pregnancy, and BI and BD

2. the effectiveness of the HSV scheme in relation to infant and child weight and body size, child morbidity, infant and child feeding, and maternal health

3. how findings differ between different populations (Scotland and UK)

4. to establish actual voucher usage and determine the reasons for uptake and non-uptake of the HSV

5. to establish the cost-effectiveness of the HSV.

Research design

The evaluation took a mixed-methods, natural experiment approach to investigate the effectiveness of the HSV scheme. The analysis used existing survey data and was enhanced with an embedded qualitative study, and the methods to conduct a health economic analysis alongside a natural experiment evaluation were established. Thus, there were three parts to this evaluation:

1. Secondary analysis of two existing data sets, including linking one to routinely collected health data (objectives 1, 2, 3).

2. Qualitative interview study of mothers, including a descriptive analysis of voucher usage (objective 4)

3. Establishing methods for a cost-effectiveness analysis (CEA), and conducting a preliminary analysis (objective 5).

As the intervention was not controlled by the researchers, we used multiple exposed and unexposed groups to assess the effectiveness of the HSV scheme. This follows Medical Research Council (MRC) guidelines for the evaluation of NEs. 32 The HSV ‘intervention’ is not randomly allocated to participants so it was difficult to have complete control for confounding; there were factors associated with both receiving HSV and the outcomes that needed to be taken into account. The use of multiple comparison/unexposed groups helped with the assessment of bias of the evaluation method.

Healthy Start voucher policy

The intervention being evaluated was the HSV scheme. There are four main aims of the scheme: to improve the nutrition of pregnant women, to increase fruit and vegetable intake, to initiate and maintain breastfeeding and to introduce foods in addition to milk as part of a progressively varied diet when infants are 6 months old. 20

It is a means-tested voucher scheme for pregnant women and families with children under the age of 4 years. If these women are in receipt of certain means-tested benefits then they are eligible for vouchers to be spent on milk, infant formula milk, fruit and vegetables. They also receive free vitamins. All mothers under the age of 18 years are eligible for the scheme. Between 2006 and 2021, vouchers worth £3.10 per week were given to eligible women; mothers with a child aged under 1 year receive £6.20. The amounts were reviewed annually and have not changed since 2006. As from April 2021, the monetary value of the voucher increased to £4.25 per week. These can be spent in neighbourhood shops and pharmacies. A health professional signs the application form confirming the pregnancy (or child under the age of 4 years) and that the applicants have received health advice. The voucher scheme was rolled out across the UK in 2006 as a replacement to the WFS.

The requirement for a health professional to countersign the application means that there is an opportunity for health promotion. The health professional who countersigns the application will have offered appropriate health, nutrition and lifestyle information to the mother.

Intervention and control groups

An issue identified from a scoping study of evaluating the HSV scheme in terms of outcomes was the lack of obvious control group. 33 There are different ways to assess the effectiveness of the HSV scheme on maternal and child outcomes. One would be to compare the WFS with the HSV scheme, but there are weaknesses in this comparison. It relies on a control group who is living at a different time period to the current HSV scheme and it would be difficult to identify and then subsequently obtain the necessary data on outcomes, exposures and confounders from existing data sets. This evaluation used contemporaneous control groups that can be identified in current existing data sources.

Guidance for the evaluation of a natural experiment advises the use of multiple comparison/unexposed groups to give multiple perspectives about sources of bias and therefore allow an assessment of these biases. 32 This evaluation used two unexposed groups.

The population of mothers and their children can be divided into four mutually exclusive groups (Figure 1) according to their exposure to the intervention (HSV scheme). Group 1 is the exposed group, and they are eligible for HSV and claim the vouchers. Group 2 is the eligible but not exposed group: they are eligible for HSV but do not claim the vouchers. Group 3 is the nearly eligible group: they are those who just miss out on the eligibility criteria for HSV. Group 4 is the rest of the population: they are those who would never qualify for HSVs.

FIGURE 1.

Population of mothers and their children split by eligibility for the HSV scheme, identifying exposed and unexposed groups.

Comparison groups

With these exposure and control groups, there were three ways to compare these groups:

1. recipients versus eligible but not claiming (group 1 vs. group 2)

2. recipients versus nearly eligible (group 1 vs. group 3)

3. all eligible versus nearly eligible (group 1 and 2 combined vs. group 3).

The comparison groups represent different aspects of HSV effectiveness. Comparison groups 1 and 2 are analogous to evaluating the effectiveness of the ‘dose’ of HSV. Recipients of HSV (group 1) were compared with those who may be similar in many characteristics (group 2), especially in the eligibility criteria, but for whatever reason, they do not claim the HSV; recipients (group 1) were also compared with those who are close to the eligibility criteria but their income is just a bit too high to be eligible for HSV (group 3). Comparison 3 was effectively an intention-to-treat (ITT) analysis; this is how the HSV policy is working in practice.

Assessing effectiveness

This evaluation used a multiple analytical approach, using both propensity score matching (PSM) and regression discontinuity (RD). Using this combination of methods allowed control for both observed and unobserved confounders. In addition, the combination of methods allowed comparison of each method with their associated biases and therefore increased the robustness of the results. PSM is a technique used to limit the bias due to confounding in an observational study. The matching accounts for variables that are associated with receiving the intervention or control and therefore allows the estimation of the effect of the intervention. In RD designs, participants are assigned to an intervention or control based on a cut-off value of an assignment variable. Despite there being no random assignment, the RD design has comparable internal validity to randomised controlled trial (RCT) designs. 34

Table 1 details the method of evaluation for each of the comparison groups and notes, which method can be used for each of the data sets. The Infant Feeding Survey (IFS) does not have a suitable forcing variable for the RD method and so all comparisons were done using PSM.

All comparisons were important, but the main comparison of interest was between exposed (group 1) versus eligible but not exposed (group 2). This was a direct comparison of similar groups intended to be recipients of this targeted intervention and was the basis for the power calculation. The other two comparisons would determine the confidence in the results. The directions, sizes and significance of the effects would be indicative of the confidence in the effect of the HSV scheme. If the other two comparisons were in the same direction, significant and of the same magnitude then that would give us the most confidence that there is an effect of the HSV scheme. There would be reduced confidence if they were in the same direction but varied in size and/or significance and least confidence if the effects were in different directions, and significantly different from one another. Again, comparing effect sizes across the data sets and evaluation methods would also determine the confidence in the effect of HSVs.

Quantitative study: study population

The population was all women eligible for HSVs, that is, pregnant women and families with children under 4 years of age with low incomes or in receipt of certain means-tested benefits, all pregnant women aged < 18 years and all mothers aged < 18 years.

The Growing Up in Scotland (GUS) study was a stratified cluster random sample of the whole of Scotland and is available from the UK Data Archive. 35 Children were recruited to the second Birth Cohort in 2011 when they were about 10 months old; it had a sample size of 6127. Pre-project and preliminary inspections of the data determined that 1350 (22%) were from families that were identified as claiming HSVs and that 413 (7%) were from families that were identified as eligible but not claiming (Table 2).

The IFS23 is a random sample of births in the UK from August to October 2010. There were three stages of data collection; we used data from stage 3, when the infant is 8–10 months old. The sample size was 10,768 with 1906 (18%) were from families that were identified as claiming HSVs and 1535 (14%) were from families that were identified as eligible but not claiming (Table 3).

The IFS is available from the UK Data Archive and provides UK-wide data. It collected a wide range of detailed infant feeding patterns but does not have the range of other secondary outcomes available from the linked GUS data. The use of both complementary data sets would show the likely generalisability of the findings from GUS to the whole of the UK.

Quantitative: confounding variables

To enable greater comparability between the studies, identical/harmonised variables from IFS and GUS were used. The variables are described below.

The following groups have been identified as having more adverse birth outcomes and therefore were confounders for the analysis: those living in the most deprived areas, those in the lowest social classes, lone mothers, primiparous women, teen mothers and those in the lowest education group. The ethnic group of the mother was also an important factor for the outcomes. 36 We planned to include ethnicity as a covariate in the analyses. However, ethnicity in GUS reflects the ethnic composition of Scotland and is 95% white and 5% non-white; owing to the small sample sizes of non-white groups, it was not possible to break the non-white group into more refined ethnic groups. The ethnicity in IFS reflects the ethnic composition of Great Britain (England, Scotland and Wales) and is 82% white. Ethnicity was not asked in the Northern Irish sample.

Confounders in GUS were:

• social class – national statistics socioeconomic classification (NS-SEC) of the mother, father and household

• household income

• educational status – education attainment of mother

• family type – single mother/lone parent, couple family

• primiparous – first born, other children

• age of mother at birth of child – age < 20, 20–4, 25–9, 30–4, 35–9, ≥ 40 years

• ethnic group – white, other ethnic background

• urban/rural classification – large urban; other urban; small, accessible towns; small, remote towns; accessible rural; remote rural.

Confounders in IFS were:

• social class – NS-SEC of the mother: 1, managerial and professional; 2, intermediate occupations; 3, routine and manual occupations; 4, never worked; 5, not classified

• educational status – age mother left full-time education – ≤ 16, 17–8, > 18 years

• single mother

• primiparous – first birth, second or later birth

• age of mother – age < 20, 20–4, 25–9, 30–4, ≥ 35 years.

When analysing GUS data and GUS/linked data, the Scottish Index of Multiple Deprivation (SIMD)37 was used. The SIMD combines information across six domains: income, employment, health, education, housing and geographical access. It provides a comprehensive picture of material deprivation in small areas within Scotland. The index ranks 6505 areas from the most deprived to the least deprived and measures the degree of deprivation of an area relative to that of other areas. The areas employed by the SIMD are data zones and are small: the 6505 data zones have a mean population of 780 people (approximately half the size of a lower super output area in England). The reason for employing small area geography at this scale is to permit identification of relatively small pockets of deprivation and minimise misclassification. Area-based deprivation was available for the IFS. It used an area deprivation measure appropriate for each country of the UK: the Index of Multiple Deprivation (IMD) for England,38 the Welsh IMD for Wales,39 the SIMD for Scotland37 and the Northern Ireland Multiple Deprivation Measure for Northern Ireland. 40

Sample size calculation

This was an analysis of existing data based on children in the GUS sample: as such the sample size was fixed. This calculation gave an indicative effect size that could be detected given the sample size. GUS has a sample size of 6127, with 1350 (22%) who were identified as being from families that were claiming HSVs and 2336 (38%) that were identified as being from families that were eligible but not claiming. GUS is a clustered sample of 199 data zones with approximately 30 respondents per cluster. Assuming an intraclass correlation coefficient of 0.01 and 63% BI, 80% power and 95% significance levels, the sample size of GUS would be able to detect a 5% difference (i.e. 68% BI).

Outcomes

Data sources and numbers available for analysis

As part of the data collection process for GUS, the participants were asked if they consented to data linkage, which would allow for the individuals in GUS to have their survey data linked with routinely collected NHS data. Figure 2 displays the data sets to which we linked the GUS data. We linked the GUS data only for those participants who had consented to have their survey data linked to routinely collected NHS data. To do this, we applied to the NHS Information Services Division eDRIS service. An application was made to the Public Benefit and Privacy Panel for Health and Social Care. We provided information about the project, including why it was important and the methods and analysis, and noting the benefits it would bring to the public. We detailed the variables we required from each data set with justifications for them and also provided the variables we used from GUS. This was to ensure confidentiality. ScotCen (Edinburgh, UK) are the data owners of GUS and they provided the GUS data to eDRIS along with notices of consent. Note that this process has been superseded by a different, but similar, process since the Information Services Division became part of Public Health Scotland.

FIGURE 2.

Diagrammatic representation of the data linkage between GUS sweeps of data collection and routinely collected NHS data. Purple indicates data already collected and available; orange indicates future data collection. A&E, accident and emergency; CHSP, child health systems programme; MIDAS, Management Information and Dental Accounting System; SIRS, Scottish Immunisation and Recall System; SMR01, Scottish Morbidity Records general/acute inpatient and day case; SMR02, Scottish Morbidity Records maternity inpatient and day case.

There was consent from 91% of GUS participants to link their survey responses with the routinely collected NHS data. Sample sizes available in GUS and the routinely linked NHS data are shown in Table 6, with comparisons of demographic information. We also used the GUS birth cohort 2, sweep 2. The sweep 2 follow-up visits were conducted in 2013/2014, which made the children around 3 years of age. The number of GUS participants matched to the GUS birth cohort 2, sweep 1, are shown in Table 5. There were differences between the comparison groups in terms of follow-up, with higher follow-up among group 3 (nearly eligible) than among groups 1 (recipients) and 2 (eligible, not claiming). By the time of sweep 2, some of the group 2 and group 3 mothers were claiming HSVs. This could have been due to changes in financial or family circumstances that meant they now qualified for HSVs. We kept participants in the original groups.

The number of GUS participants linked with routinely collected NHS health data is shown in Table 6. There were high rates of linkage across the three groups.

Statistical analysis

The socioeconomic data from GUS are more detailed than the IFS socioeconomic data and contain information to identify all three comparison groups, allowing a fuller evaluation. The data from IFS are from a larger sample and cover the whole of the UK. Both GUS and IFS allow for multiple comparison groups. Both PSM and RD can be used with the GUS data and comparison groups, but only PSM can be used for the IFS data.

The group of women who are eligible for HSVs are also those who are less likely to breastfeed and be at risk of more adverse outcomes; even within this eligible group, there will be variation in breastfeeding rates. One difficulty is how to control for these confounding characteristics and also be able to detect an effect of HSV given all these other influences on breastfeeding. There are many background characteristics that are related to both adverse outcomes and the likelihood of being low income and therefore eligible for HSV. In both GUS and IFS these can be adjusted for these influences in the analyses. The PSM gives the probability that an eligible person will have received HSV given the characteristics associated with uptake of HSV, and probabilities for the control group. Therefore the two comparison groups should be balanced in terms of the factors associated with uptake of HSV, and any effect of HSV should be unbiased.

Generalisability

The data used for this project came from different existing sources using different methods for collection and different questions. IFS is a questionnaire that subjects complete. Mothers are invited to take part via a letter and can complete the questionnaire either online or on paper and then post it back. GUS is a face-to-face interview with a researcher. Mothers are invited to take part via a letter and an interview time is agreed. The questionnaire is completed using computer-assisted personal interviewing; some parts of the questionnaire may be self-completed during the interview process. These different methods of data collection and differences in questions may cause bias in the results. To make statements about the generalisability of the broader range of outcomes available in GUS, we compared the propensity score-matched results from GUS and IFS for all comparison groups, for the primary outcomes of vitamin use in mothers and breastfeeding for infants.

Reporting of the quantitative results

We have reported the results of all pre-specified primary and secondary outcomes, and not just those that reached statistical significance. The analysis involved conducting many tests, which were not independent of each other. Rather than adjusting confidence intervals or p-values to account for this we have presented the results of all analyses and caution the reader regarding the interpretation of the results.

Qualitative interview study

To inform the understanding of the results of the quantitative study, and to understand the process of claiming and using the vouchers, we gathered qualitative data on the lived experiences of women, ascertaining reasons for uptake and non-uptake of HSVs and what the vouchers are spent on.

We carried out 40 in-depth interviews, spread across each of the following groups, mirroring the comparison groups identified in Figure 1:

1. GUS mothers who received HSVs

2. GUS mothers eligible for HSVs but who did not claim

3. mothers who were receiving HSVs

4. mothers who were eligible for HSVs but who did not claim

5. mothers who were nearly eligible for HSV, but did not meet the criteria.

This study started in 2015, when children in GUS were 5 years old. The HSV scheme is in place until the children are 4 years old. Mothers from GUS may have had difficulty recalling their thought processes from up to 5 years previously. We therefore complemented interviews with GUS mothers with interviews with mothers who were current claimers of HSVs and also those currently eligible for HSVs but who did not claim. We identified them from community groups that support families on low incomes. Those mothers nearly eligible for HSVs were also identified by this method. The mothers were selected to ensure that the sample contained women with a range of characteristics including age, number of children, ethnicity and education level.

The interviews were semistructured, face to face, approximately 1 hour in length and covered a range of topics, including what mothers knew about the scheme; how they found out about it; reservations about using it; and, if relevant, main reasons for not using it. If they were currently using the scheme or had used it previously, we asked about when they started using it, who knew they were using it, perceived extra monetary benefits, how conscious they were of extra money and how they used this extra money.

The interviews were recorded on an encrypted recorder, anonymised and transcribed verbatim. Recordings were anonymised before transferring to the transcribing service. The transcribing agency involved was subject to MRC/Chief Scientist Office (CSO) Social and Public Health Sciences Unit data confidentiality policies. Data were transferred and held in a secure environment subject to MRC/CSO Social and Public Health Sciences Unit data-sharing policies.

The transcripts were checked against the original recording before analysis started. They were coded according to prior and emergent themes using NVivo (QSR International, Warrington, UK) and summarised systematically by charting according to key themes, and emerging hypotheses were tested according to all the relevant data. Particular foci were the processes involved in the take-up, non-take-up or discontinuation of the HSV scheme, the experience of using HSVs and how the vouchers were used. Responses were compared between the five groups of women.

The 20 women sampled from GUS were sent an information leaflet and letter inviting them to take part in the semistructured interviews. A follow-up telephone call was made to get their response. For those who declined, they were replaced by other GUS participants with similar characteristics. The 20 women currently claiming/not claiming HSV were sampled from parent groups in deprived areas of Glasgow. Information leaflets were given to these groups inviting them to take part in the semistructured interviews. The participants were fully informed of the study before agreeing to participate. Once the women agreed to participate, an appointment for interview was made, with a reminder telephone call made in the week prior to the interview. The participating women were offered a £20 voucher in recognition of their time spent being interviewed.

Full details of the design, methods, analysis and results are in Chapter 5.

Health economics and assessment of the cost-effectiveness

Conducting health economic evaluations alongside NEs is relatively new. Therefore, the methodology for doing this is not well established. As part of this project, we developed and proposed methods and guidance for conducting such economic evaluations in population health using observation data from NEs. Such evaluations are subject to the inherent biases that affect analyses using observational data.

Our methodology drew on economic methods guidance50–53 and on evidence from economic evaluations carried out in similar early years contexts. 54,55 In addition, methods from previous studies incorporating economics into NEs in education and microeconomics,56–58 including a recent review on health economic evaluations and observational data,59 were used. Current guidance for economic evaluation focusses on RCT designs and therefore does not address the specific challenges for natural experiment designs. Using such guidance can lead to suboptimal design, data collection and analysis for NEs, leading to bias in the estimated effectiveness and cost-effectiveness of the intervention or policy. Fuller descriptions of the methods to develop the guidance and the health economics plan are reported in Chapter 6. The full economic evaluation is out with the scope of this report; it is subject to ongoing work and will be reported separately.

Patient and public involvement

When conducting patient and public involvement (PPI) for this project, we first defined our ‘population’ group. We identified low-income women as our PPI target group; their experiences reflected the population with the potential to be affected by the HSV scheme and the results of this project. The aim of PPI was to inform the qualitative interview schedule and refine and reflect initial analysis. As part of the PPI work, we undertook a focus group with 11 women to share our initial findings from the qualitative study and quantitative analysis. These women were recruited from third-sector organisations that support low-income families. During this process the women suggested adding the financial strain and challenges to the list of secondary outcomes, as they felt that, although HSVs represent a relatively small cash amount of £3.10, they do contribute some money to the family budget.

As part of the application process to link the GUS data to the routinely collected NHS data, lay members were part of the Public Benefit and Privacy Panel. Their opinions were fed into the decision to allow permission to use routine data for research and offer comment as to the public benefit of the research.

Ethics statement

Descriptive statistics for Growing Up in Scotland

Tables 7 and 8 show the descriptive statistics for the confounders contained in the GUS data set for each of the exposed and control groups. Group 2 had more mothers under the age of 20 years at the birth of their child than those in group 1, but fewer aged 20–4 years old. Apart from these differences the age groups were fairly similar for these two groups. Group 1 mothers were younger at the birth of their child than those in group 3 (see Table 7). For maternal social class, group 1 had more mothers who had never worked. Across all groups, the biggest proportion of mothers was in semi-routine or routine occupations. There was a large proportion (21%) of group 3 mothers in the intermediate occupation group. For group 1, the proportions for household social class were similar as for maternal social class, but groups 2 and 3 had different social class proportions for household and maternal social class. Group 3 had a large proportion (26.8%) in the managerial category. Groups 2 and 3 had similar profiles for mother’s education, but group 1 had more women in lower education categories than groups 2 and 3 (see Table 8). There was a gradient across the groups for lone parent families, with the highest proportion of lone parents in group 1 and the lowest in group 3. There were more women classified as living in the most deprived areas using both Carstairs Index and SIMD in group 1 than in groups 2 and 3. In terms of urban–rural, most women lived in urban areas across all three groups, but fewer group 1 mothers lived in accessible rural areas and more lived in small remote towns than those in groups 2 and 3.

Tables 9 and 10 show the descriptive statistics for the exposed (group 1) and control (groups 2 and 3) for the outcomes at age 10 months (see Table 9) and 3 years (see Table 10). There is a breastfeeding gradient, with the lowest proportion breastfeeding in group 1 and highest in group 3. There is also a gradient with BD: those in group 1 breastfed for a shorter time period than those in groups 2 and 3. Other outcomes in GUS sweep 1 are similar across the groups. For sweep 2, when the children were about 3 years of age, the outcomes were similar across the groups. There was a lower proportion of children who received vitamins at 3 years of age in group 1 than in groups 2 and 3.

Descriptive statistics for Growing Up in Scotland data linked to routinely collected NHS data

Descriptive statistics for outcomes available in the GUS data linked to NHS routinely collected NHS data are shown in Table 11. In groups 1 and 2 the descriptive statistics were similar across all outcomes. There were differences between groups 1 and 2 and group 3, with group 3 tending to have slightly better outcomes across the range of outcomes. There were gradients for both the number of hospital stays and the average length of stay in hospital, with group 1 subjects having more stays with longer lengths of stay and group 3 subjects having fewer stays and shorter durations of stay. The exception was the proportion of children registered with a dentist. For all groups there was an increase across all years in the proportion of children registered with a dentist. By 2016, 90% of all children were registered with a dentist.

Mothers were seen to be engaging with health services for routine tests and immunisations. This was demonstrated by 98% of children in each exposure and control group having a vision test, 100% of children in each exposure and control group having a hearing test, and 100% of children in each exposure and control group having the specified immunisations.

Descriptive statistics for the Infant Feeding Survey

Tables 12 and 13 show the descriptive statistics of confounders and outcomes across the exposure and control groups for IFS. Group 3 had consistently better levels of all confounders than groups 1 and 2. This was different from the GUS sample. There was a higher proportion of mothers in managerial and professional NS-SEC (34.4%) in group 3 than in groups 1 (6.6%) and 2 (15.8%). Group 3 had a much lower proportion of women who were aged ≤ 16 years old when they left full-time education (16.8%) than groups 1 (40.0%) and 2 (25.7%).

For the outcomes (see Table 13), there were gradients across the groups for BI and BD with group 1 having the lowest proportion and shortest duration, then group 2 and group 3 having the highest proportion breastfeeding and the longest duration. For child body weight and size outcomes, groups 1 and 2 were similar to each other. In terms of infant feeding, group 1 infants were introduced to solid food earlier than those in groups 2 and 3, but this was only by a couple of weeks.

Propensity score matching results for Growing Up in Scotland and Infant Feeding Survey

Tables 14–16 show the results for the propensity score analysis. Children were not randomly assigned to the comparison groups. Therefore, to ensure we were making fair comparisons across the groups in evaluating the effectiveness of HSVs, we used PSM. PSM is the probability of being assigned to a treatment being conditional on the observed baseline characteristics. This method tries to mimic the characteristics of a RCT when a RCT is not possible (or desirable) but there are other observational data that can be used. The propensity score is a balancing score, which means that if you condition on the propensity score, the distribution of observed baseline covariates will be similar between exposed and control subjects. In other words, this matching on key characteristics can allow for the groups to be more balanced on covariates. It is a technique that can minimise selection bias and is better at getting to the causal effect than simple covariate adjustment in models. Children were matched using propensity scores to balance the groups according to confounders (see Chapter 2, Methods). The distribution of propensity scores for matching for both the GUS and the IFS databases are shown in Appendix 1, Figures 5–10. These figures show the distributions for matching group 2 to group 1 (see Appendix 1, Figures 5 and 8), group 3 to group 1 (see Appendix 1, Figures 6 and 9), and groups 1 and 2 to group 3 (see Appendix 1, Figures 7 and 10).

For outcomes in GUS sweep 1, across all comparison groups (group 1 vs. group 2, group 1 vs. group 3, and groups 1 and 2 vs. group 3), the outcomes did not differ. There was a slight difference in birthweight for the comparison groups 1 and 2 versus group 3 (p = 0.062), but the difference was only 73 g.

For outcomes in GUS linked to routinely collected NHS health data, across all comparison groups (group 1 vs. group 2, group 1 vs. group 3, and groups 1 and 2 vs. group 3), the outcomes did not differ. There was a slight difference in average length of hospital stay for the comparison group versus group 2 (p = 0.086), but the difference was only 0.2 days.

For outcomes in IFS, outcomes differed across the different comparison groups. For group 1 versus group 2, the children in group 1 were less likely to have been breastfed than those in group 2: 57% ever breastfed in group 1 compared with 69% in group 2. Those in group 1 were also breastfed for a shorter duration than those in group 2. There was a difference in ever breastfeeding and duration of breastfeeding across all the comparison groups. Birthweight differed between groups 1 and 3, and groups 1 and 2 and group 3. All outcomes differed between groups 1 and 2 and group 3 (see Table 16).

Regression discontinuity results for Growing Up in Scotland

The forcing variable for the regression discontinuity method was household income. Categorical income was transformed into equivalised household income to make the forcing variable (see Chapter 2, Methods). It was not possible to use RD methods for most of the outcome variables as assumptions were not met. We present the results for the primary outcome of BI and BD, and the secondary outcomes of birthweight and low birthweight.

Table 17 gives the coefficients and 95% confidence intervals (CIs) for the RD analyses. There were no differences in the outcomes between those in group 1 (recipients) and those in group 2 (eligible). For group 1 versus group 3 (nearly eligible), children in the recipient group were less likely to be breastfed than those in the nearly eligible group. If they were breastfed, it was for a shorter duration (0.45 months shorter on average). Those in group 1 had a lower birthweight and were more likely to be in the low birthweight category than those in group 3. Those in groups 1 and 2 were breastfed for a shorter duration (0.33 months shorter on average) and had lower birthweight (0.018 g lighter on average) than those in group 3.

Table 18 summarises the comparisons across the two methods (propensity score with RD) and the two data sets (GUS and IFS) for outcomes available in both data sets. The main comparison group is group 1 versus group 2. For nearly all the outcomes, apart from ever BI and BD in IFS, the results indicated that there is no effect of HSV on the outcomes. For ever breastfed and duration of breastfeeding, there are differences between propensity score results from GUS and IFS, with the IFS indicating a negative effect of HSV on breastfeeding.

For birthweight and low birthweight, the effect sizes and significance vary across the data sets and methods, indicating less confidence in the results of the HSV on birthweight. For premature births and age at introduction of solid foods, there is more consistency across the data sets and evaluation methods, indicating some confidence in the results.

Propensity score matching results for Growing Up in Scotland and Infant Feeding Survey

Tables 23 and 24 show the results for the propensity score analysis. Mothers were not randomly assigned to the comparison groups. Therefore, to ensure that we were making fair comparisons across the groups in evaluating the effectiveness of HSVs, we used PSM. PSM is the probability of being assigned to a treatment being conditional on the observed baseline characteristics. This method tries to mimic the characteristics of a RCT when a RCT is not possible (or desirable) but there are other observational data that can be used. Matching on key characteristics can allow for the groups to be more balanced on covariates. It is a technique that can minimise selection bias and is better at getting to the causal effect than simple covariate adjustment in models. Mothers were matched using propensity scores to balance the groups according to confounders (see Chapter 2, Methods).

Propensity score matching shows there was no difference in vitamin use during pregnancy for any comparison (82%, group 1 vs. 86%, group 2, p = 0.103; 87%, group 1 vs. 88%, group 3, p = 0.434; and 87%, groups 1 and 2 vs. 88%, group 3, p = 0.435). In addition, there was no difference in vitamin use pre-pregnancy for any comparison (see Table 23).

There were no differences between groups 1 and 2 for any of the secondary outcomes. However, there were differences between groups 1 and 3 for mother’s current general health, current smoking and number of cigarettes, and change in financial status. There were differences between groups 1 and 2 and group 3 for current smoking and number of cigarettes, and change in financial status (see Table 24).

PSM showed that there was no difference in the number of visits to A&E or number of dental treatments across the three comparison groups (Table 25).

PSM was also used for the primary outcomes in IFS. In contrast to the GUS analysis, the results from the IFS showed differences across the comparison groups for vitamin use both before and during pregnancy (Table 26). Vitamin use pre pregnancy was lower for those in group 1 and groups 1 and 2 combined than those in groups 2 and 3 (17%, group 1 vs. 23%, group 2, p < 0.0001; 16%, group 1 vs. 26%, group 3, p < 0.0001; and 19%, groups 1 and 2 vs. 32%, group 3, p < 0.0001). However, for vitamin use during pregnancy, those in group 1 and groups 1 and 2 had higher proportions taking vitamins during pregnancy [89%, group 1 vs. 86%, group 2, p = 0.015; 89%, group 1 vs. 87%, group 3, p = 0.012; and 88%, groups 1 and 2 vs. 86%, group 3, p = 0.0023 (see Table 26)].

Regression discontinuity results for Growing Up in Scotland

The forcing variable for the RD method was household income. Categorical income was transformed into equivalised household income to make the forcing variable (see Chapter 2, Methods). It was not possible to use RD methods for most of the outcome variables as assumptions were not met. Figures 3 and 4 show that the pre and post discontinuity trends are not stable. We present the results for the primary outcome of vitamin use before and during pregnancy; and the secondary outcomes regarding financial management.

FIGURE 3.

Regression discontinuity plot with outcome as probability of taking vitamin D or folic acid before pregnancy against log₁₀ of household income as index variable. The coefficient estimate was 0.077 (z = 3.92, p = 0.0001).

FIGURE 4.

Regression discontinuity plot with outcome as probability of taking vitamin D or folic acid during pregnancy against log₁₀ of household income as index variable. The coefficient estimate was 0.11 (z = –6.255, p = 0.0001).

Figure 3 shows the mean probability of taking vitamin D or folic acid before pregnancy for groups 1 and 2. Figure 4 shows the RD for taking vitamin D or folic acid before pregnancy with the log₁₀ of household income as assignment variable. The log10 of the income was used to normalise the distribution. In this analysis the estimate for the coefficient was 0.077 (0.030–0.124), showing a change in probability of taking vitamin D or folic acid at the income threshold (£16,190) of 7.7%, which was also termed as average RD treatment effect. 22 It should be noted that for the logistic regression the odds ratio of taking vitamin D or folic acid before pregnancy for group 1 was 0.953 (0.737–1.235) (Table 27). For the RD there was a statistically significant increase in the use of vitamin D or folic acid of approximately 8% to the right of the threshold. The estimates of RD coefficients are reported in Table 27 for group 1 versus group 2, group 1 versus group 3, and groups 1 and 2 versus group 3. A negative estimate at the discontinuity means a drop just to the right of the threshold and a positive value an increase. A change at discontinuity should be interpreted in the context of the probability estimate for the y axis. For example, as shown in Table 27 there is a significant decrease (7.7%) at the discontinuity for the subjects with income just below the threshold. On the other hand, there is a negative significant drop of 11% for taking vitamin D or folic acid during pregnancy for the subjects with income just above the threshold, meaning that subjects with lower income at the threshold are benefiting (see Figures 3 and 4).

Table 27 gives the coefficients and 95% CIs for the RD analyses.

Table 28 summarises the comparisons across the two methods (PSM with RD) and the two data sets (GUS and IFS) for outcomes available in both data sets. The main comparison group is for group 1 versus group 2. For both outcomes, vitamin use pre and during pregnancy, the results indicated that there is no effect of HSVs on the outcomes. There are differences between propensity score results from GUS and IFS, with the IFS indicating a positive effect of HSVs on vitamin use. There are differences between PSM and RD results from GUS, with the RD results indicating a negative effect of HSVs on vitamin use.

Introduction

The qualitative component of the HSV scheme evaluation aimed to gain an understanding of the lived experiences of those intended to benefit from the scheme. This complements the quantitative component by providing insight into some of those findings. Qualitative interviews explored the processes involved in the take-up or non-take-up of HSVs, and how the vouchers were used and perceived by women.

Methods

Results

Summary

This qualitative research was designed to both complement the quantitative analysis and elicit women’s lived experiences and understanding of the HSV scheme. In doing so, the findings from the quantitative analysis can be situated through the experiences of women who use and do not use the HSV scheme. We sampled women in the same comparison groups used in the quantitative analysis (eligible and claiming, eligible and not claiming, and nearly eligible) and from two different time frames (those who currently use the scheme and those who used the scheme up to 5 years previously).

The rich accounts given by the participants gave rise to four key themes: knowledge, awareness, take up and use of the HSV scheme; opinions of the HSV scheme; the effect of HSVs on diet and feeding choices for their babies and children; and the broader lives of low-income women.

The HSV scheme’s aims were well understood and the scheme was valued by participants. They thought that the eligibility criteria should be widened because they had positive experiences and appreciated the scheme and the impact the vouchers had on their lives. For some mothers, it supported them to provide a healthy diet and the opportunity to give their children a range of fruit and vegetables. The HSVs were not mentioned in their decision-making around breastfeeding. Women’s choice to breast or formula feed was based on a range of other factors, such as support to breastfeed and assumptions and expectations of health professionals. The HSVs were incorporated into budgeting strategies and were seen as a good support in the context of the lives of low-income women.

The results of this qualitative analysis support some of the findings from the quantitative analysis.

Background and objective

Methods

Impact on child health

Most of the child outcomes were available in both GUS and IFS. For the main comparison group (group 1 vs. group 2) for the primary outcome of ever breastfed and duration of breastfeeding, there were differences between the propensity score results from GUS and IFS, with the IFS indicating a negative effect of the HSV scheme on breastfeeding. For all the secondary outcomes, birthweight, low birthweight, prematurity and age starting solid foods, the results indicated that there was no effect of the HSV scheme on the outcomes.

When using the other comparison groups, for birthweight and low birthweight the effect sizes and significance varied across the data sets and methods, indicating less confidence in the results of the HSV scheme on birthweight. For premature births and age at introduction of solid foods, there was more consistency across the data sets and evaluation methods, indicating some confidence in the results.

Previous studies that examined outcomes of the HSV scheme did not examine the range of birth and child outcomes, but instead compared fruit and vegetable intake and found that fruit and vegetable intake did not differ between HSV recipients and non-recipients. 29 Another programme that provides universal multisectoral health promotion interventions during pregnancy and early childhood was found to produce positive effects and was cost effective for birth outcomes. 17

Impact on maternal health

There were high rates of vitamin use during pregnancy across all groups in both data sets (group 1, 82%; group 2, 86%; and group 3, 88% in GUS, and group 1, 89%; group 2, 85%; group 3, 88% in IFS). For the main comparison of group 1 versus group 2, only the primary outcomes of vitamin use pre and during pregnancy were available in both GUS and IFS. The results indicated that there was no effect of HSVs on the outcomes. There were differences between propensity score results from GUS and IFS, with the IFS indicating a positive effect of HSVs on vitamin use. There were differences between propensity score and RD results from GUS, with the RD results indicating a negative effect of HSVs on vitamin use.

Previous studies that examined outcomes for the HSV scheme were smaller than this evaluation. They compared HSV recipients to WFS recipients and found higher daily intakes of iron, calcium, folate and vitamin C for HSV recipients than for WFS recipients. 26 Three months postpartum, the HSV mothers again reported higher iron, calcium, folate and vitamin C intake than WFS recipients, as well as a higher consumption of fruit and vegetables. 27

Further details on health outcomes, health behaviours and financial difficulties of the mothers were available in the GUS data set. For health during pregnancy, alcohol use and household managing financially, there was no effect of HSVs across all the comparison groups. There were slight differences in the mother’s current health as assessed by the SF-12 (physical and mental health domains) when group 2 was used as the control group. For smoking there were no differences for the primary comparison comparing group 1 with group 2, but there were large differences when group 2 was used as the control group. This finding is concerning as it could represent an unintended consequence of HSVs. The HSV scheme provides an opportunity to intervene for mothers with low incomes. The HSV scheme is a nutritional programme but is meant to encompass other aspects of health promotion. A report on the status of the HSV scheme 10 years after its implementation recommended that a package of health promotion measures should be incorporated into the scheme to provide this additional support for mothers and their children. 102

Understanding women’s experiences of the Healthy Start voucher scheme

The rich accounts given by the participants gave rise to four key themes: knowledge, awareness, take-up and use of HSVs; opinions of HSVs; the effect of HSVs on diet and feeding choices for babies and children; and the lives of low-income women.

The HSV scheme aims were well understood and the scheme was valued by participants. Participants had some opinions about the eligibility criteria (mainly to widen access) as they had positive experiences and appreciated the scheme and the impact that the vouchers had on their lives, supporting them to provide a healthy diet and the opportunity to give their children a range of fruit and vegetables. The HSVs were not mentioned in the decision-making around breastfeeding. Women’s choice to breast or formula feed was based on a range of other factors, such as support to breastfeed and assumptions and expectations of health professionals. The HSVs were incorporated into budgeting strategies and were seen as a good support in the context of the lives of low-income women.

A recent report on the development and planned purpose of the HSV scheme undertook a review of how the policy works in practice. 102 Their findings were similar to our study in that uptake could be low during pregnancy; the beneficiaries were not provided with consistent information on how to use the food vouchers or the importance of taking vitamins pre pregnancy, or with support to breastfeed. The report recommended including a package of information to be sent with the vouchers.

Integration of the qualitative and quantitative findings

A key feature of the mixed-methods approach is the ability to integrate the findings between the quantitative and qualitative studies. The accounts of the women in the qualitative study can be used to inform the findings from the quantitative study. There are a number of reasons that the quantitative analysis was inconclusive, and the qualitative study offered rich data on five themes that could help explain this:

Applying and developing methods

Challenges, limitations and strengths of the study

This study used a mixed-methods approach to understand the effectiveness of the HSV scheme: a quantitative analysis of existing data sets, a qualitative study of mothers’ experiences of the HSV and the development of methods to undertake a health economic evaluation to assess cost-effectiveness. The results should be read in the context of the challenges, limitations and strengths of the study.

Implications for future research

There is a need for future research to replicate some of the findings in different comparison groups and different data sets, especially with data that can assess dietary measures. Any replication study should address and overcome some of the challenges and limitations noted in this study. The GUS data linked to routinely collected NHS data provide an opportunity to assess longer-term outcomes for children and mothers in a cost-effective and non-burdensome way on participants. Further research is needed to understand the barriers to providing a healthy diet. Mothers stressed the importance of providing a good diet but were not always able to provide this. One of the reasons could be that the information and discussions come too early in the child’s life, so there is a need to understand the optimal timing for provision of this advice and support. The HSV scheme provides a good intervention point for women who are pregnant or mothers of young children. This could be used as an opportunity to undertake research to understand more about lifestyles of women of childbearing age, and what extra support could be offered alongside the HSVs.

Further research on modelling the monetary value of the HSVs would provide research evidence to understand if an increased value of the vouchers could have more of an impact on child and maternal health outcomes and offer a better mechanism for mothers to provide a healthy diet. In this study HSVs were £3.10 per week (with £6.20 for the first year of life). Recently, the value of HSV has been increased to £4.25 (and £8.50) per week.

One of the challenges of evaluating existing policies is it is difficult to ascertain a suitable control group. 33 Policy makers and researchers should identify evaluation criteria, methods and opportunities at an early stage in the policy life cycle. 109,110 That way, rigorous outcome evaluations can be embedded during policy development as well as during roll out and implementation. The evaluation can be designed optimally and suitable control groups and data can be identified. There is a need for more outcome evaluations, as well as process evaluations of policies to understand their effects on the population. 22

Research that can provide answers for the above implications will provide policy makers with options on how to optimise the HSV scheme to improve maternal and child health, and to therefore make real changes to the life chances of children.

Implications for policy

The results of this study identify inherent structural limitations of HSV. Some of these are noted in previous research, for example access and eligibility to the HSV scheme and accessing vitamins. These structural limitations limit the possible effectiveness of the HSVs but could be addressed by increasing awareness of the HSV scheme, increasing access to vitamins during pregnancy and consideration of the increased monetary value of HSVs.

Contributions of authors

Ruth Dundas (https://orcid.org/0000-0002-3836-4286) led the design and execution of the study; oversaw data linkage, study analyses and integration of methods; and led the preparation of the final report.

Massoud Boroujerdi (https://orcid.org/0000-0002-5722-9645) conducted the quantitative analysis and contributed to the methods and results chapters of the report.

Susan Browne (https://orcid.org/0000-0003-1785-1429) conducted the qualitative interview study, collected and analysed the data, and contributed to the methods and results chapters of the report.

Manuela Deidda (https://orcid.org/0000-0002-0921-6970) conducted the health economics analysis, and contributed to the methods and results chapters of the report.

Paul Bradshaw (https://orcid.org/0000-0002-8590-6156) provided expertise in the use of GUS and contributed to the data linkage application process and to chapters in the report.

Peter Craig (https://orcid.org/0000-0002-7653-5832) provided expertise in the natural experiment methods and the policy evaluation, and contributed to the design of the quantitative analysis and interpretation, and to chapters in the report.

Emma McIntosh (https://orcid.org/0000-0001-6340-3083) oversaw the health economics aspects (including design, methods and analysis) and contributed to chapters in the report.

Alison Parkes (https://orcid.org/0000-0002-4177-8041) provided expertise in the use of GUS and contributed to chapters in the report.

Daniel Wight (https://orcid.org/0000-0002-1234-3110) oversaw the qualitative study aspects (including design, methods and analysis) and contributed to interpretation and to chapters in the report.

Charlotte Wright (https://orcid.org/0000-0001-6256-6315) provided expertise in child health, feeding and nutrition, and contributed to chapters in the report.

Alastair H Leyland (https://orcid.org/0000-0003-3741-7099) provided expertise in statistics, natural experiment methods and policy evaluation in maternal health, and contributed to the design, methods and interpretation, and to chapters in the report.

Data-availability statement

The quantitative analysis used third-party data made available under licence that the author does not have permission to share. Requests to access the data should be directed to the UK Data Service DOI: 10.5255/UKDA-SN-7432-4 for GUS data; DOI: 10.5255/UKDA-SN-7281-2 for IFS data. For the qualitative data, requests should be submitted to the corresponding author for consideration. Access to anonymised data may be granted following review.

Disclaimers

This report presents independent research funded by the National Institute for Health and Care Research (NIHR). The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, the PHR programme or the Department of Health and Social Care. If there are verbatim quotations included in this publication the views and opinions expressed by the interviewees are those of the interviewees and do not necessarily reflect those of the authors, those of the NHS, the NIHR, the PHR programme or the Department of Health and Social Care.