Health Equity Indicators for the English NHS: a longitudinal whole- population study at the small-area level

Article history

The research reported in this issue of the journal was funded by the HS&DR programme or one of its preceding programmes as project number 11/2004/39. The contractual start date was in January 2013. The final report began editorial review in December 2015 and was accepted for publication in March 2016. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HS&DR 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.

Declared competing interests of authors

Richard Cookson and Peter Goldblatt report grants for related work received during this study and that they are members of the NHS Outcomes Framework Technical Advisory Group, and Brian Ferguson reports that he is Chief Economist, Public Health England.

Copyright statement

© Queen’s Printer and Controller of HMSO 2016. This work was produced by Cookson et al. under the terms of a commissioning contract issued by the Secretary of State for Health. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

Overview

This report describes the findings of independent research to develop health equity indicators for monitoring socioeconomic inequalities in health-care access and outcomes in England. Inequalities of this kind persist, raising important public policy concerns about both quality of care and social justice. However, progress in addressing these concerns is hampered because socioeconomic inequalities in health-care access and outcomes are not yet monitored systematically in England at either national or local levels. 2

We developed an integrated suite of equity indicators for two different kinds of monitoring:

1. annual monitoring of change in national health-care equity

2. annual monitoring of local within-area health-care equity against a national equity benchmark, for Clinical Commissioning Groups (CCGs) or other subnational areas comprising ≥ 100,000 people.

Our equity indicators are designed to help national and local decision-makers in England discharge the NHS health inequalities duties introduced in the Health and Social Care Act 2012. 3 The local duty for CCGs is as follows:

Each clinical commissioning group must, in the exercise of its functions, have regard to the need to –

(a) reduce inequalities between patients with respect to their ability to access health services, and

(b) reduce inequalities between patients with respect to the outcomes achieved for them by the provision of health services.

Health and Social Care Act 2012. 3 Contains public sector information licensed under the Open Government Licence v3.0

The national duty for NHS England is phrased in the same way, and the national duty for the Secretary of State is as follows:

In exercising functions in relation to the health service, the Secretary of State must have regard to the need to reduce inequalities between the people of England with respect to the benefits that they can obtain from the health service.

Health and Social Care Act 2012. 3 Contains public sector information licensed under the Open Government Licence v3.0

Our indicators will also help to monitor the health inequalities elements of the NHS duties as to promoting integration of care that were introduced in the Health and Social Care Act 2012. 3 Improving the integration of care is a central policy priority for the English NHS, including not only integration within NHS-funded services across different specialties and different primary and acute care settings, but also integration between NHS-funded services and other services that impact on patient outcomes. NHS England announced in 2013 the establishment of a ‘Better Care Fund’ for integrated care across health-care and social-care boundaries, and announced in 2014 a programme of ‘new models of care’ or ‘vanguard sites’ for integrating care between specialties and settings. 4 The relevant local duties on integration and inequalities are phrased as follows:

(1) Each clinical commissioning group must exercise its functions with a view to securing that health services are provided in an integrated way where it considers that this would —

(a) improve the quality of those services (including the outcomes that are achieved from their provision),

(b) reduce inequalities between persons with respect to their ability to access those services, or

(c) reduce inequalities between persons with respect to the outcomes achieved for them by the provision of those services.

(2) Each clinical commissioning group must exercise its functions with a view to securing that the provision of health services is integrated with the provision of health-related services or social care services where it considers that this would —

(a) improve the quality of the health services (including the outcomes that are achieved from the provision of those services),

(b) reduce inequalities between persons with respect to their ability to access those services, or

(c) reduce inequalities between persons with respect to the outcomes achieved for them by the provision of those services.

(3) In this section —

‘health-related services’ means services that may have an effect on the health of individuals but are not health services or social care services;

‘social care services’ means services that are provided in pursuance of the social services functions of local authorities (within the meaning of the Local Authority Social Services Act 1970).

Health and Social Care Act 2012. 3 Contains public sector information licensed under the Open Government Licence v3.0

The phrasing of the Health and Social Care Act 20123 makes it clear that the NHS health inequalities duties include (1) concern for reducing inequalities in the health outcomes or benefits of health care, as well as concern for reducing inequalities of access to health care, and (2) concern for improving the co-ordination of health care with social care and other public services which impact on health outcomes. These two concerns go to the heart of what it means to be a national health service, rather than a national sickness service, and are also reflected in the NHS Constitution, published in 2012. 5 The first principle of the NHS Constitution is that:

The NHS provides a comprehensive service, available to all . . . At the same time, it has a wider social duty to promote equality through the services it provides and to pay particular attention to groups or sections of society where improvements in health and life expectancy are not keeping pace with the rest of the population.

The NHS Constitution for England 2012. 5 Contains public sector information licensed under the Open Government Licence v3.0

The fifth principle is that:

The NHS works across organisational boundaries and in partnership with other organisations in the interest of patients, local communities and the wider population. The NHS is an integrated system of organisations and services bound together by the principles and values reflected in the Constitution. The NHS is committed to working jointly with other local authority services, other public sector organisations and a wide range of private and voluntary sector organisations to provide and deliver improvements in health and wellbeing.

The NHS Constitution for England 2012. 5 Contains public sector information licensed under the Open Government Licence v3.0

These concerns relate to wider health equity concern for reducing social inequality in population health. Social inequalities in life expectancy and health raise important concerns about social justice, because health is essential to human flourishing. 6 In economic terms, health is both a consumption good that people value for its own sake and a capital good that allows people to do the things they value in life. Health care is, of course, only one of many social determinants of health and survival over the life-course, along with in utero and childhood circumstances, income, education, working and living conditions, social support networks, long-term care, lifestyle factors such as smoking, poor diet and physical inactivity, and many other factors. 7–9 However, although health care cannot eliminate social inequalities in health, it can play a role in helping to reduce them. 10–12 Therefore, we sought to ensure that our equity indictors are relevant from a wider-population health perspective, as well as from a health-care perspective, and that our indicators are relevant to the integration of care across different specialties, settings and services.

Our equity indicators are intended for use by NHS and local authority decision-makers for quality improvement purposes, to help policy-makers and managers learn how to improve the delivery of health-care services including integration with social care and other health-related services. They are also intended for use by a wide range of organisations which play external scrutiny roles in helping to hold the NHS to account, including Public Health England and local Health and Wellbeing Boards, health sector regulators (such as the National Audit Office and the Health Select Committee), professional associations [such as the NHS Confederation, British Medical Association (BMA) and Royal Colleges], think tanks (such as the Health Foundation, The King’s Fund and Nuffield Trust), and national and local media organisations. Our indicators are also intended for public reporting, to facilitate more direct forms of public accountability. In principle, our equity indicators can also be used to monitor health-care equity in other high-income countries with well-developed administrative health data sets, to make international comparisons of equity in health care, and to help evaluate the health-care equity impacts of interventions in trials and quasi-experimental studies.

The aims of our study were to:

• develop indicators of socioeconomic inequality in health-care access and outcomes at different stages of the patient pathway

• develop methods for monitoring local NHS equity performance in tackling socioeconomic health-care inequalities

• track the evolution of socioeconomic health-care inequalities in the 2000s

• develop ‘equity dashboards’ for communicating equity indicator findings to decision-makers in a clear and concise format.

The main contributions of our study were as follows. First, we have developed the first indicators for local NHS equity monitoring against a national NHS equity benchmark, including new methods for national benchmarking as well as a new suite of indicators. Our approach has subsequently been adopted by NHS England in the Clinical Commissioning Group Improvement and Assessment Framework, starting with publication of an up-to-date version of one of our key local equity monitoring indicators: inequality in potentially avoidable emergency hospitalization (NHS England 2016,13 NHS Choices 2016,14 University of York 201615). This aspect of our work was also cited by the independent think tank, The King’s Fund, as being a potentially useful way of incorporating equity into routine CCG performance monitoring by NHS England, in a report commissioned by the Department of Health,16 and the University College London Institute of Health Equity are discussing piloting the use of these local equity indicators to monitor progress in vanguard sites. Second, we have developed a more comprehensive suite of national NHS equity indicators than the inequalities breakdowns currently produced in the NHS Outcomes Framework, by including indicators of inequality in health-care access as well as health-care outcomes. Third, by producing our indicators from 2001/2 to 2011/12, we have provided the first comprehensive assessment of health-care equity trends during a key period of sustained effort by the NHS to reduce socioeconomic health inequalities through primary care strengthening. Finally, we have developed a comprehensive suite of visualisation tools for presenting and communicating our equity indicator findings to decision-makers. This includes a one-page ‘equity dashboard’ presenting summary information, automated ‘equity chart packs’ providing in-depth information underpinning the dashboard, and a web-based tool that allows users to their own graphs. Visualisation is an essential component of equity monitoring, because inequality is a complex concept and judgements about ‘fairness’, ‘justice’ or ‘equity’ often involve controversial value judgements about which reasonable people can disagree. A single ‘one-size-fits-all’ headline inequality measure can therefore be misleading. So, it is essential to show people the underlying inequality patterns and trends, to help them understand the meaning and importance of the trends, and draw their own conclusions about equity based on their own value judgements.

We have developed eight general indicators of health-care equity that examine socioeconomic inequalities in health-care access and outcomes at different stages of the patient pathway: (1) primary care supply, (2) primary care quality, (3) hospital waiting time, (4) preventable hospitalisation, (5) repeat hospitalisation, (6) dying in hospital, (7) amenable mortality and (8) overall mortality. We did not include general indicators of socioeconomic inequality in health-care utilisation, such as the total number of non-emergency inpatient or outpatient hospital visits, because when diverse health-care services are grouped together it is hard to tell whether more utilisation reflects better access to care, worse quality of care or worse health.

All eight of our general indicators are potentially suitable for national equity monitoring. However, we found that the last three indicators do not fully meet the more demanding data requirements for local equity monitoring. The main issue was that there are relatively few deaths in any given local CCG area in any given year, making it hard to tell from a statistical perspective whether or not observed differences in social gradients between different local areas are merely a result of the random play of chance. We recommend three indicators as a high priority for local equity monitoring against a national equity benchmark: primary care supply, primary care quality and preventable hospitalisation. Two other indicators could also be used for local equity monitoring: hospital waiting time and repeat hospitalisation. However, as explained in Chapters 8 and 9, these indicators may require further validation and refinement before being used for routine monitoring purposes.

Our general indicators measure socioeconomic inequality across the full range of health-care activity, rather than focusing on one particular condition. General indicators can be used for local monitoring against a national benchmark, whereas at the present time disease-specific indicators can only be used for national equity monitoring of health-care outcomes. This is because the kinds of health-care outcomes we can currently measure on a comprehensive national basis involve rare events, for example hospitalisations or deaths. This is not problematic when we examine the total number of events across all disease areas, which can add up to a large number. But when we focus on one specific disease, the numbers become too small to detect statistically significant differences between local inequality and the national inequality benchmark. However, to illustrate the potential use of disease-specific indicators at a national level we have developed national disease-specific indicators of equity in the areas of coronary heart disease (CHD) and diabetes, which are presented in Appendices 1 and 2.

Our indicators can be used to assess the degree to which health-care equity in England is getting better or worse over time. They can also be used to identify local NHS areas that are performing better or worse than the national NHS average in reducing within-area socioeconomic inequalities in health-care access and outcomes. This information can be used to facilitate health-care quality improvement efforts, to understand why some areas are doing well or badly, to learn lessons, and to share good practice.

However, we would caution against using our equity indicators for setting performance targets with rewards or penalties attached, at least until further experience and understanding of equity monitoring has been built up. The Health and Social Care Act 20123 suggests the use of financial payments to reward CCGs that succeed in reducing inequalities, as one factor to be taken into account when making end of year payments to CCGs to reward quality. Specifically, section 223k of the act entitled ‘Payments in respect of quality’ states that NHS England:

may, after the end of a financial year, make a payment to a clinical commissioning group . . . For that purpose, the Board may also take into account either or both of the following factors – (a) relevant inequalities identified during that year; (b) any reduction in relevant inequalities identified during that year (in comparison to relevant inequalities identified during previous financial years).

Health and Social Care Act 2012. 3 Contains public sector information licensed under the Open Government Licence v3.0

The process of paying CCGs for quality was subsequently implemented in a process known as the ‘quality premium’, although so far health inequality has not been incorporated into this process (www.england.nhs.uk/resources/resources-for-ccgs/ccg-out-tool/ccg-ois/qual-prem/; accessed 12 July 2015). We would caution against too ambitious a timescale for incorporating our CCG-level equity indicators into decisions on this process for two reasons. The first reason is that health-care equity monitoring is still in its infancy and is less well developed than the monitoring of health-care quality for the average patient. For example, health-care decision-makers have a reasonably good idea about how to reduce average hospital waiting times, supported by a strong evidence base from decades of international policy experimentation, monitoring and evaluation. By contrast, rather less is known about how to reduce socioeconomic inequality in hospital waiting times or other forms of health-care access and outcome. The second reason is that the causal links between policy action and health-care outcome are more complex, delayed and uncertain for some of the health-care outcomes we measure, such as preventable hospitalisation and amenable mortality, compared with health-care outcomes traditionally used for performance management, such as rates of antibiotic resistant bloodstream infections in hospitals. This can make it hard straightforwardly to attribute change in inequality in these outcomes to recent actions taken by CCG managers or the services they commission. Given the current state of knowledge, therefore, the most appropriate initial uses of our indicators are (1) to hold the NHS to account, and (2) to improve quality by helping decision-makers learn how to reduce social gradients in health care and by helping researchers build a stronger evidence base, rather than (3) to set high-powered financial and managerial incentives.

Throughout the study the research team was guided by an advisory group including academic and clinical experts, NHS and public health officials, and lay members, whose membership is listed in Appendix 3. All key decisions around indicator selection and the development of analytical methods and visualisation tools were taken in consultation with the advisory group. The team is grateful for its advice and support, although the responsibility for all decisions rests with the research team.

The next two sections of this introductory chapter set out the background to this study and present the conceptual framework we developed for monitoring equity in health care. Chapter 2 of the report describes how members of the public were involved in selecting our indicators and designing our visualisation tools, through a public consultation exercise in York based on an online survey and a citizens’ panel meeting, and through the participation of the two lay members of our advisory group. Chapter 3 describes the indicator selection process, which included reviewing existing indicators used by the NHS to monitor health-care quality, consulting health indicator experts about technical feasibility and consulting NHS and public health experts about policy relevance. Chapter 4 describes the data and analytical methods used for health-care equity indicator production and visualisation at both national and local levels. Chapter 5 presents the main results for all eight of our general indicators, including national health-care equity in 2011/12, national health-care equity time trends during the 2000s and local health-care equity monitoring in 2011/12 against a national benchmark. Chapter 6 describes the NHS engagement process we undertook to develop and refine our visualisation tools. Chapter 7 presents our prototype ‘equity dashboards’. Finally, Chapter 8 discusses our findings and Chapter 9 summarises our conclusions and research recommendations.

The report also contains extensive appendices. Appendices 1 and 2 present national disease-specific health-care equity indicators for CHD and diabetes, respectively. Appendix 3 lists the advisory group members. Appendix 4 contains full technical specifications of our main general indicators. Appendix 5 presents sensitivity analysis around different ways of cleaning our data by trimming outliers. Appendix 6 contains materials from the public consultation exercise. Finally, Appendix 7 contains letters from the three key NHS organisations we consulted during the development process confirming their interest in seeing our equity indicators routinely produced and used for NHS quality improvement: the NHS England Inequality and Health Inequalities Unit, Hull CCG and Vale of York CCG.

Background on equity in health care

Conceptual framework

Our monitoring framework has the following general design objectives:

1. to monitor equity in both health-care access and outcomes, after appropriate need or risk adjustment

2. to monitor overall equity in health care for the general population, while allowing disaggregation by age, sex and disease category

3. to monitor the equity performance of the health service as a whole, including the integration of care across different specialties, different primary and acute care settings, and different health care, social care and other public services

4. to monitor equity at all main stages of the patient pathway

5. to monitor local equity performance against a national equity benchmark

6. to monitor equity trends alongside equity levels, and average performance alongside equity performance

7. to summarise all key findings in a one-page summary (‘equity dashboard’)

8. to provide visual information about underpinning inequality patterns and trends (‘equity chart packs’)

9. to provide a battery of inequality measures that are easy to understand and capture importantly different concepts of inequality that can trend in different directions

10. to ensure indicators can be understood by members of the general public.

Figure 1 illustrates our framework for monitoring inequality in health-care access and outcomes at key stages of the patient pathway and shows how our eight general indicators fit into this framework.

FIGURE 1.

Conceptual framework for monitoring inequality in health-care access and outcomes at key stages of the patient pathway.

As an example, Figure 2 illustrates how we monitor national equity trends using indicator 1, primary care supply. The top panel shows a breakdown of patients per full-time equivalent (FTE) general practitioners (GPs) by deprivation quintile group, allowing for need and population change, and the bottom panels show how this translates into two standard inequality measures that look at the whole of the social gradient in health care: the SII and the relative index of inequality (RII). These measures and graphs are explained in more detail in Chapter 4.

FIGURE 2.

National monitoring of change in equity over time. (a) Indicator 1: primary care supply; (b) RII; and (c) SII. The top panel shows a breakdown of patients per FTE GPs by deprivation quintile group, allowing for need and population change, and the bottom panels show how this translates into two standard inequality measures that look at the whole of the social gradient in health care: the SII and RII. FTE, full-time equivalent; GP, general practitioner; IMD, Index of Multiple Deprivation; RII, relative index of inequality.

Finally, Figure 3 illustrates our framework for local equity monitoring against a national benchmark. This figure shows socioeconomic inequality in preventable hospitalisation within a fictional local NHS area called ‘Any Town’. The basic idea is to compare the social gradient in health care within Any Town against the social gradient in health care within England as a whole. The social gradient shows the pro-rich link between socioeconomic status and preventable hospitalisation, after allowing for exogenous risk factors influencing preventable hospitalisation that are not under the control of the NHS, in this case, age and sex. As explained above, in Concepts of equity in health care, we would ideally also want to adjust for morbidity, or, more precisely, that part of morbidity that is not under the control of the NHS, but were unable to do so because of data limitations. The relevant NHS equity objective is to reduce the social gradient in health care, both in Any Town and in England as a whole.

FIGURE 3.

Local equity monitoring against a national benchmark: preventable hospitalisation in neighbourhoods in a fictional NHS area (Any Town). Neighbourhoods are ranked by deprivation, with more deprived neighbourhoods to the right.

Any Town has a population of approximately 200,000 people. Each dot represents one of the 125 neighbourhoods in Any Town, each containing approximately 1500 people. Neighbourhoods are ranked by deprivation, with more deprived neighbourhoods to the right. The Any Town inequality gradient is simply a regression line fitted through these 125 dots. The England inequality gradient is a regression line fitted through all 32,482 neighbourhoods in England. In this example, Any Town is doing better than England as a whole both for the average patient (a lower average line) and in terms of reducing inequality (a flatter inequality gradient). In this example, these differences are statistically significant and unlikely to be merely because of the random play of chance. The NHS may, therefore, be able to learn lessons from Any Town about how to tackle inequality in preventable hospitalisation.

Introduction

Public involvement was important to our study because one of the main purposes of our indicators is public reporting for democratic accountability, as well as facilitating quality improvement efforts by national and local decision-makers. We therefore wanted to select indicators of socioeconomic inequality in health care that members of the general public will consider meaningful and important. Before selecting our indicators, we therefore considered it important to ask the general public about what they view as the most unfair socioeconomic inequalities in health care. We also sought feedback from members of the public to help refine our visualisation tools for communicating the findings of our indicators and to ensure that members of the public are capable of understanding our indicators.

This chapter describes how members of the public were involved in this study. They were involved in two ways. First, through a small-scale public consultation exercise in York conducted at the beginning of the study to give us a better understanding of what kinds of socioeconomic inequality in health care are of most concern to members of the public. This involved both an online survey (with 155 responses) and a full-day citizens’ panel meeting (with 29 participants) to gather more in-depth views. Second, two members of the public, recruited via our public consultation exercise, gave feedback throughout the project through their membership of our advisory group.

The primary aim of the public consultation was to identify a list of priority areas for monitoring NHS equity performance. This was achieved by asking the public to consider different types of socioeconomic inequality in health and health care and assess which ones they thought were the most unfair. Our key finding was that the public are concerned to reduce inequality in health-care outcomes, but that their concern for reducing inequalities of access, specifically, for GP supply and hospital waiting times, is at least as strong. This finding influenced the selection of equity indicators for our subsequent analysis. At the inception of the project we had presumed that our indicators would focus on health-care utilisation and outcomes, which are the focus of much current academic literature on equity in health care. However, as a result of our public consultation exercise, as well as further development of our conceptual framework in monitoring equity at multiple stages of the patient pathway, we ensured that both GP supply and hospital waiting time were selected for inclusion in our suite of equity indicators.

A secondary aim of the public consultation was to identify two lay members of the public to join our advisory group to contribute further to the indicator selection process and provide feedback on the design of equity dashboards and other visualisation tools for monitoring equity performance.

This chapter is organised as follows. We start with consultation exercise methods, including the sampling approach, development of questionnaire and data collection. We then present the main quantitative results of the public consultation in terms of people’s responses to questions asking them to assess and rank different kinds of inequality in health and health care by degree of unfairness. The results are presented separately for our online survey and citizens’ panel. We then discuss the process of recruitment of lay members and their contribution to the advisory group and, in particular, the design of visualisation tools. Finally, we conclude by discussing the implications of public involvement for our indicator selection.

Methods of public consultation

Results of public consultation

Ranking of unfair inequalities

All participants responded to the question about the most unfair socioeconomic inequalities in health and health care. Figure 4 presents the full distribution of responses to the question about the most unfair inequality. The citizens’ panel group ranked socioeconomic inequality in waiting time for surgery as the most unfair (31%), whereas the online group ranked inequality in death from conditions preventable by good-quality health care as most unfair (33%).

FIGURE 4.

Choice of the most unfair type of inequality in the citizens’ panel and online samples.

However, the following three types of inequalities were identified as the most important unfair inequalities by both citizens’ panel participants and online survey respondents:

1. The richest fifth of people in England wait less time for NHS surgery than the poorest fifth (31% of citizens’ panel participants and 19% of online respondents ranked this as the most unfair inequality).

2. The richest fifth of people in England are less likely than the poorest fifth to die from conditions preventable by good-quality health care (24% of citizens’ panel participants and 33% of online respondents ranked this as the most unfair inequality).

3. The richest fifth of people in England are served by more GPs than the poorest fifth (21% of citizens’ panel participants and 24% of online respondents ranked this as the most unfair inequality).

Role of the lay members of the advisory group

Two participants at the citizens’ panel event were invited to join our advisory group as lay members, to contribute to the research design and, in particular, the process of indicator selection and design of dashboards and other visualisation tools for monitoring changes in NHS equity performance. The selection of lay members was based on their interest in the subject of inequalities in health and health care, willingness to contribute to this project, experience of using NHS health services, ability to communicate with members of the team and availability to join meetings in York and London. Based on these criteria, one male and one female participant were invited to join the advisory group.

The lay members attended all three advisory group meetings in London, were involved in additional face-to-face discussions, and reviewing and commenting on relevant documents. More specifically, the lay members contributed to the project in the following ways:

• They contributed to discussions on the choice of equity indicators that matter to the general public and, therefore, should be considered for monitoring equity performance.

• They provided useful advice about dashboard design to improve presentation and interpretation.

The lay members commented on the prototype NHS equity dashboard designs, as a result of which we revised and simplified the designs to reduce ‘clutter’ on the graphical displays, and added arrows as well as traffic-light colours to help colour-blind users and people who print out in black and white. The lay members also commented on the different types of graph in the chart pack, and reassured us that our graphs were clear and informative to non-expert audiences.

Our lay members agreed in the final project meeting that the current one-page summary dashboard style presents useful information to NHS and public health experts. However, they thought that this concise format may not be appropriate for communication to the public, as it provides too much information in a small space. They advised that public reporting would require a different kind of infographic design tailored to public audiences. They suggested that the dashboard presentation would be useful to health experts once they are familiar with the dashboard design and have read the accompanying material, but proposed that clear notes accompanying the dashboard would be useful to help interpretation.

Conclusion

We conducted a small-scale public consultation exercise in York to inform the choice of priority indicators of health-care equity. The consultation was conducted through a 1-day citizens’ panel event in York and an online questionnaire, both of which aimed to gauge the strength of public concern for different types of socioeconomic inequality in health-care access and outcomes. The consultation showed that the public considered inequalities in health behaviours, such as healthy diet and healthy level of physical exercises, to be less unfair than inequalities in health-care access and outcomes. This indicated that, at least from a public perspective, the focus of NHS equity measurement should be on indicators that are amenable to changes in the way health care is organised and delivered rather than indicators of individual lifestyle behaviour; although, of course, the former may influence the latter. The consultation found that the general public perceived the following three socioeconomic inequalities in health care as particularly unfair: supply of GPs; hospital waiting time for surgery; and death due to causes preventable by good-quality health care. Our finding suggests that the public care about inequalities in both access and outcome, and that indicators of inequality in access to health care are an essential component of NHS equity monitoring for the purpose of public transparency and accountability.

The two lay members of our advisory group also provided useful feedback and advice on indicator selection and on the design of visualisation tools. As well as helping to improve our visualisation tools in various ways, one of the most helpful pieces of feedback was a negative lesson. It is clear that our one-page equity dashboard tool is appropriate for communicating to decision-makers, but it is not appropriate for public reporting. We therefore recommend further work on public reporting of our equity indicators, involving infographic design specialists and public relations experts.

Introduction

This chapter describes the iterative process through which we selected our eight main indicators of equity and our two disease-specific indicator domains (CHD and diabetes). We produced the following prototype equity indicators at both national and local levels: (1) primary care supply, (2) primary care quality, (3) hospital waiting time, (4) preventable hospitalisation, (5) repeat hospitalisation, (6) dying in hospital, (7) amenable mortality and (8) overall mortality. The first seven are indicators of equity in health care; the eighth is an indicator of equity in health that provides useful contextual information to place the other indicators into perspective.

The definitions of these indicators are provided elsewhere in the report:

• non-technical overviews of each indicator, summarising what they mean and why they are worth measuring are in Chapter 5

• short two-line indicator definitions are in Chapter 7

• full technical indicator definitions are in Appendix 4.

We also produced prototype equity indicators at the national level for our two disease-specific domains, which are described in Appendices 1 and 2.

The indicator selection process included (1) reviewing existing indicators used by the NHS to monitor health-care quality; (2) consulting with health indicator experts about technical feasibility; (3) consulting with a diverse range of NHS and public health experts about policy relevance through one-to-one conversations and an online survey; and (4) consulting with members of the public (see Chapter 2). Indicator selection decisions were made in consultation with our advisory group, the membership of which is listed in Appendix 3.

Based on the indicator review and consultation process, we developed (1) an indicator framework around different stages of the patient pathway, (2) a long list of potential general health-care equity indicators and (3) a shortlist of potential disease-specific indicator domains, together with examples of potential equity indicators within each shortlisted domain. In consultation with our advisory group, we then selected for prototype indicator production (1) a shortlist of general equity indicators and (2) two disease-specific indicator domains (CHD and diabetes). The final set of indicators and their detailed technical specifications were subsequently selected and refined in an iterative process of data analysis and reanalysis in response to feedback from a series of presentations of emerging findings to members of our advisory group and a range of other health-care and public health experts.

All our indicators have important limitations because of data limitations. We had to do the best we could with available data that can be broken down at the neighbourhood level, and our findings must be interpreted with due caution given the data limitations. We have, as part of our research recommendations, indicated that there is a need to develop better sources of data that will enable researchers to develop improved measures of health-care access and outcome that can be broken down at the neighbourhood level or, better still, individual level. The main limitations are summarised in the paragraph below; other limitations are described in the Appendix 4.

Our measure of primary care access, GPs per patient, focuses on GPs rather than practice nurses and other primary care professionals, and also focuses on ‘inputs’ rather than outputs and outcomes from the patient perspective, for example, how long it takes to book an appointment and whether or not consultations are sufficiently long and of high quality. Data on practice nurses are now becoming available at the practice level. However, the lack of a patient perspective is less straightforward to overcome using existing data, because GP patient survey data on patient experience of access to care may be subject to sample selection bias at the practice and neighbourhood level, as explained later. Another limitation is that our measure of primary care quality focuses on the clinical process quality indicators in the Quality and Outcomes Framework (QOF), which only capture some limited aspects of primary care quality. Finally, our measures of health-care outcome were only adjusted for age and sex, not other health risk factors that increase the risk of poor outcomes in deprived neighbourhoods but may lie outside the control of health services.

The rest of this chapter contains sections on the indicator selection criteria, the indicator review and the indicator consultation. It concludes with a section describing potential quasi-experiments and additional indicators that were suggested during the indicator selection process and that may be worth considering in future work.

Indicator selection criteria

Our indicator selection criteria are listed below, together with supplementary notes on how the criteria were assessed. There are four sets of criteria: (1) general criteria for each individual indicator, (2) technical criteria for each individual indicator, (3) criteria for selecting an appropriate mix of indicators within each domain and (4) criteria for selecting the two disease-specific indicator domains.

Indicator review process

We reviewed available indicators in order to (1) identify a full range of indicators that can potentially be used for monitoring average health-care quality in the NHS, (2) identify which of these indicators could feasibly be converted into equity indicators, based on our technical indicator selection criteria, (3) select candidate indicators based on our general indicator selection criteria and (4) select candidate disease-specific indicator domains based on our indicator domain selection criteria.

We started by reviewing all indicators currently used for monitoring average health-care quality in the NHS, as published on the Health and Social Care Information Centre (HSCIC) Indicator Portal (https://indicators.ic.nhs.uk/webview; accessed 12 July 2015). This web-based portal hosts a wide range of indicators, including all of the indicators in the NHS Outcomes Framework, the PHOF and the Adult Social Care Outcomes Framework. We then considered further indicators by (1) checking an unpublished list of indicators from an internal review by the HSCIC in 2011 and (2) seeking suggestions from advisory group members, co-applicants and colleagues. Finally, we generated proposals for new equity indicators based on our own knowledge of available health data sets in England and suggestions from consultees.

Indicator consultation process

As well as the public consultation process described in Chapter 2, the indicator consultation process included (1) one-to-one conversations with NHS and public health experts, (2) an online survey of NHS experts and (3) consultation with members of our advisory group (listed in Appendix 3). Our aim was to obtain a range of views from NHS and public health experts with appropriately diverse backgrounds and perspectives, rather than a representative national sample.

Conversations were held with a range of experts, including:

• academic experts in health inequality, primary care, mental health, circulatory disease, cancer, epidemiology, health geography, sociology of health and health economics

• analysts and health inequality experts at NHS England and the Department of Health, including analysts supporting the NHS Outcomes Framework

• analysts and health inequality experts at Public Health England

• board members of two CCGs (Vale of York and Hull)

• public health directors in two local authorities (City of York and Hull City Council)

• board members of two NHS hospital trusts (York and Morecombe Bay)

• public health, health policy and performance indicator experts at leading national think tanks (The King’s Fund and Nuffield Trust).

The conversations focused on the credibility and policy importance of our proposed equity indicators, the identification of potential new indicators, the visual communication of equity indicator findings and the perceived impacts of past NHS interventions on socioeconomic inequalities in health. The questions and topics were tailored to the type and role of the respondent, rather than following a ‘one-size-fits-all’ structured interview format.

The online survey was conducted to supplement these conversations, with a range of views focusing specifically on the perceived impacts of past NHS interventions on socioeconomic inequalities in health. This specific focus was chosen to help the research team select equity indicators that NHS experts consider to be sensitive to NHS health-care delivery and that provide a platform for future ‘natural experiment’ studies of the health inequality impact of NHS interventions. Accordingly, the online survey asked two main questions:

1. In the boxes below, please list up to THREE national or local NHS interventions in the past decade or so that you think had a measurable impact on socioeconomic inequalities in health-care access or outcomes in England.

2. For each NHS intervention, what primary outcome(s) would you use to measure impact on socioeconomic health inequality?

The survey was e-mailed to 164 senior individuals from diverse organisations and clinical specialties, with an e-mail addresses identified via the personal contacts of the research team and advisory group, and web searching. Invitees could respond directly by e-mail, by filling in a Microsoft Word^® form (2010, Microsoft Corporation, Redmond, WA, USA) or by filling in an online survey form via the web-based survey tool SmartSurvey (SmartSurvey Ltd, Tewkesbury, UK; www.smartsurvey.co.uk). One reminder e-mail was sent to non-responders. Only 14 response sets were received (a 8.5% response rate). This low response rate is about average for surveys of this kind, given the challenging and time-consuming nature of these open-ended questions and that we were seeking responses from busy, senior professionals. The only impact of this low response rate on our findings was that our list of potential quasi-experiments (see Potential quasi-experiments) is not exhaustive. The most important element of our consultation was the face-to-face discussions with stakeholders, as described above, which enabled us to gather more nuanced information than is possible using a survey about stakeholder views on the selection of indicators and the visual communication of equity findings.

Potential quasi-experiments and additional indicators

Introduction

This section describes the data and methods underpinning the indicators defined in this study. The section has the following structure. It starts by (1) describing the data sources used and then goes on to cover (2) data linkage and aggregation to construct indicators; next, it discusses (3) the data cleaning methods used on the indicators, before discussing (4) standardisation methods used to adjust indicators for need and risk factors; finally, it describes (5) the estimation of absolute and relative inequality indices at the national level and (6) the estimation of absolute and relative inequality indices at the local level. At each stage alternative approaches that were considered and sensitivity analyses performed are discussed.

Data sources

Linkage and aggregation of data

The GMS and QOF data sets (see General and Personal Medical Services and Hospital Episode Statistics, respectively) collect data at the GP practice level while our basic geographical unit for our analysis is the 2001 LSOA. The ADS details the LSOAs in which the patients registered with each GP practice live. We use this information to determine the proportions of the practice-level variables in GMS and QOF to attribute to each of the LSOAs that the patients registered with the practice live in. Applying this attribution calculation to each GP practice and then aggregating the practice-level variables attributed from the different practices at the LSOA level gives us our measures of these primary care supply and quality indicators at the 2001 LSOA level. Practice-level populations were used only to apportion these practice-level variables to 2001 LSOAs; the denominators used in the indicators derived at the LSOA level based on these variables were then derived using ONS data for the LSOA-level population estimates to maintain comparability with the other indicators.

The HES data set (see Hospital Episode Statistics) provides data at the FCE level; this describes a patient’s period of care under one consultant. We further aggregate this to the CIPS level, which groups together the entire hospital stay of the patient, including transfers between consultants and between hospitals. The CIPs are described in Lakhani et al. 52

Patient-level data from HES, ONS mortality data and ONS mid-year population data are then split by sex and into age groups for ages 0–4, 5–15, 16–24, 25–39, 40–59, 60–74 and ≥ 75 years before being aggregated into 2001 LSOAs. These age groups were selected to minimise the number of subgroups while still capturing key life stages and points at which policy interventions are typically targeted in England.

Although 2001 LSOAs were the basic small-area unit of analysis, our target large-area geographies for indicator production, in particular CCGs, were defined in terms of 2011 LSOA boundaries. While 96.3% of 2001 LSOAs were unchanged between the 2001 and 2011 LSOA boundaries (‘one-to-one’ mappings), the other 3.7% (1192 out of all 32,482 of the 2001 LSOAs) needed to be mapped between these alternative LSOA definitions. When two or more 2001 LSOAs were merged to form a single 2011 LSOA (‘many-to-one mappings’), these multiple 2001 LSOAs were straightforwardly aggregated to form results at the 2011 LSOA level. When a 2001 LSOA mapped to more than one 2011 LSOA (‘one-to-many’ and ‘many-to-many’ mappings), it was assumed that the 2001 LSOA was split in equal proportions when attributed to the 2011 LSOA. The mappings for 2.7% of small areas (881 out of 32,482 of the 2001 LSOAs) required splitting in this way. We used this algorithm to produce a set of weights to map results at the 2001 LSOA level to 2011 LSOA level. These weights were then applied to all our indicator results at the 2001 LSOA level before being aggregated to higher geographical levels.

The IMD 2010 overall deprivation rank scores defined at the 2001 LSOA level were attributed to 2011 LSOAs, which were then ranked according to attributed score. These integer ranks were then normalised to produce a fractional rank ‘ridit score’53 between 0 (least deprived) and 1 (most deprived). The IMD deprivation ranks at larger geographical levels were produced by population-weighted aggregation of the 2011 LSOA-level IMD deprivation ranks to higher levels of geography, with normalisation to produce a fractional rank between 0 and 1 at the target geographical level. The IMD quintile and decile groups were defined as aggregations of appropriate sets of deprivation ranked 2011 LSOAs.

National inequality indicators

The details of the specific indicator definitions and how these indicators were constructed are provided elsewhere in this report. In this section we describe how these indicators were used to calculate inequality indices at the national level.

The primary indicator we used in our analysis is our implementation of the SII as a measure of absolute inequality; and we also used the corresponding RII. We chose this indicator as an appropriate compromise between simple indicators that are easy for users to understand but potentially misleading, such as the gap between top and bottom groups, and more sophisticated indicators that are hard for users to interpret, such as the absolute concentration index. 53,59 The slope index captures the whole social gradient, rather than selecting two arbitrary groups for comparison. It also has a reasonably simple interpretation as the modelled health-care gap between the most and least deprived neighbourhoods, allowing for the social gradient in-between. This indicator is also already used routinely in England for monitoring inequalities in health by the PHOF, and has been proposed for routine use by the NHS Outcomes Framework for monitoring national inequalities in health-care outcomes. 60

Mathematically, the absolute concentration index is the slope index multiplied by twice the variance of the socioeconomic variable, as measured on its original raw scale. 59 Thus, the main limitation of the slope index compared with the absolute concentration index is that it uses only information on the fractional rank of the deprivation variable, and does not also take account of variance in the deprivation score measured in its original raw scale. This is an important limitation when cardinal measures of socioeconomic status are available, such as income. However, the IMD is only measured on an ordinal scale in the first place, and so converting this variable into a fractional rank does not throw away any important information.

To compute the SII we start by taking the standardised indicator data at LSOA level, cleaned to ensure data quality (see Data cleaning), and for each year of data we calculated an ordinary least squares regression of LSOA-level outcome variable against LSOA-level deprivation fractional rank (measured on a 0–1 scale as described in Linkage and aggregation of data). The coefficient on deprivation fractional rank in this regression gave us the SII for the year, and the standard error on this coefficient gave us the standard error around the SII. The SII can be interpreted as the modelled difference in event count between the least deprived and most deprived LSOAs in the country, taking into account the distribution of the outcome variable across the deprivation range. For all of our indicators, a positive SII value indicates pro-rich absolute inequality in the outcome, whereas a negative SII indicates a ‘pro-poor’ absolute inequality in the outcome. This is straightforward in almost all cases, as almost all of our outcomes are defined as undesirable events. As we have defined our deprivation score to run from 0 for least deprived to 1 for most deprived, a positive linear association thereby implies that more deprived small areas have worse outcomes. The one exception is primary care quality, which is a desirable outcome. In this case, we simply invert the SII by multiplying by –1.

Our approach differs a little from some of the other commonly used definitions of the SII, in which the indicators are first aggregated to deprivation decile group level and then the slope through the decile group points is calculated. 58 We felt that given the availability of the data at the LSOA level for these indicators we were better able to capture the within-decile variation and uncertainty in our SII estimate by using LSOA-level linear regressions of outcomes against LSOA deprivation rank as opposed to national decile-level regressions.

The SII is also used to calculate a ‘real inequity gap’, based on a counterfactual situation of full equality in which all neighbourhoods do as well as the least deprived neighbourhood in terms of modelled achievement on the indicator. The real inequity gap is measured in the same units as the indicator and is calculated as 0.5 × SII × population. This is depicted by the shaded area in Figure 5, which also shows the national average and deprivation decile average values of the indicator as well as the SII slope. We did this because of feedback from our consultation process that policy-makers and members of the public find it easier to understand and relate to ‘raw’ physical units (e.g. numbers of deaths) rather than rates and proportions, as explained in Chapter 3.

FIGURE 5.

Generic national absolute inequality graph showing the SII slope, inequality gap, national and decile averages for the indicator.

We also calculated national average levels of the outcome variables and inequality trends in these variables for display in the indicator dashboards, as the year-on-year difference between this year’s figure and last year’s figure, with colour coding to show whether or not the trend was significant. We found that a simple 1-year trend was easier for users to understand than a more complex trend involving more than 1 year; although, to guard against the risk of over-reacting to a possible 1-year ‘blip’, we advise that dashboard users also consult the in-depth graphical analysis underpinning the numbers reported in the dashboards, which show the year-on-year trend over several years.

As an example, the 2011/12 SII trend can be calculated simply as:

Alternative trend measures were also considered, including differences between multiyear moving averages, linear regressions of SII against time, and pooled time series cross-section regressions involving year dummies interacted with deprivation rank. However, the simple single-year change in SII was chosen for its simplicity and the fact that the value calculated as a trend in a particular year would be fixed and not be revised as further years of data become available.

We analogously calculated the same range of indicators on a relative scale by presenting our absolute inequality results as a proportion of the national average achievement for the indicator. As almost all of our indicators are measures of undesirable outcomes, these can be seen as relative shortfall indicators rather than relative attainment indicators. 61 A relative attainment indicator could be calculated by inverting the outcomes, for example calculating the number of people having emergency hospitalisations results in a relative shortfall indicator, while calculating the number of people not having emergency hospitalisations would result in a relative attainment indicator.

Although we fitted a linear model for our base-case results, we also tested the robustness of our conclusions to using alternative non-linear functional forms including log-linear or exponential models, Poisson and negative binomial models. These non-linear models still allow the computation of a slope index, by predicting outcomes for the most and least deprived neighbourhood and taking the gap. We were reassured to see that these alternative model specifications produced the same basic national inequality trends as the linear model, and very similar patterns at the CCG level. Measuring inequality is not solely a statistical exercise in finding the best-fitting model but is also partly a normative judgement. For example, consider the situation depicted in Figure 5 in which there is a clear non-linear ‘uptick’ in adverse outcomes within the two most deprived decile groups. This is the case, for example, for preventable hospitalisation and amenable mortality. In this case, a log-linear model would give a lower estimate of the predicted absolute gap than the linear model. It would also be less sensitive to health-care indicator improvements in the most deprived decile groups. This is because a log-linear model allows a closer fit to the non-linear ‘uptick’. In effect, it assumes that more deprived areas should have more than a linear increase in event rates. This is partly a normative judgement, implying that a non-linear ‘uptick’ in event rates in the most deprived end of the spectrum is normal and so NHS organisations should not be held to account for eliminating it. By contrast, the linear model gives greater weight to inequality at the two extremes of the distribution, and so is more sensitive to changes at the most deprived end of the spectrum. Mathematically, the ordinary least squares linear model yields a weighted average of the gradient between each point and the mean point, with greater weight given to points that lie further from the mean. Gradients towards the more extreme end of the spectrum thus receive higher weight, and hence the slope index is more sensitive to change the further one moves towards the tails of the deprivation spectrum. In practice, then, use of the linear slope index rather than a non-linear slope index embodies the value judgement that the index ought to be more sensitive to change towards the most and least deprived ends of the spectrum than change in the middle.

We chose to use the linear model for our base-case estimates of the SII as this is the simplest, easiest to understand, and most widely used form of the measure in the inequalities literature. When different kinds of non-linearity are evident in the different years of the data for the different indicators, these could each in theory be modelled using the best-fitting non-linear model for that particular instance. However, we found that the linear model provides a useful general method to measure inequality and adequately captures the inequality trends across indicators and over time in a comparable manner.

We also produced results using heteroscedasticity robust standard errors and were reassured to find that these had little effect on our estimated standard errors in both the national- and CCG-level analyses. We chose to use unadjusted standard errors, in the absence of compelling empirical evidence for heteroscedasticity at the national level or theoretical reasons for expecting heteroscedasticity within some subnational areas but not others.

We did not apply LSOA-level population weights either to the computation of the deprivation ridit score or to the linear regression in our base-case results. This again was on grounds of simplicity of communication to policy-makers, as LSOAs do not vary dramatically in population size and application of population weights made little difference to the results.

Finally, we did not allow for influential outliers at the CCG level using robust regression methods that apply an iterated reweighted least squares algorithm based on a particular weighting function. Rather, we propose that individual CCGs should visually inspect their own within-CCG small-area-level and practice-level scatterplots with a view to identifying and understanding the role of ‘unusual’ local neighbourhoods and GP practices on a case-by-case basis.

Local inequality indicators

We constructed CCG-level indicators using similar methods as we used to construct the national inequality indicators. Our absolute inequality indicator at the CCG level is based on running local-level regressions using just those LSOAs that fall within a CCG, and modelling the ‘social gradient’ relationship between the outcomes of these LSOAs and its national deprivation ranks. The deprivation rank we used was the national deprivation rank rather than recomputed local within-CCG deprivation rank. We did this in order to allow us to compare the within-CCG inequality gradient with the national inequality gradient in a straightforward manner. We labelled this indicator, the AGI, to distinguish it from variants of the SII at local level that use the local deprivation rank. We also calculated a relative version of this indicator at the CCG level, analogous to the national RII, that we called the relative gradient index (RGI). To maintain comparability with the national RII, this was computed as the AGI divided by the national mean level of the indicator. Dividing by the local mean would potentially bias comparisons against the national RII benchmark by decreasing measured local relative inequality in areas with higher-than-average mean levels of the indicator outcome (e.g. relatively deprived CCGs with above-average levels of preventable hospitalisation), and vice versa. National- and local-level results were graphically combined to compare the CCG with the national-level results as shown in Figure 6.

FIGURE 6.

Absolute inequality graph for a hypothetical CCG showing the AGI slope, inequality gap, within-CCG LSOA-level results, national and CCG average for an example indicator (e.g. preventable hospitalisation per 1000 general population).

This approach differs from the PHOF approach58 to calculating within-CCG SIIs. The PHOF approach uses local deprivation ranks recalculated in CCGs, and then deprivation decile-level regressions based on these local ranks. This difference is because of the difference in purpose between health-care equity indicators and public health equity indicators. The main difference is that our health-care equity indicators aim to compare local health-care inequalities against a national benchmark, whereas the public health equity indicators focus on comparing the same local area over time. For our purpose of making comparisons against a national benchmark, using a common deprivation scale between the national- and CCG-level indicators is appropriate. A second difference is that our indicators focus on role of the NHS in reducing the link between deprivation and ill health, rather than in reducing deprivation and income inequality per se. By contrast, the PHOF indicator seeks to pick up the success of local government both in reducing the deprivation–health link and also in reducing deprivation and income inequality per se. Our more specific focus is reasonable insofar as changes in local prosperity are largely caused by factors outside NHS control; although, of course, NHS actions can have consequences for people’s wealth by protecting them against catastrophic health-care costs and keeping them economically productive. To measure the deprivation–health link specifically, we need to use the absolute national deprivation rank rather than the relative within-CCG deprivation rank. In principle, our measure will then not be sensitive to ‘gradient-preserving’ changes in local economic prosperity if this leads to precisely corresponding changes in health and health care along the national social gradient.

We also plotted the full range of CCG-level inequality results against the national inequality result on a caterpillar plot, showing data for the most recent year to help us identify areas that performed significantly better or worse than the national average in terms of inequality. An example of such a plot is shown in Figure 7.

FIGURE 7.

The CCG-level caterpillar plot comparing absolute inequality at the CCG level in terms of AGI to absolute inequality at the national level in terms of SII.

As a final analytical tool we produced plots of CCG-level average achievement and inequality achievement by deprivation to get some understanding of the contributions between-CCG and within-CCG inequality to the national inequality results. An example of such a plot is shown in the Figure 8.

FIGURE 8.

Correlations between deprivation and average and inequality performance at the CCG level. (a) Mean; and (b) AGI.

As with the national inequality indicators, we tested a range of alternative regression models to ensure the robustness of our results. We also tested using a random-effects specification of our model with CCG-level random slopes and intercepts. We found that for those indicators in which we had small event counts at the CCG level (in particular, amenable mortality and all-cause mortality), the random-effects specification had trouble converging. However, for the indicators in which the random-effects specification did converge we found that the magnitudes of inequality results were, as would be expected, shrunk towards the national average. However, the trends and rankings of CCGs in terms of inequality remained very similar to those observed with the standard linear model. Our base-case results at local level are, therefore, produced using the standard linear model as (1) this could be applied in a consistent manner across the full suite of indicators and (2) this is a simpler approach that is easier for decision-makers to understand and interpret.

Introduction

This chapter presents the main results at both the national and local level. Detailed results for each of our eight indicators of equity are presented one by one, in the same format, and then a final section summarises our findings on local equity monitoring for all eight indicators.

Detailed results for each indicator follow a common format. An introductory section first explains what the indicator means, why it was selected and how it was defined, with special attention to any departures from standard NHS indicator definitions. The main results are then presented graphically, using the same four types of graph. The key findings are then summarised. The four types of graph are as follows: first, national equity time trend graphs showing trends in (1) indicator levels by quintile group of deprivation, (2) the SII and (3) the RII. We present annual time trends for our full period of 2001/2 to 2011/12 when possible, although this is not possible for primary care supply and quality whose data series only start in 2004/5 (as explained in Appendix 4). We first present unadjusted time trends and then adjusted time trends, after allowing for need or risk factors. Second, a national equity gradient graph for 2011/12 showing adjusted indicator levels by decile groups of deprivation, the social gradient (the slope of which is the SII), and the area under the social gradient, representing the inequity gap (see Chapter 4). Third, a local equity performance graph for 2011/12 in the form of a caterpillar plot showing equity (AGI) by CCG in 2011/12 in rank order, with confidence intervals. Fourth, a local performance deprivation correlation graph for 2011/12 in the form of a CCG-level scatterplot of performance against deprivation at the CCG level, for both equity performance (e.g. the AGI in patients per GP) and average performance (e.g. the average level of patients per GP).

We start by presenting contextual information on population trends from 2001/2 to 2011/12 by deprivation group, age group and sex, before turning to the results for our eight equity indicators.

Population

Equity is a population-level concept, relating to unfair inequality within the relevant population of interest. To help interpret trends in equity over time, it is therefore important to understand the nature of changes in population size and sociodemographic composition over time. Our equity indicators are based on ONS mid-year estimates of population, which estimate the total resident population including homeless people and people in institutions such as nursing homes, prisons and barracks. We present contextual information on national population trends by age, sex and deprivation group in the form of two matrix plots. The first plot, Figure 9, has a fixed population range on the y-axes, to facilitate comparisons between age groups (rows). The second plot, Figure 10, has variable population ranges on the y-axes, to facilitate comparisons between deprivation groups (columns).

FIGURE 9.

Population matrix showing breakdown by deprivation group, age group and sex (y-axes fixed for comparison across age groups).

FIGURE 10.

Population matrix showing breakdown by deprivation group, age group and sex (y-axes free for comparison across deprivation groups).

The population matrix plots show that (1) more deprived areas have younger populations than more-affluent areas and (2) old age populations have been increasing over time in the most affluent areas, while population at younger ages has been increasing over time in more deprived areas.

The bulk of the population is between 25 and 74 years of age. There has been substantial growth over the period in the populations aged 40–59 years and 60–74 years for all deprivation groups. This growth has been most pronounced in the more-affluent groups where we also see a substantial decline in the numbers aged 25–39 years. Women and men follow largely the same trends within each age and deprivation group. Men tend to outnumber women up until the 16–24 years age group, after which women outnumber men, with the sex difference becoming more noticeable in older age groups.

These deprivation-related demographic variations and trends have important implications for risk and need adjustment. The risk of adverse health-care outcomes (such as hospitalisation and mortality) tends to increase with age. The impact of deprivation-related demographic variation, that is point (1) above, will therefore be to increase the risk of adverse events in affluent areas, and hence to reduce the unadjusted socioeconomic gradient in health-care outcomes. Similarly, the impact of deprivation-related demographic change, that is point (2) above, will be to increase the relative risk of adverse events in affluent areas compared with deprived areas, and hence to reduce the unadjusted socioeconomic gradient in health-care outcomes over time. Without age adjustment, therefore, socioeconomic inequality in health outcomes may appear to reduce over time in a pro-poor direction, even if the NHS did nothing to improve inequality. This would be misleading as the resulting inequality would reflect the impact of demographic change rather than the impact of the NHS. Therefore, we considered it relevant to adjust for age in our analysis. In a cross-sectional analysis, adjusting for age increases the SII and RII for health-care outcomes, as more deprived areas are younger and hence less at risk of poor outcomes. In a time series analysis, adjusting for age adds a growth trend to SII and RII over time, by removing the impact of ageing in affluent areas on reducing the social gradient over time.

Age is, of course, not the only factor that influences the risk of adverse health-care outcomes. In particular, morbidity may have larger and potentially opposing impacts, as deprived populations are more sick than affluent populations and more at risk of adverse events. Unfortunately, however, we were not able to adjust for morbidity because of lack of time-varying individual-level data on both age and morbidity, as explained in Chapter 4. The morbidity-unadjusted gradient in health-care outcomes that we observe will therefore be substantially larger than the ‘true’ morbidity-adjusted gradient. Furthermore, there is some evidence that socioeconomic inequalities in morbidity have widened during the 2000s. 62 If so, this would have the impact of appearing to increase socioeconomic inequalities in the non-morbidity-adjusted health-care outcomes that we observe.

Primary care supply

Access to primary care is a foundation stone of health-care quality, and makes a crucial contribution both to patient experience and improvement in population health outcomes. There is evidence that improved access to primary care can help to prevent illness, manage chronic conditions more effectively and reduced unnecessary utilisation of secondary care,12,63 although evidence on the impact of marginal changes in primary care supply on mortality in high-income countries is mixed. 12,64,65 We use a simple and objective measure of access to primary care: the number of patients per primary care physician. Measures of patients’ subjective experiences of primary care access are also available, based on the annual National GP Patient Survey. However, this survey goes back only to 2006/7 and has a response rate of around 30%, which varies substantially between practices, which may hamper comparisons in social gradients between subnational areas.

Previous studies in high-income countries, including the UK, have found significant geographical variations in the distribution of primary care physicians. 66–72 Data from England between 1974 and 2006 showed substantial and persistent geographical inequalities in supply of GPs relative to need between NHS administrative areas. 73–75 However, these studies focused on large areas, which made it difficult to accurately pinpoint primary care shortages in specific disadvantaged neighbourhoods.

Our indicator of primary care supply evaluates socioeconomic inequality in GP supply between small-area populations from 2004/5 to 2013/14. We use LSOA-level data, which allows us to capture changing patterns of socioeconomic inequality in much finer detail than previous studies. We define GP supply as the number of patients per FTE GP, excluding registrars and retainers, adjusted for age, sex and neighbourhood ill health using the Carr-Hill workload adjustment (see Appendix 4 for details). The numerator is the total number of people alive at mid-point in the current financial year, while the denominator is the number of FTE GPs attributed to each small area in the current indicator year. Further technical details of how this index was computed are presented in Appendix 4.

Unadjusted trends (Figure 11) show that there has been a significant divergence in GP supply between the most deprived fifth of areas and the other areas in the country from 2006/07 onwards. As 2006/07 the most deprived fifth of areas experienced a sustained trend of increasing GP supply (decreasing numbers of patients per GP), while GP supply in all the other areas decreased over time. However, we prefer the need-adjusted findings because in cross-section the unadjusted findings come up with the potentially misleading message that people living in deprived neighbourhoods have substantially more GP supply than others. This is a potentially misleading finding, because it fails to allow for the fact that deprived neighbourhoods tend to suffer more ill health than affluent neighbourhoods, and so have greater health-care needs.

FIGURE 11.

Unadjusted equity time trends in patients per GP. (a) Indicator 1: primary care supply (unadjusted); (b) SII; and (c) RII. Notes: (1) the solid black line shows the most deprived quintile and the solid blue line shows the least deprived quintile; (2) negative SII indicates pro-poor distribution, while a positive SII indicates a pro-rich distribution in absolute terms; and (3) negative RII indicates pro-poor distribution, while a positive RII indicates a pro-rich distribution in relative terms.

Adjusting these results for need using the Carr-Hill workload adjustment changes the levels of these lines, but we see a similar equity trend in Figure 12. We see a sustained reduction in both absolute and relative inequality, as measured by the SII and RII over the period, and by 2010/11 need-adjusted GP supply actually becomes pro-poor. This is also evident in the social gradient graph for 2011/12 shown in Figure 13, in which we see the lowest numbers of patients per GP in the most deprived areas and a negative inequity gap. The caterpillar plot in Figure 14 shows that there are substantial numbers of areas significantly more and less equal than the mean. The correlation plot in Figure 15 shows that by 2011/12 there is little evidence of a social gradient between CCGs: there is no association between mean patients per GP and deprivation at the CCG level. By contrast, there is some evidence that more deprived CCGs do better at reducing deprivation-related inequality in GP supply within their own patch; there is a clear, although weak, negative association between equity in patients per GP (AGI) and deprivation at the CCG level.

FIGURE 12.

Adjusted equity time trends in patients per GP. (a) Indicator 1: primary care supply; (b) SII; and (c) RII. Notes: (1) the solid black line shows the most deprived quintile and the solid blue line shows the least deprived quintile; (2) negative SII indicates pro-poor distribution, while a positive SII indicates a pro-rich distribution in absolute terms; and (3) negative RII indicates pro-poor distribution, while a positive RII indicates a pro-rich distribution in relative terms.

FIGURE 13.

National social gradient in patients per GP in 2011/12: adjusted. Notes: (1) dots represent decile groups, the inverted U-shape pattern indicates that neighbourhoods in the middle of the socioeconomic spectrum have less primary care physician supply than the most and least deprived neighbourhoods, after adjusting for differences in need; (2) the slope of the line is the SII. In this case, the slope is negative showing pro-poor inequality in patients per GP favouring deprived areas; and (3) the shaded area shows the inequity gap. In this case, this gap is negative indicating that bringing all neighbourhoods to the level of the least deprived would require losing some GPs in deprived neighbourhoods.

FIGURE 14.

Caterpillar plot of the AGI of inequality in patients per GP in 2011/12 at the CCG level. Notes: (1) CCGs are ranked from least equitable (left) to most equitable (right); (2) the dotted horizontal line shows the national average, CCGs to the left with confidence intervals above this line have worse than average equity performance, and vice versa; and (3) In this unusual case, there are many negative SIIs (at face value indicating pro-poor inequality) as well as positive SIIs indicating pro-rich inequality. However, as we underestimate need in deprived neighbourhoods, as explained in Chapters 4 and 6 and Appendix 4, we do not interpret negative SIIs as representing pro-poor inequality but rather as indicating no measurable pro-rich inequality.

FIGURE 15.

Scatterplots of CCG performance on patients per GP in 2011/12 against deprivation, showing both mean performance and equity performance (AGI). (a) Mean; and (b) AGI.

When reading these results it is important to keep in mind that GP supply is just one aspect of primary care access. Just because a patient is registered with a practice that has a relatively low number of patients per GP, this does not guarantee the patient will have relatively good access to adequately long primary care consultations at a convenient time, or that they will receive relatively high-quality care during their consultations.

Primary care quality

Primary care remains the most effective and cost-effective way of delivering accessible care in a time of rising prevalence of chronic conditions and multimorbidity. 76 This indicator focuses on clinical process indicators of the quality of primary care, based on the proportion of patients diagnosed with a particular condition receiving appropriate medical care for that condition. The quality of primary care can be measured using structures (such as the supply of GPs), processes (such as vaccination, investigation and prescribing) or outcomes (such as mortality, morbidity and patient satisfaction). 77 Combinations of these measures have been used in the literature to assess contribution of primary care in improving population health. 12 International studies have demonstrated that improving the process quality of primary care is associated with reduced emergency admissions, improved patient outcomes and reduced costs to the health-care system. 78–80 Moreover, small improvements in primary care process quality can have significant effects on population health at low cost. 81

Improving the quality of primary care has been incentivised in several countries, primarily using financial incentives. 82 In the UK, the QOF scheme was launched in 2004 to monitor and improve the quality of primary care. 83 This is one of the largest pay-for-performance programmes in the world with over £10B invested since inception. 65 The QOF programme rewards primary care practices based on their performance on a comprehensive set of indicators that measure primary care quality. In total, 146 indicators are used that cover the management of chronic disease, public health measures, quality and productivity of service, and patients’ experiences with respect to care. 84 To measure the public health impact of primary care quality, Ashworth et al. 85 developed a composite indicator using 20 QOF indicators weighted by their importance in terms of their potential for mortality reduction. 85 The resulting measure, termed the ‘Public Health Impact’ score, was proposed as a measure of primary care quality in terms of population health. We recognise that the QOF only captures part of primary care quality, and that there are valid concerns with using QOF as a measure of primary care quality. 86–88 We therefore advise that this indicator be interpreted cautiously, in the light of these limitations.

Our indicator evaluates socioeconomic inequality in primary care quality between small-area populations from 2004/5 to 2013/14. We selected 16 out of the 20 indicators proposed by Ashworth et al.,85 for which data were available throughout our period of analysis in a consistent format. We define primary care quality as weighted average of clinical process quality from 16 indicators in the QOF, with weights proportional to importance in terms of the estimated number of lives saved per 100,000 patients. 85 For each clinical indicator in QOF, the number of patients deemed appropriate for that indicator is the denominator and the number of patients for whom the indicator was met is the numerator. We use ‘population achievement’ which puts exception-reported patients back into the population denominator thereby assuming such patients represent poor quality, but as a robustness check we also analysed ‘reported achievement’ which excludes exception-reported patients from the population denominator. Further technical details of how this index was computed are presented in Appendix 4.

There are no ‘adjusted’ results to present for this indicator, because the population denominator for each indicator already defines the ‘at-risk’ patient population as patients diagnosed with the relevant condition. No further risk adjustment was performed, on the basis of the value judgement that age, sex and other patient characteristics are not legitimate reasons for failing to deliver high-quality care to the ‘at-risk’ patient population.

Primary care quality in terms of QOF steadily improved over the study period as shown in Figure 16. Both absolute and relative inequality in this measure of primary care quality steadily decreased, and was almost eliminated by 2011/12 as shown in Figure 17. The caterpillar plot in Figure 18 shows that in 2011/12 there are substantial numbers of CCGs significantly more and less equal to the national level of absolute inequality. The correlation plots in Figure 19 show that by 2011/12 there is no sign of a social gradient in this primary care quality measure between CCGs, if anything, more deprived CCGs tend to have slightly better GP supply (fewer patients per GP), and that CCG equity performance on primary care quality, as measured by QOF, is not associated with deprivation at the CCG level.

FIGURE 16.

National equity trends in primary care quality: adjustment not necessary. (a) Indicator 2, primary care quality; (b) SII; and (c) RII. Notes: (1) the solid black line shows the most deprived quintile and the solid blue line shows the least deprived quintile; (2) negative SII indicates pro-poor distribution, while a positive SII indicates a pro-rich distribution in absolute terms; (3) negative RII indicates pro-poor distribution, while a positive RII indicates a pro-rich distribution in relative terms; and (4) data on the number of exception-reported patients were not provided prior to 2005/06, which explains the blip in the trends between 2004/05 and 2005/06. This blip disappears when using ‘reported achievement’ after excluding exception-reported patients.

FIGURE 17.

National social gradient in primary care quality in 2011/12: adjustment not necessary. Notes: (1) dots represent decile groups; (2) the slope of the line is the SII; and (3) the shaded area shows the inequity gap.

FIGURE 18.

Caterpillar plot of the AGI of inequality in primary care quality in 2011/12 at the CCG level. Notes: (1) CCGs are ranked from least equitable (left) to most equitable (right); and (2) the dotted horizontal line shows the national average. CCGs to the left with confidence intervals above this line have worse than average equity performance, and vice versa.

FIGURE 19.

Scatterplots of CCG performance on primary care quality in 2011/12 against deprivation, showing both mean performance and equity performance (AGI). (a) Mean; and (b) AGI.

Hospital waiting time

Hospital waiting time is a major health policy issue in many countries, including the UK, and an important indicator of health system performance. 89,90 Moreover, this indicator was identified by our citizens’ panel participants as an important measure of equity in the NHS (see Chapter 3). Prolonged hospital waiting time is known to be associated with poor health outcomes, increased risk of complications, reduced quality of life and high patient dissatisfaction. For example, a systematic review of waiting time for radiotherapy found that the risk of local recurrence of cancer increased with increasing waiting time. 91 In another example, a recent English NHS study found that waiting time for hip and knee replacement surgery had a statistically significant negative impact on the health gains from surgery. 92 Similar evidence on the impact of hospital waiting time has been found for other conditions, including chronic pain,93 cataract94 and heart transplantation. 95

We measure hospital waiting time in terms of days from outpatient decision-to-treat to inpatient admission-for-treatment. This is often termed the inpatient waiting time in the literature. Another commonly used indicator is the outpatient waiting time, defined as the period between referral from a GP to the outpatient appointment with a specialist. A third and more comprehensive indicator used in the NHS since the late 2000s is the referral-to-treatment waiting time, which measures the time from referral from a GP to inpatient admission-for-treatment, including adjustment to allow for ‘clock stop’ periods of waiting attributable to patient choices (e.g. not attending an appointment) rather than NHS supply. This can be further divided into admitted and non-admitted waiting times, by distinguishing patients who are admitted for inpatient treatment from patients whose course of treatment ends at the outpatient stage without requiring inpatient admission.

However, we focus on inpatient waiting time because it is considerably quicker and easier to compute, and less subject to bias due to coding and linkage error. Computing referral-to-treatment times can be done by linking outpatient and inpatient HES at individual level across multiple years and has been done for the particular case of hip and knee replacement. 96 However, this is time-consuming in terms of both coding time and computational time, has never previously been done across all possible procedures and specialties, and would be subject to an unknown degree of coding bias and selection bias due to linkage failures. It would also be impossible, using HES data, to fully implement the complex ‘clock stop’ rules required to replicate official NHS statistics on referral-to-treatment times; and so the resulting indicator would still not precisely match official NHS statistics. The use of inpatient waiting time is also more internationally comparable, and is consistent with the definition of waiting time used in most OECD countries to measure health-system performance. 89

By all measures, average hospital waiting times have declined significantly from 2001/2 in England. 97,98 Siciliani et al. 89 report that waiting times for many procedures have more than halved, and that this can partly be attributed to the ‘targets and terror’ policy introduced from 2000 as well as increased capacity. 99,100 However, there is evidence to suggest that there remains significant socioeconomic inequalities in waiting time. For instance, Laudicella et al. 101 found that elective hip replacement patients in the poorest two socioeconomic quintiles wait about 7% longer than patients in the least deprived quintile across England. In another study, Moscelli et al. 102 found significant differences in waiting times between public hospitals in non-emergency heart revascularisation procedures in England (up to 35% difference between the most and least deprived population quintiles). These inequalities all arose within hospitals rather than across hospitals, and after allowing for differences in the number and type of diagnoses as a marker for severity.

Our indicator evaluates socioeconomic inequality in inpatient hospital waiting time between small-area populations from 2001/2 to 2013/14. We define hospital waiting time as the number of days from outpatient decision-to-treat to inpatient admission-for-treatment (i.e. the inpatient waiting time). We allow for differences in waiting times by specialty type by adjusting for the main specialty of the treating consultant. We do not additionally allow for age and sex, on the basis of the value judgement that (at least in most cases) age and sex are not a legitimate justification for making people wait longer for needed treatment. Further technical details of how this index was computed are presented in Appendix 4.

Inpatient hospital waiting time decreased substantially across all areas in the country from 2003/4 to 2008/9, after which it began to creep up again, as shown in Figures 20 and 21. Inequality appeared to be slightly pro-poor at the start of our period in 2001/2, and to become even more pro-poor up to 2003/04, after which time more affluent areas steadily started to catch up (Figures 22 and 23), with some evidence of pro-rich inequality emerging by 2011/12, as depicted in the social gradient graph in Figure 24. The caterpillar plot in Figure 25 shows that substantial numbers of CCGs are performing significantly better and worse than the national average in terms of the AGI of inequality. The correlation plots in Figure 26 show no sign of correlation between mean waiting time and inequality in mean waiting time and CCG-level deprivation.

FIGURE 20.

Matrix plot showing unadjusted trends in inpatient hospital waiting time by age, sex and deprivation (fixed axes for comparisons across age groups).

FIGURE 21.

Matrix plot showing unadjusted trends in inpatient hospital waiting time by age, sex and deprivation (free axes for comparisons across deprivation groups).

FIGURE 22.

Unadjusted national equity trends in inpatient hospital waiting time. Notes: (1) the solid black line shows the most deprived quintile and the solid blue line shows the least deprived quintile; (2) negative SII indicates pro-poor distribution, while a positive SII indicates a pro-rich distribution in absolute terms; and (3) negative RII indicates pro-poor distribution, while a positive RII indicates a pro-rich distribution in relative terms.

FIGURE 23.

Adjusted national equity trends in inpatient hospital waiting time. Notes: (1) the solid black line shows the most deprived quintile and the solid blue line shows the least deprived quintile; (2) negative SII indicates pro-poor distribution, while a positive SII indicates a pro-rich distribution in absolute terms; and (3) negative RII indicates pro-poor distribution, while a positive RII indicates a pro-rich distribution in relative terms.

FIGURE 24.

National social gradient in inpatient hospital waiting time in 2011/12. Notes: (1) dots represent decile groups; (2) the slope of the line is the SII; and (3) the shaded area shows the inequity gap.

FIGURE 25.

Caterpillar plot of AGI of inequality in inpatient hospital waiting time in 2011/12 at the CCG level. Notes: (1) CCGs are ranked from least equitable (left) to most equitable (right); and (2) the dotted horizontal line shows the national average. CCGs to the left with confidence intervals above this line have worse than average equity performance, and vice versa.

FIGURE 26.

Scatterplots of CCG performance on inpatient hospital waiting time in 2011/12 against deprivation, showing both mean performance and equity performance (AGI). (a) Mean; and (b) AGI.

Preventable hospitalisation

Preventable hospitalisation refers to emergency hospital admissions that can be prevented by timely and effective provision of primary care. This is an important indicator of primary care access and quality that is widely used in the international literature. 103–105 In England, data from 2001 to 2013 showed that preventable hospitalisations make up one in every five hospital admissions, and have increased by 48% in the last 12 years. 106 Common causes of preventable hospitalisations include urinary tract infection/pyelonephritis, pneumonia, chronic obstructive pulmonary disease (COPD), convulsions and epilepsy, and ear, nose and throat infections. Studies suggest that preventable hospitalisation can be reduced by improving primary care supply and quality. 107 These hospital visits not only result in poor outcomes, but also result in increased cost to the health-care system. 108 For instance, a recent study concluded that better management of patients in primary care could save £1.42B in England by reducing preventable hospitalisation. 109 Similar cost estimates have been published for other countries. 110,111 Studies have also found that preventable hospitalisations are associated with the socioeconomic status of patients. 112,113

Our indicator evaluates socioeconomic inequality in preventable hospitalisation between small-area populations from 2001/2 to 2013/14. We defined preventable hospitalisation as the proportion of people with an emergency admission for a chronic ambulatory care-sensitive condition, admissions that are potentially avoidable if these chronic conditions are appropriately managed in primary care. 114 This indicator could also be described as ‘emergency hospitalisation sensitive to primary care’. We depart from the corresponding NHS Outcomes Framework definition by defining the indicator numerator as the number of people with one or more events, rather than the number of events. This is because (1) we have a separate measure of repeat hospitalisation and so want to focus this measure on the incidence of hospitalisation (the proportion of people hospitalised) rather than the intensity (how many times each individual is hospitalised); and (2) following advice from the two lay members of our advisory group, we believe that members of the public find it slightly easier to understand and relate to proportions (e.g. ‘X people per 1000’ or ‘a chance of X in 100’) than event rates. We focused on chronic rather than acute ambulatory care-sensitive conditions, as the former are likely to be more sensitive to changes in primary care supply and quality. We used the same list of chronic ambulatory care-sensitive conditions as the NHS Outcomes Framework (indicator 2.3i). Our definition of preventable hospitalisation uses all ages in both numerator and denominator, as does the NHS Outcomes Framework definition. However, the international OECD definition only includes age ≥ 15 years, that is, we include children but the OECD definition does not. We then indirectly standardised each year of data for age and sex at the LSOA level. Further technical details of the standardisation procedure are in Chapter 4, and further indicator definition details are presented in Appendix 4.

There has been a slight fall in preventable hospitalisation over the study period, although substantial inequality persisted throughout. Looking at the age–sex breakdowns in the matrix plot in Figures 27 and 28, the main exception to this trend was in children aged 5–15 years within the most deprived quintile group for whom preventable hospitalisation rose during the 2000s. The unadjusted trends in Figure 29 show improvement in inequality in terms of SII and RII. However, this is misleading because of the disproportionate ageing of the affluent population, which is associated with a higher rate of hospitalisation in this quintile group. After age adjustment, the pro-rich trend disappears for both SII and RII, as shown in Figure 30. This inequality is seen both between CCGs and within CCGs, as depicted by the correlation plots in Figure 31. Inequality lines up closely with deprivation, as shown by decile points on the scatterplot Figure 32 which all lie along the social gradient line. The caterpillar plot in Figure 33 shows there are substantial numbers of CCGs performing significantly better and worse than the national average in terms of the AGI of inequality.

FIGURE 27.

Matrix plot showing unadjusted trends in preventable hospitalisation by age, sex and deprivation (fixed axes for comparisons across age groups).

FIGURE 28.

Matrix plot showing unadjusted trends in preventable hospitalisation by age, sex and deprivation (free axes for comparisons across deprivation groups).

FIGURE 29.

Unadjusted national equity trends in preventable hospitalisation. Notes: (1) the solid black line shows the most deprived quintile and the solid blue line shows the least deprived quintile; (2) negative SII indicates pro-poor distribution, while a positive SII indicates a pro-rich distribution in absolute terms; and (3) negative RII indicates pro-poor distribution, while a positive RII indicates a pro-rich distribution in relative terms.

FIGURE 30.

Adjusted national equity trends in preventable hospitalisation. Notes: (1) the solid black line shows the most deprived quintile and the solid blue line shows the least deprived quintile; (2) negative SII indicates pro-poor distribution, while a positive SII indicates a pro-rich distribution in absolute terms; and (3) negative RII indicates pro-poor distribution, while a positive RII indicates a pro-rich distribution in relative terms.

FIGURE 31.

Scatterplots of CCG performance on preventable hospitalisation in 2011/12 against deprivation, showing both mean performance and equity performance (AGI). (a) Mean; and (b) AGI.

FIGURE 32.

National social gradient in preventable hospitalisation in 2011/12. Notes: (1) dots represent decile groups; (2) the slope of the line is the SII; and (3) the shaded area shows the inequity gap.

FIGURE 33.

Caterpillar plot of the AGI of inequality in preventable hospitalisation in 2011/12 at the CCG level. Notes: (1) CCGs are ranked from least equitable (left) to most equitable (right); and (2) the dotted horizontal line shows the national average. CCGs to the left with confidence intervals above this line have worse than average equity performance, and vice versa.

In the unadjusted trends, which do not allow for age and sex, both the SII and RII decline (get better) over time. This difference compared with the adjusted trends is because of demographic change over time: affluent neighbourhoods aged during the 2000s, while there was an increase in younger populations in deprived neighbourhoods. This demographic shift increased preventable hospitalisation in richer neighbourhoods relative to poorer neighbourhoods and hence reduced pro-rich inequality in the unadjusted trends. We think the age- and sex-adjusted trends give a more accurate picture of NHS equity performance, on the basis that the NHS should not receive credit for an apparent reduction in pro-rich inequality resulting from demographic change largely outside the control of the NHS.

A final point to note is the uptick in preventable hospitalisation in 2003/4, which was particularly strong in the two most deprived quintile groups. The cause of this is not known. However, one speculation is that this may be related to change in the supply of GP out of hours care. This uptick in preventable emergency hospitalisation happened around the time of the introduction of the new GP contract which, among other things, allowed GPs to opt out of providing ‘out-of-hours’ cover for emergency care outside normal GP practice working hours. This speculation may be worth exploring in future quasi-experimental studies.

Repeat hospitalisation

Repeat emergency hospitalisation is known as an important routine indicator of health system performance. 115–118 Repeat hospitalisation may be a result of one or a combination of several factors including (but not limited to) quality of care during previous hospitalisation (including early discharge), comprehensive discharge planning,119 primary and community care after discharge (including outpatient follow-up)120 and the patients’ own social support systems and health behaviours. Therefore, repeat hospitalisation is an important indicator of the quality-of-care co-ordination between hospital care, primary care and community care settings.

Studies suggest that greater deprivation is associated with an increased risk of emergency readmission. For instance, in a study in Greater Manchester, Lyratzopoulos et al. 121 found that deprivation was significantly and independently associated with increased risk of emergency medical readmission at 3 and 12 months after initial discharge. Other studies using specific patient groups found similar socioeconomic patterns of hospital readmissions. 122 Repeat hospitalisation not only results in poor health outcomes for patients, it also significantly increases the cost of care for the health-care system. Therefore, reducing repeat hospitalisation is one of the key indicators used to assess hospital performance and the impact of health service organisation for the average patient. 116,123

Our indicator evaluates socioeconomic inequality in repeat hospitalisation between small-area populations from 2001/2 to 2013/14. We define repeat hospitalisation as a proportion of inpatients with one or more subsequent any-cause emergency readmission in the same year. We focus on all-cause rather than cause-specific repeat hospitalisation and on within-year rather than 30-day or 90-day repeat hospitalisation for the following reasons: (1) we are interested in whole-system co-ordinated care, beyond the primary cause of hospital admission and the immediate post-hospital period; and (2) all-cause repeat hospitalisation within the indicator year provides a larger number of events for the purpose of detecting statistically significant differences between CCG-level and national-level absolute inequality gradients.

The denominator for this indicator is the total number of people with an inpatient admission from any cause in a given year. The numerator is the number of people with one or more repeat hospitalisations from any cause in the same calendar year. We used repeat hospitalisation within the indicator year rather than following patients across years because this is less time-consuming in terms of coding and computational burden. In addition, 12-month readmission would result in a less up-to-date indicator by either imposing a 1-year data lag or a focus on patients admitted the year before the indicator year. The drawback of our approach is that it may produce biased estimates of the national social gradient in 12-month readmission, although this is unlikely substantially to hamper comparisons between CCGs and over time. The advantage is that this is a simpler, less computationally expensive and timelier approach. We indirectly standardised each year of data for age and sex at the LSOA level. Further technical details of how this index was computed are presented in Appendix 4.

Rates of repeat hospitalisation have increased substantially over the study period as shown in Figures 34 and 35, coupled with an increase in both absolute and relative inequality in Figures 36 and 37. By 2011/12 there was substantial inequality in repeat hospitalisation, as depicted by the national social gradient graph in Figure 38. The unadjusted trends in Figure 36 show improvement in inequality; however, this is misleading because of disproportionate ageing of the affluent population over time. After age adjustment, the underlying worsening inequality trend becomes clear for both SII and RII, as shown in Figure 37. As discussed in more detail in Chapter 8, the increase in repeat hospitalisation and associated increase in inequality may partly be a sign of success related to increasing multimorbidity as a result of people living longer, although may also partly be a result of shorter lengths of stay in hospital, hospital payment reforms that gave hospitals financial incentives to increase emergency admissions, and, especially towards the end of the 2000s, reductions in social care supply and quality because of financial pressures on local authorities. This inequality appears to be present between CCGs as well as within CCGs, as shown in Figure 39a for 2011/12. Equity performance on repeat hospitalisation shows a slightly positive association with deprivation at the CCG level, although most of this association is driven by a handful of CCGs with unusually high and low equity performance. The caterpillar plot in Figure 40 shows there are substantial numbers of CCGs performing significantly better and worse than the national average in terms of the AGI of inequality.

FIGURE 34.

Matrix plot showing unadjusted trends in repeat hospitalisation by age, sex and deprivation (fixed axes for comparisons across age groups).

FIGURE 35.

Matrix plot showing unadjusted trends in repeat hospitalisation by age, sex and deprivation (free axes for comparisons across deprivation groups).

FIGURE 36.

Unadjusted national equity time trends in repeat hospitalisation. Notes: (1) the solid black line shows the most deprived quintile and the solid blue line shows the least deprived quintile; (2) negative SII indicates pro-poor distribution, while a positive SII indicates a pro-rich distribution in absolute terms; and (3) negative RII indicates pro-poor distribution, while a positive RII indicates a pro-rich distribution in relative terms.

FIGURE 37.

Adjusted national equity time trends in repeat hospitalisation. Notes: (1) the solid black line shows the most deprived quintile and the solid blue line shows the least deprived quintile; (2) negative SII indicates pro-poor distribution, while a positive SII indicates a pro-rich distribution in absolute terms; and (3) negative RII indicates pro-poor distribution, while a positive RII indicates a pro-rich distribution in relative terms.

FIGURE 38.

National social gradient in repeat hospitalisation in 2011/12. Notes: (1) dots represent decile groups; (2) the slope of the line is the SII; and (3) the shaded area shows the inequity gap.

FIGURE 39.

Scatterplots of CCG performance on repeat hospitalisation in 2011/12 against deprivation, showing both mean performance and equity performance (AGI). (a) Mean; and (b) AGI.

FIGURE 40.

Caterpillar plot of AGI of inequality in repeat hospitalisation in 2011/12 at the CCG level. Notes: (1) CCGs are ranked from least equitable (left) to most equitable (right); and (2) the dotted horizontal line shows the national average. CCGs to the left with confidence intervals above this line have worse than average equity performance, and vice versa.

Dying in hospital

Dying in hospital is a useful indicator of the quality of end-of-life care planning. 124 It is directly related to care pathway planning and the availability of palliative care and community nursing care at home. 125,126 Moreover, socioeconomic factors including low levels of social support are associated with increased hospital deaths. 125 A recent systematic review of studies on adult preferences for place of death found that most people prefer to die at home. 127 Specifically, in England, 64% of respondents of a population survey stated that they preferred to die at home rather than in hospital. 128 However, almost half of the deaths in England occur in hospitals (47.2% in 2014/15). 129 This may be because of a combination of factors, involving health and social services and individual and social factors. Dying in hospital is also a financial concern because of high expenditure of end-of-life care in hospitals.

In keeping with our other indicators, we use a general version of this indicator that covers the full spectrum of health-care activity, rather than focusing exclusively on cancer or other specific terminal illnesses. Our indicator measures socioeconomic inequality between small-area populations in the proportion of all deaths that occur in hospital. The numerator for this indicator is the number of deaths from any cause that occurred in hospital in a given year, measured using HES data. The denominator is the total number of deaths from any cause in a given year, irrespective of the place of death, measured using ONS mortality data. The indicator was measured for years 2001/2 to 2011/12. Further technical details of how this index was computed are presented in Appendix 4.

There are no ‘adjusted’ results to present for this indicator, on the basis of the value judgement that age, sex and other patient characteristics are not legitimate reasons for differential rates of deaths in hospital.

The proportion of deaths in hospital has been falling over time in all deprivation groups as shown in Figures 41–43. However, substantial inequality between deprivation groups persists over time with those living in more deprived areas more likely to die in hospital (Figure 44). The correlation plot in Figure 45 shows that this inequality occurs between CCGs, in that more deprived CCGs have higher rates of dying in hospital. However, there is no obvious correlation between the absolute equity gradient within CCGs and deprivation at the CCG level. The caterpillar plot in Figure 46 shows that very few CCGs are statistically distinguishable from the national mean in terms of their absolute gradient in inequality.

FIGURE 41.

Matrix plot showing unadjusted trends in dying in hospital by age, sex and deprivation (fixed axes for comparisons across age groups).

FIGURE 42.

Matrix plot showing unadjusted trends in dying in hospital by age, sex and deprivation (free axes for comparisons across deprivation groups).

FIGURE 43.

National equity trends in dying in hospital – adjustment not necessary. Notes: (1) the solid black line shows the most deprived quintile and the solid blue line shows the least deprived quintile; (2) negative SII indicates pro-poor distribution, while a positive SII indicates a pro-rich distribution in absolute terms; and (3) negative RII indicates pro-poor distribution, while a positive RII indicates a pro-rich distribution in relative terms.

FIGURE 44.

National social gradient in dying in hospital in 2011/12. Notes: (1) dots represent decile groups; (2) the slope of the line is the SII; and (3) the shaded area shows the inequity gap.

FIGURE 45.

Scatterplots of CCG performance on dying in hospital in 2011/12 against deprivation, showing both mean performance and equity performance (AGI). (a) Mean; and (b) AGI.

FIGURE 46.

Caterpillar plot of the AGI of inequality in dying in hospital in 2011/12 at the CCG level. Notes: (1) CCGs are ranked from least equitable (left) to most equitable (right); and (2) the dotted horizontal line shows the national average. CCGs to the left with confidence intervals above this line have worse than average equity performance, and vice versa.

Amenable mortality

Amenable mortality is a standard indicator used internationally to monitor the performance of the health-care system as a whole, and is considered to be particularly useful for monitoring the performance of primary care and the co-ordination of care between primary and secondary services. 58,130,131 Amenable mortality refers to deaths that could be avoided by the health-care system through prevention and treatment, given medical knowledge and technology available at the time of death. 132,133 The concept was first formalised by Rutstein et al. ,134 based on treatable causes of death, and subsequently broadened to include causes preventable by health care which led to the use of the term ‘amenable mortality’. 135,136 The concept of ‘amenable mortality’ is narrower than that of ‘preventable mortality’, however, which also includes mortality preventable by public health measures outside the health-care system. 137

Amenable mortality makes up a significant proportion of total deaths, even in high-income countries. Nolte and McKee138 found that, in 2006/7, amenable mortality accounted for nearly one-quarter of all deaths among people aged < 75 years in 16 high-income countries, including the UK, where the figure was slightly above average, at 26.8%. The relationship between amenable mortality and socioeconomic status has also been investigated in several studies. 139–142 In England, socioeconomic inequality in amenable mortality increased among both men and women between 1990 and 2010; the RII for men increased from 2.21 in 1990 to 2.83 in 2010, and for women increased from 1.67 in 1990 to 2.18 in 2010. 143

This indicator measures socioeconomic inequality between small-area populations in amenable mortality. We defined amenable mortality as the proportion of people dying from causes considered amenable to health care. The numerator for this indicator is the number of deaths from causes considered amenable to health care. The denominator is the total number of deaths from any cause in a given year. We used the list of causes of death considered amenable to health care from the NHS Outcomes Framework (indicator 1.1),144 which in turn is based on a list produced by the ONS. The NHS Outcomes Framework turns the resulting mortality counts into an estimate of ‘potential years of life lost’ from premature deaths aged < 75 years. However, we have used a simple all-age mortality rate including deaths in those aged ≥ 75 years, as (1) our approach is more comprehensive (people aged > 75 years experience by far the highest rate of amenable mortality) and (2) based on advice from two lay members of our advisory group and a media expert, we believe that mortality rates are easier for the public to understand than ‘potential years of life lost’. We indirectly standardise amenable mortality for age and sex at the LSOA level. Further technical details of how this index was computed are presented in Appendix 4.

Amenable mortality has fallen in all deprivation groups over time, with some sign of an accelerated decline from 2004/5 when the primary care pay-for-performance contract was implemented (see Figures 47–50). Once adjusted for age and sex, there is a clear reduction in absolute inequality but a clear rise in relative inequality. The difference is because of the substantial declining trend in the mean over time. Relative inequality is absolute inequality divided by the mean, and so the smaller the mean, the larger the relative inequality. Inequality appears highly pronounced between CCGs (Figure 51). There is also a slight positive association between equity performance on amenable mortality and deprivation at the CCG level, although this is much weaker than the association with average levels of amenable mortality (Figure 52). The caterpillar plot in Figure 53 shows that rather few CCGs are statistically distinguishable from the national mean in terms of their absolute inequality performance on amenable mortality.

FIGURE 47.

Matrix plot showing unadjusted trends in amenable mortality by age, sex and deprivation (fixed axes for comparisons across age groups).

FIGURE 48.

Matrix plot showing unadjusted trends in amenable mortality by age, sex and deprivation (free axes for comparisons across deprivation groups).

FIGURE 49.

Unadjusted national equity trends in amenable mortality. Notes: (1) the solid black line shows the most deprived quintile and the solid blue line shows the least deprived quintile; (2) negative SII indicates pro-poor distribution, while a positive SII indicates a pro-rich distribution in absolute terms; and (3) negative RII indicates pro-poor distribution, while a positive RII indicates a pro-rich distribution in relative terms.

FIGURE 50.

Adjusted national equity trends for amenable mortality. Notes: (1) the solid black line shows the most deprived quintile and the solid blue line shows the least deprived quintile; (2) negative SII indicates pro-poor distribution, while a positive SII indicates a pro-rich distribution in absolute terms; and (3) negative RII indicates pro-poor distribution, while a positive RII indicates a pro-rich distribution in relative terms.

FIGURE 51.

National social gradient in amenable mortality in 2011/12. Notes: (1) dots represent decile groups; (2) the slope of the line is the SII; and (3) the shaded area shows the inequity gap.

FIGURE 52.

Scatterplots of CCG performance on amenable mortality 2011/12 against deprivation, showing both mean performance and equity performance (AGI). (a) Mean; and (b) AGI.

FIGURE 53.

Caterpillar plot of AGI of inequality in amenable mortality in 2011/12 at the CCG level. Notes: (1) CCGs are ranked from least equitable (left) to most equitable (right); and (2) the dotted horizontal line shows the national average. CCGs to the left with confidence intervals above this line have worse than average equity performance, and vice versa.

Overall mortality

Overall all-age all-cause mortality refers to the number of deaths for all ages and all causes in a given year as a proportion of the total number of people alive at the start of the year. We use all-age all-cause mortality as a contextual indicator of inequality in health, to help interpret levels and trends in our seven health-care equity indicators. Change in this indicator over time may partly reflect change in NHS delivery, but will also reflect change in the socioeconomic patterning of risk factors and health behaviours because of wider social determinants of health outside the health-care system.

In the past, some international studies have used all-cause mortality to measure and compare the performance of health-care systems. For instance, the WHO has reported all-cause mortality rates to compare health outcomes across countries. 145 However, it is by now well established that health care is only one of many social determinants of health,6 and so any credible measure of the role of health care in tackling these wider health inequalities has to focus on indicators that are more directly sensitive to health-care delivery.

In the UK, the ONS produces annual statistics for all-cause mortality by age and sex groups. 146 This provides an important indication of the overall mortality trend and provides the basis for exploring cause-specific mortality. 147 A number of studies have explored socioeconomic inequalities in all-cause mortality. Studies conducted in high-income countries, including the UK, found statistically significant evidence of higher rate of all-cause mortality in lower socioeconomic groups. 148–152 In the case of England, although population-level all-cause mortality rates have been decreasing, area-level deprivation is associated with higher rates of all-cause mortality. 153

This indicator measures the socioeconomic inequality between small-area populations in all-cause mortality rate. We define all-cause mortality as the number of deaths per 1000 people from all causes at all ages. The numerator for this indicator is the number of deaths from any cause that occurred in a given year. The denominator is the total number of people alive at the start of a given year. The indicator was measured for years 2001/2 to 2011/12. As the age and sex structure of each area can affect the mortality rate, using the crude mortality rate would be inappropriate. Hence, in line with the literature, we adjust the mortality rate by taking account of the age and sex structure of the population. Further technical details of how this index was computed are presented in Appendix 4.

The trend in overall mortality is similar to that in amenable mortality, having fallen in all deprivation groups over time as shown in Figures 54 and 55. Once adjusted for age and sex, absolute inequality in overall mortality shows a rise during the early 2000s followed by a fall from 2008 onwards (Figures 56 and 57). Relative inequality shows a similar pattern, except inequality merely flattens out from 2008 onwards: the difference between the two, as before, being the declining mean. Inequality appears to be highly prominent between areas as shown in Figure 58. The caterpillar plot in Figure 59 shows that very few CCGs are statistically distinguishable from the national mean in terms of their absolute inequality performance on overall mortality. As with amenable mortality, there is a positive association between equity and deprivation at the CCG level, although this is much weaker than the association between deprivation and average mortality as shown in Figure 60.

FIGURE 54.

Matrix plot showing unadjusted trends in overall mortality by age, sex and deprivation (fixed axes for comparisons across age groups).

FIGURE 55.

Matrix plot showing unadjusted trends in overall mortality by age, sex and deprivation (free axes for comparisons across deprivation groups).

FIGURE 56.

Unadjusted national equity trends in mortality. Notes: (1) the solid black line shows the most deprived quintile and the solid blue line shows the least deprived quintile; (2) negative SII indicates pro-poor distribution, while a positive SII indicates a pro-rich distribution in absolute terms; and (3) negative RII indicates pro-poor distribution, while a positive RII indicates a pro-rich distribution in relative terms.

FIGURE 57.

Adjusted national equity trends in mortality. Notes: (1) the solid black line shows the most deprived quintile and the solid blue line shows the least deprived quintile; (2) negative SII indicates pro-poor distribution, while a positive SII indicates a pro-rich distribution in absolute terms; and (3) negative RII indicates pro-poor distribution, while a positive RII indicates a pro-rich distribution in relative terms.

FIGURE 58.

National social gradient in mortality in 2011/12. Notes: (1) dots represent decile groups; (2) the slope of the line is the SII; and (3) the shaded area shows the inequity gap.

FIGURE 59.

Caterpillar plot of AGI of inequality in mortality in 2011/12 at the CCG level. Notes: (1) CCGs are ranked from least equitable (left) to most equitable (right); and (2) the dotted horizontal line shows the national average. CCGs to the left with confidence intervals above this line have worse than average equity performance, and vice versa.

FIGURE 60.

Scatterplots of CCG performance on all-cause mortality in 2011/12 against deprivation, showing both mean performance and equity performance (AGI). (a) Mean; and (b) AGI.

Summary of findings on local health-care equity monitoring

In 2011/12, in individual statistical comparisons at the 95% level, well over 20% of CCGs were found to perform significantly differently on equity than the national benchmark, including at least 10% better and 10% worse, using annual data for the following five general indicators: (1) primary care supply, (2) primary care quality, (3) hospital waiting time, (4) preventable hospitalisation, and (5) repeat hospitalisation (Table 3). This was not possible for the remaining three indicators. For indicator (6), dying in hospital, only 8% of CCGs were significantly different from average: 3% worse and 5% better. For indicator (7), amenable mortality, 11% were significantly different from average: 8% worse and 3% better. Finally, for indicator (8), overall mortality, 17% were significantly different from the national average, but most of these were significantly worse: only 3% were significantly better. Pooling additional years of data did not improve substantially the ability to detect significant differences.

Note that our overall findings on the total number of CCGs differing from the national mean must be treated with appropriate caution, as we did not perform any statistical correction for multiple testing and neither did we use statistical control limits to explore the normal range of variation in the SII in order to distinguish ‘general-cause’ variation from ‘special-cause’ variation worthy of concern. 154 Rather, we simply examine whether or not each individual CCG is statistically different from the national mean at the 95% level of statistical significance. We might of course expect that up to 5% of CCGs might pass this test by chance, because of the normal ‘general-cause’ variation, although not the 20% we observe. We leave the further refinement of our statistical methods for future research, as to our knowledge an appropriate statistical formula for setting control limits for social gradients has not previously been developed.

Introduction

We now turn from the data analytical ‘engine room’ of our equity indicators to the communication and knowledge translation ‘front-end’. This chapter describes how we developed visualisation tools for communicating our equity findings to decision-makers and health experts. We developed three main visualisation tools:

• Equity dashboards: a one-page summary for decision-makers at national and local levels, including an interactive spreadsheet tool [based in Microsoft Excel^® (Microsoft Corporation, Redmond, WA, USA)] that can display a dashboard for any CCG in England.

• Equity chart packs: a standard set of slides with tables and graphs showing the underlying inequality patterns and trends in a common format for each indicator, including a portable document format (PDF) file-creating tool [based in the free statistical programming language, R (The R Foundation for Statistical Computing, Vienna, Austria)] that can create a chart pack for any CCG in England.

• Equity custom graphs: a web-based interactive chart tool [based in free software provided by Google Charts (Google Inc., Mountain View, CA, USA)] that allows the user to draw their own customised graphs and see how equity changes over time by selecting from a wide range of variables and chart styles.

A key objective of our study was to develop visualisation tools for presenting equity findings to decision-makers and health experts in a clear, concise and informative manner. Effective communication is essential if findings are to be used in practice to inform decision-making. However, effective communication of findings about equity performance is more challenging than effective communication of findings about average performance, for two reasons. First, inequality is a more complex than the average concept, as it depends in more complex ways on the underlying distributional patterns. There are just three main ways of computing the average of a distribution (the mean, median and mode), whereas there are hundreds of different inequality indices reflecting different aspects of inequality, many of which of themselves have an infinite variety of subspecies based on one or more continuous input parameters. 155 Second, conclusions about how far inequality is ‘unfair’ or ‘equitable’ involve controversial value judgements and empirical beliefs about the causes of inequality about which reasonable people can disagree.

One key role for our equity indicators is to facilitate external NHS scrutiny, as well as to facilitate internal NHS management. In designing our visualisation tools, we therefore sought feedback from a range of intended decision-making audiences including not only NHS commissioning organisations (i.e. NHS England and CCGs) but also organisations with key NHS scrutiny and oversight roles such as Public Health England and Health and Wellbeing Boards. Our equity indicators are also intended for public reporting to enhance democratic accountability, and so we consulted the two lay members of our advisory group. Developing ‘infographics’ for public reporting requires specialised artistic and design skills beyond the skill set of our academic research team, and this was not part of the funding for the research grant. In discussions with the lay members of our advisory group, we concluded that specialised work of this kind will indeed be necessary in future to communicate equity indicator, as members of the public who are unfamiliar with using statistics and graphs may struggle to understand our dashboards and chart packs. We therefore recommend future work to develop suitable ‘infographic’ tools for public communication, which will require funding to pay for specialised media and artistic design skills.

The development of our visualisation tools has benefited from comments from many different people, including those who participated in the following presentations and meetings to national and local NHS and public health audiences:

• presentations to our advisory group in November 2013, November 2014 and September 2015 (see membership in Appendix 3)

• teleconference meeting with experts from the Royal College of General Practitioners on our GP supply indicators, September 2014

• presentation to analysts at NHS England, Quarry House, Leeds, March 2015

• presentation to analysts at Public Health England, York, June 2015

• meeting with the chairperson of Hull CCG, April 2015

• presentation to NHS and public health officials across the health system in York at Vale of York CCG, May 2015

• presentation to NHS and public health officials across the health system in Hull at Hull CCG, June 2015

• meeting with the Chief Economist, Public Health England, York City Council, July 2015

• meetings with the health inequalities lead of the Equality and Health Inequalities Unit on various occasions in 2015, including a meeting with other senior officials from NHS England, Leeds, July 2015

• meeting with analysts at Public Health England, Wellington House, October 2015.

We have also benefited from comments from health indicator experts from a range of disciplines, including those who participated in the following meetings:

• seminar at the Institute for Health Policy, Management and Evaluation, University of Toronto, Toronto, ON, Canada, May 2015

• seminar at the Canadian Institute for Health Information, Ottawa, ON, Canada, May 2015

• seminar at the WHO Collaborating Centre for Knowledge Translation and Health Technology Assessment in Health Equity, Ottawa, ON, Canada, May 2015

• seminar at the Centre for Health Economics and Policy Analysis, McMaster University, Hamilton, ON, Canada, May 2015

• conference talk at the Health Services Research Network Annual Conference, Nottingham, July 2015

• meeting with indicator experts at The King’s Fund, London, July 2015

• conference talk at the Society for Social Medicine Annual Scientific Meeting, Dublin, September 2015

• seminar presentation to the Partnership of Junior Health Analysts at the HSCIC, Leeds, September 2015

• seminar presentation to City University School of Health Sciences Seminar Series, October 2015

• conference talk at International Society for Pharmacoeconomics and Outcomes Research 18th Annual European Congress, Milan, Italy, November 2015

• seminar presentation to the Centre for Health Economics Seminar Series, University of York, November 2015.

The rest of this chapter describes the development of our three visualisation tools in turn.

Development of equity dashboards

Our basic design strategy was to review existing dashboard tools for presenting health equity indicators in the form of a one-page summary, as used by leading health organisations in the UK and internationally, to design our own tools by adapting an existing design that the research team felt would be helpful for our particular purposes, and then progressively to revise our design in the light of feedback from members of our intended audiences. Two lessons rapidly became clear following feedback from a range of professional stakeholders. First, dashboards are more likely to be useful if they summarise all the key information on a single page. Second, equity dashboards are more useful to decision-makers if they present information about averages health-care performance alongside information about social inequality in health-care performance.

The main dashboards that we considered were the Marmot indicators for local authorities in England, the Yorkshire and Humber Public Health Observatory Health Inequalities Dashboards, the WHO Handbook on Health Equity Monitoring and the AHRQ State Quality Dashboards. Of these, the research team concluded that the style of the Marmot indicators was the most suitable for our purposes. The Yorkshire and Humber Public Health Observatory style packed a lot of information into a small space using ‘sparklines’ and other compact graphing formats. However, we felt this was too compact and complex for decision-makers as opposed to analysts. The AHRQ style was the opposite extreme in being too simple for our purposes: a large dial in the middle of the page summarising overall performance across multiple indicators. The WHO and Marmot styles lay somewhere between these two extremes. However, the WHO style did not include information about trends and levels on the same page, or any benchmarking information. By contrast, the Marmot indicators included information on both current levels and trends, on both average and equity performance, and a spine plot allowing comparisons between the local area and national or other equity benchmarks. We therefore adopted the Marmot indicator style as the basis for our dashboard.

However, we made two major modifications to the dashboard design. First, in response to feedback from decision-makers about readability, we de-cluttered the dashboard and made it easier to read. We reduced the space taken up by explanatory notes which take up the entire top half of the page in the Marmot indicators. To make more horizontal room, we created three-word summary titles for each of the indicators rather than using long descriptors. We then put the explanatory notes and longer descriptors on a separate one-page set of indicator notes to be read in conjunction with the dashboard. We also enlarged the font size to 14 points, enlarged the spine plot and allowed larger margins around each cell in the table. This was done to give particular consideration to those who struggle to read text and numbers that are displayed in small font sizes or compact graphics.

Our second major design modification was to add traffic-light background colours to indicate good and bad performance, and arrows to indicate whether or not performance is getting better (an upwards arrow) or worse (a downwards arrow). This was suggested to us by a number of decision-making audiences as being a helpful way quickly to orientate users towards the key findings. Arrows were suggested by the lay members of our advisory group, as being helpful for people who are colour blind. We also received feedback that an upward arrow will naturally be interpreted as ‘improving’ equity performance, even though it implies that inequality is reducing. We experimented with a variety of colour schemes for the traffic lights, but the feedback was that standard red, amber and green colours were easier to interpret than other colour schemes.

Using our modified Marmot indicators dashboard style, we found that up to eight or nine indicators could comfortably fit on a single page in landscape orientation. Two further important pieces of feedback from the decision-makers we consulted are as follows. First, it is important to present information about average performance alongside equity performance. For example, when presenting information about socioeconomic inequality in preventable hospitalisation within a particular CCG, it is important also to present information about the mean level of preventable hospitalisation in that CCG compared with the England mean. This information helps to put the equity findings into context. For example, good equity performance may be less impressive in a context of poor average performance; and deteriorating equity performance may be less worrying in a context of improving average performance in which all social groups are becoming better off. Furthermore, decision-makers want to know this information anyway, as in reality average performance is often more important to them than equity performance. Second, it is important to present at least one equity finding in ‘real’ units rather than rates or percentages, for example numbers of GPs, hospitalisations and deaths. Non-specialists find it easier to understand real units than rates or percentages. Furthermore, decision-makers deal in real units on an everyday basis, and so presenting findings in real units helps them to understand both the scale of the equity problem and the scale of the required policy response. We therefore developed an equity measure in real units, that we call the inequity gap, as described in Chapter 4.

We re-scaled proportions in ways that are (1) easy to read and understand but also (2) help to ensure a degree of consistency across indicators. So, for indicators with proportions larger than 0.01, we re-scaled in terms of percentages, whereas in other cases we used rates per 1000 population. We also carefully considered the orientation for printing, in thinking about how the printed version would need flipping for easy reading, and we piloted the notes pages with various audiences to ensure they were clear.

Development of equity chart packs

Feedback from the professional stakeholders we consulted emphasised the importance of clear visualisation of detailed underling patterns, alongside summary indices in the dashboard. We therefore developed a suite of four main graphs to provide in-depth information about the inequality patterns and trends underpinning our dashboards. First, a matrix graph comprising a panel of line graphs presenting basic descriptive statistics on the indicator by age, sex, deprivation group and year. Second, a scatterplot at decile group level to show the basic cross-sectional shape of the social gradient in health care. Third, a panel of line charts to show equity time trends. Fourth, a caterpillar plot to show equity performance comparisons between local areas. All four types of graphs are presented in Chapter 6, and the second, third and fourth types of graph are described in more detail in Chapter 4.

For presenting basic descriptive statistics and time trends, we followed the standard practice in the health equity literature of presenting information on socioeconomic status using five quintile groups. This is generally sufficient to capture the shape of the social gradient in health care, which is usually fairly linear, although with some important exceptions for some indicators in some years, and does not vary much within particular quintile groups. However, for showing the shape of the current social gradient we opted for decile groups, as we found that for some health-care outcomes there were non-linear patterns that only became apparent within the top and bottom quintile groups.

For the descriptive statistics, we designed a matrix plot comprised a panel of time series line charts by age and deprivation group, with separate male and female lines on the line charts. This enabled us to present all of this information on a single chart. When we presented this to analysts we received positive feedback that this is a useful way of presenting a large amount of information in a small space, and that information on age–sex breakdowns is important for decision-makers.

For the time trends, we used a panel of three line charts showing trends by deprivation quintile group on top, and then trends in two inequality indicators underneath. We experimented with various ways of distinguishing the five quintile group lines using different colours, line widths, line styles, line shades and marker shapes. In the light of feedback from decision-makers and the lay members of our advisory group, we decided (1) not to overcomplicate the graph with multiple ways of distinguishing the lines and (2) to avoid use of colour in the chart packs, partly because of the risk of political overtones, partly because people may be colour blind, and partly because some people may wish to print out the chart packs in black and white. We then arrived at a fairly simple system based on different shades of grey and marker shapes, although also a different line style for the three middle quintiles. This system focuses attention on comparing the most and least deprived quintile groups, while allowing the reader to distinguish the middle three lines on closer inspection.

For the equity performance charts, we opted for caterpillar plots rather than funnel plots. This was for two reasons. First, there is evidence that clinicians, patients and members of the public generally find caterpillar plots easier to understand. 156–158 Second, funnel plots are most useful when there is a relationship between volume (on the x-axis) and outcome (on the y-axis). It is reasonable to expect a volume–outcome relationship in the case of outcomes, such as hospital surgical mortality. However, there is no reason to expect a relationship between the size of a local area and the extent of inequality; and indeed we observed no such relationship.

When presenting the results to local decision-makers, a common theme was that they would like to see scatterplots at neighbourhood level and practice level, (1) so that they can identify which neighbourhoods and practices in their local area are performing well or badly, and (2) so they can get a clearer sense of the (substantial) variation in performance that is not driven by socioeconomic status. However, we were unable to share data of this kind because it may risk disclosing individual-level personal information in which there are counts of events at the neighbourhood level of less than five. This is something that the NHS would need to consider carefully when producing these indicators, that is how to provide local decision-makers with the information they require about individual GP practices and small-area neighbourhoods, without compromising data security. One partial solution, for example, may be to create anonymised local scatterplots by censoring counts below five and/or by adding a ‘jitter’ to the scatterplot, whereby each dot is given a small random perturbation.

Development of equity custom graphs

We reviewed the purpose-built web-based tools that various large international and national organisations have created for allowing users to draw their own custom graphs, including The World Bank DataBank, the OECD Data Lab, the WHO Equity Monitor, the US Institute for Health Metrics and Evaluation, and the Public Health England public health profiles (http://fingertips.phe.org.uk; accessed 12 July 2015). We concluded that it would not be possible to replicate these tools within our limited resources. Instead, we opted to use the freely available chart development software provided by Google Charts (https://developers.google.com/chart; accessed 12 July 2015), which is based on the ‘gap minder’ tool created by Hans Rosling for displaying inequality trends over time (www.gapminder.org; accessed 12 July 2015).

We created a prototype Google Charts tool for our two primary care indicators: primary care supply and primary care quality. We did not add information on the other indicators as the purpose of this work was proof of concept rather than to create a fully comprehensive and up-to-date tool. Our prototype tool is available at http://health-inequalities.blogspot.co.uk (accessed 12 July 2015). The indicators are provided for the years 2004/5 to 2011/12 at the level of England and the four NHS regions, although we did not publish indicators at lower levels because of risk that some of the information might be disclosive. The tool includes a battery of equity measures at both quintile and decile group levels and a range of variables including individual clinical performance indicators for different types of primary care as well as the composite score. Feedback from analysts and decision-makers who viewed our graphs created in Google Charts was uniformly positive, and people particularly liked the ability of this software to show how equity patterns changed over time.

Notes on how to read the national dashboard

To understand the dashboard, it may be helpful to start by considering ‘preventable hospitalisation’, which is a classic indicator of health-care outcome. All of the other indicators of health-care outcome (indicators 4–8) can be interpreted in a similar way. The ‘Average’ columns show overall NHS performance on this indicator. The current level is 5.84 preventable hospitalisations per 1000 population and there is a downward trend since last year of –0.18. This downwards trend is coloured dark green meaning that health outcomes are getting better and that this is a statistically significant finding. The ‘Equity’ columns show equity performance on the SII. The current SII is 6.5, meaning that the most deprived neighbourhood in England has 6.5 more preventable hospitalisations than the least deprived, allowing for the gradient in between. The SII trend is –0.41, which means that the SII is lower this year than last year, that is, inequality is getting less. This box is also coloured dark green, meaning that this is a statistically significant finding. The overall equity trend arrow shows ‘not clear’, meaning that we cannot draw any clear conclusion about whether overall equity is getting better or worse. Finally, the inequity gap shows that inequality in England is associated with 171,119 preventable hospitalisations.

Now we turn to primary care supply, which is a classic indicator of health-care access. This has a similar interpretation, although there are two important differences from all the other indicators: the current SII is negative. At face value, this could be interpreted as suggesting that there is pro-poor inequality, that is deprived neighbourhoods have more GP supply relative to need than affluent neighbourhoods. However, we do not draw this conclusion because we believe that our need-adjustment underestimates need in deprived neighbourhoods, as explained in Chapter 4 and in Appendix 4. So we report the inequity gap as showing ‘no gap’ rather than a negative gap. All of the other columns can be interpreted in the same way as usual, however. So the average level of performance is 1687 patients per GP, with a significant trend in red of 13.0, showing that the number of patients per GP increased by 13.0 since last year. And the trend is –8.94 and in yellow, showing that the change in SII since last year is not statistically significant. If this change was statistically significant, we would interpret this as a beneficial reduction in pro-rich inequality, rather than a harmful increase in pro-poor inequality, because we believe that need in deprived areas is underestimated. However, assessments of need always rely on value judgements as well as empirical facts, and so we present the current negative SII so that decision-makers can draw their own conclusions based on their own value judgements about need.

Finally, we turn to primary care quality, which is different from all the other indicators in that it presents an attainment measure (more is better) rather than a shortfall measure (more is worse). This only influences the interpretation of average performance, however, as we have inverted the SII to ensure that a positive value means pro-rich inequality as with the other indicators. So, average performance is 77.4%, and the positive trend of 1.97 percentage points is coloured in dark green, an increase in quality means that overall performance is getting better, unlike all the other indicators in which an increase means overall performance is getting worse. The positive SII of 1.45 means that the most affluent neighbourhood has 1.45 percentage points more quality than the most deprived neighbourhood, allowing for the gradient in between, and the non-significant trend of 0.06 coloured yellow means that that we cannot draw any clear conclusion about the SII trend. Finally, the same applies to the overall equity trend arrow which shows that the trend is ‘not clear’.

We now turn to the example local dashboard (Figures 63 and 64).

FIGURE 63.

The CCG equity dashboard 2011/12; ‘Any Town’ CCG.

FIGURE 64.

The CCG equity dashboard 2011/12 local indicator notes. a, Adjusted for age and sex each year; b, adjusted for neighbourhood ill health in 2007; and c, adjusted for treating consultant specialty each year. CI, confidence interval.

Notes on how to read the local equity dashboard

Let us start, as before, with preventable hospitalisation. The ‘Average’ column shows overall performance for the CCG and England as a whole. This CCG has 5.89 hospitalisations per 1000, which is slightly higher (worse) than the England average of 5.84. However, this is coloured yellow and so the difference is not significant. The spine plot shows that preventable hospitalisation is not significantly different from the England average; the confidence intervals overlap with the central spine representing the England average. There is no overall equity trend. Finally, the inequity gap is 693, showing that socioeconomic inequality is associated with 693 excess preventable hospitalisations in this CCG area.

Now turning to primary care supply, the ‘Average’ column shows that this CCG has significantly worse supply than the England average: 1974 patients per GP compared with an England average of 1687. The spine plot, however, shows that this CCG is doing significantly better than the England average on equity in primary care supply; the point estimate is comfortably in the light-green zone to the left, and the confidence intervals do not overlap the England average spine.

Finally, turning to primary care quality, this CCG has slightly but significantly worse primary care quality than the England average: an average of 76.4% compared with 77.4%. Furthermore, this CCG is doing significantly worse than the national average on equity in primary care quality; the point estimate is in the dark green zone to the right, and the confidence interval does not overlap the England spine.

Summary of findings

In this study, we have developed health equity indicators to help the English NHS discharge its duty to consider reducing inequalities in health-care access and outcomes. We have developed new methods for local NHS equity monitoring against a national NHS equity benchmark. We have illustrated these methods by applying them to CCGs in the year 2011/12, although they could also readily be applied to local authorities or other geographical areas comprising more than 100,000 people. We have also developed a framework for monitoring national NHS indicators of equity at all main stages of the patient pathway. This framework goes beyond the existing inequalities breakdowns in the NHS Outcomes Framework by including indicators of access as well as outcomes. By producing these indicators from 2001/2 to 2011/12, we have provided the first comprehensive assessment of health-care equity trends during a key period of sustained effort by the NHS to reduce health inequalities through primary care strengthening. Finally, and importantly, we have developed a suite of visualisation tools for communicating equity indicator findings to national and local decision-makers, including equity dashboards providing a one-page summary of both overall and equity performance on multiple indicators, and equity chart packs providing more detailed information. Clear communication is essential in this controversial area, as inequality is a complex concept and so headline statistics are even more liable than usual to be misleading when taken out of context.

Our equity indicators and visualisation tools were selected and designed in consultation with a range of expert stakeholders, including NHS and public health officials at national and local levels and health indicator experts from a variety of disciplinary backgrounds. We also consulted members of the public, because one of the main purposes of our indicators is public reporting for democratic accountability, as well as facilitating quality improvement efforts by national and local decision-makers. Members of the public were involved through a public consultation exercise in York, based on an online survey and a 1-day citizens’ panel meeting, and the two lay members of our advisory group.

The main findings are summarised below under three headings:

1. national equity findings in 2011/12

2. national trends during the 2000s

3. local equity findings in 2011/12.

Strengths and weaknesses of the study

Comparison with other studies

Implications for clinicians, policy-makers and managers

The 2000s was a period of sustained large-scale expenditure growth in the English NHS,180 during which tackling health inequality was a high priority for the NHS. 38,106,181 As has been documented in previous studies, this decade saw substantial increases in overall NHS capacity and utilisation, and the average patient experienced significant improvements in health-care access, quality and outcomes. 182 Our study shows that the NHS also succeeded in achieving substantial reductions in inequality in primary care supply and quality from 2004/5 to 2011/12. By 2010/11, measured pro-rich inequity in primary care supply relative to need had been eliminated and measured pro-rich inequity in primary care quality had been nearly eliminated. Plausibly, these changes can partly be attributed to the substantial investments in primary care in the mid to late 2000s, including the pay-for-performance programme from 2004/5 and the additional funding for new GP practices in ‘under-doctored’ areas of the country in the form of the Equitable Access to Primary Medical Care programme announced in 2006. 166,183 However, these two measures are imperfect and so we cannot conclude there is no remaining important pro-rich inequality in primary care supply and quality. There may remain a degree of pro-rich inequity in primary care supply, because the Carr-Hill formula only allows for morbidity but does not examine how multiple morbidity and disadvantage combine to generate additional health-care needs. So, the Carr-Hill formula is likely to underestimate need in deprived neighbourhoods. There may also remain a degree of pro-rich inequality in primary care quality because QOF indicators do not capture all-important aspects of primary care quality.

The NHS also succeeded in making small reductions in absolute socioeconomic inequalities in health-care outcomes from 2003/4 to 2011/12. Absolute inequalities in preventable hospitalisation and amenable mortality decreased from 2003/4 to 2011/12, and the rate of increase in relative inequalities slowed from the mid-2000s, but substantial inequalities remained in 2011/12.

Although small, the observed reductions in absolute inequality in health-care outcomes during the 2000s are real and impressive for two reasons. First, there is some evidence of widening socioeconomic inequalities during the 2000s in the clustering of smoking, poor diet, physical inactivity and other unhealthy behaviours among lower socioeconomic groups. 62 This would have made it more difficult to reduce absolute inequality in both preventable hospitalisation and amenable mortality. Second, there were no comparable reductions in absolute socioeconomic inequality in non-amenable mortality during the period. This makes it plausible to attribute the reductions to the sustained improvements in health-care access and quality that occurred in the 2000s, rather than to wider trends in the social determinants of health outside the control of health-care services. It is hard to be certain about the causality, however, given that this is an observational study without a control group. Furthermore, there is uncertainty about how long it takes for improvements in health-care delivery to feed through into reductions in preventable hospitalisation and amenable mortality. It is reasonable to expect some short-term impact within a year or two,80 although the length of lag is likely to vary by disease and type of intervention; for example, reductions in mortality because of improved management of heart disease and diabetes may be more rapid than reductions as a result of earlier diagnosis and referral for suspected cancer.

It may not be surprising that the reductions in absolute inequality in health-care outcomes were small, given what is already known about the social determinants of health and the role of health care as just one input into the production of health. 7,9,10,36 Socioeconomic inequalities in health-care outcomes result not only from inequalities of access to health care, but also from socioeconomic-related differences in morbidity, patient self-care and lifestyle behaviour, home and work environments, social care and other public services that impact on health. 8

Therefore, our study provides further confirmation that reducing inequality in health-care outcomes is more complex and challenging than reducing inequality of access to health care. 184 Further reductions in socioeconomic inequalities in health-care outcomes are likely to require complex interventions to improve the co-ordination of care between specialties, between primary and acute care settings, and between health-care and social-care providers. There is a growing body of evidence about effective interventions to reduce preventable hospitalisation and amenable mortality. 171,184–187 Effective interventions may tend to reduce inequalities, if they disproportionately benefit ‘high-need service users’ in the more deprived end of the socioeconomic spectrum who are most in need of co-ordinated care. On the other hand, effective interventions may increase inequalities if they rely heavily on changing people’s self-care and lifestyle behaviour and if individuals in deprived neighbourhoods are less likely to change their behaviour. 188 Unfortunately, however, evidence about the impacts of interventions on socioeconomic health-care outcomes is limited. Therefore, further research is needed including rigorous evaluation of interventions designed to improve the co-ordination of care between primary care, secondary care and social care providers. The indicators developed in this study can be used to facilitate evaluations of this kind, and to help develop the evidence base for reducing inequalities in health-care outcomes through equity monitoring and quality improvement work at local, national and international levels.

Our local equity indicators and dashboards also have potentially important implications for managers, including health-care commissioners, the managers of hospital, primary and community health services, and also the managers of social services and other local government services with impacts on health outcomes. Managers do not yet know which areas of England are performing well or badly on tackling inequalities in health-care access and outcomes, or which areas are showing sustained signs of improvement or deterioration on equity. For example, managers in Liverpool currently do not know whether they are doing better, worse or about the same as managers in Manchester at tackling inequalities in health-care access and outcomes, or whether their equity performance relative to the national average has been improving or deteriorating in recent years. Producing our local equity indicators and dashboards on a routine basis could provide this information, and allow researchers and managers to start finding out why some areas are performing better than others on equity, and why some areas are improving and deteriorating faster than others. This information could also help managers to find out which deprived neighbourhoods within their own local areas are suffering the worst health-care outcomes and why. This is important for hospital managers, as well as commissioners, primary care and public health managers, as avoidable accident and emergency admissions are nearly two and a half times higher in the most deprived fifth of neighbourhoods than the least deprived fifth. Hospital managers have a direct interest and an important role to play in helping to reduce avoidable hospitalisation, through more proactive identification and follow-up of patients at risk of repeated emergency admission, many of whom will live in deprived neighbourhoods. Officials from NHS England and within CCGs have expressed an interest in our equity indicators, as described in the letters of support at Appendix 7, and national inequalities breakdowns are already starting to be included in the NHS Outcomes Framework. At the time of writing, however, no announcement had been made about whether or not, or when, the NHS would commence production of any local health-care equity indicators.

Main conclusions

We draw together our main conclusions below in this section, before turning to technical conclusions and recommendations for further research in the next two sections. The overall research question of our study was: ‘Can changes in the socioeconomic patterning of health-care utilisation and outcomes provide useful indicators of change in NHS equity performance?’ Our overall conclusion is ‘yes’. We elaborate below, with the following more specific conclusions:

• NHS actions can have measurable impacts on socioeconomic inequality in both health-care access and health-care outcomes.

• Increasing the number of primary care physicians and paying them for the quality of care they provide has been associated with small impacts on reducing inequality in health-care outcomes, although the causal link between primary care inputs and health-care outcomes has not been established in this study.

• Local NHS equity monitoring against a national NHS equity benchmark can produce useful findings both to help managers improve quality and to enhance democratic accountability.

• Currently, the most useful indicators for local NHS equity monitoring are primary care supply, primary care quality and preventable hospitalisation.

• National NHS monitoring of change over time in NHS equity can usefully be done using a much wider range of indicators of health-care access and outcomes, including disease-specific indicators.

• Equity indicators are more useful to decision-makers if they are presented together on the same page, alongside average performance indicators, and accompanied by graphs showing the underlying inequality patterns.

• Variants on our equity indicators could be used for international comparisons of equity in health care and for evaluating the impacts of interventions on equity in health care.

Technical conclusions about equity indicator production and communication

In this section, we draw technical conclusions about appropriate analytical methods for producing and communicating equity indicators. Some of our conclusions apply to equity indicator methods used in any country, although some relate to the specific kinds of data available in England, for example conclusions about how often it would be feasible to produce and report particular equity indicators given current data release cycles in England.

Recommendations for further research

Our recommendations for further research are:

1. to investigate potential explanations for variation in health-care equity performance between local NHS areas, so that health-care managers can learn quality improvement lessons

2. to perform experimental and quasi-experimental evaluations of the impacts of complex interventions on socioeconomic inequalities in health-care access and outcomes, including interventions to improve system-wide co-ordination between different specialties, health-care settings and public services

3. to make international health-care equity comparisons using these indicators of health-care access and outcomes

4. to develop broader measures of primary care access and quality that go beyond GP supply and the aspects of quality captured by the QOF

5. to develop better measures of small-area-level need for primary care, by investigating how multiple morbidity and disadvantage combine to generate additional health-care needs

6. to develop convincing methods for risk-adjusting small-area-level health-care outcomes for exogenous morbidity factors beyond the control of health-care services

7. to develop methods for monitoring other social dimensions of health-care inequality

8. to improve these indicator methods, for example by refining and adding indicators, decomposing national inequality into between-area and within-area components, and exploring the use of statistical process control methods, direct standardisation methods and non-linear functional forms

9. to develop sources of small-area-level data on the supply, utilisation, quality and outcomes of public and private social care and other goods and services that may influence health-care outcomes.

Contributions of authors

Richard Cookson (Professor, Health Economics) initiated the collaborative project, had the original idea for the study, supervised all aspects of the research, led the stakeholder liaison process, contributed to study design and interpretation of results, and led the drafting and revision of the report.

Miqdad Asaria (Research Fellow, Health Economics) accessed, extracted and assembled the data, conducted the main data analysis, contributed to study design, interpretation of results and design of the visualisation tools especially the chart packs, and contributed to drafting and revising the report.

Shehzad Ali (Research Fellow, Health Economics) contributed to the data analysis, public consultation process, study design, interpretation of results and design of the visualisation tools especially the dashboards, and contributed to drafting and revising the report.

Brian Ferguson (Professor, Public Health) contributed to the study design, interpretation of the results from a public health perspective and revision of the report.

Robert Fleetcroft (Clinical Lecturer, General Practice) contributed to the study design, interpretation of the results from a primary care perspective and revision of the report.

Maria Goddard (Professor, Health Economics) contributed to the study design, interpretation of the results from a policy perspective and revision of the report.

Peter Goldblatt (Professor, Demography) contributed to the study design, interpretation of the results from a health equity perspective and revision of the report.

Mauro Laudicella (Senior Lecturer, Health Economics) contributed to the study design, the technical indicator methods development, interpretation of the results and revision of the report.

Rosalind Raine (Professor, Applied Health Research) contributed to the study design, the interpretation of the results from an applied health research perspective and revision of the report.

All authors contributed to the design of the work and interpretation of the results, and have commented on drafts of the report and approved the final version.

Publications

Asaria M, Ali S, Doran T, Ferguson B, Fleetcroft R, Goddard M, et al. How a universal health system reduces inequalities: lessons from England. J Epidemiol Community Health 2016;70:637–43.

Asaria M, Cookson R, Fleetcroft R, Ali S. Unequal socioeconomic distribution of the primary care workforce: whole-population small area longitudinal study. BMJ Open 2016;6:e008783.

Conference presentations

• Cookson R. A Framework for Incorporating Equity into Health Care Quality Assurance: Analysis of NHS Administrative Data from 2001/2 to 2011/12. Health Services Research Network Annual Conference, Nottingham, England, July 2015.

• Asaria M. Economic Methods for Health Inequality Measurement. International Health Economics Association, Milan, Italy, 14 July 2015.

• Cookson R. A Framework for Monitoring NHS Equity Trends: Small Area Analysis of Administrative Data from 2004/5 to 2011/12. Society for Social Medicine Annual Scientific Meeting, Dublin, Ireland, 2–4 September 2015.

• Fleetcroft R. Socioeconomic Inequality in GP Supply in England 2004 to 2013. Royal College of General Practitioner’s Annual Conference, Glasgow, UK, 1 October 2015.

• Ali S. Measuring Health Care Performance on Equity: a Framework using National Administrative Data from 2004/5 to 2011/12. International Society for Pharmacoeconomics and Outcomes Research 18 Annual European Congress, Milan, Italy, 10 November 2015.

Data sharing statement

Owing to health data confidentiality requirements, we are unable to publish counts below five at the small-area level for mortality, hospitalisation or other health data. However, we will produce ‘censored’ small-area-level data sets that conform with data security requirements, check that the data providers are happy for us to release these data sets to people who do not have a licence to use the uncensored data, and then once we have acceptable ‘censored’ data sets for this purpose will share these with researchers who ask us for these data. We will request that researchers who re-use these data sets make an appropriate acknowledgement that mentions the contribution of the core research team (Richard Cookson, Miqdad Asaria and Shehzad Ali), the University of York, NIHR funding, and the original data providers. For example, an appropriate acknowledgement might be: ‘The equity indicator data were obtained from research led by Richard Cookson, Miqdad Asaria and Shehzad Ali at the University of York. The original source data were provided under licence from the Department of Health, the Health and Social Care Information Centre, and the Office for National Statistics’.

Disclaimers

This report presents independent research funded by the National Institute for Health 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, NETSCC, the HS&DR programme or the Department of Health. 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, NETSCC, the HS&DR programme or the Department of Health.

Coronary heart disease indicator definitions

Graphical results for coronary heart disease indicators

FIGURE 65.

Equity time trend in CHD primary care quality. Notes: (1) the solid black line shows the most deprived quintile and the solid blue line shows the least deprived quintile; (2) negative SII indicates pro-poor distribution, while a positive SII indicates a pro-rich distribution in absolute terms; and (3) negative RII indicates pro-poor distribution, while a positive RII indicates a pro-rich distribution in relative terms.

FIGURE 66.

National social gradient in CHD primary care quality in 2011/12. Notes: (1) Dots represent decile groups; (2) the slope of the line is the SII; and (3) the shaded area shows the inequity gap.

FIGURE 67.

Unadjusted equity time trend in emergency hospitalisation for CHD. Notes: (1) the solid black line shows the most deprived quintile and the solid blue line shows the least deprived quintile; (2) negative SII indicates pro-poor distribution, while a positive SII indicates a pro-rich distribution in absolute terms; and (3) negative RII indicates pro-poor distribution, while a positive RII indicates a pro-rich distribution in relative terms.

FIGURE 68.

Adjusted equity time trend in emergency hospitalisation for CHD. Notes: (1) the solid black line shows the most deprived quintile and the solid blue line shows the least deprived quintile; (2) negative SII indicates pro-poor distribution, while a positive SII indicates a pro-rich distribution in absolute terms; and (3) negative RII indicates pro-poor distribution, while a positive RII indicates a pro-rich distribution in relative terms.

FIGURE 69.

National social gradient in emergency hospitalisation for CHD in 2011/12. Notes: (1) dots represent decile groups; (2) the slope of the line is the SII; and (3) the shaded area shows the inequity gap.

FIGURE 70.

Unadjusted equity time trend in mortality from CHD. Notes: (1) the solid black line shows the most deprived quintile and the solid blue line shows the least deprived quintile; (2) negative SII indicates pro-poor distribution, while a positive SII indicates a pro-rich distribution in absolute terms; and (3) negative RII indicates pro-poor distribution, while a positive RII indicates a pro-rich distribution in relative terms.

FIGURE 71.

Adjusted equity time trend in mortality from CHD. Notes: (1) the solid black line shows the most deprived quintile and the solid blue line shows the least deprived quintile; (2) negative SII indicates pro-poor distribution, while a positive SII indicates a pro-rich distribution in absolute terms; and (3) negative RII indicates pro-poor distribution, while a positive RII indicates a pro-rich distribution in relative terms.

FIGURE 72.

National social gradient in mortality from CHD in 2011/12. Notes: (1) dots represent decile groups; (2) the slope of the line is the SII; and (3) the shaded area shows the inequity gap.

Discussion of coronary heart disease indicator findings

The basic pattern of our CHD findings is an improvement in average quality and a reduction in absolute inequality between 2001 and 2011. This pattern is entirely in line with our general findings for primary care quality, preventable hospitalisation and amenable mortality, although the improvements and absolute inequality reductions are even larger. However, as with the general indicators, these reductions in absolute inequality do not translate into improvements in relative inequality in the case of hospitalisation and mortality because the mean level of CHD hospitalisation and mortality is falling fast.

Diabetes indicator definitions

Graphical results for diabetes indicators

FIGURE 73.

Equity time trend in diabetes primary care quality. Notes: (1) the solid black line shows the most deprived quintile and the solid blue line shows the least deprived quintile; (2) negative SII indicates pro-poor distribution, while a positive SII indicates a pro-rich distribution in absolute terms; and (3) negative RII indicates pro-poor distribution, while a positive RII indicates a pro-rich distribution in relative terms.

FIGURE 74.

National social gradient in diabetes primary care quality in 2011/12. Notes: (1) dots represent decile groups; (2) the slope of the line is the SII; and (3) the shaded area shows the inequity gap.

FIGURE 75.

Unadjusted equity time trend in emergency hospitalisation for diabetes. Notes: (1) the solid black line shows the most deprived quintile and the solid blue line shows the least deprived quintile; (2) negative SII indicates pro-poor distribution, while a positive SII indicates a pro-rich distribution in absolute terms; and (3) negative RII indicates pro-poor distribution, while a positive RII indicates a pro-rich distribution in relative terms.

FIGURE 76.

Adjusted equity time trend in emergency hospitalisation for diabetes. Notes: (1) the solid black line shows the most deprived quintile and the solid blue line shows the least deprived quintile; (2) negative SII indicates pro-poor distribution, while a positive SII indicates a pro-rich distribution in absolute terms; and (3) negative RII indicates pro-poor distribution, while a positive RII indicates a pro-rich distribution in relative terms.

FIGURE 77.

National social gradient in emergency hospitalisation for diabetes in 2011/12. Notes: (1) dots represent decile groups; (2) the slope of the line is the SII; and (3) the shaded area shows the inequity gap.

FIGURE 78.

Unadjusted equity time trend in mortality from diabetes. Notes: (1) the solid black line shows the most deprived quintile and the solid blue line shows the least deprived quintile; (2) negative SII indicates pro-poor distribution, while a positive SII indicates a pro-rich distribution in absolute terms; and (3) negative RII indicates pro-poor distribution, while a positive RII indicates a pro-rich distribution in relative terms.

FIGURE 79.

Adjusted equity time trend in mortality from diabetes. Notes: (1) the solid black line shows the most deprived quintile and the solid blue line shows the least deprived quintile; (2) negative SII indicates pro-poor distribution, while a positive SII indicates a pro-rich distribution in absolute terms; and (3) negative RII indicates pro-poor distribution, while a positive RII indicates a pro-rich distribution in relative terms.

FIGURE 80.

National social gradient in mortality from diabetes in 2011/12. Notes: (1) dots represent decile groups; (2) the slope of the line is the SII; and (3) the shaded area shows the inequity gap.

Discussion of diabetes indicator findings

It is not possible to draw clear time series conclusions from the clinical process quality indicator for diabetes, because of the repeated changes of indicator definition which manifest themselves as repeated breaks in the time series. If anything, however, both average quality and socioeconomic inequality appear not to have changed much over the period. This contrasts with the situation for CHD and general indicators of clinical process quality, which show a clear pattern of improving quality and reducing inequality.

There is a clear pattern of increasing average emergency hospitalisation for diabetes from 2001 to 2011 along with an increase in absolute inequality. This is the direct opposite of the pattern for preventable hospitalisation in general, and for emergency hospitalisation for CHD, both of which show a pattern of falling average emergency hospitalisation and falling absolute inequality. This increase in hospitalisation may be partly because of increasing prevalence of diabetes, as we adjust for age and sex but not for disease prevalence. Reported prevalence increased during the period, although this may be partly because of changes in diagnosis and reporting behaviour leading to earlier and more complete reporting of cases as opposed to a rise in the true underlying number of cases at risk of emergency hospitalisation for diabetes. In addition, clinical guidelines introduced during the 2000s recommended interventions to achieve lower target levels of HbA_1c, related to lower blood sugars. These interventions are associated with increased use of insulin and higher doses of oral hypoglycaemics, which increases the risk of hypoglycaemia which may increase the risk of hypoglycaemic attack and hence an increase in emergency hospitalisation.

Finally, mortality from diabetes has fallen over the period, as has both absolute and relative inequality in mortality. This contrasts sharply with the pattern for emergency hospitalisation for diabetes, and raises an interesting puzzle for future research: why did mortality from diabetes fall at a time of increasing prevalence of diabetes? The reason may be that the mortality impact of diabetes depends on the presence of comorbid conditions, as well as the quality of diabetes management. Cardiovascular disease is the main cause of death in people with diabetes, accounting for between 52% and 80% of all deaths. 200 Declining morbidity from diabetes may therefore reflect declining CHD prevalence and mortality caused by changes in lifestyle behaviour and improved medical technology for the treatment of CHD. In essence, our hypothesis is that success in tackling CHD is reducing the mortality impact of diabetes while increasing its prevalence, because people who would otherwise have died prematurely from CHD are now living long enough to suffer from diabetes. This is purely a speculation, however, and would need testing in further research.

Developing indicators of change in NHS equity performance

Chairperson, Brian Ferguson.

Ray Avery, Equality and Health Inequalities Unit, NHS England.

Allan Baker, Head of Intelligence, London, Public Health England.

Chris Bentley, Independent Consultant, Health Inequalities National Support Team Associates.

Sarah Curtis, Professor of Geography, University of Durham.

Tim Doran, Professor of Health Policy, University of York.

Brian Ferguson, Director, Knowledge & Intelligence, Public Health England.

Steve Field, Chief Inspector of General Practice, Healthcare Quality Commission.

Donald Franklin, Senior Economist, Department of Health.

Chris Gale, Associate Professor, University of Leeds.

Peter Goldblatt, Deputy Director, University College London (UCL) Institute for Health Equity.

Anne Griffin, Health Inequalities Team Leader, Department of Health.

Iona Heath, Past President, Royal College of General Practitioners.

Ian Holmes, Head of Health System Alignment, NHS England.

Azim Lakhani, Former Head of Clinical Analysis Research and Development, Information Centre for Health and Social Care.

Alan Maynard, Former Chair of York CCG, University of York.

Nicholas Mays, Professor of Health Policy, London School of Hygiene & Tropical Medicine.

Lara McClure, Public member.

Mark Petticrew, Professor of Public Health Evaluation, London School of Hygiene & Tropical Medicine.

Jennie Popay, Professor, Liverpool.

Carol Propper, Chair in Economics, Imperial College London.

Wim Troch, Public member.

Primary care supply

Primary care quality

Hospital waiting time

Preventable hospitalisation

Repeat hospitalisation

Dying in hospital

Amenable mortality

Overall mortality

Recruitment materials for the citizens’ panel exercise

Consent form for the citizens’ panel event

(PDF download)

Rating question about the most unfair inequalities in health and health care

Screenshot from the online questionnaire

E-mail from Health Inequalities Lead, NHS England Equality and Health Inequalities Unit

(PDF download)

E-mail from chairperson of Hull Clinical Commissioning Group

(PDF download)

Letter from Chief Clinical Officer, Value of York Clinical Commissioning Group

(PDF download)