One of the leading continuous quality indicators that U.S. health care managers struggle with is population-wide health equity monitoring and improvement. This indicator seems to unfortunate

Contents lists available at ScienceDirect

Social Science & Medicine

journal homepage: www.elsevier.com/locate/socscimed

Health equity monitoring for healthcare quality assurance

R. Cooksona,∗, M. Asariaa, S. Alib, R. Shawc, T. Doranb, P. Goldblattd

a Centre for Health Economics, University of York, York YO10 5DD, England, United Kingdom bDepartment of Health Sciences, University of York, England, United Kingdom c Analytical Insight Resource Unit, NHS England, England, United Kingdom d Institute for Health Equity, University College London, England, United Kingdom

A R T I C L E I N F O

Keywords: Health equity Quality indicators Health care Small-area analysis Socioeconomic factors

A B S T R A C T

Population-wide health equity monitoring remains isolated from mainstream healthcare quality assurance. As a result, healthcare organizations remain ill-informed about the health equity impacts of their decisions – despite becoming increasingly well-informed about quality of care for the average patient. We present a new and im- proved analytical approach to integrating health equity into mainstream healthcare quality assurance, illustrate how this approach has been applied in the English National Health Service, and discuss how it could be applied in other countries. We illustrate the approach using a key quality indicator that is widely used to assess how well healthcare is co-ordinated between primary, community and acute settings: emergency inpatient hospital ad- missions for ambulatory care sensitive chronic conditions (“potentially avoidable emergency admissions”, for short). Whole-population data for 2015 on potentially avoidable emergency admissions in England were linked with neighborhood deprivation indices. Inequality within the populations served by 209 clinical commissioning groups (CCGs: care purchasing organizations with mean population 272,000) was compared against two benchmarks – national inequality and inequality within ten similar populations – using neighborhood-level models to simulate the gap in indirectly standardized admissions between most and least deprived neighbor- hoods. The modelled inequality gap for England was 927 potentially avoidable emergency admissions per 100,000 people, implying 263,894 excess hospitalizations associated with inequality. Against this national benchmark, 17% of CCGs had significantly worse-than-benchmark equity, and 23% significantly better. The corresponding figures were 11% and 12% respectively against the similar populations benchmark. Deprivation- related inequality in potentially avoidable emergency admissions varies substantially between English CCGs serving similar populations, beyond expected statistical variation. Administrative data on inequality in health- care quality within similar populations served by different healthcare organizations can provide useful in- formation for healthcare quality assurance.

1. Introduction

Quality of care and health equity have become two of the key issues on policy agendas worldwide. However, despite the inclusion of equity dimensions in foundational works on healthcare quality (Donabedian, 2002; Institute of Medicine, 2001) and efforts by organisations such as the Institute for Healthcare Improvement (Institute for Healthcare Improvement, 2017) and the English National Health Service (NHS) (NHS England, 2017b) to integrate equity and quality, responses to these issues have often progressed along separate lines. Efforts to im- prove quality have focused on safety and cost-effectiveness, with im- provements in equity largely a by-product of reducing variation in performance between providers (Doran et al., 2008), whereas policy responses to health equity have focused on the wider social

determinants of health rather than healthcare delivery (World Health Organization, 2014). Due to this parallel development, quality im- provement agencies (for example, the Organisation for Economic Co- operation and Development's (OECD) Health Care Quality Indicators project) (Raleigh and Foot, 2010) and quality improvement frameworks (for example, the Quality and Outcomes Framework in the UK (NHS Digital, 2017b) and accountable care organizations (ACOs) in the US (Centers for Medicare and Medicaid Services, 2017) often overlook equity. Because quality targets tend to be more difficult to achieve for socially disadvantaged populations, there are concerns that quality frameworks penalise providers serving these populations (Delgadillo et al., 2016; Doran et al., 2016; Yasaitis et al., 2016) potentially ex- acerbating existing disparities in the quality of care (Buntin and Ayanian, 2017). Adjustment for social risk factors is now being

https://doi.org/10.1016/j.socscimed.2018.01.004 Received 17 September 2017; Received in revised form 1 December 2017; Accepted 4 January 2018

∗ Corresponding author. E-mail address: [email protected] (R. Cookson).

Social Science & Medicine 198 (2018) 148–156

Available online 06 January 2018 0277-9536/ © 2018 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/BY-NC-ND/4.0/).

T

advocated (Fiscella et al., 2014; Joynt et al., 2017; National Academies of Sciences and Medicine, 2016) but this falls short of providing useful information about equity of care for vulnerable populations, which requires stratification by social risk factors. And whilst there have been isolated examples of quality improvement programs that have explicitly addressed equity (Badrick et al., 2014; Blustein et al., 2011) most are not designed to address this issue.

A major obstacle to improving equity in healthcare has been a lack of appropriate analytical tools. Performance measures in healthcare focus on a mythical “average” patient, providing insufficient informa- tion about differences in quality and outcomes that are considered unfair (Cookson et al., 2016; Fiscella et al., 2000). Periodic reports on healthcare inequalities are produced in some countries (Agency for Healthcare Research and Quality, 2016; Harvey et al., 2016; Moy et al., 2005) but these typically focus on large geographical regions (Mayberry et al., 2006) or local government areas without specific responsibility for healthcare (Remington et al., 2015) and lack the more specific equity metrics and benchmarks needed for assessing and improving the quality of healthcare organizations. To hold healthcare decision makers accountable for the equity dimension of quality, new metrics are needed which (1) speak directly to organizations with direct responsi- bility for healthcare purchasing, planning and delivery, and (2) are responsive to short-term changes in healthcare delivery. Only then will health equity metrics be incorporated into quality assurance dash- boards commanding the attention of senior healthcare executives.

To address this challenge, in 2016 the English NHS introduced a new approach to health equity monitoring for internal quality assur- ance and external public accountability purposes (NHS England, 2016a, 2016b). The initial NHS focus was on equity indicators based on rates of potentially avoidable emergency hospitalization at the neighborhood level, one of which we illustrate in this article, and consideration is being given to adding further indicators in due course. The new ap- proach can be used to construct equity indicators based on many standard indicators of healthcare structure, process and outcome quality including – but not limited to – primary care supply, primary care process quality, hospital waiting times, hospital re-admissions, hospital mortality, and mortality considered amenable to health care (Cookson et al., 2016).

The NHS chose to focus initially on potentially avoidable emergency admissions for two reasons. First, average rates of these admissions are responsive to short-term changes in health care delivery (Harrison et al., 2014; Huntley et al., 2014; Purdy and Huntley, 2013). Second, they rise steeply with neighborhood deprivation, raising concern not only about equity of access to preventive and co-ordinated healthcare (Asaria et al., 2016a) but also about cost pressures on the healthcare system as a whole (Asaria, et al., 2016b). Under the new approach, inequality in potentially avoidable emergency admissions was mea- sured within the populations of “clinical commissioning groups” (CCGs) – care organizations in England with responsibility for purchasing and planning healthcare for patients enrolled with local NHS family prac- tices. Equity within the CCG's enrolled population was then compared against two benchmarks: the national average level of inequality and the average level of inequality within ten CCG populations that are comparable in terms of deprivation, age profile, ethnic mix and rurality (NHS England, 2017a). In this article we illustrate the NHS equity in- dicator based on the sub-set of potentially avoidable emergency ad- missions for chronic ambulatory care sensitive conditions. This is an indicator of the quality of ambulatory care services in managing long- term conditions (Herrin et al., 2015; Purdy et al., 2009; Torio and Andrew, 2014) and the equity version of this indicator is intended to provide quality assurance information about the NHS duty to consider reducing inequalities in both access and outcomes of healthcare (Health and Social Care Act, 2012). In this paper, we use this indicator to il- lustrate the general analytical approach and discuss its potential ap- plication to healthcare quality assurance in other countries.

2. Methods

2.1. Data

2.1.1. Organizational geography In England in 2015 there were 209 clinical commissioning groups

(CCGs) – each serving a mean of 272,000 NHS patients registered with a local family practice (range 73,000 to 913,000). CCGs are responsible for purchasing and planning healthcare for the vast majority of their resident populations. However, the registered and resident populations do not fully overlap because residents can choose to register with a practice in a neighboouring CCG. We used registered population data from practice registers, rather than resident population data from the census, to match the legal responsibility of the CCG and to illustrate how the approach can be applied to ACOs in the US and other settings where the enrolled population does not coincide with the resident po- pulation. CCGs were introduced in April 2013. There were 211 CCGs initially, falling to 209 in 2015. Before that, there were 152 “Primary Care Trusts” (PCTs). Despite this numerical change, however, there was stability in most areas with 180 of the 211 CCGs being formed from a single PCT or part of a single PCT, and the opening and closing of practices to accommodate local population change does not cause substantial change in CCG boundaries.

2.1.2. Small area geography Our basic unit of analysis was the “CCG-LSOA” – a block of CCG

registered population residing within a neighbourhood census unit called a “lower super output area” (LSOA). Each patient has a neigh- bourhood or “Lower Super Output Area” (LSOA) in which they live. Each LSOA has a deprivation score. Patients register with a GP practice and these practices belong to CCGs responsible for their hospital care. To calculate the inequality in a CCG, we include everyone who is re- gistered with that CCG's GP practices based on the LSOA where they live. Effectively we split each LSOA into CCG blocks, as illustrated in Fig. 1.

We include all the shaded blocks for each CCG, taking the depri- vation score of the LSOA in which they are located. LSOAs have a mean population of 1650 (range 1000 to 3000), while CCG-LSOAs have a mean population of 636 (range 1–2536 from 1st to 99th percentile). Our CCG-LSOA population estimate was based on the fraction of the relevant NHS practice list attributed to the LSOA. The resulting mean number of CCG-LSOAs per CCG was 428 (range 95 to 1972). CCG- LSOAs with smaller-than-resident populations arise near CCG bound- aries, where residents of an LSOA are registered in more than one CCG with such LSOAs appearing in the analysis for more than one CCG. However, most LSOAs have a majority of their population registered with a single CCG (95.4% on average). Even among LSOAs whose po- pulations are registered with multiple CCGs, the largest proportion

Fig. 1. How CCG-LSOAs are constructed – fictional example. Note: The 3 shaded areas are CCGs, the 25 (5*5) cells are LSOAs, and the 30 shaded blocks within the cells are CCG-LSOAs.

R. Cookson et al. Social Science & Medicine 198 (2018) 148–156

149

tends to be registered with a single CCG – for example, 1748 LSOAs have population registered with five different CCGs, but among this group on average 90% of the population are registered with a single CCG.

2.1.3. Hospital admissions Data on emergency admissions were taken from NHS Hospital

Episode Statistics (HES), a data warehouse containing details of all admissions at NHS hospitals in England. We extracted data for 1 October 2014 to 30 September 2015 and calculated the indirectly age- sex standardized admission rate for each CCG-LSOA. Admissions by people with unknown age, sex, LSOA or CCG were excluded, including individuals not registered with a family practice. Approximately 0.2% of admissions had no recorded age or sex, 0.7% had no LSOA (given valid age and sex) and 0.6% had no CCG (given valid age, sex and LSOA). Approximately 1.4% of admissions were excluded.

2.2. Measures

2.2.1. Potentially avoidable emergency hospitalization We used emergency admissions for chronic ambulatory care sensi-

tive conditions as defined in existing NHS indicators (NHS Digital, 2017a). This includes diagnoses defined using ICD-10 codes for condi- tions such as asthma, bronchitis, diabetes, dementia and heart disease.

2.2.2. Neighborhood deprivation The measure of deprivation we used was the Index of Multiple

Deprivation 2015, a multi-domain index of deprivation (McLennan et al., 2011). This combines domains of deprivation including low in- come, unemployment, poor housing and crime.

2.2.3. Similar CCGs The list of similar CCGs was created by the NHS to aid bench-

marking of CCG level information, based on 10 CCGs with the lowest sum of squared differencesce on 12 indicators including age profiles, deprivation, population density and ethnicity, after first normalising the indicators by subtracting the mean and dividing by the inter-decile difference (NHS England, 2017a).

2.3. Analytic approach

Fig. 2 illustrates the general analytic approach to health equity monitoring against national and similar population benchmarks. The solid line is based on a linear regression weighted by population, and illustrates the positive association between the care quality indicator – in this case, potentially avoidable emergency hospitalization – and neighborhood deprivation within the registered population of the se- lected CCG. The slope of this line represents the CCG inequality “gra- dient” by level of deprivation within the registered population: the steeper the slope, the larger the degree of deprivation-related inequality in potentially avoidable emergency hospitalization. The dashed line is the gradient within England as a whole, and the dotted line is the gradient within a sub-set of England comprising the registered popu- lation of this CCG along with ten other CCGs with similar populations. These benchmark gradients are also based on linear regressions. In this example, the CCG gradient is less steep (better) than both the national and similar population gradients to an extent that is statistically sig- nificant, so we can conclude that this CCG is relatively equitable on both benchmarks. We can then monitor change over time against na- tional and similar population benchmarks in response to actions taken by healthcare decision makers.

The benchmark gradients play a crucial role in quality assuring the equity performance of healthcare organizations. Healthcare organiza- tions can be expected to address poor quality and equity of healthcare, but cannot address wider social determinants of the gradient in healthcare outcomes on their own (see Fig. 3). The risk of an acute ill-

health event requiring emergency hospital treatment is influenced by individual risk factors (e.g. age, morbidities) and behaviors (e.g. diet, smoking) which in turn are influenced by cumulative long-term en- vironmental risk factors (e.g. childhood circumstances; living and working conditions; and access to resources for investing in health). A residual social gradient in potentially avoidable emergency hospitali- zation would therefore remain even if the CCG achieved perfect equity by providing equal access to high quality preventive and co-ordinated care.

It is not possible to adjust for all risk factors since detailed in- formation may not be available in administrative data, especially for individuals who have limited contact with health providers. In addition, there is a danger of over-adjustment for risk factors that are highly correlated with deprivation and amenable to modification over time by healthcare services. The danger of over-adjustment increases with the breadth of policy responsibility: the broader the policy toolbox, the greater the ability to modify risk factors. Preventive and long-term care can modify individual risk factors, and wider public health and social policies can modify environmental risk factors which in turn will in- fluence individual risk factors and behaviors.

The appropriate quality assurance benchmark is therefore not zero inequality, but the residual degree of inequality expected for a similar population with a similar social patterning of unobserved risk factors. Two key benchmarks are the national gradient and the similar popu- lation gradient, which assume that the social patterning of unobserved risk factors in a CCGs is the same as, respectively, the country as a whole or as a cluster of CCGs serving similar populations. We selected ten similar populations based upon a standard analysis by the NHS of CCG population “similarity” in terms of twelve variables reflecting deprivation, health, population size and age profile, population density and ethnicity (NHS England, 2016c). The monitoring of time trends in relation to these benchmarks allows an assessment of how equity per- formance is responding to healthcare initiatives.

We measured the slope of the gradient using the “absolute gradient index” (AGI). This is the coefficient from the population-weighted linear regression of age-sex standardized avoidable hospitalization rates against fractional deprivation rank on a scale of 0–1, using all neigh- borhoods registered to the CCG. It is the same as the conventional slope index of inequality except that the AGI indices use the national depri- vation rank rather than the local deprivation rank. This difference al- lowed us to compare CCG inequality on a like-for-like basis with the national inequality benchmark and the similar population benchmark, even though different CCG registered populations can have different deprivation profiles. For a CCG serving a relatively affluent population, for example, the most deprived fifth of neighborhoods might all be fairly affluent in national terms. A low rate of hospitalization in these neighborhoods would then not reflect the same equity achievement as a low rate among nationally deprived neighborhoods. The AGI can be interpreted as the simulated gap in potentially avoidable emergency hospitalization between the most and least deprived neighborhood in England, allowing for the gradient in between, if England had the same gradient as the registered population of the CCG.

To help decision makers interpret the AGI and assess the scale of their inequality challenge, we also derived an approximate estimate of the excess hospitalizations associated with inequality, drawing on the epidemiological concept of population attributable risk. This concept represents the number of emergency hospital admissions that would hypothetically be avoided if all neighborhoods had the same admission rate as the most affluent. We estimated this using the AGI multiplied by the relevant population and divided by two. This formula is a simple approximation, based on the assumptions of a linear relationship be- tween deprivation and admissions and an evenly distributed population across the deprivation spectrum (Asaria et al., 2016b).

The analysis was carried out using R statistical software. (version 3.2.4). Full analysis code can be found at: https://github.com/ miqdadasaria/ccg_equity.

R. Cookson et al. Social Science & Medicine 198 (2018) 148–156

150

3. Results

The mean indirectly age-sex standardized rate of potentially avoidable emergency hospitalization in England was 792 per 100,000 people. The national absolute gradient index (AGI) – the estimated gap between the most and least deprived neighborhoods in England – was 927 (95% confidence interval 912 to 942), or 117% of the mean neighborhood level CCG-LSOA rate. The modelled rates for the most and least deprived neighborhoods were 1261 and 334, respectively.

Fig. 4 shows healthcare equity in 2015 for all 209 CCGs in England, with 95% confidence limits. The horizontal line is the national AGI

benchmark of 927; the similar ten benchmark is not shown as it varies by CCG. Against the national benchmark, 17% of areas (35 of 209) exhibited equity that was statistically worse-than-benchmark at a 95% confidence level – i.e. CCG inequality was larger than national in- equality – and 23% (48 of 209) exhibited significantly better-than- benchmark equity. Against the similar population benchmark, the corresponding figures were 11% worse-than-benchmark and 12% better-than-benchmark. Against both benchmarks, 9% of areas show worse-than-benchmark equity performance and 10% show better-than- benchmark equity performance.

There was moderate negative correlation between the average de- privation of a CCG and its equity performance against the relevant si- milar population benchmark, as measured by the similar population inequality gap minus the CCG inequality gap (Pearson's r −0.57). This means that English CCGs serving relatively deprived populations gen- erally performed worse on health equity than those serving relatively affluent populations, and that about one third of the variation in CCG equity performance (Pearson's r squared 0.32) was associated with average deprivation. This correlation reduced but persisted when using relative rather than absolute measures of inequality, such as the abso- lute gap as a proportion of the CCG modelled mean for an individual with the national average level of deprivation.

Fig. 5 illustrates healthcare equity in six selected CCGs. We have selected pairs of organizations serving populations with different average levels of deprivation, with each pair illustrating better-than- benchmark versus worse-than-benchmark equity (“Horsham and Mid Sussex” versus “Windsor, Ascot and Maidenhead” with low deprivation, “Ashford” versus “North Lincolnshire” with medium deprivation, and “Brent” versus “Liverpool” with high deprivation). In five of these six examples both benchmark comparisons were statistically significant. However, the comparison for “Windsor, Ascot and Maidenhead” was not statistically significant against either national or similar population benchmarks – in this dataset, there was no example of an CCG serving a low deprivation population that had significantly worse-than-

Fig. 2. Healthcare equity against national and similar po- pulation benchmarks – hypothetical scatterplot showing potentially avoidable emergency hospitalization and depri- vation for all neighborhoods within a clinical commissioning group. *Neighborhood rate of potentially avoidable emergency hospitalization, indirectly standardized for age and sex. **Neighborhood national deprivation rank from the Index of Multiple Deprivation 2015, converted into a fraction be- tween 0 (least deprived) and 1 (most deprived). Note: Dots represent neighbourhoods registered to the clin- ical commissioning group. This example shows a clinical commissioning group with an inequality gradient that is shallower than both the national benchmark gradient and the similar area benchmark gradient, indicating better-than- benchmark equity.

Fig. 3. The healthcare and non-healthcare determinants of emergency hospitalization. Note: Long-term care can include various medical and non-medical services for people with chronic mental or physical illness or disability who cannot care for themselves for long periods, including help with normal daily tasks like dressing, feeding and house- keeping.

R. Cookson et al. Social Science & Medicine 198 (2018) 148–156

151

benchmark equity. Fig. 6 illustrates how similar-ten benchmarking works and can be

presented to decision makers. It shows a CCG – Liverpool – with worse- than-benchmark equity performance. It shows the AGI for Liverpool and its ten similar CCG populations, along with the average AGI pooled across all eleven populations. Inequality is significantly worse than the similar population benchmark in Liverpool and in one other CCG (South Manchester), and significantly better in two CCGs (Brighton & Hove and Sheffield). The AGI in Liverpool was 1523 compared with a similar population benchmark AGI of 1177. This equates to 3840 excess hos- pitalizations a year associated with inequality among the Liverpool population.

Full results for every clinical commissioning group can be found online in our interactive inequality explorer: http://www.ccg- inequalities.co.uk/

4. Discussion

4.1. Summary of findings

We have illustrated the new analytical approach to health equity equity monitoring for healthcare quality assurance introduced in England in 2016, using the example of potentially avoidable emergency hospitalization. The approach aims to provide healthcare purchasing and planning organizations – in this case, English CCGs – with detailed, up-to-date information on the equity dimension of healthcare quality within their enrolled populations. It measures inequality in key in- dicators of healthcare quality within the enrolled population and then assesses equity against two benchmarks – national inequality, and in- equality within a group of care organizations with similar populations. Inequality is measured on a comparable basis using population- weighted models of the neighborhood-level relationship between healthcare quality and deprivation, allowing for differences between neighborhoods in their demographic makeup.

Using data for 2015 on potentially avoidable emergency hospitali- zation, we found that 9% of the 209 CCGs in England showed sig- nificantly worse-than-benchmark equity against both national and si- milar population benchmarks, and 10% showed significantly better- than-benchmark equity. This is considerably more than the 5% in each category expected due to chance.

4.2. Strengths of the approach

The strengths of this approach include the ability (i) to incorporate

equity metrics into mainstream quality assurance processes for orga- nizations with direct responsibility for healthcare purchasing, planning and delivery for enrolled populations as small as 100,000 people; (ii) to assess the equity dimension of quality against two relevant benchmarks: national inequality and inequality within similar enrolled populations; and (iii) to assess the scale of the health inequality challenge facing the organization using the epidemiological concept of population attribu- table risk. Particular strengths of potentially avoidable emergency hospitalization as a key equity indicator include the ability to address the equity dimension of a high profile quality issue with substantial cost implications, and to incorporate data on the quality of care for dis- advantaged people who are relatively unlikely to participate in household surveys but relatively likely to suffer emergency hospital admission.

This approach can also be used to monitor how equity changes over time in response to short-term changes in healthcare delivery by par- ticular care organizations (Sheringham et al., 2016). For example, Li- verpool has recently

Collepals.com Plagiarism Free Papers

Are you looking for custom essay writing service or even dissertation writing services? Just request for our write my paper service, and we'll match you with the best essay writer in your subject! With an exceptional team of professional academic experts in a wide range of subjects, we can guarantee you an unrivaled quality of custom-written papers.

Get ZERO PLAGIARISM, HUMAN WRITTEN ESSAYS

Why Hire Collepals.com writers to do your paper?

Quality- We are experienced and have access to ample research materials.

We write plagiarism Free Content

Confidential- We never share or sell your personal information to third parties.

Support-Chat with us today! We are always waiting to answer all your questions.