Setting and study population

Our analysis focused on local government councils with responsibility for providing adult social care services in England between 2005–2006 and 2015–2016. In 2015–2016, there were 152 English councils with responsibility for providing adult social care services. Nine of these councils were only established following structural changes to local government in April 2009. These areas were excluded from all analyses. A further three councils were missing spend data for at least 1 year, these areas were also excluded. The City of London, the Isles of Scilly and Rutland were excluded because we expected their small populations would cause instability. We excluded a further five councils that reported a year-on-year change in emergency hospital admissions for older people of >50% in one or more years between 2005/2006 and 2015/2016 on the basis that these jumps are implausible and are more likely to represent data errors. This left a balanced panel dataset of 1452 observations (n=132, T=11). The list of the councils excluded from the analysis is provided in online supplementary file 1.

Data sources

The data used in this study were collected from a number of national information systems. The main outcome variable in our analysis is counts of emergency admissions to hospitals (admissions that happen at short notice because of perceived clinical need) by people aged 65 years and over in each council area. We obtained this information from the Hospital Episode Statistics (HES) dataset. HES contains records of all admissions for patients admitted to NHS hospitals, and includes information on method of admission (eg, emergency), diagnosis codes recording the primary reason the patient is being treated and any secondary diagnoses relevant to their care.13 We hypothesised that some types of admission are more likely to be avoidable by timely access to social care. Ambulatory care sensitive conditions (ACSCs) are a well-defined set of conditions where effective community care and case management can help prevent the need for hospital admission.14 Counts of emergency hospital admissions of people aged 65 years and over for ACSC conditions were derived from HES and used as a secondary outcome measure.

Our main exposure variable was government spending on social care. We obtained financial data relating to publicly funded social care for older people (those aged 65+ years) from administrative returns provided to central government by the social services departments of councils providing adult social care services in England. Publicly funded social care in England is targeted at those with the highest needs and lowest incomes. Older people who meet needs-based eligibility criteria but whose income and assets are above a set amount are required to contribute a proportion of the cost themselves, and some with needs receive no financial support. We calculated a series for councils’ net expenditure by subtracting contributions paid by service users for their care from councils’ gross total expenditure on services for older people. These expenditure levels include income from the NHS through schemes such as the Better Care Fund.

In 2014–2015, there were changes to the financial reporting framework used by councils to collect these data that meant we needed to ‘map’ categories of council income across the two frameworks to obtain a consistent series on councils’ income from client contributions. Bridging files were published by NHS Digital to facilitate mapping exercises of this type.15

We adjusted net total expenditure, by each council in each year, to 2015–2016 prices using the Gross Domestic Product Price Deflator.16 To account for differences in population size, we used population estimates from the Office for National Statistics (ONS) to calculate a rate of expenditure per head of population aged 65 years or over per year for each council area.17

In the UK and other countries, admission rates are significantly correlated with measures of social deprivation.18 To control for socioeconomic deprivation, we used data from the Index of Multiple Deprivation 2015, the official area-based measure of relative deprivation in England. We used the Income Deprivation Affecting Older People Index (IDAOPI), a supplementary index to the overall income domain that measures the proportion of adults aged 60 years or over living in income deprived households in each council area.19 We grouped council-level IDAOPI scores into fifths, with the first quintile group representing the least deprived.

Counts of deaths by age group, gender and local authority were supplied by the ONS. Mortality rates for people aged 65 years and over were calculated by dividing the number of deaths by the population for each council and year.

Statistical analyses

We initially investigated the unadjusted trends in the rates of emergency hospital admissions and government social care spend, in total and by council.

We used multivariable regression analysis for panel data to assess the relation between emergency hospital admissions (response variable), and net spend on government-funded social care for older people (predictor variable). We used negative binomial regression models as these are appropriate when the outcome is a count variable.20 To adjust for differences in population sizes of older people, we included the size of the population aged 65 years and over as an exposure variable. Panel data models were used to help reduce the risk of omitted variable bias. Given that we were interested in the population-averaged effect, we used general estimating equations to estimate the model coefficients. Decisions relating to selecting variables and the model correlation structure were made using the quasi-likelihood under the independence model criterion (QIC).21

For each council, we decomposed the variable representing government spending on social care into two new variables: a time-invariant average spend and a time-varying difference from average spend. This allowed us to assess the influence on our outcome of variation in social care spend ‘within’ councils (eg, over time) and ‘between’ councils.

Mortality rates among people aged 65 years and over and deprivation levels (IDAOPI quintiles) were used to control for differences in levels of population need.

To adjust for the long-term trend in emergency hospital admissions and to control for unobserved in-year effects, we included dummy variables for each year. We used robust SEs to reflect the fact that populations were not sampled independently and to ensure that SEs were robust to serial correlation in the data.

Models were prepared for two outcome variables: all emergency admissions for people aged 65 years and over (our primary outcome variable) and the subset of these which were for ACSCs.

Regression analyses were performed with Stata IC V.15.1. Data preparation was done using R Statistical Software V.3.3.2 (R Foundation for Statistical Computing, Vienna, Austria).

Sensitivity analyses

We performed several analyses to test the sensitivity of model results to outlier values and to elements of the model specification. To test the effect of outliers on the model coefficients, we calculated dfbeta values for each council for the variable representing government spend on social care and reran the model removing those councils with the highest dfbeta values. To test the sensitivity of the results on the choice of a population-averaged modelling approach, we recreated the model using a random-effects approach. To test the sensitivity of the results to lagged or leading effects of the time-varying independent variables, we reran the model lagging by 1 year the government spend on social care (on the basis that the impact of a change in social care spend may have a delayed impact on emergency admissions), and leading by 1 year the mortality rate variable (given that emergency admissions are known to rise exponentially for several years prior to death).22 Based on QIC values, our model incorporates year as a series of 10 (T-1) dummy variables, rather than as a single linear covariate. However, to assess whether the treatment of time in our model influenced the relationship between our outcome variable and our variable of interest, we produced a version of the model which incorporated time as a linear covariate. Finally, we tested the sensitivity of the model results to the specification of an independent within-group correlation structure.

Patient and public involvement

Patients and member of the public were not involved in the design of this study.

Descriptive analysis

Between 2005/2006 and 2015/2016, the population aged 65 years and over grew on average by 1.9% per annum from 7.0 million to 8.4 million (see table 1). Over the same period, emergency hospital admissions of patients aged 65 years and over rose by 3.0% per annum (see figure 1). Government spend on social care for this age group reduced by 0.6% per annum after adjusting for inflation (see figure 2).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Emergency hospital admissions aged 65+ years per head of population aged 65+ years. ACSC, ambulatory care sensitive condition.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Nominal and real (inflation adjusted) government net spend on social care for people aged 65+ years (£’000 s) per head of population aged 65+ years.

There is considerable variation between councils in the levels of spend on social care and emergency admissions per head of population (see figure 3). Rate of social care spend and emergency admissions are notably higher in the most deprived quintile.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Government spend on social care (£’000 s), emergency and ACSC admissions per head of population (65+ years) by council and IDAOPI (deprivation) quintile. ACSC, ambulatory care sensitive condition; IDAOPI, Income Deprivation Affecting Older People Index.

Model results

Having adjusted for the other model covariates, we found no statistically significant relationship between the rate of government spend on social care within a council and our primary outcome variable, emergency hospital admissions (IRR 1.009, 95% CI 0.965 to 1.056)—see table 2. Likewise, we found no significant relationship with our secondary outcome, ambulatory care sensitive admissions (IRR 0.975, 95% CI 0.917 to 1.038)—see table 3.

Sensitivity analyses

We reproduced the models excluding the nine councils with the highest dfbeta values (>2 SD from 0) for the variable representing government spend on social care within councils. This did not alter the significance of the incident risk ratio for social care spend.

Similarly, reproducing the model with a random-effects formulation did not the alter significance of the incident risk ratio for social care spend.

Lagging the variable representing government spend on social care (within councils) and leading the variable representing mortality rate by one time unit did not alter the significance of the incident risk ratio for social care spend.

When we incorporated year as a linear covariate (rather than as a set of dummy variables), the incident risk ratio increased marginally and became statistically significant.

The incidence risk ratio for social care spend was not materially altered when alternative within-group correlation structures (exchangeable, unstructured and autoregressive order 1) were specified.

Online supplementary file 2 contains the model coefficient and 95% CI for our variable of interest in each of the sensitivity analyses described above.

Key findings

Across our study population, government spend on social care for people aged 65 years and over fell by £472 m (6.2%) in real terms between 2005/2006 and 2015/2016. We conclude that this was not associated with an increase in emergency hospital admissions. This finding is at odds with both intuition and the perceptions of those working in acute hospitals. We will explore a range of potential (and interrelated) explanations for this effect.

The prima-facie explanation for the observed results is that social care provision is not an effective means of preventing emergency hospital admissions for older people. It is likely that individuals who have been affected by reductions in social care spend to date are those whose need levels are close to eligibility thresholds and therefore have the lowest ability to benefit. It does not necessarily follow therefore that further reductions in social care spend will not result in increases in emergency hospital admissions. Moreover, our finding should not be taken as evidence of the ineffectiveness of social care in more general respects. The effectiveness of social care with respect to quality of life, for example, has been studied elsewhere and while this evidence base remains limited, on balance it suggests that the provision of social care leads to quality of life benefits.23

Another explanation for the findings is that social care provision may avoid hospital admissions by improving the health status of older people but that these gains are offset as social workers and social care professionals identify unmet need and trigger a healthcare intervention leading to hospital admission. This could arise, for example, if a social care professional notices a deterioration in a patient’s health status, referring the patient to a general practitioner (GP) or emergency department. While the UK has a strong primary healthcare system and might be expected to manage clinical risks in these circumstances, GP services are under pressure and access to primary care remains a problem. These direct and countervailing effects may be reduced in tandem as social care spend is reduced. This mechanism has been proposed for other interventions that aimed, but failed to reduce emergency admissions in older people.24

A third explanation focuses on the effect of substituting government funded social care for privately funded social care and informal care. While substitution may not occur in all cases, privately funded social care and informal care may be allocated more efficiently than government-funded social care, offsetting the losses associated with those individuals who now receive no care. However, the English Longitudinal Study of Ageing suggests that the use of privately funded social care fell between 2004/2005 and 2014/2015 for those reporting problems with activities of daily living (eg, walking, bathing, dressing) and instrumental activities of daily living (eg, cooking, shopping).25 While public and private provision have been falling informal care has increased substantially. A survey conducted by Department of Work and Pensions estimates that 8% (4.9 million) of people were informal carers in England in 2016.26 Since 2004 the average daily minutes of adult care provided by those aged 8 years or over has risen year on year.27 The gross value added of informal adult care in the UK increased by 45.8% between 2005 and 2014, from £39.0 billion to £56.9 billion.28 These figures demonstrate a substantial and sustained shift from public to informal care provision over the period of this study.

While healthcare services have been spared the severe funding cuts seen in other public services, healthcare services nonetheless report significant funding pressures. Many initiatives have sought to manage demand for emergency hospital admissions and a recent study demonstrated that thresholds for admission via emergency departments increased over the period from 2009/2010 to 2014/2015.29 These supply constraints and changes in clinical behaviour may explain why reductions in social care spend have not resulted in increases in emergency hospital admissions.

Finally, we consider the impact of substantial funding constraints on the efficiency of social care services. To accommodate cuts in funding, many social services departments have tightened need-based access criteria for services and have introduced reablement and preventative services to reduce long-term demand for care. This may have increased the efficiency of social care, improving the health status of service users and offsetting the losses to those who no longer receive government funded care. Local authorities facing the largest cuts may have undertaken the most radical redesign of services and achieved the greatest efficiency gains.

Limitations

Ecological studies of this type are susceptible to a number of forms of bias. However, our conservative modelling approach and the study’s statistical power imply that it is extremely unlikely that reductions in social care spend resulted in material increases in emergency admissions at a population level. None of the sensitivity analyses provided contrary evidence.

As with all observational studies, we must consider the possibility that our observed result can be explained by the omission of a key explanatory variable. Given the use of panel data, any influential omitted variable must be a time-varying effect operating within councils, which is strongly associated with our dependant variable and our variable of interest. One candidate is household income. However, income inequality in retired households has grown since 2009/2010 and over the same period councils with higher levels of deprivation have seen larger reductions in social care spend.30 We would expect these changes to amplify rather than obscure or repress any relationship between social care spend and emergency admissions.

Given the ecological nature of our study, we should consider the possibility that reductions in social care spend have indeed resulted in increases in emergency admissions, but that equivalent reductions in emergency hospital admissions have occurred in other groups of older people. We can think of no clear mechanism that might correspond to this theory.

In 2014/2015, the financial reporting framework used to collect data from councils on government spend on social care was substantially redesigned. The mapping of categories required to create consistent time series introduces the potential for error. However, our sensitivity analyses did not indicate that the results were substantially affected by the presence of outliers. A small number of councils were excluded from our analysis. While the rationale for the exclusions are clear and explicitly described, this process may have introduced bias.

While this study did not explicitly control for changes in the age profile or morbidity levels of the study population, the use of mortality rates as an independent variable adjusts for these factors indirectly.

Relation to existing literature

Unlike our study, the small number of previous studies which have explored the relationship between social spend or provision and healthcare use has found evidence of a substitution effect. A cross-sectional analysis using small area data in England found the cost effects of a transfer of resources from hospitals to care homes were broadly neutral, but this study predates the recent significant reductions in social care expenditure.8 A later study using panel data found delayed discharges from hospitals in England responded, if only weakly, to increases in the supply of care home beds.9 A Norwegian study, using patient level data, reported that after controlling for casemix, hospital and time fixed effects, higher levels of social care capacity were associated with reduced hospital length of stay.10 A 2012 study demonstrated interactions in the use of hospital care and social care for older people in England and that residents of care homes tended to use hospitals less frequently than people receiving home care.31 A working paper published by the Institute of Fiscal Studies concluded that cuts in government spending on social care had led to increases in attendances at English accident and emergency departments, but not to subsequent hospital admissions.32

Implications for policy and research

We recommend further work on this subject to verify or rebut our results, to test the credibility of the mechanisms we have proposed to explain our results or to identify and test alternative mechanisms. We have made the panel dataset available with this paper to allow others to replicate the study’s methods and explore alternative approaches to test the relationship between emergency hospital admissions and spend on social care.

Observational studies that operate at the patient level may provide insights into the relationship between social care spend and hospital admissions that cannot be obtained via ecological studies. Qualitative research could be used to explore the mechanisms by which social care influences healthcare use.

Policy makers should review schemes, such as the Better Care Fund, which are predicated on the belief that emergency hospital admissions can be reduced by increasing social care provision. Our analysis suggests that this approach is unlikely to succeed. Given the scale of the Better Care Fund, a comprehensive evaluation is warranted. Healthcare commissioners should consider alternative, evidence-based methods of moderating demand for emergency hospital care.

We note that the NHS’s latest social care funding initiative, the Improved Better Care Fund, aims to reduce hospital length of stay and delayed transfers of care rather than prevent emergency admissions. Further research should be carried out to test the limited evidence supporting this approach.