Upload
others
View
0
Download
0
Embed Size (px)
Citation preview
Does pooling health & social care budgets reduce hospital use and lower costs?Dr Jonathan Stokes (Corresponding author) a
Dr Yiu-Shing Lau a
Dr Søren Rud Kristensen a,b
Prof Matt Sutton a
a. Health Organisation, Policy, and Economics, Centre for Primary Care and Health Services Research, University of Manchester, Manchester, UK
b. Faculty of Medicine, Institute of Global Health Innovation, Imperial College London, London, UK
Keywords: Pooled budgets; Integrated care; Better Care Fund; Multimorbidity; Health financing; Health policy; Organisation of care
Highlights
Pooled budgets can theoretically provide an added incentive to integrate care. There appears to be a small additional effect of pooling on outcomes. Effect is not in intended direction, but fits with integrated care literature.
1
1
2
3456789
10111213141516171819202122
23
242526
27
28
29303132
Abstract
An increasing burden of chronic disease and multimorbidity has prompted experimentation
with new models of care delivery that aim to improve integration across sectors and reduce
overall costs through decreased use of secondary care. One approach to stimulate this
change is to pool health and social care budgets to incentivise care delivery in the most
efficient location. The Better Care Fund is a large pooled funding initiative gradually taken up
by local areas in England between 2014 and 2015. We exploit this variation in timing of
uptake to examine the short- (1 year) and intermediate-term (up to 2 years) effects of the
Better Care Fund on seven measures of hospital use and costs from a cohort of 14.4 million
patients constructed using national Hospital Episode Statistics. We test for differential
effects on people with multimorbidity. We find no effects of budget pooling on secondary
care use for the whole population. For a multimorbid patients the use of bed days increased
in the short-term by 0.164 (4.9%) per patient per year. In the short- to intermediate-term,
pooling health and social care budgets does not reduce hospital use nor costs. However,
pooling funds does appear to stimulate additional integration activity.
2
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
1. Introduction
Long-term management of patients with chronic conditions, particularly those with
multimorbidity, requires co-ordination between multiple professionals, and across multiple
sectors (e.g. primary, secondary, tertiary and social care). Yet, these sectors have
traditionally been operating independently, with separate commissioners of services,
separate funding sources and payment mechanisms (Struckmann et al., 2017). New forms of
‘integrated care’ attempt to cross these traditional boundaries with the aim of providing
more effective and efficient care in the right place and time (NHS England, 2014; World
Health Organization, 2015).
The main metric for measuring the success of integrated care in the UK has been reducing
hospital activity (Monitor, 2013). Sustainability of current healthcare spending levels have
been questioned, and there is an assumption (although not fully tested) that primary and
social care is cheaper than secondary care. Reducing hospital activity, therefore, might be an
indication of a desired care substitution, or even a care prevention (i.e. increasing health)
effect. However, shrinking the hospital sector to the benefit of a growing primary and social
care sector is likely to be difficult unless incentives are aligned.
One perceived barrier to integrating care, then, is an incentives problem (Struckmann et al.,
2017). Multiple sector providers contribute in some way to a joint patient health outcome.
For example, a patients’ probability of having an emergency admission is affected by
previous activity in all of primary, secondary, and social care (Mason et al., 2015). However,
this interdependency between providers efforts and outcomes is not accounted for with
3
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
sector-specific payments and budgets meaning each provider is incentivised separately for
different activity/outcomes (Stokes et al., 2018b).
Pooling of budgets with all partners contributing to a shared fund for spending on agreed
projects or services is a mechanism that could theoretically ease this transition (Mason et
al., 2015). With shared financial resources, sectors are incentivised to make allocation
decisions in partnership towards achieving shared outcomes, accounting for
interdependency. In theory, this should lead to better integrated service delivery, and
reduced secondary care activity should be a measurable outcome of success. (Stokes et al.,
2018a)(Mason et al., 2015).
The Better Care Fund (BCF) is a large (£5.3 billion in 2015/16, and £5.8 billion 2016/17)
pooled health and social care funding scheme, mandated in England from April 2015. The
BCF was designed to generate measurable effects within one year of implementation,
principally reducing demand for hospital services, e.g. emergency admissions, and delayed
transfers of care. While the scheme was mandated within the 2015/16 financial year, a
number of local areas (Health and Wellbeing Boards) implemented the pooled budget
mechanism from 2014/15 (NHS England, 2016a).
We provide rigorous quasi-experimental evidence on the effects of pooled health and social
care funding. We exploit the gradual roll-out of the BCF in geographical space and over time
and examine effects on secondary care utilisation and costs using a dataset recording all
secondary care use in England.
4
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
1.1 Existing evidence on pooled funding
A recent systematic review analysed the international literature on the effects of integrating
funds for health and social care. They identified 38 schemes from eight countries. However,
the findings were inconsistent, with 33% of the schemes showing no effects on secondary
care costs or utilisation, 9% significantly lower utilisation, and the remaining 58% showing
mixed or unclear evidence. None of the evidence isolated the effect of pooled funding
alone; instead assessing impact of pooled funding plus simultaneous service delivery and
organisational changes (Mason et al., 2015).
Additionally, there are a number of more recent examples of integrated care that have
implemented some pooled budgeting. For example, Gesundes Kinzigtal in Germany has
implemented a ‘virtual’ pooled budget to incentivise multiple providers through shared
savings opportunities. An evaluation found that the programme led to increased hospital
admissions, but decreased length of stay, and an overall decrease in costs per patient per
year (Busse & Stahl, 2014). However, again it was not clear to which extent this was related
to the pooling of budgets as it was part of a package of changes.
In the US, Accountable Care Organisations (ACOs) have combined multiple independent
providers into a single provider (which might alone act to incentivise shared outcomes,
accounting for interdependency) plus used pooled budgets (sometimes termed ‘global
budgets’) and pay-for-performance to shift care priorities. Overall, ACOs found savings of
2.8 percent over two years follow-up (1.9 percent in year one and 3.3 percent in year two)
compared to non-ACOs (Z. Song et al., 2012). Those formed by independent primary care
5
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
groups gained consistently greater savings than hospital-integrated groups (McWilliams et
al., 2016). In the early cohort, differential reductions in spending were greater for high risk
patients than low risk (but with similar relative reductions in both groups), whereas in those
ACOs entering later differential reductions in spending and admissions were almost entirely
among low risk patients (McWilliams et al., 2017). A study examining longer term impacts of
an ACO in Massachusetts found lower spending growth and generally greater quality
improvements after four years, although factors beyond pooled budgets were thought to
have contributed, particularly in the later part of the study period (Zirui Song et al., 2014).
In summary, these previous studies have been unable to separate the impacts of the pooled
budgeting arrangements from the other service delivery and organisational changes that
occurred simultaneously. The BCF offers a unique opportunity to study the impact of pooled
budgets over and above other integration activity.
There have been only two studies of the BCF. An investigation by the government spending
watchdog suggested that the expected reductions in secondary care activity had not been
met (National Audit Office, 2017). However, the analysis compared realised activity rates
with the predictions that had been made for the programme in advance of its
implementation. Our analysis offers a more robust research design, exploiting the gradual
roll-out of the reform.
Another recent report used aggregated data at the hospital Trust-quarter level, and
exploited differences in the amount spent on BCF activity in the two years following the
mandated implementation (Forder et al., 2018). They identified a small reduction in delayed
6
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
transfers of care (0.073% for a 1% increase in BCF expenditure per capita), but no effect on
emergency admissions. This analysis focuses on the effects of variations in the extent to
which the BCF is used rather than its introduction.
We build on the previous literature by examining the effect of pooled funding over and
above other integration activity. We use individual-level data so that we can conduct
detailed risk adjustment and estimate effects for multimorbid patients who are most likely
to benefit from integrated care.
1.2 The Better Care Fund initiative
150 local Health and Wellbeing Boards were created in 2012. They brought social care
commissioners in Local Authorities together with representatives of the NHS (Clinical
Commissioning Groups - local healthcare commissioners), public health, and patients to plan
how to meet the health needs of their local population. In 2015, the BCF required these
Health and Wellbeing Boards to pool a proportion of their health and social care budgets
(National Audit Office, 2017). The BCF was not new or additional money, but a re-allocation
of Clinical Commissioning Group and Local Authority funds (Bennett & Humphries, 2014).
Therefore, any effects associated with the introduction of the BCF will be as a result of the
new pooling mechanism rather than additional funding.
Pooled budgets are intended to deliver their effects through added stimulation of integrated
care activity. The 2012 Health & Social Care Act in England mandated all Clinical
Commissioning Groups (CCGs) to ‘promote integration’ (Department of Health, 2012).
Therefore, all local areas might be expected to be integrating care in some form prior to the
BCF. There are various types of integrated care activity, but the most ubiquitous is case
7
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
management. This consists of individualised care planning and on-going intensive
management in primary care, frequently involving a multidisciplinary team and targeting
patients at high risk of secondary care utilisation (Stokes et al., 2016a). Indeed, case
management of high risk patients was financially incentivised nationally from financial year
2013, cementing its position as the dominant form of integrated care (NHS England, 2013).
Some areas were also funded to do additional integration efforts, for example through the
national ‘Pioneer’ and ‘Vanguard’ programmes (Vanguards were funded from 2015/16,
Pioneers in two waves from 2013/14 and 2015/16). These programmes have tended to build
on their case management activity with additional activity (e.g. health coaching) covering
the broader population and wider organisational changes (NHS England, 2016b, c). In our
analysis we control for this additional integration activity to capture the effect of pooling
funds over and above existing integrated care activity.
In 2015/16, £1 billion (nearly 20%) of the Fund was ring-fenced for out-of-hospital spending.
Each local area was also asked to target reduced emergency admissions and keep an
amount equal to the value of those admissions in a pay-for-performance pot, with areas
then able to spend these funds in line with their BCF plans depending on performance. In
the first year, local areas planned to reduce emergency admissions by 106,000 saving
potentially £171m, and planned to reduce delayed transfers of care by 293,000 days,
potentially saving £90m (National Audit Office, 2017).
8
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
2. Methods
2.1 Data
Secondary care use data
We used the Hospital Episode Statistics dataset containing details of all utilisation at NHS
hospitals in England (NHS Digital, 2017). We constructed a cohort of all patients having any
planned or emergency hospital admission in the two financial years prior to the first
adoption of the BCF (1st April 2011 to 31st March 2013). This ‘high risk’ cohort is targeted for
relatively standardised integrated care service delivery activity nationally in the form of case
management. Therefore, any effects associated with the BCF will reflect its role in
stimulating sites to ‘do more’ integration.
We created an individual-level dataset over seven financial years (annually, between 1st
April 2009 to 31st March 2016), counting all hospital admissions, outpatient visits and
emergency department attendances for each cohort patient per year. We costed each
measure of utilisation using the national tariff applicable in that year (NHS Improvement,
2017), and calculated the total cost of secondary care for each individual in each year.
Activity and costs were set to zero for any year with no recorded utilisation to make a
balanced panel. We used a pseudonymised patient identifier to link in mortality data from
the Office for National Statistics (ONS, 2017). We constructed a dummy variable equal to
one if a patient died in or out of hospital within the year and exited the sample (giving us a
legitimately unbalanced panel). Co-variates were set at the last non-missing observation for
each individual.
9
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
We created seven outcome measures: (i) number of emergency admissions for ambulatory
care sensitive conditions (see Appendix for list of codes); (ii) probability of re-admission
within 30 days of discharge; (iii) total bed-days; (iv) number of days of delayed discharge; (v)
number of ED attendances; (vi) number of outpatient visits; and, (vii) total cost of secondary
care.
Intervention status data
Individuals in our cohort were assigned to Health and Wellbeing Boards based on their
Lower Layer Super Output Area (LSOA) of residence. Missing Health and Wellbeing Board
(LSOA was missing for 5% of the sample) was updated based on GP practice (NHS England,
2017b). We obtained the list of GP practices that were Vanguard’s directly from NHS
England, and we created a dummy for Pioneer status based on CCG and information
available in the literature (Erens et al., 2015; Monitor, 2015). We included dummy variables
for Vanguard and Pioneer status.
Multimorbidity status
We assigned each individual to a multimorbidity class based on all of the ICD-10 codes
recorded for that patient on hospital admissions in the two years of data on which we
constructed the cohort. We drew on a list of 30 long-term conditions used in the previous
literature (see Appendix) (Tonelli et al., 2015). We constructed: 1) a dummy variable
identifying those with two or more conditions; and 2) a categorical variable identifying those
with (a) less than two conditions, (b) those with two or more conditions (only physical
health conditions, or only mental health conditions) and (c) those with two or more
10
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
conditions (including at least one mental health condition and at least one physical health
condition).
2.3 Analysis
Our full dataset included seven years of annual, individual-level data. Models were
estimated on a pool of 14,363,471 patients that had non-missing co-variates, including age,
sex, GP practice’s Vanguard status, Pioneer status, and Health and Wellbeing Board (local
area identifier). There were 1.35 million deaths in the sample between 2012/13 and
2015/16. We have five years of data in the pre-intervention period, with the intervention
introduced in the sixth of seven years. The intervention began in the 2014/15 financial year
when seventy-five percent of the Health and Wellbeing Boards pooled health and social care
funding under the BCF intervention (see Appendix). In 2015/16, all Health and Wellbeing
Boards had implemented the intervention. As our primary analysis, we report estimates
from an ordinary least squares (OLS) regression with high-definition fixed effects (Correia,
2016).
Short-term effects
Our primary analysis uses six years of data (85,073,282 observations), including up to the
2014/15 financial year when there is a single post intervention period where seventy-five
percent of Health and Wellbeing Boards implemented the BCF intervention and remaining
Health and Wellbeing Boards are the control group. We use a standard difference-in-
differences approach:
11
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
(1 ) y ijt=β0+ β1BC F j+β2POS T t+β3BC F j∗POST t+β4mult ii+x i γ+HWB j+Vanguard j+Pioneer j+Year t+ϵ ijt
In which y ijt is the outcome for individual i in Health and Wellbeing Board area j at time t ,
BCF jis a dummy for intervention status, POST tis a dummy for whether the observation is
in the post-intervention period (i.e. 2014/15), mult ii is a dummy indicating whether the
patient is multimorbid, vector x i is a set of indicators for categories of age group (19 five-
year bands to 85+) and gender, HWB j is a set of Health and Wellbeing Board fixed effects,
Vanguard j and Pioneer j are sets of Vanguard status fixed effects controlling for ‘other
integrated care activity’, Year t a set of time fixed effects and εijt is the idiosyncratic error
term. The coefficient β3 measures the effect of pooled payment within the first year of
uptake.
Intermediate-term effects
Our intermediate-term analysis deviates from the traditional difference-in-differences
approach and uses one additional year of post intervention period (98,361,352
observations). We include a binary indicator allowing for the gradual uptake of the
intervention across areas (where POOLjt is a dummy taking the value 1 for the area and time
where the pooled funds are introduced, and 0 otherwise) combined with Health and
Wellbeing Board area and time fixed effects:
(2 ) y ijt=β0+ β1 POOL jt+β2mult ii+x i γ+HWB j+Vanguard j+Pioneer j+Yeart+ϵ ijt
The coefficient β1 measures the effect of pooled payment. The effect estimate in this model
is a weighted average of both possible two-by-two difference-in-difference estimators,
12
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
where units serve as controls for each other during periods when their intervention status
stays constant (Goodman-Bacon, 2018). The estimate relies on a proportion of sites
switching intervention status over each time period, so we are not able to examine effects
beyond two years when BCF coverage is universal.
Subgroup (multimorbidity) analysis
Our subgroup analysis (both short- and intermediate-term run separately) adds additional
interaction terms to examine any differential effects by multimorbidity status. For example:
(3 ) y ijt=β0+β1BCF j+ β2 POST t+β3BC F j∗POS T t+ β4mult ii+mult ii∗( β¿¿5 BCF j+ β6 POST t+β7BC F j∗POS T t)+x i γ+HWB j+Vanguard j+Pioneer j+Year t+ϵijt ¿
The coefficient for the triple interaction term, β7, measures the differential effect of pooled
payment on multimorbid patients compared to non-multimorbid patients. We estimate this
regression separately for the two specifications of multimorbidity.
Robustness tests
The estimated effect is identified by the earlier adoption of the BCF in some areas. We test
for selection on observable characteristics by predicting early adoption of the BCF using a
logistic regression model at the Health and Wellbeing Board level. We use data from our
cohort from 2009/10-2013/14 on all measured outcomes, number of deaths, age and sex
profile, average morbidity status, average list size and Vanguard status as predictors.
One of the underlying assumptions of the difference-in-differences methodology is that the
trend in the control group is a suitable counterfactual for the trend in the intervention
13
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
group in the absence of the intervention. To examine this assumption we test whether the
trends of the intervention and control groups are statistically significantly different in the
pre-intervention period, i.e. whether outcomes follow parallel trends. We tested this
assumption by interacting the early-adopter intervention dummy with continuous time in
the pre-intervention period (i.e. 2009/10-2013/14).
Effects might be non-linear and occur at the top of the distribution (i.e. those with non-zero
utilisation). We also run each of the above analyses using two alternative models, 1)
regression on the inverse hyperbolic sine (IHS) transformed outcome variable, and; 2) a two-
part model, first a linear probability model predicting a non-zero outcome, followed by a
regression on the log-transformed outcome variable conditional on a non-zero outcome.
We cluster standard errors at the level of the intervention, the Health and Wellbeing Board.
We additionally use the mortality data to examine the influence of death as a competing risk
to utilisation of services, attempting to predict any differential change in death rate by BCF
group. We also drop all persons who die at any point over our follow-up period as an
additional check.
3. Results
3.1 Descriptive statistics
In Table 1 below, we compare descriptive variables for early and late intervention adopters
in our analysed dataset.
14
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
Table 1: Descriptive variable statistics for analysed data, based on 2014/15 intervention status.
[insert Table 1]
Both groups show similar unadjusted trends in the outcomes over time. As expected with
regression to the mean, patients have the highest hospital use and costs during the two
years upon which the cohort was constructed based on their admissions (see Appendix).
3.2 Pre-trends
Health and Wellbeing Boards with more ACSC admissions in the pre-period were more likely
to be early BCF adopters. All other covariates were not statistically significant predictors of
early adoption at the p<0.05 level.
The pre-trends analysis also suggested a significant difference for ACSC admissions (+0.01
admissions per patient per year for early-adopters in the pre-period). We found no
statistically significant differences in pre-trends for all other outcomes. However, graphical
analysis also suggested caution in interpreting the delayed discharges outcome (see
Appendix).
3.3 Overall effects of the BCF
Table 2 summarises the overall results for all seven outcomes considered. We found no
statistically significant effects of areas implementing the pooled budget at the 5%
significance level. At the 10% significance level, total bed days per patient may have
increased very slightly (+0.06 days per patient per year, 4% of the pre-intervention mean).
15
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
Our two-part model (see Appendix) suggests this increase was driven by those patients
already utilising admitted care services (i.e. those with a non-zero value).
Table 2: Overall pooled funding results.
[insert Table 2]
3.4 Multimorbidity subgroup results
Binary multimorbidity status
Table 3 summarises the results for the first measure of multimorbidity.
Table 3: Multimorbidity dummy subgroup results.
[insert Table 3]
For multimorbid patients we identified small increases in total bed days by 0.164 (4.9% of
the mean) per multimorbid patient per year in the short-term. The estimate was only
significant at the 10% level in the intermediate-term, 0.134 (4%).
Physical & mental comorbidity
Table 4: Physical mental co-morbidity subgroup results.
16
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
[insert Table 4]
For multimorbid patients, the differential increases in bed days were for those with only-
physical or only-mental health conditions in the short-term. The estimate for those with
both a physical and mental health condition was larger, but not statistically significant
(except in the inverse-hyperbolic sine model where the increase was significant at the 10%
level for this group in both the short- and intermediate-term – see Appendix).
3.5 Robustness of results across model specifications
The overall findings of a null effect were stable across the majority of robustness checks (see
Appendix).
For multimorbidity subgroups the estimates are qualitatively similar to the OLS model, but
the statistical significance does not always hold (see Appendix). For example, the increase in
bed days for multimorbid patients holds across the inverse-hyperbolic sine model, but only
at the 10% significance level.
Analysis of death as a competing risk showed no significant differential change in deaths by
early/late BCF adopters. When we excluded all patients who died, results were qualitatively
the same as reported above (see Appendix).
17
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
4. Discussion
We found no overall effects on seven outcomes for HWBs that implemented the pooled
health and social care funding intervention. Overall health system effects on hospital use
and cost are likely to be limited, at least in the short to intermediate-term.
However, we identified differential effects for the subgroup of individuals with
multimorbidity, likely to be most sensitive to the effects of integration of care. Those with
multimorbidity experienced small increases in total bed days in the short-term compared to
patients without multimorbidities. Despite not being in the intended direction, pooled
budgeting does seem to be having some additional effect of stimulating integration activity.
Limitations
This paper primarily sets out to examine the effects of pooled health and social care funding
as part of the BCF intervention roll-out. The effect of such pooling will act through
incentivising integration activity. The responses are likely to be heterogeneous but
unfortunately this activity is not measured in national datasets. There have been additional
pooled funding arrangements in certain localities (Humphries & Wenzel, 2015), but these
are not recorded in national datasets. We have controlled for integrated care Vanguard and
Pioneer status to try and alleviate this concern. In doing so, we assume that these areas,
recognised as leaders in the advancement of new models of integrated care (NHS England,
2016b), are those most likely to have additional pooled funding arrangements and be
delivering additional integration activity.
18
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
The BCF only pooled a small amount of total health service funding but at a time when social
care budgets were experiencing substantial cuts (Erens et al., 2015). Therefore, the results
may not be generalisable to pooling of larger amounts of funding or situations when both
sectors are experiencing funding growth. The policy assumed significant savings and
reduced utilisation within a short period of time (one year). We used a very large dataset
capable of detecting small effects of the intervention, but we were not able to detect any
reduced utilisation or costs of secondary care in the short or intermediate-term (up to two
years). We were able to identify differential subgroup effects for multimorbid patients,
where we expect any effects of integration to be most pronounced, with small increases in
utilisation of bed days in the short-term. The magnitude of results might differ depending on
the extent of pooling, and different effects across the distribution. Unfortunately,
information on the extent of pooling was not available for the BCF in the first year to test
this (2014/15). In robustness checks, we sought to identify different distributional effects.
Due to the way the BCF was rolled out, we were able to record differential uptake effects
only over a single year before coverage was mandated nationally. Therefore, our primary
results relate to short-term effects of pooled health and social care funding, where any
additional effects might take longer to emerge. We have tried to maximise use of available
data by estimating an intermediate-term effect using the gradual-uptake model. However,
this model has the added limitation that it assumes an instantaneous and constant effect
over time, so averages any treatment effect heterogeneity (Goodman-Bacon, 2018).
Additionally, we had a limited selection of outcome measures in the data available to us,
where effects of integrated care might be more easily captured with measures of patient
experience, for example (Ouwens et al., 2005). We included delayed discharge as an
19
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
outcome because it was a specific policy aim, but there are criticisms of its coding in HES
data, NHS Digital outlines that due to a submission error the field contains incorrect values
(NHS Digital, 2019). Consequently, we also look at total bed days because this does not
depend on whether the hospital stay is coded as a delayed discharge.
The voluntary early adoption of the intervention that we exploit in this analysis brings the
potential for selection bias. We analysed whether early adoption was related to local area
characteristics and identified few correlations though we may underestimate any beneficial
effect on ACSC admissions. There may also have been anticipation effects in areas that were
late-adopters, biasing the results towards null. However, we again expect any anticipation
effects to have been small and equally spread across both groups, as informal evidence from
policymakers suggests that the policy was not widely expected before its introduction.
Our multimorbidity measure was constructed from ICD-10 codes present in the inpatient
records of our cohort. We constructed the measure in the pre-intervention period and held
it constant to prevent any concerns of multimorbidity as a “bad control” (Angrist & Pischke,
2008) as we might expect the intervention to act to prevent further health deterioration
(including development of new disease, and so potentially classification of multimorbidity).
We would have preferred to have constructed these measures instead from primary care
data (or a combination of the two), where recording is likely to be better for chronic
conditions (Sigfrid et al., 2006). We would also have preferred total health system costs as a
dependant variable, rather than total costs of secondary care. Unfortunately, this linked
data was not available to us. We would therefore expect our multimorbidity measure to
20
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
underestimate the prevalence of multimorbidity in our cohort (Tonelli et al., 2015), and we
are unable to estimate effects on costs in other sectors, for example primary care.
Interpretation in context of the wider literature
Previous analysis of the BCF by the National Audit Office concluded that “the Better Care
Fund did not achieve its principal financial or service targets over 2015-16” and that
estimates of effects of the BCF and integration more generally have been “over optimistic”
(National Audit Office, 2017). Our analysis likewise shows that any cost-saving expectations
from the intervention are unlikely, especially in the short-term. The report also highlighted,
however, that 90% of local areas “agreed or strongly agreed that the delivery of the Better
Care Fund plans had a positive impact on integration locally” (National Audit Office, 2017).
Likewise, the fact that we find any significant effect at all suggests that the pooled funding
has some effect on ability to integrate.
A system-level evaluation of the BCF found that those sites spending on activities classified
as ‘intermediate care’ and ‘prevention activities’ also appeared to be more effective than
other forms (Forder et al., 2018). So, there may be more effective ways of implementing the
intervention than others.
A recent systematic review found that a number of pooled funding schemes reported
improved access to care. Some of these schemes reported increased total costs as a result
of identifying substantial levels of unmet need (Mason et al., 2015). Our results would fit
with these findings, showing increased utilisation of bed days in the short-term for the most
complex patients.
21
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
The mechanism of effect for a pooled health and care funding intervention is via
incentivising additional integrated care activity. As outlined in the introduction, in England,
to date, this has been primarily through case management. The evidence for effectiveness
of case management likewise shows that increased integration could increase utilisation and
cost, particularly for the highest risk patients (McWilliams & Schwartz, 2017; Roland & Abel,
2012; Snooks et al., 2018; Stokes et al., 2016b; Stokes et al., 2017; Stokes et al., 2015). The
current evidence for integrated care more widely also matches with these results (Baxter et
al., 2018; Nolte & Pitchforth, 2014).
Implications for policy and practice
It is an unsurprising finding, then, that incentivising integration activity through the BCF
leads to similar results as we find for the integration activity itself. However, our results
show that pooled funding does appear to provide an additional effect (i.e. might help drive
more integrated care activity). The most common aim of integrated care (at least as
identified in UK-based policy) is reducing demand and costs (Hughes, 2017). There is,
therefore, the need for policymakers to identify and recommend interventions that are able
to achieve these aims.
However, current policy speculation is that to achieve increased efficiency, wider
organisational change is needed, with particular emphasis on formation of single provider
organisations, ACOs (most recently re-named integrated care systems in the NHS) (NHS
England, 2019). The BCF, or a similar pooled funding approach, might be a necessary step to
22
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
allow these ACOs to function as intended, and so ultimately allow the outcomes to be
achieved in the long-term. There is the possibility, for instance, that pooled funding could be
more useful for incentivising more preventative care with this population health
management-based approach (NHS England, 2017a), and/or other service delivery
interventions that might affect outcomes differently. However, the early evidence on ACOs
from the USA has so far been mixed (McClellan et al., 2015; McWilliams et al., 2016). Where
savings have occurred, they have not been attributed to better co-ordination but rather to
reducing waste (McWilliams, 2016).
Current results might reflect current trends in integrated care delivery, therefore, but might
allow for improvement in future if better models can be stimulated by the pooled funding
itself.
There do not appear to be beneficial overall effects of pooled health and social care funding
through the BCF. There appear to be some differential effects by multimorbidity subgroup,
with findings in line with the integrated care interventions that pooled funding currently
incentivises. In the short term, pooling health and social care budgets alone does not appear
to reduce hospital use nor costs but does appear to additionally stimulate integration
activity.
23
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
References
Angrist, J.D., & Pischke, J.-S. (2008). Mostly harmless econometrics: An empiricist's companion: Princeton university press.
Baxter, S., Johnson, M., Chambers, D., Sutton, A., Goyder, E., & Booth, A. (2018). The effects of integrated care: a systematic review of UK and international evidence. BMC Health Services Research, 18, 350.
Bennett, L., & Humphries, R. (2014). Making best use of the Better Care Fund: spending to save. London: The Kings Fund.
Busse, R., & Stahl, J. (2014). Integrated care experiences and outcomes in Germany, the Netherlands, and England. Health Affairs, 33, 1549-1558.
Correia, S. (2016). A feasible estimator for linear models with multi-way fixed effects. Duke University Preliminary Version. URL: www. scorreia. com/research/hdfe. pdf.
Department of Health. (2012). Health and Social Care Act 2012. London, UK: Department of Health.
Erens, B., Wistow, G., Mounier-Jack, S., Douglas, N., Jones, L., Manacorda, T., et al. (2015). Early evaluation of the Integrated Care and Support Pioneers Programme. London, UK: Policy Innovation Research Unit.
Forder, J., Caiels, J., Harlock, J., Wistow, G., Malisauskaite, G., Peters, M., et al. (2018). A system-level evaluation of the Better Care Fund: Final Report. pp. 1-140).
Goodman-Bacon, A. (2018). Difference-in-differences with variation in treatment timing. National Bureau of Economic Research.
24
541
542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574
Hughes, G. (2017). New models of care: the policy discourse of integrated care. People Place and Policy, 11, 72-89.
Humphries, R., & Wenzel, L. (2015). Options for integrated commissioning: Beyond Barker.
Mason, A., Goddard, M., Weatherly, H., & Chalkley, M. (2015). Integrating funds for health and social care: an evidence review. J Health Serv Res Policy, 20, 177-188.
McClellan, M., Kocot, S., & White, R. (2015). Early evidence on medicare acos and next steps for the medicare aco program (updated). Health Affairs Blog.
McWilliams, J.M. (2016). Cost Containment and the Tale of Care Coordination. New England Journal of Medicine, 375, 2218-2220.
McWilliams, J.M., Chernew, M.E., & Landon, B.E. (2017). Medicare ACO Program Savings Not Tied To Preventable Hospitalizations Or Concentrated Among High-Risk Patients. Health Aff (Millwood), 36, 2085-2093.
McWilliams, J.M., Hatfield, L.A., Chernew, M.E., Landon, B.E., & Schwartz, A.L. (2016). Early Performance of Accountable Care Organizations in Medicare. New England Journal of Medicine, 374, 2357-2366.
McWilliams, J.M., & Schwartz, A.L. (2017). Focusing on High-Cost Patients — The Key to Addressing High Costs? New England Journal of Medicine, 376, 807-809.
Monitor. (2013). Closing the NHS funding gap: how to get better value health care for patients pp. 1-23).
Monitor. (2015). Integrated Care Pioneers.National Audit Office. (2017). Health and social care
integration.NHS Digital. (2017). Hospital Episode Statistics.NHS Digital. (2019). HES data quality notes.
25
575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610
NHS England. (2013). 2013/14 general medical services (GMS) contract - guidance and audit requirements for new and amended enhanced services: Risk profiling and care management scheme. In N. Employers (Ed.).
NHS England. (2014). Five Year Forward View. pp. 1-41).
NHS England. (2016a). Better Care Fund.NHS England. (2016b). New care models - vanguard
sites.NHS England. (2016c). People helping people: Year two
of the pioneer programme. pp. 1-45).NHS England. (2017a). ACOs and the NHS
commissioning system.NHS England. (2017b). Better Care Fund.NHS England. (2019). The NHS Long Term Plan.NHS Improvement. (2017). National tariff payment
system Nolte, E., & McKee, M. (2008). Caring for people with
chronic conditions: a health system perspective. McGraw-Hill International.
Nolte, E., & Pitchforth, E. (2014). What is the evidence on the economic impacts of integrated care.
Omran, A.R. (1971). The epidemiologic transition: a theory of the epidemiology of population change. The Milbank Memorial Fund Quarterly, 49, 509-538.
ONS. (2017). Deaths.Ouwens, M., Wollersheim, H., Hermens, R., Hulscher,
M., & Grol, R. (2005). Integrated care programmes for chronically ill patients: a review of systematic reviews. International Journal for Quality in Health Care, 17, 141-146.
Roland, M., & Abel, G. (2012). Reducing emergency admissions: are we on the right track? BMJ, 345.
Sigfrid, L.A., Turner, C., Crook, D., & Ray, S. (2006). Using the UK primary care Quality and Outcomes Framework to audit health care equity: preliminary
26
611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647
data on diabetes management. Journal of Public Health, 28, 221-225.
Snooks, H., Bailey-Jones, K., Burge-Jones, D., Dale, J., Davies, J., Evans, B., et al. (2018). Predictive risk stratification model: a randomised stepped-wedge trial in primary care (PRISMATIC).
Song, Z., Rose, S., Safran, D.G., Landon, B.E., Day, M.P., & Chernew, M.E. (2014). Changes in Health Care Spending and Quality 4 Years into Global Payment. New England Journal of Medicine, 371, 1704-1714.
Song, Z., Safran, D.G., Landon, B.E., Landrum, M.B., He, Y., Mechanic, R.E., et al. (2012). The ‘Alternative Quality Contract,’ Based On A Global Budget, Lowered Medical Spending And Improved Quality. Health Affairs, 31, 1885-1894.
Stokes, J., Checkland, K., & Kristensen, S.R. (2016a). Integrated care: theory to practice. J Health Serv Res Policy, 21, 282-285.
Stokes, J., Kristensen, S.R., Checkland, K., & Bower, P. (2016b). Effectiveness of multidisciplinary team case management: difference-in-differences analysis. BMJ Open, 6, e010468.
Stokes, J., Kristensen, S.R., Checkland, K., Cheraghi-Sohi, S., & Bower, P. (2017). Does the impact of case management vary in different subgroups of multimorbidity? Secondary analysis of a quasi-experiment. BMC Health Services Research, 17, 521.
Stokes, J., Panagioti, M., Alam, R., Checkland, K., Cheraghi-Sohi, S., & Bower, P. (2015). Effectiveness of Case Management for 'At Risk' Patients in Primary Care: A Systematic Review and Meta-Analysis. PLoS ONE, 10, e0132340.
Stokes, J., Riste, L., & Cheraghi-Sohi, S. (2018a). Targeting the ‘right’ patients for integrated care: stakeholder perspectives from a qualitative study. J Health Serv Res Policy, 23, 243-251.
27
648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684
Stokes, J., Struckmann, V., Kristensen, S.R., Fuchs, S., van Ginneken, E., Tsiachristas, A., et al. (2018b). Towards incentivising integration: A typology of payments for integrated care. Health Policy, 122, 963-969.
Struckmann, V., Quentin, W., Busse, R., & Van Ginneken, E. (2017). How to strengthen financing mechanisms to promote care for people with multimorbidity in Europe? The European Observatory on Health Systems and Policies.
Tonelli, M., Wiebe, N., Fortin, M., Guthrie, B., Hemmelgarn, B., James, M., et al. (2015). Methods for identifying 30 chronic conditions: application to administrative data. BMC Medical Informatics and Decision Making, 15, 31.
Wallace, E., Salisbury, C., Guthrie, B., Lewis, C., Fahey, T., & Smith, S.M. (2015). Managing patients with multimorbidity in primary care. BMJ, 350.
World Health Organization. (2015). WHO global strategy on people-centred and integrated health services: interim report.
Table 2: Descriptive statistics (based on early/late intervention adoption)
Early adopters Late adopters Totaln = (observations) 78,576,826 19,784,526 98,361,352n = (individuals) 11,476,318 2,886,650 14,362,968Multimorbid (observations)
22,480,283 (28.6%)
5,585,771 (28.2%)
28,066,054(28.5%)
Aged over 65 years (observations)
22,319,530(28.4%)
5,418,872(27.4%)
27,738,402(28.2%)
Female (observations) 40,558,603(51.6%)
10,314,756(52.1%)
50,873,359(51.7%)
ACSC admissions (mean[SD]) 0.05 (0.34) 0.04 (0.31) 0.05 (0.33)% zeros 96.52 96.62 96.54
Bed days (mean[SD]) 1.48 (13.05) 1.41 (12.88) 1.46 (13.02)
28
685686687688689690691692693694695696697698699700701702703704705706
707
708
709
710
% zeros 82.66 82.97 82.72Re-admission within 30 days
(mean[SD]) 0.03 (0.18) 0.03 (0.18) 0.03 (0.18)
% zeros 96.71 96.77 96.72Days of delayed discharge (mean[SD]) 0.008 (10.64) 0.006 (9.43) 0.008 (10.41)
% zeros 99.97 99.98 99.98ED attendances (mean[SD]) 0.50 (1.31) 0.53 (1.42) 0.51 (1.33)
% zeros 72.20 71.40 72.04Outpatient visits (mean[SD]) 3.42 (6.42) 3.56 (7.08) 3.44 (6.56)
% zeros 44.23 44.14 44.21Total cost of secondary care
(£, mean[SD]) 1209.89(4125.88)
1205.60(5962.82)
1209.03(4555.27)
% zeros 38.83 38.78 38.82Deaths n (%) 1,089,017
(9.49%)261,770 (9.07%)
1,350,787 (9.40%)
29
Table 2: Difference-in-differences estimates of the effects of pooled funding.
ACSC admissions
Bed days Delayed discharges
ED attendances
Outpatient visits
Total cost secondary care (£)
Probability of re-admission
within 30 days
Short-termBCF*Post 4.95e-05 0.061* 0.005 0.001 -0.173 1.647 -2.4e-05
(0.0004) (0.034) (0.004) (0.009) (0.224) (17.923) (0.0004)
Intermediate-termPooled budget 0.0002 0.049 0.003 0.0004 -0.132 3.311 3.09e-06
(0.0004) (0.031) (0.004) (0.008) (0.198) (16.179) (0.0004)
OLS estimates with cluster robust standard errors in parentheses. All ‘short-term’ models based on 87,073,282 observations, ‘intermediate-term’ models based on 98,361,352 observations. All models include indicators for interactions of sex and age group, multimorbidity status, years, Vanguard status, Pioneer status, and areas. *** p<0.01, ** p<0.05, * p<0.1
30
711712
713714715716
Table 3: Difference-in-differences estimates of the effects of pooled funding. Multimorbidity dummy subgroup.
ACSC admissions
Bed days Delayed discharges
ED attendances
Outpatient visits
Total cost secondary care (£)
Probability of re-admission
within 30 days
Short-termBCF*Post 5.99e-05 0.018 -0.001 -2.81e-05 -0.066 3.192 8.61e-05
(0.0002) (0.020) (0.002) (0.009) (0.110) (7.706) (0.0003)BCF*Post -5.83e-05 0.164** 0.020 0.001 -0.426 -12.960 -0.001*Multimorbid (0.001) (0.079) (0.013) (0.007) (0.452) (44.012) (0.0009)
Intermediate-termPooled budget 0.0001 0.015 -0.001 -0.001 -0.049 2.527 8.55e-05
(0.0002) (0.019) (0.002) (0.008) (0.095) (6.632) (0.0002)Pooled budget 0.0002 0.134* 0.018 0.002 -0.343 -3.826 -0.0005*Multimorbid (0.001) (0.071) (0.011) (0.006) (0.408) (41.055) (0.0008)
OLS estimates with cluster robust standard errors in parentheses. All ‘short-term’ models based on 87,073,282 observations, ‘intermediate-term’ models based on 98,361,352 observations. All models include indicators for interactions of sex and age group, multimorbidity status, years, Vanguard status, Pioneer status, and areas. *** p<0.01, ** p<0.05, * p<0.1
31
717718
719720721722
Table 4: Difference-in-differences estimates of the effects of pooled funding. Physical mental co-morbidity subgroup.
OLS estimates with cluster robust standard errors in parentheses. All ‘short-term’ models based on 87,073,282 observations, ‘intermediate-term’ models based on 98,361,352 observations. All models include indicators for interactions of sex and age group, multimorbidity status, years, Vanguard status, Pioneer status, and areas. *** p<0.01, ** p<0.05, * p<0.1
ACSC admissions
Bed days Delayed discharges
ED attendances
Outpatient visits
Total cost secondary care (£)
Probability of re-admission
within 30 days
Short-termBCF*Post 6.01e-05 0.018 -0.001 -2.32e-05 -0.066 3.195 8.64e-05
(0.0002) (0.020) (0.002) (0.009) (0.110) (7.709) (0.0003)BCF*Post -0.0004 0.111** 0.021 0.0009 -0.332 -6.596 -0.0005*1.Physical/mental (0.001) (0.055) ( 0.014) ( 0.005) ( 0.414) ( 39.175) (0.0008)BCF*Post 0.001 0.286 0.017 -0.004 -0.671 -35.491 -0.001*2.Physical/mental (0.002) (0.197) (0.027) (0.016) (0.565) (61.414) (0.002)
Intermediate-termPooled budget 0.0001 0.015 -0.001 -0.001 -0.049 2.528 8.56e-05
(0.0002) (0.019) (0.002) (0.008) (0.095) (6.634) (0.0002)Pooled budget -6.92e-05 0.093 0.019 0.002 -0.272 5.151 -0.0004*1.Physical/mental (0.001) (0.049) (0.012) (0.005) (0.373) (37.047) (0.0007)Pooled budget 0.001 0.228 0.015 -0.001 -0.531 -32.255 -0.001*2.Physical/mental (0.002) (0.184) (0.023) (0.014) (0.510) (56.285) (0.002)
32
723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754
755