Does integrated care reduce hospital activity for patients with chronic diseases? An umbrella review of systematic reviews

Objective To summarise the evidence regarding the effectiveness of integrated care interventions in reducing hospital activity. Design Umbrella review of systematic reviews and meta-analyses. Setting Interventions must have delivered care crossing the boundary between at least two health and/or social care settings. Participants Adult patients with one or more chronic diseases. Data sources MEDLINE, Embase, ASSIA, PsycINFO, HMIC, CINAHL, Cochrane Library (HTA database, DARE, Cochrane Database of Systematic Reviews), EPPI-Centre, TRIP, HEED, manual screening of references. Outcome measures Any measure of hospital admission or readmission, length of stay (LoS), accident and emergency use, healthcare costs. Results 50 reviews were included. Interventions focused on case management (n=8), chronic care model (CCM) (n=9), discharge management (n=15), complex interventions (n=3), multidisciplinary teams (MDT) (n=10) and self-management (n=5). 29 reviews reported statistically significant improvements in at least one outcome. 11/21 reviews reported significantly reduced emergency admissions (15–50%); 11/24 showed significant reductions in all-cause (10–30%) or condition-specific (15–50%) readmissions; 9/16 reported LoS reductions of 1–7 days and 4/9 showed significantly lower A&E use (30–40%). 10/25 reviews reported significant cost reductions but provided little robust evidence. Effective interventions included discharge management with postdischarge support, MDT care with teams that include condition-specific expertise, specialist nurses and/or pharmacists and self-management as an adjunct to broader interventions. Interventions were most effective when targeting single conditions such as heart failure, and when care was provided in patients’ homes. Conclusions Although all outcomes showed some significant reductions, and a number of potentially effective interventions were found, interventions rarely demonstrated unequivocally positive effects. Despite the centrality of integrated care to current policy, questions remain about whether the magnitude of potentially achievable gains is enough to satisfy national targets for reductions in hospital activity. Trial registration number CRD42015016458.

Results: 50 reviews were included. Interventions focused on case management (n=8), chronic care model (CCM) (n=9), discharge management (n=15), complex interventions (n=3), multidisciplinary teams (MDT) (n=10) and self-management (n=5). 29 reviews reported statistically significant improvements in at least one outcome. 11/21 reviews reported significantly reduced emergency admissions (15-50%); 11/24 showed significant reductions in all-cause (10-30%) or condition-specific (15-50%) readmissions; 9/16 reported LoS reductions of 1-7 days and 4/9 showed significantly lower A&E use (30-40%). 10/25 reviews reported significant cost reductions but provided little robust evidence. Effective interventions included discharge management with postdischarge support, MDT care with teams that include condition-specific expertise, specialist nurses and/or pharmacists and self-management as an adjunct to broader interventions. Interventions were most effective when targeting single conditions such as heart failure, and when care was provided in patients' homes.

INTRODUCTION
Hospital activity continues to rise and currently accounts for almost half of annual NHS expenditure. 1 Demands on the acute sector are strongly influenced by the rapidly growing number of patients with multiple, chronic health conditions. These patients often need to access multiple health and social care settings but typically experience fragmented and poorly coordinated care. 2 3 Reducing hospital activity is seen as the key to relieving pressure on services that are rapidly approaching their limits, 4 and integrated care has become a cornerstone of the policy response to this challenge in the UK and most other developed countries. Integrated care represents an organising principle for care delivery that aims to Strengths and limitations of this study ▪ This umbrella review is the first of its kind since integrated care became central to healthcare policy. ▪ Outcomes were selected following consultation with service providers, commissioners and patient representatives to ensure relevance. ▪ We assessed a large volume of international evidence across diverse chronic conditions, interventions and outcomes. ▪ Umbrella reviews do not allow conclusions to be drawn about the detailed contexts in which interventions were implemented, but they do permit a broader overview of the evidence base than would be possible with a focus on primary research alone. ▪ Heterogeneity of intervention design, duration, intensity and follow-up prohibited meta-synthesis across reviews.
improve patient experience of services through improved coordination across and between settings. 5 By facilitating more patient contact, treatment and follow-up in primary care, in the community or in patients' homes, integration aims to reduce substantially the number of emergency and other admissions to hospital and facilitate timely and effective discharge from hospital to other settings. Following the establishment of a series of integrated care 'pioneers' in 2013, hospital trusts and commissioning organisations in England are planning and investing in a plethora of integrated services via the Better Care Fund (BCF), which aims to promote joint working at a strategic and operational level. 6 Following the NHS Five Year Forward View, 7 there are also proposals to develop and implement new models of care with integration as their central principle. 8 Integration undoubtedly has laudable aims-poor care coordination is often the main problem cited by patients when describing their experiences of health and social care services. 9 10 NHS staff also welcome integration, 11 12 yet evidence about the effectiveness of integrated care in reducing healthcare resource use, particularly within the acute sector, is limited. Integrated care programmes can have a positive effect on service quality, 13 and there is emerging evidence from recent evaluations of integrated care pilots that suggests potential for service efficiencies. 14 15 However, there is still uncertainty about which interventions are most effective and how these should be implemented, 16 alongside persistent questions over whether the aims of integration are ultimately achievable in any meaningful way. 17 Given this uncertainty, it is timely to assess the evidence. This paper reports the findings of an umbrella review of the evidence for integrated care interventions operating across health and/or social care settings for chronic disease management in order to assess: (1) whether integration reduces hospital activity, (2) which interventions are the most promising, for which patients and in which settings, and (3) what are the associated cost implications.

METHODS
Umbrella reviews synthesise evidence from multiple systematic reviews into a single 'meta review', using the findings and conclusions of included systematic reviews as the raw data. They are useful when the evidence base is broad and are of particular importance for decision makers who need a synthesis of the most current and reliable data relevant to their context. 18 The protocol was published 19 and registered on PROSPERO.

Inclusion criteria
We included systematic reviews and meta-analyses published since January 2000 that evaluated interventions designed to facilitate integrated health and/or social care services. The year 2000 was chosen following scoping searches that indicated little or no systematic review evidence for integrated care interventions before this date. Eligible reviews could include primary studies of any experimental or quasi-experimental study design, providing the authors had identified studies using systematic methods. Eligibility was limited to reviews available in English.
Participants included adult patients with one or more chronic conditions. A list of 11 specific conditions was derived following a scoping review and combined a series of conditions recommended as central to any systematic review of chronic disease 20 21 and those included in the most recent Health Survey for England. 22 The resulting conditions (hypertension, depression, diabetes, coronary heart disease, stroke, transient ischaemic attack, chronic obstructive pulmonary disease (COPD), cancer, heart failure, dementia and arthritis) covered those that are most prevalent within the adult population, most costly to manage and most likely to occur in combination with other chronic conditions.
Interventions could be implemented in any health or social care setting ( primary, secondary or community care), as long as they crossed the boundary between two or more settings. The community setting encompassed care given in the community, in patient homes or by social care professionals. Exclusion criteria were: palliative care interventions; purely psychosocial interventions or those related to spirituality, mindfulness, health literacy or the use of complementary and alternative medicines; interventions focusing solely on diet and lifestyle factors; treatment or medication adherence; the effectiveness of surgical or diagnostic techniques; caregivers; pregnancy, and interventions implemented in less economically developed countries.
Comparison groups could include usual care, no intervention or comparison to one or more other interventions.

Outcome measures
Outcome measures were selected following a scoping review, a stakeholder workshop attended by service providers and commissioners and consultation with a group of patient and public involvement (PPI) advisors. Eligible reviews assessed one or more of the following outcomes: acute sector activity (emergency hospital admissions/readmissions, length of hospital stay, accident and emergency (A&E) use) and healthcare costs.

Search strategy
The search strategy was intentionally broad and included general terms related to chronic disease, multimorbidities and long-term conditions as well as MeSH terms for the 11 specific chronic diseases identified from scoping searches. Search terms associated with integrated care and known interventions were also included. A separate search identified systematic reviews that assessed the cost implications of integrated care interventions (see online supplementary information for MEDLINE search strategy). Relevant reviews were identified by searching electronic bibliographic databases and the manual checking of each included review's reference list. We searched MEDLINE, Embase, ASSIA (Applied Social Sciences Index and Abstracts), PsycINFO, Health Management Information Consortium database (HMIC), CINAHL, Cochrane library (including the Health Technology Assessment (HTA) database, Cochrane Database of Systematic Reviews, Database of Abstracts of Reviews of Effectiveness-DARE), EPPI-Centre library, TRIP database and the Health Economics Evaluations Database (HEED). Searches were performed in July 2014 and updated in December 2015.
Eligibility assessment and data extraction Two authors (SD and SF) independently screened titles and abstracts against the inclusion and exclusion criteria, and full text copies of all potentially relevant reviews were assessed. Disagreements were resolved through the independent assessment of a third author (GC). Where multiple versions of an eligible review were available, the most recent or most comprehensive version was included. Where the same review was published more than once (eg, Cochrane Collaboration review and subsequent update), the updated version was included. Data on review characteristics (databases searched, geographical scope, healthcare settings and disease(s) focused on), methodology (aim, research questions, number of studies included, review type), study participants, interventions and outcomes of interest were extracted from each included review and crosschecked by SD and SF according to a predefined data extraction sheet. For narrative reviews, a statement summarising the authors' primary interpretation of findings was extracted. For meta-analyses, data on relative risks or ORs were extracted along with the corresponding 95% CIs.

Quality assessment
Review quality was appraised independently by SD and SF using the Centre for Evidence-Based Medicine (CEBM) tool for critical appraisal of systematic reviews (http://www.cebm.net/wp-content/uploads/2014/04/SR_ Appraisal_sheet_2005_English.doc) which scores each review between 0 (poor quality) and 5 (high quality). Using quality score as an exclusion criterion was not part of the protocol, but considerable quality differences were evident between reviews scoring 0 to 2.5 and those scoring 3 or above. Lower quality reviews had little (if any) extractable outcomes data so we decided to exclude reviews scoring 2.5 or less on the quality scale.

Data analysis
Heterogeneity in study populations, interventions assessed, follow-up periods and specification of control groups prevented pooling of intervention effects and quantitative meta-synthesis across reviews. Owing to this, and to avoid the risk of 'double counting' evidence where multiple reviews contained some of the same primary studies, our synthesis was a primarily narrative review of interventions and outcomes. The strength of evidence from each meta-analysis or narrative review was characterised according to four categories: definite positive or negative associations, mixed findings or no association (table 1). Figure 1 summarises the search. A total of 11 436 potentially eligible reviews were identified, and 50 (in 49 papers) were included (table 2). A total of 1208 individual primary studies were included in the reviews (median 19, range 4-153). Nineteen reviews did not specify patient numbers, but across the 31 that did, all Table 1 Categorisation of the strength of effect for included reviews

Category Symbol Interpretation
Positive association + At least half of a review's included primary studies showed a statistically significant increase in a particular outcome following the intervention; the authors' summary of findings (narrative reviews) demonstrated a positive association, or pooled results from a meta-analysis indicated a statistically significant positive association, eg, hospital admission rates significantly increased. Negative association

−
At least half of a review's included primary studies showed a statistically significant decrease in a particular outcome following the intervention; the authors' summary of findings (narrative reviews) demonstrated a negative association, or pooled results from a meta-analysis indicated a statistically significant negative association, eg, hospital admission rates significantly reduced.

Mixed findings ?
A review reported mixed findings, in which some primary studies may have shown a statistically significant difference between intervention and control groups whereas other primary studies showed no significant differences between groups. No association = A review where no significant differences between intervention and control groups were reported in any of the included primary studies. but three included 1000 or more patients (total 219 475, median 2692, range 857-78 590). Studies within reviews varied in duration from 2 weeks to 60 months, with most lasting up to 12 months, although 9 reviews did not specify follow-up duration. Furthermore, 21 reviews were narrative, 23-42 26 included meta-analyses 43-68 and 3 were reviews of reviews. [69][70][71] The most commonly studied condition was chronic disease (n=15), 23-28 37 39-41 43 57 69 70 followed by heart failure (n=14), 36 30 32 mental health (n=1) 51 and heart failure and COPD combined (n=1). 71 All reviews were published between 2004 and 2015. Reviews were published in Canada, 26 31 37 38 41 42 57 63 65-67 71 the UK, 24  43-45 52 56 60-62 64 the USA, 23 30 32 33 46 51 53 58 59 the  Netherlands, 25 27-29 34 48 49 68 Ireland, 39 40 Switzerland, 50 70 Norway, 54 Japan, 69 Hong Kong, 36 Spain, 47 Denmark 35 and Greece. 55 In most reviews, the comparator was usual clinical care, although a detailed description of usual care was typically not provided. Overall, 29 reviews (58%) reported a nominally statistically significant result for at least one outcome.

Quality of included reviews
The mean quality assessment (QA) score was 4/5. Twelve reviews scored 5/5 (24%). 39 66 68 The criterion for which the largest number of reviews failed to score a point related to whether a valid consideration of bias across primary studies had been undertaken. There was no discernible trend in review quality across intervention categories: the mean QA scores by the intervention group ranged from 3.4/5 (case management) to 4.2/5 (chronic care model (CCM), multidisciplinary teams (MDT), selfmanagement).

Effects by intervention type
Interventions were categorised into six broad groups (table 3), although intervention components frequently overlapped.
Eight reviews focused on case management interventions. 23-27 43-45 With the exception of one review which showed that case management was associated with significantly reduced healthcare costs, 26 and another that demonstrated a 49% relative risk reduction (RRR) in admissions for patients with heart failure, 45 all case management reviews showed mixed findings or no   Open Access Open Access  Open Access     Open Access Open Access

Continued
Open Access

Continued
Open Access 20 Liaison meetings attended by specialists and primary care staff to discuss and plan ongoing patient management; shared care record carried by the patient, computer-assisted shared care and email with data available to primary and secondary care Primary, secondary Admissions (7 studies): Mixed results. Intervention was associated with a reduction in admissions in older patients and those with higher baseline morbidity. Costs (11 studies): 3 performed full economic analyses, of which 2 reported incremental cost savings in intervention. Seven studies reported direct costs: 1 showed higher costs in intervention; 6 reported mixed results (4/6 showed intervention more expensive than control, 2/6 reported lower costs in intervention).

Continued
Open Access    association between the intervention and outcomes assessed. Of nine reviews focusing on interventions comprising one or more components of the CCM, six reported positive findings for at least one outcome. 28 46-48 50 69 All CCM reviews reported that interventions with multiple components were significantly more effective than single component interventions at reducing admission rates, 46 49 50 69 with reductions of 22-32% observed in reviews that performed meta-analysis. Multicomponent interventions were also successful in reducing readmissions by 15-30%, 47 length of hospital stay by 2-4 days 46 48 and A&E visits by 42%. 46 Fifteen reviews assessed discharge management interventions, predominantly focusing on readmission rates and length of stay (LoS). Six reviews reported significant reductions in readmission rates for patients with heart failure, 53 55 58 59 COPD 54 and general chronic diseases. 57 Reductions ranged from 15% 55 to 66%. 53 In contrast, discharge management for patients who had stroke was notably ineffective in reducing readmission rates, 32 35 52 56 although LoS reduced by 7.7 days in one stroke review. 56 Three reviews assessed complex interventions. One demonstrated a 32% reduction in A&E use, 60 another reported a 43% reduction in heart failure-related readmissions 61 and a review of reviews reported positive findings for admissions, readmissions, LoS and A&E use (no effect sizes given). 70 Ten reviews assessed MDT interventions. Although team composition varied, MDT were generally effective when used for patients with single conditions, showing a 26-31% reduction in admission rates for heart failure 62-64 and a 33% RRR for admissions in patients with COPD. 38 MDT were also associated with a 42% reduction in heart failure readmissions, 66 a 2-day reduction in LoS, 62 65 66 significantly reduced A&E use 65 and significantly lower healthcare costs. 64 Conversely, MDT for general chronic disease management showed mixed effectiveness or no significant association for any outcomes, 37 39 40 suggesting that the crucial component of an effective MDT is the inclusion of condition-specific specialist expertise in the team skill mix. Finally, five reviews assessed self-management interventions. Three showed either mixed findings 40 or no association between intervention and outcomes assessed. 41 67 The remaining two demonstrated significant reductions in readmission rates and healthcare costs for patients with heart failure 42 and significantly lower admission rates for COPD. 68 Hospital admissions Emergency admission rates were assessed in 21 reviews across five intervention categories (table 4). Eleven reviews reported significantly reduced admissions, 38   Based on implementation of a collaborative process between one or more care coordinators or case managers and the patient, to assess, plan and facilitate service delivery for patients with chronic diseases, particularly when transitions across healthcare settings are required Chronic care model (n=9) 28 29 46-51 69 Model that identifies six modifiable elements of healthcare systems: (1) organisational support, addressing organisational culture and leadership, (2) clinical information systems to organise patient, population and provider data, (3) delivery system design to address composition and function of the care team and follow-up management, (4) decision support to increase provider access to evidence-based guidelines and specialists for collaboration, (5) self-management support to provide tailored education, skills training, psychosocial support and goal-setting and (6) community resources to provide peer support, care coordination and community-based interventions Discharge management (n=15) 30-36 52-59 Interventions designed to facilitate effective transitions from hospital care to other settings. Typically includes a predischarge phase of support, transitional care for the move between the hospital and community/home setting and postdischarge follow-up and monitoring, often incorporating rehabilitation or reablement support Complex interventions (n=3) 60 61 70 Two reviews assessed a range of interventions rather than focusing on a single intervention or service model Multidisciplinary teams (n=10) 37-40 62-66 71 Interventions comprising teams composed of multiple health and/or social care professionals working together to provide care for people with complex needs. Teams typically included condition-specific expertise, nurses, occupational therapists, physiotherapists, social workers, GPs and occasionally pharmacists or case managers Self-management (n=5) 40-42 67 68 Interventions designed to provide patient support, typically via tailored education to inform the patient about their condition(s), recognising signs and symptoms of disease exacerbation, dietary and lifestyle advice and/or condition-specific education supporting medication adherence  28 ? COPD. 38 46 48 50 68 The most effective interventions were based on the CCM, for which 4/5 reviews showed statistically significant reductions in admission rates following the intervention. Multiple component strategies were associated with reductions of between 22% 46 and 32% 48 in admission rates for patients with COPD. MDT interventions were also effective, with 4/8 reviews showing significant reductions in admissions. Effect sizes ranged from 25% for a COPD MDT with formal links to primary care, 38 through 26% for teams that included specialist heart failure expertise, 64 to 31% for teams that included pharmacists as collaborators. 63 One review of structured self-management interventions demonstrated a 43% reduction in the relative risk of COPD-related admission. 68 Case management interventions were largely ineffective in reducing admission rates, with 3/4 showing mixed findings, 23 24 27 although one case management intervention for heart failure comprising intensive follow-up that gradually reduced in intensity over time showed a potential 58% reduction in admissions. 45 Most reviews reported condition-specific admissions and admissions for any cause. In all cases, potential reductions in condition-specific admissions were substantially greater than those for all-cause admissions. 38 45 62-64 68 Hospital readmissions Twenty-four reviews assessed readmissions. Eleven reported positive findings: eight for heart failure, 42 57 70 and one for COPD. 54 Discharge management was the most effective intervention, with 6/13 reviews showing significant reductions in readmission rates. 53-55 57-59 Interventions incorporating an inpatient phase and postdischarge support at home were associated with reductions in condition-specific readmission rates of 24% 32 and 49% 53 for heart failure interventions, 24% for a hospital at home intervention for COPD 54 and a 15% reduction for patients with chronic diseases. 57 Similarly, 'complex' interventions that included specialist nurse-led clinics for heart failure follow-up were associated with a 91% reduction in condition-specific readmission rates in one review, 59 and postdischarge hospital outreach coordinated by a MDT was associated with a 32% reduction in heart failure readmission rates. 55 In contrast, discharge interventions for patients who had stroke were ineffective, with 0/4 reviews assessing this intervention showing no differences between intervention and control groups. 32 35 52 56 Other interventions showed less comprehensive evidence. One of three CCM reviews that assessed readmissions found a 30% reduction in readmission rates for heart failure. 47 One self-management review in which nurses provided heart failure-specific education reported a 56% reduction in readmissions. 42 Two reviews assessing complex interventions reported significant reductions in readmission rates: one for a heart failure case management intervention 61 and another for patients with general chronic diseases. 70 One MDT review showed a 42% reduction in heart-failure specific readmission, with subgroup analysis indicating that heart failure specialist nurses could reduce condition-specific readmissions by up to 39%. 66 As with admissions, potential reductions in readmissions were substantially greater for condition-specific readmissions than all-cause readmissions, with effect sizes in the former typically double those for the latter. 47 53 56 66 Length of stay Sixteen reviews assessed LoS, across six intervention categories. Neither case management interventions 24 25 27 or self-management interventions 41 68 showed evidence of effectiveness, but there were positive findings in the CCM, 46 48 discharge management, 52 56 57 complex intervention 60 and MDT groups. 62 65 66 Two CCM interventions were associated with a significantly reduced mean LoS for COPD of 2.51 46 and 3.78 days, respectively. 48 Three discharge management reviews showed significant LoS reductions. Two were for patients who had stroke, including postdischarge support coordinated through multidisciplinary hospital outreach 52 and early supported discharge. 56 Pooled results from the early supported discharge meta-analysis suggested a mean LoS reduction of 7.7 days, rising to 28 days for the most severely impaired patients compared to 4 days for moderately impaired patients. 56 One discharge management intervention for patients with chronic diseases reported positive results, with a modest reduction of 0.91 days. 57 Finally, three MDT interventions showed significant reductions in LoS, all for heart failure patients. Again, reductions were modest at 1.9 days for an MDT that included a clinician plus specialist nurse, pharmacy, health education, dietician and social worker support, 62 a 'generally shorter' LoS for an intervention based on nurses, heart failure physicians and general practitioners (GPs) providing condition-specific patient education 65 and a MDT providing hospital outreach for at least 12 months after hospital discharge was associated with a mean reduction in LoS of 1.49 days. 66 Accident and emergency use Nine reviews measured the effectiveness of interventions in reducing Accident and Emergency (A&E) use. Five reviews included patients with chronic diseases, all showing mixed findings or no association between intervention and outcome. 24 25 27 41 70 The remaining reviews assessed single conditions, with 2/3 demonstrating statistically significant reductions in A&E use for COPD, 46 60 and one showing significant findings in patients with heart failure. 65 Case management and self-management interventions were ineffective in reducing A&E use. 24 25 27 41 Effective interventions related to the CCM, where multicomponent COPD interventions were associated with a 42% reduction in A&E use, 46 the complex intervention group, where interventions with multiple components administered by multiple professionals demonstrated a potential 32% reduction in A&E use, 60 and the MDT group, where one review found a significant reduction in A&E use when an MDT for heart failure contained condition-specific specialist expertise. 65 However, A&E use remained high overall, with 77% of patients in the intervention group having at least one emergency department visit during the 12-month follow-up period, compared to 84% of control patients.

Costs
Twenty-five reviews assessed healthcare costs but the evidence base was poor and heterogeneous-information on potential cost savings was typically qualitative and could not be compared across reviews. Ten reviews reported positive findings: five for patients with heart failure, 36 42 47 58 64 two for COPD, 46 54 two for chronic disease 26 28 and one for stroke. 35 Eleven reviews reported mixed findings, all for chronic disease 23 24 27 39 40 69 70 or stroke, 31 34 52 56 and four reported no difference in costs between intervention and control groups for chronic disease, 43 COPD, 28 mental health 51 or heart failure. 59 The most effective interventions were based on the CCM, with three reviews reporting significantly reduced costs. 28 46 47 One review reported cost savings of between 34% and 70% for CCM interventions but gave no further detail of the nature of these savings. 46 Discharge management interventions were cost-effective in some cases, 35 36 54 58 predominantly due to reduced hospitalisation costs and fewer patient bed days. MDT interventions that included specialist expertise also showed some evidence for cost-effectiveness but again, little detail was given to substantiate this. 64

DISCUSSION
The primary aim of this review was to assess whether integrated care-through interventions to coordinate care across two or more health and/or social care settings for patients with chronic diseases-can reduce hospital activity and if so, to what extent. Despite the diverse evidence base and variations within and across reviews in terms of the characteristics, duration and intensity of interventions, some positive trends were evident. Overall, the most effective interventions included discharge planning and postdischarge support for hospital inpatients, 53-55 57-59 MDT care-particularly when condition-specific specialists, specialist nurses or pharmacists were part of the team skill mix, 38 63-65 and interventions based on multiple components of the CCM, 28 46-48 50 69 although no CCM reviews reported which specific components were most likely to produce positive outcomes. Self-management showed most promise when incorporated into MDT care or when tailored patient education was included in discharge planning. 42 68 The least effective intervention was case management. Although in theory this intervention may increase health service efficiency by reducing unnecessary contacts with healthcare professionals, 43 we found little evidence of effectiveness. Some of the key features of effective interventions are outlined in table 5. This table is not intended as a 'toolkit' for effectiveness, since interventions or components that reduced hospital activity for some outcomes and/or conditions were not necessarily effective for others. Nevertheless, it summarises some of the 'ingredients' of potentially effective integrated care interventions.
All hospital activity outcomes showed some significant reductions. Proportionally, LoS was the most likely to reduce, with 9/16 reviews reporting positive findings. However, gains were typically modest: multicomponent CCM strategies could reduce LoS by 2.5-4 days, 46 48 and MDT care with specialist expertise was associated with LoS reduction of 1.5-2 days. 62 66 For admissions, 11/21 reviews demonstrated positive findings, suggesting potential reductions of between 15% and 50%. Readmission rates were significantly reduced in 11/24 reviews, suggesting a 10-30% reduction in all-cause readmission and a 25-50% reduction in condition-specific readmission could be achieved with interventions based on discharge management, 52-55 MDT 66 and the CCM. 46 48 50 69 A&E use typically reduced by 30-40% in reviews of effective interventions. 46 60 65 It has been argued that integrated care may increase hospital activity due to supply induced demand, in which integration uncovers unmet patient need. 72 73 Several reviews noted minor increases in activity following case management, 24 27 CCM, 51 discharge management 31 and MDT interventions. 39 65 However, these increases were typically restricted to one or two primary studies within a review and were rarely statistically significant.
A secondary objective was to assess the settings and patient populations for which promising interventions may be most effective. Interventions focused on single conditions showed greater effectiveness than those implemented for patients with general chronic diseases. Those that assessed MDT care or discharge management for patients with heart failure and COPD were typically effective in reducing admissions, 38 62-64 readmissions 53-55 58 59 and LoS, 62 65 66 with some positive trends evident in reducing A&E use. 65 This may reflect the difficulty of designing effective interventions for people with a broad range of conditions, in a healthcare system where care for patients with complex needs remains largely centred on single condition guidelines. Furthermore, interventions such as MDT have been an established feature of disease management for conditions like heart failure for a number of years, and the particular success of interventions focused on this patient group is likely to reflect this. Care offered in patients' homes, whether following discharge from hospital, 53 54 through MDT care, 38 62 64 or through self-management interventions 42  multidisciplinary outreach as opposed to a community in-reach model. Although these interventions were associated with significantly reduced hospital activity, the most successful were coordinated by the acute sector, suggesting that effective integrated care may still rely on the deployment of substantial hospital resources and the involvement of multiple acute sector healthcare professionals.
Our final objective was to assess the cost implications of integrated care interventions. Data were poor: the care components that cost data referred to were often unclear and effect sizes were rarely stated. Where statistical significance was described, the majority of savings appeared to come from a reduction in costs incurred through hospitalisation, whether this was because interventions allowed patients to be discharged from hospital earlier or whether interventions reduced subsequent rates of hospitalisation or rehospitalisation. As a result, interventions which included some element of home care or rehabilitation tended to be cost saving compared to care in which rehabilitation was provided within the hospital environment. 28 35 42 46 47 54 However, it is likely that substantial cost savings can only be realised if hospital capacity can be physically removed from the system, for example, through ward closures. We found little evidence of this following integrated care interventions.

Strengths and weaknesses
This is the first umbrella review of its kind and is timely given the increasing emphasis on integrated care in healthcare policy with the key aim of reducing hospital use. By undertaking an umbrella review of systematic reviews, we could assess a large volume of evidence across diverse conditions, interventions and outcomes. However, umbrella reviews have limitations. Grouping interventions in a way that allowed meaningful conclusions to be drawn about their effectiveness was challenging. Although we employed the Effective Practice and Organisation of Care Group (EPOC) taxonomy 74 as an organisational framework, few interventions were mutually exclusive and the characteristics and form of interventions frequently overlapped. For example, most discharge management interventions were delivered by MDT, and several interventions included some element of self-management support. Even for reviews which shared broadly similar intervention characteristics, the duration of follow-up, study design, complexity, intensity and mode of delivery varied. Furthermore, because the unit of analysis is the review rather than the primary study level, the re-synthesis of information at the umbrella review level that has already been synthesised at review level risks loss of detail or misinterpretation of findings and trends. But, by restricting inclusion to reviews receiving moderate, good or high QA scores, we Complex interventions with multiple components ▸ Greatest effects when treating patients with single rather than multiple conditions ▸ No reviews stated specific components that were more (or less) likely to be effective than others Postdischarge hospital outreach coordinated by a multidisciplinary team ▸ Greatest effects when treating patients with single rather than multiple conditions ▸ In contrast, community inreach interventions not effective, even when featuring MDT ▸ Not effective for patients who had stroke MDT with: A. Disease-specific specialists as core members B. Specialist nurse-led clinics C. Pharmacists as collaborative partners ▸ Greatest effects when treating patients for single conditions ▸ No reviews compared the 'added value' that a given professional or clinician may bring to a MDT, so optimal composition remains unknown Transition from hospital to home is most effective when interventions are initiated during the inpatient phase and continue postdischarge Home-based community follow-up ▸ Effective for reducing length of stay in patients who had stroke ▸ Community follow-up least likely to be effective when delivered through a case management model Self-management education combined with multidisciplinary approaches or discharge planning ▸ Greatest effects when treating patients for single conditions ▸ Simple self-management interventions were largely ineffective *MDT, multidisciplinary team.
reduced the likelihood of evidence misinterpretation and the incidence of discordant findings. Successful approaches to integrated care have highly contextspecific histories, yet by undertaking an umbrella review, we were unable to draw conclusions about the specific contexts in which interventions were implemented. Nevertheless, we believe that the methodological strengths of our approach outweigh the limitation of being unable to comment on the specific contexts in which interventions were implemented.

Implications for clinicians and policymakers
Although there was evidence that some integrated care interventions can reduce hospital activity, effects were rarely unequivocally positive. The size of gains from integration may also be modest. 17 For example, in recent years, the trend in outcomes such as length of hospital stay has been steadily reducing, largely due to improved surgical techniques and increased day case treatment. 8 This suggests that there may be limits to the absolute reductions in key hospital activity metrics that integrated care initiatives could achieve. 1 This was evident in several reviews that noted statistically significant differences in outcomes for intervention versus control patients, but which reported persistently high absolute rates of outcomes such as admissions and readmissions in each group. 48 52 54 58 65 68 This has implications for the potential success of policy initiatives designed to reduce hospital activity. In England, integration has become a central feature of the evolving healthcare policy landscape and there are high expectations of substantial benefits from integrating care. The BCF and 'Vanguard' sites 7 have been developed following recognition that radically different models of care are needed if the NHS in England is to overcome its growing challenges, and both policy initiatives involve farreaching change to health and social care services with the aim of meeting national headline targets for reduced hospital and emergency care use. 75 Interventions shown to be effective in this review have much in common with the rationale behind the BCF-care provided in the community rather than in hospitals was shown in many cases to be highly effective. Multidisciplinary care, discharge planning and self-management educating patients on identifying symptoms of exacerbation of their condition (s) all have the potential to improve outcomes and reduce activity at the 'back door' and 'front door' of the acute sector. Disease-specific expertise was also found in many reviews to be crucial to the success of integrated care interventions, as was secondary care outreach to other settings. This bodes well for BCF and Vanguard initiatives built around these interventions. However, it is of concern that many vanguard sites aim to integrate care via a case management approach, which showed the poorest evidence of effectiveness in our review. This raises questions over whether the Vanguard strategies will be able to deliver the outcome improvements they are being established to achieve. The extent to which integrated care can bring about significant cost savings in a health system beset with ongoing budgetary constraints is also highly uncertain.
Interventions designed for single conditions were substantially more effective than those designed to treat patients with chronic diseases in general terms. On one hand, this suggests that service providers can achieve some 'quick wins' by targeting interventions such as discharge planning and specialist MDT towards specific patient groups in whom the evidence for reduced hospital use is clear. On the other hand, this means that integration may not deliver the substantial reductions in acute sector activity that must be achieved if healthcare services are to remain sustainable in the longer term.
Unanswered questions and future research Integrated care poses challenges to the measurement of 'hard' healthcare service outcomes in what are often complex intervention programmes. Determining cause and effect is difficult when interventions include multiple components, yet being able to link a specific intervention to a particular observed outcome is typically central to policymaking and commissioning objectives. Research to develop a robust taxonomy for integrated care interventions and their components would make assessments of comparative effectiveness across interventions less challenging. We attempted to maximise the relevance of review findings to the English health and social care system by considering interventions implemented in developed economies, but further research is needed to determine whether interventions found to be effective in other healthcare systems can be generalised to the NHS. In particular, robust evaluations would allow the influence of local and organisational contexts to be disentangled from the effects of the intervention themselves, as although the umbrella review gives some indication about 'what' might work, it does not necessarily help our understanding of 'how' an intervention works and why it may work in some circumstances and not others.
Few reviews explicitly addressed multimorbidity, which has recently become of central importance in debates about hospital use by patients with complex needs. 4 76 Further research is needed to understand the issues faced by patients with multimorbidity when negotiating the health and social care system. 77 Similarly, despite our comprehensive search strategy, the evidence base focused little on the role of primary care, social care or the voluntary sector in providing integrated services. Given current policy drivers towards services being provided in the community by GPs and other organisations rather than acute providers, further research to assess the implications of integrated care for the organisation and delivery of services in these sectors is urgently needed.

CONCLUSIONS
This review highlights a number of potentially effective integrated care interventions to reduce hospital use for patients with chronic diseases. Interventions based on MDT that include condition specialists, those focused on discharge management that include postdischarge rehabilitation and follow-up and those based on multicomponent strategies were most likely to be associated with significant reductions in hospital use for patients with single conditions such as heart failure and COPD. Yet there was little robust evidence about potential cost efficiencies, and the effectiveness of care delivered in primary and social care settings remains largely unknown. Despite considerable fanfare accompanying efforts to integrate care across the health and social care system in England, integration does not seem to be a 'magic bullet' and the magnitude of achievable gains is unlikely to match those required by current policy targets.