Interventions for improving outcomes in patients with multimorbidity in primary care and community settings

Summary of findings for the main comparison.

Interventions aimed at improving outcomes for people with multimorbidity compared with usual care
Participant or population: Adults with multimorbidity (two or more chronic conditions) Settings: Primary care and community settings Intervention: Any intervention designed to improve outcomes for people with multimorbidity including professional‐, organisational‐ and patient‐oriented interventions Comparison: Usual care
Outcomes	Impacts	Number of studies	Quality of the evidence (GRADE)
Clinical outcomes	There is no clear effect on clinical outcomes with a range of standardised effect sizes from 0.01 to 1.6 with a minority having effect sizes > 0.5; interventions aimed at improving management of risk factors in comorbid conditions were more likely to have higher effect sizes.	11	⊕⊕⊕⊖ Moderate
Mental health outcomes	There are improved depression‐related outcomes in studies targeting comorbid conditions that include depression with a range of standardised effect sizes from 0.09 to 2.24 with 4 of 7 studies having moderate to large effect sizes (> 0.5) . Standardised mean difference of −0.41 (95% CI, −0.63 to −0.20) was calculated from combining data from 6 studies.	9	⊕⊕⊕⊕ High
Patient‐reported outcome measures (PROMs)	There are mixed effects on PROMs with only half of studies that reported these outcomes showing any benefit with a range of standardised effect sizes from 0.03 to 1.7. Only 1 of 5 studies with available data on self‐efficacy had a moderate effect size, 4 of 7 had a moderate effect size for HRQoL (> 0.5) and effect sizes for other psychosocial outcomes were generally low.	12	⊕⊕⊕⊖ Moderate
Health Service Utilisation	There were no effects on health service utilisation and changes in visits were difficult to interpret as some interventions could lead to higher numbers of visits if previous unmet need was being addressed. There was no difference in admission‐related outcomes, though numbers of admissions in most of these studies were very small.	5	⊕⊕⊖⊖ Low
Medication use and adherence	There are mixed effects on medication use and adherence with half the studies reporting this outcome showing benefit. Proportions adherent to medication were higher in intervention participants with ranges in absolute difference of 10% to 40% but all studies with available data had small effect sizes.	4	⊕⊕⊖⊖ Low
Health‐related patient behaviours	Studies measuring this outcome reported a range of effects varying from an additional 18 minutes spent walking per week to an absolute difference in kcals expenditure per week of 2516 (no studies presented data that could be used to calculate effect sizes).	7	⊕⊕⊕⊖ Moderate
Provider behaviour	The majority of studies reporting provider behaviour indicated improved provider behaviour relating to care delivery; three studies reported a range of 15% to 40% in proportions of intervention providers improving behaviours such as appropriate referral.	5	⊕⊕⊕⊖ Moderate
GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.
We downgraded the evidence for effects on clinical and psychosocial outcomes to moderate due to lack of consistency of effect across studies and small effect sizes. We downgraded the evidence for effects on provider behaviour to moderate due to limited available data for calculation of standardised effect sizes (SES) and lack of clarity regarding the clinical importance of the results. We downgraded the evidence for effects on health service utilisation and medication use and adherence to low due to variation across studies and small effect sizes.

Background

Many people with chronic disease have more than one chronic condition, which is referred to as multimorbidity. While this is not a new phenomenon, there is greater recognition of its impact and the importance of improving outcomes for individuals affected. Research in the area to date has focused mainly on descriptive epidemiology and impact assessment (Fortin 2007). There has been limited exploration of the effectiveness of interventions to improve outcomes for people with multimorbidity.

Description of the condition

There has been increasing focus on the enormous personal and societal burden of ill‐health caused by chronic disease. The World Health Organization (WHO) has emphasised the importance of organising healthcare delivery systems to improve health outcomes and has stressed the importance of building integrated healthcare systems that can address chronic disease management (WHO 2002). This can be done by focusing on generic chronic care models, as has happened mainly in the United States of America (USA), or by developing national systems focusing on single chronic conditions as has happened with the National Service Frameworks in the UK (Lewis 2004; Satariano 2013). However, many people with chronic disease have more than one chronic condition, which is referred to as multimorbidity and formally defined as the co‐existence of two or more chronic conditions (Fortin 2005). While this is not a new phenomenon, there is greater recognition of its impact and the importance of improving outcomes for individuals affected (Fortin 2007; Smith 2007).

While the accepted term for people with multiple chronic conditions is now multimorbidity, the term comorbidity has been used interchangeably in the past. It is now accepted that comorbidity should be used when there is a specified index condition or where there are defined combinations of conditions (for example hypertension and cardiovascular disease) as opposed to multimorbidity where any condition could be included (Valderas 2011). Multimorbidity is the more general term and individuals with comorbidity also have multimorbidity but the reverse does not necessarily apply. For the purposes of this review when analysing the included studies, we looked at studies based on the intervention elements but we also considered differences between studies that specifically target comorbid conditions as opposed to those targeting general multimorbidity. This is because interventions in the comorbidity studies are designed to target the specific included conditions. These distinctions are important in the context of developing and evaluating effective interventions and considering their generalisability (Fortin 2013; Smith 2013).

Individuals with multimorbidity are more likely to die prematurely (Deeg 2002; Menotti 2001; Rochon 1996), be admitted to hospital (Bähler 2015; Condelius 2008; Payne 2013), and have longer hospital stays (Bähler 2015; Librero 1999). They have poorer quality of life (Brettschneider 2013), loss of physical functioning (Bayliss 2004; Fortin 2004; Fortin 2006b), and are more likely to suffer from psychological stress (Fortin 2006a; Gunn 2012). Medicines management is often complex, resulting in polypharmacy with its attendant risks of drug interactions and adverse drug events (Duerden 2013; Gandhi 2003; Guthrie 2011). For patients, in addition to understanding and managing their conditions and drug regimes, they must also attend multiple appointments with different healthcare providers and adhere to lifestyle recommendations (Gallacher 2011; Townsend 2006).

Prevalence studies of multimorbidity have been carried out in different countries indicating that, particularly in those over 60 years, the majority of people attending family primary care services had more than one chronic condition (Fortin 2005; Fortin 2006c; van den Akker 1998; Wolff 2002). A subgroup of these service users have a debilitating combination of conditions that have a high impact on their own lives but also on their utilisation of health services and related costs (Hoffman 1996; Marengoni 2011; Parmelee 1995; Smith 2008). This emerging concept may be referred to as 'complex multimorbidity' and has been defined as people with three or more chronic conditions involving three or more body systems (Harrison 2014). These individuals can pose management difficulties, resulting in frequent health care visits, frequent emergency hospital admissions, and repeated investigations with enormous cost both for the individuals and the healthcare system involved. A UK report has examined the costs associated with this group of people who are described as 'high impact users' on the basis of their frequent emergency admissions (Rowell 2006). Fragmentation of care is a significant problem for this group, resulting from the involvement of both primary care and multiple specialists who may not be communicating with each other effectively (Wallace 2015). Starfield found that people with a greater morbidity burden have a higher use of specialists even for conditions that are normally managed in primary care, and concludes that there is a need for a better understanding of the roles of generalists and specialists in managing these individuals (Starfield 2005)

Description of the intervention

Given the complexity of managing people with multiple chronic conditions, potential interventions are likely to be complex and multifaceted if they are to address the varied needs of these individuals. We anticipated that a variety of intervention types could work to improve outcomes for people with multimorbidity and could be included within the scope of this review. Cochrane Effective Practice and Organisation of Care (EPOC) has developed a taxonomy that defines intervention types (EPOC 2002). We have used this taxonomy to define health service and patient‐oriented interventions that have been designed to improve outcomes of people or populations with more than one chronic condition.

1. Professional interventions: for example, education designed to change the behaviour of clinicians. Such interventions may work by altering professionals' awareness of multimorbidity or providing training or education designed to equip clinicians with skills in managing these individuals, thus improving their healthcare delivery.

2. Financial interventions: for example, financial incentives to providers to reach treatment targets. These interventions might work by incentivising health service delivery and providing resources to extend consultation length for people with multimorbidity.

3. Organisational interventions: these can be further divided into organisational changes delivered through practitioners or directly to patients. For example, any changes to care delivery such as case management or the addition of different healthcare workers such as a pharmacist to the healthcare team. These interventions may work by changing care delivery to match the needs of people with multimorbidity across a range of areas such as coordination of care, medicines management, or use of other health professionals such as physiotherapists and occupational therapists to address needs relating to physical and social functioning.

4. Patient‐oriented interventions: this would include any intervention directed primarily at individuals, for example, education or support for self management. These interventions might work by improving self management, thus enabling people to manage their conditions more effectively and to seek appropriate health care.

5. Regulatory interventions: for example, changes to local or national regulations designed to alter care delivery in order to improve outcomes. Such interventions might work by introducing regulatory changes that facilitate and enable the funding of care that is directed towards those with complex health needs. An example could be the introduction of free primary care for people with multimorbidity on the basis that preventive care might prevent subsequent more costly hospital admissions. While we did not find these types of interventions, we believe they could exist and would fall within the scope of this review for future updates.

How the intervention might work

We anticipated that organisational‐type interventions might predominate. We were aware that there has been a focus on case management, based mainly in health maintenance organisations in the USA (Zwarenstein 2000). Case management is defined as the explicit allocation of co‐ordination of tasks to an appointed individual or group and it is postulated that the function of co‐ordination is so important and specialised that responsibility for carrying it out needs to be explicitly allocated (Zwarenstein 2000). Our review included studies where case management was employed but only if it was specifically directed towards individuals identified as having multimorbidity.

The implementation of the Family Medicine Groups in the province of Québec, Canada, is another example of an organisational intervention as it involved new forms of shared responsibilities between physicians and nurses (MSSS 2001). Another example in the United Kingdom (UK) is the community matrons programme, which is being delivered through primary care trusts and is based on nurse‐provided case management for people with complex care needs including those with multimorbidity (London DOH 2005). It is similar to previous programmes delivered through social services in the 1990s and there have been concerns expressed as to the feasibility of achieving the programme targets without real integration of primary and specialist services (Murphy 2004).

The differences outlined earlier between multimorbidity in general and comorbidity where there are defined combinations of conditions also influences how interventions are designed. Interventions targeting specified comorbid conditions can be designed to address the specific challenges for people with those conditions. For example, an intervention that targets people with diabetes and depression will combine elements of diabetes‐focused care with psychotherapy or escalation of antidepressant medication, or both interventions, so as to address both conditions. Interventions for people with multimorbidity in general cannot have a disease focus as there are no pre‐specified conditions so the interventions might address improved coordination of care, improved medicines management or specific functional difficulties experienced by patients.

Since this review was originally planned in 2007, there has been widespread recognition of the need to address the challenge of multimorbidity across health systems with a series of articles in international medical journals highlighting the challenges involved. Two very useful resources highlighting the challenges of multimorbidity and collating research in the area are: i) the BMJ multimorbidity special collections (BMJ Multimorbidity collection) and ii) the International Research Community on Multimorbidity archive IRCMO at the University of Sherbrooke, Quèbec, Canada (IRCMO). The BMJ series includes a series of editorials, original research studies and a clinical review with a multimorbidity focus. IRCMO provides a platform for any researcher interested in multimorbidity to contribute to a regularly updated blog and also compiles a list of multimorbidity related publications.

Why it is important to do this review

This review was originally undertaken based on the clear recognition of the need for integrated care for people with multiple conditions who have complex care needs (Stange 2005). The evidence base for managing chronic conditions is based largely on trials of interventions for single conditions and individuals with multimorbidity are often excluded from such studies (Fortin 2006c; Starfield 2001; Wyatt 2014 Zulman 2011). The inadequacy of existing clinical guidelines to support clinicians in managing people with multimorbidity has been highlighted as a significant issue in delivering care (Dumbreck 2015; Wyatt 2014). Clinical guideline developers have attempted to address this issue with the consideration of certain combinations of commonly co‐occurring conditions, for example, diabetes and depression (NICE 2009). However good quality evidence is essential to inform this clinical area and in recent years focus has shifted to intervention development and the need to reorientate clinical practice and healthcare systems for the people who use them most (Satariano 2013).

Objectives

Methods

Criteria for considering studies for this review

Types of studies

We considered randomised controlled trials (RCTs), non‐randomised clinical trials (NRCTs), controlled before‐after studies (CBAs), and interrupted time series analyses (ITS), meeting EPOC quality criteria (EPOC 2013). We included NRCTs in the original protocol (Smith 2007b) as we anticipated that, given the challenges in undertaking multimorbidity research (Fortin 2007) and the likelihood that complex interventions would be tested, there would be relatively few RCTs and that non‐randomised designs might be used instead.

Types of participants

We included any people or populations with multimorbidity receiving care in a primary or community care setting. We adopted the most widely used definition of multimorbidity, that is, the co‐existence of multiple chronic diseases and medical conditions in the same individual, usually defined as two or more conditions (Fortin 2004; van den Akker 1998). We used the WHO definition of chronic disease, which is "health problems that require ongoing management over a period of years or decades" (WHO 2002). We included all studies that identified participants or sub‐groups of participants on the basis of multimorbidity, as defined by the study authors. In some studies, additional eligibility criteria were applied (for example, history of high service utilisation) in an effort to identify more vulnerable people who might benefit more from the intervention being studied.

We excluded studies where multimorbidity was assumed to be the norm on the basis of individuals' age as the interventions were not being targeted specifically at multimorbidity and its recognised challenges. This included studies where interventions were directed at communities of people based on location or age of participants in which participants could be presumed to have multimorbidity on the basis of their age or residence in a nursing home but interventions were not designed to specifically target multimorbidity.

Types of interventions

We included any type of intervention that was specifically directed towards a group of people defined as having multimorbidity. Only interventions based in primary care and community settings were included. Interventions included care delivered by family doctors, nurses, or other primary care professionals. Primary health care was defined as providing "integrated, easy to access, health care services by clinicians who are accountable for addressing a large majority of personal health care needs, developing a sustained and continuous relationship with patients, and practicing in the context of family and community" (Vaneslow 1995). However, not all countries have clearly‐defined primary care systems (Starfield 1998), so we included care delivered in community settings by individuals fulfilling the basic criteria for primary care, i.e. if they are available to treat all common conditions in all age groups and have an ongoing relationship with their patients. While some specialists may deliver components of primary care to their patients, practitioners were not included unless they fulfilled the definition of being available to treat all conditions and have an ongoing relationship with their patients.

Interventions were classified as 'simple' if they used one identifiable component or 'multifaceted' if they incorporated more than one feature.

We categorised interventions using the EPOC taxonomy presented in the Background section. Where interventions had multiple elements, we defined each element within the taxonomy and highlighted the predominant element of the intervention (see Table 1).

Table 1. Multimorbidity intervention components

Author Year	Professional	Participant	Organisational			Effect of intervention on primary outcome
			Case management or coordination of care	Reorganisation of care/team working	New team member
Predominantly organisational
Barley 2014	Nurse training	Participant information Prioritisation to create goals and health plan	Case manager provided personalised care	Regular planned participant visits Weekly team meetings	Nurse case manager	Pilot study and primary outcome was feasibility and deemed successful
Bogner 2008		Individualised programme	Case manager	Regular planned participant visits		Improved blood pressure control and depression scores
Boult 2011	Nurse training	Individual management plans Support for self‐management	Guided care nurses coordinated care	Guided care 'pods' consisting of nurse and PCP Monthly monitoring of participants		No impact on healthcare utilisation
Coventry 2015	Practice team training	Personalised goals and participant workbooks		Collaborative care using stepped care protocols Joint consultation between participant, psychologist and practice nurse	Psychologist Supervision and input from team psychiatrist	Modest reduction in depression scores
Hogg 2009		Individualised care plans		Multidisciplinary team‐based management with home based assessment Medication review	Pharmacist	Modest improvements in quality of chronic care delivery
Katon 2010		Individualised management plans and targets Support for self‐management		Team‐based care Stepped care treatment protocols Weekly team meeting	Psychologist and psychiatrist supported depression care	Improvements in composite outcome of glycaemic control, blood pressure, lipids and depression scores
Kennedy 2013	Practice training	Support for self‐management Participant guidebooks		Systems‐based approach to self‐management support with practice supports and links made with related local services		No intervention effect noted
Krska 2001		Individualised pharmaceutical care plans		Practice team‐implemented care plans	Pharmacist undertook medication review and devised pharmaceutical care plans	Reduction in pharmaceutical care issues
Martin 2013	Training for community psychologists	Cognitive behavioural therapy sessions		Psychological care programme designed for headache and depression	Community psychologists	Reduced headaches and improved depression scores
Morgan 2013	Practice nurse training	Support for self‐management Goal setting Individualised care plans	Nurse case manager	Quarterly reviews with practice nurse with GP stepping up care as needed		Improved depression scores
Sommers 2000		Risk reduction plan		Team based care with home assessment followed by team discussion, treatment plan and targets	Social worker	Reduced hospitalisation
Wakefield 2012		Participation in home telehealth monitoring	Nurse case manager using telehealth monitoring and treatment algorithms			Improved blood pressure, no effect on glycaemic control
Predominantly Patient‐oriented interventions
Eakin 2007		Support for self‐management with focus on diet and physical activity		Regular visits and follow‐up telephone calls	Health educator	Improvements in diet but not in physical activity
Garvey 2015	Occupational therapist (OT) training	OT‐led, group‐based support for self‐management programme (6 weeks) Goal setting and peer support		GP and primary care team referral	OT with input from physiotherapist and pharmacist	Improvements in activity participation
Gitlin 2006		Home‐based programme targeting functional difficulties with individualised plans and focus on falls prevention		Home visits and regular follow‐up calls	Occupational therapist and physiotherapist	Improvements in function (reduced mortality at 4 year follow‐up)
Hochhalter 2010	Training for coaches running intervention	Patient Engagement workshop (x1)		Two follow‐up phone calls	Coach who delivered workshop	No effect on outcomes
Lorig 1999	Training for volunteer lay group leaders	Chronic Disease Self Management Support Programme (six sessions) Peer support			Volunteer lay group leaders supported by study team	No primary outcome specified. Multiple outcomes reported with mixed effects
Lynch 2014		Diabetes self management support groups (18 sessions) Peer support Goal setting and behaviour skills training			Dietician led groups	No effect on primary outcome of weight reduction

The predominant intervention component is highlighted in bold text for each study

No study contained a financial‐type intervention element

We excluded the following interventions:

Professional educational interventions or research initiatives where there was no specified structured clinical care delivered to an identified group of people with multimorbidity.

Interventions including people with comorbid conditions where the intervention was targeted solely at one condition and did not address the full extent of the multimorbidity. This commonly arises in relation to chronic disease and comorbid depression, so called 'depression plus one studies'. These are increasingly common as the link between depression and most chronic conditions has now been well established (Simon 2001). They include interventions designed to address depression in participants rather than targeting all conditions identified. We therefore excluded such studies if the intervention was only targeted at the depression and did not address the full extent of the multimorbidity.

The comparison was usual care.

Types of outcome measures

We included studies if they reported any objective, validated measure of:

Patient clinical or mental health outcomes (e.g. blood pressure, symptom scores, depression scores).
Patient‐reported outcome measures (e.g. quality of life, well‐being, measures of disability or functional status).
Utilisation of health services (e.g. hospital admissions).
Patient behaviour (e.g. measures of medication use and adherence, and other objective measures such as goal attainment (Cox 2002; Gordon 1999; Kiresuk 1968), if measured with a validated scale.
Provider behaviour (e.g. chronic disease management scores).
Acceptability of the service to recipients and providers, and treatment satisfaction were included if it was reported in a study that reported objective outcome measures behaviour.
Economic outcomes (e.g. full economic analyses incorporating measures of efficiency or effectiveness in relation to costs or direct costs depending on what was reported in included studies). Where direct costs were reported alone, we indicated whether these costs related to society, the health service, or the recipients. We also reported, where possible, costs in relation to the specific year and currency presented; whether costs related to total costs or simple fees charged; what was included in the cost calculations; and over what time period costs were calculated.

We excluded attitude and knowledge outcomes.

Search methods for identification of studies

Electronic searches

We searched the following electronic databases without language restrictions up to 28 September 2015:

Cochrane Central Register of Controlled Trials (CENTRAL), The Cochrane Library, 2015, Issue 10, Wiley
Database of Abstracts of Reviews of Effects (DARE), The Cochrane Library, 2015, Issue 3, Wiley
MEDLINE, 1990 to September 2015, In‐Process and other non‐indexed citations, OvidSP
EMBASE, 1980 to September 2015, OvidSP
Cochrane Effective Practice and Organisation of Care (EPOC) Group Specialised Register, Reference Manager
Cumulative Index to Nursing and Allied Health Literature (CINAHL), 1980 to September 2015, EBSCOHost
Allied and Complementary Medicine Database (AMED), 1985 to September 2015, OvidSP
CAB Abstracts, 1973 to September 2015, EBSCOHost
HealthSTAR, 1999 to September 2015, OvidSP

We also searched the following trials registries:

We searched the IRCMO repository for unpublished/grey literature (IRCMO), and invited experts to inform us of other completed or ongoing studies

The search strategy was particularly challenging given the lack of a MeSH terms for multimorbidity. In addition, we were aware from existing epidemiological literature that the recognition of multimorbidity as a concept is relatively recent. Multimorbidity is sometimes used synonymously with the term comorbidity, though this tends to be used in relation to diseases that coexist with an index disease under study (de Groot 2004). However, comorbidity is a MeSH term, whereas multimorbidity is not, so we included both terms in our search. For pragmatic reasons we limited the MEDLINE search to articles indexed from 1990 onwards.

The search strategy published in the protocol (Smith 2007b) was not used; and the search strategy recorded for the 2007 search of MEDLINE was revised in 2009 to better capture the concept of multimorbidity. Results of the search were limited by filters for study design and an extensive list of intervention terms. Search strategies are provided in Appendix 1; Appendix 2; Appendix 3; Appendix 4; Appendix 5. The MEDLINE search strategy was used in HealthSTAR and AMED; the Cochrane search strategy was used in DARE.

Searching other resources

We also:

(a) Searched the reference lists of included papers
(b) Contacted authors of relevant papers regarding any further published or unpublished work where indicated

Data collection and analysis

Selection of studies

All citations identified by the electronic searches were downloaded to reference manager software (EndNote 2013) and duplicates were removed. Potentially relevant studies were identified by review of the titles and abstracts of search results by the lead author (SS). We retrieved full text copies of all articles identified as potentially relevant. Two review authors (SS, HS, or EW) ) independently screened all citations found by the electronic searches and assessed each retrieved article for inclusion. We resolved any disagreement by discussion and consensus.

Data extraction and management

Two review authors (SS, HS or EW ) undertook data abstraction and cross checked data abstraction forms using a modified version of the EPOC data collection checklist (EPOC 2013a). Disagreements about data abstraction and quality were resolved by consensus between the review authors or through adjudication by the Cochrane contact editor.

We extracted the following information from the included studies:
(1) Details of the intervention: a full description of the intervention was extracted as were details regarding aims; clinical protocols; use of case workers; remuneration/payment systems; providers involved; and theoretical framework on which the intervention was based; (2) Participants: patients, the nature of multimorbidity and how it was determined; providers, i.e. specialist and primary care providers, family members; (3) Clinical setting; (4) Study design; (5) Outcomes; (6) Results. Results were organised into: (i) Clinical outcomes; (ii) Mental health outcomes; (iii) Patient‐reported outcomes; (iv) Health service use (v) Recipient and provider behaviours; and (vi) Recipient and provider acceptability/satisfaction.

Assessment of risk of bias in included studies

Two review authors independently assessed the risk of bias in all included studies using standard EPOC criteria (EPOC 2015) and included the following domains: allocation (sequence generation and concealment); baseline characteristics; incomplete outcome data; contamination; blinding; selective outcome reporting; and other potential sources of bias.

Measures of treatment effect

We reported data in natural units for each study. For RCTs, we reported results as (1) Absolute difference (mean or proportion of outcome in intervention group minus control at study completion); (2) Relative percentage difference (absolute difference divided by post‐intervention score in the control group). We undertook meta‐analysis where appropriate in terms of participants, interventions and outcomes using random‐effects models. Analyses were undertaken for clinical outcomes (glycaemic control and blood pressure) and depression scores in the comorbidity studies. We also undertook meta‐analyses for HRQoL and self‐efficacy in all studies in which these were reported. All meta‐analyses apart from self efficacy had significant statistical heterogeneity so we present the figures for these analyses without the pooled estimates of effect.

Standardised effect sizes (SES) are presented in tables where possible, i.e. where studies reported relevant data for their calculation. We have reported the range of effects using SES in the text of the results and used the generally accepted convention that an SES of more than 0.2 indicates a small intervention effect, an SES of more than 0.5 indicates a moderate intervention effect and an SES of more than 0.8 is a large effect size (Cohen 1988).

For ITS we had planned to report two effect sizes:
(1) The change in the outcome immediately after the introduction of the intervention.
(2) The change in the slope of the regression lines.

However, no ITS studies were identified.

Unit of analysis issues

None of the included studies had unit of analysis errors.

Dealing with missing data

If data on multimorbidity sub‐groups were missing from potentially eligible studies, we contacted authors to obtain the information. Two studies provided additional data on sub‐groups with multimorbidity (Coventry 2015; Eakin 2007). We did not include any studies with more than 20% missing data in meta‐analyses and did not make any assumptions regarding missing data.

Assessment of heterogeneity

We assessed included studies in terms of clinical and statistical heterogeneity. Statistical heterogeneity was assessed by examining forest plots and considering the I² statistic (Cochrane Handbook). We planned to prepare tables and funnel plots comparing effect sizes of studies grouped according to potential effect modifiers (for example, simple versus multifaceted interventions) if sufficient studies had been identified but this was not possible.

If there had been enough studies, we had planned to use meta‐regression to see whether the effect sizes could be predicted by study characteristics. These could, for example, include duration of the intervention, age groups, and simple versus multifaceted interventions (Cooper 1994). We also considered formal tests of homogeneity (Petitti 1994). None of these quantitative methods were possible for this version of the review but will be considered for future review updates.

Assessment of reporting biases

We assessed incomplete reporting of outcomes, where possible, within the 'Risk of bias' tables. This was only possible for studies that had published protocols or specifically reported different results than the outcomes mentioned in the methods sections of included papers.

Data synthesis

We expected that included studies would measure similar outcomes using different methods. These included either continuous variables (such as different depression scales) or dichotomous process measures (such as proportion of people with recovery from depression). For continuous outcomes, we reported means and standard deviations at study completion with the absolute difference and relative percentage difference. We calculated standardised effect sizes for continuous measures by dividing the difference in mean scores between the intervention and comparison group in each study, by an estimate of the pooled standard deviation. For categorical outcomes, we reported the proportions in the intervention and control groups with the absolute difference and relative percentage difference.

We undertook meta‐analysis of studies that were similar in terms of settings, participants, interventions, outcome assessment and study methods. If there was a high I² indicating statistical heterogeneity, we used graphs to illustrate the results but did not present the combined effects as the heterogeneity indicates that combining the studies in a meta‐analysis is inappropriate. Where meta‐analysis was not possible we carried out a narrative synthesis of the results and presented the results based on outcome groupings. See Additional tables.

We assessed the certainty of the evidence for the main outcomes using the following GRADE (Grading of Recommendations Assessment, Development and Evaluation) criteria (Guyatt 2008); and present the main findings with our judgments in a 'Summary of findings' table

1. Study limitations (i.e. risk of bias).
2. Consistency of effect.
3. Imprecision.
4. Indirectness.
5. Publication bias.

Subgroup analysis and investigation of heterogeneity

We had planned to consider subgroup analyses based on the degree of multimorbidity of participants estimated by the number of conditions per person. These analyses were not possible due to the variation in definitions of multimorbidity and characteristics of participants across studies.

Sensitivity analysis

We planned to undertake sensitivity analyses based on intervention type or clear distinctions in studies with different risk of bias but this was not possible due to the limited number of meta‐analyses undertaken with each containing relatively few studies.

Results

Description of studies

Results of the search

The electronic searches yielded 30,296 original citations after duplicates were removed Figure 1. Of these, 30,165 citations were irrelevant and directly excluded. Full texts were retrieved for 131 studies. Of these, 74 studies were excluded with reasons Characteristics of excluded studies. Fifteen studies are on‐going (Characteristics of ongoing studies), 17 studies were duplicates or reported secondary data analyses. Eighteen studies from 21 papers were eligible for inclusion in this review and four other studies are awaiting classification.(Characteristics of studies awaiting classification).

Figure 1

Study flow diagram.

Included studies

See Characteristics of included studies table

Study design

We identified 18 RCTs eligible for inclusion in the review, 10 from the original review (Bogner 2008; Boult 2011; Eakin 2007; Gitlin 2009; Hochhalter 2010; Hogg 2009; Katon 2010; Krska 2001; Lorig 1999; Sommers 2000;) and 8 identified in this update (Barley 2014; Coventry 2015; Garvey 2015; Kennedy 2013; Lynch 2014; Martin 2013; Morgan 2013; Wakefield 2012). No other study designs with eligible interventions were identified.

Population/participants

There were a total of 8727 participants across all studies. The interventions varied in duration from eight weeks to two years, with the majority lasting 6 to 12 months. There was also variation in post intervention follow‐up, varying from immediate follow‐up to follow‐up 12 months post intervention cessation.

Nine of the 18 studies recruited participants with a broad range of conditions (Boult 2011; Eakin 2007; Garvey 2015; Gitlin 2009; Hochhalter 2010; Hogg 2009; Krska 2001; Lorig 1999; Sommers 2000), whereas the remaining nine focused on the following comorbidities: depression and hypertension (Bogner 2008); depression and diabetes and/or heart disease (Barley 2014; Coventry 2015; Morgan 2013; Katon 2010); depression and headache (Martin 2013); diabetes and hypertension (Lynch 2014; Wakefield 2012); and a sub‐group of people with at least two of diabetes, chronic obstructive pulmonary disease and irritable bowel syndrome (Kennedy 2013).

Settings

All studies were set in primary care or community settings in the USA, apart from Krska 2001 which was set in the UK National Health Service and Hogg 2009 which was set in Canada. Eight were funded by a government or university grant (Coventry 2015; Garvey 2015; Gitlin 2009; Hogg 2009; Katon 2010; Kennedy 2013; Krska 2001; Lorig 1999); and the remaining studies were funded by charitable foundations. None were funded directly by the pharmaceutical industry.

Comparison intervention

In the majority of included studies, the comparator was usual medical care which in some studies was supplemented by a newsletter or leaflet (Eakin 2007; Gitlin 2009), or involved a baseline assessment but no follow‐on intervention (Bogner 2008; Garvey 2015; Katon 2010; Krska 2001). These minimal additions to usual care could be considered as being within the variation of usual care provided in different settings. One study invited those allocated to a control group to attend a group session based on an unrelated topic (Hochhalter 2010). This was an attempt to ensure that the intervention effect did not relate to the group setting but related to the intervention content.

Description of interventions

The interventions were all multifaceted and brief descriptions for each study are provided in the Characteristics of included studies. No study specifically reported consumer involvement in the intervention design.

As outlined in the methods, we used the EPOC taxonomy of interventions to describe and categorise the interventions tested in these studies (EPOC 2002). While the interventions identified all involved multiple components they could be divided broadly into two main groups. In 12 of 18 studies, the interventions were primarily organisational, for example case management or addition of a pharmacist to the clinical care team (Barley 2014; Bogner 2008; Boult 2011; Coventry 2015; Hogg 2009; Katon 2010; Kennedy 2013; Krska 2001; Martin 2013; Morgan 2013; Sommers 2000; Wakefield 2012). In the remaining six studies, the interventions were primarily patient‐oriented, for example self‐management support groups (Eakin 2007; Garvey 2015; Gitlin 2009; Hochhalter 2010; Lorig 1999; Lynch 2014). However, there were overlapping elements with some organisational‐type studies including patient‐oriented elements such as education provided by a case manager and vice versa. No study involving financial or regulatory type interventions were identified. We have included an additional table which outlines intervention elements and indicates which elements featured in each of the included studies (Table 1)

Excluded studies

We excluded 74 studies in total, see Characteristics of excluded studies. Thirty‐four studies were excluded on the basis of ineligible participants. In some of these studies there was a potential multimorbidity sub‐group but these data were not reported or not available from authors when requested. Twenty‐six studies were excluded on the basis of an ineligible intervention. This was usually because it was conducted in a specialist setting or had a single‐condition focus despite participants having multiple conditions. The remaining studies were excluded on the basis of study design, largely due to absence of control groups.

Risk of bias in included studies

See Characteristics of included studies table, Figure 2 and Figure 3 for a summary assessment of the risk of bias of the included studies. Overall four of the 18 studies reported all elements for the risk of bias domains. Two studies reported domains with a high risk of bias and in 13 studies there were domains classified as unclear due to lack of reporting.

Figure 2

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

Figure 3

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

Allocation concealment was assessed as adequate in nine of the 18 studies (Barley 2014; Boult 2011; Coventry 2015; Garvey 2015; Gitlin 2009; Hogg 2009; Katon 2010; Krska 2001; Sommers 2000), but was assessed as unclear in the remainder. Baseline measurement of outcomes was carried out in all studies. All reported adequate follow‐up of participants except Lorig 1999 and Wakefield 2012 where the risk of bias was assessed as unclear. Lorig 1999 did not provide specific details pertaining to follow‐up for the multimorbidity subgroup, although follow‐up for the overall study was assessed as adequate. There was high risk of bias in Martin 2013 with poorer follow‐up in the intervention group (57%) compared to the control group (80%) at study completion. Objective outcomes were used in all but two studies, Krska 2001 and Hogg 2009, where this dimension was assessed as unclear. Krska 2001 used a measure detailing pharmaceutical care issues (PCIs) which was a previously developed classification system modified for the study. Hogg 2009 collected data on chronic and preventive care delivery from individuals' records but the accuracy of this process was not described. Blinding of outcome assessment was assessed as done in seven studies (Boult 2011; Coventry 2015; Gitlin 2009; Hochhalter 2010; Katon 2010; Lorig 1999; Sommers 2000). It was assessed as unclear in nine studies (Barley 2014; Bogner 2008; Eakin 2007; Hogg 2009; Lynch 2014; Martin 2013; Morgan 2013; Wakefield 2012); and assessed as not done in Garvey 2015 and Krska 2001.

Five of the 18 studies had a cluster design that ensured no contamination of control participants (Boult 2011; Coventry 2015; Kennedy 2013; Morgan 2013; Sommers 2000). Contamination of participants allocated to the control group was unlikely in a further eight studies where the intervention was directed at recipients rather than providers (Barley 2014; Bogner 2008; Garvey 2015; Eakin 2007; Gitlin 2009; Lorig 1999; Lynch 2014; Hochhalter 2010), but was possible in the remaining studies four studies (Hogg 2009; Katon 2010; Martin 2013; Wakefield 2012). However, Katon 2010 provided an appendix outlining potential contamination and indicated that it was minimal and, if it had occurred, it would have diluted rather than increased the significant effect size of their intervention. Krska 2001 stated that contamination of control participants who attended the same general practitioners (GPs) as the intervention participants could have occurred but that a cluster design would have been more problematic due to differential prescribing patterns between practices. All studies had low risk of selective outcome reporting and had no apparent other biases.

The five cluster randomised controlled trials accounted for clustering effects in their analysis so there were no unit of analysis errors (Boult 2011; Coventry 2015; Kennedy 2013; Morgan 2013; Sommers 2000).

Certainty of the evidence

See summary of findings Table for the main comparison. In general, while all the included studies were RCTs the main limitation related to a lack of consistency of effect for most outcomes. Only the mental health outcomes, largely relating to depression in the comorbidity studies, were regarded as having a high GRADE ranking. We downgraded the evidence for effects on clinical and patient‐reported outcomes to moderate due to lack of consistency of effect across studies and small effect sizes. We downgraded the evidence for effects on provider behaviour to moderate, due to limited available data for calculation of standardised effect sizes (SES) and lack of clarity regarding the clinical importance of the results. We downgraded the evidence for effects on health service utilisation and medication use and adherence to low, due to variation across studies and small effect sizes. We did not include economic outcomes in the summary of findings Table for the main comparison due to the lack of robust economic analyses, rather we summarised this outcome in Table 2.

See: Summary of findings for the main comparison

Table 2. Costs

Study	Study type	Outcome	Result	Notes
Barley	RCT	Cost‐effectiveness	The intervention demonstrated marginal cost effectiveness up to a QALY threshold of GBP 3035
Boult	RCT	Total healthcare cost	Saving of USD 75,000 per GCN and USD 1364 per participant	USD in 2007 Initial result only ns
Katon	RCT	Cost‐effectiveness	Mean reduction of 114 days in depression free days and an estimated difference of 0.335 QALYs (95% CI −0.18 to 0.85). The intervention was associated with lower OPD costs with a reduction of USD 594 per participant (95% CI USD −3241 to USD 2053).	Non‐significant but 99.7% probability that the intervention met the threshold of < USD 20,000 per QALY
Krska	RCT	Mean cost of medicines	Int: 38.83 Con: 42.61 Absol diff 3.78 Rel %diff 9%	GBP in 2000 ns SES = 0.13
Lorig	RCT	Intervention cost per completed participant	USD 70	USD in 1998 See text for assumptions made
Lorig	RCT	Cost savings per individual	USD 750	USD in 1998 See text for assumptions made
Sommers	RCT	Savings per individual	USD 90	USD in 1994 See text for assumptions made

* refers to whether original study reported statistically significant improvement in this outcome

Effects of interventions

Effects by type of interventions

We have presented an overview of intervention components for each study, highlighting the main intervention component in bold text and have included a brief summary of the intervention effect on the study primary outcomes in Table 1. The description of intervention components is based on reporting of intervention components in each paper and this is not consistent across studies. For example, most studies were likely to have included training of practitioners involved in interventions but not all studies reported this as an intervention component. We have also presented an overview of results based on whether the studies addressed general multimorbidity or comorbidity in Table 3.

Table 3. Overview of outcomes

Outcome category	Outcome	No. studies with this outcome	No. studies with p< 0.05 for this outcome
Physical Health	Hb1Ac	5	2
	BP	6	2
	Cholesterol	2	1
	Other symptom score	4	1
	Mortality	1	1
Mental Health	Depression scores	8	6
	% improved depression	1	1
	Anxiety scores	4	3
	Cognitive symptom management	1	0
Psychosocial	QoL/general health	10	4
	Functional impairment & disability	6	2
	Social (activity/support)	4	1
	Self efficacy	7	3
	Home hazards	1	0
Health service use	Visits/use service	5	0
	Hospital admission related	6	2
Patient health related behaviours	Exercise/diet	6	2
Medication adherence		5	2
Provider behaviour	Prescribing	3	2
	Disease management	3	3
Costs	Direct costs	5	Not applicable

* Multimorbidity is defined as two or more independent conditions within the same individual whereas comorbidity refers to linked conditions. In this review comorbidity studies included depression and diabetes or depression and hypertension

** The scales or measurements used in each study for the outcomes are described in the Table of included studies

Organisational interventions

Twelve of the 18 included studies had organisational‐type interventions (Barley 2014; Bogner 2008; Boult 2011; Coventry 2015; Hogg 2009; Katon 2010; Kennedy 2013; Krska 2001; Martin 2013; Morgan 2013; Sommers 2000; Wakefield 2012). These predominantly involved case management and coordination of care or the enhancement of skill mix in multidisciplinary teams in addition to delivery of patient care.

1. Clinical outcomes

Eight of the 18 organisational type studies reported clinical outcomes. These studies had a range of standardised effect sizes (SES) varying from 0.01 to 1.6. Interventions aimed at improving management of risk factors in comorbid conditions were more likely to have larger effect sizes (e.g. Bogner 2008; Katon 2010; Morgan 2013).

Five studies reported six measures of glycaemic control (five mean HbA1c and one study reported percentage achieving at least 0.5% reduction in HbA1c). Katon 2010 and Morgan 2013 reported improvements in mean HBA1c; however, Morgan 2013 had a substantial proportion of missing HbA1c data at study completion so these data were not included in the meta‐analysis of HbA1c. Hogg 2009, Lynch 2014 and Wakefield 2012 found little or no difference in HbA1c. Lynch 2014 reported that a higher proportion of intervention participants achieved an absolute reduction in HbA1c of at least 0.5%. The SES ranged from 0.05 to 1.6 but only one of these three studies had an SES greater than 0.5. The mean difference (MD) was 0.02 (95% CI −0.21 to 0.25) as outlined in Figure 4,

Figure 4

Forest plot of comparison: 1 Glycaemic control (HbA1c) Diabetes outcome: 1.1 HbA1c.

Four studies reported on systolic blood pressure (SBP). Bogner 2008 and Katon 2010 reported improvements in blood pressure, although this was of minimal clinical significance in Katon 2010. Morgan 2013 and Wakefield 2012 reported little difference. The standardised effect sizes (SES) ranged from 0.01 to1.12 but only one of these four studies had an SES greater than 0.5. The MD was −3.10 (95% CI −7.26 to 1.06) as illustrated in Figure 5.

Figure 5

Forest plot of comparison: 2 Systolic Blood Pressure: outcome: 2.1 Systolic blood pressure.

Two studies reported on cholesterol. Katon 2010 found a reduction in LDL cholesterol, whereas Morgan 2013 found no meaningful difference (SES ranges 0.22 to 0.26). Katon 2010 reported a composite primary outcome that combined three risk factors, which showed an improvement in intervention participants compared to control (see Table 4).

Table 4. Clinical Outcomes

Study	Study type	Outcomes	Results	Notes
Barley	RCT	% with angina (Rose Angina score)	Int 22/31 Con 30/37 Absol diff 8, Rel % diff 27%	ns
Bognor	RCT	Systolic BP	Int 127.3 (SD 17.7) Con 141.3 (SD 18.8) Absol diff 14, Rel % diff 10%	* SES = 1.12
Bognor	RCT	Diastolic BP	Int 83 (SD 10.7) Con 81.4 (SD 11.1) Absol diff 9.2, Rel % diff 11%	* SES = 0.8
Gitlin	RCT	Mortality	Int 9/160 (0.06) Con: 21/159 (0.13) Absol diff 7, Rel % diff 54%	*
Hogg	RCT	Systolic BP	Int 124.3 Con 124.2 Absol diff 0.1, Rel % diff < 0.1%	ns (No SDs reported)
Hogg	RCT	HbA1c	Int 7.01 Con 6.78 Absol diff 0.23, Rel % diff 3%	ns
Katon	RCT	Systolic BP	Int 131 (SD 18.4) Con 132.3 (SD 17.2) Absol diff 1.3, Rel % diff 1%	* SES = 0.07
Katon	RCT	HbA1c	Int 8.14 (SD 2.03) Con 8.04 (SD 1.87) Absol diff 0.1, Rel % diff 13%	* SES = 0.32
Katon	RCT	Cholesterol	Int 91.9 (SD 36.7) Con 101.4 (SD 36.6) Absol diff 9.5, Rel % diff 9%	* SES = 0.26
Katon	RCT	Composite: all three risk factors (BP, HbA1c and cholesterol) below guidelines	Int 36/97 (0.37) Con: 19/87 (0.22) Absol diff 15, Rel % diff 68%	*
Lorig	RCT	Pain/ physical discomfort	Int 59.8 (SD 20.1) Con 60.6 (SD 17.1) Absol diff 0.8, Rel % diff 1%	SES = 0.04 ns
Lorig	RCT	Energy/fatigue	Int 2.18 (SD 0.73) Con 2.02 (SD 0.75) Absol diff 0.16, Rel % diff 8%	ns
Lorig	RCT	Shortness of breath	Int 1.34 (SD 0.91) Con 1.58 (SD 0.83) Absol diff 0.24, Rel % diff 15%	ns
Lynch	RCT	HbA1C	Int 7.9 (SD 1.6) Con 7.4 (SD 1.6) Absol diff 0.5, Rel % diff 6.7%	ns SES = 0.31
Lynch	RCT	% with at least 0.5 absolute reduction in HbA1c	Int 15/30 (0.05) 7/31 Con (0.21) Absol diff 29, Rel % diff 138%	*
Lynch	RCT	Mean SBP	Int 135.8 (SD 21.4) Con 136.7 (SD 23) Absol diff 0.9, Rel % diff 0.6%	ns SES = 0.01
Morgan	RCT	HbA1C	Int 6.9 (SD 0.26) Con 7.4 (SD 0.36) Absol diff 0.5, Rel % diff 6.7%	* SES = 1.6
Morgan	RCT	Systolic BP	Int 134.2 (SD 3.0) Con 133.5 (SD 3.8) Absol diff 0.7, Rel % diff 0.5%	ns SES = 0.2
Morgan	RCT	Cholesterol	Int 4.22 (SD 0.14) Con 4.44 (SD 0.2) Absol diff 0.22, Rel % diff 5%	ns SES = 0.22
Morgan	RCT	Mean BMI	Int 31.2 (SD 1.0) Con 31.0 (SD 1.0) Absol diff 0.2, Rel % diff 0.6%	ns SES = 0.2
Sommers	RCT	Symptom scores	Int 17.2 Con 18.9 Absol diff 1.7, Rel % diff 9%	ns
Wakefield	RCT	HbA1c	Int 6.9 (1.1) Con 6.95 (1.1) Absol diff 0.05, Rel % diff 0.7%	ns SES = 0.05
Wakefield	RCT	Systolic BP	Int 133 (16.6) Con 137 (17.3) Absol diff 4, Rel % diff 3%	ns SES = 0.24
Martin	RCT	Mean headache rating	Int 0.63 (SD 0.5) Con 1.01 (SD 0.83) Absol diff 0.38, Rel % diff 38%	* SES = 0.58

* refers to whether original study reported statistically significant improvement in this outcome

** Total number with final data collected was 384. No final numbers of intervention and control participants presented.

Four studies reported symptom scores relating to clinical outcomes. Barley 2014, Lorig 1999 and Sommers 2000 found little or no difference whereas Martin 2013 reported improvements in mean headache rating (see Table 4).

2. Mental health outcomes

Seven studies presented data on mental health outcomes (Barley 2014; Bogner 2008; Coventry 2015; Katon 2010; Martin 2013; Morgan 2013; Sommers 2000). Five of the seven studies reported improvements in a range of depression measures whereas two showed no improvements in depression outcomes (Barley 2014; Sommers 2000). We undertook two meta‐analyses: a meta‐analysis of Patient Health Questionnaire, version 9 (HQ9) depression scores; and a meta‐analysis of standardised mean difference (SMD) in depression scores for the studies with available data where depression was a targeted condition. This suggests a modest intervention effect. The meta‐analysis for PHQ9 scores had high heterogeneity so we do not report the pooled effect (Figure 6). The SMD for other depression scores was −0.41 (95% CI −0.63 to −0.20) (Figure 7). The range in SESs for depression outcomes across these studies was from 0.09 to 2.24 with five of the nine outcomes indicating moderate to large effect sizes (i.e. SES > 0.5). These higher effect sizes were all reported in the studies in which depression was a focus of the intervention.

Figure 6

Forest plot of comparison: 3 Depression scores: 3.1 PHQ9 Depression scores.

Figure 7

Forest plot of comparison: 4 Depression scores: 4.1 Depression scores.

Three studies reported on anxiety measures, two showed improvements (Coventry 2015 and Martin 2013) whereas Barley 2014 reported little difference (see Table 5). There were small effect sizes in all studies (SES range 0.08 to 0.26).

Table 5. Mental Health Outcomes

Study	Study type	Outcome	Result	Notes
Barley	RCT	PHQ9 depression score	Int 12.6 (SD 7.1) Con 12 (SD 6.9) Absol diff 0.6, Rel % diff 8%	ns SES = 0.09
Barley	RCT	HADS depression score	Int 9.5 (SD 4.6) Con 8.8 (SD 4.8) Absol diff 0.7, Rel % diff 8%	ns SES = 0.15
Barley	RCT	HADS anxiety score	Int 9.9 (SD 7.1) Con 9.5 (SD 5.4) Absol diff 0.4, Rel % diff 4%	ns SES = 0.08
Bognor	RCT	CES depression score	Int 9.9 (SD 10.7) Con 19.3 (SD 15.2) Absol diff 9.4, Rel % diff 49%	* SES = 0.75
Coventry	RCT	SCL‐D13 depression score	Int 1.76 (SD 0.9) Con 2.02 (SD 0.9) Absol diff 2.6, Rel % diff 13%	* SES = 0.28
Coventry	RCT	PHQ9 depression score	Int 11.3 (SD 6.5) Con 13.1 (SD 6.5) Absol diff 1.8, Rel % diff 14%	* SES = 0.28
Coventry	RCT	GAD‐7 anxiety score	Int 8.2 (SD 5.8) Con 9.7 (SD 5.9) Absol diff 1.5, Rel % diff 15%	* SES = 0.26
Garvey	RCT	HADS total score	Int 15.6 (SD 8.3) Con 16.7 (SD 8.2) Absol diff 1.1, Rel % diff 6.5%	ns SES = 0.13
Katon	RCT	SCL 20 depression score	Int 0.83 (SD 0.66) Con 1.14 (SD 0.68) Absol diff 0.31, Rel % diff 27%	* SES = 0.46
Katon	RCT	Patient global improvement in depression	Int 41/92 Con 16/91 Absol diff 27, Rel % diff 150%	*
Lorig	RCT	Cognitive symptom management score	Int 1.75 Con 0.98 Absol diff 0.77, Rel % diff 79%	ns
Martin	RCT	PHQ9 depression score	Int 6.7 (SD 4.6) Con 12.6 (SD 5.3) Absol diff 5.9, Rel % diff 47%	* SES = 1.18
Martin	RCT	BDI ‐Depression score	Int 13.1 (SD 8.6) Con 28.7 (SD 9.5) Absol diff 15.6, Rel % diff 54%	* SES = 1.73
Martin	RCT	BAI Anxiety score	Int 10.5 (SD 10.8) Con 16.4 (SD 9.3) Absol diff 5.9, Rel % diff 36%	* SES = 0.1
Morgan	RCT	PHQ9 depression score	Int 7.1 (SD 0.8) Con 9.0 (SD 0.9) Absol diff 1.9, Rel % diff 21%	* SES = 2.24
Sommers	RCT	GDS score (depression)	Int 4.1 Con 4.1 Absol diff 0, Rel % diff 0%	ns

* refers to whether original study reported statistically significant improvement in this outcome

3. Patient‐reported outcome measures

Nine of the organisational‐type studies presented patient‐reported outcome measures (PROMs).

Nine of these reported a variety of HRQoL measures with a range of effects from SES of 0.03 to 1.7. Only one of the nine studies reported a large effect size (Coventry 2015). Two studies reported small effect sizes (Katon 2010; Martin 2013). The remaining six studies reported little or no effect (Barley 2014; Hogg 2009; Kennedy 2013; Krska 2001; Morgan 2013; Sommers 2000). Krska 2001 and Morgan 2013 reported that SF36 scores had been analysed across eight domains at study completion and reported little or no difference between groups, but did not present actual data. The mixed evidence regarding HRQoL is illustrated in Figure 8 which includes studies with available data but the pooled effect is not reported due to high statistical heterogeneity (I² = 73%).

Figure 8

Forest plot of comparison: 5 Health related quality of life, outcome: 5.1 HRQoL.

Five organisational studies reported on self‐efficacy with a range in SES of 0.03 to 0.11, suggesting minimal effect. (Barley 2014; Hochhalter 2010; Kennedy 2013; Wakefield 2012; Coventry 2015). We undertook a meta‐analysis of standardised mean self‐efficacy scores in comorbidity studies and found no effect, SMD −0.05 (95% CI −0.12 to 0.22) (Figure 9).

Figure 9

Forest plot of comparison: 6 Self‐Efficacy, outcome: 6.1 Self‐efficacy score.

Two of the organisational studies reported measures relating to disability or impaired activities of daily living (IADL). Hogg 2009 reported no effect of interventions on IADL, whereas Coventry 2015 reported an improvement in the Sheehan Disability score in intervention participants.

Two of the organisational studies reported measures relating to Illness perceptions and both reported no effect (Barley 2014; Coventry 2015).

A range of other PROMs were also reported with mixed effects and none had an SES greater than 0.3. These are presented in Table 6.

Table 6. Patient‐reported outcome measures

Study	Study type	Outcome	Result	Notes
Health Related Quality of Life
Barley	RCT	SF12 PCS	Int 32.4 (SD 10.7) Con 33.3 (SD 9.2) Absol diff 0.7, Rel % diff 2%	ns SES = 0.07
Barley	RCT	SF12 MCS	Int 34.5 (SD 11.6 ) Con 33.6 (SD 12.5 ) Absol diff 0.9 , Rel % diff 3%	ns SES = 0.08
Barley	RCT	HRQoL (WEMWBS)	Int 40.6 (SD 11.2) Con 39.6(SD 12.3) Absol diff 1, Rel % diff 2.5%	ns SES = 0.08
Coventry	RCT	HRQoL (WHOQOL)	Int 2.99 (SD 0.6) Con 2.91 (SD 0.6) Absol diff 0.08, Rel % diff 3%	* SES = 1.7
Garvey	RCT	HRQoL (EQ5D VAS)	Int 65.7 (SD 20.2) Con 50.5 (SD 16.3) Absol diff 15.2, Rel % diff 30%	* SES = 0.84
Hogg	RCT	SF 36 Mental Health	Int 52.4 Con 52.2 Absol diff 0.2, Rel % diff 0.3%	ns
Hogg	RCT	SF 36 Physical Health	Int 44.3 Con 41.5 Absol diff 2.8, Rel % diff 6.7%	ns
Katon	RCT	QoL score	Int 6.0 (SD 2.2) Con 5.2 (SD 1.9) Absol diff 0.8, Rel % diff 15%	* SES = 0.44
Kennedy	RCT	HRQoL (EQ5D)	Int 0.56 (SD 0.34) Con 0.57 (SD 0.32) Absol diff 0.01, Rel % diff 1%	ns SES = 0.03
Lorig	RCT	Psychological well‐being	Int 3.47 Con 3.33 Absol diff 0.04, Rel % diff 4%	ns SES = 0.21
Martin	RCT	HRQol (AQOL)	Int 26.3 (SD 4.76) Con 28.4 (SD 4.97) Absol diff 2.1, Rel % diff 7 %	* SES = 0.4
Sommers	RCT	SF36 score	Int 2.2 Con 3.3 Absol diff 1.1, Rel % diff 33%	ns
Self‐efficacy
Barley	RCT	Self‐efficacy score	Int 28.6 (SD 6.7) Con 27.9 (SD 8.1) Absol diff 0.11, Rel % diff 2.5%	ns SES = 0.09
Coventry	RCT	Self‐efficacy score	Int 5.72 (SD 1.9) Con 5.53 (SD 1.9) Absol diff 0.18, Rel % diff 3.2%	ns * SES = 0.09
Garvey	RCT	Self efficacy score	Int 6.8 (SD 1.5) Con 5.3 (SD 1.9) Absol diff 1.47, Rel % diff 28%	* SES = 0.86
Hochhalter	RCT	Self‐efficacy	Int 7.4 Con 8.0 Absol diff 0.6, Rel % diff 7.5%	ns
Kennedy	RCT	Self‐efficacy	Int 68 (SD 23.4) Con 68.7 (SD 23.1) Absol diff 0.7, Rel % diff 1%	ns SES = 0.03
Wakefield	RCT	Self‐efficacy	Int 8.1 (SD 1.9) Con 8.3 (SD 1.9) Absol diff 0.2, Rel % diff 2.4%	ns SES = 0.11
Daily function and disability
Coventry	RCT	Sheehan Disability Score	Int 5.73 (SD 2.8) Con 5.83 (SD 2.8) Absol diff 0.1, Rel % diff 2%	* SES = 0.04
Garvey	RCT	Frenchay Activities Index	Int 21.3 (SD 7.9) Con 18.9 (SD 7.2) Absol diff 2.4, Rel % diff 13%	* SES = 0.32
Garvey	RCT	Activities daily living: NEADL (total)	Int 47.2 (SD 11.9) Con 40.7 (SD 10.7) Absol diff 6.5, Rel % diff 16%	* SES = 0.58
Hogg	RCT	IADL	Int 10.6 Con 10.9 Absol diff 0.3, Rel % diff 2.7%	ns
Lorig	RCT	Disability	Int 0.86 Con 0.96 Absol diff 0.1, Rel % diff 10%	ns
Lorig	RCT	Social role/activity limitation	Int 1.91, Con 1.98 Absol diff 0.07, Rel % diff 4%	ns
Illness perceptions
Coventry	RCT	Multimorbidity illness perception scale	Int 2.1 (SD 0.9) Con 2.28 (SD 0.9) Absol diff 0.18, Rel % diff 8%	ns SES = 0.2
Barley	RCT	Illness perceptions (BIPQ)	Int 40 (SD 14.8) Con 43(SD 31.1) Absol diff 3, Rel % diff 7%	ns SES = 0.22
Social support
Coventry	RCT	Social support (ESSI)	Int 3.29 (SD 1.1) Con 3.4 (SD 1.0) Absol diff 0.11, Rel % diff 3%	ns SES = 0.11
Eakin	RCT	Multilevel support for healthy lifestyle	Int 2.7 Con 2.59 Absol diff 0.11, Rel % diff 4%	ns
Other PROMs
Barley	RCT	Patient‐reported needs (PSYCHLOPS)	Int 13.6 (SD 5.1) Con 13.4 (SD 5.4) Absol diff 0.2, Rel % diff 1.5%	ns SES = 0.04
Hochhalter	RCT	Total unhealthy days	Int 15.3 Con 14.1 Absol diff 1.2, Rel % diff 9%	ns
Hogg	RCT	Total unhealthy days	Int 7.6 Con 9.9 Absol diff 2.3, Rel % diff 23%	ns
Kennedy	RCT	Shared decision making (HCCQ)	Int 67.7 (SD 28) Con 69.3 (SD 26.1) Absol diff 1.6, Rel % diff 2%	ns SES = 0.06
Lorig	RCT	Self‐rated health	Int 3.42 Con 3.44 Absol diff 0.02, Rel % diff 0.6%	ns
Lorig	RCT	Health distress	Int 1.97 Con: 2.13 Absol diff 0.16, Rel % diff 7.5%	ns SES = 0.16
Sommers	RCT	Social activities count	Int 8.7 Con:8.6 Absol diff 0.1, Rel % diff 1%	* (when adjusted for baseline diff)
Sommers	RCT	HAQ score	Int 0.44 Con 0.5 Absol diff 0.06, Rel % diff 12%	ns

* refers to whether original study reported statistically significant improvement in this outcome

4. Utilisation of health services

Five organisational studies reported outcomes on health services utilisation (Boult 2011; Hogg 2009; Katon 2010; Krska 2001; Sommers 2000). Sommers 2000 reported improvements for intervention group participants across a variety of measures relating to hospital admissions, whereas Boult 2011, Hogg 2009, Katon 2010 and Krska 2001 found no difference in admission‐related outcomes, although numbers of admissions in most of these studies were very small.

Three studies reported data in relation to health service visits with a range of providers none of which showed clear improvements in appropriate health service use (Boult 2011; Hogg 2009; Sommers 2000) (see Table 7). No studies that included health service utilisation reported data that could be used to calculate SESs.

Table 7. Health service use

Study	Study type	Outcome	Result	Notes
Boult	RCT	No. hospital admissions	Int 0.7 Con 0.72 Absol diff 0.02, Rel % diff 3%	ns
Boult	RCT	No. days in hospital	Int 4.26 Con 4.49 Absol diff 0.23, Rel % diff 5%	ns
Boult	RCT	No. ED visits	Int 0.44 Con 0.44 Absol diff 0, Rel % diff 0	ns
Boult	RCT	No. PC visits	Int 9.98 Con 9.88 Absol diff 0.1, Rel % diff 1%	ns
Boult	RCT	No. specialist visits	Int 9.04 Con 8.49 Absol diff 0.55, Rel % diff 6%	ns
Boult	RCT	No. home healthcare episodes	Int 0.99 Con 1.3 Absol diff 0.31, Rel % diff 24%	*
Hogg	RCT	No. hospital admissions	Int 0.4 Con 0.46 Absol diff 0.06, Rel % diff 13%	ns
Hogg	RCT	Proportion hospitalised	Int 0.26, Con 0.26 Absol diff 0, Rel % diff 0%	ns
Hogg	RCT	No. ED visits	Int 0.63 Con 0.73 Absol diff 0.01, Rel % diff 14%	ns
Hogg	RCT	Proportion with ED visit	Int 0.38 Con 0.42 Absol diff 0.04, Rel % diff 9%	ns
Katon	RCT	Proportion hospitalised	Int 0.26 Con 0.22 Absol diff 0.04, Rel % diff 18%	ns
Lorig	RCT	No. doctor and ED visits	Int 6.51 Con 7.08 Absol diff 0.57, Rel % diff 8%	ns
Lorig	RCT	No. hospital stays in past 6 months	Int 0.26 Con 0.31 Absol diff 0.05, Rel % diff 6%	*
Lorig	RCT	No. nights in hospital in last 6 months	Int 1.3 Con 1 Absol diff 0.3 Rel % diff 30%	*
Sommers	RCT	No. hospital admissions per individual per year	Int 0.36 Con 0.52 Absol diff 0.16, Rel % diff 31%	*
Sommers	RCT	≥1 60 day readmission	Int 3.6 Con 9.4 Absol diff 5.8, Rel % diff 62%	*
Sommers	RCT	≥ 1 hospital admission	Int 8.8 Con 7.7 Absol diff 1.1, Rel % diff 14%	*
Sommers	RCT	No. PCP visits	Int 6.0 Con 6.1 Absol diff 0.1, Rel % diff 2%	ns
Sommers	RCT	No. office visits	Int 11 Con 12.5 Absol diff 1.5, Rel % diff 12%	*
Sommers	RCT	≥ 1 home care visit	Int 19.5 Con 18.8 Absol diff 0.7, Rel % diff 4%	ns
Sommers	RCT	No. medical specialist visits	Int 1.4 Con 1.7 Absol diff 0.3, Rel % diff 18%	ns
Sommers	RCT	No. other visits	Int 3.9 Con 4.3 Absol diff 0.4, Rel % diff 9%	*
Sommers	RCT	≥ 1 ED visit	Int 21.4 Con 16.7 Absol diff 4.7, Rel % diff	ns

* refers to whether original study reported statistically significant improvement in this outcome

5. Patient behaviour

5.1 Medication use and adherence

Four organisational studies reported measures relating to medication use and adherence. Two of these studies found an effect whereas two did not; and there was a range in SESs from 0.2 to 0.28 indicating minimal intervention effects. Bogner 2008 reported improvements in proportions of intervention participants adhering to both antidepressant and antihypertensive medication as measured using automated counting systems in the caps of medicine bottles (MEMS caps). Morgan 2013 reported a lower proportion of intervention participants were taking anti‐depressant medication. Martin 2013 reported on mean daily medication use which was not significantly different between intervention and control participants. Wakefield 2012 reported two measures of medication‐taking adherence both of which showed no significant difference; (see data in Table 8).

Table 8. Medication use and adherence and prescribing

Study	Study type	Outcome	Results	Notes
Bognor	RCT	≥80% adherence to antidepressant medication (MEMS caps)	Int 23/32 Con 10/32 Absol diff 0.41, Rel % diff 132%	*
Bognor	RCT	≥80% adherence to antihypertensive medication (MEMS caps)	Int 25/32 Con 10/32 Absol diff 0.47, Rel % diff 152%	*
Martin	RCT	Mean daily medication use	Int 2.4 (SD 3.2) Con 3.0 (SD 2.8) Absol diff 0.6, Rel % diff 20%	ns SES = 0.2
Morgan	RCT	% taking antidepressant medication	Int 34/62 Con 36/113 Absol diff 0.11, Rel % diff 34%	*
Wakefield	RCT	Adherence (Edward's scale)	Int 3.4 (SD 0.5) Con 3.3 (SD 0.5) Absol diff 0.1, Rel % diff 3%	ns SES = 0.2
Wakefield	RCT	Medication Taking Adherence Score	Int 100 (SD 1.4) Con 98.9 (SD 6.0) Absol diff 1.1, Rel % diff 1%	ns SES = 0.28

* refers to whether original study reported statistically significant improvement in this outcome

5.2 Health related behaviours

Three organisational studies provided data on a variety of outcomes relating to health behaviours by participants (Katon 2010; Morgan 2013; Sommers 2000). Katon 2010 found no difference in relation to adherence to diet and exercise. Morgan 2013 presented self‐report data on three patient‐behaviour outcomes with improvements in proportions of individuals exercising (Int 60% vs Con 29%) and in the proportions smoking (Int 8% vs Con 12%) and consuming alcohol (49% vs 64%). Sommers 2000 found no changes in a nutrition checklist score. No studies reporting health‐related behaviours reported data that could be used to calculate SESs; (see data in Table 9).

Table 9. Health‐related participant behaviours

Study	Study type	Outcome	Results	Notes
Hochhalter	RCT	PAM (patient activation measure)	Int 66.8 Con 66.2 Absol diff 0.6, Rel % diff 1%	ns
Eakin	RCT	Diet behaviour scores	Int 2.2 Con 2.41 Absol diff 0.21, Rel % diff 9%	*
Eakin	RCT	Change minutes of walking/week	Int +8 Con −10 Absol diff 18, Rel % diff 180%	*
Katon	RCT	General adherence to diet score	Int 0.86 Con 0.81 Absol diff 0.05, Rel % diff 6%	ns
Katon	RCT	General adherence to exercise score	Int 0.54 Con 0.44 Absol diff 0.1, Rel % diff 23%	ns
Lorig	RCT	Exercise: stretching and strengthening (mins/week)	Int 53.1 Con 40.4 Absol diff 12.7,Rel % diff 31%	ns
Lorig	RCT	Exercise: aerobic (mins/week)	Int 101.8 Con 88 Absol diff 13.8, Rel % diff 157%	ns
Lorig	RCT	Communication with doctor (score 1‐5)	Int 3.34 Con 3.2 Absol diff 0.14, Rel % diff 4%	ns
Lynch	RCT	Physical activity (kcal expenditure per week, CHAMPS)	Int 1913.6 Con −603 Absol diff 2516, Rel % diff 417%	*
Morgan	RCT	Smoking	Int 13/162 Con 13/110 Absol diff 0.04, Rel % diff 33%	ns
Morgan	RCT	Alcohol	Int 51/104 Con 27/42 Absol diff 0.15, Rel % diff 23%	ns
Morgan	RCT	Exercise (30 minutes/day for 5 days/ week)	Int 97/162 Con 22/75 Absol diff 0.31, Rel % diff 106%	*
Sommers	RCT	Nutrition checklist score	Int 2.0 Con1.9 Absol diff 0.1, Rel % diff 5%	ns

* refers to whether original study reported statistically significant improvement in this outcome

6. Provider behaviour

6.1 Prescribing

Two organisational studies reported measures relating to practitioner prescribing or medicines management, both of which indicated significant benefits for intervention participants. Katon 2010 reported a measure examining one or more medication adjustments for five classes of drugs related to the comorbid conditions being studied and reported differences for four of these five groups. Krska 2001 reported a reduction in pharmaceutical care issues in intervention participants; (see data in Table 8)

6.2 Other provider behaviours

Provider behaviours relating to chronic disease management or preventive care were reported in four organisational studies. Boult 2011 and Coventry 2015 both presented a validated measure called the Patient Assessment of Chronic Illness Care (PACIC) score, which is a patient assessment of the quality of care received. This score includes five elements and the aggregate quality score derived was improved in the Boult 2011 Guided Care study, and a small effect reported by Coventry 2015 study (SES 0.39). Hogg 2009 reported measures relating to chronic disease management and preventive care based on chart data and both were improved in the intervention group. Morgan 2013 reported on the proportions of particpants referred to exercise and mental health programmes which was higher in intervention than control group participants; (see data in Table 10).

Table 10. Provider behaviour

Study	Study type	Outcome	Result	Notes
Boult	RCT	PACIC score (patient measure of quality of care received)	Int 3.14 Con 2.85 Absol diff 0.29, Rel % diff 10%	*
Coventry	RCT	PACIC score	Int 2.37 (SD 1.1) Con 1.98 (SD 1.0) Absol diff 0.39, Rel % diff 20%	ns SES = 0.39
Hogg	RCT	Chronic Disease Mangement Score	Int 0.84 Con 0.77 Absol diff 0.07, Rel % diff 9%	*
Hogg	RCT	Preventive Care Score	Int 0.89 Con 0.7 Absol diff 0.19, Rel % diff 27%	*
Krska	RCT	% Pharmaceutical care issues resolved from baseline	Int 950/1206 Con 542/1380 Absol diff 0.4, Rel % diff 102%	*
Morgan	RCT	% Referred to mental health	Int 58/162 Con 10/111 Absol diff 0.27, Rel % diff 300%	*
Morgan	RCT	% Referred to exercise programme	Int 58/162 Con 24/114 Absol diff 0.15, Rel % diff 71%	*

* refers to whether original study reported statistically significant improvement in this outcome

7. Acceptability of services

Three organisational studies reported treatment satisfaction. Katon 2010 reported the proportion of participants moderately to very satisfied with treatment for depression and diabetes and heart disease at study completion. More intervention participants were satisfied with their care at study completion compared to those experiencing usual care. Boult 2011 reported on the changes in satisfaction for providers as part of an overall examination of the effect of the intervention on providers. The measure incorporated changes in 11 domains of satisfaction with service provision; five domains relating to time spent with participants; six domains relating to provider knowledge of the participant; and four domains relating to care coordination. The only changes reported in the study were improvements in 5 of the 11 domains relating to satisfaction with service provision. Coventry 2015 reported mean Client Satisfaction Scores and reported no difference between intervention and control group participants.

8. Costs

Five organisational studies provided data on costs (Barley 2014; Boult 2011; Katon 2010; Krska 2001; Sommers 2000).

Barley 2014 undertook a parallel economic analysis of the UPBEAT intervention and found that the intervention demonstrated marginal cost effectiveness up to a quality‐adjusted life‐year (QALY) threshold of GBP 3035.

Leff 2009 et al provided initial cost data from Boult 2011 and indicated a saving related to Guided Care of USD 75,000 per guided care nurse (95% CI USD −244,000 to USD 150,900) or USD 1364 per individual. However, these initial results were based on small changes in outcomes with wide confidence intervals. In addition, the final study results were subsequently published and indicated no intervention effect.

Katon 2010 reported the direct mean medical costs relating to the TeamCare intervention over a 12 month period as USD 1224 per individual. A subsequent economic analysis reported that the intervention led to an additional 114 days in depression‐free days and an estimated difference of 0.335 QALYs (95% CI −0.18 to 0.85) (Katon 2012). The intervention was associated with lower OPD costs with a reduction of USD 594 per person (95% CI USD −3241 to USD 2053). There was a 99.7% probability that the intervention met the threshold of less than USD 20,000 per QALY. The authors interpreted this as a high value intervention but this must be interpreted with caution given the wide confidence intervals in the estimates with lack of statistical significance.

Krska 2001 reported the mean medicine cost for the intervention and control groups at study completion and found a marginal benefit for the intervention.

Sommers 2000 reported a net saving per intervention participant of USD 90 though this did not include additional savings from fewer physician visits. It also excluded the costs of implementing the intervention, stated to be USD 118,950, mainly relating to salary costs; (see Table 2).

Patient‐oriented interventions

Six of the 18 included studies had predominantly patient‐oriented interventions, for example education or group‐based self‐management support courses (Eakin 2007; Garvey 2015; Gitlin 2009; Hochhalter 2010; Lorig 1999; Lynch 2014). All six aimed to address participant health‐related behaviour and did not engage or involve individuals' current health‐care providers directly. The results from these six studies were mixed and do not suggest that patient‐oriented interventions are generally effective. However, there was an indication that a focus on functional capacity and activity participation may be effective (Garvey 2015; Gitlin 2009), with one study reporting a reduction in mortality at longer‐term follow‐up (Gitlin 2006).

1. Clinical outcomes

Three of the five patient‐oriented studies reported clinical outcomes with mixed results. Gitlin 2009 published a follow‐up paper looking at long‐term effects of their intervention on mortality and found reduced mortality in intervention participants, which persisted up to three and a half years post intervention, (Int (n = 160): 6% mortality, Con (n = 159): 13% mortality, Absol diff 7%, Rel % diff 54% (Gitlin 2009). Lorig 1999 reported three measures relating to clinical outcomes all of which showed little or no difference between intervention and control. Lynch 2014 reported on glycaemic and blood pressure control in people with diabetes and hypertension. Mean HbA1c was no different but there was an increase in the proportion of intervention participants who achieved at least an absolute reduction in HbA1c of 0.5%. There was no or little difference in systolic blood pressure; (see Table 4). SESs for clinical outcomes in these studies ranged from 0.01 to 0.31 indicating minimal intervention effects.

2. Mental health outcomes

Two studies presented data on mental health outcomes (Garvey 2015 and Lorig 1999). Garvey 2015 reported Hospital Anxiety and Depression Scores (HADS) and found no overall difference in total HADS scores but modest improvements in the depression and anxiety scores. Lorig 1999 reported a mean difference of 0.77 points on a scale of 0 to 5, suggesting no difference in cognitive symptom management between groups at study completion; (see Table 5).

3. Patient‐reported outcome measures

Five studies reported PROMs (Eakin 2007; Garvey 2015; Gitlin 2009; Hochhalter 2010; Lorig 1999). Garvey 2015's primary and secondary outcomes reflected the occupational therapy basis of the intervention. The intervention was associated with improvements in all three reported occupational participation/functional ability‐type measures. Garvey 2015 also found improvements in HRQol and self‐efficacy but no improvements in the Health Education Impact questionnaire overall. The results relating to HRQol and self‐efficacy are included in the related meta‐analyses (Figure 8; Figure 9). Results of this study have to be interpreted with caution as it is reported as an exploratory trial with immediate post‐intervention follow‐up. A definitive RCT is planned (Garvey 2015). Gitlin 2009 reported six PROMs by presenting difference in adjusted means between intervention and control groups at follow‐up and two showed improvement (self‐efficacy in relation to fear of falling and improvements in control‐oriented strategies). The range in SESs across these studies, when data were available, was 0.16 to 0.86 with all higher intervention effects relating to outcomes from Garvey 2015 (see Table 6).

4. Utilisation of health services

Two studies reported outcomes on health services utilisation (Garvey 2015 and Lorig 1999). Garvey 2015 found no difference in primary care visits and hospital admissions although only examined an eight week time frame in a small sample. Lorig 1999 reported improvements for intervention group participants across a variety of measures relating to hospital admissions. Lorig 1999 also reported on primary care and emergency department visits but found no improvements (no data available to calculate SESs); (see Table 7).

5. Patient behaviour

5.1 Medication use and adherence

No study with a patient‐oriented intervention reported on medication use and adherence.

5.2 Health related behaviours

Three studies provided data on a variety of outcomes relating to health behaviours by participants (Eakin 2007, Lorig 1999 and Lynch 2014). Eakin 2007 reported improvements in diet behaviour scores and in changes in minutes of walking per week. Lorig 1999 reported three measures relating to exercise and communication with doctors and while there was moderate differences in favour of the intervention groups these were unlikely to be of clinical significance; (see Table 9). Lynch 2014 reported increased exercise measured by caloric expenditure in the intervention group. There were no data presented to calculate SESs.

6. Provider behaviour

Prescribing and other provider behaviours

No study with a patient‐oriented intervention reported on provider behaviour.

7. Acceptability of services

No study with a patient‐oriented intervention reported on acceptability of services.

8. Costs

Two studies provided data on costs (Gitlin 2009, Lorig 1999).

Gitlin 2009 reported the direct costs associated with the intervention at USD 1222 per experimental participant. This incorporated USD 439 equipment costs and USD 783 therapy costs.

Lorig 1999 reported the mean direct cost of running the course for participants who completed it, although costs did not include the cost of accommodation as this was donated to the study. The significant reduction in hospital admissions shown by the intervention translated to a saving in healthcare costs per participant of USD 750 which the authors point out is ten times the cost of the intervention. This calculation was based on a presumed cost of USD 1000 per day if admitted to hospital (see Table 2).

Discussion

Summary of main results

We have identified 18 studies eligible for inclusion in the review, 10 from the original review and 8 added in the current update. All 18 were randomised controlled trials with a generally low risk of bias. Even within this small number of studies, there was significant variation in participants and interventions. In nine of the 18 studies, the focus was on comorbid conditions, which were eligible for inclusion as their interventions had a multimorbidity focus in that they were directed at the pre‐specified comorbid conditions. The commonest combinations of conditions included depression, diabetes and cardiovascular disease. In the other studies, which included people with general multimorbidity, the focus tended to be on older individuals.

The results suggest that interventions that are targeted at specific risk factor management (for example management of vascular risk factors and depression in people with comorbid vascular disease and depression) or focused on areas where people have difficulties, such as with functional ability or medicines management, are more likely to be effective. Given the importance of developing interventions for people with multimorbidity, the review provides interesting insights into the types of intervention components that are being examined. However, the majority of interventions in included studies had multiple components incorporating different elements, making comparison of intervention effects difficult. We categorised the intervention components using the EPOC taxonomy and identified the predominant intervention element for each study and then grouped studies depending on whether they had a predominantly organisational or patient focus. When examining the effectiveness of interventions by intervention type, we concluded that organisational type interventions, for example, the introduction of clinical nurse specialists to support treatment of depression or a focus on specific risk factor management in commonly co‐occurring conditions such as diabetes and hypertension may be more effective. Interventions that target areas where people have particular difficulties, such as functional ability, are also more likely to be effective. The current evidence suggests that organisational interventions that have a broader focus, such as case management or changes in care delivery for all individuals with multimorbidity, seem less effective. Patient‐oriented interventions that are not linked to healthcare delivery also seem less effective. Two of the three patient‐oriented interventions that were delivered by professionals showed improvements in a range of outcomes including reduced mortality (Garvey 2015; Gitlin 2009) following focused and intensive interventions targeting functional difficulty, activity participation and falls prevention.

We have presented results by outcomes pre‐specified in the protocol. In general these results were mixed and inconclusive, though there was a tendency for improvements in the studies that targeted common comorbid conditions that included depression. There was not a strong focus on clinical outcomes, particularly for the multimorbidity studies and this may reflect the challenge in research in multimorbidity when disease‐specific measures cannot be used.

There was limited effect on patient‐reported health outcomes such as HRQoL and on outcomes relating to health service utilisation and mixed effects on hospital admission rates and outcomes relating to medication use, and adherence. Five studies reported patient health behaviour outcomes with a tendency for these to be improved in the studies targeting comorbid conditions. There has been ongoing interest in the potential for improved patient self efficacy to lead to better self management and improved health outcomes. Self efficacy represents an outcome that is not disease or condition focused and was examined in many of the included studies. However, the majority of studies including this outcome showed no effect.

Costs were presented in six studies but only two studies conducted cost‐effectiveness analyses and it was not possible to compare outcomes across studies. The results relating to improved prescribing and risk factor management, in some of the comorbidity trials, indicate a potential for these interventions to reduce health service costs over longer periods of time.

Overall completeness and applicability of evidence

Most of the studies in this review are relatively recent reflecting the fact that this is a new area conceptually and that research to date has focused on description and impact rather than the evaluation of the effectives of interventions. The majority of newer studies included in this update and those studies identified as ongoing, focus on common comorbidities rather than on multimorbidity in general. In the original review (Smith 2012), only two of the ten included studies had interventions that focused on comorbid conditions whereas in this updated review, this has increased to nine of the 18 included studies. The tendency towards significant improvements in mental health outcomes in the comorbidity studies is likely related to the strong focus in these interventions on targeting the specific conditions involved, particularly depression. It is more challenging to design interventions for people with a broad range of conditions. The studies that seem more effective in the general multimorbidity group are those which had interventions targeted at specific areas of concern for participants, such as improving functional ability, which is not disease specific. One of the larger multimorbidity studies included involved a large well‐designed and executed RCT, the Guided Care study, which tested a broad organisational‐type intervention targeted at high risk individuals with multimorbidity, but which found no overall effect (Boult 2011). However, a pre‐planned sub‐group analysis indicated improvements in the use of some health services in the participants enrolled in one of the participating care plans (Kaiser‐Permanente, n = 365, 43% of full sample). Boult 2011 postulated that this result may have been related to the fact that care was already more organised and structured in this system, so that the Guided Care intervention may simply have extended the existing approaches used in that setting whereas its implementation was more challenging in less organised systems. However the results of sub‐group analysis, even when pre‐planned, need to be interpreted with caution given the relatively small samples sizes involved. Nonetheless, the differences in these sub‐groups highlight the importance of the healthcare delivery setting into which new interventions are added. Indeed, some of the patient‐oriented interventions seemed to run independently of people's healthcare delivery, particularly those that recruited participants from the community rather than through healthcare providers. Most of these studies had limited effectiveness, highlighting the importance of considering the overall recipient experience and integrating interventions into the healthcare system. The results of the patient‐oriented intervention studies are consistent with the Foster 2007 Cochrane review on lay‐led self‐management support programme, which concluded that there is no evidence that these interventions improve psychological health, symptoms or health‐related quality of life, or that they significantly alter healthcare use.

The evidence from this review partially addresses the review question, i.e. what interventions can effectively improve outcomes in people with multimorbidity. It suggests that interventions such as the addition of clinical care protocols need to be targeted at populations with defined combinations of common conditions such as diabetes and depression or heart disease; or need to focus on specific problems experienced by people in multimorbidity populations, for example a multidisciplinary team intervention that addresses functional difficulties. However, even when interventions are targeted they may not be effective for appropriate use of medications. The Haynes 2008 Cochrane review of Interventions for enhancing medication adherence concludes that "current methods of improving adherence for chronic health problems are mostly complex and not very effective" and suggests further research is needed. People with multimorbidity may have more specific problems with medicines use that relate to polypharmacy and managing complex drug treatment regimens, so medicines management interventions targeting these specific difficulties may be more effective.

Most of the multimorbidity studies in this review focused on older people; however, it is important to address the needs of younger individuals as there are issues relating to employability and absenteeism. Research in Scotland has highlighted that individuals in the poorest socioeconomic groups are more likely to develop multimorbidity at a younger age and more likely to die prematurely as a result (Barnett 2012). Acting upstream for younger people with multimorbidity is preventive and has potential to bring about significant quality of life benefits for individuals as well as cost savings for healthcare systems. However, even in ageing populations, multimorbidity worsens outcomes so there is still likely to be room for improvement, at least in ambulatory care patients.

The evidence to guide intervention development for individuals with multimorbidity is increasing and evolving rapidly. A number of ongoing studies have been identified and we anticipate that future updates of the review will improve the available evidence to inform policy makers and those planning services for individuals with multimorbidity.

Quality of the evidence

All of the included studies were randomised controlled trials. Overall they were of reasonable quality with minimal risk of bias, although blinding of participants and clinicians involved in the types of interventions included in this review is often impossible. Multimorbidity is a complex area because the characteristics of participants can vary depending on definitions used. This limits the potential to reasonably combine study results for meta‐analysis which is reflected in the high heterogeneity in the meta‐analyses undertaken for the review update, and potentially limits the internal validity of the results of the review.

Potential biases in the review process

The review was carried out in accordance with EPOC guidelines and using the updated Cochrane Handbook for Systematic Reviews of Interventions (Cochrane Handbook). Potential limitations in the search process relate to the lack of a MeSH term for multimorbidity. This meant that we had to use broad search terms which led to a high yield of citations to be searched. However, the authors are active researchers in the field of multimorbidity and are unaware of any potentially eligible studies that were missed by the search. We were also unable to retrieve some missing data from authors. However, as limited meta‐analyses were undertaken this did not lead to any appreciable measurement bias.

In addition, it must be noted that when we address complex interventions in primary care, there is always a context in which those interventions take place. A systematic review does not address the context that could have influenced the results in individual studies as there was limited reporting of external validity or generalisability in individual trials. The usual limitations relating to publication bias apply but we have searched the grey literature and contacted experts in the field to try and identify published and ongoing trials in this area.

Agreements and disagreements with other studies or reviews

We were unable to group sufficient numbers of studies with similar interventions in order to comment on which elements of interventions (e.g. the use of community pharmacists) seemed most effective and compare our review to other reviews of these interventions. The most consistent intervention element across all included studies was the use of case managers, but even these varied in that some were clinical case managers and others were administrative managers. The Cochrane review of the effect of case management on health care outcomes is ongoing but does plan to address differences in effectiveness between different types of case management (Zwarenstein 2000). Systematic reviews of community‐based case management in general have indicated mixed effects with improvements in client and professional satisfaction with care and reductions in caregiver strain but no impact on healthcare utilisation (Challis 2014).

Figure 1

Study flow diagram.

Figure 2

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

Figure 3

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

Figure 4

Forest plot of comparison: 1 Glycaemic control (HbA1c) Diabetes outcome: 1.1 HbA1c.

Figure 5

Forest plot of comparison: 2 Systolic Blood Pressure: outcome: 2.1 Systolic blood pressure.

Figure 6

Forest plot of comparison: 3 Depression scores: 3.1 PHQ9 Depression scores.

Figure 7

Forest plot of comparison: 4 Depression scores: 4.1 Depression scores.

Figure 8

Forest plot of comparison: 5 Health related quality of life, outcome: 5.1 HRQoL.

Figure 9

Forest plot of comparison: 6 Self‐Efficacy, outcome: 6.1 Self‐efficacy score.

Analysis 1.1

Comparison 1 Glycaemic control (HbA1c) studies targeting diabetes, Outcome 1 HBA1c.

Analysis 2.1

Comparison 2 Systolic Blood Pressure: studies targeting hypertension, Outcome 1 Systolic blood pressure.

Analysis 3.1

Comparison 3 PHQ9 depression scores: studies targeting depression, Outcome 1 PHQ9 Depression scores.

Analysis 4.1

Comparison 4 Depression scores: studies targeting depression, Outcome 1 Depression scores.

Analysis 5.1

Comparison 5 Health related quality of life, Outcome 1 HRQoL.

Analysis 6.1

Comparison 6 Self‐Efficacy, Outcome 1 Self‐efficacy score.

Interventions aimed at improving outcomes for people with multimorbidity compared with usual care
Participant or population: Adults with multimorbidity (two or more chronic conditions) Settings: Primary care and community settings Intervention: Any intervention designed to improve outcomes for people with multimorbidity including professional‐, organisational‐ and patient‐oriented interventions Comparison: Usual care
Outcomes	Impacts	Number of studies	Quality of the evidence (GRADE)
Clinical outcomes	There is no clear effect on clinical outcomes with a range of standardised effect sizes from 0.01 to 1.6 with a minority having effect sizes > 0.5; interventions aimed at improving management of risk factors in comorbid conditions were more likely to have higher effect sizes.	11	⊕⊕⊕⊖ Moderate
Mental health outcomes	There are improved depression‐related outcomes in studies targeting comorbid conditions that include depression with a range of standardised effect sizes from 0.09 to 2.24 with 4 of 7 studies having moderate to large effect sizes (> 0.5) . Standardised mean difference of −0.41 (95% CI, −0.63 to −0.20) was calculated from combining data from 6 studies.	9	⊕⊕⊕⊕ High
Patient‐reported outcome measures (PROMs)	There are mixed effects on PROMs with only half of studies that reported these outcomes showing any benefit with a range of standardised effect sizes from 0.03 to 1.7. Only 1 of 5 studies with available data on self‐efficacy had a moderate effect size, 4 of 7 had a moderate effect size for HRQoL (> 0.5) and effect sizes for other psychosocial outcomes were generally low.	12	⊕⊕⊕⊖ Moderate
Health Service Utilisation	There were no effects on health service utilisation and changes in visits were difficult to interpret as some interventions could lead to higher numbers of visits if previous unmet need was being addressed. There was no difference in admission‐related outcomes, though numbers of admissions in most of these studies were very small.	5	⊕⊕⊖⊖ Low
Medication use and adherence	There are mixed effects on medication use and adherence with half the studies reporting this outcome showing benefit. Proportions adherent to medication were higher in intervention participants with ranges in absolute difference of 10% to 40% but all studies with available data had small effect sizes.	4	⊕⊕⊖⊖ Low
Health‐related patient behaviours	Studies measuring this outcome reported a range of effects varying from an additional 18 minutes spent walking per week to an absolute difference in kcals expenditure per week of 2516 (no studies presented data that could be used to calculate effect sizes).	7	⊕⊕⊕⊖ Moderate
Provider behaviour	The majority of studies reporting provider behaviour indicated improved provider behaviour relating to care delivery; three studies reported a range of 15% to 40% in proportions of intervention providers improving behaviours such as appropriate referral.	5	⊕⊕⊕⊖ Moderate
GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.
We downgraded the evidence for effects on clinical and psychosocial outcomes to moderate due to lack of consistency of effect across studies and small effect sizes. We downgraded the evidence for effects on provider behaviour to moderate due to limited available data for calculation of standardised effect sizes (SES) and lack of clarity regarding the clinical importance of the results. We downgraded the evidence for effects on health service utilisation and medication use and adherence to low due to variation across studies and small effect sizes.

Table 1. Multimorbidity intervention components

Author Year	Professional	Participant	Organisational			Effect of intervention on primary outcome
			Case management or coordination of care	Reorganisation of care/team working	New team member
Predominantly organisational
Barley 2014	Nurse training	Participant information Prioritisation to create goals and health plan	Case manager provided personalised care	Regular planned participant visits Weekly team meetings	Nurse case manager	Pilot study and primary outcome was feasibility and deemed successful
Bogner 2008		Individualised programme	Case manager	Regular planned participant visits		Improved blood pressure control and depression scores
Boult 2011	Nurse training	Individual management plans Support for self‐management	Guided care nurses coordinated care	Guided care 'pods' consisting of nurse and PCP Monthly monitoring of participants		No impact on healthcare utilisation
Coventry 2015	Practice team training	Personalised goals and participant workbooks		Collaborative care using stepped care protocols Joint consultation between participant, psychologist and practice nurse	Psychologist Supervision and input from team psychiatrist	Modest reduction in depression scores
Hogg 2009		Individualised care plans		Multidisciplinary team‐based management with home based assessment Medication review	Pharmacist	Modest improvements in quality of chronic care delivery
Katon 2010		Individualised management plans and targets Support for self‐management		Team‐based care Stepped care treatment protocols Weekly team meeting	Psychologist and psychiatrist supported depression care	Improvements in composite outcome of glycaemic control, blood pressure, lipids and depression scores
Kennedy 2013	Practice training	Support for self‐management Participant guidebooks		Systems‐based approach to self‐management support with practice supports and links made with related local services		No intervention effect noted
Krska 2001		Individualised pharmaceutical care plans		Practice team‐implemented care plans	Pharmacist undertook medication review and devised pharmaceutical care plans	Reduction in pharmaceutical care issues
Martin 2013	Training for community psychologists	Cognitive behavioural therapy sessions		Psychological care programme designed for headache and depression	Community psychologists	Reduced headaches and improved depression scores
Morgan 2013	Practice nurse training	Support for self‐management Goal setting Individualised care plans	Nurse case manager	Quarterly reviews with practice nurse with GP stepping up care as needed		Improved depression scores
Sommers 2000		Risk reduction plan		Team based care with home assessment followed by team discussion, treatment plan and targets	Social worker	Reduced hospitalisation
Wakefield 2012		Participation in home telehealth monitoring	Nurse case manager using telehealth monitoring and treatment algorithms			Improved blood pressure, no effect on glycaemic control
Predominantly Patient‐oriented interventions
Eakin 2007		Support for self‐management with focus on diet and physical activity		Regular visits and follow‐up telephone calls	Health educator	Improvements in diet but not in physical activity
Garvey 2015	Occupational therapist (OT) training	OT‐led, group‐based support for self‐management programme (6 weeks) Goal setting and peer support		GP and primary care team referral	OT with input from physiotherapist and pharmacist	Improvements in activity participation
Gitlin 2006		Home‐based programme targeting functional difficulties with individualised plans and focus on falls prevention		Home visits and regular follow‐up calls	Occupational therapist and physiotherapist	Improvements in function (reduced mortality at 4 year follow‐up)
Hochhalter 2010	Training for coaches running intervention	Patient Engagement workshop (x1)		Two follow‐up phone calls	Coach who delivered workshop	No effect on outcomes
Lorig 1999	Training for volunteer lay group leaders	Chronic Disease Self Management Support Programme (six sessions) Peer support			Volunteer lay group leaders supported by study team	No primary outcome specified. Multiple outcomes reported with mixed effects
Lynch 2014		Diabetes self management support groups (18 sessions) Peer support Goal setting and behaviour skills training			Dietician led groups	No effect on primary outcome of weight reduction
The predominant intervention component is highlighted in bold text for each study No study contained a financial‐type intervention element

Table 1. Multimorbidity intervention components

Table 2. Costs

Study	Study type	Outcome	Result	Notes
Barley	RCT	Cost‐effectiveness	The intervention demonstrated marginal cost effectiveness up to a QALY threshold of GBP 3035
Boult	RCT	Total healthcare cost	Saving of USD 75,000 per GCN and USD 1364 per participant	USD in 2007 Initial result only ns
Katon	RCT	Cost‐effectiveness	Mean reduction of 114 days in depression free days and an estimated difference of 0.335 QALYs (95% CI −0.18 to 0.85). The intervention was associated with lower OPD costs with a reduction of USD 594 per participant (95% CI USD −3241 to USD 2053).	Non‐significant but 99.7% probability that the intervention met the threshold of < USD 20,000 per QALY
Krska	RCT	Mean cost of medicines	Int: 38.83 Con: 42.61 Absol diff 3.78 Rel %diff 9%	GBP in 2000 ns SES = 0.13
Lorig	RCT	Intervention cost per completed participant	USD 70	USD in 1998 See text for assumptions made
Lorig	RCT	Cost savings per individual	USD 750	USD in 1998 See text for assumptions made
Sommers	RCT	Savings per individual	USD 90	USD in 1994 See text for assumptions made
* refers to whether original study reported statistically significant improvement in this outcome

Table 2. Costs

Table 3. Overview of outcomes

Outcome category	Outcome	No. studies with this outcome	No. studies with p< 0.05 for this outcome
Physical Health	Hb1Ac	5	2
	BP	6	2
	Cholesterol	2	1
	Other symptom score	4	1
	Mortality	1	1
Mental Health	Depression scores	8	6
	% improved depression	1	1
	Anxiety scores	4	3
	Cognitive symptom management	1	0
Psychosocial	QoL/general health	10	4
	Functional impairment & disability	6	2
	Social (activity/support)	4	1
	Self efficacy	7	3
	Home hazards	1	0
Health service use	Visits/use service	5	0
	Hospital admission related	6	2
Patient health related behaviours	Exercise/diet	6	2
Medication adherence		5	2
Provider behaviour	Prescribing	3	2
	Disease management	3	3
Costs	Direct costs	5	Not applicable
* Multimorbidity is defined as two or more independent conditions within the same individual whereas comorbidity refers to linked conditions. In this review comorbidity studies included depression and diabetes or depression and hypertension ** The scales or measurements used in each study for the outcomes are described in the Table of included studies

Table 3. Overview of outcomes

Table 4. Clinical Outcomes

Study	Study type	Outcomes	Results	Notes
Barley	RCT	% with angina (Rose Angina score)	Int 22/31 Con 30/37 Absol diff 8, Rel % diff 27%	ns
Bognor	RCT	Systolic BP	Int 127.3 (SD 17.7) Con 141.3 (SD 18.8) Absol diff 14, Rel % diff 10%	* SES = 1.12
Bognor	RCT	Diastolic BP	Int 83 (SD 10.7) Con 81.4 (SD 11.1) Absol diff 9.2, Rel % diff 11%	* SES = 0.8
Gitlin	RCT	Mortality	Int 9/160 (0.06) Con: 21/159 (0.13) Absol diff 7, Rel % diff 54%	*
Hogg	RCT	Systolic BP	Int 124.3 Con 124.2 Absol diff 0.1, Rel % diff < 0.1%	ns (No SDs reported)
Hogg	RCT	HbA1c	Int 7.01 Con 6.78 Absol diff 0.23, Rel % diff 3%	ns
Katon	RCT	Systolic BP	Int 131 (SD 18.4) Con 132.3 (SD 17.2) Absol diff 1.3, Rel % diff 1%	* SES = 0.07
Katon	RCT	HbA1c	Int 8.14 (SD 2.03) Con 8.04 (SD 1.87) Absol diff 0.1, Rel % diff 13%	* SES = 0.32
Katon	RCT	Cholesterol	Int 91.9 (SD 36.7) Con 101.4 (SD 36.6) Absol diff 9.5, Rel % diff 9%	* SES = 0.26
Katon	RCT	Composite: all three risk factors (BP, HbA1c and cholesterol) below guidelines	Int 36/97 (0.37) Con: 19/87 (0.22) Absol diff 15, Rel % diff 68%	*
Lorig	RCT	Pain/ physical discomfort	Int 59.8 (SD 20.1) Con 60.6 (SD 17.1) Absol diff 0.8, Rel % diff 1%	SES = 0.04 ns
Lorig	RCT	Energy/fatigue	Int 2.18 (SD 0.73) Con 2.02 (SD 0.75) Absol diff 0.16, Rel % diff 8%	ns
Lorig	RCT	Shortness of breath	Int 1.34 (SD 0.91) Con 1.58 (SD 0.83) Absol diff 0.24, Rel % diff 15%	ns
Lynch	RCT	HbA1C	Int 7.9 (SD 1.6) Con 7.4 (SD 1.6) Absol diff 0.5, Rel % diff 6.7%	ns SES = 0.31
Lynch	RCT	% with at least 0.5 absolute reduction in HbA1c	Int 15/30 (0.05) 7/31 Con (0.21) Absol diff 29, Rel % diff 138%	*
Lynch	RCT	Mean SBP	Int 135.8 (SD 21.4) Con 136.7 (SD 23) Absol diff 0.9, Rel % diff 0.6%	ns SES = 0.01
Morgan	RCT	HbA1C	Int 6.9 (SD 0.26) Con 7.4 (SD 0.36) Absol diff 0.5, Rel % diff 6.7%	* SES = 1.6
Morgan	RCT	Systolic BP	Int 134.2 (SD 3.0) Con 133.5 (SD 3.8) Absol diff 0.7, Rel % diff 0.5%	ns SES = 0.2
Morgan	RCT	Cholesterol	Int 4.22 (SD 0.14) Con 4.44 (SD 0.2) Absol diff 0.22, Rel % diff 5%	ns SES = 0.22
Morgan	RCT	Mean BMI	Int 31.2 (SD 1.0) Con 31.0 (SD 1.0) Absol diff 0.2, Rel % diff 0.6%	ns SES = 0.2
Sommers	RCT	Symptom scores	Int 17.2 Con 18.9 Absol diff 1.7, Rel % diff 9%	ns
Wakefield	RCT	HbA1c	Int 6.9 (1.1) Con 6.95 (1.1) Absol diff 0.05, Rel % diff 0.7%	ns SES = 0.05
Wakefield	RCT	Systolic BP	Int 133 (16.6) Con 137 (17.3) Absol diff 4, Rel % diff 3%	ns SES = 0.24
Martin	RCT	Mean headache rating	Int 0.63 (SD 0.5) Con 1.01 (SD 0.83) Absol diff 0.38, Rel % diff 38%	* SES = 0.58
* refers to whether original study reported statistically significant improvement in this outcome ** Total number with final data collected was 384. No final numbers of intervention and control participants presented.

Table 4. Clinical Outcomes

Table 5. Mental Health Outcomes

Study	Study type	Outcome	Result	Notes
Barley	RCT	PHQ9 depression score	Int 12.6 (SD 7.1) Con 12 (SD 6.9) Absol diff 0.6, Rel % diff 8%	ns SES = 0.09
Barley	RCT	HADS depression score	Int 9.5 (SD 4.6) Con 8.8 (SD 4.8) Absol diff 0.7, Rel % diff 8%	ns SES = 0.15
Barley	RCT	HADS anxiety score	Int 9.9 (SD 7.1) Con 9.5 (SD 5.4) Absol diff 0.4, Rel % diff 4%	ns SES = 0.08
Bognor	RCT	CES depression score	Int 9.9 (SD 10.7) Con 19.3 (SD 15.2) Absol diff 9.4, Rel % diff 49%	* SES = 0.75
Coventry	RCT	SCL‐D13 depression score	Int 1.76 (SD 0.9) Con 2.02 (SD 0.9) Absol diff 2.6, Rel % diff 13%	* SES = 0.28
Coventry	RCT	PHQ9 depression score	Int 11.3 (SD 6.5) Con 13.1 (SD 6.5) Absol diff 1.8, Rel % diff 14%	* SES = 0.28
Coventry	RCT	GAD‐7 anxiety score	Int 8.2 (SD 5.8) Con 9.7 (SD 5.9) Absol diff 1.5, Rel % diff 15%	* SES = 0.26
Garvey	RCT	HADS total score	Int 15.6 (SD 8.3) Con 16.7 (SD 8.2) Absol diff 1.1, Rel % diff 6.5%	ns SES = 0.13
Katon	RCT	SCL 20 depression score	Int 0.83 (SD 0.66) Con 1.14 (SD 0.68) Absol diff 0.31, Rel % diff 27%	* SES = 0.46
Katon	RCT	Patient global improvement in depression	Int 41/92 Con 16/91 Absol diff 27, Rel % diff 150%	*
Lorig	RCT	Cognitive symptom management score	Int 1.75 Con 0.98 Absol diff 0.77, Rel % diff 79%	ns
Martin	RCT	PHQ9 depression score	Int 6.7 (SD 4.6) Con 12.6 (SD 5.3) Absol diff 5.9, Rel % diff 47%	* SES = 1.18
Martin	RCT	BDI ‐Depression score	Int 13.1 (SD 8.6) Con 28.7 (SD 9.5) Absol diff 15.6, Rel % diff 54%	* SES = 1.73
Martin	RCT	BAI Anxiety score	Int 10.5 (SD 10.8) Con 16.4 (SD 9.3) Absol diff 5.9, Rel % diff 36%	* SES = 0.1
Morgan	RCT	PHQ9 depression score	Int 7.1 (SD 0.8) Con 9.0 (SD 0.9) Absol diff 1.9, Rel % diff 21%	* SES = 2.24
Sommers	RCT	GDS score (depression)	Int 4.1 Con 4.1 Absol diff 0, Rel % diff 0%	ns
* refers to whether original study reported statistically significant improvement in this outcome

Table 5. Mental Health Outcomes

Table 6. Patient‐reported outcome measures

Study	Study type	Outcome	Result	Notes
Health Related Quality of Life
Barley	RCT	SF12 PCS	Int 32.4 (SD 10.7) Con 33.3 (SD 9.2) Absol diff 0.7, Rel % diff 2%	ns SES = 0.07
Barley	RCT	SF12 MCS	Int 34.5 (SD 11.6 ) Con 33.6 (SD 12.5 ) Absol diff 0.9 , Rel % diff 3%	ns SES = 0.08
Barley	RCT	HRQoL (WEMWBS)	Int 40.6 (SD 11.2) Con 39.6(SD 12.3) Absol diff 1, Rel % diff 2.5%	ns SES = 0.08
Coventry	RCT	HRQoL (WHOQOL)	Int 2.99 (SD 0.6) Con 2.91 (SD 0.6) Absol diff 0.08, Rel % diff 3%	* SES = 1.7
Garvey	RCT	HRQoL (EQ5D VAS)	Int 65.7 (SD 20.2) Con 50.5 (SD 16.3) Absol diff 15.2, Rel % diff 30%	* SES = 0.84
Hogg	RCT	SF 36 Mental Health	Int 52.4 Con 52.2 Absol diff 0.2, Rel % diff 0.3%	ns
Hogg	RCT	SF 36 Physical Health	Int 44.3 Con 41.5 Absol diff 2.8, Rel % diff 6.7%	ns
Katon	RCT	QoL score	Int 6.0 (SD 2.2) Con 5.2 (SD 1.9) Absol diff 0.8, Rel % diff 15%	* SES = 0.44
Kennedy	RCT	HRQoL (EQ5D)	Int 0.56 (SD 0.34) Con 0.57 (SD 0.32) Absol diff 0.01, Rel % diff 1%	ns SES = 0.03
Lorig	RCT	Psychological well‐being	Int 3.47 Con 3.33 Absol diff 0.04, Rel % diff 4%	ns SES = 0.21
Martin	RCT	HRQol (AQOL)	Int 26.3 (SD 4.76) Con 28.4 (SD 4.97) Absol diff 2.1, Rel % diff 7 %	* SES = 0.4
Sommers	RCT	SF36 score	Int 2.2 Con 3.3 Absol diff 1.1, Rel % diff 33%	ns
Self‐efficacy
Barley	RCT	Self‐efficacy score	Int 28.6 (SD 6.7) Con 27.9 (SD 8.1) Absol diff 0.11, Rel % diff 2.5%	ns SES = 0.09
Coventry	RCT	Self‐efficacy score	Int 5.72 (SD 1.9) Con 5.53 (SD 1.9) Absol diff 0.18, Rel % diff 3.2%	ns * SES = 0.09
Garvey	RCT	Self efficacy score	Int 6.8 (SD 1.5) Con 5.3 (SD 1.9) Absol diff 1.47, Rel % diff 28%	* SES = 0.86
Hochhalter	RCT	Self‐efficacy	Int 7.4 Con 8.0 Absol diff 0.6, Rel % diff 7.5%	ns
Kennedy	RCT	Self‐efficacy	Int 68 (SD 23.4) Con 68.7 (SD 23.1) Absol diff 0.7, Rel % diff 1%	ns SES = 0.03
Wakefield	RCT	Self‐efficacy	Int 8.1 (SD 1.9) Con 8.3 (SD 1.9) Absol diff 0.2, Rel % diff 2.4%	ns SES = 0.11
Daily function and disability
Coventry	RCT	Sheehan Disability Score	Int 5.73 (SD 2.8) Con 5.83 (SD 2.8) Absol diff 0.1, Rel % diff 2%	* SES = 0.04
Garvey	RCT	Frenchay Activities Index	Int 21.3 (SD 7.9) Con 18.9 (SD 7.2) Absol diff 2.4, Rel % diff 13%	* SES = 0.32
Garvey	RCT	Activities daily living: NEADL (total)	Int 47.2 (SD 11.9) Con 40.7 (SD 10.7) Absol diff 6.5, Rel % diff 16%	* SES = 0.58
Hogg	RCT	IADL	Int 10.6 Con 10.9 Absol diff 0.3, Rel % diff 2.7%	ns
Lorig	RCT	Disability	Int 0.86 Con 0.96 Absol diff 0.1, Rel % diff 10%	ns
Lorig	RCT	Social role/activity limitation	Int 1.91, Con 1.98 Absol diff 0.07, Rel % diff 4%	ns
Illness perceptions
Coventry	RCT	Multimorbidity illness perception scale	Int 2.1 (SD 0.9) Con 2.28 (SD 0.9) Absol diff 0.18, Rel % diff 8%	ns SES = 0.2
Barley	RCT	Illness perceptions (BIPQ)	Int 40 (SD 14.8) Con 43(SD 31.1) Absol diff 3, Rel % diff 7%	ns SES = 0.22
Social support
Coventry	RCT	Social support (ESSI)	Int 3.29 (SD 1.1) Con 3.4 (SD 1.0) Absol diff 0.11, Rel % diff 3%	ns SES = 0.11
Eakin	RCT	Multilevel support for healthy lifestyle	Int 2.7 Con 2.59 Absol diff 0.11, Rel % diff 4%	ns
Other PROMs
Barley	RCT	Patient‐reported needs (PSYCHLOPS)	Int 13.6 (SD 5.1) Con 13.4 (SD 5.4) Absol diff 0.2, Rel % diff 1.5%	ns SES = 0.04
Hochhalter	RCT	Total unhealthy days	Int 15.3 Con 14.1 Absol diff 1.2, Rel % diff 9%	ns
Hogg	RCT	Total unhealthy days	Int 7.6 Con 9.9 Absol diff 2.3, Rel % diff 23%	ns
Kennedy	RCT	Shared decision making (HCCQ)	Int 67.7 (SD 28) Con 69.3 (SD 26.1) Absol diff 1.6, Rel % diff 2%	ns SES = 0.06
Lorig	RCT	Self‐rated health	Int 3.42 Con 3.44 Absol diff 0.02, Rel % diff 0.6%	ns
Lorig	RCT	Health distress	Int 1.97 Con: 2.13 Absol diff 0.16, Rel % diff 7.5%	ns SES = 0.16
Sommers	RCT	Social activities count	Int 8.7 Con:8.6 Absol diff 0.1, Rel % diff 1%	* (when adjusted for baseline diff)
Sommers	RCT	HAQ score	Int 0.44 Con 0.5 Absol diff 0.06, Rel % diff 12%	ns
* refers to whether original study reported statistically significant improvement in this outcome

Table 6. Patient‐reported outcome measures

Table 7. Health service use

Study	Study type	Outcome	Result	Notes
Boult	RCT	No. hospital admissions	Int 0.7 Con 0.72 Absol diff 0.02, Rel % diff 3%	ns
Boult	RCT	No. days in hospital	Int 4.26 Con 4.49 Absol diff 0.23, Rel % diff 5%	ns
Boult	RCT	No. ED visits	Int 0.44 Con 0.44 Absol diff 0, Rel % diff 0	ns
Boult	RCT	No. PC visits	Int 9.98 Con 9.88 Absol diff 0.1, Rel % diff 1%	ns
Boult	RCT	No. specialist visits	Int 9.04 Con 8.49 Absol diff 0.55, Rel % diff 6%	ns
Boult	RCT	No. home healthcare episodes	Int 0.99 Con 1.3 Absol diff 0.31, Rel % diff 24%	*
Hogg	RCT	No. hospital admissions	Int 0.4 Con 0.46 Absol diff 0.06, Rel % diff 13%	ns
Hogg	RCT	Proportion hospitalised	Int 0.26, Con 0.26 Absol diff 0, Rel % diff 0%	ns
Hogg	RCT	No. ED visits	Int 0.63 Con 0.73 Absol diff 0.01, Rel % diff 14%	ns
Hogg	RCT	Proportion with ED visit	Int 0.38 Con 0.42 Absol diff 0.04, Rel % diff 9%	ns
Katon	RCT	Proportion hospitalised	Int 0.26 Con 0.22 Absol diff 0.04, Rel % diff 18%	ns
Lorig	RCT	No. doctor and ED visits	Int 6.51 Con 7.08 Absol diff 0.57, Rel % diff 8%	ns
Lorig	RCT	No. hospital stays in past 6 months	Int 0.26 Con 0.31 Absol diff 0.05, Rel % diff 6%	*
Lorig	RCT	No. nights in hospital in last 6 months	Int 1.3 Con 1 Absol diff 0.3 Rel % diff 30%	*
Sommers	RCT	No. hospital admissions per individual per year	Int 0.36 Con 0.52 Absol diff 0.16, Rel % diff 31%	*
Sommers	RCT	≥1 60 day readmission	Int 3.6 Con 9.4 Absol diff 5.8, Rel % diff 62%	*
Sommers	RCT	≥ 1 hospital admission	Int 8.8 Con 7.7 Absol diff 1.1, Rel % diff 14%	*
Sommers	RCT	No. PCP visits	Int 6.0 Con 6.1 Absol diff 0.1, Rel % diff 2%	ns
Sommers	RCT	No. office visits	Int 11 Con 12.5 Absol diff 1.5, Rel % diff 12%	*
Sommers	RCT	≥ 1 home care visit	Int 19.5 Con 18.8 Absol diff 0.7, Rel % diff 4%	ns
Sommers	RCT	No. medical specialist visits	Int 1.4 Con 1.7 Absol diff 0.3, Rel % diff 18%	ns
Sommers	RCT	No. other visits	Int 3.9 Con 4.3 Absol diff 0.4, Rel % diff 9%	*
Sommers	RCT	≥ 1 ED visit	Int 21.4 Con 16.7 Absol diff 4.7, Rel % diff	ns
* refers to whether original study reported statistically significant improvement in this outcome

Table 7. Health service use

Table 8. Medication use and adherence and prescribing

Study	Study type	Outcome	Results	Notes
Bognor	RCT	≥80% adherence to antidepressant medication (MEMS caps)	Int 23/32 Con 10/32 Absol diff 0.41, Rel % diff 132%	*
Bognor	RCT	≥80% adherence to antihypertensive medication (MEMS caps)	Int 25/32 Con 10/32 Absol diff 0.47, Rel % diff 152%	*
Martin	RCT	Mean daily medication use	Int 2.4 (SD 3.2) Con 3.0 (SD 2.8) Absol diff 0.6, Rel % diff 20%	ns SES = 0.2
Morgan	RCT	% taking antidepressant medication	Int 34/62 Con 36/113 Absol diff 0.11, Rel % diff 34%	*
Wakefield	RCT	Adherence (Edward's scale)	Int 3.4 (SD 0.5) Con 3.3 (SD 0.5) Absol diff 0.1, Rel % diff 3%	ns SES = 0.2
Wakefield	RCT	Medication Taking Adherence Score	Int 100 (SD 1.4) Con 98.9 (SD 6.0) Absol diff 1.1, Rel % diff 1%	ns SES = 0.28
* refers to whether original study reported statistically significant improvement in this outcome

Table 8. Medication use and adherence and prescribing

Table 9. Health‐related participant behaviours

Study	Study type	Outcome	Results	Notes
Hochhalter	RCT	PAM (patient activation measure)	Int 66.8 Con 66.2 Absol diff 0.6, Rel % diff 1%	ns
Eakin	RCT	Diet behaviour scores	Int 2.2 Con 2.41 Absol diff 0.21, Rel % diff 9%	*
Eakin	RCT	Change minutes of walking/week	Int +8 Con −10 Absol diff 18, Rel % diff 180%	*
Katon	RCT	General adherence to diet score	Int 0.86 Con 0.81 Absol diff 0.05, Rel % diff 6%	ns
Katon	RCT	General adherence to exercise score	Int 0.54 Con 0.44 Absol diff 0.1, Rel % diff 23%	ns
Lorig	RCT	Exercise: stretching and strengthening (mins/week)	Int 53.1 Con 40.4 Absol diff 12.7,Rel % diff 31%	ns
Lorig	RCT	Exercise: aerobic (mins/week)	Int 101.8 Con 88 Absol diff 13.8, Rel % diff 157%	ns
Lorig	RCT	Communication with doctor (score 1‐5)	Int 3.34 Con 3.2 Absol diff 0.14, Rel % diff 4%	ns
Lynch	RCT	Physical activity (kcal expenditure per week, CHAMPS)	Int 1913.6 Con −603 Absol diff 2516, Rel % diff 417%	*
Morgan	RCT	Smoking	Int 13/162 Con 13/110 Absol diff 0.04, Rel % diff 33%	ns
Morgan	RCT	Alcohol	Int 51/104 Con 27/42 Absol diff 0.15, Rel % diff 23%	ns
Morgan	RCT	Exercise (30 minutes/day for 5 days/ week)	Int 97/162 Con 22/75 Absol diff 0.31, Rel % diff 106%	*
Sommers	RCT	Nutrition checklist score	Int 2.0 Con1.9 Absol diff 0.1, Rel % diff 5%	ns
* refers to whether original study reported statistically significant improvement in this outcome

Table 9. Health‐related participant behaviours

Table 10. Provider behaviour

Study	Study type	Outcome	Result	Notes
Boult	RCT	PACIC score (patient measure of quality of care received)	Int 3.14 Con 2.85 Absol diff 0.29, Rel % diff 10%	*
Coventry	RCT	PACIC score	Int 2.37 (SD 1.1) Con 1.98 (SD 1.0) Absol diff 0.39, Rel % diff 20%	ns SES = 0.39
Hogg	RCT	Chronic Disease Mangement Score	Int 0.84 Con 0.77 Absol diff 0.07, Rel % diff 9%	*
Hogg	RCT	Preventive Care Score	Int 0.89 Con 0.7 Absol diff 0.19, Rel % diff 27%	*
Krska	RCT	% Pharmaceutical care issues resolved from baseline	Int 950/1206 Con 542/1380 Absol diff 0.4, Rel % diff 102%	*
Morgan	RCT	% Referred to mental health	Int 58/162 Con 10/111 Absol diff 0.27, Rel % diff 300%	*
Morgan	RCT	% Referred to exercise programme	Int 58/162 Con 24/114 Absol diff 0.15, Rel % diff 71%	*
* refers to whether original study reported statistically significant improvement in this outcome

Table 10. Provider behaviour

Comparison 1. Glycaemic control (HbA1c) studies targeting diabetes

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 HBA1c Show forest plot	3	561	Mean Difference (IV, Random, 95% CI)	0.02 [‐0.21, 0.25]

Comparison 1. Glycaemic control (HbA1c) studies targeting diabetes

Comparison 2. Systolic Blood Pressure: studies targeting hypertension

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Systolic blood pressure Show forest plot	5	892	Mean Difference (IV, Random, 95% CI)	‐3.10 [‐7.26, 1.06]

Comparison 2. Systolic Blood Pressure: studies targeting hypertension

Comparison 3. PHQ9 depression scores: studies targeting depression

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 PHQ9 Depression scores Show forest plot	4		Mean Difference (IV, Random, 95% CI)	Totals not selected

Comparison 3. PHQ9 depression scores: studies targeting depression

Comparison 4. Depression scores: studies targeting depression

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Depression scores Show forest plot	6	1062	Std. Mean Difference (IV, Random, 95% CI)	‐0.41 [‐0.63, ‐0.20]

Comparison 4. Depression scores: studies targeting depression

Comparison 5. Health related quality of life

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 HRQoL Show forest plot	6		Std. Mean Difference (IV, Random, 95% CI)	Totals not selected

Comparison 5. Health related quality of life

Comparison 6. Self‐Efficacy

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Self‐efficacy score Show forest plot	5	3639	Std. Mean Difference (IV, Random, 95% CI)	0.05 [‐0.12, 0.22]

1.1 Studies targeting self‐efficacy	5	3639	Std. Mean Difference (IV, Random, 95% CI)	0.05 [‐0.12, 0.22]

Comparison 6. Self‐Efficacy