The impact of care farms on quality of life, depression and anxiety among different population groups: A systematic review

Abstract Care farming (also called social farming) is the therapeutic use of agricultural and farming practices. Service users and communities supported through care farming include people with learning disabilities, mental and physical health problems, substance misuse, adult offenders, disaffected youth, socially isolated older people and the long term unemployed. Care farming is growing in popularity, especially around Europe. This review aimed to understand the impact of care farming on quality of life, depression and anxiety, on a range of service user groups. It also aimed to explore and explain the way in which care farming might work for different groups. By reviewing interview studies we found that people valued, among other things, being in contact with each other, and feeling a sense of achievement, fulfilment and belonging. Some groups seemed to appreciate different things indicating that different groups may benefit in different ways but, it is unclear if this is due to a difference in the types of activities or the way in which people take different things from the same activity. We found no evidence that care farms improved people's quality of life and some evidence that they might improve depression and anxiety. Larger studies involving single service user groups and fully validated outcome measures are needed to prove more conclusive evidence about the benefits of care farming.

This review aims to understand the impact of care farming on quality of life, depression and anxiety, on a range of service user groups. It also aims to explore and explain the way in which care farming might work for different groups.
What is the aim of this review?
This Campbell systematic review examines the impact of care farming on quality of life, depression and anxiety, on a range of service user groups. It also aims to explore and explain the way in which care farming might work for different groups.

| What studies are included?
The review included randomised controlled trials (RCTs) and quasi-RCTs; interrupted time series and nonrandomised controlled observational studies; uncontrolled before and after studies and qualitative studies. Study participants were those who typically receive support at a CF. Studies conducted in a setting that met the accepted definition of a CF were included, but farming interventions that were carried out in a hospital or prison setting were excluded.
The total number of included studies in this review are 18 qualitative studies and 13 quantitative studies, one of which was a mixed-methods study.

| What are the findings of this review?
The qualitative interview studies showed that people valued, among other things, being in contact with each other, and feeling a sense of achievement, fulfilment, and belonging. Some groups seemed to appreciate different things, indicating that different groups may benefit in different ways but, it is unclear if this is due to a difference in the types of activities or the way in which people value different things from the same activity.
There is a lack of quantitative evidence that CFs improve people's quality of life, but some evidence that they might improve depression and anxiety.
Larger studies involving single service user groups and fully validated outcome measures are needed to prove more conclusive evidence about the benefits of care farming.

| What do the findings of the review mean?
There is a lack of evidence to determine whether or not care farming is effective in improving quality of life, depression and anxiety. More evidence is available for those with mental ill-health, but firm conclusions cannot be drawn.
Despite the current lack of robust evidence to support the effectiveness of care farming, there are strong arguments to support a more integrated approach to care farming as a viable alternative or adjunct to mainstream approaches for mental health problems. Lack of choice, gender inequalities, and overburdened statutory services indicate the need for a credible alternative treatment option.
There needs to be a concerted effort to increase awareness among commissioners of health care, frontline service providers and potential service users about care farming, how-and for whom-it might work.
Models across Europe that offer a more integrated approach between green care and statutory services could provide valuable learning.
The evidence for care farming for other service user groups is not as well developed as it is for those with mental health problems, but that is not to say there is not a need. Disaffected youth, adult offenders and people with dementia represent significantly large vulnerable population groups where current service provision struggles to meet demand.
The need to continue to improve and provide high quality research in these areas is therefore pressing.

| How up-to-date is this review?
The review authors searched for studies published up to July 2017.

| Objectives
The primary objective was to systematically review the available evidence of the effects of CFs on quality of life, health and social well-being on service users. Within this, we aimed to explore the size of the effect that CFs have on the health, well-being and social outcomes of different population groups. With available material we also aimed to explore the relationship between contextual data (the activities and characteristics of the farm and the nature of the service user groups) and the impact on outcomes. Finally, we aimed to understand the mechanisms of change for different population groups with a view to constructing a logic model to describe the ways in which care farming might work.

| Selection criteria
We included a broad range of study designs: RCTs and quasi-RCTs; interrupted time series and nonrandomised controlled observational studies; uncontrolled before and after studies and qualitative studies.
We excluded single subject designs, reviews, overviews, surveys, commentaries and editorials. Study participants were those that typically receive support at a CF, including but not restricted to people with mental health problems, learning difficulties, health problems, substance misuse problems, and offenders and disaffected youth. Only those attending for a single day as a visitor were excluded. Studies conducted in a setting that met the accepted definition of a CF were included, but farming interventions that were carried out in a hospital or prison setting were excluded. For the purposes of developing the logic model, we retained papers that described any theories to explain how and for whom care farming might work. These papers are not formally included in the review.

| Data collection and analysis
Each screening stage involved two independent reviewers.
Studies that were potentially eligible after title and abstract screening underwent full paper screening. Disagreements were discussed and resolved by consensus at each stage. Papers describing theories in relation to care farming were separately retained even if they did not meet the inclusion criteria for the purposes of constructing a theoretical framework to inform the logic models. The Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) was used to state the process of study selection. We stored all references in Endnote (VX7) and recorded extracted data and the outcomes of full paper screening in EPPI-Reviewer 4 (V.4.5.0.1). The data extraction form was based on the CPHG Data Extraction and Assessment Template with subsections for contextual information, and qualitative and quantitative data. We used a sequential exploratory approach to the review involving four stages: (a) developing a theoretical framework; (b) identifying the intervention components, mechanisms of change, and proximal outcomes from existing theories and qualitative data; (c) mapping the mechanisms of change and proximal outcomes to the theories to develop the logic models and (d) testing the logic models against the quantitative data. We used an adapted version of the COREQ tool to assess the quality of the qualitative studies, and the EPOC and EPHPP tools to assess the risk of bias in quantitative studies.
No studies were excluded based on quality. The nature of the studies meant that we were unable to assess treatment effect and reporting biases.

| Results
In 2015, our search methods identified 1,659 articles, of which 14 qualitative studies, 13 quantitative studies and one mixed methods study met the inclusion criteria. In addition, we identified 15 theories that had been quoted in connection with care farming. The rerun of the search of publish literature in July 2017 identified a further 391 articles, of which three qualitative studies met the inclusion criteria. The total studies in this review are 18 qualitative studies and 13 quantitative studies, one of which was a mixed-methods study. We created four logical models to explain how care farming might work: an overall one for all service user groups; one for people with either mental health problems or substance misuse problems, one for disaffected youth and one for people with learning disabilities.
These models comprised five key theoretical concepts derived from identified theories (restorative effects of nature, being socially connected, personal growth, physical well-being and mental well-being), five categories of intervention components (being in a group, the farmer, the work, the animals and the setting) and 15 categories of mechanisms derived from included qualitative studies (achievement and satisfaction, belonging and nonjudgement, creating a new identity, distraction, feeling valued and respected, feeling safe, learning skills, meaningfulness, nurturing, physical well-being, reflection, social relationships, stimulation, structure, and understanding the self). In addition, from the theories and qualitative studies, we identified 12 MURRAY ET AL. | 3o f6 1 different outcomes, both proximal (secondary) and primary, that we expected to find when testing the logic models against the quantitative studies. One key theoretical concept "restorative effects of nature" was underrepresented in the intervention components and mechanisms reported within the qualitative studies. The types of mechanisms appeared to differ according to different service user groups, suggesting that care farming may work in different ways according to different needs. Across the 13 quantitative studies (including the mixed methods study), 24 different outcome measures were reported. Eight studies (both qualitative and quantitative) reported results for mixed client groups. Only the logic model for mental illness and substance misuse was tested, due to a lack of quantitative evaluations for the other service user groups. We found a lack of evidence to indicate that CFs improve quality of life, and limited evidence that they might improve depression and anxiety. There was some evidence to suggest that CFs can improve self-efficacy, selfesteem and mood, with inconsistent evidence of benefit for social outcomes. All of the studies had a high risk of bias so the results should be treated with caution.

| Authors' conclusions
There is a lack of evidence available to determine whether or not care farming is effective in improving quality of life, depression and anxiety. More evidence is available for those with mental illhealth, but firm conclusions cannot be drawn. Small study sizes of poor design, evaluations involving mixed service user groups, the use of multiple and sometimes unvalidated outcome measures, short follow-ups, and the absence of key outcomes that fit with theory have all hampered the development of a more robust evidence base. However, we now have a set of theory-based logic models that offer a framework for research evaluations. With recommendations in place to address the current research inadequacies there is an opportunity to vastly improve the evidence base for care farming.
Despite the current lack of robust evidence to support the effectiveness of care farming, there are strong arguments to support a more integrated approach to care farming as a viable alternative or adjunct to mainstream approaches for mental health problems. Lack of choice, gender inequalities and overburdened statutory services indicate the need for a credible alternative treatment option. A concerted effort to increase awareness among commissioners of health care, frontline service providers, and potential service users about care farming, how, and for whom, it might work is needed. Models across Europe that offer a more integrated approach between green care and statutory services could provide valuable learning. The evidence f o rc a r ef a r m i n gf o ro t h e rs e r v i c eu s e rg r o u p si sn o ta sw e l l developed as it is for those with mental health problems, but that is not to say there is not a need. Disaffected youth, adult offenders and people with dementia represent significantly large vulnerable population groups where current service provisions struggles to meet demand. The need to continue to improve and provide high quality research in these areas is, therefore, pressing.
3 | BACKGROUND 3.1 | The problem, condition or issue Supporting individuals whose vulnerabilities put them at greater risk of poorer quality of life is a cornerstone of many charitable/third sector organisations. Often the support needed goes beyond that which can be provided by statutory health and social care organisations. This is partly a capacity issue relating to increasing life expectancies over the 20th century ) and increasing prevalence of long-term conditions. However, it also relates to changing needs and demands of populations within modern societies. Many of the problems presenting to health service providers are complex and are often underpinned or exacerbated by social problems such as poor education, poor housing, unemployment and social isolation, and the skills within health services to address these issues do not exist within this sector (Citizen's Advice, 2015;Popay, Kowarzik, Mallinson, Mackian, & Barker, 2007). Thus, for many such individuals inadequate support can lead to poorer quality of life, and for society as a whole, greater health inequalities (Marmot et al., 2010).
Learning disabilities is an umbrella term for a range of conditions, including Down's syndrome, fragile X syndrome and cerebral palsy.
Autism spectrum disorder (ASD) can also be included here, but not all people with ASD have a learning disability. People with learning disabilities 1 often experience poorer health and higher mortality due to increased social and health inequalities and underlying preexisting conditions (Krahn, Hammond, & Turner, 2006). Although many people with learning difficulties could reach more personal autonomy through the labour market, rates of employment are very low and social isolation is common (www.mencap.org.uk). Further, some conditions are associated with varying degrees of challenging behaviours so placement for some individuals can be difficult. Day care is available for people with learning disabilities, but ensuring that people are given a sense of purpose alongside social interaction in a place without judgement can be a challenge.
Mental illnesses, including, for example, depression, anxiety, personality disorders, schizophrenia and posttraumatic stress disorders are a leading cause of disability in the occidental cultures (Murray et al., 2012). In some countries, such as the UK, the prevalence of major depression is increasing and imposing huge personal and economic costs (Centre for Mental Health, 2010).
Likewise in Spain, although indicators of physical health have constantly improved during the last three decades, indicators of healthy habits (rates of cholesterol, diabetes, hypertension, allergies 1 Learning disabilities is an umbrella term for a range of conditions including Down's syndrome, fragile X syndrome and cerebral palsy. ASD can also be included here, but not all those with ASD have a learning disability. 4o f6 1 | and obesity) and mental health (such as the number of suicides and the number of psychological treatments) have worsened (Spanish National Ecosystem Assessment, 2013). As an early treatment, approximately 70% of depressed patients in UK primary care are prescribed antidepressant medication (Kendrick, Stuart, Newell, Geraghty, & Moore, 2015); however, adherence may be as low as one third (Bull et al., 2002). An alternative or adjunct to antidepressants is talking therapies. There are long waiting lists, and of those that complete the course around two thirds show signs of improvement and 40% recover (Department of Health, 2012). But for the many that do not take up the offer of talking therapies or who do not benefit from it, there are few alternatives. Social problems can also underpin many anxiety and depressive disorders. A more practical approach that directly targets these underpinning causes may be a more effective approach and an efficient use of resources.
Providing a safe calm environment that is nonconfrontational and offers structure and space for people to channel their energies into tasks that are mentally relaxing could provide a good fit for those who are unable to benefit from more conventional services.
Disengaged or disaffected children, defined as those who are not fully taking part in school life as they have given up trying or are resisting help (Lumby, 2013), are at high risk of exclusion from school. Exclusion from school can predispose young people to becoming a "NEET" (a person between the age of 16 and 24 and Not in Education, Employment or Training), which in turns carries an increased likelihood of committing a criminal offence, being in a lower paid job and subsequently a poorer quality of adult life compared to those who complete their education (Audit Commission, 2010;Public Health England, 2014). Evidence suggests that the numbers of children that fit within the disengaged category are increasing (McEwan et al., 2014;Robins, Cohen, Slomkowski, & Robins, 1999), and a large proportion of youth who show problem behaviour at a young age go on to develop antisocial personality disorders as an adult (Rutter et al., 1997) or can experience social exclusion (Hassiotis & Hall, 2008). Furthermore, there is also an increased risk of developing psychoactive substance use disorders, bipolar disorder and long-term smoking addictions (Biederman et al., 2008). Strategies to support children and young people in this situation are in place across a number of developed countries. For example, in the UK schools can refer pupils at risk of exclusion directly to off-site educational provisions. These can include local CFs which are contractually obliged to support teenagers to achieve National Open College Network accreditation. Importantly, classroom-based education is integrated with practical outdoor activities, which enables better student engagement.
Offenders often have mental and physical health problems (Brooker, Syson-Nibbs, Barrett, & Fox, 2009) or drug addiction and substance misuse problems (Abracen, Looman, & Anderson, 2000), and are more likely to have suffered from socioeconomic deprivation (Farrington, 1990), to have witnessed domestic violence (Caputo, Frick, & Brodsky, 1999), to have a family history of criminal violence (Farrington & West, 1990) or to have experienced harsh or neglectful parenting (Sutton, Utting, & Farrington, 2004). Poor education and a lack of skills predisposes individuals to unemployment, which itself is a risk factor for offending (Farrall, 2013). Some CFs aim to support offenders by developing self-esteem and providing work-based skills that provide hope for the future.
Being able to be physically active in nature may help to improve both the physical and mental well-being of older people (Elings, Haubenhofer, Hassink, Rietberg, & Michon, 2011). Levels of depression and anxiety are often higher among these groups than the general populations , and findings suggest that depression can cause worse health outcomes in older people when combined with chronic conditions such as arthritis, asthma or diabetes (Moussavi et al., 2007).

| The intervention
3.2.1 | Defining care farming Care farming (also called social farming) has been formally defined as the use of commercial and noncommercial farms and agricultural landscapes as a base for promoting mental and physical health, through normal farming activity (Hassink, 2003;Hassink & Van Dijk, 2006;Hine, Peacock, & Pretty, 2008a). A CF utilises the whole or part of a farm to provide health, social or educational care services, employment skills and support for different groups of people, through the provision of a supervised, structured programme of farming-related activities, rather than occasional one-off visits (Care Farming UK, 2016;Di Iacovo & O'Connor, 2009).
Care farming is a truly complex intervention. It may occupy part of a farm where farming production is the primary function (i.e., commercial agricultural units), or where the main function is provision of care services (i.e., community farms). Farms also differ in the types of farming activities undertaken (e.g., horticulture, forestry and livestock farming), other activities available (e.g., gardening, composting organic waste, medicinal plants work, conservation and woodwork), the level of support provided (e.g., health promotion, counselling, rehabilitation and skills qualifications) and the range of service user groups treated.
Given this complexity, the main defining feature of a CF is the involvement in agrarian or forestry activities for a therapeutic purpose. It is also important to highlight the farming component of the intervention, as this helps to distinguish care farms from horticultural or animal-based therapy projects. Care farms can function as a social enterprise where income gained by agricultural production is used to finance the CF (Elings et al., 2011).
• A diverse range of activities can be offered to service users at a care farm. Tasks  | 5o f6 1 php?id=33; http://www.socialfarmingacrossborders.org/seupb; http: //www.egina.eu/. In Madrid, a city farm with an urban orchard offers occupational activities and training for employment to people with learning disabilities. Among other activities they do horticultural work and raise livestock. The farm includes a one hectare urban orchard divided into 200 smaller areas that are rented to the general public. People with learning disabilities help clients to take care of their orchards and provide them with advice and support to keep orchards in a good condition. In addition, they attend school.
• In a farm in the North West of England the service users are primarily those with mental health problems, and activities are focused on horticultural production, but with some site maintenance. The service users also cook meals for themselves on site, often using produce that they have grown onsite. Service users are given work that increases in intensity as they recover. Working within a therapeutic community is the essence of the farm. • A city farm based in London runs a project on site which aims to reduce social isolation for older people living in residential homes and those using the services of older people's organisations. They specifically offer animal handling, which not only gives individuals an opportunity to touch and care, but also creates an avenue for open conversations to encourage social engagement.
• In the southern part of the Netherlands, a farmer and his wife (who works in health care) run a small scale CF with cows and arable produce. The farmer's wife provides day activities for people with learning difficulties and mental health problems. On average, eight service users access the farm each day, working together on different activities. They have coffee and lunchbreaks together with the family. Some of the service users work in the farm shop.
In addition to engaging in different activities, a small number of care farms offer service users the opportunity to interact with other professional caregivers to receive counselling or support to develop a healthier lifestyle. A recent survey of care farms in England found that, on average, 34 participants attended a CF each week. The length and duration of the CF intervention is determined by the need of the client, and this varies from one to three times a week, on average over a period of 30 weeks (Bragg, Egginton-Metters, Elsey, & Wood, 2014).
In addition, the intervention can vary depending on the wider cultural context in which the farm resides. For example, in the Netherlands, an agriculturally productive farm will offer some form of care or health promotion to their service users, whereas in Germany, care farms are frequently connected to a healthcare institution rather than being solely based on agricultural production farm (Haubenhofer, Elings, Hassink, & Hine, 2010). German care farms often function on a large scale, as government subsidies are only provided to farms with more than 300 service users (Haubenhofer, Blom-Zandstra, Kattenbroek, & Brandenburg, 2010).
The service users that utilise care farms also differ according to the setting of the intervention, for example, in Norway, the service users tend to be young children and psychiatric patients, whereas in the United Kingdom, Belgium, the Netherlands and Italy, a variety of different people use the intervention .
Individuals and communities supported through care farming include those with learning difficulties, ASD or mental health problems, plus disaffected youth, people with physical disabilities, older people, people with drug and alcohol problems, adult offenders, people with dementia, and exservice personnel (Bragg et al., 2014). In the UK, the largest service user groups are those with learning difficulties, ASD, mental health problems and disaffected youth. A similar pattern of intake is seen in the Netherlands, the country with the greatest number of care farms.
In developing countries and areas experiencing greater rural poverty, care farming is also now being used to support the longterm unemployed and empower women to become economically active (Food and Agriculture Organization, 2015).
Care farms can also provide support for offenders referred from probation services either as a rehabilitative intervention or as a way of "paying-back" to the community for crimes committed (Elsey et al., 2018). Elderly people, including those with dementia, are a more recent group to use care farming (Elings et al., 2011). Care farms can offer an alternative to day centres by providing a home from home environment that can involve some outdoor work for mental stimulation and physical activity.

| Care farming within the broader literature
The ways in which individuals interact with nature can be viewed as a continuum with overlapping categories, ranging from general everyday contact such as viewing, working or undertaking recreational activities, through to using nature deliberately as a therapeutic or treatment resource (i.e., green care) involving activities like wilderness therapy, social and therapeutic horticulture, animal-assisted therapy and care (social) farming (see Figure 1). Green care has been defined as follows: Green care utilises plants, animals and landscapes to create interventions to improve health and well-being; (i.e., it does not represent a casual encounter with nature). It also provides care and support to enable people to reach their true potential, that is, although many of the 6o f6 1 | approaches are termed "therapies" or "therapeutic", they are not necessarily directed at treating or curing conditions and diseases but, as in the case of people with learning difficulties, for example, they provide care, support, training and other opportunities to enable those individuals to develop. Such opportunities are often not available elsewhere .
Care farming is a distinct category within green care as the focus is on the use of a farm, either a commercial farms or other agricultural landscapes as a base for promoting mental and physical health, through normal farming activity (Hassink, 2003;Hassink, Zwartbol, Agricola, Elings, & Thissen, 2007;. Activities are not designed as "therapy" as they might be within a horticultural therapy or animalassisted therapy, rather they are the jobs that would need to be done on a farm to ensure successful production. Furthermore, care farms provide a range of farming activities that users can engage with. This provides a clear distinction with therapeutic horticultural activities and animalassisted interventions (AAI) which focus on a single activity such as gardening or horse riding.

| How the intervention might work
As a highly complex intervention comprising multiple activities and involving many client groups with differing needs, it is likely that multiple mechanisms and interactions will be at work to bring about changes in individuals. At the core of the intervention is the contact with nature which has value in its own right, but also provides the platform for the range of activities. Studies have shown that contact with nature has a positive effect on people's mental, physical, and psychological well-being, and spiritual beliefs (Bragg, 2013;Sempik, Hine, & Wilcox, 2010) and that engaging in nature based activities such as farming or gardening enables people to find solace . As a result, care farms may be beneficial for a wide range of service users.
A number of theories have been mentioned within the care farming literature and some of these speak specifically to the nature element such as attention restoration theory (Kaplan & Kaplan, 1989) and biophilia hypothesis (Wilson, 1984). Other theories relate specifically to the client groups that attend care farms, for example, desistence theory for offenders (McNeill & Weaver, 2010) and the recovery model for people with mental health problems (Anthony, 1993). Within these theories, mechanisms are proposed to explain how any effective intervention would be expected to bring about change. Identifying these mechanisms within the care farming interventions will indicate its fit with the theory, and therefore its likely effectiveness. For example, desistence theory suggests that interventions that lead to a reduction in recidivism involve building FIGURE 1 The different contexts in which an individual may engage with nature. Source: (Bragg & Atkins, 2016). The three columns represent the different contexts in which an individual may engage with nature. On the left, the "Everyday life" column highlights various situations in which an individual engages with nature as part of their normal lifestyle. The middle column "Health promotion" outlines a variety of existing group projects and initiatives which aim specifically to encourage individuals, communities and disadvantaged groups to benefit from nature-based activities. Funding is usually for the project as a whole and may come from public health, local authority grants or from the voluntary or private sector. On the right, the "Green care" column represents the various nature-based interventions which have been specifically commissioned for an individual with a defined health or social need as part of their care or treatment package human relationships, opportunities for reflection and change (Farrall & Bowling, 1999;Weaver & McNeill, 2007), developing self-efficacy (Maruna, 2001;McCulloch, 2005;McNeill, 2006) and social capital by learning and applying new skills to develop a new more positive identity (Farrall, 2004;Giordano, Cernkovich, & Rudolph, 2002;Laub & Samson, 1993;Maruna, 2001;McNeill & Maruna, 2007). A sense of community and the development of friendships are indeed valued aspects of a CF (Hassink, 2009). Furthermore, farmers are perceived as a role model with a strong sense of identity, thus offering an essential role model that can be emulated within a new identity (Hassink, De Meyer, van der Sman, & Veerman, 2011). Both the concepts of building human relationships and creating a new identity are clearly present within care farming interventions.

| Why it is important to do the review
With increasing pressures on the health and social care sector, commissioners and policy makers are turning to care farms as a potentially effective intervention. Farmers across Europe are becoming more multifunctional in how they use their land, and care farming may be an increasingly attractive option. As such, there is great potential to increase the use of care farms as an intervention to bring beneficial outcomes to a range of different population groups.
The growth in care farming in recent years is partly attributable to their commissioning successes with a range of health and social sector organisations through patient-referral and contracts for provision of support to health, social-care and probation clients.
Their sustainability is important given the increasing reliance that health and social care place on them. However, they remain heavily dependent on charitable funding, and policy changes over recent years have detrimentally impacted income streams.
Care farming is one of many third sector health interventions that are competing for similar funding streams. Its strength is its clear capacity to deliver care to a wide range of service users. Their ethos fits well with a number of theories relating to, for example, mental health recovery and rehabilitation of offenders. As is common for many interventions delivered by the third sector, the evidence base for their effectiveness is not well developed. This undermines the ability of the sector to move beyond being peripheral support organisations with limited core funding. In the past, the need to provide evidence was the domain of the pharmaceutical industry, but in recent decades this has expanded to cover complex health service evaluations. The methodologies for the latter are transferable to the third sector, but a lack of infrastructure and sustained income has hindered the development of a robust evidence base here.
Additionally, the complexities and multifaceted nature of care farms means that this is not an intervention that lends itself easily to a randomised controlled study design.
Nonetheless, there are a number of studies of care farms published in a wide range of journals across Europe. Although one systematic review and a small number of overviews exist (Bragg & Atkins, 2016;Elings, 2012b;, which document the extent and range of care farming initiatives and summarise the evidence for benefits, there is the need for a systematic review to capture the full range of both published and grey literature and to explore in depth the mechanisms that explain how care farms work for different client groups. Garnering this knowledge will help to clarify for policy makers and commissioners the unique contribution that care farming can make to health and social outcomes. There is the potential for care farming to improve the health and well-being of different population groups, and this is an important public health goal. If successful, they may have a role to play in reducing inequalities. Improving the lives of the most disadvantaged can have far-reaching societal benefits, for example, through enhancing social cohesion, reducing use of health and social care service usage and reducing crime (Wilkinson & Pickett, 2009).
This review aims to synthesise the existing evidence on how and for whom, care farming works, in order to improve health and wellbeing for a wide range of service users.
This systematic review is part of a feasibility and pilot study,

| OBJECTIVES
The primary objective is to systematically review the available evidence of the effects of care farms on quality of life, health and social well-being on service users.
Where possible we will synthesise the evidence in order:

| General approach
We conducted a mixed methods synthesis using a sequential explanatory approach (Pluye & Hong, 2014) that involved the development of an intervention framework based on existing theories. These theories propose how care farming might work, and our review used qualitative and quantitative evidence to test the processes and outcomes suggested by these theories. This approach is valuable in identifying possible mechanisms of change to inform the development of a logic model for care farming. An earlier scoping 8o f6 1 | review of the literature indicated a dearth of RCTs evaluating the effectiveness of care farms but instead highlighted a number of qualitative studies, a few small-scale RCTs and observational studies.
Thus a narrative approach which could synthesize the findings from a wide range of study designs was planned.
5.2 | Criteria for considering studies for this review

| Types of studies
The study designs considered for inclusion in the review were: • RCTs with randomisation at individual or cluster level.
• Quasi-RCTs and cluster quasi-RCTs, where participants are allocated by some means other than randomisation (e.g., by case number, date of birth).
• Interrupted time series that clearly define intervention points and record at least three outcome measurement points before and after (or before and during) the intervention.
• Nonrandomised observational studies that are prospective and have a control group, including: ○ Cohort studies, which ideally provide a reasonable timescale for effects to be detectable and attributable, and accurately record drop-out figures/characteristics.
○ Case control studies that report cases and controls from studies where comparability on relevant baseline characteristics and potential confounders can be judged, and comprehensively report confounders.
○ Controlled before and after studies, where data collection must be contemporaneous and groups comparable on baseline scores.
• Before and after studies that do not have a control group: The findings provided useful information on the nature and context of care farms and the mechanisms that may support effectiveness.
• Qualitative studies: All designs of qualitative studies were considered, including phenomenology, ethnography, and grounded theory. In addition, we also included qualitative studies with different methods of analysis, such as thematic analysis discourse/conversation analysis and narrative analysis.
We excluded single subject designs, reviews, overviews, surveys, commentaries and editorials. We also excluded theses with empirical data that had been subsequently published elsewhere.
In addition to these study designs, we also retained papers which described any theories offering explanations for how care farms might bring about change in the various population groups under investigation. As our interest here is purely to explore the theoretical basis by which care farming might work to initially inform the logic model(s), we do not refer to these papers as "included studies" or "excluded studies". These latter terms are for empirical data.

| Types of participants
Service users attending care farms of any age were included in the review. The list below presents the likely population groups: • Offenders serving community orders or similar sentences in the community rather than in prison; offenders "on-licence" (i.e., recently leaving prison to re-enter the community) • People with substance misuse, such as drug and alcohol problems

| Types of interventions
All care farms have some degree of "farming" (crops, livestock, woodland, etc.) and of "care" (including health care, social rehabilitation, education or training), but the balance of these elements differs from CF to care farm.
We included studies where the intervention met the definition of (b) "Providing services on a regular basis for participants": studies were included if the intervention was structured and service users attended several sessions rather than a planned "one-off" visit. The review also excluded petting farms and farms used for "one-off" educational activities.
We excluded care farming interventions that were combined with other interventions (i.e., music therapy) as we would be unable to differentiate the effects derived from actual farm work. We also excluded farming interventions that were provided in hospital or in prisons.

MURRAY ET AL.
| 9o f6 1 Eligible comparators included no intervention, wait-list controls or alternative interventions. Comparators were specific to the population group studied, for example, offenders serving their community order on a CF were compared to those serving their order cleaning public areas; or for those with addiction problems, another drug rehabilitation programme.

Primary outcomes
Care farms aim to improve a complex collection of social, educational and health outcomes for their service users. Given that the possible end impact of this complex interaction will be seen in changes in quality of life and mental health, this review included quality of life, anxiety and depression as the primary outcomes. Studies that did not use a validated instrument were not included in the analysis.

Secondary outcomes
Secondary outcomes varied according to the different populations, but we reported any mental health outcomes (in addition to quality of life, depression and anxiety as primary outcomes), social, physical and behavioural outcomes. Although we report all relevant outcomes we do not include in the analysis any secondary outcomes that had been developed in-house or failed to be defined.
The secondary outcomes included were: • Mental health outcomes: self-efficacy, self-esteem, stress, coping, mood, mental status, mental functioning, positive affect, rehabilitation and cognitive functioning, empowerment • Social outcomes: social functioning/interaction, group cohesion, recidivism, employment, school exclusion • Physical outcomes: functional performance, physical activity, and appetite and eating pattern • Behavioural outcomes: drug use, alcohol intake and smoking

| Duration of follow-up
The review included any length of follow-up of participants after their attendance at the care farm. Studies that only collected followup data at the beginning and at the end of each day were excluded.

| Types of settings
To be included, the studies need to explicitly state that activities took place on a farm that was not part of an institutional setting such as a prison or hospital. Community gardens and allotments were excluded.

| Electronic searches
Health, education, environmental, criminal justice and social science databases were searched to identify studies from a variety of disciplines.
Care farms are seen as both a health and a social intervention, and so are likely to be reported in the literature relating to these disciplines.
The selection of databases is extensive, offering a good international coverage of journals in attempt to identify relevant studies throughout the world. Further databases were added to those already identified in the protocol in order to identify studies commensurate with the range of potential outcomes and population groups. A single search strategy was used to identify both quantitative and qualitative studies. No restrictions were imposed on publication format or language in the search strategy.
In November 2014 we searched the following databases: • Applied Social Sciences Index and Abstracts (ASSIA) (ProQuest) from 1987 • CINAHL (EBSCO) from 1981 • The Campbell Library

| Search terms
The searches identified studies of care farms or agricultural-related therapies and rehabilitation practices within a farm setting.

| Selection of studies
We used a two stage screening process to identify eligible studies.
• Screening 1: Titles and abstracts Two reviewers independently screened the titles and abstracts of articles and grey literature retrieved to assess eligibility, as determined by the inclusion and exclusion criteria listed above.
• Screening 2: Full text For those studies that were selected as potentially eligible for inclusion, full copies were retrieved and two reviewers independently assessed whether studies met the inclusion criteria.
Any disagreements were discussed and resolved by consensus at each stage of the eligibility assessment. Multiple reports from the same study were coded separately before combining information across reports. We used the PRISMA chart to detail the process of study selection (Moher, Liberati, Tetzalaff, & Altman, 2009).
• Additional screening 3: Theories mentioned in care farming publications During full paper screening we also looked for theories that had been applied or mentioned within care farming studies. Even if the paper did not meet all of the inclusion criteria, it was retained so that we could use this as a source for identifying relevant theory.
The aim was to collate all theories quoted in relation to care farming, which were then explored in greater detail and used as a basis for our theoretical framework that explores the mechanisms of the intervention.

| Data extraction and management
We stored all the references identified by the search in EndNote

| 11 of 61
The data extraction for included information on the unit of analysis used in the studies, particularly where individual or cluster randomisation had occurred and whether individuals had received multiple interventions.
During data extraction, we only included qualitative themes that represented the views of the CF service users. However, in studies involving service users with communication difficulties, we included themes based on the recorded perspectives of significant others (care farmers, carers and parents) on the impact of the CF on the service users. We excluded themes from others that were about their own experiences, for example, care farmers' views on running a farm.
For papers that reported theories related to care farming, we extracted any summaries explaining how care farms might work and the anticipated outcomes. If the identified paper failed to provide an adequate description of this process we sought to identify the seminal paper. Risk of Bias tool were coded as high risk of bias; similarly, studies with two domains categorised as "weak" in the EPHPP tool were coded as high risk of bias. We did not exclude any studies based on risk of bias.

| Assessment of risk of bias in included studies
We pilot tested the tools with a sub-set of identified studies to ensure a consistent approach to assessment within the team. Two reviewers independently assessed risk of bias for each study. We resolved any disagreement by discussion or by involving an additional review team member.

| Measures of treatment effect
Where sufficient data was available we calculated effect sizes and 95% confidence intervals for each study using the

| Data synthesis of qualitative and quantitative findings
We based our data synthesis on a sequential exploratory approach (Pluye & Hong, 2014) (see Figure 2). This method involves: (a) identifying the main concepts from within theories found in relevant literature to explain why the intervention may work, (b) synthesising the qualitative data and then (c) interrogating the quantitative data to test the qualitative findings.
There were several stages within this synthesis which ultimately aided the construction of a logic model to explain how care farms might work for the heterogeneous study population as a whole and also for each population group. We based the design of our logic models on the description and definitions provided by the MRC's guidance for process evaluation of complex interventions (Moore et al., 2011). Here, a logic model is defined as: A diagrammatic representation of an intervention, describing anticipated delivery mechanisms (e.g. how resources will be applied to ensure implementation), intervention components (what is to be implemented), mechanisms of impact (the mechanisms through which an intervention will work) and intended outcomes. Reproduced from Moore et al. (2011) (p. 8).
Given the range of outcomes studied in care farms research, we designed our logic models to distinguish between "endpoint" health outcomes and proximal outcomes or mediators which are likely to be on the path to the endpoint health outcomes.
The stages of the synthesis were as follows: • Stage 1: Development of a preliminary theoretical framework to explain potential mechanisms of change

Stage 2
In Stage 2, two reviewers (N. W. and J. M. or H. E.) extracted themes from the qualitative studies to ensure that all relevant data had been captured. Where discrete themes were not presented in the papers, we looked for evidence of consensus among the participants as well as any discordant experiences. Negative as well as positive experiences were extracted. We opted for an inclusive approach to data extraction in the absence of discrete themes.
The same reviewers independently reviewed each of the extracted themes to identify which were composite and represented multiple discrete findings. These composite themes were independently deconstructed and the eventual findings were compared to ensure consensus on interpretation.
Each finding was entered into a spreadsheet, alongside its source and the client group studied. Three reviewers (J. M., N. W. and H. E.) independently categorised each finding as an intervention component (activity), mechanism, proximal and health outcome. These preidentified categories followed the definitions described in the MRC's 2011 model (Moore et al., 2011). Each finding was defined as: • Intervention: These included the facilities, activities and structure provided as part of the farm.
• Mechanism: The process by which part of the intervention might result in a particular outcome. These tended to be subjective experiences such as feelings and perceptions. For example, having physical contact with the animals (the intervention) would provide a sense of warmth and calmness (mechanism). Explicit links between mechanism and part of the intervention were not always reported. There can be multiple and linear mechanisms leading to the same outcome.
• Proximal outcome: An immediate outcome derived from a particular mechanism within the intervention. Primary health outcomes, as previously defined in this review, would not be categorised as a proximal outcome here. For example, having time away (mechanism) would lead to a sense of calm and reflection (also a mechanism) and feeling reduced stress (a proximal outcome). The key here is that there can be multiple proximal outcomes which mediate between the intervention activity and the outcome.
On agreement between the reviewers, each finding was transcribed onto a Post-It note in preparation for a clustering exercise (Backoff & Nutt, 1988). | 13 of 61 findings had not been over-interpreted (i.e., assumptions about what the mechanism might lead to) and thus placed in an unsuitable category. Given that the findings had been decontextualised during extraction and deconstruction of themes, this was an important iterative step that enabled the data to remain true to its source.
For the intervention components, one reviewer (J. M.) grouped the findings according to congruency and labelled each of the categories; this was subsequently checked by another reviewer (N.

W.).
As a gauge of the potential relative importance of each of the categories of mechanisms, we assessed the spread of the categories (across all the studies) and the frequency of the findings within each category. We carried this out for all the studies (all population groups) and for each individual population group (wherever possible).
We ordered the categories based on this assessment to explore the possibility that care farms might work in different ways for different populations.

Stage 3
The categories of, interventions, mechanisms and proximal outcomes were mapped to the theoretical concepts identified in Stage 1. This was performed by one reviewer (J. M.) and checked by a further two reviewers (N. W. and H. E.). The aim was to understand the ways in which change occurred and start testing out the theories using empirical data.

Stage 4
Two reviewers (N. W. and J. M. or H. E.) independently extracted all the quantitative results reported in the included studies. The quantitative data were summarised narratively according to the ESRC guidance (Popay, 2006). First, we assessed whether the care farms improved service user outcomes, caused harm to the service users or had no effect. If significant positive findings were

| Sensitivity analysis
To measure the robustness of the results we planned to conduct sensitivity analyses. We intended to conduct sensitivity analyses according to study design (i.e., excluding controlled before and after designs and any other non-RCTs) and according to the risk of bias, whereby we would assess sensitivity based on the inclusion and exclusion of studies with high risk of bias.

| Assessment of publication biases
We planned to use funnel plots for information about possible publication bias if we find sufficient studies (Higgins & Green, 2011).
A minimum of 10 studies with a common outcome measure is needed to be able to distinguish chance from real asymmetry (i.e., true publication bias) within the funnel plots (Higgins & Green, 2011). If asymmetry was found to be present, we would consider possible reasons for this.

| Deviations from protocol
In addition to providing a summary of risk of bias across the various domains within the studies, we had planned to summarise the overall weight of evidence that each study would contribute the review findings. However, recent Campbell reviews have tended not to use an overall quality scale. This is based on the concern that assessments of overall risk of bias may not take into consideration specific domains and are too dependent on the type of quality scale used (Brody et al., 2015).
Following the search and data extraction process, it became clear that there were several additional population groups using care farms which we had not been identified when writing the protocol. Given that our review aimed to understand how care farms may "work" for disadvantaged groups we decided to include any group that could be considered disadvantaged in some way. In light of this we included "socially isolated older people" but added an exclusion for "participants not from a vulnerable or disadvantaged population".
The process of review and data extraction helped us to further reflect on the definition of care farms. The definition of a CF used within the protocol was: "use of commercial farms and agricultural landscapes as a base for promoting mental and physical health through normal farming activities. Specifically, providing a structured supervised programme of health, vocational, social and/or farm related activities for vulnerable people." Within the review process, the importance of the "normal farming activities" became clearer and helped us to distinguish between interventions that were specifically designed as a "therapy", for example, horticultural therapy or equine therapy, and care farming which primarily focused on farming activity to sustain the farm and production, rather than primarily as therapy.
The review process identified a diverse range of primary and proximal outcomes. The protocol stated that the primary outcome was "quality of life". However, the review process identified a large number of studies (nine were included) that measured depression and anxiety. As these outcomes are frequently considered as "endpoint" health outcomes, we included these as primary outcomes in our presentation of results and the logic models.
The proximal and secondary outcomes identified during the review were varied and numerous. As described in the protocol we included any outcomes that used a recognised measure of health or wellbeing or behavioural factor and were assessed using self-report or objective measures. This helped us to identify pathways to change for different disadvantaged groups and develop a logic model to explain these relationships. Being too restrictive in the secondary outcomes for our review would have limited our understanding of these potential mechanisms.
In addition to the extraction fields specified in the protocol, we also extracted data on "duration of follow-up" (5.2.5) and "types of settings" (5.2.6). This enabled us to understand the importance of the setting and the sustainability of the impacts of cares farms on participant outcomes.
The protocol included a broad outline of the qualitative synthesis process. The detailed process of qualitative analysis using the four steps described in this report developed following further training of the review team on mixed methods systematic reviews. 6 | RESULTS 6.1 | Description of studies 6.1.1 | Results of the search We found 2,176 articles through searching of electronic databases and 125 via grey literature retrieval methods (see Figure 3). After duplicates were removed, we screened 1,659 references based on title and abstract. We obtained full copies of 215 articles and, of these, 31 studies (reported in 42 papers) met the inclusion criteria.
In a separate screening process, we were able to identify theoretical and contextual information about care farming interventions in 26 publications. Seven of these theory publications also reported empirical work, six had used qualitative | 15 of 61 methods and one was an uncontrolled before and after study.
These seven studies were screened and included in the subsequent stages of the review, that is, in the 31 studies mentioned above. Those that were purely theoretical or did not meet our inclusion criteria for empirical studies, were used only for Stage 1 of the review process.

| Included studies
A total of 31 studies were included. Eighteen qualitative studies (reported in 21 papers) (Table 1), 13 quantitative studies (reported in 21 papers) ( Table 2), and one mixed methods study (Elings et al., 2011) met the inclusion criteria for this review.  (Table 3). The most commonly applied theoretical concept mentioned in studies was the recovery model (mentioned in four studies) (Anthony, 1993). Two concepts were philosophical rather than theoretical and did not offer a mechanistic explanation for how care farming might contribute to well-being; namely, "existential   Steiner, 1925). These were excluded from the process of developing a theoretical framework.

| Characteristics of included qualitative studies
All of the included qualitative studies (See Table 1 the second study, limited information was gathered from service users (Elings, 2004), and in the third study, accounts of farmers, carers and parents supplemented the visual elicitation methods adopted by the researcher (Kaley, 2015). Ten studies failed to provide information on the age of the study participants, and gender was not reported in five studies. Excluding those studies where gender was not reported, there were almost twice as many male service users participating in the studies as females (ratio of 1.8:1).

| Characteristics of included quantitative studies
The 13 studies were conducted in five different countries: four in Norway; four in the UK; three in the Netherlands, and one each in Pakistan and the United States (see Table 2). There were two RCTs and three controlled before and after studies (CBAs), with the remaining nine using an uncontrolled before and after design (UBAs).
The two RCTs involved single target groups, both focusing on mental illness. Ten studies evaluated the effects of care farming on a targeted single client group: six were on service users with mental   (2015), six car farm staff and seven carers (this data only supplemented the interviews with service users; de Bruin et al. (2015), 12 people on a waiting list for the CF and 17 people attending regular day care services. b These were farmers who provided information on behalf of the service users.  (one of which provided percentages rather than numbers); there were more than twice as many males compared to females (n = 261 males; 117 females). The mean ages of participants in the studies ranged from 9 to 78 years. However, age was not reported in two studies (Hine, Barton, & Pretty, 2009;Marshall & Wakeham, 2015).
The intensity and duration of interventions varied, but most commonly involved half day (1.5-3 hr) or full day (5-6 hr) sessions two to three times per week over a 12 week period. In the two studies involving disadvantaged youth Suprise, 2013), the duration of intervention was substantially longer, with one study mentioning 6 months and the other with an open-ended contract. Studies involving service users with mental health problems most commonly stated a 12 week intervention period.

Data collection time points
One CBA study involving offenders on a community order completed follow-ups mostly just prior to the end of the intervention to maximise retention in the study (Elsey, Murray, & Bragg, 2016). Four UBA studies (Hine et al., 2009;Hine, Peacock, & Pretty, 2008b;Lambert, 2014;Pedersen, Nordaunet, Martinsen, Berget, & Braastad, 2011) performed follow-ups immediately after the intervention. The RCTs reported follow-ups at 6 months (from baseline) (Berget, Ekeberg, Pedersen, & Braastad, 2011;Pedersen et al., 2012b). The remaining studies reported outcomes at 12 months (four studies), 6 months (two studies) and 3 months (one study). Only one study did not report the time point of follow-up (Suprise, 2013). The longest follow-up period reported was three years from a UBA study (Javed, Chaudhry, Suleman, & Chaudhry, 1993) involving service users with mental health problems; however, the duration of the intervention was not provided.  These outcomes were not included in the logic models as the measure was not defined or the outcome had been modified without adequate description or validation.  (Ulrich, 1983) [1 study: Leck et al., 2015] Ulrich argues that being in contact with nature can reduce stress. He argues that affective reactions (i.e., feelings) precedes cognitive responses. An affective reaction is an immediate emotional response, that is naturally triggered such as joy, like or dislike. The affective reaction shapes the subsequent conscious processing, physiological responding and behaviour. According to Ulrich, natural settings triggers positive affective reactions, followed by positive physiological response or positive behaviour

Outcomes
Restorative effects of nature Biophilia (Wilson, 1984) [2 studies: Pedersen et al., 2012a andLeck et al., 2015] Biophilia is a fundamental and biologically based human need and a propensity to affiliate with life and lifelike processes. According to Wilson biophilia is inherent in every person, put another way, it is a biological need. Biophilia is part of people's evolutionary heritage (i.e., our ancestors evolved in natural environments) Restorative effects of nature Presence Theory (Baart, 2001;Droës & van Weeghel, 1994;Kal, 2002) Caring involvement in response to the need for intimacy and involvement. People thrive on company but feel isolation if they lack intimacy. In presence approach, the "carer" offers a way out of isolation through being a caring presence. There are no hierarchical differences, no particular goal or intervention/treatment route… Care worker is just "attentively present". It requires trust, meaningful relationships, where client feels seen and counted. It is about being there, being together, doing things together Being socially connected; mental Well-being Social Support and Social Interactions (Cobb, 1976;House, 1981) [1 study: Ellingsen-Dalskaua et al., 2016] There are four main domains of support. Informational support includes giving advice, information and instructions. Emotional support is about having concern, listening and providing trust. Appraisal support involves affirmation and feedback and is likely to be a part of the contact between the farmer and the participant. Instrumental support is practical support and in the case of care farming the provision of, for example, tools, food and equipment. Social support is information which lets us feel cared for and loved; esteemed and valued; a member of a network of mutual obligations. Having social support facilitates coping with crisis and adaptation to change. Since humans are innately drawn to animals, animals serve as a medium through which social interactions can transpire   (Antonovsky, 1979(Antonovsky, , 1996 [1 study : Schreuder et al., 2014] Having a positive outlook or optimistic attitude contributes to better health. The SOC is used to explain why some people remain healthy under stress. The SOC includes three dimensions: Personal growth; mental wellbeing • Comprehensibility: believe that the challenge is understood • Manageability: believe that resources are available to cope • Meaningfulness: believe that the challenge is worthy of commitment It is hypothesised that people with higher SOC scores are more able to remain health under stress Spiritual Experience Process Funnel (Fox, 1999) When people start to feel relaxed in wilderness they become open to opportunities for spiritual experience and become more connected to nature. Over time this spiritual experience can develop into spiritual growth which can contribute towards significant changes in attitude and adoption of new behaviours Restorative effects of nature Recovery Model (Anthony, 1993) [3 studies: Granerud and Eriksson, (2014) and Elings et al. (2011);Hassink, 2009;Hassink et al., 2010, Iancu et al., 2014Kogstad et al., 2016) This is a person-oriented perspective whereby people with mental disorders go through a personal journey and adapt to a new status quo and learn to find personal meaning despite and beyond the limitations imposed by their mental ill-health: Being socially connected; personal growth; mental wellbeing • Moratorium: denial of the mental diagnosis, confusion, helplessness • Awareness: awareness of a possible identity beyond that of a "sick person" • Preparation: focus on one's values, strengths and weaknesses • Rebuilding: actively pursuing a positive identity, stablishing goals and taking responsibilities • Growth: living beyond disability and being resilient Ecological Model of Aging (Lawton & Nahemow, 1973) Through providing an environment that is a good fit with needs/abilities. Purports that this is achieved through an environment that is compensatory, constant, predictable and stimulating (Lawton, 1989) Being socially connected; physical well-being   (Table 4). For example, the physicality of the work is likely to vary according to age, physical ability and mental health, but some studies only mentioned working with animals as an activity.
The types of activities reported on care farms fell into four categories: • Horticultural or land maintenance work-in addition to the more traditional growing of vegetables and fruit, activities also included hedge cutting, conservation work, tree planting and mending fencing. All client groups were reported to have participated in these types of activities.  Attachment Theory (Bowlby, 1969) Aims to address trust and security issues. Through the use of animals to create healthy attachments and promote development of prosocial behaviours by restoring a sense of trust and security in interpersonal relationships Being socially connected Intentionally designed experiences (Sheard & Golby, 2006) Taken from adventure playground literature but considered that green are activities are examples of IDEs with engagement with the natural world working at all levels: looking at nature, being active in nature, shaping nature and interacting with animals and the IDEs conceptualise how activities provide a chain of events where care farms are vectors for health benefits including first order outcomes achievement, restoration, resilience and empowerment and second order outcomes stress reduction, self-efficacy, identity formation and social support Restorative effects of nature; mental well-being; being socially connected, personal growth Therapeutic Landscape Concept (Gesler, 1992) [1 study : Kaley, 2015] A therapeutic landscape is one win which "physical and built environments, social conditions, and human perceptions combine to produce an atmosphere which is conductive to healing…healing induces cure in the biomedical sense (physical healing), a sense of psychological well-being (mental health) and feelings of spiritual renewal (spiritual healing)" Restorative effects of nature; mental well-being Behavioural theory (Lewinsohn, 1974) Certain environmental changes and avoidant behaviours inhibit individuals from experiencing environmental reward and reinforcement and subsequently leads to the development of depressive symptoms. By encouraging individuals to take part in activities that create a sense of pleasure or mastery, avoidant behaviours can be reduced Personal growth; mental wellbeing Abbreviations: IDE, intentionally designed experience; SOC, sense of coherence.  Most activities on the farms were related agricultural production; training of users for integration into the labour market in two farms and other daytime activities for people living under supported housing (n = 1). On the 12 private farms, supervision was provided by farmers (n = 3), by farmers previously trained as mental health nurses or social workers (n = 4), by professional activity supervisors (n =3)or by both trained farmers and professionals (n =2) One care farm was owned by a mental health organisation, and employed a farmer and several professional activity supervisors for the guidance of users.   Aim to provide an adequate day structure and a meaningful day programme to frail and/or community dwelling elderly people, so as to prevent social isolation and to offer respite care to informal caregivers at home Activities do not contribute to agricultural production. They include farm or animal related activities (watching or feeding animals, cleaning pens and cages, picking eggs); garden or yard related activities (sweeping yards, gardening, working in greenhouse); games (party games, memory games, quizzes, billiards, shovelboard); crafts (flower arranging, decorating postcards, knitting, making nest boxes, sanding or painting fences); other leisure and recreational activities (dancing, singing, gymnastics, going for a walk, reading, participating in group discussions); domestic activities (peeling potatoes, chopping fruit and vegetables, laying the table, dish washing, shopping); sitting or pottering while watching and/ or chatting (no involvement in organised activity); resting (sleeping or napping in chair or in bed) Farms are often co-operatives with regular health care institutions. Their services are financed by the Dutch national insurance system

Mixed groups
Little Gate Farm (2015), UK To enrich the lives of children with special needs and give them the opportunity to gain independence and confidence, while at the same time having a lot of fun and learning lot of new things, such as farming, animal care and where food comes from.
To support learning disabled adults to learn practical farm and woodland skills Farming, animal care, animal feeding and handling, making our own pizza dough bases and topping; chick cleaning and holding, craft (making bird feeders and bird cake, decorating a flower pot and planting a sunflower); woodland den building; animal cleaning and feeding. Animal care, horticulture, woodland management, traditional skills, enterprise and conservation Charity funding with animals, depending on the abilities of the client (Berget et al., 2007).
• Additional farm animal-based activities-beekeeping, fish farming, maintaining a mini zoo and working with donkeys.
• Other activities-these included working in the shop, outdoor recreational activities (camping, campfires, outdoor trips and den building) and indoor activities (baking, meal preparation, crafts, games, general household work and tractor driving).
There was a general lack of information regarding contractual arrangements of care farms. A range of models were in place: care farms as part of a nursing home or mental health care organisation; privately-owned farms working in collaboration with health care organisations (the Netherlands and UK) or the welfare sector (Norway) or probation (UK); and privately owned farms with income generated through personal budgets, charitable donations or grants.

| Excluded studies
One hundred fifty-one studies were excluded after examining the full text. Four excluded studies consisted of single subject studies. Eight studies were excluded because the participants were not from a vulnerable or disadvantaged population; for instance, the participants were school children visiting a farm for educational purposes.
Twenty-four studies were excluded because the studies did not meet the care farming definition. Some studies classified activities as "therapy" rather than activities that are therapeutic, so we excluded four studies. Twelve studies were excluded on the grounds of setting; these studies were not delivered at a farm, but instead at a prison or a hospital. Four studies were excluded because the intervention exclusively consisted of single activities such as gardening or horse riding. Some studies combined different interventions, for example, care farming activities combined with learning music at a recreation centre. For these studies, it was difficult to separate the true effect of the care farms, so three studies were excluded. Two studies consisted of "one-off" educational visits to the CF and were excluded.
Eighty-five studies were excluded because they were reviews, overviews, surveys, commentaries or editorials. Five PhD theses were excluded because their findings had been subsequently published elsewhere and the peer-reviewed publication was included in this review.

| Qualitative studies
Nine studies (50%) fully met more than 50% of the 37 quality assessment criteria ( Note: AWZB accreditation-The Dutch "general law on exceptional medical expenses" (AWBZ) provides general insurance covering special health care needs. Care is either provided "in kind" through certified health care institutions or can be hired by clients through a personal budget. Some farms in the Netherlands are registered as certified health care institutions (Elings, 2007). Studies in which no details about the intervention included: Leck et al. (2015); Di Acova (2013) pre-existing or newly established relationships were only addressed by one study each. Two criteria fundamental to all research practice are evidence of ethical approval and of informed consent. These were not reported in nine (50%) studies.
We observed that eight of the ten studies that met (fully or partially) more than 50% of the quality criteria used a theoretical framework. Conversely, only one  of the eight studies scoring <50% in the quality assessment used a theoretical framework. The implications this might have on the quality of the results are unclear. Studies involving service users with mental health problems that used the recovery model reported greater variability in the extracted findings, specifically the range of mechanisms, compared to those who did not use a framework.

| Quantitative studies
All the included quantitative studies had many limitations and were assessed as having a high risk of bias. A summary of the risk of bias of the quantitative studies can be found in Tables 6 and 7.

Randomised controlled trials
Allocation. The method of random sequence generation was described clearly in both RCTs. For example,  used computer-generated random numbers. However, only one study clearly described the allocation concealment.  did not address allocation concealment whereas in Pedersen et al.
(2012b) randomisation was conducted by a researcher blinded to farm and participants.
Baseline outcomes. Patient outcomes were measured at baseline in both studies, and one study reported no important differences across intervention groups . However, Pedersen et al.
(2012b) reported higher depression scores and anxiety scores in the control group at recruitment, and higher self-efficacy scores in the intervention group at recruitment, but these differences at baseline were not adjusted in the analysis.

Baseline characteristics. Pedersen et al. (2012b) reported differences
in baseline characteristics between the intervention and control groups. For example, there were more men, and participants were older and better educated in the intervention group. It is unclear whether the baseline characteristics were similar in the study conducted by . For instance, some characteristics are mentioned in text, but no data were presented for the intervention and control groups separately.
Incomplete outcome data. Both studies reported attrition rates and the number of participants excluded from the analysis. In both studies, proportionally more people dropped out of the CF arm than in the control arm: 32% and 50%  versus 7% and 15% (Pedersen et al., 2012b). It should be noted that in the latter study (Pedersen et al., 2012b), the number of included participants were very small (n = 29), the control group was a waitlist group, and half of those dropping out of the CF arm did so before the intervention started. The reasons for drop out were little interest in animals and boredom . Furthermore, it was reported that significantly higher drop-out rates were observed in those using sleeping medication (p = .05), and hospitalised patients (p = .006) .
Blinding. Primary outcomes variables were not assessed blindly in both studies (Berget et al., 2007;Pedersen et al., 2012b). This was reported as a limitation in the discussion section of both studies.

Contamination. Pedersen et al. (2012b) used a wait-list control
group and it is unlikely that the wait-list control group received the intervention prior to the intervention group. However, it is uncertain whether there was contamination in Berget and colleagues' study.
They report that the control group received treatment as usual, but do not give any additional description.
Selective outcome reporting. There was no evidence that the outcomes were selectively reported in both studies; for instance, all the outcomes described in the methods section were reported in the results section Pedersen et al., 2012b). Neither study published a protocol detailing outcomes to be measured a priori.
Controlled before and after studies and uncontrolled before and after studies Selection bias. Only one study had selected individuals that were likely to be representative of the target population. Three studies had selected individuals that were somewhat likely to be representative of the target population, for example, through referral from clinicians in a systematic way. Seven studies did not use a systematic process to select individuals.
Study design. We assessed the likelihood of bias due to the allocation process; all eleven controlled before and after studies and uncontrolled before and after studies were rated at moderate risk of bias as the investigators did not use a robust process to select participants.
Confounders. Only one study controlled for at least 80% of relevant confounders. Four studies controlled for approximately 60-79% of relevant confounders. Six studies either controlled <60% of relevant confounders or did not report any confounders.
Blinding. In the majority of studies (nine studies), the outcome assessors were aware of the intervention status of participants. Two studies did not describe blinding.
Data collection method. Five studies used valid and reliable tools to collect data. Four studies did not describe the reliability of the data MURRAY ET AL.

| 31 of 61
collection tools and two studies did not describe the data collection tools used to measure outcomes.
Withdrawals and drop-outs. Only one study reported a follow-up rate >80%. Four studies reported follow-up rates between 60-79%. Six studies either reported follow-up rates <60% or failed to report withdrawals and drop-outs rates.
6.3 | Synthesis of results 6.3.1 | Stage 1: development of a preliminary theoretical framework

Theories and theoretical concepts
Theories (see Table 3 for a complete list and descriptions) differed in scope and in the extent to which they explained causation, thus contributing to the development of a theoretical framework in different ways. With regards to scope, some theories provided a rich, focused, description for how multiple but seemingly disparate dimensions of life could combine to produce a specific outcome. For example, the recovery model (Anthony, 1993) describes aspects of identity, achievement and social connectedness for improved mental health. In comparison, the ecological model of aging (Lawton & Nahemow, 1973) starts with a broader premise and offers a more superficial description of possible mechanisms drawing again on disparate entities, but with a set of defined outcomes relating to cognition, psychology and physiology rather than just one.
Some theories were complex, and so mapping the farming mechanisms derived from the qualitative studies to them in their original state was impractical. Instead we distilled out the key concepts from each

Proximal outcomes suggested by the theories
Proximal outcomes suggested by theories relate to confidence (SHIFT desistance theory), stress (attention restoration theory; psychoevolutionary theory; intentionally designed experiences), coping (social support theory; SHIFT desistance theory) and self-efficacy (self-efficacy theory), prosocial behaviours (attachment theory; SHIFT desistance theory).
This list of outcomes and proximal outcomes is not definitive since arguably many supposed outcomes might actually be part of the mechanisms contributing to the theory. For example, the recovery model talks about being "in work" as part of the recovery from mental illness rather than necessarily seeing it as an outcome in its own right. The aim here is to look at the role of various theories in explaining how care farms might work rather than defining the developing logic models by the theories themselves.
6.3.2 | Stage 2: identification of care farming components, mechanisms and proximal outcomes from qualitative studies Through the process of deconstruction of reported themes, we identified 85 intervention components (grouped into four categories), 164 mechanisms (grouped into 15 categories) and 24 proximal outcomes.

Care farming components
Five categories of components were identified (see Table 8): • Being in a group-comprised mostly positive findings about the benefits of working with other people. Findings included "relatively stable and informal group working", "working together" and "interacting with different people". This category also included two negative findings (from two different studies both involving people with mental health problems) about this aspect of the care farming intervention, and these included "not wanting to interact with others" and "finding it challenging to deal with disabled users".
• The farmer-all findings were positive and related to how the farmer and farm staff supported the service users through the activities they provided and individually. Findings included "being able to express how they felt", the farming "seeing them as normal" and "providing practical experience".
• The work-findings relating to the actual activities revealed commonalities, but also diversity in preferences. The pressure of the work was valued in some studies, while in others being able to do work at one's own pace was expressed as important. Doing "real" and varied work was also reported as a benefit. There was one negative finding about "not enjoying some of the tasks because it was a working farm".
• The animals-none of the findings about animals were negative experiences. Being able to touch, be responsible for and overcoming fear of animals were reported findings in this category.
• The setting-quietness and space to be alone were common features of the setting that service users identified. Being outside and experiencing nature were also reported. There was one The work 29 (32) 20 (28) 7 (28) 10 (48) The animals 12 (13) 11 (15) 3 (12) 0 The setting 11 (12) 9 (13) 6 (24) 1 (5) All 92 71 25 21 Note: Twenty-two of mental ill-health and substance misuse findings also included disaffected youth and service users with learning disabilities and older people. a Only five of 25 findings were solely disaffected youth. b Sevenof 21 findings included service users from other groups.
negative finding involving mental health service users who felt that they were "on display" because of the educational visits.
Overall, care farming intervention components relating to the farmer and the work appeared to be prominent features in the findings (Table 8). Despite the fact that data were infrequently reported for single client groups we did observe some differences in the types of components mentioned that may indicate differences in either the types of activities made available to disparate client groups or the level of importance of those activities to types of service users.
For example, studies involving predominantly people with learning disabilities did not mention activities relating to the animals or the setting. However, studies involving disaffected youth reported a preponderance of work and setting related activities.

Mechanisms
Through the iterative clustering exercise, mechanism based findings were organised into 15 categories of mechanisms (Table 9 for description of each category). Across the studies the number of findings relating to mechanisms ranged from 3 to 22. In general, theory-based studies identified more mechanism findings (Table 10).
In terms of frequency and spread of findings, "understanding the self", "social relationships" and "belonging and non-judgement" represented the most common categories across all studies (represented in bold in Table 10). "Creating a new identity" and the farm as a "distraction" were least often observed across the studies.

Comparing mechanisms across client groups
Where there were sufficient data, we ordered the categories of mechanism according the frequency with which they were reported for each client group (Table 11).
As all of the substance misuse findings were reported with mental illness findings, we report these as one client group. As the largest group, with 105 findings from 10 studies, a similar pattern to the overall findings was present in the mental health problems and substance misuse group. No findings relating to "reflection" or "creating a new identity" were found in this combined client group.
For disaffected youth, "feeling safe" was more frequently reported than "belonging and non-judgement". "Achievement and satisfaction" was frequently mentioned in both the mental health problems/substance misuse group and the learning disabilities group, but it was reported less often in the disaffected youth group.
"Reflection" was also reported more often in the disaffected youth group compared to the others.
In the learning disability client group, "understanding the self" was reported less frequently than "social relationships", "belonging and non-judgement", "social relationships" and "meaningfulness".
"Physical health" was also reported much less frequently in this client group than in the others. "Creating a new identity", which described how people with learning disabilities aligned themselves with the farmer, was the seventh most often reported category, but did not appear in either the mental illness/substance misuse or the disaffected youth groups.
As there were only 10 and five findings from the older people and autistic spectrum disorder client groups respectively, we did not order the mechanisms according to frequency of reporting.

Proximal outcomes
We extracted 24 proximal outcomes (Table 10), identified by participants in the qualitative studies as benefits of being on a care farm. Most (n = 11) related to emotions, such as increased confidence and self-esteem, which mainly arose from studies underpinned by the recovery model for mental health. Improved coping and feelings of well-being were also mentioned in numerous studies, as was independence. In five studies there were no reported outcomes.
There were many more benefits reported by service users than those explicitly proposed by the theories, but as already mentioned in Section 6.3.1 (Stage 1), this may reflect the emphasis on theories on the mechanisms. In a study involving disaffected youth, only two proximal outcomes (happiness and changing behaviours) were reported.

| Stage 3: mapping of qualitative data to theoretical framework and creation of logic models
The categories of mechanisms from the qualitative studies were mapped to the five theoretical concepts (Table 12). Some of the categories fit across more than one concept. So, for example, "belonging/non-judgement" included findings such as "being in an inclusive environment" and "animals are safe and do not judge".W e considered that the former example fitted better with the theoretical concept of "being socially connected" while the latter finding fit with "mental well-being" (Table 13).
Only four single findings within the mechanism categories of "reflection", "stimulation" and "feeling safe" appeared to map to the theoretical concept of "restorative effects of nature". These findings were "silence in nature", "peace", "enjoying the sensory experience of being with animals" and "cuddling the animals gives a sense of security". We considered that these primarily mapped to the theoretical concept of "mental well-being", but had links to the "restorative effects of nature". The dearth of findings that map to this theoretical concept occurred despite "the setting" of a farm, which could be considered as "nature", being mentioned frequently in the qualitative studies as an important component of the intervention.
The theoretical concepts of "mental well-being", "being socially connected" and "personal growth" were best represented by the qualitative mechanisms overall. Across the three main client groups (mental health problems/substance misuse; disaffected youth; learning difficulties), there were some differences. In the mental health problems/substance misuse group, the number of mechanism findings that mapped to "mental well-being" was almost double that of any other theoretical concept. In the other client groups, "being socially connected" and "mental well-being" were similarly represented by the mechanisms. The categories of mechanisms were then combined with the intervention components and proximal outcomes to create a logic model for the following client groups: Feeling valued and respected Service users feel valued, appreciated and needed by the farmer (and the animals) and consider that they are respected "for who they are"

15
Feeling safe The atmosphere at the farm creates a feeling of safety and security, providing a mental shield between illnesses and addictions. For some service users this experience is enhanced through physical contact with the animals but for others there is a need to overcome fear of animals which can then lead on to a feeling of safety 12 Learning skills Care farms give service users the opportunity to learn new skills ranging from growing crops to looking after animals which enables some to gain qualifications enabling then to (re)enter the work place 11 Meaningfulness Service users perceive tasks as meaningful because they are judged to be useful to others and are needed to conduct day to day activities at the farm. service users also see their role as personally meaningful, contributing to society giving them a sense of purpose, happiness and fulfilment 13 Nurturing Through helping each other and caring for the animals/plants service users become consider of other peoples' needs and recognise they are doing good for other living creatures 5 Physical well-being Through physical activity on the farm service users improve their physical strength. There is a sense of "good" tiredness from physical work. service users start to feel more independent and healthier 10 Reflection The care farm environment is quiet and peaceful allowing service users to stop and reflect about their problems, their social influences and also the progress they have made. For young people, working at the farm gives space and time away from their family and friends 4 Social relationships Care farms provide opportunities for participants to interact with the farmer, and other service users. For instance, often service users were working together in groups which helped them to develop their communication skills. As the intervention progressed the service users deepened their relationships with the farmer and considered him as a role model. Once service users gained social confidence, their social networks grew. In particular, they found that in social functions talking about their farm work was more interesting rather than talking about their illness. However, a few service users did not want to interact with others and found it difficult to deal with the diverse range of service users at the farm 17 Stimulation Service users find tasks stimulating giving them more energy, encouraging a mindful approach to work especially around animals which are unpredictable. Working with animals offers a sensory experience and the energy derived from the work enables them to work through their own problems better. The experience of being in nature is energising 7 (Continues) • All client groups (Figure 4) • Mental health problems and substance misuse ( Figure 5) • Disaffected youth (but includes some qualitative findings from other client groups) ( Figure 6) • Learning difficulties (but, as above, includes some qualitative findings from other client groups) (Figure 7) In general, there was a lack of sufficient evidence detailing which intervention component linked to which categories of mechanisms, and thereafter which proximal outcomes and outcomes. Therefore, the logic models only provide a single connecting arrow between each of these aspects.

| Quantitative results
The quantitative evidence was mapped onto both the proximal outcomes and the endpoint health outcomes in the logic models to.
Based on our overall logic model built from theory and the qualitative evidence, we expected to find empirical evidence suggesting that care farms would improve: Endpoint outcomes: • Quality of life (primary outcome identified from theory) • Anxiety (primary outcome identified from theory and qualitative studies) • Depression (primary outcome identified from theory) Proximal outcomes: • Self-efficacy (theory) • Confidence (theory and qualitative studies) • Coping skills (theory and qualitative studies) • Independence (qualitative studies) • Social activity (qualitative studies) • Self-esteem (qualitative studies) • Self-image 2 (theory and qualitative studies) • Physical well-being (including having more active lifestyles and being physically tired (all from qualitative studies) • Happiness or well-being (qualitative studies) • Vocational skills (qualitative studies) • Stress (theory) • Negative behaviours (theory and qualitative studies) • Medication usage (qualitative studies) No quantitative studies were found that evaluated the impact of care farms on confidence, personal identity and physical well-being (including tiredness). Changes in negative social behaviours were measured, but only one form (reduction in reoffending) was clearly defined. Additionally, vocational skills may have been measured in the form of occupational functioning and work abilities. However, as these outcomes were either not defined or incorporated highly subjective measurements, there is no clear result.
We found evidence relating to quality of life, self-efficacy, coping skills, independence, social activity, well-being, anxiety, depression, stress and medication usage. In addition to the outcomes identified from theory and qualitative evidence in the logic model, four further outcomes were found from the quantitative studies, namely cognitive functioning, improvements in psychiatric status (from chronic psychiatric illness), positive affect and appetite and eating pattern.
These were added to the logic models.
The majority of the evidence was derived from studies involving service users with mental health problems and substance misuse problems. This meant that quantitative results relating to disaffected youth and users with learning difficulties could not be mapped against these logic models.

Mechanism categories Description
Frequency of findings in each category (all groups from across all qualitative findings)

Structure
The daily farming activities provided a predictable work environment to the service users. This consistency helped the service users to gain a normal rhythm. Moreover, the farmers also allowed service users to work at their own pace as they understood that the service users can have a "bad day" and may not be able to work at full capacity. Similarly, farmers involved participants in deciding tasks for the day 8 Understanding the self The care farm environment has allowed service users to better understand themselves. Participant's self-awareness grew while at the care farm. For example, learning to master an activity at the farm increased their self-respect and positive self-image. At the farm, participants were free to be themselves, they also had the opportunity to learn and when they made mistakes they were given time and guidance to learn from their mistakes. This gave them the understanding that tasks at the farm are manageable which enhanced their self-efficacy and self-confidence. Some found caring and cuddling animals helped them to deal with problems 26   Three studies (two RCTs and a UBA study) assessed participants' anxiety at two follow-up points (see Table 2). The first RCT found no significant change in anxiety between groups at 12 week follow-up ( B e r g e te ta l . ,2 0 1 1 ) .H o w e v e r ,a t6 -month follow-up, they found a statistically significant positive effect of the intervention in reducing anxiety compared to the control group. The authors reported that this positive effect is also clinically significant because the participants were diagnosed with severe anxiety at baseline, which improved to moderate anxiety at 6-month follow-up. In the second RCT, Pedersen et al. (2012b) found no significant change in anxiety between groups at the end of the intervention follow-up (12 weeks) and 3 months after the intervention (Pedersen et al., 2012b). Gonzalez et al. (2011aGonzalez et al. ( , 2011b) reported a statistically significant but transient reduction in Understanding the self 1 1 2 5 Abbreviations: ASD, autism spectrum disorder; DY, disaffected youth; LD, learning difficulties; MH, mental ill-health; SM, substance misuse. *Rank represents the frequency of the findings in each category and the spread of the findings across the studies for that client group. **Older people and autism spectrum disorder not separately represented due to very low numbers of findings. The most common categories across all studies are highlighted in bold. Mental well-being Feeling safe, structure, belonging/nonjudgement, meaningfulness, reflection, feeling valued and respected, achievement and satisfaction, stimulation and distraction anxiety at 12 week follow-up, but anxiety levels were still within the clinically severe range (remaining above the estimated clinical cut-off of ≥45) (Spielberger, 1983). At 3-month follow-up, change in anxiety scores were no longer statistically significant.
Four studies reported depression outcomes immediately after completion of the intervention. Both RCTs reported no significant change in depression between groups at 12 week follow-up Pedersen et al., 2012b). A UBA study found a statistically significant reduction in depression at the end of the intervention (12 weeks), and 3 months after the intervention (Gonzalez et al., 2011a(Gonzalez et al., , 2011b. The results were clinically significant as the participants BDI scores moved from moderate to mild depression between baseline and first follow-up. However, the results at second follow-up were no longer clinically significant as the participants returned to baseline moderate level (Beck, Steer, & Brown, 1996). In a further UBA study , a statistically significant decrease in the depression scores of participants from the start to the end of the intervention was reported; however, no further follow-ups were reported.
Overall, the studies did not indicate that care farms can improve quality of life for people with mental health problems. Also, the evidence on the effectiveness of care farms to reduce anxiety and depression within mentally unwell service users and those with substance misuse problems is inconsistent and therefore inconclusive.
Proximal outcomes. Two RCTs measured self-efficacy and both found no significant change in self-efficacy, between groups, at 12 week follow-up Pedersen et al., 2012b). However, at 6month follow-up,  found a statistically significant improvement in self-efficacy.
Self-esteem was measured in one UBA studie (Hine et al., 2008b) the authors claim a statistically significant improvement in selfesteem at the end of the intervention, with no further follow-ups reported. A statistically significant reduction in stress was also found at the end of the intervention (12 weeks); however, this effect was not maintained 3 months after the intervention (Gonzalez et al., 2011a(Gonzalez et al., , 2011b. In addition,  reported no significant effect on coping, compared to the control group, at 12 week and 6-month follow-up. Additionally, Gonzalez et al. (2011aGonzalez et al. ( , 2011b) measured positive affect, which is the extent to which participants experienced the following affects: interested, strong, enthusiastic, inspired, proud, alert, strong and active. At 12 week follow-up, there was a statistically significant improvement in positive affect, but this was not maintained 3 months after the intervention.
Social outcomes were measured in two studies. Social functioning (including social engagement, interpersonal communication, independence and competence) was measured in one CBA study and at 12month follow-up, there was no effect on social functioning between the participants that went to care farms compared to participants that attended day activity projects (Elings et al., 2011). Gonzalez et al. (2011a, 2011b) assessed participants' group cohesion using the Therapeutic Factors Inventory Cohesiveness Scale which captured a person's sense of belonging to the group and experience of acceptance, trust, and group cooperation. During the length of the intervention (12 weeks), they found that the participants' group cohesion significantly improved.
One study measured participants' appetite and eating patterns and at 12-month follow-up, found no differences in appetite and eating patterns between service users attending care farms versus those at day activity projects (Elings et al., 2011).
Overall, across all secondary outcomes there is inconsistency in the findings at immediate, 3 months, 6 months and longer-term follow-ups. Most studies measured immediate follow-up with few addressing longer-term impacts. The impact of care farms on psychological, social and physical outcomes in service users with mental health problems or substance misuse problems remains unclear.

Mapping outcomes to the disaffected youth logic model
Three outcomes were reported for disaffected youth both at 6-and 12-month follow-ups . The authors reported a significant positive effect (MD = 1.05) on problem behaviours (i.e., internalising problems, anxiety/depression, being reserved, externalising problems, and delinquent behaviour) at 6-month follow-up.
Four of seven aspects of coping questionnaire showed significant, positive improvements, including: seeking social support, passive expectancy, self-esteem and active problem solving. No difference was found in self-determination at both follow-ups. The evidence on the impact of care farms for disaffected youth is scant.
Evidence for other client groups/mixed groups Lambert (2014) observed a 17.08 points improvement in quality of life as measured by the EQ-5D health state score from baseline to end of the intervention for the mixed client group. However, the author did not report whether this overall score is statistically significant, or provide a standard deviation. Nevertheless, Lambert (2014) conducted subgroup analyses and found statistically significant improvement in quality of life among people with anxiety or depression, personality or social issues, and psychosis, but not for people with learning difficulties.
In a CBA study involving older people, de Bruin (2009) reported no significant change in cognitive functioning at 6-month follow-up between those attending care farms compared to a control group that attended day care facilities.
In a very small UBA study, Marshall and Wakeham (2015) reported a 65% reduction in expected 12-month reoffending rates for offenders attending a CF as part of their community order. become physically healthy. These mechanisms are a good fit with a number of theories, and this review provides the first attempt to map evidence from quantitative and qualitative studies against the concepts of these theories in relation to care farms.
Although we ordered mechanisms based on frequency and spread, we do not suggest that any one mechanism is any more important than any other at an individual level. However, based on available data, we observed potential differences in the way care farms work for particular client groups. While this may reflect differences in the focus of the topics covered in the qualitative methods used by different authors, these differences are worth further exploration. For example, a sense of achievement and satisfaction appeared to be more important to the substance misuse/mental illness and the learning disabilities service users groups compared to the disaffected youth service group, where feeling safe may be a priority. In this latter client group, having the opportunity to reflect seemed to be valued. While we do not have sufficient data to be able to robustly link the intervention components to the mechanisms, we do tentatively suggest that in the disaffected youth group the emphasis on reflection appears to fit with the greater focus on the "setting" aspect of the intervention. As with the causal pathway between intervention components and mechanisms, the relationship between many of the mechanisms and proximal outcomes/outcomes is unclear. For example, "understanding the self" (a mechanism category), which included findings such as increasing self-respect and understanding of tasks that are manageable, could potentially be linked to proximal outcomes relating to self-efficacy and improved confidence. However, with others which were seemingly important mechanisms such as "belonging and nonjudgement", the connection to outcomes is less clear. It is likely that many of these mechanisms interact in a way that is not yet understood to influence outcomes. These hidden features of complex interventions are commonly observed within logic models.
A key finding within this aspect of the review was that the theoretical concept "restorative effects of nature" was represented by the intervention components (but to a notably lesser extent than "the work" and "the farmer" components), but was not represented at all in the categories of mechanisms. This was somewhat surprising given that, informally at least, one of the most lauded attributes of care farming is its nature-based approach. Only four findings of the 164 that mapped to the theoretical concept about mental well-being could potentially relate to nature. We suggest that the absence or near absence of "the restorative effects of nature" is not a true absence; rather, nature is the essential platform which allows other more overt mechanisms to be acted out. Thus, as individuals recall their experiences on the farm, it is primarily the mechanisms Despite being able to develop the logic model for the disabilities client group, the lack of quantitative studies with this group meant that we could not map outcome data to the model. While more quantitative data was available for the substance misuse/mental illness groups and the disaffected youth logic models, very limited mapping of secondary outcomes was possible with the latter group.
Based on limited quantitative evidence from only two small RCTs, we did not find sufficient evidence to conclude any significant positive effects of care farms in improving quality of life. We did find some limited and inconclusive evidence to suggest that care farming can reduce anxiety. For depression, while there appeared to be significant reductions following the intervention as assessed in UBA studies, the RCT found no significant differences between intervention and control groups, however the small sample size may have undermined the power of this study to detect a difference.
For proximal/secondary outcomes, there were no significant positive effects for self-efficacy and coping (measured in the RCTs) at the end of the intervention. However, a significant improvement in self efficacy (but not coping) was reported at follow-up. The possibility that there may be some delayed benefits (as with anxiety) for self-efficacy requires confirmation by future studies. A number of UBAs reported significant improvements in self-esteem, stress, affect, mood and group cohesion at the end of the intervention.
However, only stress and affect were measured at follow-up (3 months after the intervention ended), and improvements were not sustained. Most of the primary and secondary or proximal outcomes were limited to immediately postintervention, with only three (social functioning, eating and appetite, and mental status) reported beyond 6 months. With respect to disaffected youth, there was some suggestion that coping might be improved, but no impact identified on self-esteem.

| Overall completeness and applicability of evidence
Most of the studies were conducted within three European countries, in particular in the Netherlands (n = 12). This was followed by Norway (n = 9) and then the UK (n = 7), with two studies in the United     This was the case with the mental health problems logic model.
While there was a reasonable body of qualitative evidence relating to mechanisms for disaffected youth, findings on proximal or endpoint outcomes were very limited, with only two found. Only one theory (attachment theory) (Bowlby, 1969) was specifically mentioned in relation to adolescents and applied within an excluded overview about animal assisted therapy (Geist, 2011 Likewise, standard good practice of obtaining informed consent and ethical approvals was only reported in six studies. Although not specifically a quality criterion, we observed a clear connection between study quality and the use of contextual theories to guide the research question and analysis. Those that used a theory much more often met more of the quality criteria. Again, provision of basic demographic data (age and gender), which was also not a specific quality criterion, was often absent in studies. Six of the qualitative studies were not published in academic journals and missed the opportunity for rigorous external peer-reviewing. Some were locally commissioned without the intention of publishing in a journal and this may explain the lack of good quality reporting.
In the qualitative studies, the vast majority of themes did not separate the experiences of different client groups.

| Quantitative studies
There was much heterogeneity across the studies in terms of the client groups, duration and intensity of the intervention, outcomes and outcome tools, periods of follow-up and overarching study design; hence, we were unable to conduct a meta-analysis.
Heterogeneity was also observed in the outcomes and measures applied in the quantitative studies. Twenty three different outcomes were measured over 12 studies, probably reflecting the range of client groups and the varied way in which care farms might be considered to impact on lives. Quality was also compromised by the use of unvalidated outcomes within a number of studies. The majority of quantitative studies in general did not offer a theoretical basis or even suggest a mechanism by which the intervention might work, questioning the basis of decisions on types of outcomes.
Most of the quantitative evidence was derived from UBA studies, which do not control for threats to internal validity and thus causal inferences cannot be made from these studies. Furthermore, most of the outcome data were restricted to immediately after the intervention, potentially offering inadequate latency for observed effects.
Only three outcomes were reported at 6 months, and a further three for 12 months and beyond.
All of the quantitative studies had a high risk of bias. In the two RCTs, three and four of the seven quality assessment domains were unclear. Studies did not demonstrate any evidence of bias in the selection of outcomes reported, and all data on attrition was reported. However, neither study blinded outcome measurement, and one of the two studies lacked clarity about potential contamination between the groups and about differences in baseline characteristics. Furthermore, in one study the differences in baseline outcomes were not adjusted for in the analysis.
Similarly all other CBA and UBA studies were found to be at high risk of bias. In particular, only one study reported data on attrition.
As with the qualitative studies, six (one CBA and five UBA) of the 13 quantitative studies (including the mixed methods study in the total) were reports that were not published in a peer-reviewed journal, and therefore were not subjected to the rigors of an external review processes. In general, samples sizes across most of the studies were small and so were likely underpowered, thus increasing the risk of type II error.

| Agreements and disagreements with other studies or reviews
We only know of one other published review that has specifically targeted care farming as an intervention for people with mental health problems , which included five studies, three of which were RCTs. One of the RCTs was excluded from our review because the intervention was horticulture therapy delivered by a health care professional, rather than therapeutic horticulture delivered by a care farmer . The other UBA study (Cerino, Cirulli, Chiarotti, & Seripa, 2011) was not found by our search, but would not have met our eligibility criteria, being a single activity (therapeutic horse riding). Overall, for the included studies, the reviews are in agreement in so far as quality, scope of outcomes and findings. We agree with Iancu's (2013) view that care farming as a work-based intervention should be evaluated as a form of vocational rehabilitation, and yet as a robust measure this is lacking from the studies. Iancu (2013) also found three key qualitative themes from three studies relating to disability (distraction, stress release and participation), recovery (viewing the self differently and being socially included), and specific farm experiences (absorption in work and connecting with nature). Our synthesis was more in-depth and involved more studies but we did find the themes to which Iancu (2013) refers.
Other reviews (one systematic and the other a simple literature review) with a broader nature-based remit (Annerstedt & Währborg, 2011;Bragg & Atkins, 2016), and also with a narrow but overlapping focus on conservation or horticulture therapy and gardening, exist (Clatworthy, Hinds, & Camic, 2013;Kamioka et al., 2014;Lovell, Husk, Cooper, Stahl-Timmins, & Garside, 2015). One of the broader reviews, involving 38 papers, included nature-assisted interventions, wilderness and horticulture therapies, but not care farms, and focused on a wide range of vulnerable groups (Annerstedt & Währborg, 2011), but mostly related to disaffected youth, and those with mental health problems or dementia. The main difference here is the application of a "therapy", implying the delivery of an intervention by a professional (often health-based), rather than offering an intervention that is "therapeutic", as is the case with our review. Some of the studies also included an additional therapeutic component such as psychotherapy or cooking activities, mostly for participants with addiction problems. The contribution of the nature element in these interventions is unclear. As with our review, the authors found that the quality of the studies was mostly low, with often small sample sizes and short term follow-up (at the end of the intervention). However, most studies reported finding positive outcomes, and the authors conclude that there is a small body of evidence to support the use of nature-assisted therapies for a range of conditions and social circumstances. The second broad literature review looked at social and therapeutic horticulture, care farming and environmental conservation (Bragg & Atkins, 2016). These interventions were separately covered by the other reviews so are not discussed here. The systematic review on conservation involved volunteers, so did not specifically address impacts of nature-based interventions on vulnerable populations. The review on horticulture therapy (an intervention that can be included within care farming) included four RCTs involving people with dementia, severe mental illness such as schizophrenia, bipolar disorder, and major depression, as well as frail elderly people in nursing homes and hemiplegic patients after stroke. As with all the reviews reported here, including our own, meta-analysis was not possible due to heterogeneity in outcomes and across the interventions. Again, the studies were found to be of low quality, but overall there was evidence of effectiveness for improved mental health and behavioural outcomes.

| Implications for practice and policy
By far the most studied client group in care farming research is people with mental health problems. In the UK currently, there are more care farms providing support for people with learning difficulties (93% of farms) and ASD (84% of farms) than there are for those with mental health problems (75%) (Bragg et al., 2014).
However, only four of the 18 qualitative studies explored the experience of care farming for learning disabilities and autistic spectrum disorder. Similarly, disaffected youth who are supported by around 64% of UK care farms (Bragg et al., 2014) were again the focus of only four studies, with two being quantitative.
Reasons for the intense research interest in mental health problems above other client groups likely reflect a growing concern about increasing mental health problems in modern society (Murray et al., 2012), a lack of choice and availability of treatment options (MIND, 2013) and the impact on the economy through benefit support, absenteeism and unemployment (Centre for Mental Health, 2010). Although the use of nature to support recovery from a range of mental health conditions is not new, the way it is used has evolved over time. Once an adjunct to institutional psychiatric care, it has become part of a community-based multifunctional "green care" service. However, the evidence for nature as a mental health "treatment option" has not evolved at the same rate as for other more medical approaches. Only recently, through the application of social prescribing, have health care providers and commissioners started to translate the longstanding knowledge that many mental health problems are underpinned by social circumstance (Marmot et al., 2010) and begun to commission services that provide social interventions (CRD, 2015).
Yet even within this approach, green care services are used relatively infrequently when compared to traditional approaches (Bragg & Leck, 2016). Given that, in the UK at least, care farms are underutilized relative to the spaces available on the farms' structured programmes (Bragg et al., 2014), lack of capacity across the broader green care service is not the issue. Lack of access may contribute specifically within more urban areas with fewer green spaces, higher deprivation and lack of transport. Lack of understanding and awareness is however likely to be a major factor. In countries such as Norway, Sweden and the Netherlands, where care farming is wellestablished and research is most active, there is greater integration with statutory services (Elsen and Finuola, 2013). In the North and Republic of Ireland there has been an active push to market care farming directly to commissioners combined with the establishment of a network of farms supported by EU funding (Social Farming Across Borders, 2015), and this could be an option in areas where engagement has been low. In addition, in other countries access to care farming has been written into their constitution (https://www. cliclavoro.gov.it/Normative/Legge_18_agosto_2015 n.141).
However, the need to communicate how care farms work and who they are appropriate for is needed in the UK, where healthcare commissioners lack awareness and understanding about care farming and who might benefit (Bragg, Egginton-Metters, Leck, & Wood, 2015), but this is just one side of the problem. In addition to securing funding through commissioners, there is the dual task of communicating directly to frontline providers, specifically primary care staff and social prescribing facilitators, who have the role of identifying interventions for patients with complex social needs that present as mental health problems. Here, the skill is matching needs to service response, and while some interventions have a clear fit (e.g., debt services, housing support and relationship counselling), others, particularly care farming, may be more challenging to place. There is also a lack of awareness and understanding from patients as to the potential benefits of green care, including care farms, and so as a client-led approach, green interventions may not be a considered an option. Having developed a theoretical framework and a set of logic models to describe potential mechanisms behind care farming, we now have a basis upon which to inform health and social care commissioners how care farms may work theoretically and for whom they might be suitable.
The studies included in the review had twice as many male as female participants. Reports on the care farming sector in the UK (Bragg et al., 2014) indicate that this is a reflection of the use of care farms by men and women. This preference for care farming by men is of interest to commissioners of mental health services. There is a gender inequality in utilisation of mental health services, where usage is much higher among females (Health and Social Care Information Centre, 2014). It may be that green care interventions are perceived by men to be a less intrusive, and therefore a more acceptable form of support. Findings from the qualitative studies included in this review would seem to support this, where service users refer to the benefits of not forcing early social interactions, where conversations centre on work rather than illness, and where distraction is welcome. This may mean that care farms may be preferred by those wanting less intense personal interventions to improve their mental health.
In regards to "treatment" costs and duration of the intervention, studies included in the current review suggested an intervention duration (averaging around 12 weeks) that is representative of practice and comparable with talking therapies. Although not considered in the current review, the costs of talking therapies are also not dissimilar to care farming (Bragg et al., 2014;MIND, 2013).
There is a need to identify a wider range of interventions to address mental ill-health and allow tailoring to individuals' personal treatment needs. Providing a greater range of intervention options, such as care farming, would provide choice where there is currently little on offer and has the potential to reduce waiting lists for talking therapies (MIND, 2013). Furthermore, it could help redress gender inequalities in terms of accessing support for mental health problems.
Further studies are needed to explore the effectiveness of alternative mental health interventions, such as care farming, with exploration of who they may work for and how.
For the other client groups, the implications for policy and practice from this review are limited. Disaffected youth, particularly those at risk of exclusion from school, potentially represent those most likely to offend, are more likely to have future mental and physical health problems and fewer employment prospects (Parker et al., 2014). Care farms could potentially offer an alternative form of education with qualified educators supporting the delivery of qualifications such as Open College Network Qualifications (Bragg et al., 2014). While this review did not search specifically for CF studies with educational outcomes, in those studies included here none had measured educational outcomes alongside health outcomes. Understanding the impacts on young people's education, behaviour and any inter-relationships with health would be a valuable future area for study.
The European studies included in the review indicate that systems appear to be in place that allow people with learning disabilities to access green care where it is wanted, or where it is accessible, with funding often provided through local authority personal budgets. It is interesting that people with learning disabilities is the largest client group attending care farms in the UK, but the question of benefit accrued has not been explored in great depth. It is unclear whether individual carers who are in pursuit of support actively seek out organisations such as care farms or whether local authorities are more informed about services available in the community. Regardless, there appears to be a working mechanism that enables those with learning disabilities to have the opportunity to benefit socially and physically from farm work, and this seems to be supported by the qualitative literature.
The most recent patient group to engage with care farms is people with dementia. Although we found little research, we are aware of a number of programmes throughout the UK that are starting to engage people with dementia in nature-based activities. initiative in London has also recently been involved in piloting a scheme to deliver animal handling sessions in an attempt to reduce social isolation in older people and in those with dementia (http:// furry-tales.org.uk/). There are also opportunities within social prescribing schemes to refer older people experiencing social isolation and those with perhaps the earlier stages of dementia to attend care farms, but as with mental health problems, the benefits are yet to be demonstrated.

| Implications for research
Contextual descriptions revealed a wide range of activities provided for service users on care farms (see Table 4); however, there was insufficient information to establish whether effects differed according to these. Information was not sufficiently detailed to allow us to determine client specific activities, although logic dictates that some more vulnerable and less independent service user groups are less likely to be involved in heavy traditional farming activities that contribute to productivity. Knowledge about this is important for helping to understand the ways in which care farming might work for different client groups; this is clearly of value to commissioners and other funders of care farms. We know from the qualitative studies that there might be some differences in the intervention components as interpreted by the service users and that there may be differences in the mechanisms of change, but because many studies include mixed client groups and failed to report separate themes, we have limited information.
Care farming research has become an active field in recent years; however, well designed studies are still lacking. There is some evidence, albeit inconsistent, that as a theoretically underpinned intervention, care farming might improve mental health outcomes.
The need for a robust evidence base seems most urgent in the mental health field where there is growing concern about the increasing individual and economic burden that mental illness imposes and the limited range of interventions available (Centre for Mental Health, 2010). To progress the evidence, the quality of the research needs to improve.
Our review highlighted how different population groups experience and may benefit from care farms differently. Going forward, research studies should collate data on single population groups so as to provide answers to health and social care commissioners who tend to commission services for specific client groups. We recognise that for care farms, working with only one single population group or not combining groups in activities may be challenging and impractical.
However, research can be designed to build the evidence base relevant to different population groups. Evidence on the impact on health is particularly important to the care farming sector as well as health commissioners. Often situated in the third sector, care farms balance income from a range of sources, including grants from charities and private organisations, revenue from selling farm produce, but an important source of income for many care farms in Europe is through funding from public health and social care. Thus demonstrating their contribution to health and social outcomes to secure one of their potentially long term funding sources is important.
One of the aims of this review was to understand how care farming worked for these different client groups. We have observed some differences across the groups with "achievement and satisfaction" and "feeling safe" being potentially more or less important in some groups compared to others. How these convert or contribute to outcomes is unclear, and indeed the general conversion of mechanisms to outcomes is an invisible part of all logic models. What we can glean from these logic models is a sense of which outcomes might be most appropriate for which client group. The mental illness/ substance misuse logic model provided the most obvious path from theory to mechanisms and then to outcomes. However, vocational rehabilitation was not adequately addressed and only "work ability" (Lambert, 2014) was measured, but without adequate clarity about its reliability. Returning to work/taking up work could offer important individual financial and well-being gains, but also, from an economic perspective, can potentially reduce the burden on society from a reduction in health service utilisation and benefits; however, included studies lacked data on these outcomes. This is an area in which commissioners are becoming increasingly interested, so care farming research needs to demonstrate its impact more broadly.
More reliable and objective proxy measures for returning to work would be of interest. In addition to broadening its impact in line with anticipated outcomes that fit with explanatory theories, longer-term follow-ups beyond 6 months are required. There was some indication that positive outcomes, such as improvements in anxiety and selfefficacy, may take time to manifest, but this needs to be confirmed.
For disaffected youth, the path from theory to outcomes was not followed, as measured outcomes did not adequately fit with the model. We would suggest that care farming interventions involving disaffected youth use these models to determine the most appropriate outcomes.
The disaffected youth client group was the only one to report findings relating to "reflection". Children at risk of exclusion from school are at high risk of entering into an adult criminal lifestyle (Audit Commission, 2010), and desistance theory suggests that a period of reflection is a critical early step in the rehabilitation of offenders (Cusson & Pinsonneault, 1986;Farrall & Bowling, 1999), but only if it is supported with interventions that take them beyond this. In this respect, care farming may have the capacity to rehabilitate young people who are at risk of committing offences later in life. In line with this, the other category of mechanism that was present in this client group but not the mental health problems group was "creating a new identity" which again fits with desistance theory. This category was also found in the learning disabilities group and related more to how this client group envisaged themselves as a farmer.
Studies included in the current review used a wide range of measures and concurs with the findings from a previous review of care farming interventions . In an area of research where individual studies tend to be underpowered, there is a greater need to be able to combine findings in a meta-analysis. In the current review, the most commonly applied mental health outcome measures were the Beck Depression Inventory (Beck et al., 1996) and the State-Trait Anxiety Inventory (Spielberger, 1983), both of which appear to be acceptable to the population group. More fundamentally, this review identified a number of small scale evaluations which used tools that had not undergone psychometric evaluation. We would suggest that researchers select existing reliable and validated tools.
Adopting robust study designs must be matched with capacity to undertake the research, and this is where care farming studies may need to compromise. A lack of service infrastructure across the care farming sector and peripheral relationships with statutory services means that methodically robust large RCTs are very difficult to perform, particularly where income for the intervention is not guaranteed and single client groups at individual farms are quite small in number. In the absence of available studies where data can be combined, larger studies that involve multiple care farms, possibly operating in a network, are an option. These would ideally require agreed standardised criteria for referrals across multiple healthcare organisations.
In general, we recommend that a more cohesive approach to care farming research be adopted. This means understanding the needs of commissioners and thinking beyond individual CF research studies.
Green care has potentially much to offer, but currently cannot prove its worth until more robust methodologies and strategically aligned research are conducted.