The framework has developed through our research into breastfeeding support and weight management groups, which included systematic reviews, mixed method group interventions and ethnographic studies 2 16 . Studies included a randomised controlled trial of a policy to provide breastfeeding groups across Scotland 17 with a mixed method evaluation of implementation processes 18 ; a controlled intervention study of individual and group peer support for breastfeeding 19 20 and an ethnographic study of participant and provider experiences of weight management groups 21 . All of these studies conducted in-depth qualitative interviews with participants and providers and group observations.

In addition, our studies and this paper were informed by a literature review of health improvement interventions for patient groups. We searched Medline, CINAHL, Embase, The Cochrane Library, National Institute for Health and Clinical Excellence (NICE), Psycinfo and the International Bibliography of the Social Sciences using group$, class$, club$, workshop$ and program$ as title words for papers published between 1980 until 2009. This was a pragmatic search strategy to limit the number of papers identified and was necessary because of the widespread use of the word "group" in differing contexts in research papers. We identified systematic reviews in relevant areas, for example smoking and alcohol cessation; chronic diseases like diabetes, cancer support and heart disease and searched their reference lists. We also searched our personal reference archives and hand searched references in key papers. We excluded educational groups for teaching students or staff; organisational and management literature on work groups and teams, and treatment groups in mental health, where there is an extensive literature.

The framework evolved through an iterative process of mixed method data analysis, reference to the literature, reflection and research team discussion over a period of 6 years. It builds on reviews of group processes 9 , small group work in education and work settings 22 and self-help or support groups 23 .

The evidence for health improvement interventions in group settings is varied, reflecting their heterogeneity and complexity and some argue that policy for group interventions to encourage self care has raced ahead of the evidence 24 . Systematic reviews often focus on an individual disease, the type of treatment, or behaviour change theory and they often inappropriately combine the results of interventions delivered individually and in group settings in a meta-analysis 13 . For example, a Cochrane review of additional support for breastfeeding differentiates between whether the support is lay or professional, but not between interventions delivered to individuals or groups 25 . In a systematic review of smoking cessation groups, theories of behaviour change validated for individual delivery, for example stages of change or cognitive behavioural theory, have been assumed to be transferable to delivery in a group setting 1 . This systematic review of smoking cessation interventions 1 is one of few which has specifically analysed outcomes for interventions in group settings and compares them with self-help materials, individual counselling, nicotine replacement or no intervention. It concludes that group programmes are more effective than no intervention or self help interventions, but there is insufficient evidence to evaluate whether groups are more effective than individual counselling. Another systematic review comparing group versus individually delivered interventions for weight management in adult obesity found that group interventions report significantly more weight loss at 1 year follow-up compared with the same intervention individually delivered 2 . A systematic review of group education programmes for adults with type 2 diabetes compared with routine treatment, waiting list control or no intervention, found group programmes were effective at lowering glycated haemoglobin, fasting blood glucose, blood pressure, weight and medication use 4 . For cardiac rehabilitation, interventions delivered individually at home were as effective as those delivered in a centre to a group 6 .

The aims of a group intervention, the underlying behavioural change theories and group processes can determine who attends and both the group and the individual health outcomes. A one size fits all approach seldom meets everyone's needs and it has been argued that for self care in chronic disease, groups should only be considered for simple standardised messages, where peer support is beneficial or preferred and where a group will save money 24 . Importantly, some groups may increase health inequalities by attracting more educated and higher income participants 17 26 27 . It has been suggested that support groups are more likely to be sought for diseases viewed as stigmatizing, like AIDS, alcoholism, breast and prostate cancer rather than less stigmatizing, yet important, diseases like heart disease 28 . There are inconsistencies in self care group definitions, for example in a survey of American self care groups 60% of those identified had professional facilitators and the clinical potential for social support in combination with professional guidance compared to peer only support is largely unknown 28 .

The potential for cost savings with group compared to individual health improvement interventions can appear attractive; however the health economic evidence is mixed and often weak. In a review of individual or group self care interventions, there was insufficient evidence to support the claim that group interventions are cheaper and there are often trade-offs between the numbers of patients treated and the quantity of intervention each individual receives 24 . Smoking cessation groups are no more cost effective than intensive individual counseling 1 and breastfeeding groups provided as part of routine care would have similar costs to individual health visitor home visits to group participants 17 . Group interventions for weight management in obesity are potentially more resource saving in terms of total health professional-hours involved per participant 2 . Few self care group interventions for arthritis have measured service use and findings are conflicting 29 30 .

In the literature discussed above, it is clear that the jury is still out when it comes to deciding whether group or individual interventions are more effective and cost-effective at improving health outcomes. Group processes and interactions have received less attention than dyadic or individual behaviour change mechanisms and little is known about which components of groups contribute to effectiveness. As a result there are many unanswered questions for practitioners and policy makers who aim to establish patient groups to improve health outcomes. The NICE behaviour change guidelines recommend being as specific as possible about the content of the intervention, spelling out what is done, to whom and in what social and economic context 10 . A challenge with groups is to unpick the extent to which outcomes are determined by leadership style, personality, participants' characteristics or more complex interactions. Having a framework for design and process evaluation is one step towards producing this evidence and we describe this below.

The first and crucial stage in designing and evaluating a group intervention is to consider how aspects of the setting will impact on all aspects of group processes, composition and outcomes (Table 1). In a trial which randomised primary care organisations to deliver a policy to provide breastfeeding support groups, the breastfeeding outcomes were explained by the characteristics of the primary care organisation, including health inequalities and deprivation, the amount of organisational change taking place and multi-disciplinary teamwork 18 . The breastfeeding outcomes could not be explained by the amount of intervention delivered or the number of people attending the groups.

Evaluating how different environmental settings influence behaviour has been relatively neglected in research compared to the predominant approach of measuring the psychological variables of individuals receiving a behaviour change intervention. Theorising interventions as events in complex systems is particularly relevant to health improvement groups, where the context, the change in relationships and resources over time are important 31 . The importance of the behavioural setting 32 or the activity setting 33 as a unit of analysis is a cross disciplinary feature of psychological, social and biological ecology 32 34 . Behavioural or activity settings refer to time and space bounded patterns of behaviour, and the concept originates from research which characterised all behaviour settings in a small American community 32 . This demonstrated that behavioural settings have effects on individuals that extend beyond the specific behaviour demanded by the activity.

It can be helpful to consider the context in which the group intervention takes place at macro, meso and micro levels, as proposed in ecological models of health improvement 35 . The macro level considers how wider policy, economic and socio-cultural factors interact with the group, for example media scares or the cultural values of participants. The meso level includes the interrelations between the group, the setting and the surrounding environment, for example holding the group in an affluent or disadvantaged area can influence perceptions and who attends a group 18 . Importantly, the impact of parallel activity settings 33 which may target the same population should be considered. For example, the impact that general parent craft groups open to all women can have on participation in breastfeeding only groups 18 . Similarly, consideration should be given to assessing the impact of activities displaced by the group intervention 31 . What providers and participants stop doing when a health improvement group starts is likely to be crucial to outcomes, yet this is seldom systematically described or evaluated. The micro level is the interaction between group participants and the space where the group meets. Community settings like church halls or family centres convey different meanings for people compared to health settings and health service group leaders may behave differently outside health service settings 18 . Group resources can vary between purposely designed venues with state of the art props to "make do" multi-purpose clinically cluttered spaces. Consideration should also be given to sensory perceptions including comfort, temperature, noise, smell and visual appearance 20 .

Table 2 focuses on the design of the intervention, the theory underlying the choice of intervention, the target population and choosing the relevant behavioural outcome to measure.

Table 3 proposes a series of questions examining the micro-level of interactions between the group leader, new and existing group members and the group setting. It is closely related to the ecological, social and behavioural theories described above. Group processes and their interactions with the group setting can determine group survival and the intended outcomes, but they have been under-researched in the context of health improvement groups and the relative importance of individual components is unknown. Group Environment Scales (GES) have been developed to systematically measure the norms, values and psychosocial characteristics of social environments 47 . The underlying theory is that environments like people have unique personalities, climates or atmospheres which are important determinants of behaviour. GES have three basic dimensions: relationship dimensions which assess the extent to which individuals are involved, for example cohesion, support; personal development dimensions, for example autonomy, goal attainment; system maintenance or system change dimensions for example order, organisation, clarity, control. GES have been developed in the field of mental health treatment, education and work and little is known about their generalisablity to health improvement groups. They have been used to compare self-help groups 23 and these three dimensions seem salient in a wide variety of social settings, although the characteristics of each dimension may vary 34 48 . The dimensions we use in Table 3 are adapted from GES to provide a better fit with our data. In particular the personal development dimension seems more relevant to educational or treatment groups and we have reframed this as attributes of the leader and the group participants. As groups are complex systems there is some overlap and interaction between dimensions.

We have used the term "group leader" to embrace a range of observed styles which others have defined 9 22 . Variation in the delivery skills of group leaders can theoretically determine the success of the intervention and it is important to consider the statistical effects of such clustering in the design 13 . Education, support and behavioural interventions in a group setting are characterised by different communication styles. For example, a group leader may be an advocate of a method, philosophy, activity, seller/deliverer of a "product" or a facilitator of person centred approaches, for example empowerment, counselling or support 49 . How leaders with personal experience of a condition, for example in commercial weight management groups 21 and the expert patient programme 7 , influence group outcomes is largely unknown. Some groups are multifaceted with communication interspersed with activities like relaxation, physical activity, weighing or physiological monitoring. Health service training in group leadership is variable depending on the type of group and the nature of the intervention. For example, exercise groups have standardised qualifications for leaders, whereas in some primary care groups for weight loss, practice nurses are self-taught and learn on the job 21 .

The NICE behaviour change guidelines 10 recommend that community interventions should: improve self-efficacy; develop and maintain supportive social networks; promote resilience; build skills; promote participation in voluntary activities; promote involvement in planning and delivery of services and have access to the financial and material resources needed to facilitate behaviour change. Of fundamental importance is having a clearly stated purpose with aims and objectives that are agreed by everyone. However, even with clear aims and objectives, differing interpretations will arise. For example, in a trial of breastfeeding support groups, leaders interpreted the word "support" in different ways. Despite a clear protocol stating that support should be woman-centred and based on informed choice, some interpreted "support" as promoting and encouraging exclusive breastfeeding as you might support a football team 16 . Some experts classify groups into either task orientated groups, where individuals or the group have specified goals, or socio-emotional groups where feelings and inter-personal relationships are paramount 9 . However, our group observations suggest that this binary classification may not apply to the complexity of health improvement interventions, which usually combine task and socio-emotional objectives 16 21 .

A study of cancer support groups illustrates how an analysis of group processes can provide important evidence and highlights the difficulties with pre-selecting group composition 50 . Education and discussion groups combining distressed and undistressed cancer patients were compared with a group of distressed patients. The heterogeneous group increased the social comparison opportunities for distressed patients and benefited them. However, undistressed women with breast cancer who had high levels of social support showed a slight deterioration in physical functioning when attending the heterogeneous group, which raised ethical concerns. The authors suggest training undistressed participants to maximise their benefit for distressed participants but to minimise negative consequences for themselves.

Individuals tend to conform to the attitudes and behaviours of the majority within a group, and cliques can develop. For example, some breastfeeding groups can be dominated by mothers breastfeeding older infants, which may be off-putting to new mothers joining, but an over-all "feel-good atmosphere" can override personal differences 20 . Studies suggest that conformity and uniformity in groups increases with group size and over time and groups tend to exhibit more extreme attitudes, beliefs and behaviours than the individual group participants 9 .