Impact of a social prescribing intervention in North East England on adults with type 2 diabetes: the SPRING_NE multimethod study

How is social prescribing defined and operationalised in the UK?

Despite this rapidly emerging policy context, there remains no agreed single definition of social prescribing within the UK or internationally; however, there is a broad consensus that social prescribing helps patients access non-clinical sources of support, predominantly in the VCSE sector, but also draws on resources supplied by health and local government services. To enable this, front-line health staff, primarily but not solely in primary care, refer patients to a link worker. Once the patient is connected to a link worker, which pre COVID-19 was predominantly face to face, a conversation takes place to identify problems and goals, thereby ‘co-producing a social prescription’ designed to find solutions that will lead to improved health and well-being. 11

Currently within UK practice there is no single model of social prescribing. Rather, social prescribing models cover a spectrum12 that varies according to whether programmes focus on individuals, communities or a combination of both; the level of support offered to patients following referral; and the actual activities offered. Usefully, Husk et al. 4 identified multiple pathways from primary care to activities undertaken and reported that current policy supports link worker-based models. Recognising that patients who are simply given information about a service will not necessarily take it up,13 in most schemes the link worker acts as a ‘facilitator’ and offers personal support, although the level of ongoing support varies considerably. Services into which patients are referred vary and can include clubs offering physical activities, such as gyms, walking groups, gardening clubs and dance clubs, and those offering weight management and healthy eating activities, such as cooking clubs. Addressing wider economic and social issues can involve referral into services that address welfare, debt, housing and employment issues. Support and self-help groups, such as those targeted at people with specific LTCs, for example chronic obstructive pulmonary disease (COPD) or diabetes, may also be accessed via social prescription. 1

The link worker role is central to social prescribing as currently practised in the UK1,14,15 and elsewhere. 16,17 A 2019 survey of UK CCGs reported 75 different terms for this ‘connector’ role,18 and a recent scoping review identified 18 separate terms that described this position. 19 The link worker role is recognised to be ‘complex and demanding’15 and requires a wide range of skills. 20 Executing the link worker role requires time, interpersonal skills and strong community networks and relies on the existence of onward referral services. 15 There is no clear professional pathway into the link worker role, and link workers come from a range of backgrounds, including the social work, teaching, health-care and VCSE sectors,21 which is likely to affect how they perceive and carry out their role. Previous life and work experience are regarded as important;20 link workers employed via the NHS are required to have a National Vocational Qualification (NVQ) at level 3 or above. 13 The relatively limited peer-reviewed literature on the role indicates that link workers can feel inadequately trained to deal with highly vulnerable clients with complex issues that involve time-consuming case management; therefore, link workers can lack capacity to meet the needs of large numbers of referrals and can also face problems related to service cuts or unacceptably long waiting lists for onward referrals. 20,22 There is now a professional membership network for UK link workers23 and, although link working is recognised as a rewarding role,21 lack of clinical supervision and support is commonly reported among link workers based in primary care and is cited as a reason for leaving. 24

Introducing new services into primary care, irrespective of how appropriate they may be, is not straightforward and requires belief in the benefits and ‘buy-in’ from those expected to refer to the service. 14,25,26 Moreover, referrals to PCN link workers can come from hospitals, allied health workers, the police, the fire service, job centres, social services and the VCSE sector, among others. 13 The role of the general practitioner (GP) and other members of the primary care team in relation to social prescribing is not well understood,27 given that overall a relatively small number of GPs have been included in studies about social prescribing26–30 and even fewer studies include other members of the primary care team. 31,32 Enabling linkage between primary care and social prescribing relies on procedures that facilitate appropriate referrals of patients who meet the social prescribing service criteria. This requires co-operation and trust between members of the primary care team and the social prescribing service(s). 32 A recent qualitative study of the barriers to, and facilitators of, social prescribing for patients with mental health problems based on the views of 17 GPs found that most GPs in the study were supportive of social prescribing and reported that it enabled their patients to be better connected to their local VCSE organisations. 27 Link workers were seen to have more time with clients than the GP and, therefore, were able to act as a bridge between the GP and the community. This contrasts with previous findings from interviews with GPs (n = 3) and district nurses (n = 8) in the west of Scotland, who saw their focus as predominantly clinical and expressed concerns about being held accountable for the actions of unknown or unverified organisations that their patients were linked into. 32 In an interesting overlap between the views of GPs,27 link workers33 and the wider VSCE sector,22 capacity, resources and precarity of the VCSE organisation in the context of austerity were seen as significant challenges to social prescribing. 25

Social prescribing has been the topic of a number of editorials in the medical press, with some arguing the case strongly for social prescribing34 and others being more circumspect. 35,36 Concern has also been raised that the use of the term ‘prescribing’ further medicalises problems that require non-medical support. 37,38 Most apparent is unease about the current level of evidence for the effectiveness of social prescribing on health and well-being outcomes and resource use. 35–37,39 It would appear that there is a strong sense of the potential for social prescribing to improve patient well-being and an awareness that conducting robust research on social prescribing is challenging,40 but that obtaining this evidence is necessary for health professionals to more fully engage with social prescribing. 39

Quantitative evaluations of the impact of social prescribing

Bickerdike et al. ’s41 systematic review of 15 social prescribing evaluations, published in 2017, included evaluations published between 2000 and January 2016 in which the primary outcome of interest was any measure of health and well-being and/or usage of health services. The authors concluded that there was:

. . . little convincing evidence for either effectiveness or value for money . . . most evaluations are small scale and limited by poor design and reporting . . . common design weaknesses include a lack of comparators, loss to follow-up, short follow-up durations and lack of standardised and validated measuring tools . . . a distinct failure to consider and/or adjust for potential confounding factors, undermining the ability to attribute any reported positive outcomes to the intervention (or indeed interventions) received.

Reproduced with permission from Bickerdike et al. 41 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by/4.0/

A subsequent evidence synthesis by Public Health England,42 which included eight studies, examined the effectiveness of social prescribing on (1) contact with primary health care and (2) changes in physical and/or mental health. It concluded that there was no clear evidence for effectiveness.

Although the number of peer-reviewed publications on social prescribing has increased considerably since 2017,19 the number of UK peer-reviewed studies using quantitative health-related outcome measures is still relatively small. Focusing on the UK, between the end of Bickerdike et al. ’s41 review period in 2016 and December 2020, we used similar search terms (see Report Supplementary Material 1). We identified 10 peer-reviewed studies that had quantitative outcome measures (five quantitative and five mixed methods) and evaluated the impact of nine UK social prescribing initiatives on clients’ health, well-being or resource use. Our focus was on peer-reviewed studies; however, we did not include a ‘museums on prescription’ intervention43 because this did not involve a link worker-type facilitator and, because of our focus on social prescribing in primary care, we also excluded a social prescribing initiative in secondary care mental health services. 7 The shared characteristic of the social prescribing interventions evaluated (see Appendix 1, Table 36) is that they link primary care and local assets and services via a link worker, connector or facilitator. Thereafter, the similarities end and each study evaluated a different model of social prescribing, funded via different sources and delivered by a different type of ‘facilitator’, including paid staff and a combination of paid staff and volunteers. 44,45 The interventions varied considerably in ‘intensity’, for example six sessions,46 12-week programmes5,45,47,48 or no limits on time with a community links practitioner (CLP). 49 The intervention target populations varied: they were predominantly aimed at adults aged ≥ 18 years, although one intervention was aimed at those aged ≥ 14 years. 46 Most interventions were targeted at people in mid to later life living in areas of high socioeconomic deprivation and experiencing LTCs, multimorbidity, mental health problems, loneliness and social isolation; these people typically have frequent attendance in primary care and polypharmacy.

The specific aims of the evaluated interventions varied. They include supporting people with complex needs and living in deprived inner-city localities;49,50 improving patient well-being, increasing personal self-efficacy, reducing primary care use/health and social care use, addressing polypharmacy;5,44,48 reducing social isolation/loneliness;45,46,51 improving mental health;48,52 and increasing physical activity. 47 A range of study designs and outcome measures were used. Most studies used a before-and-after design. 5,45–48,50–52 Two studies included matched comparison groups: a before-and-after postal questionnaire study design44 and a quasi-experimental cluster-randomised controlled trial (RCT) design. 49 The latter study49 is the most methodologically robust study to date.

Outcome measures included depression and anxiety,44,50 energy expenditure,47 physical activity,49 general/subjective well-being,44,46,48,50,52 loneliness,45 social support,51 social networks46 and health-related quality of life (HRQoL) [as measured using the EuroQol-5 Dimensions, five-level version (EQ-5D-5L)]. 46,49,51 However, in some studies it was not clear what the primary outcome was. Three studies examined changes in primary health-care resource use:5,44,48 one examined self-reported GP usage,46 two studies examined accident and emergency (A&E) use44,48 and one study examined change in repeat medication. 5

The two studies with control groups44,49 found no statistically significant differences in patient-reported outcomes between baseline and 8 months44 or in health-related quality of life (as measured using the EQ-5D-5L) between baseline and 9 months. 49 However, in the latter study,49 some improvements were found in those patients who engaged with the programme as opposed to those simply being referred. 49,53 Outcomes for the uncontrolled studies showed a varying picture, with high rates of attrition and relatively small sample sizes at follow-up being common. In one study,48 statistically significant improvements in measures of health, well-being, patient activation and frailty were identified at the 12-month follow-up (n = 86, constituting 57% of the sample). 48 In another study,52 a statistically significant improvement in subjective well-being was reported from baseline to post intervention. However, this study suffered from high rates of attrition, with 87.8% of patients either lost to follow-up or not engaging with the intervention. 52 Statistically significant improvements in subjective well-being, social networks and EuroQol-5 Dimensions (EQ-5D) overall health rating were reported in another study. 46 However, the number of clients in this study completing the baseline and post-intervention assessments (n = 265) was considerably smaller than the 2250–3750 users reported as being in contact with the service. Furthermore, these results included only individuals who had engaged with the full six sessions. In a national study in which the primary outcome was loneliness,45 over 70% of clients (n = 1634) reported reduced loneliness (p < 0.0001) immediately post intervention; however, 3-month follow-up data from a subsample of users (n = 101) showed that 60% were experiencing increased loneliness. 45 With regard to primary care use and polypharmacy,5 prescribed medication44 and self-reported health-care utilisation,46,49 no differences were found in these measures from baseline to post intervention. However, a statistically significant decrease in the median number of GP consultations was identified,44 although the authors did not rule out regression to the mean. Overall, this body of peer-reviewed research did not robustly demonstrate beneficial effects on quantifiable health outcome measures or health-care resource use, reinforcing Bickerdike et al. ’s41 systematic review conclusions. However, the challenges in identifying which outcomes to measure and in which ways should not be underestimated. 39

Qualitative evaluations of the impact of social prescribing

Qualitative research with patients who accept their referral and engage with services generally indicates positive evaluations of social prescribing services. 1,53,54 In three of the previously mentioned studies with quantitative components,5,44,49 linked qualitative data about the perceived impact of social prescribing by clients indicated positive narratives that were not reflected in the quantitative data. Of these linked studies, the most detailed qualitative analysis identified considerable variation in the perceived impact of the Glasgow ‘Deep-End’ Link Worker Programme, ranging from no change to moderate/major improvement. 53 Twelve adults aged 34–67 years with a combination of social, psychological and physical problems were interviewed. Some saw the CLP as a catalyst for behaviours and activities that improved their well-being. In contrast, others reported that the CLP had provided practical support, which improved psychological well-being but had no effect on health-related behaviours, or that the intervention resulted in no benefits. 53 Feeling understood and developing a supportive relationship with the CLP were identified as important, but not on their own sufficient to facilitate improvements. A further requirement for improvement was the development of ‘wider social relationships and developing competence to engage in the wider community’. 53

A qualitative study of 30 patients aged 40–74 years with LTCs who engaged with a social prescribing intervention identified that most patients experienced multimorbidity combined with mental health problems, low self-confidence and social isolation, and all were adversely affected by their health problems. 1 The social prescribing intervention, with which it was possible to be engaged for up to 3.5 years, engendered feelings of control, self-confidence and reduced social isolation; had a positive effect on health-related behaviours and LTC management; and improved mental health. 1 A qualitative follow-up study54 of 24 individuals, who had been with the social prescribing intervention for between 12 and 24 months, found that participants reported improvements in condition management and health-related behaviours and reduced social isolation. However, setbacks related to multimorbidity, family circumstances and social, economic or cultural factors were also identified, which highlighted the importance of long-term support and link worker continuity. 54 As has been reported elsewhere,25,53,55,56 the availability of a supportive and easily accessible link worker able to deliver a personalised service that reflect individual goal-setting priorities, enable gradual change and deal with issues beyond health was both positively appraised and seen as important to maintaining positive changes. Hanlon et al. 53 and Wildman et al. 54 emphasise the importance of an approach that addresses both individual behavioural change and wider structural barriers in the context of socioeconomic deprivation. Furthermore, as has already been highlighted by GPs and organisations, the availability of suitable onward referral services is an important consideration for social prescribing at a time of severely constrained public spending. 54,55

Qualitative research that has explored challenges in delivering social prescribing services55 and the uptake of and adherence to social prescribing56 further highlights the importance of the link worker–client relationship, the complexity of the link worker role, patient expectations, the availability and cost of onward referral services, fear of stigma about psychosocial problems, short-term programmes and the need for strong community infrastructure. Husk et al. ’s4 recent realist review indicates that patients are more likely to accept a social prescribing referral if the intervention matches their needs and expectations and is presented to them in an acceptable way; patients are more likely to engage if they are supported to connect with accessible activities; and patients are more likely to adhere to the intervention because of skilled facilitators or condition/symptom change.

Quantitative and qualitative research highlights the complex phenomenon that is social prescribing, which is described as a ‘pathway and series of relationships’4 and not a single intervention, making it challenging to evaluate. As yet, it is not possible to make inferences about the efficacy of any particular model over another. 4 Current research into social prescribing indicates the need for longer-term outcome evaluation using robust routine data and mixed-methods design. If quantifiable effects of social prescribing are identified, the important questions are why, how and for whom these effects are seen. If there are no quantifiable effects, questions remain about why this is the case. The literature also highlights many gaps in our understanding of social prescribing, including the link worker role, service user experiences, the role of VCSE, integration into primary care, impacts on secondary services use, cost implications and cost-effectiveness.

Work package 1

Work package 1 (WP1) had the following objectives:

• to measure the short-term (i.e. 1-year) and long-term (i.e. 2- and 3-year) effects of the social prescribing intervention targeting adults with T2DM on levels of glycated haemoglobin (HbA_1c; primary outcome), body mass index (BMI), blood pressure (BP), cholesterol levels, smoking status and health-care utilisation

• to provide a range of estimated intervention effects based on comparing the intervention group with a number of relevant control groups

• to examine differential intervention effects in subgroups by gender, age, ethnicity, multimorbidity and deprivation level

• to measure HRQoL as change in EQ-5D-5L score at the 12-month follow-up in the social prescribing intervention clients.

Work package 2

Work package 2 (WP2) had the following objective:

• To measure the cost-effectiveness of the social prescribing intervention for health-care utilisation and each of the outcomes. From the perspective of a health service provider, costs and benefits will be compared for individuals in the intervention group and a cohort of individuals who did not receive the intervention. The robustness of the results will be investigated using sensitivity analysis.

Work package 3

Work package 3 (WP3) had the following objectives:

• to use ethnographic methods to examine –

  • link workers’ experiences of delivering social prescribing

  • patients’ engagement with the social prescribing intervention and whether or not and how social prescribing leads to changes in patients’ lives

• to examine the impact of COVID-19 on people with LTCs and the role of social prescribing during the pandemic.

Patient and public involvement methods

Patient and public involvement results

Reflections on patient and public involvement

There were aspects of PPI that were helpful to the research endeavour. Our ongoing engagement with link workers enabled us to provide feedback about response rates that acted as regular reminders to link workers about data collection and enhanced the baseline response rate. Equally, we realised that link worker-administered questionnaires would be logistically difficult at follow-up and switched to researcher-administered follow-up, which was likely to have enhanced the follow-up response rate.

The inclusion of two ethnographies meant that we were foregrounding the voices and experiences of those receiving and delivering the intervention. Over the period of the ethnographic fieldwork, we did not feel that it was appropriate to share information derived from the ethnographies with clients or link workers.

On reflection, we experienced some ‘blurred boundaries’ as a result of the involvement of link workers both as participants and in their role administering questionnaires. Furthermore, the constraints imposed by lockdown hampered the planned engagement with broader stakeholders throughout the course of the study. Further stakeholder engagement will take place once the research is fully completed, enabling the research team to include a wide range of perspectives, including study participants, providers, VCSE organisations, primary care, public health and local government, and providing input into the findings, interpretation and implications of the research.

Difference-in-differences models

Treatment and control groups

To apply the DiD model, it is important that we determine the treatment and control groups. With the intervention introduced into geographically determined general practices, there are a number of ways to consider the compositions of the treatment and control groups. Within our data, we have numerous ways to define treatment and control groups, leading to different possible estimates of the treatment effect. Table 3 outlines the possible treatment groups and the different estimates to which they give rise.

Each of the possible comparisons lead to different potential estimations of the treatment effect and the potential ‘bias’ associated with estimated treatment effect. We give most attention to the value τ_3b, the intention-to-treat (ITT) estimate of the treatment effect. The ITT estimate overcomes a number of problems associated with the estimation of treatment effects using observational data. First, any treatment is based on a referral process, so that those individuals who are the most ill/most in need, or incurring the highest costs, are those who are referred first. With regression to the mean, it is possible that these individuals see improvements in their health regardless of the intervention and, therefore, the effect of treatment may be overestimated. The ITT analysis mitigates this somewhat by considering all individuals to be treated at the time that treatment becomes available. Second, the ITT analysis accounts for the fact that the intervention will suffer from non-compliance and wide individual heterogeneity in the intervention composition and level of participant engagement. A final advantage of the ITT analysis is that it provides an estimate of the expected effect of the intervention on an individual drawn at random from the treatment practices. However, the ITT estimate cannot distinguish between the effect of the intervention on the individual and the effect of the general practice. It is possible to specify all of the possible combinations of treatment effects, as outlined in Table 3, as ITT analysis. Individuals in the respective treatment groups are deemed to be treated once treatment is available, despite that it may be some time, even years, before individuals actually receive treatment.

As outlined above, there are good reasons for taking an ITT approach. In addition, in recent years it has been noted that the two-way fixed-effects model is potentially biased when treatment assignment varies across time and there is treatment heterogeneity. 78,82 Essentially, the two-way fixed-effects estimator becomes a weighted average of all of the possible two-by-two combinations of treatment and control groups that are available in the data. This leads to the possibility that even when the treatment effect is negative the final estimated treatment effect from two-way fixed effects can be positive. This result depends on the size of effects and the size of the respective groups. To overcome this problem, de Chaisemartin and D’Haultfoeuille78 propose an estimator that is based on the time that treatment occurs, so comparing treatment changes across consecutive time periods with observations whose treatment status did not change. This allows the researcher to investigate the evolution of treatment across time and for the estimation of placebos that are similar to the event-study approaches outlined in Equation 3. Owing to the small number of switchers into treatment, at some of our time points we estimate these models for the main effects only and not for the subgroups. It follows from these recent modelling developments that it is not possible to rank the direction and sign of the potential treatment effects that can be estimated. It is possible that the direction of the estimates is affected by the relative sizes of the groups, the potential treatment heterogeneity and the possibility of heterogeneity in treatment over time. 78

Two-part models for secondary care expenditure outcomes

One of our aims is to estimate the impact of the intervention on health-care expenditure, taking account of usage and cost. Health expenditure is a mixture of distributions because the costs arise from a two-stage procedure. First, there is health-care use and, second, the costs of the health-care usage. As a result, there is considerable density in the distribution at zero because most individuals, including within our data, do not use secondary health-care services in a given year. A further modelling complication arises because, whereas most secondary health service use is of low cost, there are a significant number of individuals who incur very high health-care costs, giving the distribution of the data a long right-hand tail. The consequence of these distributional problems is that the traditional linear model in Equation 2 is not appropriate for investigating expenditure data.

To accommodate the distributional challenges of expenditure data, we follow83 and estimate two-part models (TPMs). The TPM splits the decision into two stages: a use stage and a cost stage. There were four main modelling decisions to be made: first, whether to apply a logit or a probit for the first part of the model (although the consequences of this decision are trivial). We chose a logit model. Second, we had to decide on the link and distribution family for the second-stage generalised linear model (GLM). We made these choices on the basis of the Box–Cox test and the modified Park test. 83 Following this procedure, we chose a log-link function and the gamma distribution. The final decision relates to which variables to include in the linear index. We tried a range of controls to assess their impact on the estimated treatment effect; all models included age and age squared.

The TPM can formally be considered as:83

In this formulation, f₀ is the density of y_i when y_i = 0 and f₊ is its conditional density when y_i > 0. There has previously been concern that there may not be independence between the two parts of the mode; however, Drukker84 has formally demonstrated that E(y_i|x_i) can be identified even if there is dependence between the two parts.

The TPM is a non-linear model and, therefore, the derivation of the treatment effect is slightly different from that of a standard two-way fixed-effects model. Following Deb et al. ,83 we can specify a non-linear conditional expectations function for a simple model as:

The control variables (x′_i where bold text is used to denote vectors), and the treatment dummy (D_i) are a linear function that is transformed by a non-linear function g(*). Now the treatment effect is the difference between two expected values based on the treatment status:

Therefore, in this case, the treatment effect of interest can be obtained from the sample as:

where N is the sample size. Essentially, as shown by Puhani,85 the treatment effect is the difference between the expected value of the outcome in the post-treatment period and the hypothetical expected value in the post-treatment period had the individual not received treatment. In practical terms, the average impact of treatment on the treated can be estimated by taking the average marginal effect from the TPM.

Data

We use data from three different sources: the QOF, SUS and intervention provider. All of these data were provided by NECS.

The QOF is an incentive programme that financially rewards English general practices for quality of patient care and helps to standardise improvements in the delivery of primary care. Under this programme, sets of targets are used to allocate funds, meaning that there can be excellent reporting of a variety of health outcomes for individuals. During the period of our study, there was only one major reform to the QOF targets that may be an issue for our analysis: a reduction in the number of targets by 40 in 2014–15. However, as these changes affected all practices, we would not expect this reform to affect our identification. The QOF provides data on our health outcomes: levels of HbA_1c (our primary outcome measure), blood pressure, cholesterol levels, glucose levels, BMI and smoking status. Where individuals had more than one outcome measure per year as a result of repeat visits to their GP, we took the yearly average as the outcome. The level of HbA_1c (a measure of glycaemic control) was chosen as our primary outcome measure because it is a key measure of T2DM control. Data on levels of HbA_1c are well collected (i.e. there are clear QOF targets for levels of HbA_1c) and glycaemic control has a significant effect on long-term health and even small reductions in HbA_1c levels can reduce long-term macrovascular and microvascular complications. 86–88

Quality and Outcomes Framework data are populated with measures and Read codes describing the data collected. There are possibilities of errors in routine data collection. For example, HbA_1c level measured using International Federation of Clinical Chemistry (IFCC) units can be recorded in the Read codes as DCCT (Diabetes Control and Complications Trial) units. For these reasons, some top coding was required. HbA_1c level was top coded at 120 mmol/mol, BMI was top coded at 50 kg/m² and bottom coded at 20 kg/m², and cholesterol level was top coded at 10 mmol/l.

The QOF also provides information on patient characteristics, including age, ethnicity (although there are some recording issues with this characteristic leading to inconsistent categorisation and small sample sizes), sex at birth, area of residence and the presence of additional comorbidities. All of the health outcomes and characteristics were extracted by NECS. The QOF was used to identify our sample of patients: individuals within the eligibility age range (40–74 years), registered at a treatment or control GP practice and with a diagnosis of T2DM on 1 April 2015. The data observation period ran from 1 April 2011–12 to 30 March 2018–19. Under our data-sharing agreements, only comorbidities that were part of the eligibility criteria for the intervention were included in our data sample, that is COPD, asthma, heart failure, coronary heart disease, epilepsy and osteoporosis. The area of residence was given as LSOA (lower-layer super output area) and these LSOA markers were linked to the IMD deciles, with 1 being the most deprived decile. Permission to access patients’ QOF register data were given by 13 out of the 16 treatment practices, covering 80% of those referred into the intervention and 11 out of the 15 control practices, providing information on approximately 8300 individuals, with approximately 58% of individuals in treatment practices. The exact number of observations depends on the outcome under consideration, as can be seen in the results.

Secondary uses services data are a record of secondary health-care use in NHS providers. For our study, we used information on the use and cost of outpatient services, inpatient elective services, inpatient non-elective services and A&E services. The costs were measured as the national tariffs that are paid for services at 2019 costs. To prevent extreme values affecting overall results, all costs were capped at the top percentile value.

Finally, we were able to observe who had actually received treatment from the intervention data that provided dates that an individual completed their initial goal-setting meeting with their link worker. We define this date as the point at which an individual started treatment. For the linear health outcome models, we also test the sensitivity of our results by constructing a new treatment group, where the treated are those individuals who have completed at least two goal-setting meetings with their link worker. These meetings typically occur at least 6 months apart.

The three separate data sets were pseudonymised by NECS, using the same key, and linked to form the complete data set. Table 4 shows the variable names and definitions. The pre-treatment summary statistics for the whole sample, and broken down by treatment and control group, are presented in Tables 5 and 6. Summary statistics for the whole time period are given in Appendix 3, Table 37, and Appendix 4, Table 38.

The data presented in the tables demonstrate that, in terms of average health outcomes, the treatment and control groups are similar, with the treatment group having, on average, slightly higher levels HbA_1c. The tables also indicate that the sample size for blood glucose is much smaller than that for the other outcomes; for this reason, we do not analyse the blood glucose outcome. In terms of health-care costs (and, to some degree, health-care use), the treatment group appears to incur, on average, lower costs (and use) than the control group. Finally, Table 7 shows the characteristics of the sample. Compared with the treatment group, the control group is slightly older, more likely to have an extra multimorbidity and more likely to be male. There are considerable differences in ethnicity, with the treatment group more likely than the control group to be non-white. Large proportions of both the control group (55%) and the treatment group (60%) live in areas in the top quintile of deprivation.

Health outcomes

Figures 2 and 3 and Appendix 5, Figures 11–13, provide the unconditional means for the health outcomes for the treatment and control groups across the whole time period. These are based on the ITT analysis, with individuals classed as being in the treatment group if they are registered with a general practice offering the intervention. We mark the point at which the intervention was available, year 15/16. These graphs help to provide evidence to support our DiD approach. Ideally, the two groups would have similar trends in outcomes in the pre-treatment period. If there is an impact of treatment, we would expect some divergence in outcomes post treatment. However, even if the figures themselves do not demonstrate parallel trends, because these are unconditional trends, it may be that once age, time and individual fixed effects are accounted for the parallel trends assumption holds.

FIGURE 2.

Glycated haemoglobin (HbA_1c) level: control vs. treatment by year. Mean outcomes plotted against year for the treatment and control practices, with the orange line indicating the year that the intervention became available.

FIGURE 3.

High blood pressure: control vs. treatment by year. Mean outcomes plotted against year for the treatment and control practices, with the orange line indicating the year that the intervention became available.

The trends suggest a mixed picture for the intervention. The average level of HbA_1c is lower in the control group than in the treatment group in each year prior to treatment (see Figure 2). The small dashed lines represent the linear trend. Pre treatment, the linear trends are parallel. Following treatment, the average level decreases to below the average for the control group and remains there, suggesting that, on average, the intervention has a role in reducing the levels of HbA_1c. However, the range over which levels of HbA_1c are changing is very small.

For the probability of having high blood pressure (see Figure 3), we see a similar pattern. Pre treatment, the trends (the small dashed lines) are parallel and the treatment and control groups largely move together. Post treatment, there is a divergence between the treatment group and the control group, with the treated group showing improvements in probability of having high blood pressure. Although, again, this finding is tempered by the fact that the changes are very small in magnitude.

For the other health outcomes, that is BMI, cholesterol level and smoking status (see Appendix 5, Figure 11), there is very little evidence that the intervention had any effect. The trends are largely parallel pre and post treatment. It is, of course, possible that these unconditional averages hide important effects and potential heterogeneity in outcomes. To consider these issues, we need to turn to the regression analysis.

However, before considering the regression analysis, we present the results from the falsification tests. These are event-study approaches, including the leads and lags of the treatment, and provide a test of whether or not there are significant differences in outcome between the treatment group and the control group across the whole time period. The omitted time period is the year prior to treatment. Ideally, pre treatment, the estimated coefficients will be close to zero and insignificant. Post treatment, if treatment is effective, then we would expect to see large estimates that are significantly different from zero.

In Figures 4 and 5 (and Appendix 6, Figure 14), the graphs for levels of HbA_1c (see Figure 4) and blood pressure (see Figure 5), and to a lesser extent cholesterol level (see Appendix 6, Figure 14), suggest that we can be confident of the parallel trends assumption and the impact of the intervention. Post treatment, the signs of the estimated coefficients mostly change sign and there is some evidence of significance. For BMI (see Appendix 6, Figure 15) and current smoking status (see Appendix 6, Figure 16), the results are less reassuring. There is no clear evidence of any treatment effect and most of the estimates are insignificant in both the pre-treatment and the post-treatment period.

FIGURE 4.

Health outcomes: falsification results – HbA_1C coefficients. Plots of the coefficient [and 95% confidence interval (CI)] estimating the difference in health outcome between the treatment and the control practices for each year, with the orange line indicating the year that the intervention became available.

FIGURE 5.

Health outcomes: falsification results – probability of high blood pressure. Plots of the coefficient [and 95% confidence interval (CI)] estimating the difference in health outcome between the treatment and the control practices for each year, with the orange line indicating the year that the intervention became available.

Health-care costs

In this section, we present the results from the ITT analysis of health-care costs using TPMs. A priori, we may expect the intervention to reduce A&E and inpatient non-elective use and costs, as individuals become better at managing their conditions with link worker support. For inpatient elective, outpatients and overall costs, the direction of effect could be either positive or negative. More planned use of health care may result from better condition management and link worker support. Although the assumption of parallel trends is more complicated for non-linear models, it can still be of interest to consider the pre-treatment and post-treatment trends. Appendix 7, Figures 17–21, and Appendix 8, Figures 22–26, show the unconditional plots for use, expenditure (see Appendix 7, Figure 17) and for the estimated falsification treatment effects from the TPM (see Appendix 8, Figure 22).

For the regression models, we alter our subgroup analysis slightly because of the reduced sample sizes that arise owing to the low number of individuals using health-care services during the year, which.

Time-varying treatment allocation and treatment heterogeneity

As discussed in Methods, Treatment and control groups, the standard two-way fixed-effects model cannot be used to estimate treatment effects when the treatment is allocated at different points in time and there may be treatment heterogeneity. For this reason, thus far we have used ITT-based estimates in which it is assumed that treatment occurs at the same time for all individuals. With linear models, such as those used for the health outcomes, it is possible to estimate models that account for timing of treatment and potential treatment heterogeneity. However, this approach cannot be applied to the TPMs and, therefore, we cannot address this issue for the health-care use and costs elements.

We implemented the de Chaisemartin and D’Haultfoeuille78 estimator to investigate the impact of treatment timing and treatment heterogeneity on the health outcomes. The results are presented in Table 16. The results show treatment effects relative to the time that the treatment first changed, alongside a pre-treatment placebo estimate. There are no estimates for the early compared with the late treated for time period T + 3 because of the small sample size for this comparison. The t-statistics are estimated using a bootstrap procedure with 50 replications.

The results are qualitatively similar and generally larger than the estimates from the ITT analysis. For example, for levels of HbA_1c, the estimated treatment effect can be as high as –7.59 mmol/mol; however, this is not significant, probably because of the small number of observations in this switching period. This is considerably higher than any of the ITT analysis estimates. Where sample sizes are larger, the estimates are still around –1.51 to –4.57 mmol/mol, suggesting that the intervention has made a significant impact on the level of HbA_1c. There are similar results for the probability of having high blood pressure, with estimates 3 years after treatment ranging between –0.10 and –0.11, suggesting a 10- to 11-percentage-point reduction in the probability of having high blood pressure. There are some small effects for cholesterol levels, but no clear effects for BMI or smoking status, for which the results are far more mixed.

Model description

The United Kingdom Prospective Diabetes Study Outcomes Model 2^© (UKPDS-OM2) is a well-established simulation model in the area of T2DM, which can be used to estimate life expectancy, quality-adjusted life-years (QALYs), and the cumulative cost of complications among patients with T2DM. 98 The model was developed using patient data from individuals with newly diagnosed T2DM who participated in the UKPDS. Data from 5102 UKPDS patients and 4031 survivors entering a 10-year post-trial period were used to inform model equations forecasting the occurrence of a range of diabetes-related complications and death. The model is capable of predicting event rates for T2DM-related complications [i.e. ischaemic heart disease (IHD), myocardial infarction (MI), heart failure, stroke, amputation, blindness, renal failure, ulcer, death) based on defined demographic characteristics of individual patients, as well as data on relevant risk factors (smoking status, total cholesterol, HDL cholesterol, LDL cholesterol, systolic blood pressure (SBP), levels of HbA_1c, peripheral vascular disease (PVD), atrial fibrillation (AF), weight, albuminuria, heart rate, white blood cell (WBC) count, haemoglobin levels, estimated glomerular filtration rate (eGFR)]. Adjustments to risk factor values and, therefore, demographic characteristics, will impact the complication rate for an individual modelled patient, as well as impacting their utility (quality of life).

The UKPDS-OM2 was selected for use in this analysis because its structure and equations are based on a single, large UK-based RCT in the area of T2DM,98 and because its structure is also consistent with T2DM guidelines provided by NICE. 99 The population considered was patients with T2DM aged 40–74 years either undergoing the intervention (link worker social prescribing as part of the WtW programme) or receiving current practice. The key characteristics of the cost-effectiveness analysis are presented in Table 24, while the model structure is shown in Figure 6.

FIGURE 6.

Flow chart showing the structure of UKPDS-OM2. Reproduced with permission from Hayes et al. 98 Reprinted by permission from Springer Nature Customer Service Centre GmbH: Springer Nature, Diabetologia, ‘UKPDS Outcomes Model 2’. CHD, coronary heart disease.

In each cycle, which represents 1 year, individuals face a risk of dying or experiencing complications, where this risk is calculated from the individual characteristics. Each complication that occurs will affect costs and lead to a decrease in utility. To allow for random events, the calculated risk for each is compared, in a random order, with a random number drawn from a [0,1] uniform distribution. If a complication is predicted, it will inform the remaining equations in that cycle. The nature of the complications, if they occur, will affect the underlying risk of death. Each year that an individual survives leads to an appropriate updating of risks, either from existing data sets or from the risk factor time equations. When an individual dies, they exit the simulation and their total costs and QALYs are calculated. Given that a patient-level data set was unavailable for this analysis, the modelled population was sampled based on a combination of baseline data available from the WtW study and from the literature (see Estimation of model parameters).

Estimation of model parameters

Point estimate values for the clinical effectiveness, utility and cost parameters are presented in Appendix 11, Tables 39–41. In addition, uncertainty around the point estimate and a distribution of uncertainty are presented where appropriate and available to allow for a probabilistic analysis to be undertaken.

The model requires the user to define patients’ baseline demographic characteristics and risk factor values (for each risk factor) for individual patients entering the model and on a year-by-year basis. Where these data are unavailable, the model automatically provides two methods to estimate yearly risk factor values for continuous risk factors (HDL cholesterol, LDL cholesterol, SBP, HbA_1c, weight, heart rate, WBC count, haemoglobin and eGFR) and two methods for binary risk factors (smoking status, PVD, AF, albuminuria). For continuous risk factors, (1) hold the initial value constant for the simulation period or (2) populate using linear regression (y = mx + c), where y is the result for a given year and x is the value from the previous year or the initial value from the inputs worksheet for the first year. Values for m and c can then be specified by group. For binary risk factors, hold the initial value constant for the simulation period or populate each year with a specified value for each group. 98 For this analysis, binary risk factors were held constant for the simulation period. However, an alternative approach was utilised to populate yearly continuous risk factor values for individual patients based on a distribution of uncertainty surrounding the baseline value. Annual values for each risk factor were sampled from a normal distribution, with a mean value equal to the value of the corresponding risk factor in the previous year and a standard deviation (SD) equal to 10% of that value. All sampled values were then compared with their corresponding range of plausible values, before plugging them into the model.

Assessment of costs and outcomes

The model was run to obtain the expected values for each strategy over a lifetime time horizon. The analysis was designed to generate the incremental cost per QALY gained (per patient) associated with introduction of the intervention. Overall results are presented in the form of an incremental cost-effectiveness ratio (ICER), which is calculated by dividing the difference in cost between strategies by the difference in effectiveness. The ICER can then be assessed by comparing the calculated value with the NICE-recommended willingness-to-pay (WTP) threshold for QALYs (lower threshold of £20,000 used to assess cost-effectiveness). 106 In addition, the cost and QALY difference between strategies, as well as the potential cost savings related to avoidance of complications, are presented based on a cohort of 10,000 patients (5000 in each group) to account for the broader population impact.

Where available, and when appropriate, data were entered into the model as distributions to fully incorporate the uncertainty around parameter values so that a probabilistic sensitivity analysis (PSA) could be undertaken. The PSA was run with 1000 simulations and appropriate probabilistic outputs were produced, that is, cost-effectiveness acceptability curves (CEACs) to show the probability of the intervention being cost-effective across a range of WTP thresholds, as well as cost-effectiveness planes. The combination of 10,000 patients and 1000 simulations were chosen as the numbers where conversion was achieved in test runs. Results of the probabilistic analysis also present the incremental net monetary benefit (NMB) of the intervention. The NMB represents the value of an intervention in monetary terms when a WTP threshold for a QALYs is known, calculated as: (incremental benefit × threshold) – incremental cost. 107 Probabilistic outputs were produced using the heRvis visualisation builder tool in R Shiny (RStudio Inc., Boston, MA, USA). 108 Deterministic sensitivity analyses (see Sensitivity analysis) were also conducted to explore individual and multiple parameter variation and the impact that this would have on the results. All costs and health effects were considered from an NHS and a Personal Social Services (PSS) perspective and were discounted at an annual rate of 3.5%. 109

It should be noted that the base-case analysis considers only the impact of the intervention on two of the baseline risk factor values (SBP and HbA_1c level), given the data that are available from the QOF data. However, in all likelihood, the intervention will impact additional parameters that we attempt to explore through sensitivity analyses.

Sensitivity analyses

Base-case results

Probabilistic results

The base-case results for the probabilistic analysis are presented in Table 27. These results differ slightly from the results of the deterministic analysis because they take into account the uncertainty associated with individual model parameters and patients. The results indicate that the intervention increases QALYs by 0.004. Costs associated with the intervention arm are higher than those associated with current practice (£1309) (primarily because of the cost of the initial intervention itself – otherwise the intervention would be marginally cost saving). The incremental NMB of the intervention is also presented in Table 27. The results indicate that the NMB of the intervention is negative at both £20,000 and £30,000 WTP thresholds.

TABLE 27 - Base-case results: probabilistic

CI, confidence interval.

Strategy	QALY (95% CI)		Cost (£) (95% CI)		Incremental net benefit (£) (95% CI)		ICER (cost per QALY) (£)
	Mean	Difference	Mean total	Difference	£20,000 WTP threshold	£30,000 WTP threshold
Current practice	12.091 (9.140 to 14.713)		15,441 (11,852 to 19,636)
Intervention	12.096 (9.148 to 14.721)	0.004 (–0.022 to 0.029)	16,750 (13,181 to 20,935)	1309 (1192 to 1440)	–1224 (–1874 to –611)	–1182 (–2089 to –314)	327,250
Difference based on a cohort of 10,000 patients		40		13,090,000

Further outputs from the probabilistic analysis are shown in Figures 7 and 8. The cost-effectiveness plane (see Figure 7) shows individual results (points on the scatterplot) from the multiple simulations of the model. All points are above the horizontal line of the plane, indicating that the intervention is likely to be more costly than the comparator. However, there are also more points in the north-east quadrant of the plane, indicating that the intervention is likely to be more effective than the comparator. Results from the CEAC (see Figure 8) indicate that the probability of the intervention being cost-effective increases as the WTP threshold increases, although it is highly unlikely to be cost-effective at existing WTP thresholds. Results from Figures 7 and 8 (as well as from the NMB analysis) suggest that the intervention is likely to be more effective than the comparator, and that as the WTP threshold increases the likelihood of the intervention being cost-effective also increases (because of the additional cost of the intervention in the base-case analysis).

FIGURE 7.

Cost-effectiveness plane base case: results of probabilistic sensitivity analysis.

FIGURE 8.

Cost-effectiveness acceptability curve base case: the probability of each preferred intervention being most cost-effective against WTP for each QALY threshold.

Sensitivity analysis results

Participant observation

The researcher (Jayne Jeffries) spent 2 days per week over a period of 6 months between August 2019 and February 2020 at the provider organisations in offices shared by link workers, initially gaining a broader understanding of the structure and ethos of each organisation and understanding the professional practice(s) of link workers. This included attendance at meetings, as well as any scheduled training events. All link workers received participant information leaflets and those participating completed informed consent sheets (see Report Supplementary Material 2). Twenty link workers completed a demographic questionnaire (see Report Supplementary Material 2). During this time, the researcher also invited link workers to take part in individual interviews and focus groups to reflect on their experiences of the intervention.

In addition, the researcher spent a more focused period of time between October 2019 and February 2020 shadowing a smaller number of link workers (n = 8) as they went about their daily working routines. This included observing link workers meeting with clients at provider offices, GP surgeries or in the community, and accompanying link workers when they were visiting clients in their homes. Practice staff and patients were informed of the researcher’s (Jayne Jeffries) presence via letters to GP surgeries, GP surgery posters and leaflets (see Report Supplementary Material 2). At opportune times during the working day, short informal interviews were conducted with link workers, during which they were asked to reflect on their practice.

Field notes were recorded after each period of observation in line with standard ethnographic procedures. 110,111 Field notes described scenes and allowed us to explore meaning and reflect on positionality. They supplemented interviews by capturing context and information that was not audio-recorded, such as non-verbal communication.

Interviews and focus groups

During face-to-face fieldwork, interviews were conducted with six link workers using a semistructured interview guide (see Report Supplementary Material 2) and were recorded. Two focus groups were conducted at provider A (n = 7 and n = 3 participants). One focus group was conducted at provider B (n = 6 participants). Topics discussed included link worker roles, referrals, and relationships with GP surgeries, clients and the community sector (see Report Supplementary Material 2). Questions focused on ‘what works’ for clients and how ‘what works’ varies for different clients. Focus groups were audio-recorded with consent.

At the start of the COVID-19 pandemic, face-to-face fieldwork had to cease and the researcher proceeded to undertake a further 13 semistructured interviews with senior link workers, link worker managers, social prescribing stakeholders from both provider organisations and staff from the intervention’s umbrella body. Owing to the differing nature of these roles, topic guides were personalised and focused on professional histories and roles alongside organisational responses to the rapidly changing circumstances. These were conducted by telephone and audio-recorded. These data were also used as part of our exploration of the impact of COVID-19 on the experience of clients during the pandemic (see Chapter 7).

Interview and focus group recordings were transcribed (using intelligent verbatim transcription) and transcripts were checked for accuracy. Transcripts were ‘cleaned up’ during transcription, with dropped consonants and vowels added (e.g. ‘gonna’ to ‘going to’). Some participants often substituted whole words with regional dialect words. These words were not changed during transcription and their meaning is indicated in square brackets. Data were anonymised and pseudonyms applied to preserve anonymity. Data presented are labelled by participant code/pseudonym, provider, source and date as appropriate.

Data analysis

The final qualitative data set comprised interview and focus group transcripts, as well as ethnographic field notes. Thematic content analysis was developed subsequent to data collection and was achieved via coding all data and synthesising codes and categories into emerging themes of understanding. Interview narratives (‘what was said’) were compared with the observations (‘what was done’). Line-by-line coding was conducted using NVivo 12 (QSR International, Warrington, UK) to identify themes in the data, and memos were used to assist in the process of moving from content-based descriptive themes that are relevant to the research questions to more conceptual themes that incorporate differences, similarities, inconsistencies and contradiction in the data. 112 Analysis involved synthesising and theorising the data, and re-contextualising this new knowledge by considering it in the context of other research. Analysis was led by Bethan Griffith, who met regularly with Tessa M Pollard and Suzanne Moffatt to review the coding framework, discuss emerging themes and develop the analysis.

Work routines

Link workers operated at different sites. Their ‘base’ was at the offices of their provider organisations and in both instances included a shared office space with a kitchen. Each link worker was also allocated a GP surgery at which they had designated ‘surgery time’. They mostly worked with clients from their surgery, but if caseloads were high or staff left, then this could change. Less commonly, link workers visited clients in their homes or in a community setting. It was generally felt that home visits were conducted much less often at the time of fieldwork than at the start of the intervention. One encounter took place in a local library, which was observed, and link workers reported that they could meet clients in cafes, but this was not observed. At one team meeting, it appeared that this could become problematic when clients were unable to afford coffee and provider policy dictated that link workers could not pay from petty cash.

Link workers’ interactions with clients could often begin before formal referral, as they were increasingly responsible for generating referrals, which could include telephoning potential clients to offer the intervention. Following referral from any source, link workers and clients usually met face to face. The first encounter centred on completion of the first WBS and could routinely take upwards of 1 hour. These were seen to take place mostly at GP surgeries but also in clients’ homes and at one of the provider offices. At the end of this interaction, link workers would encourage clients to set goals, often offering onward referral or signposting to relevant organisations to facilitate meeting them. Subsequent contacts were to assess progress, conduct WBSs and set new goals. The expectation was that when WBS scores were stable, or there were no new goals, clients moved to KIT before decisions were made about ending journeys.

During the COVID-19 pandemic, link workers had to adapt their work routines considerably, with all link workers having to work from home. It was, however, apparent that, in addition to supporting existing clients and delivering the supplementary COVID-19 response, they were still being encouraged to meet targets.

The next section presents the four main work routines identified above and explores how these activities were differentiated and shaped within the intervention.

The referral process

As part of the intervention’s original logic, referrals were to be generated by primary care staff, but differential levels of engagement meant low referral rates had to be supplemented by link workers recruiting new clients into the intervention themselves. Meeting referral targets meant that new WBSs could be completed to secure payment.

Some primary care sites allowed link workers access to their patient lists so that they could ‘cold call’ eligible individuals. At other sites, link workers used IT systems to identify eligible patients, then sent a paper request to clinical staff to generate a referral. The process of generating referrals was an unpopular administrative burden and was also felt by many to be an uncomfortable interaction:

Kate [link worker] emphasises the issues regarding list work . . . the unease at ‘cold calling’ NHS patients asking if they need the [intervention]. She says there are issues with staff feeling harassed by their caseload and the volume of work that they have to conduct/balance, by feeling that they are harassing potential clients. ‘There is a lot of pestering’.

Field notes, B, November 2019

Of course, as well as the work of generating the referral, each new client contributed to the link workers’ caseload. Increasing workload was frequently highlighted and attention was drawn to the incongruity of recruiting new clients and not having time to meaningfully engage with them. The following extract from the researcher’s field notes illustrates just how laborious generating referrals could be, potentially obscuring other tasks, such as face-to-face contacts:

It is only in the preceding 10 minutes, chatting away as we travel together that I became aware of the link worker’s role within this [GP] surgery. I already know that [this link worker] is ‘one of our best, he generates loads of referrals . . . and stars’, a comment made by the service manager during a team meeting in November. However, I did not realise that his time at the surgery is strictly for generating referrals; [he] does not see clients face to face at this location.

Sam, B, shadowing, February 2020

After arriving at the surgery, Sam proceeded to a shared office space to call potential clients:

‘My name is . . . , I am calling from [the GP surgery] from [WtW]. Is it possible to speak to [patient]?’. The link worker then goes on to explain the following benefits of the service – GP offers short appointment time, WtW can offer up to an hour, we provide support, information and guidance to patients with LTCs, we can focus on lifestyles, managing your symptoms and social aspects too. Can I send you a letter about our service?

Following the call, Sam had to send a message to a GP to complete the referral:

‘I have spoken to the patient, explained the WtW service, and s/he is happy to have an appointment. Please can you refer this patient?’, The link worker tells me that he has 10 referrals pending in the task list, the time lag is often because he is waiting for the secretaries to action a task, which can take up to 10 days.

Sam, B, shadowing, February 2020

Woven through many of these examples was evidence of the differentiated relationships that link workers had with primary care staff. Engaging primary care to increase awareness of the intervention and secure more referrals was an important part of the link worker’s role and could be challenging:

As the conversation continued, it became clear that link workers had been tasked with shadowing a GP in their surgery/ies, and that a handout was designed to enable link workers to raise questions with staff within the surgery/ies. This was an exercise designed to develop working relationships and to understand the role of different actors within social prescribing, breaking down barriers that might, for example, prevent referrals, by engaging in conversation . . . However, one of the link workers stated that she felt uncomfortable doing this, and it was clear that there were different relations of power and interactions experienced by link workers depending on the surgery they worked in.

Field notes, A, December 2019

During the COVID-19 pandemic, link workers were still encouraged to generate referrals. One link worker, Abby, described how practices had given them lists of vulnerable patients whom they could call as part of the intervention’s COVID-19 response. If they were eligible, they could also offer individuals the intervention.

At the start of the pandemic, there were examples of relationships with primary care being renegotiated and new working practices emerging. The intervention responded rapidly, providing practices with the opportunity to deploy link workers in activities they felt to be the most useful:

There have been opportunities with supporting around the shielded list, and some partners that maybe haven’t necessarily understood our role or utilised the link workers in the past have utilised them.

Janice, A, COVID-19 interviews, June 2020

This experience was not consistent, and link worker managers were able to identify that increasing pressure on primary care teams during the pandemic had occasionally exacerbated existing difficulties with engaging them in the referral process. Nevertheless, it was clear that new possibilities could be negotiated and that strong working relationships with primary care could reshape performative pressures and change existing work routines.

The Well-being Star

Following referral, a face-to-face meeting was arranged between clients and link workers. Following introductions, link workers would usually outline the intervention’s consent and data-sharing procedures. Through repeated observation, it became clear that the remainder of this first meeting was heavily structured around completing the WBS and, although different practices were observed, the sequential movement through all eight domains seemed to permeate all of the encounters. The WBS was popular with link workers as a tool to guide conversations and identify issues that clients may not otherwise mention; however, the need to complete it could, on occasion, be at odds with the natural flow of conversation. Link workers had to navigate this carefully, striking a balance between listening to the client’s concerns and gathering the information for the star.

Between stars, clients ideally worked towards goals, and link workers acted on tasks generated. The aim was to complete stars every 6 months until things were stable, at which point clients were expected to move to KIT:

What do typical journeys look like? The team leader picks up a marker pen turns to the flip chart paper and starts to write, from the point people get referred, people stay in engagement phase until first star, this period of time can be days, weeks or months, we don’t give up on people (stress, bereavements, etc.) – there is no official time limit for the amount of time people can be on the engagement phase it is just what they are classed as being in until the first appointment. Intervention phase begins once client moves between first and second star, typical journey last 18 months – 2 years, over the course of 18 months set goals, supported clients with goals and most goals have been achieved. People are steady, scores have shifted from low to as good as they can get (diabetes under control, exercising, moved house), therefore nothing else needed, so move to KIT (a safety net for the client and a way that link workers can make sure all changes made by clients in 1.5 years are going to be sustained over a long term without the need of a link worker). If we call them in 3 months it gives client the opportunity to say if things are going well, or if things have changed. Hopefully stay in KIT for star five and six, and all scores stay same, have not gone backwards, therefore, we can close them.

Field notes, B, December 2019

This extract charts a linearity, imposed by completing sequential stars, that was often at odds with the routines of link workers and overlooked the complex reality of clients’ lives. This same linearity could also obscure some aspects of practice:

Then you’ve also got the people who have catch-up between their Well-being Stars. I have people that want to be seen every 6 weeks just to keep them on target, they need that space to come and talk through everything that’s going on in their heads. So, I have quite a big handful of people like that. You’ve got them that you’re seeing, then you’ve got your stars every 6 months, then you’ve got your generic catch-ups, which can be anywhere between 2 or 3 months between stars, and then they’re expecting you to get more new ones in as well.

Anna, A, focus group 2, October 2019

During the same focus group, another link worker recounted her experience with a client with whom she had been working:

I did a referral for her for talking therapies. She just turned up at the surgery one day and she wanted to see me, she says, ‘I’ll wait . . . If she’s got people in I’ll wait’. I sometimes work until 6.00 [p.m.], even 6.30 [p.m.] if I have to. I don’t mind, because, to me everybody’s different so it’s that flexibility you give your client. We had a bit of a chat, then I did something else for her and did another referral for her.

Anita, A, focus group 2, October 2019

This incongruity was brought into even clearer focus during the COVID-19 pandemic, when all clients and link workers experienced a significant disruption to their circumstances. The following extract paints a picture that sits in stark contrast to the linear one drawn above:

Life, regardless of whatever interventions, whether it’s us, GPs, or us, it’s waves. It tends to be waves that kind of occur. I guess our job is trying to help try to smooth some of the waves so it’s not as rocky and people don’t get quite as seasick with them, [laughter] for a bit of an analogy. We can’t necessarily . . . We can’t achieve that kind of . . . Right, everybody is okay for the rest of their lives now. That’s not really realistic.

Alice, A, COVID-19 interviews, July 2020

Understanding how the WBS shaped client encounters and imposed a sense of linearity is important because link workers were under a performative pressure to complete them to secure payment, as illustrated by these field notes from a training event led by the team leader at provider B and by a quotation from a link worker from the same provider:

The team leader asks staff for their thoughts on the money associated with WBSs and their awareness of this information. One of the newer link workers states, ‘I find it difficult that we place a monetary value on people. I tend not to think about that side of it’. The team leader states that link workers are in a funny position, ‘at interview we say it is person-centred, it’s holistic, we do it at a client’s pace, that’s why we have the luxury of such a long time working with them’. She also states that link workers have to represent the client-facing caring aspect of the service, without letting them know about the financial implications of the WBS. Getting funding from social investors, that we can pay back, so it is in our interest to do so, but should not let clients know or push them to do stuff.

Field notes, B, December 2019

I know you don’t feel as if you think about the money, but that was one of the first things I was told when I started. So, Jo [link worker] had said to me, ‘We get x-amount for a second star’. And I thought that was commission. That was how it came across and I said, ‘Oh, I didn’t know that it was commission-based’. And she said, ‘It’s not. It’s for the company’. And I was like, ‘Right’.

Abby, B, focus group, September 2019

Once completed, data from the stars had to be put into the IT system; the following extract shows how failing to do this meant that they were of no value:

The team leader moves on to cover the importance of adding a goal for a client, noting that MIS [management information system] order means that if ‘add goal’ is completed first then payment for WBS is missed and therefore MIS will NOT generate a second WBS reminder or show on claims form. The team leader provides instructions about how to reorder the information using ‘input’, stating that all client journeys will stay red until the order is changed.

Field notes, B, December 2019

Once the WBS was completed, link workers were able to move on to the ‘linking’ inherent in their job title.

‘Linking’

Link workers described how they used the WBS to structure goal-setting and offer support to the client. What this meant in practice varied considerably. Goals could be anything from stopping smoking to attending the next appointment. Onward referrals were observed to be lifestyle interventions, such as cookery classes and exercise sessions, with some signposting to more tailored groups, such as fishing clubs and veteran support, although these examples were already known to the clients. Link workers also made referrals to the voluntary sector for welfare advice, wrote letters on behalf of clients, and liaised with GPs and other medical staff to resolve queries about medications. Onward referral and signposting could be difficult, requiring the link worker to align the client’s social world, the requirements of the intervention and local community provision. Pauline, for example, worked with a client who was keen to do more exercise as one of his goals:

The client had previously been referred to the gym, a 10-week session that had now finished. He was waiting for a phone call to move to another class. LW explained that he will still be able to move to another class but that there was a waiting list. The client enjoyed swimming but [his local pool] is currently closed for refurbishment. LW suggested [another pool], but client said it was no good as he would have to use two buses to get there, and two buses to get back.

Pauline, B, shadowing, October 2019

Few direct interactions were observed with VCSE actors. Where they did occur, they involved a link worker contacting an individual they already knew, such as a named contact at the local food bank. It was acknowledged that community sector funding was a key part of social prescribing, with concerns that already precarious funding streams would be further threatened in the post-pandemic landscape. One manager alerted the researcher to concerns about VCSE capacity locally, citing a report that likened the roll-out of NHS link workers to ‘lots of travel agents and no holidays’. 113

In addition to onward referral and signposting, link workers also provided informal and less tangible types of support. For example, one link worker, Lucy, said that she would often text clients in her own time to check on them and remind them of important appointments. Another link worker said that sometimes she just listened:

And you’re just . . . And I’ve had so many clients burst into tears on me at the first meeting going, ‘Finally, somebody’s listening to me. Somebody is going to help me’. I think that’s really important. I think that’s the biggest thing.

Kate, A, focus group 2, October 2019

Variability was apparent between link workers in what they felt the boundary of their role was. Hilary made a clear distinction between support work and behaviour change, citing the following example:

My previous link worker was very much a support worker . . . she did 17 home visits, whereas I brought it down to one. She went to the homes, and once you are in somebody’s home they’ll say, ‘Oh, I can’t do this’. She was very much a, ‘I’ll do it’. Hence the dog washing. This man couldn’t wash his dog. She said, ‘Oh, I’ll do it’. Then it became regular. Then every week she was going round to have coffee, dog washing. I think the problem was is she had lost sight of the role. She was a support worker, not a link worker, and that’s the really big difference for me.

Hilary, A, interview, November 2019

We also touched on a sense that over time link workers were being encouraged more towards a focus on promoting individual behaviour change. Inherent in this was a sense that clients had to be ready for the intervention. For some link workers, this created contradictions between wanting to help clients navigate their complex social reality and acknowledging their reduced capacity to benefit from an intervention that required them to change their behaviour:

Really, I guess, if we really want to be part of social change, then those are the [complex] people that we really want to really try and win over. They’re the people who also DNA [do not attend] for the doctors and nurses, who don’t come and get vaccinations or whatever might help support their health. So, it does make a difference, but those are going to be very miniscule movements. Whereas you get someone further along who’s more motivated and, yes, they’ll go to [exercise classes] or anything that you send them to, and they’ll have a bigger change because you’ve caught them at the right time and they’re ready for it.

Lucy, A, focus group 2, October 2019

We touched on how the proposed linearity of the intervention could be at odds with the complexity of clients’ needs, and here we see how this complexity was poorly aligned with the increasingly dominant narrative of behaviour change.

Decisions to discharge

The decision to end a client’s journey was not always easy. We saw in the linear account of the intervention given above that clients in KIT who no longer ‘needed’ the link worker’s support were ready to end their journeys. However, the observed reality could deviate from this and appeared to be shaped by the performative pressure of meeting targets. There were examples of link workers being encouraged to end journeys to balance capacity but, equally, discharge could be delayed until after a star was completed to secure payment.

Closure. The link workers refer to their ‘core purpose’, to work with as many people as possible, stating that closure is needed to make room for more. The discussion focused on being able/or not, to free up time to get a new cycle of clients going.

Field notes, A, December 19

Some clients did not feel ready to be moved into a KIT phase or to be discharged. Indeed, one link worker suggested that some clients reported lower scores on their WBS to avoid discharge. It appeared that for some the lived reality of ending a client’s journey could be a complex process, embedded in the therapeutic relationship. In one interaction with her client, Tracy (who we meet again in the next chapter), Lucy introduces the possibility of ending her journey soon:

The link worker states that Tracy has been a WtW client for a while now, and that it won’t be yet but eventually she will have to ‘stand on her own two feet’, she says that Tracy will have more time to focus on herself after all of her appointments subside. Tracy says that she is scared of losing her support networks if she gets too healthy. The link worker recognises her genuine concerns, encouraging her to think about the social benefits of going to the gym, and not to worry about stuff disappearing as ‘you will fill your time with things that are more challenging’.

Lucy, shadowing, February 2020

This echoes earlier narratives about patients not needing the intervention any more and resonated with other accounts where link workers acknowledged that the end point was not to ‘fix’ clients but to help them do things for themselves.

However, primarily because of caseload and regular changes in staff, the process of ending a journey could not always be so carefully negotiated. The start of the pandemic forced many link workers to review client journeys and discharge some clients to balance capacity and workload.

Reflecting on success

When considering what had been successful, link workers often gave nuanced and detailed descriptions of what they had achieved for clients. These were often less tangible than the outcome metrics recorded by the intervention and that generated payment, and were also intertwined with the narratives around ending journeys when clients could ‘stand on their own two feet’:

I think from working and having a caseload, it is seeing people go through the whole process, make big changes in their life and get to that point where they don’t need us and you just think, ‘You’ve done your job properly and you’re in a much better place’, even if it’s just through minor, little changes or really big house moves or whatever it is.

Marie, B, interview, December 2019

You see massive character change, you see confidence change, like, they’re willing to engage change. I’ve got a guy, who when I first started seeing him, was having five or six panic attacks a week, now it’s two or three and he says that’s because of me.

Molly, A, focus group 2, October 2019

In another example, one link worker described how she had worked tirelessly with a cancer patient who was homeless and struggling with her immigration status. She was proud to have arranged accommodation, secured a regular income, worked with the GP to stabilise the client’s medication and had positive correspondence with the Home Office; these are all outcomes that can be traced to the work of ‘linking’ described above.

There are also other examples that could be included, often detailing clients’ complex social circumstances and the wide variety of routes that link workers took to help them. Success was seen variably as an increase in clients’ confidence, making things a bit easier for them, or simply getting clients to attend a follow-up appointment.

Participant recruitment

Participants were purposively sampled on the basis of gender, service provider and length of time with the intervention (see Appendix 13, Table 43, for sample demographic data). Based on these specifications, a manager at each provider organisation was asked to select the sample for us to approach. This was to avoid link workers selecting ‘good’ patients, although link workers were asked to use their personal knowledge of clients to advise if any of those selected should not be approached because participating in the study would be likely to cause the client distress (e.g. because of a recent bereavement or a serious mental health problem). During recruitment, the sample was continually monitored for age, area of residence, ethnicity, employment status and household structure to ensure diversity across those demographic categories. As fieldwork proceeded, additional younger and more intensively engaged participants were purposively sampled to populate the younger age range of the sample and to obtain more data about the intervention.

Some participants had been referred into WtW soon after the intervention began when it was delivered by the original four providers (n = 8, referred between September 2015 and May 2017); an additional participant was referred in September 2017, shortly after the number of providers decreased from four to three. The remainder of the sample was referred into WtW under the current arrangement of two providers delivering the intervention (n = 10, referred between January 2018 and March 2019).

We recruited friends and family opportunistically during participant observation with key participants. Family members were approached for interview only with the permission of the relevant key participant.

Participants were provided with an information sheet (see Report Supplementary Material 2) and were asked to provide written informed consent (see Report Supplementary Material 2). Other individuals encountered during the ethnographic observations (but who did not take part in interviews) were offered information sheets (see Report Supplementary Material 2) and asked for verbal consent for the researcher to carry out observations. Informed consent was revisited with participants at regular intervals throughout fieldwork.

Interviews

Initial (flexible) interview guides (see Report Supplementary Material 2) asked about participants’ health, their experiences of the intervention and whether or not and how they made changes to their everyday practices as a result of their interactions with link workers. Final interviews were tailored to individual participants and pursued questions and phenomena identified during participant observation, and discussed findings from observations where necessary to avoid misinterpretation. In addition, with participant consent, data documenting client journeys (e.g. notes made by link workers following client meetings) were provided by the intervention (n = 15) (see Report Supplementary Material 2). These records were discussed with key participants at the final interview. Semistructured interviews with family and friends explored health and their perspectives on the intervention (see Report Supplementary Material 2).

Translation

Some participants did not speak English fluently. Language barriers were overcome by either working alongside paid translators or, at the participant’s request, using a family member. We did not pursue participant observation where language barriers posed problems, as the presence of a translator is unhelpful for this form of data collection. Interviews with non-English-speaking participants were conducted in their first language with Kate Gibson and an experienced translator fluent in the relevant language. Interviews were recorded and simultaneously transcribed and translated into English. The translator participated in discussions about the interpretation of the data to ensure that meanings were correctly understood.

Recording and analysing data

Procedures for recording, transcription and analysis were as described in Chapter 5. Kate Gibson led analysis, and Tessa M Pollard and Suzanne Moffatt contributed through regular meetings to discuss emerging findings. Pseudonyms are used throughout.

Compensation for participants

Key participants received a token of thanks (a £20 shopping voucher) at their first and last interviews, and if they participated in a photo-elicitation interview; family members were also given a £20 voucher for participating in a family interview.

Referrals

The WtW logic model (see Appendix 15, Figures 21 and 31) stipulates that general practices identify and refer patients who have ‘expressed willingness/ability to change’. This appears to be the experience of participants who were referred into WtW during its early stages. For instance, despite an extensive period of time passing, Andy clearly recalls a conversation with his GP about his referral into WtW:

It was offered through the doctor. The programme – I don’t know how long it had been going – but they just suggested, ‘Well, we’ve actually got this. Do you want to give it a try?’.

Andy, referred September 2015, first interview, 6 February 2019

A conflicting picture emerges for participants referred into WtW more recently. Although some participants did first hear about WtW from a HCP, as the following quotation demonstrates, they were sometimes slightly reluctant recruits. Patrick recalls with detail a conversation that he had with a health-care assistant:

. . . she [the health-care assistant] went, ‘Mr P, I think you could do by seeing [a link worker]’ . . . And I went, ‘Well, I don’t really know’. She went, ‘Well, let them come and talk to you and see if they can help in any way, with all these problems you’ve got’. But they’re not really problems when I think . . . I don’t class them as problems. And I went, ‘Fair enough’.

Patrick, referred November 2018, first interview, 16 March 2019

Patrick’s conversation with the health-care assistant clearly allowed him to prefigure his referral. However, contrary to following a conversation about ‘readiness to change’ between a patient and a primary HCP, Patrick appears to have been persuaded or ‘recruited’ by the health-care assistant to consider WtW, despite his hesitancy around classifying his ‘problems’ as ‘problems’.

Other accounts of recent referrals suggest a lack of primary HCP involvement in the referral process, and several participants were first alerted to WtW when they received a telephone call from a link worker. For instance, Shirley’s quotation below indicates that she was ‘cold-called’ by a link worker:

She [the link worker] said that my name had been passed on. She’s not at the doctor, but she works in conjunction or something with the doctor.

Shirley, referred September 2018, first interview, 4 February 2019

This experience of being ‘cold-called’ by a link worker confirms the previous chapter’s findings that the introductory stages of clients’ WtW pathways may be shaped by the need to generate referrals, a task often taken in hand by link workers. This temporal ‘drift’, where clients’ different experiences of WtW (in this case, its introduction) relate to the changing trajectory of the intervention as it is operationalised, is something that resurfaces throughout clients’ experience of WtW, as the following sections will elaborate.

Supported linking

Participants in the KIT phase who were in active intervention during the early stages of WtW often reflected on being actively linked into a variety of onward services and activities. These included interventions providing support in seeking employment; advice on benefits; LTC management sessions; small community groups, such as information technology (IT), photography, cookery, gardening, exercise classes and walking groups; social groups; VCSE and NHS-run mental health support; local council fitness programmes; and volunteering opportunities. Moreover, clients received a range of practical support; for instance, link workers often made enquiries or advocated on clients’ behalf, reporting back to the client thereafter (e.g. with GPs, the council and social services). Elsewhere, participants recalled link workers conducting home visits to assist with IT connectivity, supporting clients to navigate housing and benefits system by helping with form filling, sourcing information about grants, and one participant borrowed a link worker’s disabled toilet key while she waited for hers to be issued. Importantly, this was achieved via sustained and continuous interactions with a link worker. For instance, in the following quotation, Brenda recalls an interaction with her link worker following a gym referral. Not only does her link worker follow up on the referral but also the exchange is reciprocal and becomes an opportunity for the link worker to learn more about the gym – an added benefit of their established rapport:

He [LW38] used to ask all about it [the gym], and how it made you feel and what you thought about this, and would you recommend it. Then he used to ask me to ask them questions.

Brenda, referred December 2015, first interview, 5 February 2019

Some participants received telephone reminders concerning impending appointments and activities. As Eddie (referred April 2016) sums up, ‘they would ring me, or send me a text message to remind me that my appointment was getting closer and that kind of thing’. Masood also illustrates this dynamic (referred July 2015); his journey data record detailed activity surrounding several links. After expressing an interest in learning how to ride a bike, Masood was supported to attend a one-to-one bike session. When on the day of the session he indicated he was hesitant to attend, his link worker persisted and accompanied him to the session. Ultimately, Masood was not comfortable with cycling. A few weeks later, Masood was advised by the GP to lose weight and so the link worker put in motion a referral to the local gym and, once again, accompanied him to his first session. For the following weeks thereafter, the link worker called Masood to remind him about his weekly gym sessions. When his attendance dwindled, the link worker liaised with the gym instructor to restart the referral and supported him to attend once more. Alongside this, they continued to send Masood reminders to attend a local IT group.

Masood was not alone in being accompanied to groups. Those in the sample who were in ‘active involvement’ in the early stages of the intervention indicated that they were often accompanied to activities, or at the very least accompaniment was offered. For instance, Zaheer (referred February 2016) was supported by his link worker to apply for Personal Independence Payment (PIP) and his link worker then accompanied him to his appointment:

He [the link worker] ordered the forms, helped me to fill them and then when I got the appointment, he met me at the office where we needed to go, the PIP office. I went in the taxi and I got paid for that. He went with me for the interview. He was a really good support because sometimes you want to say things but you don’t want to say. Where I was, sort of, lacking, he would say something on his own behalf, what he has . . . I mean it wasn’t choreographed or anything. He was there and if he thought I was not saying enough. He would elaborate.

Zaheer, first interview, 9 January 2019

Zaheer’s mention that the link worker ‘was there’ alludes to a form of practical support that resurfaces time and time again in the narratives – and the WtW journey data – of those referred into the intervention in its early stages.

Importantly, the nature and intensity of support appeared to reflect the different needs of clients. Those participants who were enduring more complexities, for example Masood, who was living with multiple LTCs, or Zaheer whose depression and T2DM were exacerbated by redundancy, seemed to have triggered more support from the intervention. Likewise, those with fewer difficulties, for example Geetha (referred August 2016) and Andy, required less support to access activities and, therefore, it is unsurprising that instances of supported linking are relatively absent from their data. As Geetha pointed out, ‘I don’t need support’, and Andy indicated he just needed a ‘point in the right direction’. Furthermore, the WtW journey data show that this group of participants often experienced face-to-face contact with their link worker. The regularity of this face-to-face contact again varied according to clients, but occurred continuously regardless of whether the participant needed support or linking. For instance, even Andy, who just needed a point in the right direction, had face-to-face contact with his link worker every couple of months while in active intervention.

Intervention continuity

Some participants in active intervention during fieldwork experienced a personalised trajectory tailored to their circumstances, similar to participants referred to WtW in its early stages. Particularly important is that these participants reported experiencing regular contact with the same link worker. For example, Patrick had regular contact with his link worker despite saying that he did not really have a need for the intervention (recall he was ‘recruited’ into WtW by the health-care assistant). Nonetheless, he appreciated his link worker ‘checking in’ with him from time to time:

She rang me last week, ‘Are you alright Patrick?’. And all this. I think it’s nice when people . . . You get to know them, and they call you by your first name.

Yes.

It’s nice to know they’re there if they could do anything, if there was anything that I wanted but I don’t really.

Patrick, first interview, 16 March 2019

Throughout fieldwork, Patrick’s link worker ‘checked in’ by telephone or by home visit every couple of months. Patrick’s link worker also got to know his partner well, who said:

He’s got somebody to fall back on because she said, ‘Any time . . . Anything you want’. I suppose that gives you a bit of a boost because you’ve got a bit of safety net.

Patrick’s partner interview, 4 September 2019

Towards the end of the fieldwork, Patrick learnt that his health condition had seriously deteriorated. Because of the relatively frequent contact that he had with his link worker, his change in circumstance was picked up by the intervention soon after his diagnosis; therefore, when Patrick indicated that he required bathroom adaptations, his link worker was on hand to offer support. His link worker got in touch with the appropriate service:

When she came back and said, ‘I’ve been in touch and it’s going to be 18 months or 2 years’. I went, ‘No. I could be dead by then’. . . . I was joking and laughing, and she was. Then she phoned that woman back . . . She said, ‘But I told her you were a lovely man’.

Patrick, exit interview, 7 July 2020

Importantly, the lack of continuity in Patrick’s care did not relate to a lack of contact with his link worker, but was a consequence of inadequately funded public sector services.

Sandeep also experienced regular contact with his link worker. He offered a clear evaluation of his experience at first interview:

I’ve seen him [link worker] probably once. I talk to him quite a few times, just over the phone. He rang me and asked me how I was doing with my daily activities and well-being. I feel like a friend, when he talks. I don’t normally talk to any strangers about how I’m feeling unless it’s a professional that I need to talk to. So, that’s the only time I talk . . . I feel that someone is taking notice and asking me. I feel that if somebody’s helping you once, you try to feel proud of it.

Sandeep, referred January 2018, first interview, 11 March 2019

When Sandeep, who drove for a living, found it increasingly difficult to sit for long periods, he began to reduce his working hours. As with Patrick, because Sandeep was in regular contact with his link worker, his change in circumstance was quickly attended to, and his link worker supported him with an application to PIP and referrals to two other services. Because an established rapport had been fostered with his link worker, Sandeep, who does not ‘normally talk to strangers’, felt comfortable in seeking out the support of his link worker, who felt like ‘a friend’.

Jill also illustrates this dynamic (referred August 2018). At first interview, Jill had no understanding of WtW. She confused the link worker with the diabetes nurse and after several prompts recalled ‘I went to see somebody at the doctor’s surgery’. During fieldwork, and 10 months after their first meeting, Jill received a telephone call from her link worker. Coincidently, the call came soon after Jill’s father passed away, at a time when her mother’s health was deteriorating and her own health issues were in flux. Importantly, the telephone call, to conduct a WBS, acted as a catalyst for activity and triggered a series of four face-to-face meetings over many months. As well as offering emotional support, the focus of these interactions centred on providing practical support for Jill’s mother, such as help with form filling and accessing a disability badge. As Jill’s husband reported:

Sort of practical help that he was offering as well as the just having somebody to talk to as well I think and just reinforce the healthy eating side of things.

Jill’s husband interview, 23 September 2019

Contrary to Jill’s lack of clarity about WtW at first interview, by the final interview her narrative shifted from a complete lack of awareness to a clear evaluation of her link worker:

He offered support, he said . . . ‘Well all these forms we were filling in for my mum’ . . . If we had any problems he would help with them, and he was really – he’s a nice lad. Easy to talk to. But yes, he offered us these walking groups and stuff, but again it’s a group. I didn’t do stuff like that. Yes, I like [link worker].

Jill, exit interview, 3 July 2020

Evidently, although it took 10 months for their relationship to be initiated, the continuity of having the same ‘nice lad’, who was ‘easy to talk to’, together with the relevance of a series of face-to-face interactions to Jill’s context, rendered the intervention memorable for Jill in ways that did not surface in many others, as the following sections will explore.

Unsupported linking

Participants in active intervention at the start of fieldwork reported a tendency to be signposted to services and activities. Interestingly, none of these participants engaged with the signposted activities during the fieldwork period. In addition, participants experienced far less supported linking than previously observed. For instance, at first interview, Shirley discusses an impending gym referral, which her link worker had set in motion following her first WBS some months ago. When Shirley attends her gym induction, she is disappointed to learn that her high blood pressure resulted in the referral being postponed. When she finally commences the gym, it is clear that her link worker had been absent from the referral process bar the initial referral 6 months previously:

I ask if she’s heard from the link worker. She says, ‘Nope, I’ve never heard anything’. Apart from meeting her once in September and chatting on the phone about her gym referral before Christmas, Shirley’s had no contact. She says ‘I’d like to hear from her to find out what happens next’. I say, ‘What do you mean ‘next?’. She replies, ‘What happens after the gym?’.

Shirley, referred September 2018, meeting for coffee, 25 February 2019

At final interview, some 17 months later, Shirley explained, ‘I heard nothing after that’.

Shirley’s referral experience is in sharp contrast to, for example, Zaheer’s and Brenda’s experience of linking, as described above. Their retrospective narratives allude to a depth of support around linking, which is relatively absent from Shirley’s recollection that she ‘heard nothing’ despite her early expectations. This lack of supported linking appears not to relate to client need but to reflect a lack of contact, which in turn means that the intervention often failed to capture changes in circumstances, such as the postponement of Shirley’s referral. Furthermore, during fieldwork, it became apparent that frequency of contact varied according to provider and/or link worker rather than client. For instance, participants with one provider appeared to have less contact overall. Some participants referred to this provider were contacted only to complete a WBS, which was at times conducted late, meaning that there was a considerable gap between contacts. Moreover, a number of these participants had no face-to-face contact with a link worker following their first meeting. The most extreme example of this experience is Bobby, whose referral in July 2018 came just 8 days after his T2DM diagnosis. Following his first face-to-face link worker meeting, Bobby had two further telephone interactions with his link worker, each time to complete a WBS (WBS2, 15 minutes; WBS3, 10 minutes). By the end of fieldwork, some 24 months after his initial referral, he received a total of 1 hour 25 minutes from the intervention.

Labani’s experience also echoed this finding (referred November 2018). In our first interview (19 March 2019), Labani’s daughter, who was helping with translation, after several prompts recalled that they had met ‘someone in the GP’, in turn prompting Labani to recall that the link worker had offered support with her finances: ‘It is cold and we spend a lot and it isn’t enough, so they said they will help us’ [translated from Bengali]. Her daughter added that they had been expecting some information through the post. At the end of the interview, Labani asked Kate Gibson for benefits advice. On learning that they did not know how to contact the link worker, Kate signposted them to the surgery. When Kate returned to conduct the photo-elicitation interview (13 May 2019), Labani again asked for benefits advice. Kate texted Labani the provider’s office number, and Labani called Kate the following week to ask the name of her link worker. When Kate telephoned 1 month later, Labani explained that she had been in touch with the link worker and that she had been sent a ‘benefits pack’: ‘you can come and I will show it to you’ she added. The field notes below describe Kate’s visit the following week:

Labani has been to the GP surgery to meet with the link worker and the link worker had asked about her health and Labani had asked about benefits. I ask if they received help with benefits and she gets out the contents of an envelope for me to look at. I notice a WtW compliments slip which reads, ‘As requested, please find enclosed information . . .’. Labani shows me two A4 printed pages. . . . Her husband is also in the room by this point and on learning that they do not know what to do with the information, I explain to them that ‘Welfare Rights’ refers to benefits advice and that I have brought something similar. I show them the list that I have brought. I point to [two local community resource centres] and ask if they know how to get there? They know both.

Labani, home visit, 24 June 2019

Three months later, neither the link worker nor Labani and her family had followed up on the welfare advice information:

I ask if she still has to go to the doctor’s every couple of weeks for a sick note for benefits. ‘They have stopped the benefits’ she says. I learn that because they were [abroad] for 5 weeks their benefits were stopped and don’t start again until mid-October. ‘How are you managing?’ I ask. ‘We have borrowed some more money from here and there’, she tells me. I ask if they managed to do anything about the debt advice places I gave them last time. Her husband tells me they lost the paper.

Labani, home visit, 24 September 2019

Rather than receiving support to access benefits advice, Labani, like several others, was signposted. However, the conversation that Labani and her family have with Kate around accessing benefits advice suggests that, with continuity and persistence, this link could have come to fruition. Particularly important is the fact that, unlike the examples of Sandeep, Patrick and Jill described above, whose link workers clearly responded to their biographical disruptions, Labani’s link worker was not ‘there’ to know that her benefits had been stopped. This lack of continuity is particularly interesting given that Labani was in touch with the same link worker throughout her journey: a rare experience across the sample. The following section will explore the absence of embedded, continuous and structured link worker–client relationships.

Intervention interruptions

Given that participants sometimes engaged with the intervention for a lengthy period, it is unsurprising that link worker change was a common experience. However, in analysing the impacts of changing link workers, a theme of temporal drift once again emerges.

Many participants did not report having an established rapport with their link worker and, therefore, a change in link worker often had no negative impact on their experience. On the contrary, for some, a change in link worker resulted in a more effective intervention. For example, Steve (referred November 2018), who lives with multiple LTCs and was experiencing housing difficulties, received one telephone call from his link worker shortly after his first meeting. He was then contacted 10 months later by a new link worker, who organised a talking therapies referral:

He went to see a ‘different lass’ a while back. ‘She was quite good actually. She got us a . . .’. He spends several minutes trying to recall the word he was looking for. Eventually it comes to him and he tells me that he’s been seeing a counsellor. The link worker had asked if seeing someone would help him and he told her that he had seen people in the past but that had stopped now. So, she referred him to [service] and within a couple of weeks he was seeing a counsellor.

Steve, telephone call, 18 May 2020

A change of link worker initiated intervention activity for Brenda also. Brenda had previously understood that she was no longer in WtW and was, therefore, surprised to receive a telephone call ‘out of the blue’ from a new link worker. She explains below:

Brenda told her [the new link worker] that the last person she’d seen had said that there was nothing more WtW could do for her so had been discharged. The new link worker explained that she wasn’t sure what happened there because they don’t normally discharge people like that. So they did everything over the phone, including a star [WBS6]. Afterwards the link worker asked if there is anything she’d be interested in doing and Brenda had said ‘Yes, I’d like to join [name of gym] again’. So the new link worker has obviously organised it she tells me as she received a letter inviting her to go to the gym.

Brenda, telephone call, 24 April 2019

Neither Brenda nor Steve position a change of link worker as having a disruptive effect on their WtW journey. For Brenda, who was referred in December 2015, this lack of impact is in contrast to her early experiences of changing link worker, who we note above supported her during her first gym referral:

He [the link worker] was really nice. He was keen, he was interested, and then he just said to us [me], ‘I’ll not see you on your next appointment’, and then he was explaining, and I said, ‘Well, it’s stupid. If you’re here and your clients are happy with you, why are they moving you?’.

Brenda, first interview, 5 February 2019

In fact, many participants referred into WtW in its early stages alluded to a break in continuity triggered by a change in link worker, as the following excerpt demonstrates:

Do you feel like you know your link worker?

The first one and the second one, because I spent quite a bit of time with the first one and the second one. The first one, I spent more time with. I got to know that lady a bit more, and I felt at ease and everything with her. It was the same, as time went on, with the second one, LW17. LW23, it’s like a one-off thing. Sometimes he calls me after 2 months or 3 months or 6 months. It’s not like a progression every week, kind of thing, you know.

Eddie, exit interview, 6 July 2020

Although this account highlights the ways that link worker change resulted in a break in established support, as Eddie indicates, with time and continuity new relationships were built, thus delineating the importance of attending to the link worker–client relationship. Moreover, in addition to underlining the break in continuity that occurred when changing link workers, Eddie’s retrospective narrative further emphasises the extent to which over time link workers have become less embedded in client journeys. It is notable that he alludes to a lack of rapport with his later link worker. To some extent, this is to be expected as these clients gradually move towards the later stages of their journey; however, as we have noted above, a lack of contact also surfaced in the experiences of some participants who had been with the intervention for a shorter time.

Some participants had been with WtW for a long period of time; indeed, six participants were referred into WtW during its first year of operation. For some, this could be attributed to their particular contexts and their long tenure reflected the intensity of their ‘active intervention’ stage. For example, Masood required considerable support to be linked into activities. However, not all participants required such long-term support and, therefore, the long duration of their time with WtW was surprising. For example, following Geetha’s first WBS, she consistently scored 40 out of 40 for the duration of her 3-year journey. Furthermore, for this group of participants, the overall decline in contact was magnified during their later years with the intervention. Rather than a ‘natural decline in frequency of contact’,15 contact became unpredictable and declined sharply. For example, in contrast to their experience of active intervention, Zaheer and Masood, both discharged at WBS7, had not been contacted since completing WBS6 12 months previously. In the extract below, it is clear that Zaheer’s discharge came as a surprise:

I didn’t even know I had to be discharged. It’s just like, you know, when you go to the doctor and you are fine and you have no contact until the next time you go.

Zaheer, exit interview, 3 July 2020

Tracy also illustrates this dynamic. After completing WBS6 (when she scored lower than her previous star), she learnt that she would be discharged in the near future:

She [LW9] phoned to say that they won’t be finishing yet, but they will be finishing soon. Like I said to her, am I getting pushed to the sidelines? Am I not getting any support anymore? . . . I just feel like I’ll go back to square one.

Tracy, referred May 2017, exit interview, 9 July 2020

Particularly striking is that Tracy and Zaheer’s telephone calls were during the first wave of COVID-19. At that time, both Tracy and Zaheer were struggling to manage their LTCs, and their mental health was in decline because of deteriorating domestic relationships and the loss of their limited social connections outside the family home. These narratives about discharge, together with the apparent shift towards contacts driven by WBS completion, further highlight the ways in which the intervention has drifted away from its original provision to deliver holistic, personalised social prescribing tailored to individual circumstances. Furthermore, the finding that many participants were not in regular contact but appeared to be ‘kept open’ until the final possible WBS (WBS7) suggests that their ‘journey’, and especially their exit, was defined by measurable outputs that generated funding for providers (i.e. completed WBSs) rather than their personalised requirements. To conclude this first section, we now turn to focus on ‘the group’ as an alternative means of fostering a continuous experience of WtW.

Onward referrals

On the whole, in contrast to the recalled experiences of participants who had been with the intervention prior to fieldwork, more recently engaged participants were not linked into smaller community groups during fieldwork. Although some participants were signposted to such groups, as we note above, participants were unlikely to engage with signposted activities. Instead, participants were offered formal referrals into larger organisations. These referrals were often time limited and either entailed a one-off appointment, for instance to discuss welfare rights or debt advice, or involved participants participating in another intervention, for instance talking therapies, exercise on referral, and LTC management courses.

Some participants had attended weekly groups that were hosted by each provider: an exercise group and a social support group. These groups were central to these participants’ understanding and experience of WtW. In fact, for these participants, the group was the intervention. As one participant stated, ‘groups like that, WtW, the community needs them. Not wants them, needs them’. In the data presented in the remainder of this section, we have further anonymised participants to break links with their biographies and intervention histories.

One provider had previously hosted exercise groups, which were valued as a means to foster social connections. As one participant summed up:

I like to go to the group. It reduces isolation, when you sit by yourself and then you meet other people.

During fieldwork, Kate Gibson frequently attended two social support groups run by the providers. For one participant, this group was a lifeline:

Before WtW [the group], I didn’t have a life. So, it fits in quite well. Now, I’m introducing, I start to get to know people. I didn’t have anything. I just felt like a recluse.

The social support groups were an important link to further activities and other services. For instance, a local organisation delivered a healthy eating workshop and another local charity delivered a session about welfare benefits. Nevertheless, the interventional focus of behaviour change was highly apparent:

[Facilitator] gets the group started by reading out the welcome statement. She reads ‘Hello my name is . . .’, and then reads the aims of the group: ‘to help each other to help ourselves to take individual responsibility for our health and to motivate each other to make changes and to follow our WtW’.

Support group, A, field notes, 20 May 2019

Participants were encouraged to follow the aims of the group and, apart from initial check-ins, they often took part in various activities. For instance:

[Facilitator] hands out the Balanced Life exercise. She writes her answers down on the board as a starting point and we work separately to add our activities into three columns: self-maintenance, self-development and fun. [Facilitator] has gardening in her fun column – [attendee 1 (A1)] says that he and his wife used to love gardening but people would steal the ornaments, even in the daytime. [A2] says that his wife used to have pots out the front of his house and they were stolen too. [A3] says that she tries to research information about hearts because she doesn’t know anything about them.

Support group, A, field notes, 20 May 2019

Attendees often ‘tinkered’125 with health advice and shared stories, which reinforced, encouraged and sometimes challenged each other. For instance:

A7 says that she was doing ok, but is anxious about the weather turning; last winter she was stuck in the house for 3 weeks. She can’t breathe in the cold air so she is making sure she has tins in. Her sister had gone to the shops for her last winter, but it worries her because she shouldn’t leave the house especially because she’s asthmatic now. A6 asks if she is still worried about choking. A little bit, A7 says. She says following the advice of A5 she had tried lying down and tilting her head back to open her airway and it worked. A5 adds that his brother has emphysema and swears by sticking his head in the fridge before he leaves the house. The rationale being that you inhale the cold air and it’s not such a shock for your lungs when you leave the house. A7 says she’ll definitely try it. She finishes her check-in by adding that she had managed an extra minute on the treadmill last week. Although it’s only a minute she is chuffed with that because it puts her at four minutes now. The rest of the group agree that it was a fantastic achievement.

Support group, B, field notes, 20 November 2018

As a space in which health and sociality were inter-related (sometimes explicit, sometimes not), group members felt listened to and, for this reason, these settings were an important space for peer support. Moreover, they were a space in which peer support and interventional support could come together, as the following field notes detail:

The conversation moves to [name] pool opening and the link worker says ‘I thought of you [A3] because I know you like swimming’. I thought [A3] had nodded off again but she says ‘I used to. I used to have my lifesavers’. ‘What would you need to help you to go?’ the link worker asks. ‘I don’t know’, she says, ‘not right now’. The only other female attendee [A4] says she’d go with A3 and the link worker suggests they call in for a coffee and see what it’s like before they go. A2 interrupts with another story and while he’s talking A4 quietly asks A3 what size swimmers [bathing costume] she is and offers to see if she has anything at home she can dig out for her.

Support group, A, field notes, 29 July 2019

Not only does the quiet and tactful conversation between the only two female attendees about body size illuminate their shared and mutual recognition of the potential discomfort of going to a swimming pool in a large female body, but the supportive role of the link worker is integral to opening up the possibility of swimming in the first place.

These groups enabled the clients who attended them to have regular and embedded contact with a link worker and, consequently, the link worker was continually abreast of participants’ contexts. It is notable that the only participant to be accompanied to an activity during fieldwork was an attendee at a social support group. Echoing the findings thus far, this participant had experienced a gradual drift towards unsupported signposting. For instance, while she recalled ‘feeling safer’ when a link worker accompanied her to a home appointment for debt advice in 2017, she also recalled a recent experience of being signposted to information by a different link worker. As with other participants, the participant did not follow up on the signposts. Nevertheless, through the regular contact that this participant had with a link worker at the social support group, it became apparent that she was keen to attend a local community centre but had never ‘dared’ go in because she was ‘too nervous’. Having fostered a close rapport with the link worker through the social groups, the link worker then accompanied her to the community centre.

Ironically, despite their capacity to both create a synergy with intervention and foster community connectivity, both providers withdrew from these groups. When the exercise group ended, this came as a great disappointment to one participant who had attended regularly:

We were so close to the worker, actually, who did the exercise. Upstairs is the office. They usually go downstairs, do the exercise and go back home. Maybe they haven’t got enough staff as well. I don’t know.

For the social support groups, the link workers handed over the reins to trained volunteers and gradually withdrew. The effects of this differed according to group. One group that had been established for 2 years was well attended by 25 regulars, all of whom had developed close bonds. This group identified the withdrawal of WtW as a positive experience because it left them free of restrictions. Even during the COVID-19 pandemic, members of this group continued to keep in digital contact via WhatsApp (Meta Platforms, Inc., Menlo Park, CA, USA). The other group fared less well. Having been set up only 6 months prior to the provider’s withdrawal, apart from the two volunteer co-ordinators it was attended by just four regulars. There was a genuine concern that the group would ‘fold’, and indeed a lack of social connections established there meant that during lockdown attendees did not remain in contact. There was a feeling that a valuable connection to WtW had been lost. As one participant summed up:

Well, that has stopped, hasn’t it, the WtW? I don’t know why. Since the pullback from the group, there’s not much of the support left there. I feel like they’re forgetting.

Stable contexts

We first present the stories of Andy, Geetha and Jill, who we suggest engage with their health and WtW from an advantageous social context.

Jill

Jill also owns her own home in a quiet residential area. Jill, who works part time, lives there with her husband. As with Andy and Geetha, Jill was referred into WtW shortly after her T2DM diagnosis and, as outlined in Intervention continuity, WtW was supportive when her mother’s health deteriorated shortly after her father died.

Jill’s diagnosis of T2DM prompted her to exercise more and re-join Slimming World^® (Alfreton, UK). She is well supported by her husband, whose use of ‘we’ in the excerpt below emphasises that Jill’s health is a shared family project:

We’re trying to improve her health through exercise, diet. It hasn’t been helped recently by the death of her dad and her mam’s hospitalisation. Hopefully that will get better as time goes on.

Husband, family interview, 23 September 2019

Jill’s journey through dieting and exercise was interrupted by her caring responsibilities for her ageing parents. Her weekly ‘menu’, which she forward planned with Slimming World recipe books, became increasingly difficult to follow when her parents were in hospital. Her father was in hospital at the time of her photo-elicitation interview:

With being at the hospital most days, the menu that we tend to do has gone out of the wall; however, we are cooking tonight what we should be doing. But we had a Chinese on Friday, because it was yon time [late], when we got back. Well, I was in the hospital from half past eight [morning] until about eight o’clock at night, and it was just . . . didn’t want to start cooking then.

Photo-elicitation interview, 29 April 2019

Increasingly, Jill found that she was having ‘bad days’, as indicated by her photograph (Figure 9) and quotation:

That’s quiche and tomatoes. That was probably a bad day. In fact, it was: having quiche. And bread and butter. That would probably be one of my bad days.

FIGURE 9.

Jill’s photograph of quiche and tomatoes.

By her exit interview, when Jill’s mum had settled into a nursing home, Jill had started to make progress with her diet:

I’m still on Slimming World, I’ve lost 11 lb up to now.

Oh brill.

Well yes, but it’s not great considering we’re in lockdown and I should be able to do better.

Exit interview, 3 July 2020

For Jill, prioritising caring for her parents meant that her everyday and routine eating activities were disrupted, which, in turn, affected her diabetes management. It is notable that she no longer had the time (or energy) to plan and cook her healthy ‘menu’ during this unstable time. In spite of her priorities being elsewhere, Jill appears simultaneously to blame herself for individual and moral shortcomings attached to having a ‘bad day’. In addition, it is notable that, although Jill’s self-care practices ‘slip’, once she had established appropriate care for her mum she was able to re-establish her routinised healthy eating practices; despite that she chastised herself for not ‘doing better’. With the support of her husband and the link worker, Jill regained her healthy momentum and returned to prioritising her health. Jill’s experience of the disruption of a well-established routine underlines the importance of stability in managing future health. Not all participants had the privilege of prioritising their self-care most of the time, as the following section will show.

Uncertain contexts

Eddie

Eddie lives alone in a one-bedroom flat in a social housing complex. He is unemployed, suffers from depression and anxiety, and has had T2DM since 2005. He tries to ‘eat the right foods’, although sometimes ‘it gets on top’ of him because the ‘right foods’ ‘are expensive to buy’. Referred into WtW in 2015, Eddie’s WtW journey has involved a number of activities and referrals. For instance, he was referred to a local gym; however (and despite several text reminders), he never went. He was encouraged to attend a social group, but after a series of cancellations, the link worker concluded that he was ‘uncommitted’. Eddie was then referred for counselling sessions, which he regularly attended.

Eddie’s photograph (Figure 10) and accompanying words emphasise the weighty reality of his everyday life:

FIGURE 10.

Eddie’s photograph of the ground.

Sometimes when I’m out, you see, I don’t like to look ahead. I look down.

Photo-elicitation interview, 7 May 2019

That Eddie ‘doesn’t like to look ahead’ is partly because of the benefits system, which fixes him as powerless, as indicated in the following excerpt:

I can’t do anything at this time until I hear from these people [Department for Work and Pensions]. Like, I’m in limbo, you know, I’m not one thing or the other, so I just have to wait until they come with their decisions on yes or no. So, it’s like they’re controlling my life at the moment.

First interview, 6 March 2019

As fieldwork unfolded, Eddie increasingly struggled to survive on Universal Credit. He contacts the intervention requesting food bank vouchers on an almost weekly basis and, with the threat of eviction, becomes increasingly reliant on food banks. Ultimately, however, the intervention questions the frequency and long-term usage of the vouchers and appears reluctant to issue them so unequivocally. The field notes below detail an occasion when Eddie’s link worker telephoned him after he contacted the intervention requesting a voucher having just collected one the week before:

He looks at the screen, holds up his finger as if to pause our conversation and says it is the link worker. ‘Actually I don’t like going to that one, I usually go to the one up the top . . . OK, see you in a bit’. He tells me with a raised eyebrow that the link worker wants to meet to look over his expenditures with him after the Debt Relief Order (DRO) is sorted – he doesn’t look too pleased. For now though he can go up to the office and get the foodbank voucher.

At the café with Eddie, 14 November 2019

As with Aisha, the ‘present-focused’, reactive nature of Eddie’s everyday practice appears at odds with WtW’s expectation that clients co-produce a future-facing ‘action plan’ to improve their health. Poverty, and its associated lack of stability, interrupts his engagement in ‘sustained healthy behaviour change’ (see Appendix 15, Figures 31 and 32). We next turn to Tracy, who also has limited resources to hand.

Background

During the time of this additional study (April–July 2020), participants were living and providing services under changing government COVID-19 restrictions. On 23 March 2020, the UK government moved into a nationwide ‘lockdown’ as a result of increasing numbers of cases of COVID-19. The ‘stay at home’ restrictions (23 March–12 May) directed people to stay at home throughout this period except for essential purchases, essential work travel (if remote work was not possible), medical needs, one exercise session per day (alone or with household members) and providing care for others. Employees and employers who could work remotely were asked to do so, and others had to cease working. The ‘stay alert’ restrictions (13 May–1 June) stated that anyone who could not work from home was actively encouraged to go back to work, and people were allowed to take ‘unlimited’ outdoor exercise as long as they did not meet with more than one other person from another household. Restrictions were relaxed from 1 June, which included enabling up to six people to meet outside and in private gardens, and clinically extremely vulnerable people to spend time outdoors. ‘Support bubbles’ were allowed from 13 June, with single-adult households able to join with one other household and, for this reason, many of our participants had not experienced a support bubble at the time of interview. 138

Worldwide, COVID-19 is more likely to be experienced severely in older people and those with LTCs, particularly multimorbidities: this is the target group of WtW. 139,140 Consequently, the UK Government’s COVID-19 response included advising people with certain single or multimorbidities (2.2 million clinically extremely vulnerable people) to ‘shield’ in their homes for 12 weeks from 22 March 2020 and to avoid all in-person contact with others, even in their own households. 141 Within this climate, social prescribing has received attention as an intervention that could be remodelled to meet exacerbated public health challenges, including social isolation. 12 For this reason, it was valuable to extend the SPRING_NE study to explore how the intervention functioned during the pandemic, from client, link worker and managerial perspectives.

Aims and objectives

This chapter explores how the intervention changed during the first wave of the COVID-19 pandemic from the service provider perspective and how the intervention was experienced by clients during this period. The objectives are to:

1. describe what changes, challenges and opportunities occurred in delivering the intervention to clients during the first wave of the COVID-19 pandemic

2. explore how clients experienced the intervention during the first wave of the COVID-19 pandemic and what role it played in clients’ overall experience of the pandemic.

Sampling and recruitment

Client participants included those already in the EQ-5D-5L study who were contacted to complete their 12-month follow-up questionnaire during the lockdown period and who agreed to participate in an additional telephone interview about their experiences during the pandemic (n = 29) and those already in the client ethnography (n = 15) (see Chapter 6). Service provider participants were those already in the link worker ethnography (see Chapter 5), and additional link workers and managerial staff recruited via local connections and snowballing. Participants received information leaflets (see Report Supplementary Material 2) and consent (see Report Supplementary Material 2) was recorded verbally prior to the interview, because of lockdown restrictions.

Data collection

Telephone interviews were recorded with permission. Interviews with clients comprised administering a third iteration of the EQ-5D-5L questionnaire and conducting a semistructured interview. The topic guide (see Report Supplementary Material 2) covered experiences of ill-health; impacts of COVID-19 on interviewees, their LTCs and their household members; support needed or received including social prescribing; and feelings about life during and after lockdown. Demographic data were collected on age, gender, ethnicity, employment, education, household income, housing tenure/type/composition and availability of private outside space (see Report Supplementary Material 2). Service provider interviews followed topic guides that were tailored to each individual but included questions regarding their employing organisation’s philosophy, adaptations to their services since March 2020, and organisational and individual challenges and opportunities. Demographic data were not collected. Data were collected by Stephanie L Morris, Jayne Jeffries, Kate Gibson, Suzanne Moffatt and Tessa M Pollard.

Data analysis

Interview data were transcribed, read and re-read, and an inductive coding framework was developed. Line-by-line coding was used to identify descriptive and conceptual themes, using NVivo to assist with data management. Data were analysed as a ‘snapshot’ in time during the changeable circumstances of the pandemic. Data analysis was led by Stephanie L Morris, in discussion with Josephine M Wildman, Tessa M Pollard, Suzanne Moffatt, Bethan Griffith and Kate Gibson.

Service providers

The social prescribing service providers interviewed were eight service managerial staff and five link workers or senior link workers and a managerial staff member from the WtW umbrella organisation. The quotations in this section are attributed by anonymous codes using the following template: pseudonym name, provider (A/B), interview date.

The data reported here provide a ‘snapshot’ of the temporary WtW service changes that inevitably altered over time as the pandemic continued. This section builds on Chapter 5 by providing an account of changes, challenges and opportunities, as reported by service providers in semistructured interviews between May and September 2020.

Clients

The following sections detail how the intervention was experienced by clients and draw attention to some of the overlapping themes across the client and service provider data sets. We begin by describing the sample and providing a contextual description of how COVID-19 restrictions affected the health and well-being of WtW clients, before presenting the key findings regarding the role WtW played in clients’ experiences of the first wave of the pandemic.

The shaping of a social prescribing intervention

Ways to Wellness was the ‘UK’s first health SIB’ set up to deliver social prescribing. 73 The SIB was considered vital in facilitating the commissioning of the service prior to the UK Government’s commitment to fund social prescribing within the NHS. The Deep Dive Report154 on the SIB, commissioned by the Commissioning Better Outcomes Fund, which aims to encourage the development of SIBs, identified key opportunities and challenges faced by the SIB, including the ‘immense’ challenge of identifying appropriate outcome metrics. Our ethnographic data indicate clearly that the choice of outcome metrics used to allocate payments to the provider organisations had an impact on intervention delivery.

Payments to providers were based on engagement with clients, as measured by the completion of WBSs, so as to avoid incentivising providers to exaggerate the impact of their services and to enable small VCSE organisations to deliver the service by minimising financial risk. 73 As we have seen (see The intervention), the payment mechanism was rebalanced after 2 years, coinciding with the final withdrawal of two of the original provider organisations by March 2018. Over time, the payment structure generated a focus on obtaining referrals into the service and, for clients within the service, on ensuring the completion of WBSs at the cost of other elements of the intervention. The fact that some of the GPs in practices affiliated to the intervention did not directly refer to WtW contributed to an increased emphasis on the generation of referrals by the providers themselves. At the time of the ethnographic fieldwork (November 2018–July 2020), link workers were observed to spend much of their time generating referrals or chasing up WBSs with existing clients, and many link workers expressed concerns about the need to focus on meeting targets for the completion of ‘stars’. For many participants, this focus played out in a feeling of being ‘recruited’, rather than referred, into the intervention. This ‘out of the blue’ contact contrasted with the experiences of participants who had been referred in the early stages of the intervention, who recalled more intensive interactions with their link workers.

It is important to note that more intensive support was by no means absent from the intervention during the period of our study, illustrated by the case of Gill, whose link worker was ‘the kind of person that I could actually talk to’ in the months prior to the COVID-19 pandemic and was particularly vital during the pandemic when Gill’s partner died (see Chapter 7). However, it is also noteworthy that during the pandemic the need for a different set of priorities was recognised. To enable a shift in the service, the payment model was changed to explicitly allow the intervention to respond flexibly to patients’ needs.

We did observe differences between provider organisations in terms of link worker caseload and client contact, and across the period of fieldwork it became apparent that the second provider was increasing its focus on referrals and the completion of ‘stars’, with training days focused on these aspects of their work.

The funding structure was, therefore, fundamentally responsible for a shift over time towards a less intensive, less responsive and more linear intervention, as link worker time was invested in generating referrals and ensuring the completion of WBSs. These findings accord with previous research showing that pressure to generate outcome metrics shapes the operationalisation of SIBs. 155,156 Similarly, our findings resonate with Edmiston and Nicholl’s157 findings that a SIB-funded service will inevitably prioritise achieving its outcome metrics rather than continuity. In this case, the performative pressure emerged from the use of an outcome metric linked mainly to the number of clients recruited into and maintained within the intervention. Although this approach was generated partly by a concern not to impose a singular or narrow measure of success on the providers, the focus on numbers (output) itself distorted the intervention. 122 In essence, this is an example of the unintended consequences of an incentive that diverts providers ‘away from activities that are most needed towards activities that are most measurable’. 156 A focus on measurable outcomes is not unique to SIBs of course and is known to have had unintended consequences more generally in health service provision. 158

The original vision of the social prescribing intervention included an inherent tension between connecting and supporting individuals, tackling (as far as possible) the social determinants of health, and an emphasis on empowering individuals to manage their health and change their ‘lifestyle’. This was partially resolved over time by a drift towards prioritising individual-level behaviour change that aligned with the drift towards a focus on generating WBSs.

From the start, there was a recognition that different providers would bring different perspectives and strengths to the implementation of the intervention and, as described in Chapter 5, the two providers delivering the intervention at the time of our fieldwork each had their own ethos, with one explicitly oriented towards individual-level behaviour change. Link workers employed by this provider were encouraged towards a behaviour change approach and often described their role as motivating and empowering clients to achieve behaviour change. Often this meant focusing on encouraging clients to ‘take control’ of their health by co-creating lifestyle goals and signposting (often what we term ‘unsupported linking’) into local gyms and diet-related services. Other link workers were more oriented to offering wider support to their clients. The tension between these approaches surfaced repeatedly through our fieldwork, for example in the concerns expressed by some link workers about the need to avoid being a ‘support worker’. In responding to those experiencing lockdown in the spring of 2020, the contrast between the two approaches was brought into sharp focus. Working with clients through the threat of COVID-19 and social isolation of lockdown, the intervention’s umbrella organisation and the providers felt that they could be most valuable by aligning with the necessarily ‘present-focused’ concerns of clients (many of whom were already well rehearsed in living in the here and now), but this frustrated some link workers who were committed to the promotion of individual-level behaviour change.

We observed that clients recruited from later periods were less likely to experience the frequent contact and supported linking recalled by those who were recruited in the early years. Observations of link workers in the later years also suggested the dominance of a behaviour change approach. The individual behaviour change model aligned with the ‘lighter touch’ approach to working with clients that the funding model necessitated. Thus, both ethnographies found that the intervention had drifted away from the ‘holistic’ approach that previous qualitative research identified,1,54 towards an approach that shifts responsibility for change on to the individual. Such a drift towards lifestyle-focused approaches has previously been identified within other interventions,31,159,160 and has been observed to be driven partly by the pressures of targets and workload,31 although it is also a reflection of dominant discourses that construct ‘healthy lifestyles’ as the responsibility of individuals. 161 This is not to say that support was absent from the intervention as we observed it, but in general there was limited time available for such work. The average link worker caseload was 100, and an explicit emphasis on setting clients up to independently manage their own health that was narrower than the original vision for the scope of the intervention.

Social prescribing, social determinants of health and health inequalities

Resources shape access to health162–164 and participants with access to resources were more able to engage with the increasingly dominant unidirectional model of individual behaviour change. Warin et al. 135 suggest that people living in circumstances of deprivation are often constrained by ‘short horizons’. We too found that many disadvantaged participants were more inclined to concentrate on the immediacy of their precarious social circumstances, often at the expense of more future-focused health-enhancing practices.

Our findings, therefore, suggest that a drift towards signposting and an emphasis on individual-level behaviour change risks excluding some participants. NHS England states that ‘Active signposting works best for people who are confident and skilled enough to find their own way to services after a brief intervention’. 126 We suggest that this individualised focus ignores the wider context of people’s lives, especially the lives of those living with the greatest disadvantage. The intervention’s emphasis on ‘motivating’ clients to adopt healthy behaviour change and ‘have improved attitudes towards challenges’ (see Appendix 15, Figures 31 and 32) positions local context as an external factor to be addressed only inasmuch as it facilitates or constrains the achievement of the intervention’s objectives. 165 Hence, although we observed a number of referrals made by link workers to lifestyle interventions, such as local gyms and healthy eating programmes, which may have important impacts on individual health and well-being outcomes, they fall far short of addressing structural social determinants of health or impacting on health inequalities. 162

Furthermore, although we do find examples of link workers tailoring the intervention to address individual circumstances and supporting clients to navigate the social determinants of health, the intervention could not remove such structural barriers. Just as significant upstream factors, such as austerity-driven cuts, shaped clients’ priorities to engage with the intervention, so too did they shape what support link workers could offer. In this way, our qualitative findings demonstrate the implausibility of addressing health inequalities via a downstream intervention which emphasises individual lifestyle change. 31

In summary, the ethnographic WPs paint a detailed picture of a social prescribing service that was shaped by the systems in which it was embedded, and of clients whose ability to respond to the intervention was in turn affected by the way it was delivered, and by how it embedded within their own social worlds. Our findings include many examples of link workers providing a service that was greatly valued by clients, but there was also considerable heterogeneity within the intervention, and some clients received a ‘light touch’ version of social prescribing when they required a more intensive form of the intervention. Thus, we suggest that the drift towards a form of social prescribing offering limited time and support to clients and focusing on behaviour change constrained further an already limited opportunity to address the social determinants of health via a downstream intervention.

Work package 1 (health outcomes)

The quasi-experimental ITT design reflects the ‘real-world’ application of this heterogeneous and personalised intervention that measures the impact across the 4-year period for which data were available. The ITT analysis provides an estimate of the expected effect on outcomes for an individual drawn at random from the treatment practices. Taking account of issues associated with the referral process, non-compliance and wide individual heterogeneity, and by considering all individuals to be treated at the time that treatment becomes available, the strength of the ITT approach is that it overcomes a number of problems associated with observational data. The limitations of the study are (1) reduced sample size due to the refusal of seven general practices to share data (three treatment and four controls); (2) the incompleteness and unreliability of some of the QOF data (although levels of HbA_1c and blood pressure are well recorded thanks to QOF targets and incentives, this is not the case for BMI, cholesterol, smoking status or ethnicity data); (3) the fact that the only available measurement of treatment intensity was number of completed WBSs, which we know from the qualitative data is not an accurate reflection of the ‘dose’ of interaction with the intervention; and (4) the fact that data about comorbidity were restricted to those who determined intervention eligibility and did not, therefore, indicate the true extent of comorbidity, which we know to be greater. Despite these limitations, this is a robust design that includes large numbers of observations with sufficient power to detect effects.

Work package 1 (health-related quality of life)

The other parts of the evaluation focused only on people with a T2DM diagnosis, and a strength of the HRQoL study was the inclusion of everyone referred into the intervention over a 12-month period, irrespective of their diagnoses. An important finding was the overall poor self-rated health of this subgroup in comparison with national norms. The limitations are the (1) lower than anticipated baseline response rate of just under 55%, which reduced the study power; (2) weakness of a before-and-after design; and (3) requirement to account for the impact of COVID-19 and need to divide the sample into pre- and post-COVID-19 groups, further reducing the power of the study.

Work package 2 (cost-effectiveness)

To our knowledge, this is the first robust cost-effectiveness analysis of a UK social prescribing intervention to date using a well-established T2DM simulation model98 to provide a cost per QALY gained associated with the introduction of the intervention. The limitations are (1) reliance on an exploratory analysis with significant sensitivity analysis due to data limitations on the impact of the intervention on relevant risk factors; (2) the fact that effects were captured over a 4-year period only, rather than over the longer term; and (3) the inability to capture non-clinical outcomes related to social determinants of health.

Work package 3 (qualitative research)

A key strength of ethnography is the use of participant observation (complemented by interviews and focus groups) to gain a holistic understanding of practices within the context in which they take place, over an extended time period (i.e. 20 months for the client ethnography and 10 months for the link worker ethnography). Our use of an ethnographic approach allowed us to ‘see’ the intervention from the perspective of link workers and clients in the settings in which they worked and lived, rather than relying on self-report. 168 The use of two ethnographies focusing on experiences of the intervention from both link worker and client perspectives, in conjunction with interview data during the initial lockdown period, afforded a considerable degree of triangulation and assurance about the reliability of our interpretation of the observational data. A further strength was in the extended time frame that the intervention was observed from the client perspective. The limitations are (1) the reliance on relatively small numbers of research participants; (2) the premature curtailment of face-to-face fieldwork due to COVID-19, particularly impacting on the ‘shadowing’ fieldwork; (3) limited VCSE perspectives; and (4) not including those were eligible for the intervention but who did not participate, along with a lack of data from those involved in the early stages of the intervention but who did not remain engaged, potentially leading to a biased perspective on the early years of the intervention.

Journal articles

Wildman J, Wildman JM. Combining health and outcomes beyond health in complex evaluations of complex interventions: suggestions for economic evaluation. Value Health 2019;22:511–17.

Gibson K, Pollard TM, Moffatt S. Social prescribing and classed inequality: a journey of upward health mobility? Soc Sci Med 2021;280:114037.

Wildman J, Wildman JM. Evaluation of a community healthworker social prescribing program among UK patients with type 2 diabetes. JAMA Netw Open 2021;4:e2126236.

Gibson K, Moffatt S, Pollard TM. ‘He called me out of the blue’: an ethnographic exploration of contrasting temporalities in a social prescribing intervention. Sociol Health and Illn 2022, 1–18.

Morris SL, Gibson K, Wildman JM, Griffith B, Moffat S, Pollard TM. Social prescribing during the COVID-19 pandemic: a qualitative study of service providers’ and clients’ experiences. BMC Health Serv Res 2022;22:258.

Wildman JM, Morris S, Pollard T, Gibson K, Moffatt S. ‘I wouldn’t survive it, as simple as that’: syndemic vulnerability among people living with chronic non-communicable disease during the COVID-19 pandemic. Soc Sci Med Qual Res Health 2022;2:10032.

Morris SL, Wildman JM, Gibson K, Moffat S, Pollard TM. Managing disruption at a distance: unequal experiences of people living with long-term conditions during the COVID-19 pandemic. Soc Sci Med 2022;302:114963.

Conference presentations

Gibson K, Pollard T, Moffatt S. Social Prescribing and Classed Inequalities in Health: Exploring a Complex Relationship Using Ethnographic Methods. Presented at Society for Social Medicine & Population Health Virtual Conference, 9–11 September 2020.

Moffatt S. Why is NHS England so Keen on Social Prescribing and How Does it Impact on Health and Wellbeing? Presented at British Geriatrics Society Virtual Annual Conference, 27–28 November 2020.

Gibson K. Social Prescribing and Class: A Journey of Upward Health Mobility? Presented at Chronic Living Virtual Conference, 4–6 March 2021.

Griffith B, Moffatt S, Pollard T. Two Sides of the Same Coin? An Ethnographic Exploration of Primary Care Staff and Link Worker Perspectives on Social Prescribing. Society of Academic Primary Care 49th Annual Scientific Meeting (Living and Dying), 30 June–1 July 2021.

Griffith B, Moffatt S, Pollard T. Link Working at the Intersections: And Ethnographic Explorations of Delivering Social Prescribing in Primary Care. Society for Social Medicine and Population Health 65th Annual Scientific Meeting, 15–17 September 2021.

Morris S, Wildman J, Gibson K, Moffatt S, Pollard T. Managing Disruption at a Distance: Unequal Experiences of People Living with Long-term Conditions During the COVID-19 Pandemic. e-Poster P109 with 3-minute oral presentation. Society for Social Medicine and Population Health 65th Annual Scientific Meeting, 15–17 September 2021.

Quantitative data

Owing to the confidential nature of the quantitative data that were provided for the sole purpose of this study, these data cannot be shared.

Qualitative data

The ethnographic and interview data are not suitable for sharing beyond that contained within the report.

For queries relating to either data set, further information can be obtained from the corresponding author.