Effects of participating in community assets on quality of life and costs of care: longitudinal cohort study of older people in England

Objectives Improving outcomes for older people with long-term conditions and multimorbidity is a priority. Current policy commits to substantial expansion of social prescribing to community assets, such as charity, voluntary or community groups. We use longitudinal data to add to the limited evidence on whether this is associated with better quality of life or lower costs of care. Design Prospective 18-month cohort survey of self-reported participation in community assets and quality of life linked to administrative care records. Effects of starting and stopping participation estimated using double-robust estimation. Setting Participation in community asset facilities. Costs of primary and secondary care. Participants 4377 older people with long-term conditions. Intervention Participation in community assets. Primary and secondary outcome measures Quality-adjusted life years (QALYs), healthcare costs and social value estimated using net benefits. Results Starting to participate in community assets was associated with a 0.017 (95% CI 0.002 to 0.032) gain in QALYs after 6 months, 0.030 (95% CI 0.005 to 0.054) after 12 months and 0.056 (95% CI 0.017 to 0.094) after 18 months. Cumulative effects on care costs were negative in each time period: £−96 (95% CI £−512 to £321) at 6 months; £−283 (95% CI £−926 to £359) at 12 months; and £−453 (95% CI £−1366 to £461) at 18 months. The net benefit of starting to participate was £1956 (95% CI £209 to £3703) per participant at 18 months. Stopping participation was associated with larger negative impacts of −0.102 (95% CI −0.173 to −0.031) QALYs and £1335.33 (95% CI £112.85 to £2557.81) higher costs after 18 months. Conclusions Participation in community assets by older people with long-term conditions is associated with improved quality of life and reduced costs of care. Sustaining that participation is important because there are considerable health changes associated with stopping. The results support the inclusion of community assets as part of an integrated care model for older patients.

I, the Submitting Author has the right to grant and does grant on behalf of all authors of the Work (as defined in the below author licence), an exclusive licence and/or a non-exclusive licence for contributions from authors who are: i) UK Crown employees; ii) where BMJ has agreed a CC-BY licence shall apply, and/or iii) in accordance with the terms applicable for US Federal Government officers or employees acting as part of their official duties; on a worldwide, perpetual, irrevocable, royalty-free basis to BMJ Publishing Group Ltd ("BMJ") its licensees and where the relevant Journal is co-owned by BMJ to the co-owners of the Journal, to publish the Work in this journal and any other BMJ products and to exploit all rights, as set out in our licence.
The Submitting Author accepts and understands that any supply made under these terms is made by BMJ to the Submitting Author unless you are acting as an employee on behalf of your employer or a postgraduate student of an affiliated institution which is paying any applicable article publishing charge ("APC") for Open Access articles. Where the Submitting Author wishes to make the Work available on an Open Access basis (and intends to pay the relevant APC), the terms of reuse of such Open Access shall be governed by a Creative Commons licence -details of these licences and which Creative Commons licence will apply to this Work are set out in our licence referred to above.
Other than as permitted in any relevant BMJ Author's Self Archiving Policies, I confirm this Work has not been accepted for publication elsewhere, is not being considered for publication elsewhere and does not duplicate material already published. I confirm all authors consent to publication of this Work and authorise the granting of this licence. their health and wellbeing.' 6 This idea has recently been given new impetus with a 2 commitment in the Long Term Plan for the NHS in England to have over 1,000 trained social 3 prescribing link workers in post by March 2021 and to expand provision so that over 900,000 4 people will have been referred to social prescribing schemes by March 2024.

5
This rapid expansion of formal provision will occur without a strong evidence base. Although 6 reviews and qualitative work have suggested that that community assets improve the health 7 of participants 7,8 , there is limited quantitative evidence. 9 The evidence base for social 8 prescribing is equally limited and has yet to arrive at a consensus. 10 9 We previously evaluated an integrated care programme for older people which included a 10 programme to improve use of community assets. 9 We used data from a cohort of older 11 people to analyse cross-sectional associations between community asset participation, 12 health and health care utilisation. The evidence suggested that community asset 13 participation was associated with significant improvements in health and not significant 14 reductions in health care costs. However, the cross-sectional nature of the data meant that 15 we could not interpret the relationships as causal.

16
In this study, we analyse the relationships between community asset participation, health 17 and health care utilisation longitudinally, to provide a more rigorous assessment of the 18 causal impact of community asset participation. Using administrative health records further 19 strengthens the analysis presented here as it removes the reliance on recall. As well as 20 considering the uptake of community assets as a possible health enhancing activity, we 21 additionally examine the possibility of there being health decrements associated with 22 ceasing to participate in community assets. A priori, it is not expected that there will be equal 23 gains and equal reductions.  Table A2 (supplementary appendix).

4
Statistical methods 5 We used double-robust estimation 23 to estimate the impact of community asset participation 6 on (i) health related quality of life, (ii) costs of formal health care services, and (iii) net social 7 benefit. 22 8 Double-robust estimation is a form of treatment effects estimator that accounts for 9 observable factors that could influence treatment. The method combines a propensity score 10 model with a regression adjustment. The propensity score is obtained from a logistic 11 regression of community asset participation on baseline covariates. The inverse of this 12 propensity score is then used to weight the regression model for the outcome. 23 As long as 13 one model is correctly specified, the double-robust estimator produces unbiased results. 24,25 14 If both models are correctly specified, then double-robust estimator is both unbiased and 15 efficient. 26 16 The choice of control variables for both models is important. We provide a full list of all 17 variables included in both the treatment (propensity score) equation and the outcome

18
(regression adjustment) model in an online appendix Table A2.

19
Primary analysis 20 Our primary analysis focuses on the individuals who provided information on their 21 participation in community assets in all four waves of the survey. To assess if initial 22 community asset participation was associated with whether the respondent remained in the 23 sample, we ran a logistic model of drop-out as a function of baseline characteristics, 24 including health and community asset participation. We interacted baseline community asset 25 participation with all of the covariates to see if there were differential associations between drop-out and the covariates between those who did or did not participate in community 2 assets at baseline.

3
Uptake analysis 4 For the 6-month analysis, we defined the comparator group as those individuals who did not 5 participate in community assets at baseline and continued to not participate at the 6-month 6 follow-up. The treatment group consists of those individuals who did not participate in assets 7 at baseline but did report participation at 6-months. This is comparison A ( Table 1).

TABLE 1 HERE
9 For the 12-month and 18-month analyses the definition of the treatment group was more 14 In the 12-month and 18-month analyses, the comparator group is those individuals who 15 never participated (NNN or NNNN). The primary definition of treatment in the 12-month 16 analysis was NYY (comparison C) and in the 18-month analysis was NYYY (comparison E).
17 Cessation analysis 18 We followed a similar logic for estimating the effects of ceasing to participate in community 19 assets. For the 6-month analysis we defined the comparator group as those who always 20 participate and the treatment group as those individuals who initially participated at baseline 21 and then stopped by the 6-month follow-up; comparison F. The 12-month and 18-month 22 analyses followed a similar pattern, and are shown as comparisons H and J in Table 1. In a secondary analysis we relaxed the restriction that an individual had to remain in the 3 sample for all four waves. We included data from all individuals in the respective waves.

4
In another secondary analysis, we additionally considered the effects of participating in 5 community assets at the 12 or 18-month follow-up, regardless of what happened in the 6 interim periods. For the uptake analysis, these were comparisons B and D in Table 1. For 7 the cessation analysis, these were comparisons G and I. 8 9

RESULTS
10 Selected characteristics of the respondents at baseline are available in Table 2. Further 11 detail is provided in Table A2.
12 Participation in community assets over time 13 Figure 2 shows how many people participated in community assets at each wave. Participation in community assets increased over time ( Table 2). The largest increase in 16 participation occurred between baseline (53%) and the 6-month follow-up (57%). Mean 17 levels of health-related quality of life decrease over time for both participants and non-18 participants.

TABLE 2 HERE
20 Attrition analysis 21 The only significant predictors of drop-out from the cohort were older age and education.

22
However, the magnitude of their effects on drop-out were not significantly different between those who initially participated and those who initially did not participate in community 2 assets. The full regression results are presented in a supplementary appendix (Table A3). 3 Statistical tests of suitability of the propensity score 4 Figure A1 (supplementary appendix) shows the distributions of the propensity scores before 5 and after matching. Panel (a) shows the distributions for the uptake analysis and panel (b) 6 shows the distributions for the cessation analysis. In both cases, the matching considerably 7 improves the similarity between the control and treatment groups. There is a positive and statistically significant effect of starting community asset participation 10 on health-related quality of life (Table 3, panel (a)). The benefit of starting to participate in 11 community assets is a 0·017 QALY gain (95% CI: 0·002 to 0·032) compared to those who 12 never participate in assets at the 6-month follow-up. The effect of starting to participate in 13 community assets is a QALY gain of 0·030 (95% CI: 0·005 to 0·054) at the 12-month follow-14 up and a QALY gain of 0·056 (95% CI: 0·017 to 0·094) at 18 months.

15
Starting to participate in community assets reduced costs in the 6-month period by £96 (95%

6
When we considered the total costs of health-care utilisation, we found that stopping 7 participating in community assets led to large and statistically significant increases in health 8 care utilisation costs. In the 6-month period this increase was £689 (95% CI: £162 to £1216) 9 whereas in the 12-month and 18-momnth follow-ups these increases were £857 (95% CI: 10 £252 to £1463) and £1335 (95% CI: £113 to £2558), respectively.

11
Additionally, there were negative net-benefits (assuming a £20,000 NICE threshold) 12 associated with cessation. In the 6-month period this potential loss was £624 per-participant 16

17
The results using all available data on respondents are qualitatively similar in terms of 18 magnitude and statistical significance (Table A4).

19
Use of less strict definitions of uptake and cessation also produces similar results, but the 20 effects are typically smaller in magnitude (Table A5).

22
Our study involved a large sample of patients recruited and followed up over an 18-month 23 period. Although there was loss to follow-up, the overall rate of retention was reasonable. 24 We collected detailed data on asset use and health, with objective data on health care costs available from administrative records. We adopted rigorous methods for the estimation of 2 causal effects and found the main results were robust to several assumptions. 3 We additionally performed many robustness/sensitivity analyses where we changed the 4 variables include in the matching model. Our main results remained qualitatively similar in all 5 cases, and we concluded that our main findings were not sensitive to the choice of variables 6 used in the matching equation. 7 However, the study was conducted in a single region in the United Kingdom, in a population 8 of older people living in an area undergoing transformation of older people's services. Care 9 must therefore be taken in generalising from this context.

10
As we highlighted in previous work, objective data on the impact of increasing use of 11 community assets is limited 9 , and this paper therefore makes a significant contribution to this  Ageing (ELSA). They found that current use of social groups significantly predicted better 17 cognition. Their study differs from ours in that we are interested in health and health care 18 utilisation and we model the decision to partake in social groups and community assets.

19
Also using ELSA, Steffens et al 28   2 participation and depression. 29,30 They show, using various data sources, that that 3 membership of more clubs was associated with a lower probability of future depression and 4 that identification with a social group predicts recovery from depression. Our results are 5 consistent with this in that depression has been shown to be a major driver of health related 6 quality of life 31 and health care utilisation. 32 7 Social prescribing schemes play a key role in the NHS Long Term Plan. Although popular 8 with services and policy makers, a recent review of such schemes found significant issues 9 with the quality of the evidence base 10 , with only 2/15 evaluations having any sort of 10 comparator.

11
Our analytical methods provided a comparator group to better assess the impact of changes 12 in asset use. We assessed naturalistic changes in asset use in the context of a wider 13 integrated care initiative, which saw some patients starting to use assets, and others ceasing 14 use. It is plausible that at least some of this increased use reflected the wider integrated care

15
initiative that was being undertaken in the area, but this cannot be determined reliably. Our

16
analysis used a large sample and robust analytic methods, and was able to assess the 17 effects of starting and stopping asset use. However, we were not testing the impact of new 18 referrals to community assets, and we cannot be sure that the benefits of the changes we 19 assessed would necessarily translate to patients in formal social prescribing schemes.

20
Nevertheless, our results make an important contribution, given the policy interest in these 21 approaches and the limited evidence base.

22
Our results highlight that the effects of starting and stopping asset use are not symmetrical,

23
which suggests that equal attention needs to be given to these different processes. The

24
focus of social prescribing tends to be on the former, but our data suggests that it is 25 important to identify people whose use of assets stops. If such people can be identified and supported, the gains might be even greater, but it is not clear that the same schemes would 2 be suited for increasing use and maintaining use.
3 Unanswered questions and future research 4 As noted previously, the study was conducted in a single region of the UK, and the results 5 would need replication. Given that the benefits of asset use seemed to increase with time, 6 further long-term evaluation would also be indicated. Exploration of the reasons why people 7 stop using assets, and whether it can be identified and managed more effectively, would 8 also be a research priority.

9
Our results provide a robust assessment of the impacts of changes in the use of community 10 assets, and provide further impetus to calls for robust evaluation of their effects. There is a 11 legitimate debate as to whether the standard controlled trial is optimal for the assessment of

17
We used quasi-experimental methods to explore the impact of changing patterns of the use 18 of community assets in a population of older people living in an area that introduced an 19 integrated care initiative which sought to increase asset use.

20
We found that increasing use of community assets was associated with increased health  The effects of starting to use assets were not symmetrical with ceasing use, with the latter 2 associated with larger losses. This is important, as encouraging use among those who do 3 not currently use assets may require different policy and patient-level interventions to those 4 designed to encourage continued use.

5
The results support the inclusion of community assets as part of an integrated care model for

Consent for publication
Not applicable.

Data sharing statement
The data that support the findings of this study are available from the Principal Investigator of the original study but restrictions apply to the availability of these data, which were used under license for the current study, and so are not publicly available. Data are however available from the authors upon reasonable request and with permission.

Competing interests
None of the authors have any competing interests to declare.

19
Health and social care organisations were advised to support the development and use of 20 such assets among their populations, by mapping community assets and engaging in a 21 process of Asset Based Community Development 5 , to help the community increase the 22 health and well-being of its population using activities, skills, and assets within the 23 community.

24
The way in which health and social care organisations engage with community assets has 25 subsequently become more direct. In several areas, health and care professionals (as well 14 This rapid expansion of formal provision will occur without a strong evidence base. Although prescribing is equally limited and has yet to arrive at a consensus. 10 However, it is worth 20 noting that the evidence is still developing in this field, with ongoing qualitative and quantities 21 studies.

22
We previously evaluated an integrated care programme for older people which included a 23 programme to improve use of community assets. 9 We used data from a cohort of older 24 people to analyse cross-sectional associations between community asset participation,

25
health and health care utilisation. The evidence suggested that community asset 26 participation was associated with significant improvements in health and not significant 3 In this study, we analyse the relationships between community asset participation, health 4 and health care utilisation longitudinally, to provide a more rigorous assessment of the 5 causal impact of community asset participation. Using administrative health records further 6 strengthens the analysis presented here as it removes the reliance on recall. As well as 7 considering the uptake of community assets as a possible health enhancing activity, we 8 additionally examine the possibility of there being health decrements associated with 9 ceasing to participate in community assets. A priori, it is not expected that there will be equal 10 gains and equal reductions.

12
Data: cohort description 13 The data used in this analysis were made available as part of the National Institute of Health      Health-related quality of life (HRQoL) was measured using the Euro-QoL 5D-5L. 12,13 The

12
EQ-5D-5L is a generic preference-based measure of HRQoL covering five domains
14 Participants completed the EQ-5D-5L in the baseline, 6-month, 12-month, and 18-month EQ-5D-5L. 15 In a robustness check we used the newly developed algorithm for directly 20 calculating utility scores from the EQ-5D-5L. 16 21 Quality-adjusted life years (QALYs) were then calculated at the individual level using the

24
Health care utilisation  16 We applied a discount rate of 3.5% to the costs and benefits. 21 17 Net-benefit 18 As in our earlier work 9 , we defined net-benefits as an individual's QALY gain minus the cost 19 of their healthcare utilisation. 22 We used the two thresholds used by the National Institute for 20 Health and Care Excellence; namely £20,000 and £30,000 but focus mainly on the £20,000 21 threshold for reasons of brevity.

22
Community asset participation 23 Community asset participation was defined as a binary variable equal to one if an individual 24 reported participating in any one of a list of activities, and zero otherwise. The list of 1 community assets is included in a supplementary appendix, along with reported participation 2 rates over time (Table A1). 3 Demographic and socioeconomic characteristics 4 We controlled for gender and age using a series of 5-year age categories (ranging from 65-5 69 years, up to 85+years). The reference age group is 65-69 years. We also controlled for  Table A2 (supplementary appendix).

14
We used double-robust estimation 23 to estimate the impact of community asset participation propensity score is then used to weight the regression model for the outcome. 23 As long as 22 one model is correctly specified, the double-robust estimator produces unbiased results. 24  drop-out and the covariates between those who did or did not participate in community 17 assets at baseline.

18
Uptake analysis 19 For the 6-month analysis, we defined the comparator group as those individuals who did not 20 participate in community assets at baseline and continued to not participate at the 6-month 21 follow-up. The treatment group consists of those individuals who did not participate in assets 22 at baseline but did report participation at 6-months. This is comparison A ( Table 1). participation and non-participation, respectively. We focused on the 'best case scenario' in 5 the primary analyses.

6
In the 12-month and 18-month analyses, the comparator group is those individuals who   Table 1.

Secondary analyses
18 In a secondary analysis we relaxed the restriction that an individual had to remain in the 19 sample for all four waves. We included data from all individuals in the respective waves.

20
In another secondary analysis, we additionally considered the effects of participating in 21 community assets at the 12 or 18-month follow-up, regardless of what happened in the 22 interim periods. For the uptake analysis, these were comparisons B and D in Table 1. For 23 the cessation analysis, these were comparisons G and I.  Table 2. Further 2 detail is provided in Table A2.

RESULTS
3 Participation in community assets over time 4 Figure 2 shows how many people participated in community assets at each wave. Participation in community assets increased over time ( Table 2). The largest increase in 7 participation occurred between baseline (53%) and the 6-month follow-up (57%). Mean The only significant predictors of drop-out from the cohort were older age and education.

13
However, the magnitude of their effects on drop-out were not significantly different between 14 those who initially participated and those who initially did not participate in community 15 assets. The full regression results are presented in a supplementary appendix (Table A3). shows the distributions for the cessation analysis. In both cases, the matching considerably 20 improves the similarity between the control and treatment groups.

18
When we considered the total costs of health-care utilisation, we found that stopping 19 participating in community assets led to large and statistically significant increases in health 20 care utilisation costs. In the 6-month period this increase was £689 (95% CI: £162 to £1216)

23
Additionally, there were negative net-benefits (assuming a £20,000 NICE threshold) 24 associated with cessation. In the 6-month period this potential loss was £624 per-participant  3

4
The results using all available data on respondents are qualitatively similar in terms of 5 magnitude and statistical significance (Table A4).

6
Use of less strict definitions of uptake and cessation also produces similar results, but the 7 effects are typically smaller in magnitude (Table A5).

9
Our study involved a large sample of patients recruited and followed up over an 18-month 10 period. Although there was loss to follow-up, the overall rate of retention was reasonable.

11
We collected detailed data on asset use and health, with objective data on health care costs 12 available from administrative records. We adopted rigorous methods for the estimation of 13 causal effects and found the main results were robust to several assumptions.
14 We additionally performed many robustness/sensitivity analyses where we changed the   there was more integration of care within Salford, particularly during the study period.
2 Therefore, the results need to be interpreted in this context, where there has been significant 3 investment in community assets locally.

4
As we highlighted in previous work, objective data on the impact of increasing use of 5 community assets is limited 9 , and this paper therefore makes a significant contribution to this  Ageing (ELSA). They found that current use of social groups significantly predicted better 11 cognition. Their study differs from ours in that we are interested in health and health care 12 utilisation and we model the decision to partake in social groups and community assets.

13
Also using ELSA, Steffens et al 28  comparator. This evidence base is continually evolving, and we expect this to change given 5 a number of ongoing and planned evaluations.

6
Our analytical methods provided a comparator group to better assess the impact of changes 7 in asset use. We assessed naturalistic changes in asset use in the context of a wider 8 integrated care initiative, which saw some patients starting to use assets, and others ceasing 9 use. It is plausible that at least some of this increased use reflected the wider integrated care 10 initiative that was being undertaken in the area, but this cannot be determined reliably. Our

11
analysis used a large sample and robust analytic methods, and was able to assess the 12 effects of starting and stopping asset use. However, we were not testing the impact of new 13 referrals to community assets, and we cannot be sure that the benefits of the changes we 14 assessed would necessarily translate to patients in formal social prescribing schemes.

15
Nevertheless, our results make an important contribution, given the policy interest in these 16 approaches and the limited evidence base.

17
Our results highlight that the effects of starting and stopping asset use are not symmetrical,

18
which suggests that equal attention needs to be given to these different processes. The

19
focus of social prescribing tends to be on the former, but our data suggests that it is 20 important to identify people whose use of assets stops. If such people can be identified and 21 supported, the gains might be even greater, but it is not clear that the same schemes would 22 be suited for increasing use and maintaining use.

23
Unanswered questions and future research 24 As noted previously, the study was conducted in a single region of the UK, and the results 25 would need replication. Given that the benefits of asset use seemed to increase with time,

26
further long-term evaluation would also be indicated. Exploration of the reasons why people 3 Another potential limitation is that we do not observe the timing of events. For example, in 4 the cessation analysis we know that individuals ceased participation in community assets 5 and they experiences a decline in QALYs. We assume that the former caused the latter, but 6 it may be possible that declining HRQoL led to a cessation in asset participation. The

11
In our analysis, we are unsure if individuals chose to start (or stop) using community assets 12 because they were referred to them by a link worker (a social prescriber), or if they chose to 13 do so for other reasons (including friend referrals, more exposure, etc.). Therefore, whilst we 14 demonstrate that community assets have considerable benefits, we cannot be completely 15 confident that this is all attributable to social prescribing.

16
Further, we cannot confidently demonstrate which type of community assets are most 17 beneficial, as our definition of utilisation is based on self-reports.

18
Our results provide a robust assessment of the impacts of changes in the use of community 19 assets, and provide further impetus to calls for robust evaluation of their effects. There is a 20 legitimate debate as to whether the standard controlled trial is optimal for the assessment of  We used quasi-experimental methods to explore the impact of changing patterns of the use 2 of community assets in a population of older people living in an area that introduced an 3 integrated care initiative which sought to increase asset use.

4
We found that increasing use of community assets was associated with increased health 5 related quality of life, reduced costs, and positive societal net-benefit. The reduction in costs 6 and positive net-benefits were sustained over time and indicated substantial benefits from 7 prolonged community asset use. 8 The effects of starting to use assets were not symmetrical with ceasing use, with the latter 9 associated with larger losses. This is important, as encouraging use among those who do 10 not currently use assets may require different policy and patient-level interventions to those 11 designed to encourage continued use.

12
The results support the inclusion of community assets as part of an integrated care model for 13 older patients.

Consent for publication
Not applicable.

Data sharing statement
The data that support the findings of this study are available from the Principal Investigator of the original study but restrictions apply to the availability of these data, which were used under license for the current study, and so are not publicly available. Data are however available from the authors upon reasonable request and with permission.

Competing interests
None of the authors have any competing interests to declare.            Figure A1: Density plots of propensity scores before and after matching   Figures 1 and 2 (pages 6 and 11)  *Give information separately for exposed and unexposed groups. Improving outcomes for older people with long-term conditions and multimorbidity is a priority.

4
Current policy commits to substantial expansion of social prescribing to community assets, 5 such as charity, voluntary or community groups. We use longitudinal data to add to the limited 6 evidence on whether this is associated with better quality of life or lower costs of care.   20

21
The way in which health and social care organisations engage with community assets has 22 subsequently become more direct. In several areas, health and care professionals (as well as 23 other front-line professionals) have begun to make referrals to such community assets as part 14 This rapid expansion of formal provision will occur without a strong evidence base. Although 15 reviews and qualitative work have suggested that community assets improve the health of 16 participants 7,8 , there is limited quantitative evidence. 9 Outcomes that have been identified in 17 qualitative studies have included a sense of involvement and better well-being 8 , whereas 18 outcomes that have been identified in quantitative studies have included health related quality

19
of life and health care costs 9 . The evidence base for social prescribing is equally limited and 20 has yet to arrive at a consensus. 10 However, it is worth noting that the evidence is still 21 developing in this field, with ongoing qualitative and quantitative studies.

22
We previously evaluated an integrated care programme for older people which included a 23 programme to improve use of community assets. 9 We used data from a cohort of older people  costs. However, the cross-sectional nature of the data meant that we could not interpret the 2 relationships as causal. 3 In this study, we analyse the relationships between community asset participation, health and 4 health care utilisation longitudinally, to provide a more rigorous assessment of the causal 5 impact of community asset participation. Using administrative health records further 6 strengthens the analysis presented here as it removes the reliance on recall. As well as 7 considering the uptake of community assets as a possible health enhancing activity, we 8 additionally examine the possibility of there being health decrements associated with ceasing 9 to participate in community assets. A priori, it is not expected that the absolute size of the 10 gains from starting will equal the size of the reductions from stopping.

12
Data: cohort description 13 The data used in this analysis were made available as part of the National Institute of Health

14
Research funded Comprehensive Longitudinal Assessment of Salford Integrated Care

15
(CLASSIC) study. 11 CLASSIC is an evaluation framework designed to evaluate the Salford    Health-related quality of life (HRQoL) was measured using the Euro-QoL 5D-5L. 12,13 The EQ-

13
Information from primary care records contained a count of the number of times an individual 14 visited their GP. We then applied the PSSRU Unit Cost (in 2014/15 values) of £65 per visit. 20

15
We applied a discount rate of 3.5% to the costs and benefits. 21 16 Net-benefit 17 As in our earlier work 9 , we defined net-benefits as an individual's QALY gain minus the cost 18 of their healthcare utilisation. 22 We used the two thresholds used by the National Institute for

19
Health and Care Excellence; namely £20,000 and £30,000 but focus mainly on the £20,000 20 threshold for reasons of brevity.

21
Community asset participation

22
Community asset participation was defined as a binary variable equal to one if an individual 23 reported participating in any one of a list of activities, and zero otherwise. The list of community 24 assets is included in a supplementary appendix, along with reported participation rates over 25 time (Table A1). Demographic and socioeconomic characteristics 2 We controlled for gender and age using a series of 5-year age categories (ranging from 65-3 69 years, up to 85+years). The reference age group is 65-69 years. We also controlled for 4 living situation, coded as 'live with spouse', 'live with other' or the reference category 'live 5 alone'. We included binary variables for each of the following qualifications: 'one or more  Table A2 (supplementary appendix).

12
We used double-robust estimation 23 to estimate the impact of community asset participation then used to weight the regression model for the outcome. 23 As long as one model is correctly 20 specified, the double-robust estimator produces unbiased results. 24,25 If both models are 21 correctly specified, then double-robust estimator is both unbiased and efficient. 26 22 The choice of control variables for both models is important. We provide a full list of all 23 variables included in both the treatment (propensity score) equation and the outcome

Uptake analysis
14 For the 6-month analysis, we defined the comparator group as those individuals who did not 15 participate in community assets at baseline and continued to not participate at the 6-month 16 follow-up. The treatment group consists of those individuals who did not participate in assets 17 at baseline but did report participation at 6-months. This is comparison A (Table 1).

19
For the 12-month and 18-month analyses the definition of the treatment group was more  In the 12-month and 18-month analyses, the comparator group is those individuals who never 2 participated (NNN or NNNN). The primary definition of treatment in the 12-month analysis was 3 NYY (comparison C) and in the 18-month analysis was NYYY (comparison E).
4 Cessation analysis 5 We followed a similar logic for estimating the effects of ceasing to participate in community 6 assets. For the 6-month analysis we defined the comparator group as those who always 7 participate and the treatment group as those individuals who initially participated at baseline analyses followed a similar pattern, and are shown as comparisons H and J in Table 1.

Secondary analyses
11 In a secondary analysis we relaxed the restriction that an individual had to remain in the 12 sample for all four waves. We included data from all individuals in the respective waves.

13
In another secondary analysis, we additionally considered the effects of participating in periods. For the uptake analysis, these were comparisons B and D in Table 1. For the 16 cessation analysis, these were comparisons G and I.

18
Selected characteristics of the respondents at baseline are available in Table 2. Further detail 19 is provided in Table A2.
20 Participation in community assets over time 21 Figure 2 shows how many people participated in community assets at each wave.  The only significant predictors of drop-out from the cohort were older age and education.

7
However, the magnitude of their effects on drop-out were not significantly different between 8 those who initially participated and those who initially did not participate in community assets.

9
The full regression results are presented in a supplementary appendix (Table A3). shows the distributions for the cessation analysis. In both cases, the matching considerably 14 improves the similarity between the control and treatment groups.

15
Multivariate analysis: Uptake analysis 16 There is a positive and statistically significant effect of starting community asset participation 17 on health-related quality of life (Table 3, panel (a)). The benefit of starting to participate in 18 community assets is a 0·017 QALY gain (95% CI: 0·002 to 0·032) compared to those who 19 never participate in assets at the 6-month follow-up. The effect of starting to participate in 20 community assets is a QALY gain of 0·030 (95% CI: 0·005 to 0·054) at the 12-month follow-

22
Starting to participate in community assets reduced costs in the 6-month period by £96 (95%

13
When we considered the total costs of health-care utilisation, we found that stopping 14 participating in community assets led to large and statistically significant increases in health 15 care utilisation costs. In the 6-month period this increase was £689 (95% CI: £162 to £1216) 16 whereas in the 12-month and 18-momnth follow-ups these increases were £857 (95% CI:

18
Additionally, there were negative net-benefits (assuming a £20,000 NICE threshold) 19 associated with cessation. In the 6-month period this potential loss was £624 per-participant  The results using all available data on respondents are qualitatively similar in terms of 2 magnitude and statistical significance (Table A4).

Secondary Analyses
3 Use of less strict definitions of uptake and cessation also produces similar results, but the 4 effects are typically smaller in magnitude (Table A5).

6
Our study involved a large sample of patients recruited and followed up over an 18-month 7 period. Although there was loss to follow-up, the overall rate of retention was reasonable. We 8 collected detailed data on asset use and health, with objective data on health care costs 9 available from administrative records. We adopted rigorous methods for the estimation of 10 causal effects and found the main results were robust to several assumptions.

11
We additionally performed many robustness/sensitivity analyses where we changed the

22
The SICP programme also ensured that there was more integration of care within Salford,

23
particularly during the study period. Therefore, the results need to be interpreted in this 24 context, where there has been significant investment in community assets locally. As we highlighted in previous work, objective data on the impact of increasing use of 2 community assets is limited 9 , and this paper therefore makes a significant contribution to this 3 area. Our broad results are consistent with the published work in this field, while adding value 4 due to the methodological strengths of the work.

5
Haslam et al 27 undertook a longitudinal study of the relationship between engagement with 6 social groups and cognitive function using data from the English Longitudinal Study of Ageing 7 (ELSA). They found that current use of social groups significantly predicted better cognition. 8 Their study differs from ours in that we are interested in health and health care utilisation and 9 we model the decision to partake in social groups and community assets.  This evidence base is continually evolving, and we expect this to change given a number of 2 ongoing and planned evaluations.

3
Our analytical methods provided a comparator group to better assess the impact of changes 4 in asset use. We examined non-experimental changes in asset use in the context of a wider 5 integrated care initiative, which saw some patients starting to use assets, and others ceasing 6 use. It is plausible that at least some of this increased use reflected the wider integrated care 7 initiative that was being undertaken in the area, but this cannot be determined reliably. Our 8 analysis used a large sample and robust analytic methods, and was able to assess the effects 9 of starting and stopping asset use. However, we were not testing the impact of new referrals 10 to community assets, and we cannot be sure that the benefits of the changes we assessed 11 would necessarily translate to patients in formal social prescribing schemes. Nevertheless,

12
our results make an important contribution, given the policy interest in these approaches and 13 the limited evidence base.
14 Our results highlight that the effects of starting and stopping asset use are not symmetrical,

15
which suggests that equal attention needs to be given to these different processes. The focus 16 of social prescribing tends to be on the former, but our data suggests that it is important to 17 identify people whose use of assets stops. If such people can be identified and supported, the 18 gains might be even greater, but it is not clear that the same schemes would be suited for

19
increasing use and maintaining use.  4 Another potential limitation is that we do not observe the timing of events. For example, in the 5 cessation analysis we know that individuals ceased participation in community assets and 6 they experiences a decline in QALYs. We assume that the former caused the latter, but it may 7 be possible that declining HRQoL led to a cessation in asset participation. The statistical 8 matching on baseline characteristics should somewhat mitigate against this if we assume that 9 initial levels of HRQoL and health indicate similar rates of decline, conditional on age and other 10 factors. However, without detailed dates of when community asset participation stopped, we 11 cannot be certain of the sequence of events.

12
In our analysis, we are unsure if individuals chose to start (or stop) using community assets 13 because they were referred to them by a link worker (a social prescriber), or if they chose to 14 do so for other reasons (including friend referrals, more exposure, etc.). Therefore, whilst we 15 demonstrate that community assets have considerable benefits, we cannot be completely 16 confident that this is all attributable to social prescribing.

17
Further, we cannot confidently demonstrate which type of community assets are most 18 beneficial, as our definition of utilisation is based on self-reports.

19
Our results provide a robust assessment of the impacts of changes in the use of community 20 assets, and provide further impetus to calls for robust evaluation of their effects. There is a 21 legitimate debate as to whether the standard controlled trial is optimal for the assessment of  We used quasi-experimental methods to explore the impact of changing patterns of the use 2 of community assets in a population of older people living in an area that introduced an 3 integrated care initiative which sought to increase asset use.

4
We found that increasing use of community assets was associated with increased health 5 related quality of life, reduced costs, and positive societal net-benefit. The reduction in costs 6 and positive net-benefits were sustained over time and indicated substantial benefits from 7 prolonged community asset use. 8 The effects of starting to use assets were not symmetrical to those from ceasing use, with the 9 latter associated with larger losses. This is important, as encouraging use among those who 10 do not currently use assets may require different policy and patient-level interventions to those 11 designed to encourage continued use.

Consent for publication
Not applicable.

Data sharing statement
The data that support the findings of this study are available from the Principal Investigator of the original study but restrictions apply to the availability of these data, which were used under license for the current study, and so are not publicly available. Data are however available from the authors upon reasonable request and with permission.

Competing interests
None of the authors have any competing interests to declare.