Results of the CLASSIC cohort

Figure 5 shows the flow of patients into the cohort.

FIGURE 5 Flow of patients in the cohort. (Figure not reproduced here; see the original article.)

Owing to current word limits on the report, we do not present detailed descriptive data on the cohort and restrict the main presentation to quasi-experimental analyses of SICP mechanisms of integration (community assets and care plans).

We present basic descriptive data on patient experience items in Appendix 3. Analyses using the cohort data to explore other aspects of care for older people can be found in published papers,114,115 and more will be reported in due course.

A limitation of existing analyses of integrated care is that they are too large in scope or rapid in delivery to allow setting up data collection to capture effects, restricting analyses to routine data that lack patient-reported outcomes. The CLASSIC trial used the cmRCT design to develop a cohort, which provided the ‘context’ into which the SICP and its mechanisms of integration would be introduced. The cohort had two functions:

1. to provide a sampling frame for the cmRCT within CLASSIC for formal experimental analyses (full details are provided in Chapters 8 and 13)
2. to provide a sample of the total eligible population of older people, which could be used to track the impact of mechanisms of integration on patients through variation in exposure to those mechanisms among patients in the cohort.

This chapter will focus on the second function. Two mechanisms of integration suited to evaluation through the cohort are community assets and care plans. Community assets were a specific mechanism of integration within the SICP, with its own dedicated workstream (see Chapter 2, Box 1). Care plans have long been seen as critical to effective management of long-term conditions,15,116 and a major feature of health policy in the UK.117 We assessed use of both community assets and care plans in the cohort and used variation in use to explore their impact on patient outcomes.

Community assets

In 2010, the UK government stressed the need for a ‘Big Society’, whereby individuals engage more with the facilities in their local community. A critical component was community assets, defined as:

. . . the collective resources which individuals and communities have at their disposal, which protect against negative health outcomes and promote health status.

Community assets (such as charity, voluntary or community groups) may have a role in improving health and reducing inequalities. Reviews119–121 and qualitative work suggest that community assets improve health,122 but quantitative evidence is sparse.

The community assets workstream of the SICP was not a primary focus of the CLASSIC research programme. The community assets workstream involved a number of different parts, including increasing voluntary work, delivery of well-being plans, digital inclusion, falls prevention and malnutrition awareness. A community assets working group brought together multiple partners locally as well as older people to develop better co-ordination of current assets. We used the cohort to answer the following research questions:

1. How do people in the cohort use community assets and how does that change over time?
2. Is community asset use associated with outcomes (quality of life and health-care utilisation)?

Care plans

Care plans are seen as critical to quality care for long-term conditions,15,117 but, despite their importance in policy, implementation is inconsistent.74,123 In the SICP, care plans were important in both the MDG (shared care plans) and community asset (well-being plans) workstreams.

We used the CLASSIC cohort to answer the following research questions:

1. What is the prevalence of self-reported care plans in the cohort and how does that change over time?
2. Is the self-report of a care plan associated with outcomes (feelings that care is integrated, patient activation and health-care utilisation)?

Methods

Details of the CLASSIC cohort are provided in Chapter 4.

Community asset use

Individuals were asked ‘Have you attended or used any of the following community groups, activities and services in the last 6 months?’ and provided with the following options: (1) trade unions, (2) group for the elderly or older people (e.g. lunch club), (3) environmental groups, (4) youth groups (e.g. Scouts, Guides), (5) parent–teacher association or school association, (6) Women’s Institute, Townswomen’s Guild or women’s groups, (7) residents’ group or neighbourhood watch, (8) social club (including working men’s clubs, Rotary Clubs), (9) education, arts, music or singing groups, (10) sports club, gym, exercise or dance groups, (11) religious group or church organisation, (12) other group or organisation and (13) charity, voluntary or community group. This represented only a part of the SICP community assets programme.

An individual was classified as using community assets if they ticked ‘yes’ to one or more. The list of assets was less comprehensive in waves 2 and 3, as it included only (2) and (8)–(12). However, results were robust to these changes.

We obtained the costs of outpatient attendances, ambulance use and A&E visits from NHS reference costs,111 and cost of a GP visits from a PSSRU costs publication.110 In 2014/15 prices were as follows: GP appointment £65.00, outpatient attendance £134.22, ambulance callout £96.35 and A&E visit £131.92.

Analytic methods

How do people in the CLASSIC cohort use community assets and how does that change over time?

We present descriptive data from the cohort on community asset use over time.

1. Is community asset use associated with quality of life and health-care utilisation?

We used variation in use of community assets to explore its causal impact. In the absence of randomisation, determining causal associations is complex, as cross-sectional associations may reflect reverse pathways (health influencing asset use rather than vice versa) or unmeasured confounders. We conducted the following analyses:

1. examined cross-sectional associations between asset use and outcomes (quality of life, health-care utilisation), controlling for measured confounders
2. used an instrumental variable approach (using distance to community assets as an instrument) to test whether or not the association was maintained, to provide a more rigorous test of a causal relationship
3. used longitudinal data to assess whether or not change in asset use was associated with outcomes.

Associations between asset use and outcomes

We compared quality of life, health-care utilisation and costs between users and non-users of community assets. We estimated three multivariate models for quality of life and utilisation, including community asset use and:

1. sex, age and socioeconomic characteristics only
2. sex, age, socioeconomic characteristics and the presence of 23 long-term conditions
3. sex, age, socioeconomic characteristics and the presence of 23 long-term conditions limiting activity.

We combined the effects on health-related quality of life and costs to produce estimates of the societal value of 1 year’s participation in community assets using the net benefit framework.124 This net benefit combines both the benefit and the cost into one single metric. For example, it would be better equipped to deal with a scenario in which an individual’s health improved at the same time as they used more NHS resources (i.e. having higher costs). This involves multiplying EQ-5D scores by a value for a QALY and subtracting the annual cost of health-care utilisation. We use a range of threshold values (£20,000, £30,000 and £12,936)125 and multiplied 6-month costs to obtain annual costs. Net benefit was then used in three further regression models, which included sex, age, socioeconomic characteristics and 23 long-term conditions limiting activities.

Instrumental variable analyses

We received council data containing postcodes of community assets (May 2014–15). These assets were defined by the council as:

. . . buildings or amenities that play a vital role in local life . . . community centres, libraries, swimming pools, village shops, markets or pubs.

We included a range of asset types (‘health and well-being’, ‘skills and training’, ‘food’, ‘outdoor space’, ‘physical exercise’, ‘religious’, ‘shopping’, ‘social’, ‘volunteering’ and ‘other’), but excluded those coded as ‘older teenage’.

Data from patient postcodes enabled the calculation of distance between households and assets, by translating postcodes to co-ordinates and calculating distances using the Stata^® (StataCorp, College Station, TX, USA) ‘vincenty’ command via the law of cosines. We calculated (1) the minimum distance to the nearest asset (in miles) and (2) the number of assets within a given radius (1, 2, 5, 10, 15 and 20 miles).

As noted earlier, regression of community asset use on outcomes can produce bias via reverse causality. To overcome this bias, we used an instrumental variables approach, defined as ‘an analytical technique, traditionally used in non-randomised research studies, which uses a variable associated with the factor under study but not directly associated with the outcome variable or any potential confounders’.127 Instruments are variables that are related to asset use, but not directly to outcomes. We used the distance to nearest asset and the number of assets in a certain radius as instruments. We argue that the way in which assets impact on health is through use only. That is, the distance to an individual’s nearest asset can affect their health only if that individual then uses that asset. Just because an individual lives close to an asset is not sufficient for their health to improve, they must make use of that asset.

Instrumental variable models have two stages:

1. the prediction of asset use conditional on minimum distance and number of assets in the radii (and other cofounders)
2. effect of asset use on health (the first stage prediction of asset use is used in the second stage).

As asset use is binary, we used linear regression with endogenous treatment effects:128,129

1. we estimated three separate multivariate models for the EQ-5D score, which included asset use and sex, age and socioeconomic characteristics only
2. sex, age, socioeconomic characteristics and the presence of 23 long-term conditions
3. sex, age, socioeconomic characteristics and the presence of 23 long-term conditions limiting activity.

We then estimated the same models for utilisation (GP, outpatient and total costs). Finally, we combined the effects on quality of life and costs to produce estimates of societal value.

Longitudinal analyses

We estimated the effect of asset use using a matched control group analysis,130 whereby the effect is defined as the mean difference between the observed and the imputed potential outcome for each individual. We use nearest ‘neighbour matching’ to impute potential outcomes, which takes an average of the outcome of similar subjects that are not treated. We further matched on baseline EQ-5D to better capture baseline health. As well as considering the effects of starting asset use, we estimate effects of stopping. We estimated short- (baseline to 6 months) and long-term (baseline to 12 months) effects (Table 12) of starting and stopping use of assets.

In all models, we matched on baseline age, sex, living arrangements and the presence of limiting conditions. We additionally matched on baseline EQ-5D scores in a later specification. For utilisation and net benefits outcomes, we further matched on baseline levels of these outcomes.

Care plans

The analysis of the effects of care plans used analogous methods to the assessment of community assets, using change in use of care plans over time to explore potential impacts on patient outcomes.

The measure of care plans was the question ‘Do you have a written care plan?’ at baseline and 18 months.

Outcomes were:

1. patient perception of integration; we used a single item from the unpublished QIPP measure, ‘Do you think the support and care you receive is joined up and working for you’, scored on a four-point Likert scale
2. patient activation as measured by the PAM (see Chapter 4)
3. EQ-5D.

Results

Participation in community assets

Fifty per cent (1829/3686) of respondents at baseline reported using community assets. Most reported using one, and those using at least one community asset reported use of an average of two assets.

Table 13 shows participation over time among those reporting data at three waves of the cohort.

Associations between community asset use and outcomes

On average, community asset users had an EQ-5D score of 0.690 compared with 0.596 for non-users (difference 0.094; p < 0.001) (Table 14). Table 14 shows that there were few differences between the two groups in other characteristics, but 9% of community asset users had university qualifications, compared with 4% of non-users.

On average, community asset users had visited a GP three times in the previous 6 months, 0.33 visits fewer than non-users (p < 0.001). Community asset users reported an average of 2.2 hospital outpatient appointments in the last 6 months, 0.3 visits fewer than non-users. Users also reported fewer ambulance call-outs and casualty visits. The average health-care costs over a 6-month period in users was £97 (p = 0.003) lower than non-users (£448 vs. £545).

Instrumental variable models

Table 15 shows the average distance, with users of assets closer to assets [difference –0.026, 95% confidence interval (CI) –0.039 to –0.013]. There is little difference in the number of assets within given radii for users and non-users, suggesting that asset availability is not responsible for the higher quality of life in users.

We ran regressions with different combinations of distance and radii variables. All met validity tests:

1. instruments predict asset use (first stage chi-squared statistic of 316.47; p < 0.0001)
2. they are not overidentified, as shown by a Sargan score of 1.425 (p = 0.84).

The best-fitting model using Bayesian information criterion131 was one with five instruments: (1) distance to nearest asset, (2) this distance squared, (3) number of assets within a 1-mile radius, (4) number of assets within a 10-mile radius and (5) number of assets within a 15-mile radius. Table 16 presents results from the regression. Column 1 shows the first-stage results. Longer distances to the nearest asset reduce the probability of use (as expected). If the distance to the nearest asset increases by 1 mile, then individuals are 66% less likely to use assets. The number of assets in a local area is statistically significant only if that area is defined as within a 15-mile radius. In second stage results (the effects of asset use on quality of life), we observe a significant positive effect of use. Controlling for health conditions and their effects on activity reduces the effect, but it remains statistically significant.

In analyses of the effects of asset use on utilisation and costs, asset use does reduce health-care utilisation and total costs, but the reductions are not statistically significant (Table 17).

Using the current NICE threshold values of £20,000–30,000 per QALY, we estimated that the net benefits of community asset use were £3624.20 (95% CI £1473.52 to £5054.83) to £4665.90 (95% CI £2061.96 to £7269.78) per participant per year (see Table 17). Using the threshold value proposed by Claxton et al.125 gave a net benefit estimate of £2288.30 per participant per year (95% CI £1067.94 to £3508.70).

Longitudinal analyses: short-term effects (0–6 months)

We present a range of models estimating the effects of starting to use assets (Table 18), with the primary analysis being ‘nearest neighbour with five matches’ (see Table 18). Starting using assets increases EQ-5D scores (0.058, 95% CI 0.031to 0.085), but has no significant effect on GP or hospital outpatient visits. Analysis of net benefit showed large and statistically significant gains at all three thresholds.

Table 19 shows effects for stopping asset use, whereby the reduced benefits associated with stopping were smaller than the benefits of starting and the change in net benefit was not statistically significant.

Longitudinal analyses: long-term effects (0–12 months)

Table 20 shows long-term effects of asset use were similar to the short term. Effects on health-care utilisation remained non-significant, but net benefit increases were significant.

The effects of stopping asset use were not statistically significant for EQ-5D and health-care utilisation, and were significant only in terms of net benefit, whereby the reductions were considerably larger in the longer term than the short term (Table 21).

Use of care plans

The numbers of people reporting a care plan increased over time (baseline 6.4%, 12 months 8.9%, 18 months 10.7%). A similar pattern was observed when restricted to people with data at all three time points (baseline 5.4%, 12 months 8.5%, 18 months 10.4%).

We used a quasi-experimental approach and estimated the effect associated with gaining a care plan. As before, we used the nearest neighbour matching method to match older people who gained a care plan to similar individuals who did not get a plan between baseline and 18 months.

In all models the outcome is either perceptions of integration, patient activation or quality of life at 18 months. We tested the following models:

• model 1 – matched on age, sex and living arrangements at baseline, 1 : 1 matching
• model 2 – model 1 plus presence of limiting health conditions at baseline, 1 : 1 matching
• model 3 – model 2 plus baseline value of outcomes, 1 : 1 matching
• model 4 – model 3, 5 : 1 matching
• model 5 – model 3, propensity score matching.

Results

Gaining a care plan was associated with significant increases in the proportion of patients reporting that their care was ‘joined up and working’ (Table 22). There were no associations between gaining a care plan and either patient activation (Table 23) or quality of life (Table 24).

Summary

We used cohort data to assess the causal relationship between care plans, use of community assets and patient outcomes.

Only a minority of patients reported a care plan at baseline. The numbers doubled over the period of the cohort, although the rates remained small (around 10%). Reporting a care plan was associated with increased feelings that care was integrated, but did not affect patient activation or quality of life. These broad outcomes are similar to a smaller cross-sectional study conducted as part of the evaluation of the North West London Integrated Care Pilot.132

Self-reported use of care plans may not always accord with service measures. For many older patients, care plans are a relatively new clinical tool, and our previous work has highlighted the potential difficulty of measuring the presence or absence of care plans.133 Although the rate of care plans reported here accords with other studies,74 it is possible for care plans to be created for patients who may either be unaware of the care plan or unfamiliar with the terminology used. Nevertheless, awareness of the plan and full involvement in the production of the plan is supposedly an important part of the process.

Asset use was associated with quality of life in cross-sectional analyses, and the relationship remained significant in analyses using instrumental variables and longitudinal approaches. Although confidence in the internal validity of that causal relationship cannot approach that of a formal randomised trial, the analyses make a significant contribution to the literature concerning the health benefits of asset use.

It is noteworthy that the benefits (and net benefits) of asset use are largely restricted to improvements in quality of life, as impacts on utilisation were small and largely non-significant.

There are methodological limitations to consider. The cohort was self-selected and attrition further reduced the sample size and representativeness. All measures were self-reported, which is appropriate for feelings of integration, activation and quality of life, but potentially less accurate for health-care utilisation. There is evidence that self-reports are reasonably accurate compared with routine data on utilisation,134 and since the analysis is largely focused on associations rather than absolute levels, the overall results may be reasonably robust to any self-report bias.

We highlight two issues in relation to the measurement of assets.

First, we used a simple binary measure of asset use, without any details as regards the amount or pattern of use. Although a more detailed assessment of asset use would have been preferable, our survey was limited in space to reduce respondent burden, and a more detailed measure was not possible. Further research using more detailed measures of asset use may report different findings, but we would expect the relationships demonstrated with a less precise measure to be conservative.

Second, we used a generic measure of community asset use. There is a lack of consensus over the scope of the term. Also, understanding among patients of the meaning of terms and how they relate to assets they use are likely to be complex. We used a scale that has been used before, but which was not directly linked to the interventions undertaken by the SICP community assets programme, which included a range of other activities (including well-being champions, volunteers in care homes, well-being plans, digital inclusion, falls prevention and malnutrition awareness). This is because the exact nature of those programmes was not clear when the cohort started, and it was felt that using specific terms for the assets used by the SICP would be potentially unreliable as they might not be known to patients. Therefore, linking changes in overall asset use (the 6% increase in those reporting use at waves 1 and 3) with the specific SICP investments is open to challenge. We can be more confident that changes in asset use over time are driving changes in health outcomes, but can only infer that the benefits seen here are only caused by the SICP.