Article Text

other Versions

Download PDFPDF

Effect of exercise referral schemes upon health and well-being: initial observational insights using individual patient data meta-analysis from the National Referral Database
Free
  1. Matthew Wade1,2,
  2. Steven Mann3,
  3. Rob J Copeland4,
  4. James Steele1,5
  1. 1 Research Institute, ukactive, London, UK
  2. 2 St Mary's University Twickenham, Twickenham, UK
  3. 3 Places for People Leisure, Camberley, UK
  4. 4 The National Centre for Sport and Exercise Medicine, Sheffield Hallam University, Sheffield, UK
  5. 5 School of Sport, Health, and Social Sciences, Solent University, Southampton, UK
  1. Correspondence to Dr James Steele, Research Institute, Ukactive, London WC1R 4HE, UK; jamessteele{at}ukactive.org.uk

Abstract

Objectives To examine if exercise referral schemes (ERSs) are associated with meaningful changes in health and well-being in a large cohort of individuals throughout England, Scotland, and Wales from the National Referral Database.

Methods Data were obtained from 23 731 participants from 13 different ERSs lasting 6 weeks to 3 months. Changes from pre- to post-ERS in health and well-being outcomes were examined including body mass index (BMI), blood pressure (systolic (SBP) and diastolic (DBP)), resting heart rate (RHR), short Warwick Edinburgh Mental Wellbeing Scale (SWEMWBS), WHO Well-Being Index (WHO-5), Exercise Related Quality of Life scale (ERQoL), and Exercise Self-Efficacy Scale (ESES). Two-stage individual patient data meta-analysis was used to generate effect estimates.

Results Estimates (95% CIs) revealed statistically significant changes occurred compared with point nulls for BMI (−0.55 kg.m2 (−0.69 to −0.41)), SBP (−2.95 mmHg (−3.97 to −1.92)), SWEMWBS (2.99 pts (1.61 to 4.36)), WHO-5 (8.78 pts (6.84 to 10.63)), ERQoL (15.26 pts (4.71 to 25.82)), and ESES (2.58 pts (1.76 to 3.40)), but not RHR (0.22 f c (−1.57 to 1.12)) or DBP (−0.93 mmHg (−1.51 to −0.35)). However, comparisons of estimates (95% CIs) against null intervals suggested the majority of outcomes may not improve meaningfully.

Conclusions We considered whether meaningful health and well-being changes occur in people who are undergoing ERSs. These results demonstrate that, although many health and well-being outcomes improved, the changes did not achieve meaningful levels. This suggests the need to consider the implementation of ERSs more critically to discern how to maximise their effectiveness.

  • exercise
  • physical activity
  • health behaviour

Statistics from Altmetric.com

Introduction

Engagement in physical activity is widely considered to be effective in the prevention, management and treatment of many chronic health disorders.1 2 Despite this, in the UK a recent survey has shown that population levels of physical activity have remained unchanged in recent years, with a large proportion of the population still classed as inactive, presenting potentially serious repercussions for population health.3 The costs of physical inactivity to the National Health Service (NHS) were estimated as £900 million in 20154 which, despite relatively stable levels of physical activity, had risen to £1.2 billion in 2017.5 Though it is hoped that population wide increases in physical activity are possible they are difficult to achieve and thought to require complex interventions aimed at several socio-ecological levels.6–9 Therefore it has been argued that the promotion of physical activity must be a key component of a healthcare system within a ‘whole systems approach’.10

For many, achievement of sufficient levels of physical activity evidently does not arise from their typical activities of daily living in our present socio-ecological environment. Instead, as opportunities for spontaneous physical activity during daily life have likely reduced in our modern environment despite our bodies still ‘expecting’ this stimulus, directed exercise 11 may be a necessity for health and fitness in this current era.1It is worth differentiating physical activity from exercise here. Physical activity has been adopted as a term for those activities which are unstructured and occur as a result of tasks of daily living, occupation, or leisure without the explicit goal of improved health and/or fitness, whereas exercise, though falling under the broad category of physical activity, involves tasks often specifically engaged in for the pursuit of improved health and/or fitness.11 Both may present potential disturbances to homeostasis as a result of muscular action, thus providing stimulus for adaptation including improved health and fitness.11 ,1 Yet, compared with replacing sedentary activity with just light physical activity, structured exercise may have differential effect health effects.12 Further, for many conditions, network meta-analyses have shown exercise interventions are similarly—and in some cases more—effective than drug treatments for secondary prevention.13 14 In this respect, it has been argued that exercise should perhaps be prescribed to patients much like a drug,15 and as noted may be similarly effective.12 For some, recommendation from a physician/general practitioner (GP) may be a key facilitator in participating in exercise in the first place.16

Considering a ‘whole systems approach’, where physical activity is argued to be a key component,10 management and treatment of chronic diseases with exercise have been built into public health pathways and healthcare models through the use of exercise referral schemes (ERSs). In ERSs, a primary care professional, most commonly a GP, will refer a patient with, or at risk of, a chronic health disorder into an exercise based intervention.17 18 ERSs were first introduced in the 1990s in primary care settings to facilitate exercise participation in those suffering from chronic disease.19 20 Delivery of ERSs usually occurs through leisure/sport centre pathways and often includes both cardiorespiratory and resistance training modalities, though despite typically poor reporting of the specific exercise intervention employed there is likely considerable heterogeneity in programme implementation.17 18 ERSs typically last from 10 to 12 weeks in England and Ireland,18 or 16 weeks in Wales, the latter being more cost effective.21

Considering the current issues with the extant literature regarding ERSs, including inconsistent and weak evidence regarding their effects on health, well-being and quality of life outcomes,17 18 22 23 there is a need for continued evaluation to help inform guidelines regarding them.17 24 A recent updated systematic review has highlighted that ERSs in the UK may be effective and that longer schemes may be more effective than shorter schemes.25 However, most schemes focus on physical activity changes solely, likely considering them appropriate surrogates for improved health and well-being. As such, it has been argued that ERSs, and evaluation of their benefits, should extend beyond merely increasing physical activity levels and consider other health and well-being outcomes.26 Thus the aim of this paper is to describe the initial insights obtained from individual patient data meta-analysis of the National Referral Database27 with a primary focus on the effects (ie, change from pre- to post-ERS), including point estimates and precision of those estimates, of ERSs on health and well-being outcomes.

Methods

​Study design

An accompanying pre-print manuscript27 describes the database formation, data cleaning, and structure in detail in addition to the key issues and limitations of the database. For the sake of space this is not replicated here but the reader is encouraged to refer to this pre-print27 to interpret the findings presented in their appropriate context. In essence, this database represents a retrospective cohort longitudinal study design following individuals entering and exiting ERSs following referral from a range of organisations and referrer types (primary, secondary, and tertiary) across the UK. The database includes ERSs broadly speaking and unfortunately at present data are unavailable regarding their specific delivery, though work is underway to retrospectively determine details of this conforming to current reporting standards (described in Steele et al 27). Due to the inclusion of different schemes within the database with varying and unspecified characteristics we used individual patient data meta-analysis with a two stage approach to account for this in analysis in providing point estimates and precision.

​Outcome measures

Outcome measures varied across the ERSs but were taken at both pre- and post-intervention across all ERSs. The majority of schemes included height and weight and from this body mass index (BMI) could be calculated. Blood pressure,28 systolic (SBP) and diastolic (DBP), and resting heart rate (RHR) were also included as outcomes in the majority of schemes. Mental well-being was included and measured in a small number of ERSs using the short Warwick Edinburgh Mental Wellbeing Scale (SWEMWBS),29 and general well-being and quality of life were measured using the WHO Well-Being Index (WHO-5)30 and Exercise Related Quality of Life scale (ERQoL).31 Lastly, exercise related self-efficacy was collected in one scheme using the Exercise Self-Efficacy Scale (ESES).32

​Statistical analyses

Two stage individual patient data meta-analysis was performed on the change scores (ie, post- minus pre-ERS scores) for each outcome measure. For stage 1, change scores and their standard errors were derived for each scheme. For all measures, change scores were calculated in their raw units of measurement. The second stage involved performing a random effects meta-analysis using the ‘metafor’ package in R (version 3.5.0; R Core Development Team, https://www.r-project.org/) across all schemes in order to derive a final point estimate and precision of estimates (95% confidence intervals (95% CI)). Estimates were weighted by inverse sampling variance and restricted maximal likelihood estimation was used in all models. Schemes without sufficient participants (n <4) were excluded from analysis in order to maintain sufficient sample sizes for precision of a one sample effect estimate calculated at 1 σ unit for the margin of error (or 95% CI half width; ESCI 10 to 13, La Trobe University, Australia). Robustness of main effects were considered through sensitivity analyses by removal of individual schemes and re-analysis of the random effects model. Where significant estimates became non-significant and vice versa, in addition to where there were considerable changes in the magnitude and/or precision of those estimates, the results of sensitivity analyses are reported. Although initially intended, the effects of scheme length as a moderator were not examined due to the inclusion of only one scheme 6 weeks in length. It was considered that the slight difference between 12 weeks and 3 months (6 days) was unlikely to have any meaningful impact on the analysis and any statistically significant findings might arise from type I errors.

Where it was possible to use an informed null interval based upon minimal clincally important changes (MCICs), for each outcome measure we also calculated second generation p values33 as supplementary statistics for the point estimate and precision of estimate of the random effects meta-analysis. The null interval for changes in each outcome were as follows: BMI = −1 to +1 kg/m2 34; RHR = −5 to +5 f c 35; SBP and DBP = −2 to +2 mmHg28; WHO-5 = −10 to +10 pts30; SWEMWBS = −2.77 to +2.77 pts.36 These null intervals were plotted on to forest plots for visual interpretation. Published data regarding either MCICs, reliability, etc, were not available for ERQoL or ESES and so second generation p values were not calculated for these outcomes.

An α level of 0.05 was used to determine statistical significance; however, results were not interpreted dichotomously based purely on this, or whether the 95% CIs crossed zero. As noted, supplementary statistics in the form of second generation p values (pδ) were also calculated to determine the proportion of data supported hypotheses that fall outside the null interval determined by the MCICs where a value of 1 suggested the data support only null hypotheses (ie, no meaningful change in outcome), a value of 0 suggests the data support only alternative hypotheses (ie, a meaningful change in outcome), and a value between these suggests the data are inconclusive at this stage (ie, the possibility of both meaningful or non-meaningful changes are supported by the data). In essence, pδ described the degree of overlap between the interval estimate (95% CIs) for the changes in outcome measures with the null intervals (MCICs) noted above. That is to say the analyses performed were with the intention of reporting broadly the following: do we observe a meaningful effect in people who are undergoing ERSs?

Additional supplemental meta-regression analyses were conducted to examine the association between changes in physical activity levels and changes in health and well-being outcomes. The results of these supplementary analyses are reported in the supplementary materials.

Results

Table 1 shows the final samples included in the analysis. Table 2 shows the pre-ERS scores across the schemes for each of the outcome measures examined. Results from the supplementary meta-regressions are reported in the accompanying online supplementary materials.

Supplemental material

Table 1

Characteristics of all schemes included in analysis

Table 2

Pre-ERS scores for all outcome measures for each scheme included in analysis

​Body mass index

For BMI one scheme was excluded due to small sample size (scheme 5028) and a total of 11 schemes including 4834 participants were included for analysis. Change in BMI differed significantly from the point null of zero (−0.55 kg.m2 (−0.69 to −0.41), p<0.0001). Figure 1 shows the forest plot for BMI. Significant heterogeneity was evident among the schemes (Q(10) = 161.34, p<0.001; I 2=96.5%); however, sensitivity analysis did not reveal any influential schemes. The second generation p value for the estimate from the random effects model meta-analysis was pδ=1.00 due to the interval estimate being entirely within the null interval, suggesting that the data supported only null hypotheses.

Figure 1

Forest plot of change in body mass index (BMI) across schemes. Note: dashed red lines represent the MCIC derived null intervals for interpretation of the meaningfulness of changes.

​Resting heart rate

For RHR one scheme was excluded due to small sample size (scheme 5028) and a total of nine schemes including 4287 participants were included for analysis. Change in RHR did not differ significantly from the point null of zero (−0.22 f c (−1.57 to 1.12), p=0.7448). Figure 2 shows the forest plot for RHR. Significant heterogeneity was evident among the schemes (Q(8) = 185.46, p<0.001; I 2=97.5%); however, sensitivity analysis did not reveal any influential schemes. The second generation p value for the estimate from the random effects model meta-analysis was pδ=1.00 due to the interval estimate being entirely within the null interval, suggesting that the data supported only null hypotheses.

Figure 2

Forest plot of change in resting heart rate (RHR) across schemes. Note: dashed red lines represent the MCIC derived null intervals for interpretation of the meaningfulness of changes.

​Systolic blood pressure

For SBP two schemes were excluded due to small sample size (schemes 5028 and 5144) and a total of 11 schemes including 7389 participants were included for analysis. Change in SBP differed significantly from the point null of zero (−2.95 mmHg (−3.97 to −1.92), p<0.0001). Figure 3 shows the forest plot for SBP. Significant heterogeneity was evident among the schemes (Q(10) = 55.38, p<0.001; I 2=89.26%); however, sensitivity analysis did not reveal any influential schemes. The second generation p value for the estimate from the random effects model meta-analysis was pδ=0.039 due to the interval estimate being only partly overlapped with the null interval, suggesting that the majority of data supported the hypothesis that SBP decreases meaningfully, though the data are not wholly conclusive.

Figure 3

Forest plot of change in systolic blood pressure (SBP) across schemes. Note: dashed red lines represent the MCIC derived null intervals for interpretation of the meaningfulness of changes.

​Diastolic blood pressure

For DBP two schemes were excluded due to small sample size (schemes 5028 and 5144) and a total of 11 schemes including 7451 participants were included for analysis. Change in DBP differed significantly from the point null of zero (−0.93 mmHg (−1.51 to −0.35), p=0.0016). Figure 4 shows the forest plot for SBP. Significant heterogeneity was evident among the schemes (Q(10) = 48.25, p<0.001; I 2=85.39%); however, sensitivity analysis did not reveal any influential schemes. The second generation p value for the estimate from the random effects model meta-analysis was pδ=1.00 due to the interval estimate being entirely within the null interval, suggesting that the data supported only null hypotheses.

Figure 4

Forest plot of change in diastolic blood pressure (DBP) across schemes. Note: dashed red lines represent the MCIC derived null intervals for interpretation of the meaningfulness of changes.

​Short Warwick Edinburgh Mental Wellbeing Scale

A total of three schemes including 1625 participants were included for analysis. Change in SWEMWBS differed significantly from the point null of zero (2.99 pts (1.61 to 4.36), p<0.0001). Figure 5 shows the forest plot for SWEMWBS. Significant heterogeneity was evident among the schemes (Q(2) = 11.04, p<0.001; I 2=77.97%). Sensitivity analysis did not reveal any influential schemes that altered whether the estimate was statistically significant (ie, shifted from significant to non-significant). Removal of scheme 5002 resulted in an increase in estimate to 3.81 (2.73 to 4.89). The second generation p value for the estimate from the random effects model meta-analysis was pδ=0.422 due to the interval estimate being overlapped with the null interval suggesting that the data are inconclusive. When scheme 5002 was removed this was reduced to pδ=0.018 due to the interval estimate being only partly overlapped with the null interval, suggesting that the majority of data supported the hypothesis that SWEMWBS increases meaningfully, though the data are not wholly conclusive. Considering the weighting of scheme 5002 in the random effects model meta-analysis, it seems more likely that the original estimate and confidence intervals including all schemes is a better estimate of the true effect.

Figure 5

Forest plot of change in Short Warwick Edinburgh Mental Wellbeing Scale (SWEMWBS) across schemes. Note: dashed red line represents the MCIC derived null intervals for interpretation of the meaningfulness of changes.

​WHO Well-being Index

Only one scheme (5131) collected the WHO-5 as an outcome measure and included 449 participants. Point estimate for change in WHO-5 score was 8.78 pts (6.84 to 10.63). The second generation p value for the estimate was pδ=0.834 due to the interval estimate being only partly overlapped with the null interval, suggesting that the majority of data supported hypotheses are null hypotheses, though the data are not wholly conclusive.

​Exercise Related Quality of Life Scale

For ERQoL one scheme was excluded due to small sample size (scheme 5119) and a total of three schemes including 777 participants were included for analysis. Change in ERQoL differed significantly from the point null of zero (15.26 pts (4.71 to 25.82), p=0.0046). Figure 6 shows the forest plot for ERQoL. Significant heterogeneity was evident among the schemes (Q(2) = 254.21, p<0.001; I 2=99.1%). Sensitivity analysis did not reveal any influential schemes with respect to altering whether the estimate statistically significantly differed (ie, shifted from significant to non-significant). However, removal of scheme 5156 resulted in a reduction in the point estimate but an increase in the precision (9.83 pts (8.70 to 10.95), p<0.001).

Figure 6

Forest plot of change in Exercise Related Quality of Life Scale (ERQoL) across schemes.

​Exercise Self Efficacy Scale

Only one scheme (5131) collected the ESES as an outcome measure and included 252 participants. Point estimate for change in ESES score adjusting for pre-ERS score was 2.58 pts (1.76 to 3.40).

Discussion

Despite widespread adoption, research exploring the effect of ERSs on health outcomes from ecologically valid datasets is scarce. The results presented here describe the initial findings from first analysis of the National Referral Database, a UK wide database of ERS based on the Standard Evaluation Framework for physical activity interventions. The analyses performed here were with the intention of considering broadly the following: do we observe a meaningful effect in people who are undergoing ERSs? The present results demonstrate that, although many health and well-being outcome changes are statistically significant when compared with point null estimates (ie, the results are unlikely under the assumption of a change of zero), our analysis revealed a general lack of meaningful change over time in participants undergoing ERSs lasting from 6 weeks to 3 months.

For BMI our results suggested that the change over time was statistically significant, yet unlikely to be meaningful (−0.55 kg.m2 (−0.69 to −0.41), p<0.0001, pδ=1.0). This is perhaps unsurprising as physical activity or exercise based interventions as standalone approaches seem to lack effectiveness for eliciting changes in body mass.37 Indeed, in their systematic review and meta-analysis, Pavey et al 18 reported no significant (or meaningful considering the criteria used here) differences for the effects of ERSs compared with other comparator interventions on BMI. Studies of ERS ranging from 8 weeks to 4 months show similarly little changes in BMI (ie, ranging from no change to ~0.6 kg/m2 22 38–40). Considering that participants undergoing ERSs in our sample were obese (BMI 31.53±6.53 kg/m2), lack of effectiveness could be considered an issue, particularly as ERSs have been noted as being tied to the current ‘obesity agenda’.41 With multiple lines of evidence suggesting that exercise alone may not result in meaningful weight loss, it may be argued that ERSs might benefit from combination with dietary guidance. There is perhaps potential here for wider social prescribing efforts regarding nutritional knowledge to coordinate with physical activity based interventions such as ERSs—particularly as a BMI classified as overweight or obese, although a crude measure, is on a population level associated with all-cause mortality risk.42 One explanation for the lack of effect observed here could be that positive changes in body composition occurred resulting in similar BMI post ERS (decrease in body fat and increase in lean mass). Such results have been shown to occur with supervised resistance training based ERSs43; so, despite the lack of meaningful change in BMI, participants in the ERSs examined here may have experienced positive changes in body composition. Indeed, the supplementary meta-regression performed revealed a significant positive coefficient for change in BMI with change in vigorous minutes as a moderator (0.47 kg/m2 (0.18 to 0.76), p=0.0013; see online supplementary materials), whereas this was not the case for total MET (metabolic equivalent) minutes, moderate minutes, walking minutes, or sitting minutes. This may reflect positive body composition changes as a result of increased vigorous activity.

Measures of cardiovascular health, including RHR and blood pressure, were common outcomes in many of the ERSs observed here. High RHR is known to be a risk factor for both cardiovascular and all-cause mortality,44 yet no change was observed over time in those participating in ERSs according to our analysis. Only one paper reporting the results of three small studies (n’s ranged from 11 to 65) has reported changes in RHR from ERS, and our data are supportive of these. Webb et al 40 reported point estimates ranging from +2.0 to −6.8 f c as a result of an 8 week ERS. Heart rate data from FitBit covering 150 billion hours of activity suggest that in those who are 55 years of age (similar to the sample in the current database), an increase of 180 min physical activity is needed to see an improvement in RHR of around ~6.5 f c.45 Previous meta-analysis of controlled trials has also shown an average change after endurance training of −6 f c.46 As noted, the exact ‘exercise prescription’ for ERS observed here is unknown and so it is difficult to know whether the lack of change in RHR might result from the implementation of exercise components that lack efficacy (ie, either of insufficient volume or intensity of effort), or are indeed due to poor fidelity of their implementation.

The change in SBP reported here was statistically significant, although its clinical meaningfulness should be treated with caution (−2.95 mmHg (−3.97 to −1.92), p<0.0001, pδ=0.039). Recent network meta-analysis shows that structured exercise interventions can produce similar reductions in SBP compared with antihypertensive medications,13 though their effect estimates were greater than those reported here (−4.84 mmHg (−5.55 to −4.13) across all populations). This may be due to interventions being examined in the context of randomised controlled trials as opposed to the observational data examined here from ‘real-world’ interventions. Reduction in SBP to a range of 120 to 124 mmHg may provide the greatest reductions in risk of cardiovascular disease and all-cause mortality, particularly if the reduction to this level is large.47 The mean baseline SBP for participants in the database was 131.70±16.91 mmHg; thus, considering the results of Bundy et al 47 highlighting that large SBP reductions produce the greatest mortality risk reduction, in combination with the small reduction reported here, it is unclear whether sufficient reduction might have occurred to have meaningfully reduced risk. Though other studies of ERSs also report reductions in SBP, these are either similarly small (−2.84 mmHg (−6.57 to 0.82) and −3.53 mmHg (−7.31 to 0.25)22; −3.2 mmHg (−4.6 to 1.7) and −2.9 mmHg (−4.4 to 1.4)38) or variable in their point estimates (−6.1 to +4.8 mmHg40). This is perhaps unsurprising as meta-analysis48 suggests reductions are small for endurance type exercise (−3.5 mmHg (−4.6 to −2.3) and dynamic resistance training (−1.8 mmHg (−3.7 to −0.011)). The exception to this appears to be isometric exercise which produced far larger reductions (−10.9 mmHg (−14.5 to −7.4)48), suggesting that this form of exercise should perhaps be incorporated into ERSs. Cornelissen and Smart48 also reported that reductions in blood pressure may be greater in prehypertensive or hypertensive participants; therefore, considering also the results of the recent network meta-analysis by Bundy et al,47 it could be that the typical participant referred to an ERS may be unlikely to benefit meaningfully from it.

A statistically significant reduction in DBP was observed, yet this was unlikely to be clinically meaningful (−0.93 mmHg (−1.51 to −0.35), p=0.0016, pδ=1.00). Though a reduction in SBP to a range of 120–124 mmHg may reduce cardiovascular and all-cause mortality risk,47 recent evidence also suggests that, when SBP is in the range of 120–140 mmHg, a DBP of 70–80 mmHg is optimal for reduced morbidity and mortality risk.49 Participants in the database had baseline DBPs at the high end of the optimal range on average (80.13±9.91 mmHg) and so the small reduction found here might be considered meaningful in context; yet, it is unlikely that such a dichotomous consideration is accurate. As with SBP, prior studies typically show either no effects, or very small and variable effects, on DBP with ERSs (0.77 mmHg (−2.07 to 3.61) and 1.55 mmHg (−1.02 to 4.11)22; −2.5 mmHg (−3.5 to −1.5) and 0.2 mmHg (−3.8 to −0.8)38; −4.1 mmHg to 3.0 point estimates40). Again this is perhaps unsurprising given the small effects reported in meta-analysis48 for endurance exercise (−2.5 mmHg (−3.2 to −1.7)) and dynamic resistance training (−3.2 mmHg (−4.2 to −2.0)), or combined training (−2.2 mmHg (−3.9 to −0.48)), though again isometric exercise appears more effective than traditional approaches (−6.2 mmHg (−10.30 to −2.0)). As noted, we were unable to consider the impact of different exercise approaches used within the ERSs examined here; therefore, again it may be that the relatively small change in DBP might result from the implementation of exercise components that lack efficacy (ie, either of insufficient volume or intensity of effort).

Other outcomes included a variety of questionnaires relating to mental well-being, quality of life, and self-efficacy relating to exercise. Only three schemes examined SWEMWBS, and though the improvement was statistically significant, initial analysis suggested that it was unclear as to whether the change was meaningful (2.99 pts (1.61 to 4.36), p<0.0001, pδ=0.422). Sensitivity analysis did reveal one influential scheme and removal of that scheme improved the estimate from meta-analysis (3.81 (2.73 to 4.89), p<0.0001, pδ=0.018). That scheme, however, was considerably larger than the other two included (n=1479 vs 107 and 39). Thus it is possible that the point estimate from this scheme is better representative of μ and indeed the 95% CIs are far narrower, suggesting better precision of the estimate. Only one study to our knowledge has examined SWEMWBS specifically in response to exercise based intervention. Skinner et al 50 reported a point estimate for change in SWEMWBS of 1.5 pts (−1.0 to 4.0) after a short 4 week supervised exercise intervention once a week in prostate cancer survivors. Malcolm et al 51 used the full WEMWBS and reported a change of 3 pts (0.0 to 6.0) and, though both measures subtly different components of mental well-being, considering the relationship between both the SWEMWBS and WEMWBS52 it seems as though this reflects a similar sized effect. Recent large scale observational research53 highlights that physical activity is significantly and meaningfully associated with mental health (1.49 fewer days of poor mental health in the past month). However, randomized controlled trial evidence from ERSs54 initially appear less supportive of a role for physical activity and exercise in mental health, in this case depression; though, recent analysis suggests that many studies and media portrayals of the role of physical activity and exercise are misrepresented.55 Other studies have shown reductions in anxiety and depression from ERSs in both those with and without prior mental health issues.22 56 Despite a lack of agreement within the extant literature it would appear that physical activity and exercise can have positive effects on mental health. Indeed, recent work shows that resistance training, for example, improves both anxiety and depressive symptoms regardless of prior health status,57 58 and meta-analysis of broad exercise approaches for depression controlling for publication bias support an effect.59 Our results suggest this effect may also be present in those undergoing ERSs, yet the clinical meaningfulness of the effect is less clear.

Global well-being and quality of life related outcomes were also explored in some ERSs. One scheme collected the WHO-5 wellbeing index where there was a statistically significant, yet likely not meaningful, change (8.78 pts (6.84 to 10.63), pδ=0.834). Few studies have used the WHO-5 as an outcome measure,30 though one has compared exercise to wake therapy in those with major depression,60 reporting a point estimate for improvement from exercise of 46 pts. A study in older adults performing a 6 month progressive high effort resistance training intervention61 found an improvement of 7.26 pts (5.78 to 8.74). It would seem that responsiveness of the WHO-5 to exercise based interventions may be at least in part dependent on the population as large effects have been reported for major depression,60 whereas the effects reported for older adults,61 and here in ERS participants, do not typically exceed the minimal clinically important change.30 A relatively novel outcome measure designed specifically for evaluation of quality of life in those participating in ERSs31 was also included in some schemes. Results from the meta-analysis suggested that there was a statistically significant improvement in this outcome (15.26 pts (4.71 to 25.82), p=0.0046), and that sensitivity analysis revealed an increase in the precision of estimate for the effect when one influential scheme was removed (9.83 pts (8.70 to 10.95), p<0.001). However, though specifically designed for use with ERSs, this outcome lacks research regarding what constitutes a minimal clinically important change and so it is difficult to determine whether the improvements seen are indeed meaningful.

The final outcome examined, ESES, was only collected by one scheme. ESES showed a significant change of 2.58 pts (1.76 to 3.40), though the lack of prior literature examining this as an outcome measure makes it difficult to determine the meaningfulness of this change. Exercise interventions have been shown to improve general self-efficacy.62 Furthermore, the extent to which self-efficacy is affected by an intervention may impact on their ability to produce behaviour change (eg, for web based health interventions63). It has also been shown that changes towards higher exercise self-efficacy predicts continuation of exercise behaviour once an intervention has ended.64 Thus it might be that the results here suggesting an impact of ERS on ESES might lead to longer term maintenance of behaviour.

As noted, currently there is a lack of agreement of what constitutes ‘impact’ with respect to the evaluation of ERSs41 and the evidence presented here from one of the largest databases of ERS does little to support the use of ERSs, broadly speaking. It should be noted that the database and analyses presented here are not without limitations (observational data without control group for counterfactual, follow-up bias due to high proportion of dropouts, selection of null intervals for meaningfulness of effects, details of specific components of ERS including fidelity, or consideration of participant characteristics such as referral reasons, etc) and these are detailed further in the accompanying manuscript describing the initial overview of the database.27 Indeed, in considering factors such as lack of appropriate controls and follow-up bias, it may be that the overall effects reported may be less in reality. However, given the considerable heterogeneity seen between schemes (I 2 >78% across outcomes), some schemes may be more effective than others due to characteristics either relating to the scheme specifically, or perhaps characteristics of the types of participants undergoing that scheme. Given that scheme level characteristics regarding delivery are not presently available in this dataset, it is unfortunately difficult to explore the effects of personal level characteristics (eg, BMI pre-ERS).

Conclusion

The data presented here represent the initial findings from first analysis of the National Referral Database health and well-being data. The analyses performed were with the intention of considering broadly the question “do we observe a meaningful effect in people who are undergoing ERSs?”; the findings revealed a general lack of meaningful change over time in participants undergoing ERSs lasting between 6 weeks and 3 months in length. These findings suggest the need to consider the implementation of ERSs more critically to discern how best to maximise their potential in light of the wider literature supporting the efficacy of physical activity and exercise, and the extensive reach of ERSs across the UK.

What is already known on this topic

  • Physical activity is widely considered to be effective in the prevention, management, and treatment of many chronic health disorders, yet population physical activity levels are relatively low and have changed little in recent years.

  • Sufficient physical activity levels for health and well-being often do not arise as a result of typical activities of daily living; thus, it has been argued that specific exercise is necessary for many, and one approach to providing this has been through exercise referral schemes.

  • These are aimed at increasing physical activity levels in sedentary individuals with chronic disease; however, evidence is currently lacking as to whether exercise referral schemes are effective as currently implemented.

What this study adds

  • Our findings suggest that, though exercise referral schemes are associated with statistically significant changes in most health and well-being outcomes, the size of the changes was not as clinically meaningful as would be hoped for.

  • These findings support the need to consider exercise referral schemes, particularly their implementation, more critically using real world data to understand how best to maximise their potential, particularly considering the known benefits of exercise and the reach of ERSs across the UK.

Acknowledgments

The authors would like to offer their thanks to Alan Batterham for his insight regarding the statistical approaches taken to analysing the data presented here.

References

Footnotes

  • Twitter @jamessteeleii

  • Contributors The corresponding author attests that all listed authors meet authorship criteria and that no others meeting the criteria have been omitted. JS/MW/RC/SM had the idea for the article. JS/MW wrote the initial draft. JS conducted the statistical analysis. All authors contributed additional writing to the first draft, critical review and editing, and signed off on the final draft.

  • Funding The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.

  • Competing interests None declared.

  • Patient consent for publication Not required.

  • Ethics approval As per the Health Research Authority and Research Ethics Committee section 11 of Standard Operating Procedures, ethical approval is not required for research involving patient data that are not identifiable.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Data availability statement Data are presently available upon reasonable request. We are working to establish the database as an open resource with continuous version controlled datasets (see Steele et al. for details

    https://osf.io/preprints/sportrxiv/rgywq/).

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.