A systematic review and economic evaluation of exercise referral schemes in primary care: A short report

Identification of studies

The search strategy comprised the following main elements:

• searching of electronic databases

• contact with experts in the field

• scrutiny of bibliographies of retrieved papers.

The search strategies used in the Pavey et al. 11 systematic review of the evidence were used in this review. These consisted of two search strategies (stage 1 and stage 2), details of which are included in Appendix 1 of this report. The stage 1 search was a focused phrase search, with stage 2 being a more sensitive search combining the terms for exercise referral with study type and setting terms (primary care). Searches were limited by English language and a publication date of October 2009 to current (8 May 2013 for stage 1 and 17 June 2014 for stage 2). SPORTDiscus was not available to the research team; therefore, Scopus via Elsevier was used, and the stage 1 search was conducted in this data source. Key sports and exercise science journals such as Medicine and Science in Sports and Exercise and International Journal of Sports Psychology have been covered, as they are indexed in one or more of the databases listed below. The Journal of Aging and Physical Activity was found not to be indexed in the databases searched, hence this journal was hand-searched by scanning the electronic table of contents available at http://journals.humankinetics.com/japa-contents (2009–current and in-press articles as of September 2013).

Sibling studies

In order to identify any sibling studies (qualitative studies conducted as part of a mixed-method evaluation of the intervention), two searches were undertaken. First, the names of authors and project names of the included trials papers were searched for in Google Scholar (Google, Mountain View, CA, USA). To augment this search, citation searches of the included trials were undertaken in the Science Citation Index and proceedings and Social Science Citation Index and proceedings [via Web of Science Thomson Institute for Scientific Information (ISI)].

The following electronic databases were searched: MEDLINE and MEDLINE In-Process & Other Non-Indexed Citations (via Ovid); EMBASE (via OvidSP); PsycINFO (via OvidSP); Scopus (via Elsevier); The Cochrane Library, including the Cochrane Database of Systematic Reviews (CDSR), Cochrane Central Register of Controlled Trials (CENTRAL), NHS Health Technology Assessment (HTA) Database, NHS Economic Evaluation Database (NHS EED), Database of Abstracts of Reviews of Effects (DARE); Science Citation Index and proceedings and Social Science Citation Index and proceedings (via Web of Science Thomson ISI), UK Clinical Research Network Study Portfolio (UKCRN) portfolio database; Current Controlled Trials; and ClinicalTrials.gov.

An example of the stage 1 and 2 search strategies is shown in Appendix 1.

Inclusion/exclusion criteria

Quantity and quality of research available

Our search of electronic databases and relevant journals yielded 9627 titles, of which one primary study was judged to meet the inclusion criteria. Figure 1 summarises the process of identifying the inclusion and exclusion process. In two studies,22,23 additional data were supplied by the authors. One study,24 identified in bibliographic searching, could not be retrieved. The main reasons for excluding studies included non-RCT design (n = 44), participants recruited from primary care for inclusion in an exercise programme but without referral from a health-care professional (n = 20), the intervention was a prescription to undertake exercise but not a referral to a third-party exercise provider (n = 5), the population was not appropriate (n = 1), participants were already part of an intervention prior to randomisation (n = 1) or randomisation occurred prior to the baseline assessment (n = 1). Appendix 2 provides the full list of excluded studies. The included studies are summarised in Table 1, which includes the studies incorporated in the Pavey et al. review. 11 The new evidence is highlighted in bold type within the tables.

FIGURE 1.

Flow diagram demonstrating the process of identifying new studies for inclusion in the review.

Characteristics of included studies

The characteristics of the included ERS studies are summarised in Table 2. All of the included studies were RCTs. The Pavey et al. 11 review included seven trials. 12,15,18,26–28,31 The data from these studies are included in the tables of this review, with new data emboldened. One additional study was identified (four publications). 23,25,26,32 This study was a mixed-methods evaluation, incorporating RCT and qualitative evidence, undertaken in Wales. It was larger than previous studies, with 2160 participants. The earlier studies ranged in size from 52 to 943 participants. The total number of participants in all eight studies was 5190.

The studies of Duda et al. 17 and Gusi et al. 28 used cluster allocation, with the other studies using individual-level randomisation. Follow-up duration ranged from 2 to 12 months. The general practitioner (GP) was the main referrer, usually using a bespoke referral form to a fitness or exercise instructor/officer. The Murphy et al. 23 study included referrals by health professionals working in a range of health-care settings.

Five studies12,15,26,23,28 compared ERSs with a usual-care group, which consisted of no exercise intervention or simple advice on physical activity. Sorensen et al. 27 compared ERSs with motivational counselling aimed at increasing daily physical activity. The Isaacs et al. study18 also included an instructor-led walking programme. The Duda et al. study17 compared two forms of ERSs, that is, standard ERS versus a combined ERS plus self-determination theory (SDT)-based intervention.

Characteristics of participants

A total of 5190 participants were included in the eight trials. 12,15,17,18,23,26–28 A summary of the characteristics of participants is presented in Table 3. All of the studies recruited participants who were sedentary or who were believed by their GP to be able to improve health by an increased physical activity level. They all also excluded individuals who had poorly controlled hypertension, diabetes mellitus or heart disease. Gusi et al. 28 also excluded those with severe obesity or major depression. Sorensen et al. 27 included participants who were willing to pay 750 Danish krone for the intervention.

The participants included in the Murphy et al. study23 shared a similar profile to those already included in earlier studies. Most of those recruited were middle-aged white adults, with at least one medical condition, who might benefit from an increased level of physical activity (Table 4). In the Murphy et al. study,23 72% of participants had CHD risk factors and 24% had mental health problems.

Characteristics of interventions

The characteristics of the interventions are summarised in Table 5. Interventions ran for between 10 weeks14,21,33 and 6 months28 and included instructor-led exercise classes or walks,28 some form of consultation aimed at increasing activity22,33 or a combination of the two. 14,17,21,23,34 The intervention by Sorensen et al. 27 was based on the transtheoretical model of behaviour change, which categorises people into stages of readiness for change and postulates 10 experiential and behavioural processes of change, pro and con beliefs and self-efficacy beliefs (beliefs in personal ability to carry out the behaviour in question). 22 The Duda et al. intervention17 was based on SDT, which emphasises the determinants and consequences of different reasons for behavioural engagement, which vary in their degree of self-determination. 23 The intervention aimed to promote autonomy support, which is assumed to foster participants’ feelings of competence, autonomy and relatedness and, as a result, enhance autonomous motivation for physical activity and associated mental health outcomes. 17 The only study added since the Pavey HTA,11 by Murphy et al. ,23 was also the only study to explicitly use motivational interviewing. However, fidelity to this intervention method during consultations was considered to be low. Motivational interviewing is ‘a collaborative, person-centred form of guiding to elicit and strengthen motivation for change’. 35 In those studies that held supervised exercise sessions or walks, the number of sessions per week varied between one32 and three,28 with the most common being two sessions per week,14,17,21 although one study27 held two sessions per week for the first 2 months and then one session per week for the following 2 months. Supervised exercise sessions or walks lasted between 30–40 minutes12 and 1 hour. 17,34 Two interventions were group interventions28,34 and four were at group and/or individual levels. 14,17,22,23

Risk of bias

Table 6 summarises the risk of bias for each of the included studies. Most included a power calculation and allocated participants using an appropriately generated random number sequence. However, the reporting of concealment of trial group allocation was poor, although there was good evidence of participant characteristics of intervention and control groups at baseline. Although blinding of participants and intervention providers in these studies was not feasible, blinding of outcome assessment was possible. Outcome blinding is particularly important in preventing assessment bias in the case of outcomes that require observer judgement or involvement (e.g. blood pressure measurement or exercise testing). Two studies17,23 reported outcome blinding. Recall Questionnaire was assessed via telephone to maintain blinding. The reporting and handling of missing data were detailed for most studies, and all studies, except one,12 reported the use of ITT analysis. The level of missing data at follow-up ranged across studies from 16.5% to 50%. Most studies used imputation methods (last observation carried forward or complete case average values) to replace missing data values at follow-up. Overall, three studies12,15,26 were judged to be at moderate overall risk of bias and five17,18,23,27,28 to be at low overall risk of bias.

Exercise referral scheme eligibility, uptake and adherence

There was a considerable range in the proportion of individuals randomised compared with those deemed eligible (Table 7). In both the Sorensen et al. 27 and Duda et al. 17 studies, of those deemed eligible for ERSs, a substantial number refused participation in the trial. In the Sorensen et al. 27 study this low number may be reflective of the 750 Danish krone payment by patients as part of a standard Danish Exercise on Prescription. In the Duda et al. 17 study, this may be related to the workload and training needs of the health and fitness advisors at the time of recruitment.

The terms ‘adherence’ and ‘uptake’ can be used variably within the literature. ‘Uptake’ refers to initial attendance, take-up or enrolment. ‘Adherence’ describes the level and duration of participation, and the threshold for determining adherence may vary in different studies. 34

Rates of uptake varied in the included studies. Taylor et al. ,12 Isaacs et al. ,18 Sorensen et al. 27 and Murphy et al. 23 reported uptake rates in excess of 85%, and in the Stevens et al. study26 only 126 (35%) of the 363 randomised to ERSs attended the first consultation. Stevens et al. 26 discussed how the low uptake they experienced may have been reflective of the nature of the invitation letter sent to participants and the point of randomisation (pre-invitation letter). Furthermore, they hypothesise that a change in the format of the letter (e.g. including a specific appointment date for the first ERS appointment) would have improved participation. Uptake was not reported by Duda et al. 17 or Gusi et al. 28

Adherence was assessed differently between the trials, and levels of adherence also varied. Stevens et al. 26 and Gusi et al. 28 reported ERS programme completion rates of 25% and 86%, respectively. However, these rates do not reflect the number of sessions attended, only those who attended a second consultation26 or follow-up assessment. 28

Sorensen et al. 27 reported that an average of 18 out of a total of 24 ERS exercise sessions were attended and 68% and 75% of participants attended the counselling sessions at 4 and 10 months, respectively. Both Taylor et al. 12 and Isaacs et al. 18 provide a detailed description of ERS programme adherence. Taylor et al. 12 reported 13% attending no exercise sessions and 28% attending 75–100% of exercise sessions, with an average of 9.1 out of 20 prescribed exercise sessions attended. Isaacs et al. 18 reported 7.6% attending no exercise sessions and 42% attending 75–100% of exercise sessions in the leisure centre group. In the walking group, 23.5% attended no exercise sessions, with 21.5% attending 75–100% of exercise sessions. As shown in Table 8, there was no consistent difference in attendance rates between those in at-risk groups and the overall study population in the studies of Taylor et al. 12 and Isaacs et al. 18 In the Isaacs et al. study,18 the 60–69 years age group had the highest adherence in both the ERS (53.3%) and the walking (24.2%) groups. There were no significant differences in attendance rate based on employment status, educational level, socioeconomic status, ethnicity or relationship status. However, Murphy et al. 23 did find differences in uptake and adherence between participants from deprived and less-deprived areas. Adherence was lower for those without access to private transport in both the ERS and walking groups. Harrison et al. 15 and Duda et al. 17 did not provide information on participants’ adherence to the ERS intervention.

Murphy et al. 23 found that participants already active at baseline were the most likely to enter the ERS, but that they were also most likely to partially attend a programme. Men and younger participants were slightly less likely to enter the scheme, and women were less likely to complete the programme. Participants in the least-deprived areas were more likely to take up the scheme, although the differences in adherence between the most- and least-deprived areas were smaller. Non-car owners were less likely to take up the scheme or to adhere to it if they joined. Those referred for mental health reasons were more likely not to enter the ERS, and only one in three mental health patients completed the programme.

In a univariate regression analysis of predictors of uptake and adherence, the only significant correlates of uptake in the Murphy et al. study23 were car ownership and deprivation. Those in moderately deprived areas were less likely to enter the ERS than those in the least-deprived areas. Car owners were significantly more likely to enter the ERS than non-car owners. Older participants, those referred for non-weight-related CHD risk factors, non-mental health patients and those already moderately active at baseline were most likely to complete the programme.

In a multivariable regression analysis, the significant difference between low and medium deprivation areas remains and car ownership remains predictive of uptake. Associations of CHD risk factors with adherence become non-significant and associations of age and mental health status remain significant. Associations of baseline activity with adherence are strengthened in the multivariable analysis, with the contrast between inactive and moderately inactive participants becoming more significant (see Barriers and facilitators of referral, uptake and adherence to exercise referral schemes for further discussion of factors influencing uptake and adherence).

Assessment of effectiveness

Only Isaacs et al. 18 reported all outcome domains applicable to this systematic review (Table 9). New data added to the Pavey et al. 11 review are emboldened within the tables.

Physical activity

All studies, with the exception of Gusi et al. ,28 provided a measure of self-reported physical activity. Self-reported measures included the validated 7-Day Physical Activity Recall (7-Day PAR) questionnaire12,23,26,27 and the validated Minnesota Leisure Time Activity questionnaire. 18 None of the studies reported methods of measuring physical activity using an objective method of measurement, and all relied on self-report tools.

Exercise referral schemes versus usual care/advice only

The most consistently reported physical activity outcome across studies was the proportion of individuals achieving 90–150 minutes of at least moderate-intensity activity per week. Data for this outcome in the Murphy et al. study23 were supplied by the author (Professor Simon Murphy, Cardiff University, 2013, personal communication). When pooled across studies the relative risk (RR) was 1.12 [95% confidence interval (CI) 1.04 to 1.20] of achieving this outcome with ERS compared with usual care at 6–12 months’ follow-up (Figure 2). There was no evidence of heterogeneity in this analysis (I² 0%). This analysis draws on data published by Pavey et al. 11 Three studies12,15,23 reported this outcome based on the number of individuals who were available at follow-up. These results show a decrease in the RR found by Pavey et al. 11 (RR 1.16, 95% CI 1.03 to 1.30). In order to assess the potential (attrition) bias in using completers, the denominators of these three studies were adjusted to all individuals randomised in order to perform an ITT analysis (Figure 3). It was assumed that all missing cases did not meet the physical activity threshold. In the pooled ITT analyses, the proportion achieving the physical activity threshold in the ERS group compared with usual care was RR 1.08 (95% CI 1.00 to 1.17). There was no evidence of heterogeneity in this analysis (I² 0%). This is also a reduction on the RR found by Pavey et al. 11 (RR 1.11, 95% CI 0.99 to 1.25).

FIGURE 2.

Number achieving 90–150 minutes physical activity/week (updated meta-analysis). df, degrees of freedom; M–H, Mantel–Haenszel; PA, physical activity.

FIGURE 3.

Number achieving 90–150 minutes physical activity/week (ITT analysis) (updated meta-analysis). df, degrees of freedom; M–H, Mantel–Haenszel; PA, physical activity.

Total minutes of physical activity were reported by Isaacs et al. 18 and Murphy et al. 23 When these data were pooled, there was a significant increase in the number of minutes of physical activity per week in the ERS group; mean difference 55.10 minutes (95% CI 18.47 to 91.73 minutes) (Figures 4–6).

FIGURE 4.

Minutes spent in at least moderate-intensity physical activity per week at 6–12 months’ follow-up. 11 df, degrees of freedom; IV, inverse variance; PA, physical activity; SD, standard deviation.

FIGURE 5.

Minutes of total physical activity/week at 6–12 months’ follow-up ERS vs. advice only (updated meta-analysis). df, degrees of freedom; IV, inverse variance; SD, standard deviation.

FIGURE 6.

Minutes of total physical activity/week at 6–12 months’ follow-up ERS vs. alternative physical activity. 11 df, degrees of freedom; IV, inverse variance; PA, physical activity; SD, standard deviation.

Exercise referral schemes versus alternative physical activity intervention

Sorensen et al. 27 reported a higher level of energy expenditure with ERS than with physical activity counselling. In contrast, the study by Isaacs et al. 18 observed a higher level of physical activity (minutes of total and moderate-intensity activity, energy expenditure) in those in the walking programme than in the ERS group. When pooled across studies, there was no significant difference in the total amount of physical or energy expenditure between ERSs and alternative physical activity interventions (Figures 7 and 8).

FIGURE 7.

Energy expenditure (kcal/kg/day) ERSs vs. usual care at 6–12 months’ follow-up. 11 df, degrees of freedom; IV, inverse variance; SD, standard deviation.

FIGURE 8.

Energy expenditure ERSs vs. alternative physical activity intervention at 5–12 months’ follow-up. 11 df, degrees of freedom; IV, inverse variance; SD, standard deviation.

Exercise referral schemes versus a self-determination theory informed delivery of exercise on referral

In the Duda et al. study,17 the proportion of patients achieving at least 150 minutes of moderate physical activity per week increased in the standard ERS group from 27% at baseline to 63% at 3 months and 46% at 6 months. There were no significant differences in these proportions between the standard ERS and a SDT-informed delivery of ERSs (Table 10).

TABLE 10 - Summary of physical activity data at follow-up

PA, physical activity; SD, standard deviation.

Sorensen et al. 27 metabolic equivalent (METS)/hour/day.

Numbers of individuals with complete data/questionnaires.

Between-group difference not statistically significant at p-value ≤ 0.05.

The p-value calculated by authors of the present report.

Between-group difference statistically significant at p-value ≤ 0.05.

All randomised participants.

Provides b (patients achieving PA guidance) and f (all other PA measures).

Mean change score.

New evidence is highlighted in bold in the table.

Study and time of follow-up	Patients achieving PA guidance (90–150 minutes/at least moderate intensity per week)		At least moderate intensity PA (minutes per week)				Total PA (minutes per week)		Energy expenditure (kcal/kg/day)a
	ERS, n/N	Control, n/N	ERS, mean (SD)		Control, mean (SD)		ERS, mean (SD)	Control, mean (SD)	ERS, mean (SD)	Control, mean (SD)
			Moderate	Vigorous	Moderate	Vigorous
ERS vs. usual care
Taylor et al.12
8 weeksb	51/63	20/31c,d	247 (174)	49 (60)	145 (178)e	21 (61)e	Not reported	Not reported	34.6 (1.2)	33.7 (1.7)e
16 weeksb	51/57	18/31d,e	226 (252)	59 (72)	160 (262)c	21 (72)e	Not reported	Not reported	34.6 (1.2)	33.9 (1.7)e
26 weeksb	39/47	18/31d,e	183 (234)	56 (108)	206 (251)c	34 (111)c	Not reported	Not reported	34.4 (1.8)	34.3 (1.2)c
37 weeksb	39/57	19/31c,d	158 (228)	42 (96)	162 (245)c	23 (106)c	Not reported	Not reported	34.1 (2.4)	33.9 (2.2)c
Stevens et al.26
8 monthsf	204/363	174/351c,d	Not reported	–	Not reported	–	Not reported	Not reported	Not reported	Not reported
Harrison et al.15
6 monthsb	38/168	22/162d,e	Not reported	–	Not reported	–	Not reported	Not reported	Not reported	Not reported
9 monthsb	36/149	31/140c	Not reported	–	Not reported	–	Not reported	Not reported	Not reported	Not reported
12 monthsb	40/155	32/157c	Not reported	–	Not reported	–	Not reported	Not reported	Not reported	Not reported
Isaacs et al.18
10 weeksg	48/164	29/157d,e	93 (115)	–	79 (114)c	–	584 (479)	668 (555)c	34 (26)	36 (32)c
6 monthsg	70/179	66/200c,d	65 (106)	–	58 (98)c	–	692 (496)	647 (463)c	38 (27)	35 (27)c
Gusi et al.28
6 months	Not reported	Not reported	Not reported	–	Not reported	–	Not reported	Not reported	Not reported	Not reported
Murphy et al. 23
12 months	431/724	409/755	–	–	–	–	335.53 (442.47)	277.42 (371.70)	–	–
ERS vs. alternative PA intervention
hSorensen et al.27
4 monthsb	Not reported	Not reported	Not reported	–	Not reported	–	63 (114)	23 (107)c	43 (2.4)	41 (4.8)
10 monthsb	Not reported	Not reported	Not reported	–	Not reported	–	20 (124)	20 (152)c	41 (2.1)	40 (5)
Isaacs et al.18
10 weeksg	48/164	53/92d,e	93 (115)	–	113 (291)c	–	584 (479)	863 (1026)e	34 (26)	43 (38)e
6 monthsg	70/179	62/141c,d	65 (106)	–	89 (150)e	–	692 (496)	759 (539)c	38 (27)	42 (27)c
ERS vs. SDT-informed ERS
Duda et al.17
3 monthsg	Not reported	Not reported	319 (338)c	–	331 (336)c	–	Not reported	Not reported	Not reported	Not reported
6 monthsg	66/156	83/169c,d	249 (356)c	–	246 (343)c	–	Not reported	Not reported	Not reported	Not reported

Subgroup analysis exploring the impact of patient variables on effects of exercise referral schemes on levels of physical activity

Pavey et al. 11 reported the following analysis of subgroups within seven studies: Harrison et al. 15 reported no statistical significant interaction effects between the ERS effect and pre-specified baseline variables (i.e. CHD risk factors, sex and age). Comparing high adherers (> 75% attendance at ERS) with low adherers (< 75% attendance at ERS) in the Isaacs et al. 18 study, 32 high adherers and 16 low adherers were achieving > 150 minutes of moderate physical activity per week at 10 weeks. At 6 months, 41 high adherers and 29 low adherers were achieving > 150 minutes of moderate physical activity per week. However, these proportions were not significantly different. In the Duda et al. study,17 age, sex, deprivation (Index of Multiple Deprivation score), ethnicity, depression at baseline and level of physical activity at baseline were assessed by regression methods as predictors of physical activity at 6 months. Only physical activity at baseline was associated with physical activity at 6 months’ follow-up (p < 0.001). Murphy et al. 23 also found that effectiveness was highly dependent on adherence, with significantly greater differences in all outcomes among those who completed the 16-week programme compared with those who attended only partially or not at all.

Murphy et al. 23 reported that referral and participation in ERSs increased physical activity significantly for those referred for CHD risk factors [odds ratio (OR) 1.29, 95% CI 1.04 to 1.60]. However, among those referred for mental health reasons, either solely or in combination with CHD, there was no difference in physical activity between the ERSs and normal-care participants at 12 months’ follow-up. The effect of being in the ERS group on all referrals was an increase in levels of physical activity at 12 months, but this finding was of borderline statistical significance (OR 1.19, 95% CI 0.99 to 1.43).

Physical fitness

No additional data were available for this outcome; the following results are taken from Pavey et al. 11 The studies by Taylor et al. ,12 Isaacs et al. 18 and Sorensen et al. 27 reported physical fitness outcomes (Table 11).

TABLE 11 - Summary of physical fitness data at follow-up in included ERS trials11

N/A, not applicable; PA, physical activity; SD, standard deviation.

Numbers of individuals with complete data/questionnaires.

Between-group difference not statistically significant at p-value ≤ 0.05.

Between-group differences statistically significant at p-value ≤ 0.05.

Study and time of follow-up	Mean predicted heart rate at a workload of 150 W		VO2max (ml/kg/minute)		Submaximal bike ergometer (minutes)		Submaximal shuttle walk (m)		Isometric knee strength (N)		Leg extension power (W)
	ERS, mean (SD)	Control, mean (SD)	ERS, mean (SD)	Control, mean (SD)	ERS, mean (SD)	Control, mean (SD)	ERS, mean (SD)	Control, mean (SD)	ERS, mean (SD)	Control, mean (SD)	ERS, mean (SD)	Control, mean (SD)
ERS vs. usual care
Taylor et al.12
16 weeksa	138.6 (23.0)	147.2 (29.7)b	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported
26 weeksa	136.3 (22.6)	142.3 (28.5)b	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported
37 weeksa	134.2 (19.0)	146.0 (24.2)c	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported
Isaacs et al.18
10 weeksa	Not reported	Not reported	Not reported	Not reported	9.65 (1.5)	8.87 (1.5)c	456 (102)	434 (104)b	277 (54)	265 (56)b	174 (31)	165 (31)b
6 monthsa	Not reported	Not reported	Not reported	Not reported	8.86 (1.7)	9.08 (1.7)b	445 (96)	434 (97)b	265 (58)	267 (66)b	173 (66)	167 (68)b
ERS vs. alternative PA intervention
Sorensen et al.27
4 monthsa	Not reported	Not reported	23.8 (7.1)	21.7 (11.0)b	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported
10 monthsa	Not reported	Not reported	23.0 (8.2)	22.4 (12.7)b	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported
Isaacs et al.18
10 weeksa	Not reported	Not reported	Not reported	Not reported	9.65 (1.5)	8.92 (1.7)b	456 (102)	437 (100)b	277 (54)	275 (58)b	174 (31)	166 (32)b
6 monthsa	Not reported	Not reported	Not reported	Not reported	8.86 (1.7)	8.92 (1.8)b	445 (96)	448 (95)b	265 (58)	264 (66)b	173 (66)	164 (68)b

Exercise referral schemes versus usual care

Taylor et al. 12 reported a lower (more favourable) submaximal heart rate (at 150 W) for ERS compared with usual care. Isaacs et al. 18 reported no significant differences in any of the physical fitness measures (submaximal bike and shuttle walk, isometric knee strength, leg extension power) between the ERS and usual-care groups at follow-up, except at 10 weeks for the submaximal bike ergometer test. Pooling of the cardiorespiratory measures (mode: cycle ergometer or cycle/walking) showed no difference between ERSs and usual care (Figure 9). There was considerable evidence of statistical heterogeneity.

FIGURE 9.

Physical fitness at 6–12 months’ follow-up. df, degrees of freedom; IV, inverse variance; PA, physical activity; SD, standard deviation.

Exercise referral schemes versus alternative physical activity intervention

Isaacs et al. 18 and Sorensen et al. 27 reported no significant differences in any of the physical fitness measures between the ERS and the alternative physical activity intervention groups at follow-up (see Figure 9).

Exercise referral schemes versus exercise referral scheme plus self-determination theory

The study of Duda et al. 17 did not assess physical fitness.

Clinical factors

Five studies provided information on clinical outcomes, that is, CHD risk factors (Table 12), weight and obesity measures (Table 13) and respiratory function (Table 14).

TABLE 12 - Summary of CHD risk factors in included ERS trials

DBP, diastolic blood pressure; N/A, not applicable; PA, physical activity; SBP, systolic blood pressure; SD, standard deviation.

Numbers of individuals with complete data.

Between-group difference not statistically significant at p-value ≤ 0.05.

Study and time of follow-up	SBP (mmHg)		DBP (mmHg)		Cholesterol (mmol/l)		High-density lipoproteins (mmol/l)		Low-density lipoproteins (mmol/l)		Triglycerides (mmol/l)
	ERS, mean (SD)	Usual care, mean (SD)	ERS, mean (SD)	Usual care, mean (SD)	ERS, mean (SD)	Usual care, mean (SD)	ERS, mean (SD)	Usual care, mean (SD)	ERS, mean (SD)	Usual care, mean (SD)	ERS, mean (SD)	Usual care, mean (SD)
ERS vs. usual care
Taylor et al.12
16 weeksa	130 (14.5)	130 (14)b	84 (8)	84 (8)b	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported
26 weeksa	130 (14)	131 (14)b	84 (8)	84 (8)b	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported
37 weeksa	130 (17)	131 (18)b	85 (9)	83 (9)b	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported
Isaacs et al.18
10 weeksa	133 (10)	132 (10)b	82 (6)	83 (6)b	5.68 (0.53)	5.71 (0.42)b	1.35 (0.18)	1.35 (0.18)b	3.41 (0.46)	3.44 (0.47)b	2.12 (0.71)	2.14 (0.71)b
6 monthsa	133 (12)	133 (12)b	82 (6)	82 (7)b	5.65 (0.50)	5.60 (0.50)b	1.37 (0.25)	1.38 (0.17)b	3.40 (0.48)	3.37 (0.50)b	2.04 (0.74)	2.00 (0.84)b
ERS vs. alternative PA intervention
Isaacs et al.18
10 weeksa	133 (10)	134 (10)b	82 (6)	84 (6)b	5.68 (0.53)	5.69 (0.53)b	1.35 (0.18)	1.33 (0.17)b	3.41 (0.46)	3.45 (0.46)b	2.12 (0.71)	2.05 (0.76)b
6 monthsa	133 (12)	134 (12)b	82 (6)	83 (6)b	5.65 (0.50)	5.56 (0.57)b	1.37 (0.25)	1.37 (0.16)b	3.40 (0.48)	3.36 (0.48)b	2.04 (0.74)	1.95 (0.74)b
ERS vs. ERS plus SDT
Duda et al.17
6 monthsa	130 (17)	127 (16)b	82 (11)	79 (11)b	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported

TABLE 13 - Summary of weight and measures of obesity outcomes in included ERS trials11

BMI, body mass index; N/A, not applicable; PA, physical activity; SD, standard deviation.

Numbers of individuals with complete data.

Between-group difference not statistically significant at p-value ≤ 0.05.

Between-group differences statistically significant at p-value ≤ 0.05.

All randomised participants.

Taylor et al. 26 sum of four skinfolds (mm).

Mean change score.

Study and time of follow-up	Weight (kg)		BMI (kg/m2)		Body fat (%)		Waist-to-hip ratio (cm)
	ERS, mean (SD)	Usual care, mean (SD)	ERS, mean (SD)	Usual care, mean (SD)	ERS, mean (SD)	Usual care, mean (SD)	ERS, mean (SD)	Usual care, mean (SD)
ERS vs. usual care
Taylor et al.12
16 weeksa	Not reported	Not reported	27.5 (0.6)	27.6 (0.6)b	70 (8)	76 (8)c	0.87 (0.08)	0.83 (0.09)b
26 weeksa	Not reported	Not reported	27.3 (1.3)	27.5 (1.1)b	70 (11)	75 (11)c	0.87 (0.08)	0.83 (0.09)b
37 weeksa	Not reported	Not reported	27.5 (1.3)	27.6 (1.1)b	71 (13)	76 (13)c	0.87 (0.08)	0.84 (0.09)b
Isaacs et al.18
10 weeksd	81 (3)	81 (3)b	30.2 (0.8)	30.1 (1.5)b	37.4 (1.9)	37.5 (1.9)b	0.88 (0.06)	0.89 (0)b
6 monthsd	82 (3)	82 (3)b	30.5 (1.1)	30.4 (1.1)b	37.8 (2.4)	37.8 (2.4)b	0.88 (0)	0.88 (0)b
Gusi et al.28
6 monthsa	Not reported	Not reported	29.7 (4.2)	30.6 (4.3)c	Not reported	Not reported	Not reported	Not reported
ERS vs. alternative PA intervention
fSorensen et al.27
4 monthse	–1.1 (4)	–1.1 (4)b	–0.3 (1.3)	–0.04 (1.6)b	Not reported	Not reported	Not reported	Not reported
10 monthse	–0.3 (4.4)	–0.3 (4.4)b	–0.1 (1.9)	–0.6 (2.8)b	Not reported	Not reported	Not reported	Not reported
Isaacs et al.18
10 weekse	81 (3)	81 (3)b	30.2 (0.8)	30.2 (1.6)b	37.4 (1.9)	37.1 (1.9)b	0.88 (0.06)	0.88 (0.06)b
6 monthse	82 (3)	82 (3)b	30.5 (1.1)	30.5 (1.1)b	37.8 (2.4)	37.8 (1.1)b	0.88 (0)	0.88 (0)b
ERS vs. ERS plus SDT
Duda et al.17
6 monthse	Not reported	Not reported	32.8 (6.9)	32.8 (6.4)b	Not reported	Not reported	Not reported	Not reported

TABLE 14 - Summary of respiratory function outcomes in included ERS trials

FEV, forced expiratory volume; FVC, forced vital capacity; PA, physical activity; PEF, peak expiratory flow; SD, standard deviation.

All randomised participants.

Between-group difference not statistically significant at p-value ≤ 0.05.

Study and time of follow-up	FEV : FVC ratio		PEF
	ERS, mean (SD)	Usual care, mean (SD)	ERS, mean (SD)	Usual care, mean (SD)
ERS vs. usual care
Isaacs et al.18
10 weeksa	0.86 (0.0)	0.86 (0.06)b	417 (58)	409 (58)b
6 monthsa	0.86 (0.09)	0.86 (0.09)b	407 (115)	411 (117)b
ERS vs. alternative PA intervention
Isaacs et al.18
10 weeksa	0.86 (0.0)	0.85 (0.06)b	417 (58)	407 (61)b
6 monthsa	0.86 (0.09)	0.85 (0.09)b	407 (115)	416 (117)b

Exercise referral schemes versus usual care

Taylor et al. 12 reported percentage of body fat in ERS participants compared with usual care at follow-up and found no statistically significant difference between the groups. Gusi et al. 28 reported a lower body mass index (BMI), with no other between-group differences in weight and body fat outcomes for the other measured clinical factors (Figures 10 and 11). There was no significant difference in resting blood pressure, serum lipids or respiratory function between ERSs and usual care at follow-up (Figures 12 and 13).

FIGURE 10.

Systolic blood pressure at 6–12 months’ follow-up. 11 df, degrees of freedom; IV, inverse variance; PA, physical activity; SD, standard deviation.

FIGURE 11.

Diastolic blood pressure at 6–12 months’ follow-up. 11 df, degrees of freedom; IV, inverse variance; PA, physical activity; SD, standard deviation.

FIGURE 12.

Body mass index at 6–12 months’ follow-up. 11 df, degrees of freedom; IV, inverse variance; PA, physical activity; SD, standard deviation.

FIGURE 13.

Body fat at 6–12 months’ follow-up. 11 df, degrees of freedom; IV, inverse variance; PA, physical activity; SD, standard deviation.

Exercise referral schemes versus alternative physical activity intervention

In both the studies by Isaacs et al. 18 and Sorensen et al. 27 there were no significant between-group differences at follow-up in resting blood pressure (see Figures 10 and 11), BMI (see Figure 12), body fat outcomes, serum lipids and respiratory function. The Sorensen et al. trial27 reported reduced levels of glycosylated haemoglobin (HbA_1c) in both the ERS group (mean –0.26%, 95% CI –0.79% to 0.27%) and the physical activity counselling group (mean –0.23%, 95% CI –0.47% to 0.02%) at 4 months’ follow-up, although there was no difference between groups.

Exercise referral schemes versus exercise referral scheme plus self-determination theory

Duda et al. 17 reported no significant difference between standard ERS and ERS plus SDT in BMI or resting blood pressure.

Psychological well-being

Four studies17,18,23,25,28 reported psychological well-being outcomes and the results are summarised in Table 15.

TABLE 15 - Summary of psychological well-being data at follow-up in included ERS trials

HADS: Hospital Anxiety and Depression Scale; PA, physical activity; SD, standard deviation.

Significant difference in change from baseline between groups.

All randomised participants.

Between-group differences statistically significant at p-value ≤ 0.05.

Only mean values available.

Numbers of individuals with complete data/questionnaires.

Between-group difference not statistically significant at p-value ≤ 0.05.

Study	Psychological well-being		Anxiety		Depression		Anxiety/depression
	ERS, mean (SD)	Usual care, mean (SD)	ERS, mean (SD)	Usual care, mean (SD)	ERS, mean (SD)	Usual care, mean (SD)	ERS, mean (SD)	Usual care, mean (SD)
ERS vs. usual care
aTaylor and Fox25
16 weeksb	2.31 (0.79)	2.31 (0.67)c	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported
37 weeksb	2.41 (0.79)	2.42 (0.54)c	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported
dIsaacs et al.18
6 monthse	Not reported	Not reported	6.9	7.1f	4.8	4.9f	Not reported	Not reported
Gusi et al.28
6 monthse	Not reported	Not reported	14.1 (9)	22.2 (9.8)c	1.8 (2.3)	2.9 (2.5)c	1.2 (0.4)	1.5 (0.7)c
Murphy et al.23
12 months	Not reported	Not reported	HADS 7.82 (95% CI 7.39 to 8.25) (n = 472)	HADS 8.35 (95% CI 7.92 to 8.77) (n = 502)	HADS 6.14 (95% CI 5.73 to 6.54) (n = 471)	HADS 6.93 (95% CI 6.53 to 7.32) (n = 506)
ERS vs. alternative PA intervention
dIsaacs et al.18
6 monthse	Not reported	Not reported	6.9	7.5f	4.8	5.1f	Not reported	Not reported
ERS vs. ERS plus SDT
Duda et al.17
3 monthsb	Not reported	Not reported	7.7 (4.4)f	8.89 (4.3)	5.9 (4.2)f	6.68 (4.1)	Not reported	Not reported
6 monthsb	Not reported	Not reported	7.9 (4.8)f	8.86 (4.7)	6.1 (4.4)f	6.65 (4.3)	Not reported	Not reported

Exercise referral schemes versus usual care

Taylor and Fox25 reported physical self-perceptions measures, with improvements shown in physical self-worth, and perceptions of physical condition and physical health-collected physical self-perceptions data, and reported significant differences favouring the ERS group compared with usual-care group at 16 and 37 weeks. Isaacs et al. 18 reported no differences between the ERS and usual-care groups in the anxiety and depression scores using the Hospital Anxiety Depression Scale (HADS) at 6 months. In the Gusi et al. study,28 all measures [Geriatric Depression Scale, State Trait Anxiety Inventory and the anxiety/depression subscale of the European Quality of Life-5 Dimensions (EQ-5D)] at 6 months were found to favour ERS participants compared with those receiving the usual care. In the Murphy et al. study,23 in participants who were referred for mental health reason or in combination with CHD, there were significantly lower levels of anxiety (OR –1.56, 95% CI –2.75 to –0.38) and depression (OR –1.39, 95% CI –2.60 to –0.18) but no effect on physical activity. Murphy et al. 23 also found significant interactions with sex for both mental health outcomes, with the beneficial effect of the intervention apparent only among women. There was a suggestion that the intervention was more effective on mental health outcomes among the youngest age group (18–44 years), although this was not statistically significant. Effects did not vary significantly by deprivation status.

Pavey et al. 33 in a review updating his earlier review,11 pooled HADS score data from Murphy et al. 23 and Gusi et al. 28 (Figure 14). This showed a significant reduction in depression (standardised mean difference –0.82, 95% CI –1.28 to –0.35) but not in anxiety (standardised mean difference –4.12, 95% CI –11.52 to 3.28) for ERSs compared with usual care.

FIGURE 14.

Meta-analysis of depression and anxiety in patients, at 6–12 months’ follow-up: fixed-effects model used. (Pavey et al. 33 reproduced with permission, ©2011 BMJ). df, degrees of freedom; IV, inverse variance; SE, standard error.

Exercise referral schemes versus alternative physical activity intervention

Isaacs et al. 18 reported no differences between the ERS and walking programme in anxiety or depression outcomes at the 6-month follow-up.

Exercise referral schemes versus exercise referral scheme plus self-determination theory

Duda et al. 17 reported no difference between groups in anxiety or depression outcomes at either the 3- or 6-month follow-up.

Quantitative results of randomised controlled trial and observational studies

Quantitative analyses of the predictors of uptake and adherence to ERSs have been explored by Pavey et al. 11 It is beyond the scope of this short report to update this element of the review. However, as new data have become available we have added these to the data from the existing Pavey et al. 11 review.

Pavey et al. 11 identified five RCTs and 14 observational studies for inclusion in their review. One additional trial23 and two observational studies34,36 were identified for inclusion in this update, although our search for new observational data was not exhaustive. See Table 21 for a summary of uptake and adherence to ERSs across studies.

Sample sizes ranged across studies from 30 to 6610 participants in the observational studies and from 97 to 2068 in the RCTs. Mean age ranged from 44.9 to 51.9 years across the observational studies and from 53.0 to 59.1 years for the RCTs. Uptake and adherence were described differently in the included studies. Uptake was defined in one of two ways: attendance at the initial consultation with the ‘exercise professional’ or attendance at least one exercise session.

Most studies provided a definition of adherence – completion of a set number of exercise sessions, either numerically (e.g. completed 20 sessions) or a percentage (e.g. > 80% attendance). For four studies,37–40 attendance at a post-ERS consultation was also required to meet the definition of adherence.

The uptake of ERSs across the RCTs ranged from 35% to 100%. Murphy et al. 23 had an uptake of 85%, which was slightly above the pooled result (80%, 95% CI 61% to 98%) reported in the Pavey et al. 11 review. Adherence to the scheme throughout compares favourably with the pooled rate of 37% (95% CI 20% to 54%) across schemes assessed by trials in the review.

Levels of uptake to ERSs ranged from 28% to 81% in the observational studies and adherence ranged from 12% to 70%. Two additional studies36,48 reported adherence levels of 58% and 54%.

Qualitative evaluation of the discussion sections of included studies

In order to explore a further source of data to improve understanding of the factors that will influence the uptake of and adherence to ERSs, we undertook a qualitative analysis of the discussion and conclusion sections of the included studies. Often authors will report their perceived reasons for success and failure of interventions. These may not be reported in the measurable outcomes described within the results sections of published papers and these views can give valuable additional insights that are particularly relevant to the studies included in the review.

As a result of this analysis we created a logic model to summarise and describe the complexity of factors that might impact on referral to ERSs, uptake, adherence and sustained change in levels of physical activity.

Qualitative sibling studies

We searched for qualitative studies that were undertaken as part of a mixed-methods evaluation of an ERS included in this review.

One study23,25,26,32 undertook a mixed-method process evaluation, using structured observation, implementer interviews and routine data to assess the extent to which the ERS was implemented as intended. Semistructured patient interviews explored processes of change and the emergence of social patterning in responses to the scheme. 52

The features of the intervention were as follows: motivational interviewing – initial consultation with an exercise professional (including introduction to leisure-centre facilities and goal-setting), 4-week telephone contact, 16-week consultation and then an 8-month telephone contact and 12-month review, plus discounted access to one-to-one and/or group exercise classes (group and/or individual).

The intervention recruited participants who were sedentary and who had at least one CHD risk factor or suffered a mental health problem (mild anxiety, depression or stress).

The findings of the semistructured patient interviews are summarised.

Population, intervention and comparator

The population, intervention and comparator are the same as those defined in Chapter 4, Comparison of the Anokye adapted to exercise referral scheme model and the Pavey exercise referral scheme model.

Horizon and perspective

The analysis was conducted for a hypothetical cohort of 100,000 patients with a mean age of 50 years receiving either exercise referral or usual care. A lifetime horizon was adopted in order to capture the potential long-term benefits and cost savings associated with physical activity. Mean life expectancy was based on UK interim life tables55 and the analysis did not consider males and females separately. The economic perspective of the model is the NHS and PSS in the UK. Costs and health benefits were discounted at an annual rate of 1.5% as recommended by the Methods for the Development of NICE Public Health Guidance. 56

Model structure

The model structure is described in full elsewhere but we provide a brief description here. 53 The model has a Markov structure and considers a cohort of individuals aged 50 years who present in a physically inactive state and are given a referral to a service designed to increase physical activity that includes a physical activity or exercise programme compared with a control group with no referral to an exercise service. The age of the population was selected to reflect the populations enrolled in the studies providing evidence on the effectiveness of ERS. 11,23 The model estimates the likelihood of becoming physically active and the consequent risk reduction this has on CHD, stroke and type 2 diabetes mellitus.

In the first year of the model individuals enter either an ‘inactive, healthy state’ or an ‘active, healthy’ state; this is considered a ‘run-in’ period where individuals reach a stable level of activity. From year 2 onwards, a proportion of individuals transit from the initial state into one of the following states: event free, CHD, stroke and type 2 diabetes mellitus. Stroke and CHD are then subdivided into non-fatal and fatal. Patients can also die from other causes. Figure 16 shows the model structure.

FIGURE 16.

Model structure from year 2 onwards.

Model parameters

Deterministic results

Table 33 shows the cost-effectiveness results per individual person based on point estimates of parameter values. The deterministic analysis indicates that ERSs are expected to produce a small health gain [0.003 quality-adjusted life-years (QALYs)] at an additional cost of £225 compared against usual care. This resulted in an incremental cost-effectiveness ratio (ICER) of £76,059 per QALY gained.

Probabilistic results

Table 34 shows the PSA results per individual person. The PSA results are very similar to the deterministic results, with an ICER of £76,276.

Figure 17 shows a cost-effectiveness plane of the probabilistic results. Although all of the PSA estimates indicate a positive health gain and increase in costs, there is uncertainty in the magnitude of that cost and QALY gain. However, as can be seen from the cost-effectiveness acceptability curve in Figure 18, the probability that ERSs are cost-effective at a willingness-to-pay threshold of £30,000 per QALY gained is only 0.004.

FIGURE 17.

Cost-effectiveness plane.

FIGURE 18.

Cost-effectiveness acceptability curve.

Comparison of the Anokye adapted to exercise referral scheme model and the Pavey exercise referral scheme model

The aim of this short report was to adapt the Anokye et al. 21 brief advice model into a model comparing ERSs with usual care. The parameters we changed were replacing the RR of brief advice versus usual care with the same RR of ERSs versus usual care used by Pavey et al. ,11 changing the starting age from 33 years to 50 years to reflect the Pavey et al. data and inflating all costs to 2011–12 values. The adapted model appears to generate results that are less favourable to ERSs, even though the effectiveness estimate applied in the updated model is slightly more favourable. We will attempt here to explain why the model results differ in a manner not explained solely by the update of efficacy evidence. We do not have access to the Pavey et al. 11 model and so our explanation of these discrepancies is based on the Pavey et al. health technology report.

Systematic review of exercise referral schemes

This is an update of an existing review,11 and, therefore, this review uses the search results, data extraction tools, data and findings of that review in this update. One additional RCT,31 additional qualitative data49 and two additional observational studies36,48 have been included in this update.

A total of eight RCTs14,17,21,22,28,32–34 met the inclusion criteria and were included in the review. Six RCTs14,17,21,28,32,33 compared ERSs with usual care, which in all cases included some sort of advice regarding physical activity. Two RCTs21,34 compared an alternative physical activity-promoting intervention and one RCT31 compared an alternative form of ERS (i.e. ERS plus SDT intervention) with usual care.

There was considerable heterogeneity in the nature of the exercise/physical activity intervention across studies. Studies recruited predominantly sedentary middle-aged adults who had evidence of at least one lifestyle risk factor, and five of the studies also included individuals with a medical diagnosis (e.g. hypertension, depression). ERSs usually took place at a leisure centre and involved 10–12 weeks of exercise intervention and where there was follow-up it took place at 6 and/or 12 months post randomisation. Studies were judged to have a moderate or low overall risk of bias.

The most commonly reported outcome was self-reported physical activity. When compared with usual care, there was weak evidence of an increase in the number of ERS participants who achieved 90–150 minutes of at least moderate-intensity physical activity per week at 6–12 months’ follow-up (pooled RR 1.12, 95% CI 1.04 to 1.20). There was no difference in physical activity between ERSs versus alternative physical activity promotion interventions or ERSs versus ERSs plus SDT at 6–12 months’ follow-up. We found no evidence to support differences across subgroups (e.g. age, sex) in terms of the impact of ERSs on physical activity. Murphy et al. 32 found that the intervention was effective in increasing levels of physical activity in those referred for a CHD risk factor. There was no consistent evidence for a difference between ERS and any of the comparator groups in the duration of moderate/vigorous intensity and total physical activity, physical fitness, blood pressure, serum lipids, glycaemic control, obesity indices (body weight, BMI, percentage fat), respiratory function, psychological well-being (perception of self-worth, or symptoms of depression or anxiety) or HRQoL.

We found considerable variation across studies in the level of uptake (i.e. attendance at the first induction visit) and adherence to ERSs (i.e. completion of the programme) across the 19 included studies (16 observational studies and six RCTs). Uptake levels were higher, on average, in RCTs than in observational studies, although there was no clear difference in adherence between the two. In bivariate and multivariate analyses, women and older people were more likely to take up ERS. In addition, although older people were also more likely to adhere, women were less likely to adhere than men. Very few studies reported associations between ERS uptake or adherence and participant psychosocial factors or programme-level predictors.

A consistent finding in the quantitative studies reporting adherence and uptake is the association between increasing age and likelihood of uptake and adherence to ERS. This reverses the trend seen in the general population, which is a decrease in physical activity with advancing age. A number of factors have been proposed to explain this including that older people are less time-constrained, are more likely to value the social interaction offered by the group-based approaches and may find it easier to incorporate the scheme exercise activities (such as walking, swimming and cycling) into their everyday life. The only group to show a statistically significant increase in physical activity in the Murphy et al. 23 study were those referred for CHD risk. This group would also have been likely to have been older. Increasing age or CHD risk factors may be confounding factors. It may be that these factors work in a complementary manner, with physical activity perceived to be a means of reducing CHD risk and CHD risk a perceived greater threat to those of advancing age.

One issue that may impact on trial results, as suggested in the qualitative review of discussion sections of included trials, is the possibility that those volunteering to participate in the trial in the first place may have been more motivated to become active or increase their physical activity. This may have contributed to the overall finding that physical activity levels increased in both intervention and comparison groups. It is therefore possible that trial participants (and perhaps also ERS participants in general) may be more intrinsically motivated initially to become physically active and thus may differ from the population as a whole in terms of their response to such a programme; according to SDT,⁷¹ intrinsic motivation is more likely to lead to behaviour change that is maintained over the longer term. Therefore, ERSs may be inadvertently creating inequality in service delivery and failing to reach those most in need of intervention, through the element of self-selection that could creep in in terms of uptake and adherence to the scheme.

In two recent systematic reviews (Orrow et al. 72 and Campbell et al. 70) exploring the effectiveness of advice or counselling delivered by health professionals within primary care to promote physical activity, positive benefits were seen with significant increases in self-reported physical activity at 12 months. In Campbell et al. where interventions were limited to brief advice the RR was 1.30 (CI 95% 1.12 to 1.50), favouring brief advice over usual care. Neither review found any evidence to support ERS over advice or counselling interventions.

This review did find benefit of ERSs over simply giving advice to promote physical activity. In all of the control groups, advice was given on the value of increasing levels of physical activity. This may have had the effect of reducing the apparent effectiveness of ERSs if compared with usual care which might mean no advice to promote levels of physical activity. The quality of advice may not have been as high as that delivered in the brief intervention studies, hence a difference was found between the effectiveness of ERS schemes and advice alone. The presence of bias may also influence the findings in these studies, of which only two22,32 attempted to blind at outcome assessment.

Clinical effectiveness

The strength of this update review is that it was able to use the robust findings of a previous systematic review. 11 The additional data included in this update supported the existing findings.

A rigorous search was carried out to identify any additional RCTs and qualitative studies done alongside the RCTs (sibling studies). The narrow focus of this review in terms of its definition of ERSs meant that similar interventions which may have yielded valuable insights were excluded from the review.

Measuring physical activity by methods of self-report has an obvious risk of self-report bias. There is no gold standard self-reporting measure of adherence to physical activity or physical activity levels. Many traditional instruments have shortcomings from a clinical perspective. Physical activity levels are often scored on scales that are not easily converted into a counselling message. It can also be difficult to assess small but clinically significant changes in physical activity levels in a practice situation. Problems with these instruments underscore the challenge of translating research findings into clinical practice. Self-report levels of physical activity are the only outcomes that show a significant effect. The reliance on outcomes that are subject to recall bias is a weakness of this review.

Cost-effectiveness

The scope for the economic analysis of ERSs for this brief report was to update the Anokye et al. 53 brief advice model, with evidence from the updated systematic review on the effectiveness of ERS, and to update the costs. As such, we did not conduct a systematic review of the existing evidence on the cost-effectiveness of ERSs, which limits our ability to place these results into a broader context. However, we have conducted a detailed comparison with the model used by Pavey et al. 11 to explore why the ICERs differ so much from those previously reported. Several of the assumptions made in our economic evaluation are more conservative than the assumptions made by Pavey et al. 11 In particular, the restriction of benefits to 10 years, which has had a substantial impact on the ICERs, may be considered to be a more reasonable assumption than the lifetime benefits assumed by Pavey et al. 11 (This issue is explored in more detail in Appendix 6.)

The estimate of process utility gain associated with physical activity, which is a particularly important driver of cost-effectiveness within the model, was based on cross-sectional data. 11 It was included in our base case analysis, as this was the approach taken in the economic evaluation of brief interventions to increase physical activity,53 but was actually excluded from the base case analysis in the economic evaluation of ERSs by Pavey et al. 11 In our clinical effectiveness review, some but not all of the RCTs that reported HRQoL found a statistically significant difference between ERSs and usual care, suggesting that this short-term benefit is still relatively uncertain. (This issue is explored in more detail in Appendix 7.)

One of the main limitations of the economic evaluation is that it does not fully explore the potential for cost-effectiveness to vary according to the exact nature of ERSs or the characteristics of the population to whom it is offered. The cost of an ERS is likely to be highly dependent on the exact nature of that scheme. The incremental cost is also dependent on the provision made to those who received usual care, which we assumed to have zero cost. There was substantial variation in both of these factors across the included studies, although our threshold analysis suggested that large changes in the incremental cost of ERS compared with usual care would be needed to bring the ICER under £30,000. (This issue is explored in more detail in Appendix 7.) Although we have provided some estimates of the cost-effectiveness of ERSs for patients who might be referred for ERSs because of a pre-existing health condition, this has been achieved by assuming that some data from the general population, such as the relative risks of being physically active, are transferrable to those patients with pre-existing conditions.

Another key limitation of the model is that it only captures the impact of physical activity on three health conditions and it does not allow for individuals to have multiple conditions. This fails to capture the many aspects of health that may be influenced by physical activity and the complex relationships that exist between different exercise-related conditions, such as the impact of type 2 diabetes mellitus on cardiovascular risk. In addition, the model does not account for the fact that stroke patients are at a higher risk of having recurrent strokes and, thus, the utility loss and additional costs associated with this are not taken into account. The impact of these limitations on the cost-effectiveness of ERSs is difficult to estimate.

Clinical effectiveness of exercise referral schemes

A number of uncertainties remain regarding the clinical effectiveness of ERSs:

• the potential value of different components of the ERS programme on promoting physical activity

• the long-term changes in physical activity behaviour as a result of participating in these schemes

• the extent to which people in the advice-only groups changed their levels of physical activity

• whether or not the small increases in self-reported physical activity are clinically significant or lead to clinical significant differences.

Cost-effectiveness of exercise referral schemes

Good-quality data on the sustainability of activity levels, and the magnitude and sustainability of any associated process utility gains, would lessen the uncertainty in the model.

Database: Ovid MEDLINE(R) In-Process & Other Non-Indexed Citations and Ovid MEDLINE(R) <1946 to present>

Database: Ovid MEDLINE(R) In-Process & Other Non-Indexed Citations and Ovid MEDLINE(R) <1946 to present>