Description of studies

 See Characteristics of included studies; Characteristics of excluded studies; and Characteristics of ongoing studies (in the online version of this Cochrane review).

 

Results of the search

 For this update, a total of 240 potentially new studies were identified by the Cochrane Peripheral Vascular Diseases Group, and the hand search yielded four potential studies for inclusion. After title and abstract evaluation, 48 articles on supervised exercise therapy in participants with PAD remained for full-text analysis. The eight included and 19 excluded (27 in total) clinical trials identified in the initial review of 2006 were re-evaluated.

 

Selection process

 One review author (HJPF) obtained full-text copies of the 75 articles (48 new articles and 27 originally documented in the 2006 re- view) for further assessment. We (HJPF, BLWB, JAWT, MHP) excluded 53 articles of 53 studies (Characteristics of excluded studies) and included 20 articles based on 14 studies (Characteristics of included studies). In addition, we found one ongoing trial (two articles) with potentially usable outcome variables (Frans 2012).

 

Unpublished data

 To avoid publication bias, we contacted six authors of included studies to obtain unpublished data for the assessment of secondary outcome variables. From four authors, we received additional data (Collins 2011; Nicolai 2010; Sanderson 2006; Treat- Jacobson 2009). We were unable to obtain appropriate or adequate data for two studies (Cheetham 2004; Gardner 2011).

 

Included studies

 For this update, nine additional studies were included (Collins 2011; Gardner 2011; Gardner 2012; Hodges 2008; Nicolai 2010;Parr 2009; Sanderson 2006; Stewart 2008; Treat-Jacobson 2009), making a total of 14 included studies. Three of the eight studies (Degischer 2002; Nielsen 1975; Nielsen 1977) that were included in the previous version of this review were excluded in this update for the reasons given in Characteristics of excluded studies. The 14 included trials involved a total of 1002 participants; 589 received supervised exercise therapy (SET), and 413 received non- super- vised exercise therapy. The trials were conducted in the United States (7), the UK (4), Australia (1), the Netherlands (1), and South Africa (1). Nine of the trials were relatively recent (2006 to 2012). Five trials were conducted from 1997 to 2005 and were reported in the original review.

 

Some of the included trials were small; seven included fewer than 30 participants (Hodges 2008; Kakkos 2005; Parr 2009;Regensteiner 1997; Sanderson 2006; Savage 2001; Treat-Jacobson 2009), and four others included fewer than 80 participants (Cheetham 2004; Gardner 2011; Patterson 1997; Stewart 2008). Three trials included more than 140 participants (Collins 2011; Gardner 2012; Nicolai 2010); the largest trial, which was con- ducted in the Netherlands, consisted of 304 participants and was the only multicenter-based study (Nicolai 2010).

 

In three studies (Parr 2009; Sanderson 2006; Treat-Jacobson 2009), three different modes of exercise regimens were investi- gated. For this review, we used data from the supervised walk- ing groups and the regular non-supervised groups only. One trial was designed to investigate the effects of home-based exercise, but participants were treated with a supervised session once a week (Collins 2011). Therefore this study was included in the super- vised group. Another trial (Gardner 2011) investigated the effects of supervised exercise in a three-armed randomized study. In this trial, a control group with "go home and walk" advice was com- pared with a group given home-based or supervised-based exer- cise against a background of standard medical care. We (HJPF, BLWB, JAWT, MHP) decided to use the results of the non-su- pervised exercise groups ("go home and walk" advice and home- based exercise) separately in the subgroup analysis. The results of the supervised group therefore were equally split and were used as comparator.

 

Inclusion criteria

 In six of the trials, in order to be eligible for inclusion, participants with intermittent claudication had to be stable for several months (Cheetham 2004; Kakkos 2005; Patterson 1997; Regensteiner 1997; Sanderson 2006; Stewart 2008). This criterion was not men- tioned in the other trials. In all trials, both males and females were included (Characteristics of included studies), except for one trial, which included only male participants (Regensteiner 1997).

 

The mean age of participants in the SET and control groups was 65.3 years (range 57 to 69) and 66.3 (range 61 to 70), respectively. In the four trials that used age restrictions, participants had to be older than 18 years of age (Treat-Jacobson 2009), older than 40 years of age (Collins 2011), older than 50 years of age (Savage 2001), or between 50 and 75 years of age (Patterson 1997). A diagnosis of PAD was an essential inclusion criterion in all trials. In one trial (Savage 2001), the diagnosis of intermittent claudication was based on a clinical history. In two other tri- als (Kakkos 2005; Parr 2009), participants were included when intermittent claudication was clinically determined and a pe- ripheral arterial lesion was found on ultrasound or angiogra- phy, or both. The remaining trials (Cheetham 2004; Collins 2011; Gardner 2011; Gardner 2012; Hodges 2008; Nicolai 2010; Patterson 1997; Regensteiner 1997; Sanderson 2006; Stewart 2008; Treat-Jacobson 2009) used a low ankle-brachial index (ABI) or a drop in ankle pressure after exercise to determine the clinical diagnosis of PAD. Collins et al included participants with PAD and diabetes mellitus type 1 or 2 (medication use, diet control, or hyperglycemia) (Collins 2011).

 

Exclusion criteria

 The exclusion criteria used by the included trials were variable. In general, ischemic rest pain, comorbid illness with limitations in an exercise therapy program, and recent vascular surgery or percutaneous transluminal angioplasty (PTA) were mentioned. In one study, smoking habits (Savage 2001), and, in another, the pres- ence of diabetes mellitus (Regensteiner 1997) were exclusion criteria. Kakkos 2005 and Nicolai 2010 excluded participants with a maximal claudication distance greater than 300 or with a maximal claudication distance less than 50 meters and greater 500 meters, respectively. Two trials excluded cases in which the results of the treadmill test differed by more than 25% in terms of absolute claudication distance derived from three screening tests (Kakkos 2005; Treat-Jacobson 2009). Collins excluded participants with- out a phone (Collins 2011), and Sanderson 2006 did not include participants who lived farther than 50 km from the research center. Gardner et al excluded participants who used cilostazol or pentoxifylline or both (Gardner 2011; Gardner 2012).

 

Supervised exercise regimens

 In general, the supervised exercise programs consisted of three exercise sessions a week for durations of six weeks (Parr 2009; Sanderson 2006), three months (Gardner 2011; Hodges 2008; Patterson 1997; Regensteiner 1997; Treat- Jacobson 2009), six months (Gardner 2012; Kakkos 2005), and 12 months (Nicolai 2010). However, in three trials (Cheetham 2004; Collins 2011; Stewart 2008), the supervised program was given as a weekly session during six months, a weekly session with phone call sup- port during three months, and biweekly supervised therapy during three months, respectively. Generally, a training session involved walking on a treadmill with varying intensity until moderate or intense pain occurred; this was followed by a short period of rest. Three studies (Cheetham 2004; Parr 2009; Patterson 1997) described an alternative training regimen with walking training as the dominant exercise but with additional exercises for lower limb strengthening. In one trial, treadmill exercises were not performed and exercises were described as based on training of the lower extremity, but unfortunately this was not specified further (Stewart 2008). In all trials, the duration of each supervised exercise session varied between 30 and 70 minutes, except in Patterson 1997, in which a session lasted 20 to 40 minutes. Four trials with an initial supervised period of three months and an additional three months' unsupervised follow-up were included (Patterson 1997; Savage 2001; Stewart 2008; Treat-Jacobson 2009). Cheetham and Kakkos treated participants with six months of SET with an additional six months' unsupervised follow- up (Cheetham 2004; Kakkos 2005). The remaining trials did not have a follow-up period without supervised therapy. In two trials (Patterson 1997; Savage 2001), participants were additionally educated weekly or monthly about PAD, and Treat- Jacobson 2009 described all non-supervised exercises in reports.

 

Non-supervised exercise regimens

In ten trials, participants in the non-supervised exercise therapy groups were advised to walk (Cheetham 2004; Collins 2011 Gardner 2011; Gardner 2012; Hodges 2008; Kakkos 2005; Nicolai 2010; Parr 2009; Sanderson 2006; Stewart 2008). This walking advice was given once and predominantly consisted of walking at least three times weekly with walking intensity rang- ing between tolerance and intense pain. Besides receiving this advice, participants in four trials were treated with standard medical therapy (Cheetham 2004; Hodges 2008; Nicolai 2010; Sanderson 2006). Nicolai et al provided an additional brochure as well (Nicolai 2010). In five other trials, participants in the control group were treated with a home-based exercise program (Gardner 2011; Patterson 1997; Regensteiner 1997; Savage 2001; Treat-Jacobson 2009).

 

Outcome measurements

All trials used a treadmill walking test to measure the pain-free treadmill walking distance or time and the maximal treadmill walking distance or time. The treadmill tests used varied between trials and are reported in the table of Characteristics of included studies. Three trials used a fixed protocol (Cheetham 2004; Kakkos 2005; Stewart 2008), and the other eleven used a graded proto- col. Seven of these studies (Collins 2011; Gardner 2011; Gardner 2012; Hodges 2008; Nicolai 2010; Parr 2009; Savage 2001) used the Gardner Skinner protocol (Gardner 1991). Data on the maximal treadmill walking distance of Cheetham (Cheetham 2004) were calculated on the basis of the P value, and it was assumed that the standard deviations of both groups were equal. Data from Patterson (Patterson 1997) were extracted from the accompanying figures. Seven trials (Cheetham 2004; Gardner 2011; Gardner 2012; Kakkos 2005; Parr 2009; Stewart 2008; Treat-Jacobson 2009) measured adherence to the exercise program by registering absence in the supervised groups. In the non- supervised groups, attendance was noted in three trials on the basis of self-reported logbooks (Cheetham 2004; Gardner 2011; Treat-Jacobson 2009). Participant-reported outcomes were taken into account in nine trials (Cheetham 2004; Collins 2011; Gardner 2011; Gardner 2012; Kakkos 2005; Nicolai 2010; Patterson 1997; Regensteiner 1997; Savage 2001). The Medical Outcomes Study (MOS) Short Form (SF)-36 was used in seven trials (Cheetham 2004; Collins 2011; Gardner 2011; Kakkos 2005; Nicolai 2010; Patterson 1997; Savage 2001); the MOS SF-20 was used in the trial of Regensteiner (Regensteiner 1997). Additionally, the Walking Impairment Questionnaire (WIQ) was used in four trials (Gardner 2011; Gardner 2012; Kakkos 2005; Nicolai 2010), the Intermittent Claudication Questionnaire (ICQ) in one trial (Kakkos 2005), the Geriatric Depression Scale (GDS) in one trial (Collins 2011), and the Charing Cross Claudication Questionnaire (CCCQ) in one trial (Cheetham 2004).

 

Excluded studies

 For this update, 34 additional studies were excluded (Ciuffetti 1994; Collins 2005; Crowther 2008; Dahllo?f 1976; Degischer 2002; Fakhry 2011; Fowler 2002; Gardner 2005; Greenhalgh2008; Grizzo 2011; Hobbs 2007; Krause 1974; Kruidenier 2011; Leicht 2011; Leon 2005; Manfredini 2008; McDermott 2009; Mene?ses 2011; Mika 2005; Mika 2006; Mika 2011; Murphy 2012; Nielsen 1975; Nielsen 1977; Ritti- Dias 2010; Saxton 2011; Schlager 2011; Schlager 2012; Spronk 2009; Tebbutt 2011; Tew 2009; Walker 2000; Wood 2006; Zwierska 2005). A summary of the excluded studies can be found in the table entitled Characteristics of excluded studies. Overall, 53 studies were excluded. Thirty-two studies compared SET with a non-exercising group or a group that maintained usual care with or without usual medical care. Seven trials (Ciuffetti 1994; Gardner 2005; Grizzo 2011; Krause 1974; Lepantalo 1991; Mene?ses 2011; Ritti-Dias 2010) compared SET with another kind of supervised therapy, and six studies compared SET with an invasive procedure (Gelin 2001; Kruidenier 2011; Murphy 2012; Spronk 2009; Taft 2001; Taft 2004). Four trials (Fowler 2002; Leon 2005; Manfredini 2008; Tebbutt 2011) compared home-based exercise therapy with walk advice. In another study (Tisi 1997), participants were treated with a 4-week SET program. Five trials (Degischer 2002; Fakhry 2011; Nielsen 1975; Nielsen 1977; Wind 2007) were excluded because they used a non-randomized study design. Three of these had been classified as included studies in the original review (Degischer 2002; Nielsen 1975; Nielsen 1977). However, because a greater number of eligible studies were available, we decided to include RCT designs only in the current update.

 

Ongoing studies

 One trial (Frans 2012) was a description of a study protocol.

 

Risk of bias in included studies

 See Figure 1; Figure 2.

 

Figure 1. Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

 

Figure 2. Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

 

Allocation

Sequence generation

 Four studies did not detail the process of sequence generation (Patterson 1997; Regensteiner 1997; Savage 2001; Treat-Jacobson 2009). All others (Cheetham 2004; Collins 2011; Gardner 2011; Gardner 2012; Hodges 2008; Kakkos 2005; Nicolai 2010; Parr 2009; Sanderson 2006; Stewart 2008) described an adequate sequence generation.

 

Concealment of allocation

 Five studies described adequate concealment of allocation, using sealed opaque envelopes or a computer program with or with- out block randomization (Gardner 2011; Gardner 2012; Nicolai 2010; Sanderson 2006; Stewart 2008). Allocation concealment was not detailed in the remaining nine studies (Cheetham 2004; Collins 2011; Hodges 2008; Kakkos 2005; Parr 2009; Patterson 1997; Regensteiner 1997; Savage 2001; Treat-Jacobson 2009).

 

Blinding

 In trials that compared exercise programs with walking distance as the primary outcome, blinding of staff and participants is not possible and therefore was not considered. Two trials were judged to have a low risk of detection bias because a blinded analysis of the outcomes was performed (Cheetham 2004; Nicolai 2010).

 

Incomplete outcome data

 All included studies reported results of predefined primary out- come variables. Nicolai et al reported additional outcome variables, which were not published (Nicolai 2010). However, through contact with the author, these unpublished data were obtained. Five trials reported an intention-to-treat analysis (Collins 2011; Gardner 2011; Gardner 2012; Kakkos 2005; Nicolai 2010), and six other trials reported no dropouts or had minimal loss to fol- low-up and therefore did not perform an intention- to-treat analysis (Cheetham 2004; Regensteiner 1997; Sanderson 2006; Savage 2001; Stewart 2008; Treat-Jacobson 2009). Overall, 159 participants were lost to follow-up (15.9%). Two trials were unclear in their attrition bias description (Parr 2009; Patterson 1997), and Hodges et al reported no details on attrition (Hodges 2008).

 

Selective reporting

 Reporting bias was difficult to determine. No study protocols of trials were obtained. All included studies reported commonly used primary outcome parameters, so a low reporting bias was assumed for these outcomes. In one trial, we assessed a high risk of reporting bias (Cheetham 2004). In this study, quality of life was measured with Medical Outcomes Study Short Form 36; however, we could not obtain usable data from the figures. In another trial, reporting bias was just as unclear because quality of life was reported in a "Physical function score" only (Gardner 2011). Only one author replied with unpublished SF-36 data (Nicolai 2010).

 

Other potential sources of bias

 To detect publication bias, we analyzed the two best represented outcome variables (maximal walking distance at three and six months) with the use of funnel plots (Figure 3; Figure 4). No asymmetrical plots were observed, leading to the conclusion that publication bias was minimal. Seven trials were judged to have an unclear risk of potential bias. The work of Collins 2011 was funded by the American Diabetes Association; however, no potential con- flicts of interest relevant to this article were reported. Two studies by Gardner et al (Gardner 2011; Gardner 2012) could be influ- enced by participation bias. Participants in these trials were vol- unteers; therefore, they may represent those more interested in ex- ercise. Four other trials reported data in interquartile range (IQR) format, which had to be recalculated; this could have led to po- tential bias (Kakkos 2005; Nicolai 2010; Patterson 1997; Stewart 2008). All other trials were judged to have low risk (Cheetham 2004; Hodges 2008; Parr 2009; Regensteiner 1997; Sanderson 2006; Savage 2001; Treat- Jacobson 2009).

 

 

Figure 3. Funnel plot of comparison: 1 Walking distances: Supervised exercise therapy versus non- supervised exercise therapy, outcome: 1.2 Maximal treadmill walking distance after 3 months.

 

Figure 4. Funnel plot of comparison: 1 Walking distances: Supervised exercise therapy versus non- supervised exercise therapy, outcome: 1.3 Maximal treadmill walking distance after 6 months.

 

Effects of interventions

Primary outcome

Maximal treadmill walking distance or time

 Data on maximal treadmill walking distance after six weeks of exercise therapy were available in five trials (Gardner 2012; Hodges 2008; Kakkos 2005; Parr 2009; Sanderson 2006) with a total sample size of 234 participants. This outcome was repeated after three months in nine trials (n = 592) (Cheetham 2004; Gardner 2011; Hodges 2008; Nicolai 2010; Patterson 1997; Regensteiner 1997; Savage 2001; Stewart 2008; Treat-Jacobson 2009), after six months (n = 686) in nine trials (Cheetham 2004; Collins 2011; Gardner 2012; Kakkos 2005; Nicolai 2010; Patterson 1997; Savage 2001; Stewart 2008; Treat-Jacobson 2009), after nine months (n = 308) in two trials (Cheetham 2004; Nicolai 2010), and after 12 months (n = 321) in three trials (Cheetham 2004; Kakkos 2005; Nicolai 2010). We calculated the effect estimates after standardization of mean differences. At six weeks, the maximal treadmill walking distance was increased with an overall effect size (ES) of 0.52 (95% confidence interval (CI) 0.24 to 0.81) when the fixed-effect model was used in favor of the SET group. At three months, the overall effect size increased to 0.69 (95% CI 0.51 to 0.86). This translates to a difference in favor of the supervised group of approximately 180 m in increased walking distance, corresponding with an approximately 35% to 40% difference in increase between the groups. At six months, this effect was maintained with an overall effect size of 0.48 (95% CI 0.32 to 0.64). In two trials in which the outcomes were repeated, the effect was present at nine months (ES 0.60, 95% CI 0.36 to 0.85) (Cheetham 2004; Nicolai 2010) and 12 months after exercise therapy (ES 0.61, 95% CI 0.37 to 0.84) (three trials: Cheetham 2004; Kakkos 2005; Nicolai 2010). We performed a subgroup analysis at three months and six months of follow-up by splitting the non-supervised group into a walking advice group and a home-based exercise group. At three months, the difference favored the supervised group, with an ES of 0.76 (95% CI 0.56 to 0.96) and 0.50 (95% CI 0.17 to 0.83) for the walking advice group and the home-based group, respectively. No significant difference was found between the two subgroups (P = 0.19). At six months, the supervised exercise group was still in favor compared with both non-supervised groups. In addition, we found no significant differences between the two subgroups (P = 0.78). At six weeks, non-significant heterogeneity was noted with an I2 of 27% (P = 0.24). At three months, moderate heterogeneity was observed (I2 = 43%; P = 0.07). Use of the random-effects model for the results after three months resulted in a similar overall effect size of 0.69 (95% CI 0.43 to 0.94), indicating that this heterogeneity is of non-importance. At six months, significant heterogeneity was seen (I2 = 71%; P = 0.0006). With a random- effects model, the effect size was increased from 0.48 to 0.58 (95% CI 0.24 to 0.91). This heterogeneity is likely due to the negative trial of Collins (Collins 2011). When a sensitivity analysis was performed by excluding this trial, a non-important heterogeneity was achieved (I2 = 25%; P = 0.23). At other time points, moderate heterogeneity was noted at nine months (I2 = 82%; P = 0.02) and at 12 months (I2 = 64%; P = 0.06).

 

Secondary outcomes

Pain-free treadmill walking distance or time

 Data on pain-free treadmill walking distance after six weeks were available in four trials (Gardner 2012; Kakkos 2005; Parr 2009; Sanderson 2006) with a total sample size of 171. This outcome was repeated after three months in seven trials (n = 479) (Gardner 2011; Nicolai 2010; Patterson 1997; Regensteiner 1997; Savage 2001; Stewart 2008; Treat-Jacobson 2009), at six months in eight trials (n = 630) (Collins 2011; Gardner 2011; Kakkos 2005;Nicolai 2010; Patterson 1997; Savage 2001; Stewart 2008; Treat- Jacobson 2009), and at 12 months after exercise therapy in two trials (Kakkos 2005; Nicolai 2010) (n = 266). We calculated the effect estimates after standardization of the mean differences. At six weeks, the pain-free treadmill walking distance was increased, with an overall effect size of 0.51 (95% CI 0.18 to 0.84), when the fixed-effect model was used in favor of the SET group. At three months, the overall effect size increased to 0.70 (95% CI 0.52 to 0.89). This translates to a difference in favor of the supervised group of approximately 150 m in increased pain- free walking distance. At six months and twelve months, this effect was maintained, with an overall effect size of 0.52 (95% CI 0.35 to 0.69) and 0.50 (95% CI 0.24 to 0.76), respectively.We again performed a subgroup analysis in the non-supervised group at three months and six months of follow-up. At three months, the difference favored the supervised group, with an ES of 0.72 (95% CI 0.49 to 0.94) and 0.65 (95% CI 0.29 to 1.01) for the walking advice group and the home-based group, respectively. No significant differences were found between the two subgroups (P = 0.76). At six months of follow-up, the SET group was still in favor, with an ES of 0.45 (95% CI 0.27 to 0.63) and 1.04 (95% CI 0.54 to 1.53) compared with the walk advice group and the home-based exercise group, respectively. However, a significant difference (P = 0.03) between the non-supervised subgroups was found in favor of the home-based group. No statistical heterogeneity was observed at six weeks (I2 = 0%; P = 0.45), at three months (I2 = 0%; P = 0.56), and at 12 months (I2 = 0%; P = 0.54). However, at six months, substantial heterogeneity was reported (I2 = 64%; P = 0.007). After a sensitivity analysis that excluded the negative trial of Collins (Collins 2011), we found an unimportant heterogeneity (I2 = 16%; P = 0.30).

 

Mortality

 In five of the fourteen trials (Cheetham 2004; Kakkos 2005; Nicolai 2010; Patterson 1997; Stewart 2008), a total of thirteen participants died during the course of the study. In the trial of Patterson 1997, nothing is mentioned about the cause of death of the two participants who died during follow-up (both from the supervised group). Cheetham 2004 registered two deaths (one from each group), neither of which could be attributed to a vascular event. Kakkos (Kakkos 2005) described one death. This person died as the result of acute leg ischemia after undergoing an operation for bladder cancer. Nicolai et al mentioned three deceased people in the control group (coronary heart disease (n = 2), renal cell carcinoma (n = 1)) and four deaths in the supervised group (complication lower extremity bypass surgery (n = 1), lung carcinoma (n = 1), ruptured abdominal aortic aneurysm (n = 1), pancreatic cancer (n = 1)) (Nicolai 2010). One death occurred in the control group of Stewart (Stewart 2008). No deaths were reported in the remaining trials.

 

Adherence to exercise program

 Five trials (Cheetham 2004; Gardner 2011; Gardner 2012; Parr 2009; Stewart 2008) studied adherence to exercise training. Cheetham 2004 simply asked participants whether they walked "less than three times", "three times", or "more than three times" a week. Twice as many participants in the SET group as in the non- supervised group claimed to be walking more than three times a week. Gardner 2011 provided participants with a step activity monitor, and they were instructed to wear it for each exercise session. Additionally, participants received an exercise logbook in which they were to record their walking sessions. Adherence to home- based exercise and supervised exercise was similar (82.5% vs 84.8% completion of exercise sessions, respectively). Unfortunately, no adherence data were available from the group that re- ceived walking advice at baseline only. The adherence ratio from the exercise group in Gardner 2012 did not differ greatly (74% completion) from that obtained in their earlier research, although a progressive decline in attendance was evident from the first two months to the final stage at six months (from 86% to 63%). Kakkos 2005, Parr 2009, and Stewart 2008 noted the attendance of the supervised group only (mean attendance: 60%, 80%, and 89%, respectively). Treat-Jacobson 2009 revealed an adherence of 100% in 61% of participants in the SET group, and 97% of participants completed at least 75% of the prescribed training. Conversely, 75% of control group participants reported participation in outside exercise at least three times a week.

 

Patient-reported outcomes (quality of life)

 In this update, we calculated the effect estimates after standardization of mean differences in SF-36 outcomes after three and six months. The SF-36 was used in seven trials after three months (Cheetham 2004; Collins 2011; Gardner 2011; Kakkos 2005; Nicolai 2010; Patterson 1997; Savage 2001), and the MOS SF-20 in the trial of Regensteiner (Regensteiner 1997). However, Collins 2011 and Cheetham 2004 published SF-36 results as a change from baseline, but they were not usable. We contacted the study authors but were not able to obtain raw or unpublished data. Results of the physical function role were presented in five trials (Gardner 2011; Nicolai 2010; Patterson 1997; Regensteiner 1997; Savage 2001); both social function and the emotional role were noted in four trials (Nicolai 2010; Patterson 1997; Regensteiner 1997; Savage 2001); and the remaining roles were mentioned in three trials (Nicolai 2010; Patterson 1997; Savage 2001). The role "general health" showed significant improvement with supervised exercise (ES 0.30, 95% CI 0.05 to 0.55) only. The other roles showed a positive trend toward the intervention; however, these changes are not statistically significant. A sub analysis of non-supervised exercise programs could not be made because of the small number of studies. SF-36 outcomes at six months were obtained in four trials (Kakkos 2005; Nicolai 2010; Patterson 1997; Savage 2001). Again, almost all roles showed a positive trend toward supervised exercise, except for the role of "pain", which revealed a significant positive change (ES 0.25, 95% CI 0.00 to 0.49). SF-36 outcomes at other time periods were not used because they could be obtained only from unpublished data of Nicolai 2010 and summary data of Kakkos 2005. No statistical heterogeneity was noted in all quality of life analyses, except for the vitality role, at three months' follow-up (I2 = 79%; P = 0.008). Use of random-effects models resulted in a somewhat increased overall effect size from 0.05 to 0.30 (95% CI -0.43 to 1.03), indicating that no important heterogeneity was present. Using the CCCQ, Cheetham 2004 reported a statistically significant difference between the two groups (P < 0.05) after nine months of exercise in favor of the SET group. Intragroup analy- ses of the WIQ and the ICQ showed less improvement or even reduction in the non-supervised group compared with an in- crease in score in the supervised group (Kakkos 2005; Regensteiner 1997). However, these changes were not compared between groups (Kakkos 2005; Regensteiner 1997).