Summary of main results

Primary outcome

 We included data from 14 randomized trials enrolling a total of 1002 participants. Statistically significant and clinically relevant differences in improvement in walking distance were consistently demonstrated in favor of SET compared with non-supervised exercise therapy regimens during the first year of treatment. A sub- analysis was performed to investigate the influence of two types of non-supervised exercise regimens. No significant differences were found between "go home and walk" advice and a more structural home-based exercise program. Despite the variability of prescribed exercise training and treadmill walking tests used to measure the outcome, heterogeneity was present only for the results of the maximal treadmill walking distance after six months. This is likely a result of the negative trial of Collins (Collins 2011). Indeed removing this trial from the analysis resulted in the absence of statistical heterogeneity. In addition to chance alone, several factors could contribute to the observed heterogeneity. First, the supervised exercise regimen used by Collins was described as home-based exercise and consisted of once-a-week supervised training only, thereby reducing the intensity of supervision compared with the other trials, which trained three times a week or biweekly (Stewart 2008). Furthermore, in the trial of Collins (Collins 2011), all participants had diabetes mellitus. Reported data on walking time and distance were standardized to allow calculation of the difference in increase between the two treatment groups. When the standardized data were translated back to walking distances, summary estimates of maximal tread- mill walking distance showed a difference in increase of approximately 180 m in favor of the SET regimens over non-supervised therapy for three months. This difference was maintained at six, nine, and twelve months.

 

Secondary outcomes

 In line with maximal walking distance, pain-free walking distance was more increased by supervised exercise than non-supervised exercise during the first year of treatment, with a maximal effect size of 0.76 (95% CI 0.57 to 0.95) at three months. This effect was maintained at six and twelve months. A significant difference was found in the sub analysis between "go home and walk" advice and a more structured home-based exercise program, probably as a result of the SET negative trial of Collins (Collins 2011). Furthermore, significant heterogeneity was found at six months. A sensitivity analysis excluding the trial of Collins (Collins 2011) resulted in non-important heterogeneity and no significant difference between the two non-supervised exercise types. No obvious results in mortality registration were obtained. Thirteen of the 1002 participants died, but none of these deaths were related to exercise therapy. Participant-related outcomes (SF-36) were not significantly different between the two exercise regimens. This could be explained by an underpowered analysis. We obtained complete SF-36 data from three studies (n = 245) at three months of follow- up (Nicolai 2010; Patterson 1997; Savage 2001) and from four trials at six months of follow-up (n = 258) (Kakkos 2005; Nicolai 2010; Parr 2009; Savage 2001). Five other studies did record SF- 36 outcomes; however, we were not able to obtain the raw data. This could have led to a potential publication bias.

Overall completeness and applicability of evidence

Participants

 Both inclusion and exclusion criteria were variable across stud- ies, but they did not differ very widely from daily practice. Regensteiner 1997 included solely male participants, and Collins 2011 included participants with diabetes mellitus only.

 

Interventions

 Most included studies were performed in hospital settings all over the world. One trial provided supervised exercise in part at the hospital and in part at home (Collins 2011). Another used com- munity-based supervised exercise (Nicolai 2010). Kruidenier concluded that community-based SET appeared to be as effective as hospital-based SET, so it seems unlikely that these potential factors of heterogeneity limit the applicability of the results (Kruidenier 2009). In three trials, training sessions were complemented by strength training (Cheetham 2004; Parr 2009; Patterson 1997), and one trial was based on exercise therapy of the lower extremity without treadmill walking (Stewart 2008). There were no indications that these supervised sessions were less stringent than in the other studies. The effectiveness of different modes of exercise will be compared in another Cochrane review (Lauret 2012b). Six trials had a follow-up period without supervision. After a sensitivity analysis was performed by removing four trials with a non-super- vised follow-up period (n = 350; Patterson 1997; Savage 2001; Stewart 2008; Treat-Jacobson 2009), effect size at six months did decrease slightly from 0.48 (95% CI 0.32 to 0.64) to 0.45 (95% CI 0.27 to 0.63). This suggests that exercise therapy with a period of supervision could have a prolonged effect on increasing walking distance and could be as effective as that described in trials with a full follow-up trajectory.

 

Comparator

 This review compared supervised with non-supervised exercise therapy. We found that non-supervised regimens showed a lot of heterogeneity from simple walking advice at the start of treatment to a structured home-based exercise program with some form of re- mote "supervision". Therefore, we performed a sub analysis within the non- supervised exercise group, which was divided into a "go home and walk" advice group and a home-based exercise group. In one three-armed trial (Gardner 2011), the results of the control group and those of the home-based treated exercise group were compared with results of the supervised group by splitting the supervised group into two groups.

 

Outcomes

 All trials used treadmill exercise to investigate the effectiveness of supervision. Only three trials used a fixed protocol. Furthermore, we used standardized mean differences to reduce potential het- erogeneity caused by differences in treadmill testing. To reduce heterogeneity in secondary outcomes, we used SF-36 values only for participant-reported outcomes.

 

Applicability

 To determine the clinical relevance of the ability to walk 180 m farther, one should realize that a mean maximal walking distance at baseline is approximately 300 m with an even shorter pain-free walking distance of approximately 200 m. Regular exercise, such as walking, has been shown to significantly reduce cardiovascular events (Izquierdo 2000; Wannamathee 2001). Because claudicants have high cardiovascular morbidity and mortality (30% mortality at five years) (Aquino 2001; Dormandy 1999; Hooi 2002), a habit of regular walking with increased exercise capacity could be of considerable importance. Results of our review reveal increased ability to perform on a treadmill. However, such an increase might not have any influence on walking behavior in daily life because some studies have found that supervised exercise does not affect day-to-day walking behavior (Crowther 2008; McDermott 2009).

 

Quality of the evidence

 Unfortunately, the raw data from one study (Cheetham 2004) were not directly available and had to be derived by working backward from the reported P values. We assumed that the standard deviations of both groups were equal. Compared with other trials that reported equal standard deviations, this seemed reasonable. The data of Patterson (Patterson 1997) were extracted from the accompanying figures. Previously included non-randomized studies were excluded (Degischer 2002; Nielsen 1975; Nielsen 1977). Other detailed information of quality is presented in the Assessment of risk of bias in included studies. We noted two important limitations in the quality of the evidence. The nature of the interventions made participant blinding effectively impossible. In addition, participation bias may have influenced the results, in that enrollment in a study motivated participants to walk (Collins 2011).

 

Potential biases in the review process

 Unfortunately, heterogeneity was noted in the non-supervised comparator group. We tried to minimize this by introducing a subgroup analysis. Furthermore, because of differences in inclusion and exclusion criteria, such as age or gender restrictions or the presence of diabetes mellitus and smoking behavior, generalization of the results of this meta-analysis could be a topic of discussion. Not all outcome data were available for inclusion. This potentially introduced bias into our review and may reflect publication bias. Therefore, we conducted a formal analysis of publication bias by using a funnel plot. It is suggested from this analysis that publication bias is not a matter of importance (Figure 3; Figure 4).

 

Agreements and disagreements with other studies or reviews

 Several other systematic reviews have been performed that are in line with our findings. Watson et al suggested that exercise therapy should play an important part in the care of selected patients with intermittent claudication, to improve walking times and distances. Effects were demonstrated after three months of supervised exercise, although some programs lasted longer than one year (Watson 2008). Wind et al concluded that SET increases pain-free and absolute walking distances in participants with intermittent claudication as well (Wind 2007). A recent review of Fakhry et al. compared supervised exercise with no interventional observation. The authors concluded that supervised exercise yielded significant clinical benefit (Fakhry 2012).

 

Several plausible mechanisms might explain the beneficial results of SET over non-supervised regimens. A first explanation could be the fact that the exercise regimen of the SET group mainly consisted of treadmill walking, which involves a higher workload than level ground walking at "normal" speed as used by the non-super- vised group (Degischer 2002). It is very difficult to measure the intensity of training, but it is generally assumed that home train- ing cannot be considered to be performed with the same energy as training under supervision (Nielsen 1975). Second, a higher workload will lead to a larger positive effect of the general physical condition of the patient, possibly as the result of increased cardiovascular stress, providing a better stimulus for exercise-induced adaptations (Hamburg 2011). Furthermore, a supervised program could offer additional encouragement and motivation to patients resulting in a higher adherence rate, which can be explained in part by the Hawthorne effect as mentioned by Wind et al (Wind 2007). The Hawthorne effect describes the effect that the awareness of being under observation can alter the way a patient behaves or can positively influence the outcome. Nevertheless, when adherence is taken into account, and on the basis of the contrasting results between two trials (Cheetham 2004; Gardner 2011), nothing can be suggested about the influence of a supervised exercise regimen on adherence. However, adherence in a supervised exercise setting is effortlessly measurable during the session in contrast to adherence during home-based intervals or in home-based settings.