https://orcid.org/0000-0003-3022-8893
Abstract
The aim was to determine the comparative benefits of structured high-pain exercise, structured low-pain exercise, and usual-care control, to identify which has the largest effect on walking ability in people with intermittent claudication (IC).
A network meta-analysis was undertaken to assess two outcomes: pain-free walking ability (PFWA) and maximal walking ability (MWA). Nine electronic databases were searched. Trials were included if they were: RCTS; involved adults with IC; had at least two of the following arms—structured low-pain exercise, structured high-pain exercise or usual-care control; and a maximal or pain-free treadmill walking outcome.
Some 14 trials were included; results were pooled using the standardized mean difference (MD). Structured low-pain exercise had a significant large positive effect on MWA (MD 2.23, 95 percent c.i. 1.11 to 3.35) and PFWA (MD 2.26, 1.26 to 3.26) compared with usual-care control. Structured high-pain exercise had a significant large positive effect on MWA (MD 0.95, 0.20 to 1.70) and a moderate positive effect on PFWA (0.77, 0.01 to 1.53) compared with usual-care control. In an analysis of structured low- versus high pain exercise, there was a large positive effect in favour of low-pain exercise on MWA (MD 1.28, −0.07 to 2.62) and PFWA (1.50, 0.24 to 2.75); however, this was significant only for PFWA.
There is strong evidence in support of use of structured high-pain exercise, and some evidence in support of structured low-pain exercise, to improve walking ability in people with IC compared with usual-care control (unstructured exercise advice).
Introduction
Intermittent claudication (IC) is the most common manifestation of peripheral arterial disease (PAD)1, and manifests itself as leg pain during exercise caused by ischaemia secondary to flow-limiting atherosclerosis in the arteries of the lower limbs2. As the severity increases, people with IC become progressively more sedentary, with lower physical activity levels and poorer walking ability3–6. People with IC who have low levels of physical activity have been associated with negative quality of life, depression, and increased risk of all-cause mortality, independent of disease severity and age7,8. Therefore, improving walking ability is viewed as one of the most important outcome measures of intervention to clinicians and patients, measures of which include pain-free and maximum walking distances (or times) obtained during standardised walking assessments1.
Walking ability can be improved by structured exercise (adhering to the frequency, intensity, time, and type (FITT) principle9,10) in people with IC11, in which pain is often prescribed in place of intensity. Furthermore, supervised exercise programmes are more effective than home-based exercise programmes which employ methods of observation12. The National Institute for Health and Care Excellence13 recommends structured exercise programmes which involve walking to maximum claudication pain for 2 h per week, for 12 weeks; these are similar to international guidelines9. Nevertheless, research suggests that improvements in walking ability are achievable across a range of exercise modalities, whether exercising to mild, moderate, or maximal claudication pain in people with PAD and IC14,15, and the benefit of structured low-pain exercise may be overlooked.
Despite the benefits of low-pain exercise interventions, there is little published evidence of the comparative efficacy of low- and high-pain exercise interventions on walking ability in people with IC. When direct comparisons are lacking, a network meta-analysis allows comparison of multiple treatments when studies use a common comparator, such as a usual-care control group. The aim of this study was to perform a systematic review and network meta-analysis of RCTs to determine the comparative effects of structured high-pain exercise, structured low-pain exercise, and usual-care control (unstructured exercise advice only) to identify which has the largest effect on walking ability in people with IC.
Methods
This meta-analysis was conducted and reported in line with PRISMA guidelines16; the checklist is available in the supplementary material. Ethical approval was not required. Methods of the analysis were specified in advance and the protocol registered with PROSPERO (CRD42020172421).
Data sources and searches
Records, without language restriction and published from database inception to 21 January 2021, were identified by an experienced clinical librarian in MEDLINE, Embase, Emcare, the Cochrane Library, Cochrane Central Register of Controlled Trials, PEDro, OpenGrey, ClinicalTrials.gov, and the Cumulative Index to Nursing and Allied Health Literature (CINAHL) bibliographic databases. The MEDLINE, Embase, Emcare, and CINAHL searches were limited to RCTs using validated filters17–19. Search terms for IC and exercise were defined in part using strategies from two systematic reviews14,20 and a Cochrane review12, with an additional cluster for walking assessment. Additional trials were identified by hand-searching bibliographies from included studies, relevant reviews, and meta-analyses. Results from each database were combined using EndNote version X9 for Windows® (Clarivate Analytics™, Philadelphia, Pennsylvania, USA). The MEDLINE search strategy is shown in the supplementary material.
Study selection
Following removal of duplicates, article titles and abstracts were reviewed independently by two authors. Potentially eligible articles were retrieved for full-text review where possible, and any disagreements were resolved by discussion and consensus. Studies were included if they met the following criteria: adult men and women with IC (Fontaine stage 2 or less, or Rutherford 3 or less; or exercise-induced aching/cramping, pain affecting the lower limbs or buttocks, which subsided with rest; or IC confirmed by validated questionnaire); and ankle : brachial pressure index less than 0.9 to 1.00 or over 1.4 at rest, or over 20 per cent decrease after exercise; RCTs with at least two of the following arms—structured exercise which is prescribed at low pain, structured exercise which is prescribed at high pain, or usual-care control (unstructured exercise advice only); studies with completed measures of pain-free or maximal walking time or distance before and after intervention measured using a treadmill test, and reported or retrieved as mean(s.d.) or mean(s.e.m.); and structured exercise adhering to FITT9,10.
For the purpose of this review, levels of pain either achieved or prescribed in the exercise interventions were used to classify studies into low pain (below 50 per cent on a claudication pain scale) or high pain (over 50 per cent on an even scale). Studies that used a scale with an odd number of scale points, the mid-point fell outside of classification, and studies were classified in the same way. For two studies21,22 with multiple publications describing the same cohort, one follow-up study21 was excluded as there was a subgroup of participants from the original study23; another22 was excluded as the follow-up was shorter (6 versus 12 months).
Data extraction and quality assessment
Data extraction was completed independently by two researchers using a customized Microsoft® Excel spreadsheet (Microsoft, Redmond, WA, USA) and checked for agreement. Data extracted for each eligible trial included: bibliographic information (author, publication year); baseline and post-intervention measures of maximal walking distance or time (maximal walking ability, MWA), or pain-free walking distance or time (pain-free walking ability, PFWA) (mean(s.d.) or mean(s.e.m.)); sample characteristics (age, number in each trial arm); and details of interventions (duration of intervention, supervision, treadmill test, claudication pain level and scale, training modality, control group activities, and exercise frequency and volume). The assumptions of transitivity and consistency were assessed comparing Patient/Population, Intervention, Comparator, Outcomes (PICO) across the included studies. If the required data were not reported in the article, the corresponding authors were contacted for further information; if the data were still unavailable, mean(s.d.) or mean(s.e.m.) values were extracted using ImageJ image analysis software by two independent investigators24. To ensure the accuracy of this method, validity was assessed by comparing extracted values from published figures with known true values. Reliability was assessed by analysing each published figure three times, recalibrating the software each time. Inter-rater reliability (IRR) was then assessed using a two-way mixed, absolute agreement, average-measures model to assess the degree of consistency between researchers using the IRR package25 in RKWard, a graphical front end to R (R Foundation for Statistical Computing, Vienna, Austria)26. For the assessment of validity and reliability using ImageJ software, investigators displayed excellent validity and reliability (investigator 1: ICC 1.00, r = 1.00; investigator 2: ICC 0.99, r = 1.00). The investigators also displayed excellent inter-rater reliability (ICC 0.99). It was therefore very likely that only minimal error (if any) was introduced using this method. Risk of bias was assessed by at least two authors using the Cochrane Risk-of-Bias Tool 2.0; any disagreement was resolved by consensus.
Data synthesis and analysis
Data for exercise and usual-care control groups were extracted at baseline (before intervention) and follow-up (after intervention) as mean(s.d.) or mean(s.e.m.) for MWA and PFWA. Alongside the sample size, this information was used to estimate the standard deviation of the within-arm mean difference with correlation coefficients (r) of 0.1, 0.5, and 0.9, as suggested in the Cochrane Handbook for Systematic Reviews of Interventions27. This analytical approach was used because, within each arm, the standard deviations of the difference were not available as they were reported for MWA or PFWA only before and after the intervention. For both MWA and PFWA (either time or distance), the effect of the intervention was estimated as a standardised mean difference (MD), calculated as Hedges g28. Effects were considered trivial if the g value was below 0.2, small if 0.2–0.5, moderate if 0.5–0.8, and large at over 0.8, in relation to the common thresholds applied when interpreting Hedge’s g29. In one study30, two high-pain arms reported MWA and PFWA, and a fixed-effect meta-analysis was used to combine effect sizes across the two arms.
Random-effects pairwise meta-analyses were undertaken, using restricted maximum likelihood and an r value of 0.5, for MWA and PFWA, presented as Hedge’s g with 95 per cent confidence interval and 95 per cent prediction interval (PI). Heterogeneity was quantified using the I2 statistic; I2 values of 25, 50, and 75 per cent are generally considered as indicative of low, moderate, and high heterogeneity respectively31. Despite the small number of studies, funnel plots were produced to assess the risk of a small-study effect, which was tested formally using Egger’s regression test, and possible publication bias.
A random-effects network meta-analysis was then performed with a frequentist approach, based on multivariate random-effects meta-analysis or meta-regression32. Network plots were drawn to determine the network of comparisons, with the thickness of lines between nodes and size of the nodes based on the number of studies in each comparison and treatment respectively. Comparisons across the three interventions are presented as Hedge’s g with 95 per cent confidence interval separately for MWA and PFWA, using within-arm estimates for r of 0.5. Sensitivity analyses using r values of 0.1 and 0.9 were also undertaken and are presented in the same way.
All analyses were conducted using Stata® version 16.1 (StataCorp, College Station, TX, USA).
Results
Study characteristics
Electronic searches yielded 7379 records and an additional 24 articles were identified from manual searches of bibliographies of relevant reviews, meta-analyses, and other article publications (Fig. 1). After removal of duplicates, 1419 titles and abstracts were screened for eligibility, and 82 articles were reviewed in full. A total of 14 studies were included in the analysis, with a combined sample size of 657 patients. The final analysis included the comparison of 9 high-pain arms23,30,33–39, 4 low-pain arms40–43, and 13 usual-care control arms23,30,33–43 for MWA; and 7 high-pain arms23,30,33,35,37–39, 4 low-pain arms41–44, and 11 usual-care control23,30,33,35,37–39,41–44 for PFWA.
PRISMA flow chart showing selection of articles for review
Characteristics of the included studies are presented in Tables S1–S3. Ten of 14 studies had an intervention length of 12–14 weeks, two had 6 months, and two had 12 months. All studies used walking as a training modality and used a treadmill walking test to measure MWA and PFWA. Twelve studies reported use of a progressive treadmill test to measure walking outcomes. Two studies used a single-stage treadmill test. Most studies fully supervised the intervention. Eleven of the interventions implemented an exercise frequency of three times per week with a duration of up to 60 min per session. One study implemented an exercise frequency of five times per week with undisclosed session durations, one study implemented an exercise frequency of twice per week with 30-min sessions, and another implemented an exercise frequency of two to three times per week with 60-min sessions.
Risk of bias
Ten studies23,34,35,37,39–44 were deemed to have a high risk of bias. Only three studies30,33,36 met intention-to-treat principles in line with CONSORT guidelines45. Eleven studies23,34,35,37–44 that had not included participant data for drop-outs, and those lost to follow-up after randomization in the main analysis, were considered per-protocol analyses, and most of these were considered to be at high risk of bias owing to missing outcome data for this reason. Furthermore, there were some concerns of risk of bias because blinding procedures were not stated in nine studies23,33–36,38–40,43, and a high risk of bias owing to assessor non-blinding in one study37. For all studies in this review, there were either some concerns23,30,33–38,40–42,44 or a high risk of bias39,43 regarding selection of the reported result, particularly due to the lack of available prespecified statistical analysis plans. Details of these assessments are provided in Tables S4 and S5.
Pairwise comparisons
In pairwise comparisons, the MDs for structured low-pain exercise versus usual-care control were 2.18 (95 per cent PI −7.47 to 11.83; 95 per cent c.i. 0.16 to 4.19; I2 = 94.8 per cent) across four studies40–43 for MWA (Fig. 2), and 2.21 (−5.87 to 10.29; 0.51 to 3.91; I2 = 93.4 per cent) across four studies41–44 for PFWA (Fig. 3). Corresponding estimates for structured high-pain exercise versus usual-care control were 0.85 (0.58 to 1.11; 0.63 to 1.07; I2 = 0 per cent) across nine studies23,30,33–39 for MWA (Fig. 2), and 0.70 (0.40 to 1.01; 0.47 to 0.94; I2 = 0 per cent) across seven studies23,30,33,35,37–39 for PFWA (Fig. 3).
Pairwise meta-analysis of effect of structured high-pain exercise versus usual-care control and structured low-pain exercise versus usual-care control on maximal walking ability
A random-effects restricted maximum likelihood model was used for meta-analysis. Hedges’ g values are shown with 95 per cent confidence intervals.
Pairwise meta-analysis of effect of structured high-pain exercise versus usual-care control and structured low-pain exercise versus usual-care control on pain-free walking ability
A random-effects restricted maximum likelihood model was used for meta-analysis. Hedges’ g values are shown with 95 per cent confidence intervals.
The investigation for small-study effects suggested no evidence of publication bias for the comparison of high-pain exercise versus control for MWA (P = 0.356) (Fig. S1) and PFWA (P = 0.505) (Fig. S2). Conversely, there was evidence of publication bias for the comparison of low-pain exercise versus control for both MWA (P < 0.001) (Fig. S1) and PFWA (P = 0.044) (Fig. S2).
Network meta-analysis
Treatments were grouped into common nodes based on high-pain exercise, low-pain exercise, and usual-care control. Networks of included trials were connected at C (usual-care control) and A (structured high-pain exercise) or B (structured low-pain exercise) for MWA and PFWA. There were 9 high-pain arms, 5 low-pain arms, and 14 usual-care control arms in total. No studies compared high-pain with low-pain exercise, so quantitative tests of inconsistency were not possible (Figs S3 and S4).
The network meta-analysis of structured low-pain exercise versus usual-care control showed a large positive effect in favour of low-pain exercise on MWA (Hedges g 2.23, 95 per cent c.i. 1.11 to 3.35) (Fig. 4) and PFWA (Hedges g 2.26, 1.26 to 3.26) (Fig. 5). For high-pain exercise versus usual-care control, there was a large positive effect in favour of high-pain exercise on MWA (Hedges g 0.95, 0.20 to 1.70) (Fig. 4), and a moderate positive effect on PFWA (Hedges g 0.77, 0.01 to 1.53) (Fig. 5). For low-pain versus high-pain exercise, there was a large positive effect in favour of low-pain exercise on both MWA (Hedges g 1.28, −0.07 to 2.62) (Fig. 4) and PFWA (Hedges g 1.50, 0.24 to 2.75) (Fig. 5); however, the result was significant only for PFWA.
Summary of network meta-analysis of effect of structured exercise on maximal walking ability
Hedges’ g values are shown with 95 per cent confidence intervals. Effects were considered trivial at g less than 0.2, small at 0.2–0.5, moderate at 0.5–0.8, and large at over 0.8. H, structured high-pain exercise; L, structured low-pain exercise; C, usual-care control.
Summary of network meta-analysis of effect of structured exercise on pain-free walking ability
Hedges’ g values are shown with 95 per cent confidence intervals. Effects were considered trivial at g less than 0.2, small at 0.2–0.5, moderate at 0.5–0.8, and large at over 0.8. H, structured high-pain exercise; L, structured low-pain exercise; C, usual-care control.
Network meta-analysis results were consistent in sensitivity analysis with r values of 0.1 and 0.9 (Figs S5 and S6). For the comparison of structured high-pain exercise versus control, there was a moderate-to-large positive effect in favour of high-pain exercise on MWA and a small-to-large positive effect in favour of high-pain exercise on PFWA. For the comparison of structured low-pain exercise versus control, there was large positive effect in favour of low-pain exercise on MWA and PFWA. Finally, for the comparison of low-pain versus high-pain exercise, there was a large positive effect in favour of low-pain exercise for both outcomes.
Discussion
This systematic review and network meta-analysis aimed to determine the comparative benefits of structured high-pain exercise, structured low-pain exercise, and usual-care control (unstructured exercise advice only) to identify which has the largest effect on walking ability in people with IC. The analysis of RCTs revealed three important findings: there was an overall positive effect of both low- and high-pain structured exercise on MWA and PFWA compared with usual-care control in people with IC; structured low-pain exercise had a larger positive effect than structured high-pain exercise for MWA and PFWA compared with usual-care control; and comparison of structured low-pain with high-pain exercise revealed a large positive effect in favour of low-pain exercise on walking ability in people with IC, although there was a level of uncertainty, with only PFWA reaching statistical significance and wide PIs which crossed the null. With little published evidence on the comparative efficacy of low- and high-pain exercise interventions on walking ability in people with IC, this analysis provides the most robust estimate to date, and adds to a growing body of literature that supports structured low-pain exercise as a non-pharmacological treatment for IC.
The present study adds to contrasting findings in the literature exploring the effect of low- and high-pain exercise on walking ability in people with PAD and IC. An earlier systematic review15 reported that improvements in walking ability and peak oxygen uptake were achievable when exercising to varied levels of claudication pain. By comparison, a recent RCT46 in people with PAD showed that high-intensity exercise (moderate–severe pain) was associated with superior walking ability outcomes than low-intensity exercise (no pain); those in the low-intensity exercise group performed no better than those in the control group, even though participants in the low-intensity group reported greater adherence than those in the high-intensity group. However, it is worth highlighting that less than 20 per cent of the participants in this study experienced classical claudication symptoms and the prescription of pain in the high-pain group would not be classified as a high level of pain according to the pain scale cut-off points used in the present review. In addition, data for treadmill walking distance were not available in the claudication subgroup; therefore, this study was not included in the present meta-analysis.
A recent systematic review20 reported greater adherence to low-pain compared with high-pain exercise interventions in people with IC (overall adherence rate 93.4 (range 80–100) versus 77 (57.1–100) per cent respectively; P = 0.004). Greater programme adherence to low-pain exercise may coincide with greater cumulative exercise volume over the intervention, and thus a greater effect on walking ability compared with structured high-pain exercise and usual-care control. In the present analysis, it was not possible to pool results by adherence as only one low-pain study40 reported this outcome, yet this is an interesting possibility and requires further research. A study by Murrow and colleagues47 compared the effect of traditional walking exercise (walk to pain) and walking exercise that attained a 15 per cent reduction in skeletal muscle oxygenation, which was less painful than pain-guided exercise (range 1–6.5 on a 10-point pain scale), on walking ability and mitochondrial oxidative capacity. Both training programmes improved PFWA and mitochondrial oxidative capacity, which supports the notion that a repeated ischaemic stimulus, however modest, may contribute to improvements in walking ability and mitochondrial oxidative capacity in people with PAD and IC. However, there was a significant interaction in favour of traditional walking (which was more painful) for mitochondrial oxidative capacity, and adherence was similar between groups. This suggests that the most ischaemic stimulus may be better at improving PFWA when adherence is held constant. Nevertheless, only 50 per cent of the randomized cohort completed the trial, suggesting that these results may have been underpowered. If heterogeneity in walking ability outcomes following exercise training is influenced by the levels of ischaemia reached, and adherence to exercise, more research is needed to confirm whether structured low-pain exercise may be a viable alternative to high-pain exercise.
Although the large effect size estimates in the present review suggest that low-pain exercise has a greater effect than high-pain exercise on walking ability, there is a level of uncertainty. Using the network analysis to compare low-pain with high-pain exercise indirectly, the MD crosses the zero threshold in the MWA comparison (Hedges g 1.28, 95 per cent c.i. –0.07 to 2.62) (Fig. 4), although there is a trend towards significance. The overall effect size for structured low-pain exercise is highly influenced by two studies41,42 in particular, which had a low standard deviation and large mean difference, inflating the effect size estimate. Alongside this, there are wide PIs which cross the null for the pairwise comparison of structured low-pain exercise versus control. The estimates are imprecise as only a few small studies could be included in the analysis, and, when calculating PI, there is an assumption that τ2 and the study effects are normally distributed48. Nevertheless, structured low-pain exercise appears to be superior to unstructured exercise advice only. Therefore, the authors encourage further research into the potential use of structured low-pain exercise as a non-pharmacological treatment for IC.
The present review included more high-pain studies, which showed no evidence of heterogeneity. By comparison, high heterogeneity was present among the low-pain studies, likely driven by the small number of such studies included in the analysis. A further limitation is the possible publication bias for low-pain studies; conversely, there was no evidence of publication bias for high-pain studies. In addition, there was a high risk of other biases in most of the included studies, with only three studies adhering to CONSORT guidelines45. This is a weakness of the studies included in this review and in the literature as a whole. The objective cut-off points used to classify high- and low-pain exercise have potential limitations. To address these issues, large head-to-head RCTs that follow CONSORT guidelines are needed, to confirm the efficacy of different structured exercise programmes prescribed using pain on walking ability in people with IC. This would also allow assessment of consistency in the estimates, as the networks were tree-shaped (without loops) for both outcomes in the present analysis, hampering the possibility for this assessment49.
Current exercise guidelines for people with IC suggest exercising up to maximum claudication pain. The present analysis has demonstrated that there is strong evidence in support of structured high-pain exercise, and some evidence in support of structured low-pain exercise, to improve walking ability in people with IC, with both performing better than usual-care control (unstructured exercise advice only). There is a clear lack of studies on structured low-pain exercise, despite some positive effects on walking ability in people with IC; as a result, structured low-pain exercise may have been overlooked in national guidelines.
Funding
This work was supported by the George Davies Charitable Trust (registered charity number 1024818).
Acknowledgements
The authors thank G. Davies and the George Davies Charitable Trust for the generous charitable donation that funded this work. The data underlying this article will be shared on reasonable request to the corresponding author. T.Y, J.S.M.H., and A.T.O.N. receive support from the National Institute for Health Research Leicester Biomedical Research Centre, and T.Y has received investigator grants from Astra Zeneca. J.P., J.S.M.H., and A.T.O.N. are funded, and R.S, part-funded, by a charitable donation from the George Davies Charitable Trust.
Disclosure. The authors declare no other conflict of interest.
Supplementary material
Supplementary material is available at BJS online.
References
