Potential studies were identified by a systematic search of Medline (Ovid) (1950-December 2012), Embase.com (1974-December 2012), Cochrane Central Register of Controlled Trials and CINAHL (1981-December 2012).

The search strategy included a mix of MeSH and free text terms for the search terms chronic renal failure, chronic kidney disease, hemodialysis and exercise training, aerobic exercise, resistance exercise, physical activity, strength training and these were combined with a sensitive search strategy to identify randomized controlled trials. Reference list of papers found were scrutinized for new references. All identified papers were assessed independently by two reviewers (NS and JM) and a consensus reached. Searches of published papers were conducted up until December 3rd 2012.

Randomized, controlled trials of exercise training in adult (>18 years) chronic kidney disease patients receiving dialysis were included. Included studies reported the post-training mean and standard deviation (SD) or standard error (SE) in exercise and control group of at least one primary or secondary outcome measure (listed below). There were no language restrictions. Acute exercise response studies were not included.

Animal studies, review papers and non-randomized controlled trials were excluded. Studies that did not have an exercise intervention, desired outcome measures, participants who were not chronic kidney disease patients or not receiving dialysis in either treatment or control groups were also excluded. Authors were contacted to provide missing data or clarify if information was duplicated in multiple publications from the same authors. At the time of submission no additional data was forthcoming. Incomplete data or data from an already included study resulted in exclusion.

Data relating to chronic kidney disease patient characteristics and exercise training protocols were reviewed. Information was archived in a database and separated into studies that recruited patients undertaking dialysis. Where studies randomized patients to more than two groups only exercise training versus usual care (CKD versus control) data was analyzed. When reviewers suspected data had been duplicated between studies, only study data from the largest patient cohort was analysed.

Outcome measures were peak VO₂, 6 minute walk distance, knee extensor strength, cardiac function (left ventricular ejection %), left ventricular mass index, serum albumin and potassium, inflammatory markers (serum interleukin-6, C-reactive protein), quality of life (SF-36, Beck Depression score), daily energy intake, dialysis adequacy index (Kt/v) and body mass index. Data on exercise adherence and safety (number of adverse events were also collected. Measures of study quality were assessed by Jadad score which examined details of randomization, investigator blinding and study withdrawal [11].

For continuous data, we took the approach of assuming randomization would adjust for baseline imbalance and used end-point data only as advised by the Cochrane Collaboration [12]. A random effects inverse variance was used with the effects measure of mean difference (MD). For dichotomous data Mantel-Haenszel fixed effects odds ratio was calculated. Heterogeneity was quantified using a Cochran Q test [13]. We conducted two sub-group analyses; first we compared home versus outpatient exercise training and second we compared those studies of 6 months or greater versus those less than 6 months. We used a 5% level of significance and 95% confidence interval, figures were produced by Review Manager version 5.0.

Forty-nine randomized controlled trials were identified. Twenty five studies were excluded, leaving 24 full manuscripts. Included studies can be seen in table 1. Assessment of study quality can be seen in table 2. Reasons for study exclusion are seen in table 3.

Twenty four studies yielding data on 879 HD patients, 493 were exercising participants and sedentary 386 controls. Generally exercise and control groups were well matched at baseline for age, gender, EPO use and peak VO₂. Data available from 24 studies of HD patients suggested mean time of receiving haemodialysis was 4.8 (exercise) and 4.5 years (control) and both exercise and control participants demonstrated a mean of two other co-morbid chronic diseases.

Nine studies (one study examined two types of exercise, hence 10 datasets in figure 1) [4,6,7,14-19] measured peak VO₂ in 400 patients. Significantly greater posttraining peak VO₂ values [+5.03 mlO₂·kg⁻¹·min⁻¹ (95% CI 3.73; 6.33, p < 0.00001)] were exhibited following exercise versus control (figure 1). Mean post-intervention peak VO₂ for both exercise and control participants were 23.1 ± 2.6 and 18.1 ± 3.3 mlO₂·kg⁻¹·min⁻¹ respectively.

A further analysis was conducted to examine the effect of

Table 1. Summary characteristics of included studies.

ExT: exercise training; RT: resistance training; AT: aerobic training; CT: combined training; ND: exercise on non-dialysis days; ID: exercise on dialysis days; HRV: heart rate variability; LVEF%: left ventricular ejection fraction; SF-36: short form 36 health questionnaire; BDI: beck depression inventory; Cr: serum creatinine; Hb: haemoglobin; Kt/V: dialysis adequacy index; CRP: C-reactive protein; Chol: total serum cholesterol; LVMI: left ventricular mass index; Kcals: daily energy intake (kilocalories); BP: blood pressure; 6MWD: six minute walk distance.

Table 2. Assessment of study quality—JADAD score.

study duration on change in peak VO₂ in 9 studies. First, we removed the four studies that employed an exercise training program of less than 6 months from the analysis [15-18], mean change in peak VO₂ for the five studies of 6 months for longer was 32.1% ± 8.9%, while the four shorter duration studies yielded a significantly lower peak VO₂ change of 16.2% ± 5.3% (p = 0.002).

Four studies of combined aerobic (AT) and strength (ST) training [6,7,14,18] appear to convey a weighted mean 28.6% ± 11.4% improvement in peak VO₂ compared to five isolated AT studies [4,15,16,17,19] which produced a weighted mean 22.2% ± 9.7% improvement in peak VO₂ (p = 0.37). None of the isolated resistance studies reported peak VO₂. Four studies [6,15,17,18] used interdialytic training (ID) and also reported weighted mean change in peak VO₂ was 17.8% ± 7.7% while the five studies [4,7,14,16,19] using exercise training on nondialysis days showed 30.8% ± 9.6% improvement in peak VO₂ (p = 0.07). One study[20] directly compared exercise training induced improvements in peak VO₂ at 12 months with outpatient (ND) and (ID) training programs 38% versus 31% respectively (p = 0.07). Only two studies [14,21] examined home versus outpatient exercise with conflicting results. The former study [21] showed a greater improvement in six minute walk distance in the home group, while the latter study [20] reported a larger peak VO₂ improvement in the out-patient versus home exercise patients.

Data from three studies [21-23] in 80 haemodialysis patients showed an improvement in six-minute walking

Table 3. Excluded randomized controlled trials.

distance MD 60.7 metres (95% CI 18.9, 102.5 metres, p = 0.004) see figure 2.

Three studies [23-25] reported knee extensor muscle strength in 95 HD patients, improvements were seen in exercising patients versus sedentary controls, MD 2.99 kg (95% CI 0.46, 5.52, p = 0.02), see figure 3.

As EPO use may have had significant impact on exercise training adaptations, particularly peak VO₂, relevant sub-analyses may have been informative, but this was not justified as thirteen of the HD studies did not provide details of patient EPO use during the studies. Three studies did however keep EPO use stable and the remaining eight studies targeted EPO use to haemoglobin levels, but only four of these 11 studies reported peak VO₂ rendering sub-analyses impractical.

Three studies [15,26,27] reported post-training body mass index in 72 HD patients, BMI was not different between exercise and control participants MD 1.15 kg·m⁻² (95% CI - 0.33, 2.63, p = 0.13), see figure 4. Three studies [15,27,28] including 66 patients, employed dietary recall to estimate daily energy intake, exercise participants increased their intake by 4% from baseline. Meta-analysis, showed daily energy intake to be significantly higher in those who exercised versus controls MD

238 Kcal·day⁻¹ (95% CI 94, 383, p = 0.001), see figure 5. Figure 6 shows a forest plot of 6 studies [15,26-30] reporting serum albumin in 126 HD patients. Albumin was significantly lower in exercising versus sedentary controls MD −0.07 g·L⁻¹ (95% CI −0.16, −0.01, p = 0.08). Two studies [26,31] reported post-training serum levels of interleukin-6 (pg/ml) in 49 patients, IL-6 was significantly lower in exercising HD patients versus sedentary controls MD-0.58 pg·ml⁻¹ (95% CI −1.01, −0.15, p = 0.008), see figure 7. Two studies [26,30] reported post-training serum levels of C-reactive protein in 38 patients, CRP was significantly higher in exercising HD patients versus sedentary controls MD 0.92 mg/L⁻¹ (95% CI 0.29, 1.56, p = 0.004), see figure 8.

Following analysis of three studies [6,14,29] in 119 patients, a non-significant trend towards improved left ventricular ejection fraction was observed post-intervention in exercise training patients MD 2.8% (95% CI −1.4, 6.9%, p = 0.20). 

Three studies [4,5,23] reported Beck depression score (BDI) in 86 HD patients, BDI score was significantly lower (improved) in exercising patients versus sedentary controls MD −6.9 (95% CI −9.7, −4.1, p < 0.00001), see figure 9. Neither the physical or mental components of SF-36 general health questionnaire were improved following exercise training.

Figure 10 shows four studies [23,28-30] reported dialysis adequacy index in 54 patients, Kt/V was significantly lower in exercising patients versus sedentary controls MD −0.25 (95% CI −0.34, −0.17, p < 0.00001). Figure 11 shows 6 studies [4,6,7,14,26,32] reporting serum potassium in 253 patients, potassium was higher in exercising