Effects of Circuit Training on Patients with Knee Osteoarthritis: A Systematic Review and Meta-Analysis

The most prevalent joint disease is osteoarthritis (OA), which affects an estimated 240 million individuals worldwide. Knee osteoarthritis (KOA) is one of the top 10 causes of disability worldwide. The aim of this study is to systematically evaluate the effect of circuit training (CT) on patients with KOA. We searched through PubMed, Scopus, ScienceDirect, Cochrane, and Google Scholar up to 12 February 2022. We used random-effects statistical analysis for continuous variables and reported the results as a standardized mean difference (SMD) with 95 percent confidence intervals (CI). Seven trials involving 346 patients were included. A significant improvement in the intervention group was observed for the parameter, pain level (SMD −0.96, 95% CI −1.77 to −0.14; p = 0.02; seven trials, 346 participants; high quality evidence), while no significant improvement was found in physical function (SMD 0.03, 95% CI −0.44–0.50; p = 0.89; five trials, 294 participants; high-quality evidence), quality of life (SMD −0.25, 95% CI −1.18–0.68; p = 0.60; three trials, 205 participants; high-quality evidence), the activity of daily living (SMD 0.81, 95% CI −0.85–2.48; p = 0.34; three trials, 223 participants; high-quality evidence), and knee stiffness (SMD −0.65, 95% CI −1.96–0.66; p = 0.33; two trials, 71 participants; high-quality evidence). The findings in this meta-analysis suggest that CT could effectively complement the conventional treatment of KOA, particularly in alleviating pain. However, comprehensive data on the guidelines for the CT approach would be needed to adequately examine the effects of CT on quality of life and biochemical markers in patients with KOA.


Introduction
Osteoarthritis (OA) is the most common joint disease and the main source of chronic pain and disability [1]. Pain is the predominant symptom of OA [1]. Knee osteoarthritis (KOA) is one of the top ten causes of disability worldwide [2][3][4], accounting for more than 80% of the entire OA disease burden [5]. Clinical OA, as defined by symptoms and physical

Data Extraction
After reading the full manuscripts, two authors (S.B.A. and A.A.I.) performed independent sampling and data extraction from eligible studies. The studies that were included contained substantial data that were extracted. The data includes the name of the first author, population, year of publication, gender, age of patients, number of patients, and method (exercise name, duration, intensity, sets, frequency, intervention timing, study duration, and outcome measures).

Assessment of Risk of Bias
The risk of bias was checked based on random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessors, completeness of outcome data, selectivity of outcome reporting, and other bias, as described in the Cochrane Handbook for Systematic Reviews of Interventions [35].

Measurement of Treatment Effect
To draw forest plots for trials with categorical outcomes, we used relative risk ratios (RR) and 95% confidence intervals (CI), as well as risk differences (RD), estimates and %CI. We intended to analyze continuous data using mean differences (MD) or standardized mean differences (SMD) and 95% of CI where applicable.
The heterogeneity of the studies was determined through two steps. First, we screened the demographics, contexts, treatments, and outcomes to determine whether there was any noticeable variability. Second, we used the I 2 statistic [36] to analyze statistical heterogeneity. We performed a subgroup analysis on the duration of intervention when it was feasible.

Sensitivity Analysis
In the studies that were included, we performed a sensitivity analysis to evaluate how the risk of bias influenced sequence generation and allocation concealment.

Summary of Findings Table
To assess the quality of evidence, we used the GRADEpro technique developed by the Cochrane Collaboration. The GRADEpro system assigns four degrees of quality, the highest being randomized trial evidence. It might be downgraded to moderate, low, or even extremely poor-quality evidence depending on the presence of the following four elements: (i) constraints in study design and implementation; (ii) indirectness of evidence; (iii) unexplained heterogeneity or inconsistency of results; (iv) imprecision of outcomes. The GRADEpro application was used to present the evidence quality for each specific outcome, and the evaluation is being phased in alongside the summary of findings (SoF) table [36].
The SoF table is made up of the following elements: • Key findings that were summarized (participants, comparative, and baseline data, and results) [37]; • Statistical results that have been condensed; • A summary of the evidence's quality, the degree of the effect, and the source of information utilized in the assumed risk.

Included Studies
A total of 402 studies were retrieved from the following databases: PubMed, Scopus, Science Direct, Cochrane, and Google scholar as indicated in Figure 1. After identifying duplicate articles, 256 studies were screened for further selection. After reading the titles and abstracts of the articles, a total of 226 were excluded according to pre-set inclusion and exclusion criteria. The full text of the remaining 30 articles was assessed thereafter, of which  23 were excluded. Finally, data were extracted from the seven trials with 346 participants that met the eligibility criteria [33,[38][39][40][41][42][43].

Included Studies
A total of 402 studies were retrieved from the following databases: PubMed, Scopus, Science Direct, Cochrane, and Google scholar as indicated in Figure 1. After identifying duplicate articles, 256 studies were screened for further selection. After reading the titles and abstracts of the articles, a total of 226 were excluded according to pre-set inclusion and exclusion criteria. The full text of the remaining 30 articles was assessed thereafter, of which 23 were excluded. Finally, data were extracted from the seven trials with 346 participants that met the eligibility criteria [33,[38][39][40][41][42][43].

Participants Characteristics
Three of the seven trials were from high-income countries [41][42][43], and four were from middle-income countries [33,[38][39][40]. Two out of the seven trials recruited their respondents from hospital settings [42,43], while three trials reported enrolling their participants through an informative text or email distributed on social media platforms (Facebook), radio, and newspapers, all of which included interviews with members of the research team [35,43,44]. In one trial, the participants were recruited from the residents of the city of Ribeirao Preto, Sao Paulo, Brazil [39]. Meanwhile, in one trial, information regarding the recruitment of the participants was not provided [38]. Five of the seven trials performed the exercise at healthcare sites [32,[39][40][41][42], while one conducted the exercise at both the healthcare site and the participants' homes [43]. Meanwhile, in one trial,

Participants Characteristics
Three of the seven trials were from high-income countries [41][42][43], and four were from middle-income countries [33,[38][39][40]. Two out of the seven trials recruited their respondents from hospital settings [42,43], while three trials reported enrolling their participants through an informative text or email distributed on social media platforms (Facebook), radio, and newspapers, all of which included interviews with members of the research team [35,43,44]. In one trial, the participants were recruited from the residents of the city of Ribeirao Preto, Sao Paulo, Brazil [39]. Meanwhile, in one trial, information regarding the recruitment of the participants was not provided [38]. Five of the seven trials performed the exercise at healthcare sites [32,[39][40][41][42], while one conducted the exercise at both the healthcare site and the participants' homes [43]. Meanwhile, in one trial, information regarding the exercise site was not stated [38]. In relation to comorbidity, one trial included patients with stable comorbidities such as type 2 diabetes, cardiovascular or respiratory disease, or lower, back, or upper limb pain [42]. Meanwhile, another trial used the Charlson comorbidity index and reported that 45 patients had a single comorbidity, while multi-comorbidities were found in 15 patients [43]. Table 1 describes the characteristics of the included trials.

Intervention Characteristics
Patients in the included studies were randomly assigned to intervention and control groups. In all seven trials, the intervention was CT [33,[38][39][40][41][42][43]. There was a difference in the duration of the intervention among the included studies. In two trials, the intervention was for eight weeks [39,41]. In another trial, the intervention was for one month [38]. In two trials, the intervention was for 14 weeks [33,40]. In one trial, the intervention was for 12 months [42]. In the last trial, the intervention was for 12 weeks [43]. In five trials, the CT was performed at healthcare sites [33,[39][40][41][42], while one trial conducted the exercise at the healthcare site and home-based exercise [43]. Meanwhile, one trial did not state information regarding the exercise site [38].

Risk of Bias in Included Studies
The risk of result assessment bias is depicted in Figures 2 and 3. Figure 2 displays the proportion of studies categorized as low or unclear risk of bias for each risk of bias indicator. Figure

Risk of Bias in Included Studies
The risk of result assessment bias is depicted in Figures 2 and 3. Figure 2 displays the proportion of studies categorized as low or unclear risk of bias for each risk of bias indicator. Figure

Random Sequence and Allocation Concealment
In four trials, the randomization method was described, and the random sequence generation was judged to have a low risk of bias [33,40,41,43]. In the remaining three trials, the randomization method was not explained; thus, we judged random sequence generation to have an unclear risk of bias [41,42,45]. Allocation was concealed in two trials by central randomization and was only revealed after baseline assessment [38,42]. In two trials, the allocation was performed by randomization and balance distribution [33,40]. In the remaining three trials, the allocation numbers were concealed in opaque envelopes prepared by a staff member who was independent of the study [42,44,46]. Thus, we judged allocation concealment to have a low risk of bias (Supplementary Materials).

Blinding of Participants, Personnel, and Outcome Assessment
In three trials, participants were blinded throughout the research procedure, and the trials were deemed to have a low risk of bias [42,44,45]. In two trials, the information regarding the blinding of the participant was not provided and was therefore deemed an unclear risk of bias [38,43]. In two trials, the participants were aware of all exercise procedures and, thus, they were judged to have a high risk of bias [33,40]. The assessors were blinded in six trials and blinding of the outcome assessment was judged to have a low Healthcare 2022, 10, 2041 9 of 17 risk of bias [33,[39][40][41][42][43]. Meanwhile, one trial did not report if the assessors were blinded, and blinding of the outcome was judged to have an unclear risk of bias [38].

Random Sequence and Allocation Concealment
In four trials, the randomization method was described, and the random sequence generation was judged to have a low risk of bias [35,[43][44][45]. In the remaining three trials,

Incomplete Outcome Data
Six trials reported that all participants completed the study, and the bias due to incomplete outcome data was judged as low risk [33,[38][39][40][41]43]. However, in two trials, a total of four participants, two in each of the intervention and control groups, did not complete the post-treatment assessment because they refused to participate [33,40]. Moreover, in one trial, three participants in the intervention group did not complete the post-treatment assessment. Of the three participants, one died, one canceled, and one was no longer interested. In the control group, six participants did not complete the post-treatment assessment. Of these, one died, two were no longer interested, one canceled, one was unhappy with group allocation, and one had personal or health issues [43]. Meanwhile, the intention to treat analysis was applied. One trial reported that five participants in the intervention group did not complete the post-treatment assessment [42]. One had a hip complication, one had knee surgery, one moved away, and one discontinued the post-treatment assessment. However, the trial did not mention why the fifth participant had not completed the assessment, and the bias due to incomplete outcome data was judged as an unclear risk. Meanwhile, intention to treat analysis was applied and the trial measured the primary outcome at 12 months [42].

Other Potential Sources of Bias
We did not detect any other potential source of bias.

Outcomes
The primary outcomes in this review were pain level and quality of life. Seven trials reported pain levels. Out of the seven, three trials measured the pain level post-intervention using the Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) [35,42,45]. In two trials, the pain level was measured using the Visual Analogue Scale (VAS) [38,40]. Another two trials measured the pain level using the Knee Injury and Osteoarthritis Outcome Score (KOOS) [41,43]. Three trials reported the quality of life using KOOS-quality of life [41,43], and WOMAC-quality of life [38]. The secondary outcomes were physical function, the activities of daily living, health-related quality of life, sport and recreation function, anxiety, depression, stiffness, KOA symptoms, triglycerides, and high-density lipoprotein. Five trials reported on physical function using the WOMAC-physical function [39,42] timed up-and-go test [43], a maximum number of knee bends in 30 s [41], and a 40-m walk test [33]. Three trials reported on the activities of daily living post-intervention using KOOS [41,43] and aggregated functional performance time (AFPT) of four common activities of daily living [42]. Two trials reported on health-related quality of life post-intervention using the EQ-VAS from the EQ-5D 5 Dimensional form 3 level version (EQ-5D-3L) [42,43]. Two trials reported sport and recreation activities post-intervention using KOOS-sport and recreation [41,43]. One trial reported on anxiety post-intervention using the hospital anxiety and depression scale (HADS-A) [42]. One trial reported depression post-intervention using the hospital anxiety and depression scale (HADS-D) [42]. Two trials reported on stiffness post-intervention using WOMAC-stiffness [33,39]. Two trials reported on knee symptoms post-intervention using KOOS [41,43]. One trial reported on triglycerides and high-density lipoprotein using serum samples [33].

Knee Pain
There was a significant difference in knee pain outcomes (SMD −0.96, 95% CI −1.77-−0.14; I² statistic = 92%; p = 0.02; seven trials, 346 participants; high quality evidence) [33,[38][39][40][41][42][43] (Figure 4, Table 2) between the CT group and the standard treatment group.    ⊕⊕⊕ Moderate CI: confidence interval, MD: mean difference, SMD: standardized mean difference, RCT: randomized control trials, a : participants were aware of all exercise procedures, b : there is considerable heterogeneity in the studies, c : there is substantial heterogeneity in the studies, d : the included studies recorded a small sample size for both the control and intervention group.

Quality of Life
There was no difference in the quality-of-life outcomes (SMD −0.25, 95% CI −1.18-0.68; I 2 statistic = 90%; p = 0.60; three trials, 205 participants; high-quality evidence) [38,41,43] ( Figure 5, Table 2) between the CT group and standard treatment group. CI: confidence interval, MD: mean difference, SMD: standardized mean difference, RCT: randomized control trials, a : participants were aware of all exercise procedures, b : there is considerable heterogeneity in the studies, c : there is substantial heterogeneity in the studies, d : the included studies recorded a small sample size for both the control and intervention group.

Summary of Main Results
The current review was designed to incorporate all randomized controlled trials evaluating the effectiveness of CT among patients with KOA. The activities of daily living, depression, physical function, quality of life, and knee stiffness were not different between the CT and standard treatment groups. There was no difference in the anxiety, sports recreation, HDL, triglyceride, and knee symptom outcomes between the CT and standard treatment groups for the limited number of trials included. Yet, the pain level, depression, and health-related quality of life significantly differed between the CT and standard treatment groups.

Overall Completeness and Applicability of Evidence
We conducted an extensive and elaborate literature review to evaluate the effectiveness of CT among patients with KOA. The RCT included in this review comprehensively illustrate CT outcomes among OA patients. Seven trials were included in the meta-analysis. We found a significant improvement in the intervention group for pain levels.

Quality of the Evidence
The quality of trial evidence ranged from moderate to very low certainty. In many domains, the risk of bias was uncertain or low for most trials. No evidence of selective reporting bias was found. A lack of adequate random sequence generation in the original study and subsequent review may have contributed to treatment effect bias in the original trial and subsequent review. The risk of performance bias was present in two trials. Performance bias was high in two trials because the participants were aware of all the exercise procedures. Two trials reported that four participants (two in each group) did not complete the post-treatment assessment. One trial stated that three participants in the intervention group did not complete the post-treatment assessment. One trial indicated that five participants in the intervention group did not complete the post-treatment assessment. However, in all the aforementioned trials, intention-to-treat analysis was carried out. The study's random-effects meta-analysis revealed low to moderate heterogeneity. There was no shift in the effect estimate where the random-effects meta-analysis was performed, and although the 95% CI was wider in all cases, the overall quality of evidence contributing to this review, as assessed using the GRADE approach, was moderate to very low.

Potential Biases in the Review Process
We aimed to reduce publication bias by searching different databases without language restrictions and analyzing all relevant papers' reference lists for extra information. We cannot claim to have identified all the studies in this area with absolute certainty. We were unable to create a funnel plot for publication bias relative to each outcome because there were seven studies included. All included papers satisfied all the inclusion criteria, and we did not introduce any bias through the review process; all studies were thoroughly evaluated, and secondary citations were examined. Though all the studies included in this meta-analysis showed a similar trend, we identified substantial heterogeneity in the knee pain, quality of life, and daily living activities outcomes. Due to limited trials, we were unable to explain this in our analysis.

Agreements and Disagreements with Other Studies or Reviews
To the best of our knowledge, this is the first systematic review and meta-analysis carried out to determine the effectiveness of CT among patients with KOA. Three different reviews examined the effects of exercise on patients with KOA [45][46][47]. Xie, Wang [47] evaluated the effects of a rehabilitation program on patients with KOA and showed significant improvement in pain levels. Meanwhile, there was no significant improvement in physical function among the patients who participated in the rehabilitation program compared with conventional rehabilitation. The study included four trials with a total of 791 patients with KOA. Li, Su [45] included 17 trials with a total of 1705 patients and found that resistance exercise relieves pain, alleviates stiffness, and improves physical function in patients with KOA. Hall, Castelein [46] conducted a systematic review and meta-analysis to evaluate the effects of combined diet and exercise and found moderate improvement in physical function and pain in overweight or obese people with KOA.

Implications for Practice
In this meta-analysis, CT was found to have a significant effect in improving pain levels in individuals with KOA. Hence, it could effectively complement the conventional treatment of KOA. However, comprehensive data on the guideline for the CT approach would be needed to adequately examine the effects of CT on quality of life and biochemical markers in patients with KOA.

Implications for Research
If further studies are carried out to investigate the application of CT on patients with KOA, they should comprise a detailed pain-assessing test/outcome and outlined safety information. Data on aerobic exercise or combined aerobic and resistance exercise for patients with KOA and other joint problems should also be collated. If studies are carried out in isolated and under-developed regions or settings with no or little access to standard clinical care, the adjuvant treatment should include a comprehensive designed CT program of tolerable duration to improve KOA.

Supplementary Materials:
The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/healthcare10102041/s1. Table S1: Risk of bias assessment for studies included in Table 1; File S1: Research question and search strategy employed in the study.