The studies were eligible if: (1) the research design was an RCT; (2) the participants were undergoing TKA for knee OA; (3) preoperative exercise intervention or postoperative exercise intervention was performed; (4) the researchers assessed the participants’ body function and/or activity using parameters such as pain, strength, ROM, QOL, balance, and gait speed; and (5) the paper was published in English. Eligibility criteria for the control group were not set. Regarding the selection of each article, the choice of the two researchers were independent. According to the international classification of function, disability, and health proposed by the World Health Organization,8 we defined body function as physiological functions of body systems (including psychological functions) and activity as the execution of a task or action by an individual.

We used the following search terms to search all trial registers and databases: “arthroplasty, replacement, knee,” “osteoarthritis, knee,” “exercise,” and “exercise therapy.” The search strategy consisted of a combination of free text words and medical subject heading terms. The search strategy is shown in Table 1. All studies published until July 2017 were included in the search. The terms “population” and “intervention” were combined with the word “AND” as an operator. Population was defined as participants with OA of the knee on a waiting list for TKA or who had undergone TKA. This RCT intended to achieve the most valid information regarding intervention effectiveness. For each concept, synonyms and Medical Subject Headings terms were combined with the “OR” operator.

The PubMed, Cochrane Central Register of Controlled Trials, Physiotherapy Evidence Database (PEDro), and Cumulative Index to Nursing & Allied Health databases were searched.

Two reviewers independently screened the titles and abstracts using the predetermined eligibility criteria. Disagreements were resolved by discussion. Full-text copies of articles that were not definitively excluded based on the title and/or abstract were retrieved, and the criteria were reapplied. Uncertain cases were discussed by the reviewers to achieve a consensus.

Predesigned spreadsheets were used to extract data regarding the participants, interventions, outcome measurements, and results.

The database search was supplemented by a manual search of the reference lists of past systematic reviews.

Two researchers independently applied a validated scale (PEDro) to rate the methodological quality and statistical reporting of each trial.9 The 11 items are based upon the Delphi list.10 Each item is scored “yes” or “no” with a maximum score of 10 as one criterion is not scored. The PEDro score has demonstrated moderate inter-rater reliability (intraclass correlation coefficient=0.68 [95% confidence interval (CI), 0.57–0.76]) for clinical trials.11 A trial with a score ≥6 was considered to demonstrate high quality consistent with previous reviews.12,13

Standardized mean differences (SMDs) (effect sizes) and 95% CI were calculated based on the postintervention means and standard deviations (SDs).14 When the standard error or 95% CI was provided, the data were converted to the SD. The p values were used to estimate the SD.

Positive SMD values were used to indicate that the outcome favored the intervention group. A value of 0.2–0.5 indicated a small effect size; 0.5–0.8, a moderate effect size; and >0.8, a large effect size.15 A meta-analysis was performed using the inverse variance method and a random effects analysis.16 The Review Manager Version 5.1 (The Cochrane Collaboration, Freiburg, Germany) software program was used for the meta-analysis. Before combining data, trials were categorized into subgroups based on intervention, outcome, and intervention period (4, 8, or 12 weeks). Combining the data in meta-analyses was planned in which a minimum of two trials in a subgroup were clinically homogeneous. A trial was considered clinically homogeneous if a common population and outcome measurement were used. Regarding interventions, there were no restrictions in type or intensity.

The Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach17 was applied to each meta-analysis to determine evidence quality. This approach entails the downgrading of evidence from high to moderate to low and very low quality based on certain criteria. Downgrading the evidence one step (e.g., high to moderate quality) occurred if: (1) the PEDro score was ≤5 for the majority of trials (>50%) in the meta-analysis; (2) statistical heterogeneity between the trials was greater than the accepted low level (I2>25%)18; or (3) there were large confidence intervals, indicating a small number of participants. If there were serious issues with the methodological quality, for example, all trials in the meta-analysis had a PEDro score <6 without allocation concealment or the use of blinded assessors, a double step downgrade occurred (e.g., high to moderate quality). A footnote was used to explain the reason for the grade applied to each meta-analysis.

The combined database search yielded 132 trials. After the adjustment for duplicates, 79 trials were considered. Of them, 34 were eliminated after abstract review for not meeting the selection criteria. The complete text of the remaining 45 studies was examined in detail. Eighteen studies did not meet the inclusion criteria as described. Finally, 27 studies19–45 fulfilled the inclusion criteria (Fig. 1).

Figure 1.

Study flow diagram.

(0.19MB).

The included studies involved 2432 participants (1208 interventions and 1224 comparisons). Based on the available data of the interventions, the participants were 60.6–74.6 years of age and 27.0–95.2% were female. A summary of the included trials is shown in Tables 2 and 3.

Measurements used to assess body function (or impairment) varied and included pain, physical function, stiffness, muscle strength, and ROM. Pain was rated using the WOMAC, visual analog scale, Functional Status Index, and Knee injury and Osteoarthritis Outcome Score. Self-reported information regarding physical function and stiffness was evaluated using the WOMAC. Muscle strength was measured for the knee extensors or flexors. The ROM was measured at maximum active knee extension or flexion. Other than these measurements, measures regarding walking ability (50-m timed walk, gait speed) were included in 6 trials.22,24,26,32,37,40 Health-related QOL was assessed using the Short Form 36 Health Survey. Depressive symptoms were rated using the Hospital Anxiety and Depression score in two trials.22,24 The Oxford Knee Score or timed up and go (TUG) test was used to assess activity limitations.

There were 27 high-quality trials (PEDro score >5/10), with a mean score of 6.6/10 across all trials. The total PEDro scale scores were 8 points for five trials,24,29,31,32,37,39,43 7 points for eleven trials,18,21,23,26,27,34,35,40–42,44 6 points for three trials,25,28,33,38 5 points for two trials,19,36 4 points for two trials,20,22 and 3 points for one trial.32 The most adhered to items on the PEDro scale were random allocation, the use of similar groups at baseline, measurements of variability for at least one key outcome, and between group comparisons, which were evident in almost all trials. None of the trials blinded the participants or therapists, which is expected given that these items are the most difficult to adhere to in trials of interventions involving exercise. Sixteen trials reported employing an intention-to-treat analysis, while 21 used allocation concealment, 17 used blinded outcome assessors, and 19 used measurements of at least one key outcome for >85% of the participants.

From among the outcomes observed as having a significant effect in our study, those that were not confirmed in preceding studies included pain, physical function and stiffness evaluated by the WOMAC, knee extension strength, active knee flexion ROM, TUG, and gait speed by exercise for 8 weeks after discharge in addition to standard postoperative intervention. The effect size of each was small, and the quality of evidence was low to moderate. Our meta-analysis showed that an exercise-intervention period ≥8 weeks is needed after discharge to improve body function and activity. These results have not been reported in preceding studies. Regarding active knee flexion ROM, Minns Lowe et al.5 reported a significant effect due to exercise intervention by integrating knee joint flexion and extension. In contrast, our study integrated the data of knee flexion ROM only. The results obtained by Minns Lowe et al.5 and in our study indicate that the improvement in knee ROM due to exercise starting after discharge in addition to standard postoperative interventions was led by improved flexion ROM and required at least 8 weeks.

In our study, the following types of intervention were not observed as having any significant effects in several outcomes: 4- or 8-week preoperative exercise intervention for pain, physical function evaluated using the WOMAC, knee extension strength; early exercise intervention for pain; 4- or 8-week early versus late standard postoperative interventions; 4-week exercise starting in the hospital added to the standard postoperative intervention for pain, physical functions, and stiffness evaluated using the WOMAC and active knee extension/flexion ROM; 8 weeks exercise starting after discharge in addition to standard postoperative intervention for active knee extension ROM. The results of our meta-analysis did not suggest that the early start of exercise demonstrated an effect on body function and activity before and/or after TKA. These evidences were discovered for the first time in the current systematic review and meta-analysis. However, the quality of this evidence ranged from very low to high, so the degree of effect may vary in subsequent studies.

The results of this study suggest that exercise intervention for 8 weeks after discharge would improve pain, physical function, and stiffness measured by the WOMAC, knee extension strength, active knee flexion ROM, TUG results, and gait speed. These results are an important fact for clinicians since insurers may worry about an increase in medical costs due to the continuous exercise intervention. However, the physical function in patients who performed exercise for 8 weeks was more improved than that of controls. Femoral fractures occur at a rate of 2.5% after TKA.46 The risk of this fracture can be decreased by using exercise to improve one's physical function.47,48 Therefore, our findings might be used by clinicians to support the claim that insurers should accept the need for an exercise intervention period ≥8 weeks after discharge.

The strengths of our study are as follows. Prior to our study, Wallis and Taylor,3 Gill and McBurney,4 Minns Lowe et al.,5 and Artz et al.6 conducted systematic reviews and meta-analyses regarding the effect of exercise intervention in patients who underwent TKA. However, Wallis and Taylor3 and Gill and McBurney4 did not investigate the effect of postoperative exercise intervention. Additionally, the research design by Gill and McBurney4 includes primary studies other than RCT. Moreover, Gill and McBurney4 and Minns Lowe et al.5 used data other than the mean value and standard deviation of the primary study for their meta-analysis and did not investigate evidence level. Artz et al.6 conducted a meta-analysis based on the differences in the follow-up period. Accordingly, these reports have room for examinations regarding the effective exercise intervention period. Meanwhile, regarding our findings, a meta-analysis was performed based on the subgroup of similar intervention periods. Consequently, the results obtained from our study are the effective intervention period of exercise started after discharge. Our findings will be helpful to clinicians who are planning the exercise program after discharge in patients with knee OA undergoing TKA.

The current study has three main limitations. The first limitation involves the evidence level. The GRADE system used in our study determines the evidence level based on the results of investigations regarding all five items. One such item is the presence of publication bias, which is investigated by a visual analysis using funnel plotting or statistical testing. However, in the case of a statistical test, if the primary study has a score of ≤10, then a β error is suspected in the investigation results.49 The primary study included in the meta-analysis of our study had a maximum score of 3, and performing a detailed investigation regarding publication bias was difficult. Therefore, the evidence level determined in our study is provisional, with the possibility that the evidence may have been downgraded by one level due to the results of future studies depending on the outcome thereof. The second limitation regards the individual effect of the exercise intervention. The results of the exercise intervention are expected to differ depending on the time of commencement and program content. Accordingly, in our study, an investigation was performed by dividing the primary study into the following four items in terms of content: preoperative exercise intervention; postoperative exercise intervention; early postoperative exercise intervention; and exercise intervention after discharge. Thereby, the data from primary studies with similar exercise content is more fully integrated than in past reviews,3–6 and it may be said that the range of generalization is clearer than in past reviews. However, because most primary studies combine a plurality of exercise interventions, an investigation into the effect of each individual exercise could not be performed. Therefore, the effects of individual exercise intervention remain unknown. The third limitation regards the frequency of the exercise intervention. The primary studies extracted in our study did not have a completely matching frequency. Regardless, because there were only a few articles, the data could not be stratified and investigated. Therefore, the influence of the differences in frequency on the effect cannot be eliminated from the analysis results. Finally, fourth limitation is that some studies not published in a non-English language were not included in our review. A systematic review of intervention researchers working in non-English speaking countries might have published some of their research in local journals. However, English reports are more likely to have better methodological quality than reports written in other languages.50 Thus, in this review the results are considered to be largely unchanged if some studies not published in a non-English language were included. To overcome these four limitations, it is necessary to carry perform a statistical analysis to investigate the presence of publication bias, stratify the primary studies included in the meta-analysis by content and frequency, and add investigations after the number of articles on RCT increases in the future.