Intended for healthcare professionals


Effectiveness of physiotherapy exercise after knee arthroplasty for osteoarthritis: systematic review and meta-analysis of randomised controlled trials

BMJ 2007; 335 doi: (Published 18 October 2007) Cite this as: BMJ 2007;335:812
  1. Catherine J Minns Lowe, research fellow1,
  2. Karen L Barker, director2,
  3. Michael Dewey, special lecturer3,
  4. Catherine M Sackley, professor of physiotherapy research1
  1. 1Department of Primary Care and General Practice, University of Birmingham
  2. 2Physiotherapy Research Unit, Nuffield Orthopaedic Hospital NHS Trust, Oxford
  3. 3School of Community Health Sciences, University of Nottingham
  1. Correspondence to: C J Minns Lowe, Physiotherapy Research Unit, Nuffield Orthopaedic Centre NHS Trust, Oxford OX3 7LD catherine.minnslowe{at}
  • Accepted 8 August 2007


Objective To evaluate the effectiveness of physiotherapy exercise after elective primary total knee arthroplasty in patients with osteoarthritis.

Design Systematic review.

Data sources Database searches: AMED, CINAHL, Embase, King's Fund, Medline, Cochrane library (Cochrane reviews, Cochrane central register of controlled trials, DARE), PEDro, Department of Health national research register. Hand searches: Physiotherapy, Physical Therapy, Journal of Bone and Joint Surgery (Britain) Conference Proceedings.

Review methods Randomised controlled trials were reviewed if they included a physiotherapy exercise intervention compared with usual or standard physiotherapy care, or compared two types of exercise physiotherapy interventions meeting the review criteria, after discharge from hospital after elective primary total knee arthroplasty for osteoarthritis.

Outcome measures Functional activities of daily living, walking, quality of life, muscle strength, and range of motion in the knee joint. Trial quality was extensively evaluated. Narrative synthesis plus meta-analyses with fixed effect models, weighted mean differences, standardised effect sizes, and tests for heterogeneity.

Results Six trials were identified, five of which were suitable for inclusion in meta-analyses. There was a small to moderate standardised effect size (0.33, 95% confidence interval 0.07 to 0.58) in favour of functional exercise for function three to four months postoperatively. There were also small to moderate weighted mean differences of 2.9 (0.61 to 5.2) for range of joint motion and 1.66 (−1 to 4.3) for quality of life in favour of functional exercise three to four months postoperatively. Benefits of treatment were no longer evident at one year.

Conclusions Interventions including physiotherapy functional exercises after discharge result in short term benefit after elective primary total knee arthroplasty. Effect sizes are small to moderate, with no long term benefit.


Osteoarthritis is the commonest cause of disability in older people,1 with painful knee osteoarthritis affecting 10% of people aged over 55 in the United Kingdom.2 Over 80% of patients experience limitations in performing activities of daily living, such as mobility outside the home, household chores, and work duties.3 In 2005, patients with osteoarthritis accounted for at least 55 495 primary knee joint arthroplasties in England and Wales.4 As the length of hospital stay after joint arthroplasty surgery has markedly and rapidly decreased,5 and given that patients who undergo knee arthroplasty may still experience considerable functional impairment postoperatively,6 the effectiveness of physiotherapy after discharge is a valid question. The present uncertainty regarding effectiveness makes it difficult for commissioning organisations, healthcare practitioners, and patients to make decisions regarding such physiotherapy. We systematically reviewed randomised controlled trials to determine the effectiveness of physiotherapy exercise after discharge in terms of improving function, quality of life, walking, range of motion in the knee joint, and muscle strength for patients with osteoarthritis after elective primary unilateral total knee arthroplasty.



In March 2005 and in April 2007 we identified randomised controlled trials by simultaneously searching AMED (from 1985), CINAHL (from 1982), Embase (from 1974), Kings Fund database (from 1979), and Medline (from 1966) via Knowledge Access 24/7 (KA24). We also searched the Cochrane library, PEDro physiotherapy evidence database, and the Department of Health national research register. In July 2005 and April 2007 we handsearched Physiotherapy (1985- March 2007 inclusive) and Physical Therapy (1985-April 2007 inclusive) to double check for trials. The conference proceedings in the Journal of Bone and Joint Surgery (Britain) (1985-2006 inclusive) were also handsearched, as were the reference lists of included trials.

As it is difficult to locate physiotherapy trials, we considered that using multiple general searches was the optimum method. This review is part of a series that included both knee and hip search terms. Table 1 summarises the searches. No language restrictions were applied.

Table 1

 Search strategy for systematic review

View this table:


We sought randomised controlled trials of patients undergoing elective total knee arthroplasty for osteoarthritis who received an intervention of physiotherapy exercise after discharge from hospital. We used broad definitions of “physiotherapy” and “exercise” to include any exercises or exercise programme advised or provided by physiotherapists or physical therapists during the rehabilitative period after discharge from hospital after surgery in the outpatient, community, or home setting. We excluded trials in which the intervention consisted of an electrical adjunct to physiotherapy, such as the use of continuous passive motion. Physiotherapy exercise interventions included outpatient physiotherapy sessions and functional physiotherapy programmes, in which exercises are based on functional activities. Trials were included if they investigated a physiotherapy intervention compared with usual or standard care or compared two different types of relevant physiotherapy intervention. Usual or standard care refers to the continuation of home exercise programmes provided to patients during a stay in hospital. These programmes usually consist of isometric or simple strengthening exercises, exercises to regain range of movement, and stretches. Effectiveness outcomes were measures of functional activities of daily living, walking, self reported measures of quality of life, muscle strength, and range of motion in the knee joint. As most trials use functional measures rather than specific pain outcomes, we did not include pain as an effectiveness outcome. Two reviewers (CML and CS) assessed and agreed on study eligibility.

Validity assessment, data abstraction, and quality assessment

We developed and piloted a data extraction form using quality indicators from the CONSORT statement7 and the CASP guidelines8 (table 2). Similar analysis of individual quality components has previously been used in reviews of physiotherapy9 and is advocated to avoid known problems associated with existing composite scores.10 Items could be marked as yes, no, unclear, or partial. Items were marked as yes only if they fully and explicitly met the detailed criteria laid out in the CONSORT standards.7 Two reviewers (CML and KB) independently extracted the data. KB was masked to the key details of each paper and the extent to which masking was successful was assessed. The masking rates were 80% for authors, 20% for journals, 80% for author affiliations, and 80% for funding sources, all of which except journal of publication were considered successful. The level of agreement between reviewers was 69.09% (κ 0.524, intraclass correlation coefficient (2,1) 0.49, 95% confidence interval 0.30 to 0.63).

Table 2

 Quality component checklist and quality evaluation of six trials included in review

View this table:

We resolved initial disagreements regarding study quality by discussion until consensus was reached. Major disagreement was rare; usually disagreement was the more minor “yes” to “partial/unclear” or “no” to “partial/unclear” and 100% agreement was obtained. A third reviewer (CS) was available in the event of consensus not being reached, but this was not required. Where key study details were absent or unclear we contacted authors for further information.

We considered studies to be of good quality if they were good enough to include in meta-analyses. Table 2 presents quality assessment findings for each study. We excluded one studyw1 from the meta-analysis because participants were allocated by alternation. All trial outcomes were measured by assessors masked to allocation. As table 2 shows, most studies clearly reported the flow of participants through the trial, justified the sample size, and included intention to treat analyses. Several quality indicators were not fully discussed in all papers, such as allocation concealment and details regarding the implementation of randomisation methods.

Quantitative data synthesis

We carried out meta-analyses for knee function, walking, range of joint motion, and quality of life with R2.3.1 and the rmeta package.11 Our outcome was the score at the chosen time point rather than the change in score as this maximised the number of comparable studies. The time points used were three to four months after surgery and 12 months after surgery. If the same measure was reported we used weighted mean differences, otherwise we used standardised effect sizes (small (0.2), medium (0.5), and large (0.8)12). We used fixed effect models and 95% confidence intervals throughout and performed tests of heterogeneity (χ2) at a 5% significance level, though we accept these have low power because few studies were available for meta-analyses. We also calculated I2 to give a measurement of the degree of heterogeneity between the trials in the meta-analysis. Random effects models were not considered as there was no compelling evidence of heterogeneity and estimating the variation between studies is difficult with such low numbers. The differences were calculated so that positive differences indicate that the effect favoured treatment and negative differences that the effect favoured control or usual care. We considered it inappropriate to assess publication bias because of the small number of trials.


We identified and screened 27 potentially relevant studies. Of these, six studiesw1-w6 were included in the systematic review and fivew2-w6 in the meta-analysis (fig 1). Table 3 gives details of excluded studies.w7-w27 Table 4 provides the results of the analysis of heterogeneity.


Fig 1 Trial flow diagram to summarise the stages of systematic review

Table 3

 Studies excluded from the review

View this table:
Table 4

 Heterogeneity χ and test* results for range of outcomes

View this table:

Study characteristics

Table 5 summarises the characteristics of the included studies and provides information regarding the participants, interventions, main outcomes, and conclusions reached by authors.

Table 5

 Study characteristics of trials evaluated in systematic review

View this table:

Summary of the interventions and comparisons

With the exception of one trial,w6 in-depth details of the intervention and comparison groups were available from the papers and authors (table 6).

Table 6

 Summary of trial interventions and comparisons included in the knee replacement trials (n=6)

View this table:

The trial interventions were similar to each other in that they provided additional physiotherapy exercises or treatment after discharge after total knee arthroplasty, often involving programmes of functional weight bearing exercise. The study by Rajan et al provided few details regarding the intervention.w6 Though most interventions included functional weight bearing exercises, Codine et al investigated the effect of eccentric isokinetic muscle strengthening with a CYBEX dynamometer.w1 Interventions usually started within two weeks of discharge. Outpatient programmes generally lasted up to 12 weeks, while home exercise programmes were recommended for up to a year or indefinitely in one case.w3

The comparison groups were mainly control groups in which no additional outpatient physiotherapy was organised. Patients were expected to continue with the traditional home exercise programme—namely, isometric strengthening and range of movement exercises plus gait training or re-education provided to all patients during their stay in hospital.

Quantitative data synthesis

Measures of function (five trials)

Five of the studies contained a measure of function.w1-w5 The measures used included the 12 item Oxford knee score,w4 which measures functional ability, including pain, (scores 12-60, low score indicates high function) (Frost et al used one item of this scorew2); the American Knee Society clinical rating score,w1 w3-w4 which measures pain, movement, stability, and functional activity (scores 0-100, high score indicates favourable); the 24 item Western Ontario and McMaster Universities osteoarthritis index (WOMAC),w3 w5 which has domains for pain, stiffness, and function (scored as a percentage by Moffet el alw5 and out of 0-170 for function by Kramer et alw3 (low scores are favourable)); and the Bartlett patellar score,w4 which measures anterior knee pain, quadriceps strength, and function (scores 3-30, high scores are favourable).

Within the individual trials, three found no significant differences between groups.w1-w3 Frost et al found significant differences within groups for the treatment arm, indicating a benefit of treatment.w2 Mockford and Beverland presented no results in their published abstract but supplied summary statistics for their outcomes,w4 allowing us to include their study in the meta-analysis. Moffet et al found significant differences between the two groups, in favour of the intervention, at four and six months after arthroplasty but not at 12 months.w5

Figure 2 shows the three studies with data on functioning at three to four months and 12 months after surgery. Where studies included more than one measure of function we decided to use the Oxford knee and the WOMAC scores as these encompassed all component trials. No trial included both these scores. At three to four months the standardised effect size was 0.33 (95% confidence interval 0.07 to 0.58), which is considered small to moderate.12 At 12 months, with one additional study, the effect size was close to zero at −0.07 and the confidence interval (−0.28 to 0.14) included zero.


Fig 2 Forest plot of standardised effect sizes with confidence intervals for function and results of test for heterogeneity

Walking (three trials)

Three knee arthroplasty trials used some form of outcome measure for walking.w2 w3w5 The measures reported included walking speed over a 10 metre distance, measured in m/sec,w2 and a six minute timed walking test, measured in metres.w3w5 The study by Moffet et alw5 reported on time walking over a 50 metre walkway.

The results from these trials were mixed. One trial found no significant differences between groups,w3 another found differences approaching significance,w5 and the third trial found significant differences within intervention groups.w2 Figure 3 shows that the intervention had no overall influence on walking at either three or 12 months.


Fig 3 Forest plot of standardised effect sizes with confidence intervals for walking

Range of joint motion (five trials)

Five of the total knee arthroplasty trials used the range of motion in the knee joint as an outcome measure.w1-w4w6 Although all measurements were provided in degrees, the method of achieving results varied. Codine et al used a goniometer integrated into a dynamometer to measure knee flexion and extension,w1 while Mockford and Beverland used a goniometer to measure active and passive flexion and extension.w4 Frost et alw2 and Kramer et alw3 both measured active flexion only, and Rajan et al measured range of motion in the knee as a single value.w6

Once again, the results were mixed. Codine et al found a significant difference in knee extension between the two groups at 10 days,w1 though, despite randomisation, extension was different in the two groups at baseline. Another study concluded that there was a significant difference in active knee movement in favour of the intervention group but not in the passive range.w4 In the pilot study by Frost et alw2 there was a trend for less loss of range in the functional group than in the traditional exercise group but the study was small and the difference was not significant. Two other studies also found no significant differences.w3 w6

All the studies on range of movement in the knee joint used the same measure (degrees); therefore figure 4 shows the weighted mean differences and confidence intervals. The three month summary shows an increase of 2.9° (0.61° to 5.2°), which is considered small to moderate. At 12 months the effect was smaller, about 1°, and the confidence interval (−1.10° to 3.00°) included zero.


Fig 4 Forest plot of weighted mean differences with confidence intervals for range of motion (in degrees)

Quality of life (three trials)

Three trials included measures of quality of life.w3-w5 The SF-36 health survey provides an eight scale profile of functional health and wellbeing scores with low scores indicating poor health. Kramer et alw3 used the SF-36, and Moffet et al used a French translation of the same score.w5 Moffet et al also provided the physical component and mental component scores of the SF-12,w5 as did Mockford and Beverland.w4

One trial found no significant differences between the groups.w3 One other trial has not yet presented statistical analyses for this measure.w4 The final trial found small significant differences in favour of the intervention group for the role-physical dimension of the SF-36 and the physical and mental component scores at six month follow-up but not at 12 month follow-up.w5

Figure 5 presents the studies with data on quality of life. At three to four months the studies used the same measure, the SF-12, and so we have presented weighted mean difference results. At 12 months after surgery, however, not all studies used the same measure and therefore we used standardised effect sizes.


Fig 5 Forest plot of weighted mean difference (three to four months) and standardised effect size (12 months) with confidence intervals for quality of life

At three to four months after surgery the weighted mean difference was 1.7 (−1.0 to 4.3), indicating a small effect in favour of the intervention. At 12 months the effect was close to zero with a standardised effect size of 0.03 (−0.20 to 0.25).

Muscle strength

None of the trials included in the review directly measured muscle strength.


This systematic review provides support for the use of physiotherapy exercise interventions with exercises based on functional activities after discharge, rather than traditional home exercise and advice programmes, to obtain short term benefit after elective primary knee arthroplasty. There was a small to moderate standardised effect size in favour of functional exercise for function three to four months postoperatively. Small to moderate weighted mean differences, in favour of functional exercise interventions, were seen for range of joint motion and quality of life three to four months postoperatively. Any benefits seen after treatment did not persist to one year follow-up.

Strengths and weaknesses of review procedures

Physiotherapy literature remains a difficult area to search, with numerous bibliographic databases and unindexed journals.13 While we made every attempt to identify studies in any language, other studies might exist. We believe, however, that this review remains the most comprehensive to date.

Trial quality was good overall. Of the five adequately randomised studies included in the meta-analyses, most were sufficiently powered with adequate strategies to conceal allocation and outcome measurements obtained by assessors blinded to treatment allocation.7 Yet, like most physiotherapy trials,14 studies were relatively small, with 554 participants in the five trials included in the meta-analyses and 614 participants included overall in the review.

The most commonly used outcomes were function, predominantly subjective measures of functional ability, and range of joint motion as an objective measure. While range of joint motion is important, its usefulness as an outcome measure of physiotherapy interventions is limited as other factors, such as prosthetic design, preoperative knee motion, and surgical technique, also influence postoperative range of joint motion.15 None of the trials directly measured muscle strength, although one included leg extensor power,w2 instead studies used objective measures like walking.

There were no apparent problems with our data extraction processes. Although many quality checklists and scales exist, there is no accepted ideal score; component approaches are often preferred as the wide variety of scores and weighting systems available mean that the same trial may score as both high quality and low quality depending on which score is used.10 Additionally, many scoring systems downgrade the quality rating of a trial if it is not double blinded. For many physiotherapy trials, such as those in this review, patients and therapists inevitably know the treatment allocation and this is not an indication of low or high trial quality. For these reasons we used a component approach, although we accept this is controversial.

The χ2 tests did not indicate major problems with heterogeneity in any of the eight analyses, but these were limited by low power. The I2 results also indicated no observed heterogeneity.16 The number of available studies, and their size, does limit this review and prevents its findings from being conclusive. It is perhaps surprising that so few published trials exist for such a common practice. This may be partially attributable to the general lack of research on rehabilitation in orthopaedic surgery patients after discharge, rather than knee arthroplasty patients as such, as we also found few existing trials investigating exercise and rehabilitation after elective hip arthroplasty.

Clinical implications

Presently, given the reduction in length of hospital stay, compressed inpatient rehabilitation, and the limitations of the available evidence, it seems reasonable to refer patients for a short course of physiotherapy after discharge to provide short term benefit. While range of motion may be limited as an outcome measure of physiotherapy, the small to moderate standardised effect size obtained for function, which favours the intervention, is considered clinically important. This reflects actual improvements in one or more important aspects of function reported by patients after they received the treatment intervention. The type of physiotherapy provided also needs consideration. In the short term physiotherapy exercise interventions with exercises based on functional activities may be more effective after total knee arthroplasty than traditional exercise programmes, which concentrate on isometric muscle exercises and exercises to increase range of motion in the joint.

Future directions

Although there were few studies and they were not large, they are still likely to have detected most worthwhile effects. These tentative findings suggest that further research would be worthwhile to reduce the current level of uncertainty.17

There seemed to be no benefits related to treatment at one year, though the evidence is not conclusive. The content of the intervention could be better designed and further tested. Interventions to date have largely consisted of exercise programmes and gait rehabilitation, mainly targeting impairment and helping patients to recover from the effects of surgery rather than specifically targeting limitations in activity or restrictions in participation. From the wider field of rehabilitation as a whole, however, such task training seems highly relevant. A recent systematic review, which assessed physiotherapy on functional outcome after stroke, found that effective studies contained focused exercise programmes within which the relevant functional tasks were directly trained.18 Research is currently underway to determine whether a brief feasible physiotherapy intervention of this type, supplied after discharge, affects patient's functional ability one year after knee arthroplasty. An investigation into the health economics is also included.

What is already known on this topic

  • Osteoarthritis is the commonest cause of disability in older people, and total knee joint arthroplasty is a common orthopaedic procedure

  • Uncertainty exists regarding whether physiotherapy after discharge should be routinely provided to patients after elective primary knee arthroplasty for osteoarthritis

What this study adds

  • Functional physiotherapy exercise soon after discharge results in short term benefit after elective primary knee arthroplasty

  • No benefit was seen at one year


  • We thank Robert Bourne, David Beverland, P Codine, Helen Frost, Patricia Humphreys, John Kramer, and Brian Mockford for providing additional data for the review. Mike Clarke and students on the “Systematic Reviews” Module, May 2005, University of Oxford Department for Continuing Education, commented on the design of the review during its planning.

  • Contributors: CJML designed the review, undertook the review searches, screened trials for eligibility, assessed the quality of the trials, assisted with data analysis, and drafted the paper. She is guarantor. KLB supervised the review, assessed the quality of trials, and reviewed the draft paper. MD designed and undertook the meta-analyses for the review and reviewed the draft paper. CMS supervised the review, screened trials for eligibility, and cowrote the paper.

  • Funding: CJML is funded by a nursing and allied health professional researcher development award from the Department of Health and NHS research and development. CMS is funded by a primary care career scientist award from the Department of Health and NHS research and development.

  • Competing interests: None declared.

  • Ethical approval: Oxford local research ethics committee (AQREC No A03.018).

    Provenance and peer review: Not commissioned; peer reviewed.


View Abstract