The guideline development committee was composed of 16 experts from four medical disciplines (primary care two, rheumatology 11, orthopaedics one, and evidence-based medicine two) and six countries in Europe and North America (France, Netherlands, Sweden, UK, Canada and the USA). All members of this guideline development team participated in: (1) a critical appraisal of existing treatment guidelines; (2) a Delphi exercise to generate consensus recommendations; and (3) an exercise to grade the strength of recommendation for all modalities of therapy recommended. Three additional experts were invited to undertake critical appraisals of existing guidelines in languages other than English.

A systematic literature search for existing guidelines for the management of hip and/or knee OA published in any language between 1945 and October 2005 was undertaken using MEDLINE (1966–), EMBASE (1980–), CINAHL (1980–), AMED (1985–) and the Science Citation Index (1945–). The search strategy consisted of two basic components: guidelines in any term (e.g., guidelines, recommendations, standards, algorism, or expert consensus, etc.) and hip or knee OA in any possible terms in the databases (Appendix 1). In addition, Google (the first 100 hits) and seven Guideline Websites were searched, including the National Guideline Clearinghouse http://www.guidelines.gov/, Primary Care Clinical Practice Guidelines http://medicine.ucsf.edu/resources/guidelines/, the Scottish Intercollegiate Guidelines Network http://www.sign.ac.uk/, the Canadian Medical Association Infobase for Clinical Practice Guidelines http://mdm.ca/cpgsnew/cpgs/index.asp, the Guidelines International Network http://www.g-i-n.net/, Evidence Based Medicine Guidelines http://www.ebm-guidelines.com/, and the National Institute for Clinical Excellence http://www.nice.org.uk/.

Guidelines developed for the management of hip and/or knee OA using consensus or evidence-based methods were included. The latest version was included if the guidelines had been updated. Guidelines developed for OA in other joints or for aspects of OA other than treatment were excluded, as were narrative reviews, commentaries and appraisals of implementation.

English language guidelines were randomly assigned to three groups of four committee members for appraisal of quality and content. Three guidelines published in German and Dutch were appraised by three additional experts who were fluent in these languages. The quality of the guidelines was assessed using the AGREE instrument²², in which 23 criteria in seven domains are evaluated. These include the scope and purpose of the guidelines, stakeholder participation, methodological rigour, clarity, applicability, editorial independence and overall quality. The content was extracted using a comprehensive reference list of treatment modalities. Each appraiser scored the guidelines independently and results were collected and analysed by the lead investigator (WZ) and the co-chairs (GN and RM), who did not take part in the assessment.

A systematic search of the literature published between 31 January 2002 and 31 January 2006 was undertaken using MEDLINE, EMBASE, CINHAL, AMED, the Science Citation Index and the Cochrane Library databases. Research evidence prior to January 2002 was not sought systematically as this was available from the systematic literature review conducted by EULAR¹⁷. Separate searches for research evidence for each treatment modality were undertaken. Each search was conducted sequentially according to the evidence hierarchy (SRs/MAs, followed by RCTs/controlled trials (CTs), quasi-experimental and uncontrolled studies) (Table II)²³. An example of how this search strategy was employed to obtain the best available research evidence for the efficacy of acetaminophen (paracetamol) is shown in Appendix 2. The same strategy was used for searching MEDLINE, EMBASE, CINHAL and AMED. For the Science Citation Index, however, a key word search was used and all possible terms and combinations of terms were tied in order to obtain relevant citations. Medical subject heading searches (MeSH) were used for all databases and key word searches were used if a MeSH search was not available. All MeSH search terms were exploded. The reference lists of SRs were examined and any additional studies meeting the inclusion/exclusion criteria were included.

The search in the Cochrane Library included MeSH searches of Cochrane reviews, abstracts of Quality Assessed Systematic Reviews, the Cochrane Controlled Trial Register, the National Health Service (NHS) Economic Evaluation Databases, the Health Technology Assessment Database and the NHS Economic Evaluation Bibliography Details Only. In addition, a comprehensive search for all articles including the term OA regardless of treatment was undertaken.

Only studies with clinical outcomes for hip and/or knee OA were included. The main focus was on SRs/MAs, RCTs/CTs, uncontrolled trials, cohort studies, case–control studies, cross-sectional studies and economic evaluations. Studies of OA at other sites such as the hand or spine, and other chronic joint diseases were excluded, apart from studies in which adverse effects of relevant pharmacologic treatments were being investigated as a primary outcome. Case reports, animal studies, non-clinical outcome studies, narrative review articles, commentaries and guidelines were excluded.

The efficacy of any modality of treatment was determined by using the best available evidence. For example, when the efficacy of an intervention could be confirmed by category Ia evidence (MA/SR of RCTs), then studies lower in the evidence hierarchy such as individual RCTs (category Ib) were not reviewed (Table I). If there was more than one study in the same evidence level (e.g., four SRs for NSAIDs), the study with the best quality score was used. Information concerning side effects was obtained from both RCTs and observational studies. While the efficacy of each therapeutic intervention was assessed separately for hip and knee OA, side effects were evaluated for each intervention regardless of the OA therapy and the target joint. For determination of cost effectiveness, only cost-utility analyses were included.

The quality of SR/MAs was assessed using the Oxman and Guyatt checklist²⁴ and the quality of RCTs was evaluated using the Jadad method²⁵. All quality scores were converted into percentages of the maximum score attainable. Quality assessments were not undertaken for other types of study designs, such as cohort or case–control studies. For cost-utility analysis, study perspective, comparator, time horizon, discounting, modelling and uncertainty were evaluated.

Effect sizes (ESs) and 95% confidence intervals (CIs) compared with placebo or active control were calculated for continuous outcomes such as reduction of pain from baseline or improvement in function²⁶. ES is the standard mean difference, i.e., the mean difference between a treatment and a control group divided by the standard deviation of the difference. It is expressed as a number without units and can be used for comparisons across all interventions. From the clinical standpoint ESs of 0.2 are considered small and 0.5 moderate, while an ES>0.8 indicates a large clinical effect²⁷. Statistical pooling was undertaken, as appropriate, when SRs were not available²⁸. For dichotomous data, such as the percentage of patients with moderate to excellent (or more than 50%) pain relief or symptomatic improvement, the number needed to treat (NNT) was estimated²⁹. The NNT is the estimated number of patients who need to be treated to achieve the target effect. Thus the smaller the NNT the better the treatment effect. The 95% CI for the NNT was calculated using Altman's method³⁰.

The relative risk (RR) of side effects was calculated from RCTs or cohort studies for the incident risk, and from cross-sectional studies for prevalent risk. Odds ratios (ORs) were calculated from case–control studies³¹. Both RR and OR provide information on how many times more likely (or less likely) it is that a subject who is exposed to a treatment modality will have an adverse event, when compared with a subject who is not exposed. An RR/OR=1 indicates no increased risk, whereas an RR/OR>1 or <1 indicates increased or decreased risk, respectively.

The systematic literature search yielded 1462 citations (MEDLINE 276, EMBASE 413, CINAHL 81, AMED 27 and SCI 553, Google and Guidelines Websites 112). Of these, 23 met the inclusion and exclusion criteria specified^{16, 17, 18, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51}. Six guidelines were predominantly based on opinion, five primarily based on evidence and 12 based on both (Table II). Whilst the majority of the guidelines¹⁴ did not separate hip and knee, eight were specific for knee but only one for hip OA. Thirteen guidelines had been developed for specific care settings (five for primary care, three for rheumatology, three for physiotherapy and two for orthopaedics); but 10 did not specify target users.

Scores for overall quality of guidelines were 28%, 41% and 51% for opinion-based, evidence-based and hybrid guidelines, respectively (P<0.001) (Fig. 1). Scores for different quality criteria varied but apart from applicability, opinion-based guidelines tended to have lower scores (Table III).

Fig. 1

Overall quality score of guidelines (mean±s.e.m.).

 

Fifty-one treatment modalities were addressed in the 23 guidelines. Twenty of these modalities were recommended by all (100%) of the guidelines in which they were addressed (Table IV), but the strength of agreement for any modality appeared to be related to the number of guidelines that addressed that modality. For example, while regular telephone contact and knee fusion were recommended in 100% of the guidelines in which these modalities of therapy were considered, this was actually in only two guidelines for each modality. By contrast, although weight loss was not universally recommended, it was in fact recommended in 13/14 of the guidelines where this modality was considered.

Evidence to support recommendations ranged from Ia (SR of RCTs) to IV (expert opinion), and did not necessarily reflect the extent of agreement (Table IV). For example, while canes/sticks, total joint replacement and osteotomy were not supported by RCTs, they were still universally recommended in the guidelines which addressed them. In contrast, despite evidence from SRs of RCTs for the efficacy of chondroitin sulphate and ultrasound, they were recommended by <50% of the guidelines in which these modalities were considered (Table IV).

With the exception of combination therapy, the use of a cane/stick and referral, all the non-pharmacologic and pharmacologic therapies recommended universally by existing guidelines were supported by recent SRs of RCTs (Ia) or RCTs (Ib) published after 2002. By contrast, there were no placebo controlled trials of surgical modalities of treatment such as total joint replacement and osteotomy, and supporting evidence came from uncontrolled or non-experimental observational studies (Table V). Overall quality scores for evidence ranged between 40% and 100% but 24/40 studies (60%) scored 100% (Table V).

The ES for pain relief scores varied from small (e.g., education ES=0.06, 95% CI 0.02, 0.10) to moderate (e.g., aerobic exercise ES=0.52, 95% CI 0.34, 0.70). No modality of therapy had an ES as high as 0.80 – the accepted criterion for a large clinical effect²⁷ (Fig. 2). ESs for pain relief score with oral analgesics such as acetaminophen (ES=0.21 95% CI 0.02, 0.41) and NSAIDs (ES=0.32, 95% CI 0.24, 0.39) were small (Fig. 3 and Table V).

Fig. 2

ES for pain relief with non-pharmacological therapies.

 

Fig. 3

ES for pain relief with pharmacological therapies.

 

ESs for improvement in function were also generally small, and very similar to those for pain relief, for a number of modalities of non-pharmacological therapies (Table V). However, the ES for improvement in function for >10% weight reduction was 0.69 (95% CI 0.24, 1.14) compared with the ES for pain relief (0.13, 95% CI −0.12, 0.38). ESs for reduction in stiffness were also available for a few modalities of treatment (Table V).

Some studies provided data, which allowed calculation of NNTs. For example, weight reduction (>10%) was associated with an NNT of three (95% CI 2, 9), i.e., one in three patients with knee OA who achieved this loss of weight would have more than 50% reduction in the total Western Ontario and McMaster Universities (WOMAC) Osteoarthritis index⁵². The NNT for topical NSAIDs was also three (95% CI 2, 4), indicating that one in three patients with pain associated with knee OA treated with a topical NSAID would be expected to experience moderate to excellent pain relief⁵³.

In general, non-pharmacologic therapies had numerically smaller ES (ES=0.25, 95% CI 0.16, 0.34) than pharmacological therapies (ES=0.39, 95% CI 0.31, 0.47) (Fig. 2, Fig. 3). Among surgical treatments, ES could only be calculated for arthroscopic lavage and debridement. An SR of four RCTs showed that arthroscopic joint lavage and debridement were no more effective than placebo⁵⁴. One placebo controlled RCT (with a quality score of 100%) included in this review demonstrated that the ES for arthroscopic lavage and debridement vs placebo were 0.09 (95% CI −0.27, 0.44) and −0.01 (95% CI −0.37, 0.35), respectively⁵⁵. Similar results were obtained for improvement in function (Table V). Although there are no placebo controlled RCTs of total joint (knee or hip) replacement or osteotomy, two recent SRs of uncontrolled trials and cohort studies confirmed that they were highly effective in relieving pain and improving quality of life^{56, 57}.

Evidence for side effects of treatments has been mainly investigated in pharmacologic therapies. Oral NSAIDs were associated with 3–5 times the risk of gastrointestinal (GI) side effects when compared with placebo or non-exposure⁵⁸, whereas treatment with topical NSAIDs resulted in no more adverse GI events than placebo (RR=0.81, 95% CI 0.43, 1.56)⁵³ or non-exposure (OR=1.45, 95% CI 0.84, 2.50)⁵⁹ (Table VI). Whether or not long-term treatment with acetaminophen 4 g daily is associated with GI and renal side effects remains inconclusive (Table VI). Treatment with cyclooxygenase-2 (COX-2) selective drugs or conventional non-selective NSAIDs together with proton pump inhibitors (PPIs) or misoprostol has been shown to be associated with a reduction in the risk of NSAID-induced upper GI side effects. However, treatment with rofecoxib has been shown to be associated with an increased risk of cardiovascular (CV) events (RR=2.24, 95% CI 1.24, 4.02)⁶⁰ and treatment with misoprostol with an increased risk of diarrhoea (RR=1.81, 95% CI 1.52, 2.61)⁶¹. Following the withdrawal of rofecoxib, a number of RCTs and SRs of the CV safety of other coxibs and conventional non-selective NSAIDs have been undertaken. While the increased risk of CV side effects with rofecoxib was confirmed, the evidence for similar CV toxicity with celecoxib, valdecoxib and conventional non-selective NSAIDs was inconsistent (Table VI). However, the overall CV risk associated with COX-2 selective inhibitors was not significantly greater than that associated with conventional non-selective NSAIDs (RR=1.19, 95% CI 0.80, 1.75)⁶² (Table VI).

The critical appraisal of the 23 existing guidelines showed that of 51 treatment modalities addressed, 20 were universally recommended in those guidelines in which they were considered (100% agreement in Table IV). These included recommendations for non-pharmacological modalities of therapy such as education, exercise, patient contact by telephone and provision of walking aids and pharmacological treatments such as acetaminophen, non-selective NSAIDs with co-prescription of gastroprotective agents or selective COX-2 inhibitors, opioids and some herbal remedies. Surgical treatments recommended in all the guidelines in which they were considered included knee aspiration and joint lavage as well as osteotomy, knee fusion and total joint replacements. Self-management and the combination of non-pharmacologic and pharmacologic treatments were also uniformly recommended core recommendations. It is apparent that this core set of recommended therapies must reflect the availability of treatments. The less than universal recommendation for some modalities of therapy may have been a consequence of them not being universally available, e.g., topical NSAIDs and avocado soybean unsaponifiables are available in Europe but not in the USA. It is also important to consider the number of guidelines, which considered any particular modality of therapy in ones interpretation of the reliability of the strength of agreement for that treatment. Clearly, the confidence one can have in the universal recommendation for exercise, where this modality of treatment was considered and endorsed in 21/21 guidelines, is likely to be greater than the confidence one has in the recommendation for knee fusion, which was only considered and endorsed in 2/2 guidelines.

It was also apparent that some of the core set of universally recommended therapies were not supported by evidence from RCTs. For example, while exercise of various types was supported by SR of RCTs (level Ia), total joint replacement was only supported by uncontrolled or cohort studies (level III) and the recommendations for knee aspiration and knee fusion were based on expert opinion (level IV). The extent to which RCTs should be the gold standard for the recommendation of all treatments has been the subject of previous discussion and controversy^{71, 72}. Nevertheless, the level of research evidence and clinical effectiveness have been important considerations in the development of recent guidelines for the treatment of knee and hip OA^{17, 18} and in the development of the OARSI recommendations. Clearly guidelines based on recommendations for treatments for which there is proven evidence of benefit should at least have the potential for improving clinical outcomes and the quality of health care for patients, although success is certainly not guaranteed and evidence-based guidelines are only one option for improving the quality of health care.

A pilot survey of the perceived usefulness of the treatment modalities addressed by the existing guidelines was conducted among physicians and other health care professionals attending a New York University – OARSI Rheumatology Symposium in 2006. The purpose of the survey was to collect the users' opinions on the usefulness of current treatment guidelines. The usefulness of each recommended treatment modality was assessed by the participants using a 5-point categorical scale (not useful, slightly useful, moderately useful, very useful and absolutely essential). Votes (%) on “very useful or absolutely essential” were calculated. Of 19 participants who completed the questionnaire (four general physicians, eight rheumatologists, one physiotherapist, one orthopaedic surgeon, one pharmacist and four other health professionals), 94% perceived total joint replacement to be very useful or essential therapy for both knee OA and hip OA. Combination therapy was judged to be very useful or essential by 79% for knee OA and 72% for hip OA. Weight reduction was perceived to be more useful for knee than hip OA by 68%, whereas NSAIDs, NSAID plus PPIs, COX-2 inhibitors, self-management, education and exercise were considered useful for both hip and knee OA. Although this survey was far from being truly representative of all potential guideline users and only involved a very small number of participants, most of whom were from the United States, the views expressed about the usefulness of various modalities of treatment were at least consistent with the appraisal of existing guidelines that has led to the definition of a tentative core set of recommended treatment modalities. It also points to a possible way of assessing the potential applicability of any future recommendations for other modalities of therapy being considered as additions to this core set.

The methodology involved in the development of treatment guidelines for OA has evolved considerably in the last decade. Between the publication of the first guidelines for the treatment of OA by the Royal College of Physicians in 1993⁴⁹ and the publication of the EULAR recommendations in 2005¹⁸, the paradigm has shifted from purely opinion-based guidelines⁴⁹ to entirely evidence-based guidelines such as the Prodigy Guidance³⁴ and subsequently to hybrid guidelines based on both research evidence and clinical expertise such as the EULAR recommendations^{17, 18}. However, no attempt had been made to try and assess the quality of these guidelines. We have therefore used the AGREE instrument to evaluate the quality of all existing guidelines for scope and purpose, stakeholder participation, methodological rigour, clarity, applicability, editorial independence and overall quality²². Overall quality was better in evidence-based than opinion-based guidelines, and significantly better still in the hybrid guidelines that combined research evidence with expert opinion (Fig. 1). This is mainly attributable to the improved scores for scope and purpose (P=0.007), rigour of development (P<0.001) and editorial independence (P=0.013) in the hybrid guidelines (Table III). There is a tendency for evidence-based guidelines to have lower applicability, although the differences are not statistically significant (Table III). This may, in part, reflect the gap that exists between RCTs which demonstrate that an intervention works (“efficacy”) and how often and well the intervention works in clinical practise (“clinical effectiveness”). Hybrid guidelines can be expected to demonstrate improved applicability as clinical expertise can temper the rigidity of research data and close the gap between research and clinical practise.

In the development of hybrid guidelines by the EULAR OA Task Force, expert consensus on the most important propositions was followed by a systematic search for published supporting research evidence, prior to assigning a strength and confidence of recommendation for each treatment proposition. These were based on combined consideration of the research evidence and clinical expertise after also considering risks and benefits, including potential adverse effects and the cost of each treatment modality¹⁸. This method is clinically driven and evidence supported. The sequence of steps has been modified slightly for the development of the OARSI Treatment Guidelines. An initial SR of research evidence was followed by the development of expert consensus based on a combined consideration of the research evidence and the clinical expertise of the members of the committee. This was then followed by assignment of strength and confidence of recommendation for each proposition as before. This current method is evidence-driven and clinically supported. Another important difference in the methodology used in the development of the OARSI recommendations has been that the committee has not arbitrarily restricted the number of treatment options that it would consider, as was the case in the development of the EULAR guidelines^{17, 18}.