Osteoarthritis and Cartilage (1998) 6, 371–3737 1998 Osteoarthritis Research Society 1063–4584/98/060371 + 03 $12.00/0

EDITORIAL

Design of clinical trials in knee osteoarthritis: practical issues fordebate

By Michael Doherty and Adrian Jones

Rheumatology Unit, City Hospital, Nottingham NG5 1PB, U.K.

Keywords: Osteoarthritis, Clinical trials, Design.

A number of guidelines on the conduct of clinicaltrials in osteoarthritis (OA) have been published inrecent years [1–3]. Such guidelines are useful insuggesting, amongst others: appropriate casedefinitions, inclusion and exclusion criteria andoutcome measures. These guidelines can be used intrials of drugs that may modify symptoms and/orstructural change. Such consensus statementsshould encourage more thoughtful, more consist-ent and hopefully better study designs. Greaterhomogeneity also permits wider inclusion of studydata within systematic reviews and meta-analyses.A number of practical and theoretical issues stillneed to be addressed, and subjected to widerdiscussion. Particularly important we feel arethose which may adversely affect patient recruit-ment, drop-out rate, and generalizability of studyfindings. We believe that the following aspects ofknee OA drug clinical trials, in particular, requirefurther debate.

Should only one (signal) joint receivetreatment and/or assessment?

In rheumatoid arthritis (RA) information onmany joints is pooled to give a single summatedscore for one individual, expressed, for example, asthe swollen joint count or tender joint count [4].Pain scores in RA are also given as a single overallscore with no reference to a specified ‘signal’ orworst joint [4].

By contrast, in OA studies the primary outcomeassessment is inevitably focused on a single jointsite, specifically the knee, hip or hand [1, 2]. Thisappears reasonable in that:

(1) Risk factors for development and progressionof OA are site specific so that focus on one site

encourages homogeneity of the study popu-lation,

(2) Unlike RA, OA usually causes symptoms at justone or a few sites at any particular time.

However, the majority of patients with knee orhand OA, and many patients with hip OA, havebilateral symptoms. In these cases, it is advisedthat the most symptomatic knee or hip is selectedas the signal joint for symptom-modifying studies[1, 2]. Assessment of the contralateral joint can beincluded as a secondary outcome measure [1]. Forhands, however, it is apparently acceptable to pooldata from right and left sides, although moreattention might be given to the dominant side [1].

This approach presents several practical andtheoretical problems with respect both to assess-ment and drug treatment. For example:

, Many patients report that both knees areequally symptomatic, or that the ‘worst’ kneechanges from week to week. Should suchpatients be excluded from study? Should oneknee be picked at random as the signal joint?Should both knees receive equal treatments andbe assessed equally for outcome? The latteroption would facilitate recruitment to studiesand also produce more generalizable data thatreflects clinical practice—how many physicianswould treat just one of two symptomatic knees?

, Many patients find it hard to give separatescores for pain and function for bilaterallysymptomatic knees. Pain experience on one sidemay influence pain reporting on the other [5].There are no data to suggest that pain severityis halved if one rather than two equallytroublesome knees are involved. A composite(right and left) knee pain assessment wouldfairly represent the pain severity experienced byan individual.

, If intra-articular treatment is given to just oneknee, exacerbation of pain in the contralateral

Address correspondence to: Professor Michael Doherty,Rheumatology Unit, City Hospital, Nottingham, NG5 1PB.Tel: + 44 115 9691412; Fax: + 44 115 9627709.

371

Doherty and Jones: Practical issues in clinical trials372

untreated knee during the study period mayresult in patient withdrawal irrespective of theeffect of treatment on the signal knee. Thisnecessitates inclusion of additional patients intrials. If both symptomatic knees were given thesame treatment (active or placebo) this wouldhopefully decrease the drop-out rate, and mightreduce the numbers of patients required for study.

, If comparison is being made between an oralanalgesic drug (which may benefit both knees)and an intraarticular drug/device (which willsolely or predominantly influence just one knee)there could be bias in favour of the oralcompound. Equal intra-articular treatment ofboth knees would redress this bias.

, Is it rational to recommend a pooled index forthe small joints of one or both hands but not forboth knees or both hips? Each knee has threecompartments that may contribute to intracap-sular pain, so that pooling of information on oneknee has already occurred.

, Symptom causation in knee OA may result fromvarious mechanisms. Should we aim for consist-ency, at least in differentiating intracapsularfrom periarticular pain through simple examin-ation techniques?

Inclusion of two symptomatic knees (or hips) inassessment and in treatment could, therefore:facilitate recruitment; reduce drop-out rates;eliminate certain bias in between treatment com-parisons; and produce more generalizable, ‘realworld’ treatment data. Stratification for one or twojoint involvement could easily be included inrandomization. Intra-articular treatments shouldobviously be the same (treatment or placebo) forthe two knees. Because multiple joints from onesubject cannot be treated as independent variables(‘two knees, one person’ consideration [6, 7])outcomes would need either to be expressed as asingle score for one individual, or more complexanalytical techniques are required [6, 7]. If data ispooled, this might involve: use of single patientresponses for pain and function for both knees indisease-specific instruments (eg the WOMAC orLesquesne index [1]); assessments could be under-taken for both sides and then pooled into a singlemean score, or a joint chosen at random. For moregeneric arthritis, health status and quality oflife/utility instruments laterality is not an issue.

Should there be restricted radiographic entrycriteria?

It is clearly reasonable to require definiteradiographic change at the site of interest to

confirm a diagnosis of structural ‘OA’ [1, 2].However, marked structural change on X-ray isregarded as an exclusion for clinical trials of kneeor hip OA [1, 2], and at the knee the (medial)tibiofemoral compartment is often given soleprominence with no consideration of the patello-femoral compartment [1].

It is clearly important to select patients withmild–moderate radiographic change for study ofstructure modifying drugs, so that progression ofjoint space narrowing can be assessed. Forsymptom modifying drugs the situation is different.The frequent discordance between X-ray changesand symptoms in OA, even at the knee and hip, iswell recognized [8]. The premise that inclusion ofsevere structural OA is an unfair test ofsymptom-modifying drugs [2] requires testing.Furthermore, the patello-femoral compartment is acommon target site for OA and probably a commonsite for symptom causation [10]. Studies that focussolely on anteroposterior tibiofemoral X-rays mayin fact be including patients with severepatellofemoral structural change, even thoughthey are selecting for mild–moderate tibiofemoralOA. If the primary outcome measure is pain,assessed over a period as short as four months,should we unduly worry about the degree orlocation of X-ray change? Such restriction limitsrecruitment, especially from hospital-based clinics.Furthermore, data are produced that relate only tomild–modest structural OA, reducing their gener-alizability to the OA population. Inclusion of anyX-ray grade, with assessment of all three kneecompartments, would increase recruitment and thegeneralizability of the study data. Severity andlocation of radiographic change might then beexamined as a possible predictor of response. Wewould favour a broad rather than narrowdefinition of patient radiographic eligibility [1].

The most important entry criterion, and primaryoutcome measure, for studies of symptom-modify-ing drugs should be the level of reported pain.Although the lower limit required is defined( r 25 mm on a 100 mm visual analogue scale, orr 1 on a 5 point categorical scale [1]), the upperlimit also requires definition—very high painscores are biased to change only in a downwarddirection towards improvement. We would favorentry levels in the range of r 25– R 75 mm (100 mmvisual analogue scale), or r 1– R 4 (5 point ordinalscale).

How should study results be expressed?

Current guidelines are careful not to define aminimum clinically important response in

Osteoarthritis and Cartilage Vol. 6 No. 6 373

symptom-modifying drug trials. Although someinvestigators do give their own definition of apositive clinical response (often an improvement ofq 20 mm on a 100 mm visual analogue painscale—the same effect size that is often used inpower calculations) variability in definitions andoutcome measures impedes comparison betweenstudies. In the absence of response criteria, studyresults are usually expressed as relative benefit ofthe treatment group compared to the placebogroup (i.e., comparisons of means or medians).However, even if the mean difference between atreatment and a control group is less than thesmallest change that is considered important, thetreatment could have had an important impact onmany of the study patients. For this reason it is anadvantage to express results also in terms of the‘number needed to treat’ to obtain good benefit[10, 11]. Despite certain caveats, agreed site-specific response criteria for symptom relief of OAwould greatly facilitate such reporting and assistbetween study comparisons of effect size oftreatments.

In line with other biomedical journals, editorsof most rheumatology journals now conform tothe CONSORT (Consolidation of Standards forReporting Trials) format for the reporting ofclinical trials [12]. This requires explicit detailson all important aspects of a randomized con-trolled trial. If adhered to, such presentation willenable the data to be appropriately judged. It willalso encourage good quality scores (13) andinclusion of the data in any systematic review.Guidelines on OA studies should support suchpresentation.

These and other practical issues require contin-uing discussion. There is no single correctapproach but Osteoarthritis and Cartilage is anideal forum for such debate. Guidelines are neverset in stone and regular review and revision is partof the undertaking of the groups that publish them.The participation of those that will have toimplement them is vital in this process.

References1. A Task Force of the Osteoarthritis Research Society.

Design and conduct of clinical trials in patientswith osteoarthritis. Osteoarthritis Cart 1996;4:217–43.

2. Group for the Respect of Ethics and Excellencein Science (GREES). Recommendations for theregistration of drugs used in the treatmentof osteoarthritis. Ann Rheum Dis 1996;55:552–7.

3. Bellamy N, Kirwan J, Boers M, Brooks P, Strand V,Tugwell P, et al. Recommendations for a core setof outcome measures for future phase III clinicaltrials in knee, hip, and hand osteoarthritis.Consensus development at OMERACT III. JRheumatol 1997;24:799–802.

4. Felson DT, Anderson JJ, Boers M, Bombardier C,Chernoff M, Fried B, et al. The American Collegeof Rheumatology preliminary core set of diseaseactivity measures for rheumatoid arthritis clinicaltrials. Arthritis Rheum 1993;36:729–40.

5. Creamer P, Hunt M, Dieppe P. Pain mechanisms inosteoarthritis of the knee: effect of intraarticularanesthetic. J Rheumatol 1996;23:1031–6.

6. Zhang Y, Glynn RJ, Felson DT. Musculoskeletaldisease research: should we analyze the joint orthe person? J Rheumatol 1996;23:1130–4.

7. Sutton AJ, Muir KR, Jones AC. Two knees or oneperson: data analysis strategies for paired joints ororgans. Ann Rheum Dis 1997;56:401–2.

8. Hadler NM. Knee pain is the malady—not osteo-arthritis. Ann Intern Med 1992;116:598–9.

9. Cicuttini FM, Baker J, Hart DJ, Spector TD.Association of pain with radiological changes indifferent compartments and views of the kneejoint. Osteoarthritis Cart 1996;4:143–7.

10. Cook RJ, Sackett DL. The number needed to treat:a clinically useful measure of treatment effect.BMJ 1995;3:452–4.

11. Laupacis A, Sackett DL, Roberts RS. An assessmentof clinically useful measures of the consequencesof treatment. N Engl J Med 1998;318:1728–33.

12. Begg CB, Cho M, Eastwood S, Horton R, Moher D,Lokin I, et al. Improving the quality of reportingof randomised controlled trials. The CONSORTstatement. JAMA 1996;276:637–9.

13. Jadad AR, Moore RA, Carroll D, Jenkinson C,Reynolds DJM, Gavaghan DJ, et al. Assessingthe quality of reports of randomised clinicaltrials: is blinding necessary? Cont Clin Trials1996;17:1–12.