During the last few years, major advances have been made in the treatment of osteoarthritis.¹ The current management includes non-pharmacological and non-surgical measures, the use of pharmacological agents, and surgery.^2,3 According to recommendations from expert panels and regulatory authorities, drugs used for the treatment of osteoarthritis can be classified as symptom modifying or structure modifying agents, depending on their ability to control the symptoms of the disease or the progression of joint structural changes.^2–,⁵ Structural modification of the joint is considered to be the most important determinant of disease progression. Structure modifying agents have thus been assimilated into disease modifying agents. Change in joint space width, measured on standard x rays, is currently the recommended primary end point in trials assessing new chemical agents for treating osteoarthritis, for both scientific organisations^5,6 and regulatory agencies in the USA and Europe.^2,3

However, to date, very few prospective long term studies have confirmed the surrogacy of joint space narrowing for relevant hard clinical end points—that is, the long term incidence of osteoarthritis related joint surgery. Subsequently, the exact clinical relevance of the results obtained in studies showing the ability of an anti-osteoarthritis drugs to delay, reduce, or prevent joint space narrowing remains equivocal. Knee surgery is widely recognised as the single most relevant outcome in knee osteoarthritis. The need for such a procedure usually reflects the failure of non-surgical treatment and it is only undertaken in patients with painful and debilitating disease.^7,8

Our aim in the present study was to confirm prospectively the long term clinical relevance of joint space narrowing—that is, the ability of this radiological feature to predict the risk of future knee surgery.

METHODS

RESULTS

Of 139 patients who completed the three year initial trial, 13 (9.3%) could not be reached at the time of the follow up evaluation. Their baseline demographic characteristics did not differ from those of the overall cohort (table 1⇓). This left a total of 126 patients (90.7%) assessable in the present study. During a median follow up of 5.0 years (mean 3.8, maximum 8.0) after the end of the original study, 16 patients (12.7%) had received knee surgery. These surgical procedures included 11 total knee replacements (68.8%) and five joint debridement/meniscectomies (31.2%). While some patients had surgery in both knees, this intervention occurred more often (in over 60%) in the target knee in the initial study.

The areas under ROC curves measuring the respective value of mean and minimum JSN for the prediction of the incidence of knee surgery are reported in fig 1⇓. Minimum (p = 0.006), but not mean (p = 0.51), JSN after three years was highly predictive of the risk of requiring osteoarthritis related knee surgery within a further period of five years.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

 Area under the receiver operating characteristic (ROC) curves evaluating the predictive value of mean and minimum joint space narrowing over three years for future knee surgery within five more years.

The relative risk of undergoing knee surgery, and the specificity, sensitivity, and positive and negative predictive values for various cut off points of mean or minimum JSN, are presented in tables 2⇓ and 3⇓. While a cut off value of 0.7 mm for minimum JSN after three years provides the best numerical overall efficiency for predicting the incidence of future knee surgery, no meaningful differences were observed for cut off values between 0.5 and 0.8 mm (relative risks for future surgery all above 4.5, and all overall efficiencies above 70%).

DISCUSSION

Whereas clinical outcomes for the assessment of symptomatic improvement in osteoarthritis of the lower limb (that is, the Western Ontario and McMaster Universities index, the Lequesne algo-functional index, and so on) are well established, there is still debate over the most appropriate way of assessing a drug aimed at reducing, delaying, or preventing the structural progression of osteoarthritis. There is a large consensus that the incidence of lower limb surgery is the most relevant clinical outcome, but on practical grounds the assessment of this particular end point would request studies involving massive cohorts of patients over several years.⁶ Furthermore, the decision process for osteoarthritis related surgery may be affected by various non-medical conditions, including sex, race, and socioeconomic status.¹³ In fact, there are no widely accepted guidelines defining the exact profile of a patient undergoing osteoarthritis related surgery.¹⁴ All these elements make the design of a clinical trial to assess a potential structure modifying agent in osteoarthritis on the basis of the incidence of lower limb osteoarthritis surgery at least hazardous, and at any rate impractical.

There is now a consensus within scientific organisations and regulatory authorities for using radiographic changes—that is, JSN over a period of two to three years—as a surrogate for a hard clinical end point.^2,3,6 However, as of today no long term prospective clinical study has ever unequivocally confirmed the surrogacy of JSN at the knee for lower limb surgery. To be fully validated, a surrogate end point should show a natural course closely related to that of the relevant end point (thus a decrease in JSW would be linked to an increase in the incidence of joint surgery). The magnitude of the change in the surrogate responsible for a subsequent increase or decrease in the end point should be clearly defined (for example, a 0.5 mm decrease in JSW would be linked to a twofold increase in the risk of joint surgery). Eventually, this numerical relation should also be validated following an external intervention, such as a reduction in JSN by 50% resulting in a 25% decrease in the incidence of joint surgery. In the present study, we show for the first time from a long term prospective study that the assessment of minimum JSN fulfils the two first requirements of the validation of this particular surrogate end point.

Previous studies have suggested that a 0.5 mm threshold of JSN at the hip would be an appropriate cut off value, mainly because this value corresponded to the smallest difference between two measurements of JSW that exceeded the measurement error.^15,16 A decrease of 0.2 mm and 0.4 mm in hip JSW after one or two years has also been significantly associated with an increased risk of further hip arthroplasty after five years, with corresponding ranges of sensitivity and specificity of 68–75% and 67–78%, respectively.¹⁷ The figures reported in the present study for a minimum JSN of the knee between 0.5 and 0.8 mm after three years (overall efficiency of between 73% and 75% for predicting the occurrence of knee surgery five years later) compare favourably with these results. On an individual basis, the negative predictive values of our suggested thresholds (that is, the probability of surgery in a patient with JSN below the cut off value) are high (92–94%), but the positive predictive values (the probability of surgery in a patient with JSN above the cut off value) remains marginal (25–30%). When we look only at total knee replacement, the negative and positive predictive values of our suggested thresholds are 94–95% and 13–18%, respectively. Thus our results re-emphasise the fact that clinical considerations may be more important that radiological features in the discussion process to undergo surgery.

We also, for the first time, provide a head to head comparison of the respective value of mean and minimum JSN assessment as predictor for the risk of future knee surgery. While cut off values of minimum JSN between 0.5 and 0.8 mm after three years are all linked to a four- to fivefold increase in the risk of future knee surgery (p = 0.003 to 0.004), none of the selected thresholds of mean JSN (0.2 to 0.8 mm) was significantly related to the clinical outcome (p values between 0.09 and 0.49). We have no definite explanation for such a finding. It has previously been reported that although the minimum joint space width could be more sensitive for identifying changes in cartilage than the mean joint space width, the latter is less sensitive to the influence of variations in radiographic procedures and patient positioning. Recently, the rate of tibial cartilage loss over two years—assessed by magnetic resonance imaging—has been found to be an independent predictor of knee replacement at four years.¹⁸ We suggest that minimum joint space width changes are significantly correlated with tibial cartilage volume changes. However, more studies are needed to confirm these particular findings.

We also show in the present study that, even though other radiographic views of the knee joint might be more appropriate today than the fully extended approach, this particular view remains of significant interest as the radiographic changes observed with this technique after three years are significantly predictive of future knee surgery.