The OAI is a longitudinal cohort study that has among its aims to assess the validity of MRI-derived morphology biomarkers for OA disease progression. We used state-of-the-art analysis methods to analyse numerous morphological biomarkers in articular cartilage from MRI. This paper presents descriptive results of cartilage morphometry and its change at the 1-year time point from the first substantive MRI data release from the OAI progression subcohort. In general, the annualised rates of change are small with the central medial femur showing the greatest consistent change.
The longitudinal change from the participants in this study is smaller than the majority of values published in previous longitudinal studies, which typically demonstrated about 5% loss of knee cartilage volume per year (range of about −1 to −8%) in knees with OA5
This then raises questions regarding differences between the study protocols or analysis technologies, which may explain very different rates of progression. The methods in previous studies differ markedly from those described here, including that they use a completely manual method of tracing boundaries for segmentation, analysis is read unblinded to time point order, have acquired images using 1.5T scanners and a different pulse sequence to that used here. However, our results are consistent with recent data from other studies using similar cartilage quantification techniques that showed cartilage volume loss of about −1 to −3% per year (including MAK,17
and more recently data by Eckstein et al
These more recent studies have generally found rates of loss similar to what we have found in the medial tibia and femoral plates. If these more recent estimates of cartilage volume change are confirmed, then there are important implications for future clinical trials of disease-modifying treatments for OA using MRI techniques.
While significance tests are frequently used to assess change they do not indicate the magnitude of change. To give greater meaning to the amount of change the concept of responsiveness was introduced. The term responsiveness is used as an indicator of sensitivity to change. There are many responsiveness indicators that result in different effect size indices. Most of these indicators agree on the numerator (change from baseline to follow-up); however, there is little agreement on the appropriate denominator. Effect sizes greater than 0.2 are generally held to be clinically detectable.20
came up with conventions for these values that constitute a ‘‘trivial’’ (ES <0.20), small (ES ≥0.20 <0.50), medium (ES ≥0.50 <0.80), and a large effect (ES ≥0.80). There is more controversy over the interpretation of other indices of responsiveness, including SRMs; however, if we were to generalise, the SRMs in this analysis are trivial to small.
The percentage change is calculated as change divided by baseline measure, so change and percentage change always have the same sign for each individual. However, their means in the population may not necessarily have the same sign. When the mean of change was calculated, each subject’s change was weighted with 1/sample size and summarised. However, when the mean of percentage change was calculated, each subject’s change was weighted with 1/(baseline measure×sample size) and summarised.
Several structure modification studies have based their sample size estimates on MRI-based rates of volume change of about 5% with an SD of 5% per year in knee cartilage volume. Projected sample size depends on: (1) the expected rate of progression in participants treated with placebo; (2) the minimum magnitude of the drug effect, or rate of progression expected in the active treatment arm(s); (3) the variation in progression rate that occurs between participants; and (4) the precision of the measurement technique. In a simple analysis of change from baseline, to detect a 50% reduction in loss of baseline cartilage volume over 1 year requires evaluable data on 64 participants per arm if the expected background progression is 5% (SD 5%), but 250 per arm if it is 5% (SD 10%) (for 80% power, a (two-sided) 0.05).9
A modest within-subject correlation of ρ = 0.7 would reduce the sample size by a factor of 2.3. However, change on the order of 1% (SD 10%) as observed in this study would require a prohibitively large sample size (N~6400/arm) until the correlation reaches values on the order of 0.99. If these estimates of cartilage change are confirmed, then for MRI to be a useful tool with which to study OA progression then it will be necessary to develop more sensitive algorithms to detect structural change in the joint, and identify study populations undergoing more rapid disease progression.
How can we explain that the cartilage volume of some regions remains the same while the denuded areas appear to be increasing? Cartilage volume and thickness changes aggregate areas of cartilage swelling with separate areas experiencing a reduction in thickness, reducing the ability of these summary measures to identify change. In early OA, cartilage may not be thin but instead is thicker and swollen with water, which is imbibed by cartilage when the collagen network is disrupted and the role of proteoglycans is altered.21 22
Increasing thickness may also reflect a healthy trophic response to focal loading for normal cartilage as distinct from early disease. Thus measuring cartilage volume or mean thickness in regions of the knee (eg, medial tibia) and regional mean thickness may provide a very different measure of important pathological change when compared with focal measures of change centred around focal defects in diseased joints.23
Further, these measures cannot assess the composition of cartilage that can be measured using MRI techniques to ascertain alteration in proteoglycan and collagen content that may accompany swelling of cartilage.24
Distinguishing the MRI measures that are the most sensitive to change and are correlated with clinical symptoms is essential if we are to utilise them appropriately. Future work with this data set will investigate the association of MRI morphology measures with clinical symptoms. Moreover, we plan to estimate the cartilage loss involving partial and complete thickness focal cartilage defects.
As seen in previous studies, the variability in femoral cartilage volume measures (especially the posterior femoral condyles) is greater than other knee compartments potentially reflecting difficulties with lack of contrast among tissues in this region or due to partial volume effects in this curved surface.
This study has several limitations. The MRI postprocessing technique used every other slice as opposed to every slice. We anticipate a slight improvement in precision if each slice is used in the analysis. Another potential limitation is that we quantified cartilage morphometry using the DESSwe sequence rather than the more standard FLASH sequence, although the former has been cross-validated with the latter.25
The automated, pairwise image segmentation process we used imposes a bias on cartilage thickness and volume measurements, and is a unique feature of our methodology. Paired image analysis is typically more precise than unpaired image analysis, and biases imposed by these processes need to be presented and accounted for in analysis. The relative advantages of our analysis methods will require independent segmentation and quantification of these images by alternative image analysis techniques.
The separation of the cartilage tissue into several compartments introduces noise into the region of interest measurements. Without this subdivision it is impossible to report localised changes in the central femur, and the analysis, therefore, the ability to detect changes will be minimised. On the other hand, the segmentation of cartilage plates was done using an automated process that only increases marginally the measurement error.
The original description of the K&L grade was made and developed on weight bearing, fully extended films not on films that were semi-flexed such as in this study.
The sample used in this analysis had a smaller cartilage change than reported in other populations. One likely explanation is that cartilage change is greater in subjects with more advanced OA. 24 of 150 (16%) of our sample were participants with K&L grade 0 and 1, effectively a population without definitive radiographic OA (but with persistent knee pain). In other studies the greatest change in cartilage morphology has been seen in those with K&L grade 3 disease. Further, the sample in this study is heterogeneous with respect to whether the medial or lateral compartment is primarily involved in OA. Mediolateral frontal knee alignment has been shown to be a significant risk factor for disease progression in the primarily loaded tibiofemoral knee compartment.26
We plan to conduct further studies of this population and evaluate whether enriching the sample for certain features that predict progression such as meniscal damage, bone marrow lesion and alignment will facilitate identification of a cohort of persons at risk for greater progression.
In summary, the rates of change of cartilage morphometry in people with knee OA in this study are small. These results need to be replicated and published by other investigators using other image analysis tools and software algorithms. Our results should stimulate discussion as to which MRI parameters should be measured in longitudinal studies of OA progression and how best to perform these measurements. The greater complexity and cost of MRI joint morphology will need to prove its value.