We showed that in 78 subjects with symptomatic knee OA, the rate of tibial cartilage loss was ongoing beyond 2 years, at a rate of between 3.1% and 4.8% per year over 4.5 years. Higher initial pain scores were associated with diminished medial and total tibial cartilage loss. There was a moderately strong positive correlation between the percentage loss of medial and lateral tibial cartilage. In addition, the use of the random coefficient modelling structure enabled an assessment of the relationship between the true rate of cartilage loss and initial cartilage volume – an assessment that is plagued by regression to the mean when using only the observed rates of change and initial cartilage volumes in the sample. For example, the finding for the medial compartment that the relationship between initial volume and rate of change was diminished after multivariate adjustment would not have been obtained if the simple observed rates of change had been regressed on the observed initial volumes, and a spurious effect of initial volume as a predictor of change would have emerged.
This is the first study to examine whether average annual cartilage loss in subjects with symptomatic knee OA is similar over the medium term, with biannual loss stable over two periods. We found the average total tibial cartilage loss to lie between 3.1% and 4.8% per year over 4.5 years. These results are consistent with the magnitude of loss observed in previous studies that examined subjects over 2 years [10
]. Over 2 years, in the initial cohort of 123 subjects with symptomatic knee OA, we showed the annual rate of loss of total tibial cartilage to lie between 4.4% and 6.2% [10
]. Over 24 months, another group, examining 32 subjects, found similar results, with the annual total tibial cartilage loss between 2.2% and 6.6% [11
]. The only study examining change in 11 subjects over about 3 years found no mean significant change over this time period [24
]. However, the 95% confidence intervals for annual medial tibial cartilage change included a loss of 3.9% per year to a gain of 3.0% per year in that study. We found that the distribution of rate of change was normal over 2 years, and also at 4.5 years [10
]. The sensitivity to change of the MRI volume measurements over the various periods using the Standardized Response Mean index [25
] can be obtained simply from Table by dividing the average change by the SD of the changes. These are of the order of 0.50, 0.75 and 0.90 for the periods 0 to 2 years, 2 to 5 years and 0 to 5 years, indicating that MRI volume measures display increased sensitivity to change with longer follow-up of subjects.
The moderately strong correlations seen between cartilage volume measured at each time point suggests that tracking of individual change in cartilage volume occurs in those with knee OA. This means that subjects maintain their relative ranking over time in terms of cartilage volume, compared with other subjects within the study. This phenomenon has been described in adult women aged 30 to 94 years, regarding tracking of bone mineral density [26
]. It has also been shown to occur in children and adolescents with regard to height and also to the accrual of bone mineral content [27
]. It has been suggested that this is under genetic control, unless strong environmental factors intervene. Similarly, change in cartilage volume has also been shown to have strong genetic determinants in healthy adult children of those who have had knee arthroplasties [28
Our results for predicted rates of change for individuals suggested that only one subject gained lateral cartilage volume beyond prediction uncertainty over the course of the study. It is possible that this increase in volume was related to a gain of cartilage or it might have been due to swelling related to early OA in that compartment [29
]. Quantification of cartilage volume is unable to differentiate between these two possibilities.
Some cross-sectional studies have suggested an effect of age on cartilage volume [8
], although others have not [31
]. Only a longitudinal study, such as this, is able to differentiate between the within-individual cartilage loss of individuals (for instance, expected change in one person over one year) as they age and the cross-sectional effects of age across different subgroups of people (for instance, comparisons between people who differ by one year of age only). The average rate of medial cartilage loss within an individual over 1 year was estimated as 61.2 μm3
/year (longitudinal effect of age), whereas the difference in average volumes of two groups of people differing by 1 year in age (cross-sectional effect of age) is expected to be 6.6 μm3
(Figure ). The longitudinal rate of loss in lateral tibial cartilage was 71.3 μm3
/year compared with the cross-sectional difference of 14.2 μm3
per year of age. This illustrates that, in this study population, the longitudinal rates of change are far greater than cross-sectional differences, which is similar to findings relating to bone mineral density [32
]. This suggests that interpreting cross-sectional age effects in studies with similar populations as representing change in an individual may be unhelpful and misleading, The implication is that classification systems based on these assumptions may be flawed.
There is increasing evidence that biomechanical effects are important in the progression of knee OA [34
]. We found lower levels of knee pain to be associated with higher rates of cartilage loss in the medial tibial cartilage. This contrasts to previous findings, with our initial report of this cohort (123 subjects) [10
] and a similar study (110 subjects) [37
], both over 2 years, of no effect of baseline pain over 2 years. The present study took place over 4.5 years and showed that for every increase in WOMAC pain score of 10 (of a maximum of 500), there was an reduction in annual cartilage loss of about 10% (6.7 mm3
/year). The previous studies, over a shorter period, may not have had power to show this effect. Our findings may be explained by subjects with painful knee OA adapting their gait in response to pain, to reduce pain, by reducing external adduction moment [38
]. This reduces compressive force on the cartilage, which may reduce cartilage loss [1
]. Conversely, pain reduction increases external adduction moment, which has been associated with increased disease progression [34
]. Joint loading has been shown to affect articular cartilage in knee OA [34
The main study limitation relates to the loss to follow-up of 27 subjects, with a completion rate of 74%. The subjects who did not undergo knee joint replacement surgery and were lost to follow-up were similar to those who completed the study, including disease severity, and lost cartilage at a similar rate during the first 2 years of the study. Indeed, of the 123 subjects who participated in the initial study, 63% underwent MRI at the 4.5 years follow-up. The 18 who underwent knee joint replacement have already been shown to have lost cartilage more rapidly than those who did not undergo joint replacement [40
]. It is possible that our study might have underestimated somewhat the average rate of cartilage loss in all subjects with OA, given the exclusion of subjects who underwent knee replacement over the study period, who are known to lose cartilage more rapidly [40
]. The presence of swelling in early OA may also reduce the cartilage loss seen.
We found few predictors of rate of change in cartilage volume. This may be because there is a strong genetic component to cartilage loss [28
] or may simply reflect a lack of power related to limited sample size, reducing our ability to detect modest determinants of change. Other factors such as meniscal pathology and cruciate ligament integrity have been shown to affect the incidence of OA and subsequent cartilage loss [41
]. We were unable to examine for these pathologies because of the limited MRI sequences that were used in this study.
Another limitation is that the model was restricted to assuming that change was linear because only three time points were available. This may be a reasonable assumption over a short period (less than five years) because we found no evidence to the contrary. However, the assessment of linearity in this study was complicated by the large amount of variability observed in cartilage volume measurements for individuals around their predicted change. Observing a greater number of measurements over time will be needed to confirm whether the pattern of cartilage loss is linear. The within-individual variability in cartilage volumes not accounted for by linear loss may be due to other factors that may affect change in cartilage volume varying with time, such as effusion, trauma, inflammatory change and pain. It may also be that loss is not truly linear, in contrast with reduction in bone mineral density [26
], but shows mild exponential 'decay', with reduced activity. Finally, despite our attempts to minimise measurement error by means of reader training and blinding, it is possible that this still represents a substantial component of the unexplained within-individual variability.