As far as we know, this study provides the first detailed longitudinal data from a large sample describing the natural history of individual BMLs in articular subregions, as well as at the level of the knee as a whole, and relates different types of longitudinal change in individual BMLs to the risk of cartilage loss in the same subregion of the articular surface. We found that over 30 months a substantial proportion of existing BMLs decreased in size (many resolving completely) while increase in size, including development of new BMLs, was also common. Most knees with prevalent BMLs at baseline showed either no change or changes in only one direction (increase or decrease), but a smaller proportion of knees simultaneously displayed combinations of stable, increasing and decreasing BMLs. In comparison with subregions with BMLs that were stable during the study, the risk of cartilage loss was substantially increased for enlarging or new BMLs. The risk of cartilage loss was substantially decreased when a subregion had no BMLs at both baseline and follow-up. Subregions in which BMLs decreased in size or resolved had the same risk of cartilage loss as those with stable BMLs.
Studying the natural history of BMLs, we found that 66% of the pre-existing BMLs showed a change in size over 30 months. This is a higher percentage than that obtained by the GARP study that reported change in only 24% of pre-existing BMLs when analysing individual lesions.11
In particular, we found that the proportion of baseline BMLs that decreased in size or disappeared (about 50%) was strikingly higher than previously reported.4 11
Hunter et al
reported regression in only 0.6% of their subjects; one can only speculate on the reasons for this discrepant finding in comparison with our results. The subjects in the BOKS study had symptomatic knee OA, were predominantly male and older than our cohort. Cartilage scores were summed for the medial and lateral tibiofemoral compartments in the BOKS publication and it is not clear if compartmental analysis was also applied to BMLs or if individual lesions were assessed.
We also evaluated prevalence and changes over time in individual BMLs summarised over the whole knee. Although our sample was heterogeneous and included knees with and without OA, the proportion of knees with a BML at baseline (65%) was similar to that in studies of patients with symptomatic OA (57–75%), consistent with the high risk profile, including obesity, of our sample.4 11 12
We found that about one-quarter of knees with any BML at baseline simultaneously displayed combinations of stable, increasing and decreasing BMLs at follow-up. The importance of this finding remains unclear, but it lends support to the idea that BMLs are closely linked to highly localised biomechanical stresses and changes in the integrity of the adjacent cartilage.
This study confirms previous findings that prevalent subchondral lesions of the bone marrow are a predictor of cartilage loss.4 17
A cross-sectional MRI study found that higher grades of cartilage defects are associated with higher prevalence and larger size of adjacent subchondral BMLs.7
In comparison with subregions with stable BMLs, subregions with progressive and new BMLs showed a significantly higher risk of cartilage loss in the same subregion at follow-up MRI. Subregions with regression and resolution of BMLs exhibited the same risk of cartilage loss as the subregions with stable BMLs. One possible explanation for this finding is the relatively short follow-up interval. A longer observational period might clarify whether persistent regression of BMLs is associated with a lower risk of adjacent cartilage loss than size-stable lesions. Absence of BMLs at baseline and follow-up was associated with a decreased risk of cartilage loss.
As the main aim of the study was the structural relation between BMLs and adjacent cartilage over time we did not include data on pain or function. Felson and coworkers showed recently that development of knee pain is associated with an increase in BMLs as revealed on MRI.18
Whether or not regressing or resolving BMLs are associated with an improvement in pain or function remains to be shown.
Our study has a number of limitations that need mentioning.
- Concerns have been raised that image quality of 1.0 T extremity systems might be inferior to commonly used 1.5 T or 3.0 T systems. For this reason a validation exercise was performed, which showed good agreement, sensitivity and specificity for cartilage and BML scoring. Thus, we believe that image quality in our study was adequate to obtain reliable readings.
- As the main aim of our study was to assess change in BMLs and cartilage over time, the MRIs were not read blinded to time point. This might result in a slight tendency to read more change in comparison with a blinded reading. However, it has been shown that scoring without knowing the chronological sequence substantially decreases sensitivity in the detection of clinically relevant changes in comparison with scoring in chronological order.19 20 These studies showed that blinding to time point can lead to misclassification of the longitudinal change in a feature and that it may compromise the assessment of the relation of that feature and its outcome. However, we have to acknowledge that to date longitudinal OA studies comparing semiquantitative MRI assessment blinded and non-blinded to chronological order have not been carried out.
- Unfortunately, semiquantitative MRI assessment of OA does not allow for separate assessment of two adjacent features such as BMLs and cartilage as both features are visualised within the same image and one feature cannot be blinded for the other.
- It is difficult to prove that BMLs actually precede cartilage loss. We could prove a strong association between the new development of BMLs and cartilage loss at follow-up, but the chronological order of these structural changes remains unclear. Only repeated examinations with markedly shorter time intervals might ultimately answer these questions. Results of a recently published animal study investigating the development of cartilage defects in a small number of cruciate-deficient dogs suggest that newly developing cartilage defects are preceded by BMLs in the same subregions.21 Whether progressive cartilage loss predicts incident or progressive BMLs should be tested in future studies.
- Previous studies have shown that the presence of BMLs may be associated with loading.4 8 22 We adjusted our analysis for BMI as one of the risk factors of structural OA progression. Malalignment is another important predictor of structural progression in OA, and in the BOKS study explained the association of progression of BMLs and cartilage loss. However, the direct subregional association of BML status at baseline and follow-up with cartilage status at follow-up will persist even when malalignment might be the actual trigger for progression.
- Finally, there are radiological differential diagnoses for subchondral BMLs, the most common being traumatic bone contusion.23 Although histologically similar, bone contusions should be regarded as distinct from non-traumatic BMLs in conjunction with cartilage defects and OA.10 24 We excluded knees with unequivocal radiological differential diagnoses before analysis.
In summary, we found that subchondral BMLs, unlike cartilage, are a highly variable feature in patients with knee OA or at risk of developing it and that the majority of prevalent BMLs showed regression or resolution over time, which has not been reported before. Second, absence of BMLs was associated with a decreased risk of adjacent cartilage loss while subregions with new and progressive BMLs exhibited a high risk of cartilage loss at follow-up. Whether BML change might become a target for treatment remains to be shown but the underlying mechanical abnormalities such as knee mechanical alignment or meniscal pathology influencing cartilage loss and BMLs need to be fully understood.