Correlation of the development of knee pain with enlarging bone marrow lesions on magnetic resonance imaging




Results of cross-sectional studies have suggested that bone marrow lesions (BMLs) visualized on magnetic resonance imaging (MRI) are related to knee pain, but no longitudinal studies have been done. This study was undertaken to determine whether enlarging BMLs are associated with new knee pain.


Subjects ages 50–79 years with knee osteoarthritis (OA) or at high risk of knee OA were asked twice at baseline about the presence of knee pain, aching, or stiffness (classified as frequent knee pain) on most days; absence of knee pain was the baseline eligibility criterion. At 15 months' followup, subjects were again queried twice about frequent knee pain. A case knee was defined as absence of knee pain at baseline but presence of knee pain both times at followup. Controls were selected randomly from among knees with absence of pain at baseline. All MR images were scored for volume of BMLs in the medial, lateral, and patellofemoral compartments. We focused on the maximal change in BML score among the knee compartments from baseline to 15 months. Multiple logistic regression, with adjustments for demographic and clinical variables, was used to assess whether an increased BML score is predictive of the development of knee pain.


Among case knees, 54 of 110 (49.1%) showed an increase in BML score within a compartment, whereas only 59 of 220 control knees (26.8%) showed an increase (P < 0.001 by chi-square test). A BML score increase of at least 2 units was much more common in case knees than in control knees (27.5% versus 8.6%; adjusted odds ratio 3.2, 95% confidence interval 1.5–6.8). Among case knees with increased BMLs, most already had BMLs at baseline, with enlarging BMLs at followup, but among the subset of knees with no BMLs at baseline, new BMLs were more common in case knees (11 [32.4%] of 34) than in control knees (9 [10.8%] of 83).


Development of knee pain is associated with an increase in BMLs as revealed on MRI.

Approximately one-fourth of adults ages ≥55 years have frequent knee pain (1). Although many of these individuals have a diagnosis of knee osteoarthritis (OA), at least half do not. Furthermore, even in those subjects with knee OA, the source of the joint pain is unknown.

Whereas cartilage is aneural, bone is richly innervated and the involvement of bone in the pathologic features of OA is well recognized (2–4). Because of this innervation, bone could be a source of pain in at least some subjects with knee OA. In a previous cross-sectional study, we compared the findings on magnetic resonance imaging (MRI) of the knee in subjects with pain-free knee OA with the findings in subjects with knee pain and radiographic knee OA (5). We found that the subjects with knee pain were much more likely to have bone marrow lesions (BMLs) (5).

Subsequent studies of the association of BMLs with knee pain, all of which have been cross-sectional, have either corroborated our initial findings (6) or contradicted them (7, 8). This has left uncertainty as to the relationship between BMLs and knee pain. The differing study findings could also be attributed to differences in populations, since study subjects have ranged from a cohort of older, mostly male subjects to a cohort of women in midlife. BMLs could be a source of pain in some individuals with knee OA who report having knee pain, but BMLs are not the source of the pain in all individuals. Therefore, the varying results depend on the proportion of patients whose knee pain could be secondary to the presence of BMLs.

One problem with all of these studies is that, because of their cross-sectional design, study investigators rely on comparisons among subjects to make inferences. There is individual variability in pain thresholds and differing levels of disease severity between subjects, and this variability may make it difficult to detect the source of the joint pain. The relationship of a structural feature with knee pain is better investigated by examining disease-related changes within an individual subject. Evidence to demonstrate a relationship between BMLs and knee pain would be stronger if subjects with no knee pain who then developed pain also had a concurrent increase in BMLs. We therefore performed a study of subjects who initially had no knee pain but who were at high risk of developing knee pain over 15 months of followup.


Study design and subjects.

The Multicenter Osteoarthritis Study (MOST) was a prospective epidemiologic study of individuals ages 50–79 years, in which the goal was to identify risk factors for incident symptomatic knee OA and progressive knee OA. Study subjects either had knee OA or were at high risk of developing the disease. Factors considered to contribute to a high risk of knee OA included being overweight or meeting the criteria for obesity, having knee pain, aching, or stiffness on most of the preceding 30 days, a prior knee injury that made it difficult to walk for at least 1 week, or previous knee surgery. Weight eligibility for MOST was defined as having a body weight higher than the median weight for each age- and sex-specific group based on the data from the Framingham OA Study (9). For example, the body weight cutoff values were as follows: in women, ages 50–59 years, 154 pounds, ages 60–69 years, 151 pounds, and ages 70–79 years, 148 pounds; in men, ages 50–59 years, 194 pounds, ages 60–69 years, 187 pounds, and ages 70–79 years, 182 pounds. For body weight determinations, each subject was weighed without shoes and heavy jewelry and wore a standard gown or lightweight clothing. In addition, each subject's height (without shoes) was measured using a stadiometer.

All MOST subjects were recruited from 2 communities in the US, Birmingham, Alabama and Iowa City, Iowa, through mass mailing of letters and study brochures, supplemented by media and community outreach campaigns. Each center also recruited ethnic minorities according to their representation in the recruitment population. Subjects were excluded from the MOST if they screened positive for rheumatoid arthritis (10), had ankylosing spondylitis, psoriatic arthritis, or reactive arthritis, experienced problems with the kidneys that resulted in the need for hemo- or peritoneal dialysis, had a history of cancer (except for nonmelanoma skin cancer), had undergone bilateral knee replacement surgery, were unable to walk without the help of another person or walker, or were planning to move out of the area in the subsequent 3 years. The study protocol was approved by the institutional review boards at the University of Iowa, University of Alabama, Birmingham, University of California, San Francisco, and Boston University Medical Center.

In the present study we used a subsample of subjects from the parent study. Using a nested case–control study design (11), we investigated knees that were considered eligible for MRI and that had been examined both at baseline and at 15 months' followup. At baseline, all study subjects were asked a question regarding knee pain, as follows: “During the past 30 days, have you had pain, aching, or stiffness in your knee on most days?” This question was posed to subjects both by phone interview and during a clinic visit 1 month thereafter. If the subject answered no regarding the presence of pain, aching, or stiffness in the knee at each of these time points at baseline, that knee was considered eligible for analysis in the current nested case–control study.

At 15 months' followup, this same question regarding knee pain was posed to subjects, both in a phone interview and at a clinic visit. If the subject answered yes to the question at both of these followup time points, the knee with new pain was considered to be a case knee (n = 110). Because this question was intended to identify pain, aching, or stiffness “on most days,” and because the 2 time points were, on average, 1 month apart, we characterized a positive response at both time points as “consistent frequent knee pain.” Furthermore, although our question included the symptoms of stiffness and aching, we labeled a positive response as related to pain. Control knees (n = 220) were selected randomly from among the knees examined by MRI at baseline and for which the subject indicated absence of pain at both baseline time points.

In addition, at the baseline clinic examination, subjects completed surveys on medication use and filled out a questionnaire on symptoms of depression, the Center for Epidemiologic Studies Depression Scale (CES-D) (12). In addition, subjects were weighed (without shoes) on a balance-beam scale to determine the body mass index (BMI), computed as weight (kg)/height (m2). The same survey on medication use was administered at the 15-month visit. Subjects also completed the Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC), a survey of joint pain, stiffness, and limitation of physical function (13), for each knee at both the baseline and 15-month visits.


MR images were obtained with a 1.0T dedicated MR system (OrthOne; Oni, Wilmington, MA) with a circumferential extremity coil. All MRIs were performed using fat-suppressed, fast spin-echo, proton density–weighted sequences in 2 planes, the sagittal plane (repetition time [TR] 4,800 msec, time to echo [TE] 35 msec, slice thickness 3 mm, interslice gap 0 mm, field of view [FOV] 14 cm, matrix 288 × 192 pixels, number of excitations [NEX] 2) and the axial plane (TR 4,700 msec, TE 13.2 msec, slice thickness 3 mm, interslice gap 0 mm, FOV 14 cm, matrix 288 × 192 pixels, NEX 2), and a STIR sequence in the coronal plane (TR 7,820 msec, TE 14 msec, inversion time 100 msec, slice thickness 3 mm, interslice gap 0 mm, FOV 14 cm, matrix 256 × 256 pixels, NEX 2).

Two musculoskeletal radiologists (AG and FR), who were blinded to the case/control status and clinical data, read the MR images for evidence of BMLs and effusions, according to the Whole-Organ MRI Score (WORMS) (14). Using the WORMS, BMLs were scored 0–3 based on the size, or volume, of the lesion in each of 5 subregions of the medial and lateral compartments and in 4 subregions of the patellofemoral compartment. Similarly, effusion volume was scored 0–3. The MR images were read as paired images from both the baseline and followup visits. In addition, bone attrition was scored 0–3 for each subregion, with results expressed as the maximal score for a compartment. Periarticular lesions (e.g., anserine bursitis) have also been cross-sectionally associated with knee pain (15); however, no change in these lesions was observed in our subjects' knees.

Given that the results of previous studies (5, 6) have suggested that the size of BMLs is associated with pain, we focused on the change in size, or volume, of the lesion as our measure of BML change. We summed the BML score for all 5 subregions of the medial compartment (score range 0–3 in each subregion; across all 5 subregions, score range 0–15), all 5 subregions of the lateral compartment (score range 0–15), and all 4 subregions of the patellofemoral compartment (score range 0–12). This yielded 3 BML scores for each knee.

In each knee compartment, change was defined as change in the BML score from baseline to followup. Interreader agreement for this change score was an intraclass correlation coefficient (ICC) of 0.93 (P < 0.001). Given the compartment-specific nature of OA and the likelihood that a compartment would develop new or worse disease, we focused on change in BML score in each compartment. For each subject, we thus obtained 3 compartment-specific BML change scores, and of these 3 scores, we focused on the change score that showed the maximal unit increase; we have previously used this compartment-specific approach for scoring of BMLs to identify a strong relationship between BMLs and knee malalignment (16). In a sensitivity analysis, we summed all of the BML scores for the whole knee and created a global knee change score by subtracting the baseline score from the followup score. Results were the same as those obtained using the compartment-specific approach.

Radiographic assessment.

At baseline, all subjects underwent weight-bearing, posteroanterior (PA), fixed-flexion radiographic evaluation of the knee using the protocol described by Peterfy et al (17). Body weight was equally distributed between the 2 legs, and the big toes of the feet and the front of the thighs were placed in contact with the front plate of the plexiglass frame. The external rotation of the feet was fixed at 10 degrees, using a V-shaped foot angulation support on the frame. The central radiographic beam was directed to the midpoint between the back of the knees, at a caudal angle of ∼10 degrees, to allow the anterior and posterior lips of the medial tibial plateau to be optimally superimposed (film-focus distance 183 cm).

A musculoskeletal radiologist (PA) and a rheumatologist (DTF), who were experienced in reading study films and were blinded to both the case/control status and clinical data, graded all of the PA radiographs according to the Kellgren/Lawrence (K/L) scale of radiographic knee OA (18). Radiographic OA was considered present if knees were assessed a K/L grade ≥2.

Full-limb radiographs of both legs were obtained at baseline, using the method of Sharma et al (19). The mechanical axis was defined as the angle formed by the intersection of a line from the center of the head of the femur to the center of the tibial spines, and a second line from the center of the talus to the center of the tibial spines. The interobserver ICC for determination of the mechanical axis was 0.99 (P < 0.0001).

Statistical analysis.

To test the frequency of change in BMLs in the case knees as compared with the control knees, we used a chi-square test, both comparing knees and comparing subjects (given the occurrence of 8 cases in which both the left and the right knee of a subject were affected). We used multiple logistic regression with generalized estimating equations (to adjust for the correlation of knees) to evaluate the relationship between an increase in BML score and incident knee pain, in models adjusted for age, sex, race (white versus African American), BMI, CES-D score, baseline quadriceps strength, K/L grade, effusion score, baseline BML score, and change in effusion score, and with or without adjustment for mechanical knee alignment. Additional analyses in which we examined change in pain medication use (including use of nonsteroidal antiinflammatory drugs) as a covariate yielded similar findings, as did additional analyses in which we adjusted for the presence or absence of bone attrition (a rare finding in the knees studied). All P values (2-tailed) were calculated using SAS for Windows, version 9.1 (SAS Institute, Cary, NC). P values less than or equal to 0.05 were considered significant.


Case knees (n = 110), i.e., knees with incident pain, aching, or stiffness at the 15-month followup visit, were identified in study subjects who were slightly older and more often female than were the subjects in the control group (n = 220). In addition, case knees were identified as having radiographic OA at baseline more frequently than control knees (Table 1).

Table 1. Characteristics of subjects with knee osteoarthritis (OA) among the case and control groups*
CharacteristicCases (n = 110 knees; n = 102 subjects)Controls (n = 220 knees; n = 220 subjects)P
  • *

    All values are from the baseline examination, except change in effusion score. Control knees were randomly selected from the same source population as case knees but were without pain, aching, or stiffness at baseline. Case knees were those reported to have frequent pain, aching, or stiffness at 15 months' followup. BMI = body mass index; CES-D = Center for Epidemiologic Studies Depression Scale; K/L = Kellgren/Lawrence.

  • Calculated by chi-square test or 2-sample t-test, as appropriate.

  • Calculated using the Whole-Organ Magnetic Resonance Imaging Score (scale 0–3).

  • §

    Calculated from findings on magnetic resonance imaging (scale 0–3). Change was defined as an increase in score.

Age, mean ± SD years62.9 ± 8.361.2 ± 8.40.08
Female, no. (%)79 (65.5)132 (60.0)0.04
African American, no. (%)12 (10.9)30 (13.6)0.29
BMI, mean ± SD kg/m229.5 ± 4.629.7 ± 4.30.60
CES-D score, mean ± SD7.3 ± 7.56.7 ± 6.60.43
Quadriceps strength, mean ± SD newtons84.5 ± 39.096.0 ± 42.90.02
Knee OA (K/L grade ≥2), no. (%)33 (30.0)47 (21.4)0.08
Bone marrow lesion score at baseline, mean ± SD2.1 ± 2.31.7 ± 2.10.07
No. (%) of knees with effusion score ≥1 at baseline§40 (36.4)65 (29.6)0.21
No. (%) of knees with change in effusion score from baseline to followup§27 (24.6)23 (10.5)0.0008

In analyses focusing on the knee compartment with greatest change in BML score, we found that 54 of 110 case knees (49.1%, 95% confidence interval [95% CI] 40.2–58.9%) showed an increase in BML score in a compartment, whereas only 59 of 220 control knees (26.8%, 95% CI 21.0–32.7%) showed an increase in BML score in a compartment (P < 0.001 by chi-square test, between knees and between subjects). Since each compartment had 4 or 5 subregions, we compared case knees and control knees according to the number of subregions of a compartment that showed an increase in BML score. We found that an increase in BML score occurred in 1 subregion in 27.3% of case knees (95% CI 19.0–35.6%) compared with 23.6% of control knees (95% CI 18.0–29.3%). Of note, however, the BML score increased in more than 1 subregion in 25.5% of case knees (95% CI 17.3–33.6%) but in only 8.2% of control knees (95% CI 4.6–11.8%).

Compared with control knees, case knees showed larger increases in BMLs (P = 0.0003), even after adjustment for age, sex, race, BMI, baseline quadriceps strength, CES-D score, baseline K/L grade, baseline BML score, effusion score, and change in effusion score (Table 2). There were no knees or compartments within a knee in which the BML score, either at baseline or at followup, was at the highest possible (ceiling) level. Additional adjustments for alignment of the mechanical axis across the knee yielded similar results.

Table 2. Association of 1-unit increase in BML score with incident knee pain*
 OR (95% CI)Adjusted OR (95% CI)
  • *

    Increase in bone marrow lesion (BML) score was defined as the maximal increase in BML score among the compartments of a knee. OR = odds ratio; 95% CI = 95% confidence interval.

  • Adjusted only for Kellgren/Lawrence (K/L) grade of radiographic knee osteoarthritis.

  • Adjusted for age, sex, race (white versus African American), body mass index, quadriceps strength, Center for Epidemiologic Studies Depression Scale score, K/L grade, baseline BML score, and presence of knee malalignment, all at baseline. Also adjusted for change in effusion score determined on magnetic resonance imaging.

  • §

    P = 0.0003 versus controls.

Controls1 (referent)1 (referent)
Cases1.74 (1.38–2.18)1.61 (1.24–2.09)§

We then assessed whether change in BML score (as well as magnitude of change in score) might be associated with new knee pain (Table 3). The results showed that new knee pain was modestly, but nonsignificantly, associated with a 1-unit increase in the BML score in any compartment of the knee (adjusted odds ratio [OR] 1.5, 95% CI 0.8–3.1). However, new knee pain was strongly and significantly associated with a ≥2-unit increase in BML score (adjusted OR 3.2, 95% CI 1.5–6.8). In fact, among the control knees, only 8.6% had a BML score increase of ≥2, as compared with 27.5% of case knees showing this magnitude of increase in BML score.

Table 3. Incident knee pain in relation to incremental change in bone marrow lesion (BML) score over 15 months' followup*
 Control knees, no. (%)Case knees, no. (%)OR (95% CI)Adjusted OR (95% CI)§
  • *

    OR = odds ratio; 95% CI = 95% confidence interval.

  • One case had missing information on bone marrow change because we had difficulty assessing change given the lack of comparable images at baseline and followup.

  • Adjusted only for Kellgren/Lawrence (K/L) grade of radiographic knee osteoarthritis.

  • §

    Adjusted for age, sex, race (white versus African American), body mass index, quadriceps strength, Center for Epidemiologic Studies Depression Scale score, K/L grade, presence of knee malalignment, and BML score, all at baseline. Also adjusted for change in effusion score on magnetic resonance imaging.

  • P for trend = 0.002.

No increase in BML score161 (73.2)55 (50.5)1 (referent)1 (referent)
Increase of 1 in BML score40 (18.2)24 (22.0)1.6 (0.9–3.0)1.5 (0.8–3.1)
Increase of ≥2 in BML score19 (8.6)30 (27.5)4.1 (2.1–8.1)3.2 (1.5–6.8)

Most of these findings were accounted for by an increase in BML score in knees that were already reported to have BMLs at baseline. Nevertheless, among the subset of knees with no BMLs in any compartment of the knee at baseline, new BMLs were more common in case knees (11 [32.4%] of 34 knees, 95% CI 17.3–49.4%) than in control knees (9 [10.8%] of 83 knees, 95% CI 4.2–17.5%).

In analyses focusing on the site of increase in BMLs, we found that increases in BML scores of ≥2 units occurred in all compartments of the knee more often among case knees than among control knees. There was no predilection for these increases to occur in any compartment.

We then examined whether the increase in BML score involved single large lesions or multiple small lesions. The results showed that more than one-half of the BML increases were a reflection of the development of single large lesions, in which the BML score would change from a score of 0 (absence of lesions) to scores of 2 or 3 or from a score of 1 to a score of 3.


In the present study, which we believe is the first to examine the relationship between BMLs on MRI and presence of new knee pain in subjects over time, we found that an increase in the size of BMLs in a knee compartment was associated with the development of knee pain. The relationship persisted after statistical adjustment for a large number of factors that have been linked to knee pain, including effusion score (20). Our observations provide strong evidence that BMLs are one source of pain in knees with or without radiographic OA.

Although our data suggest a consistent and moderately strong association of both the development of BMLs (new lesions) and enlargement of existing BMLs with the development of frequent knee pain, the findings do not identify BMLs as being the sole source of knee pain. For example, of those subjects in the incident knee pain group, only 49.1% had an increase in BML score, compared with 26.8% of control knees, suggesting that such an increase in BMLs is frequent and, in some knees, may not be associated with pain. Thus, although we provided additional evidence of the importance of BMLs as a likely source of knee pain, our results suggest that they are not the sole source, and many subjects may develop knee pain without any increase in BMLs in their knee and must therefore develop pain for other reasons. One of these reasons might be inflammation within the joint or enlargement of effusions.

To meet criteria for incident knee pain, subjects must have reported consistent knee pain at followup and no knee pain at baseline. However, those subjects who noted no frequent knee pain at baseline often reported some knee pain on the baseline WOMAC questionnaire. Thus, incident knee pain in such subjects would represent an increase in the frequency and severity of the knee pain and not necessarily completely new knee pain.

There were important limitations to this study. First, we were unable to tell whether BMLs ought to be defined by the largest lesions or the sum of all lesions within a knee. The strong relationship between knee pain and an increase of least 2 units in the BML score suggests that the size of the lesions is strongly related to knee pain. Moreover, since most subjects who developed incident pain did not have WOMAC knee pain scores of 0 at baseline, their “incident” knee pain actually represents the development of more frequent pain. We cannot say that incident knee pain was truly incident in our subjects. Many of our subjects could have had knee pain before the baseline examination and could have experienced fluctuating pain thereafter. We can conclude, however, that BMLs are likely to appear or get larger when pain occurs.

A similar relationship between changing BMLs and pain has been reported in an observational manner in subjects with transient bone marrow edema syndrome, a phenomenon sometimes seen in diabetics, in which the resolution of BMLs is associated with the resolution of pain (21). Transient BMLs also occur in subjects with acute knee trauma (22, 23). Another limitation of our study is that we did not study the resolution of pain, but rather its increase. We have previously reported that BMLs are unlikely to disappear in knees with OA, but rather, BMLs often tend to either increase or appear over time (24, 25), and this has been confirmed by other investigators (26). We therefore did not believe that it would be productive to focus on the resolution of pain and its association with the resolution of BMLs, since little such resolution occurs.

Observational studies investigating pain in OA will continue to be constrained by the cooccurrence of multiple pathologic features, each of which may be associated with pain. The OA knee is a joint that is failing, with pathologic features in many structures, including the bone, ligaments, and joint capsule, as well as inflammation in the synovium. The pathologic features in these structures are not independent, but rather reflect the whole disease process, and therefore it will continue to be difficult to distinguish the contribution of different structures to pain. One approach is to assess the disease earlier in its process in subjects who have little structural disease. Many of our subjects had minimal, if any, radiographic disease and had no BMLs at baseline. In this subgroup, we similarly found a strong relationship of the development of new bone lesions with new knee pain. We anticipate that future studies will focus on earlier stages of the disease, at which point it might be easier to identify specific structural pathologic features and their relationship to symptoms.

In conclusion, in this study evaluating the development or enlargement of BMLs in individuals and the association of change in BMLs with knee pain, we found that subjects without frequent knee pain who developed knee pain in followup were more likely to experience an increase in BMLs on MRI than were a control group of subjects drawn from the same cohort. This provides further evidence that bone marrow lesions are likely to be a source of the knee pain in some subjects.


Dr. Felson had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study design. Felson, Torner, Lewis, Nevitt.

Acquisition of data. Felson, Guermazi, Roemer, Aliabadi, Clancy, Torner, Lewis.

Analysis and interpretation of data. Felson, Niu, Guermazi, Roemer, Torner.

Manuscript preparation. Felson, Guermazi, Roemer, Torner, Lewis, Nevitt.

Statistical analysis. Niu.


We are indebted to the MOST staff members at Iowa City and Birmingham, and to the study subjects for their participation.