Introduction 1 2 3 3 1 4 10 11 14 15 16 9 7 17 Despite the recent literature on BME, the implication of these findings on MRI and how they relate to clinical outcomes is unknown. Most studies have used the subjective complaint of pain as the only clinical correlate. There has been no measure of an objective clinical outcome related to the finding of BME. The purpose of this study was to determine if a correlation exists between MRI findings of BME of the knee joint and the incidence of total knee arthroplasty (TKA) within a follow-up period of 3 years. Materials and methods The entire database of knee MRI studies from 1995–1997 (over 4,000 studies) conducted at a large urban hospital system was used to select individuals with knee osteoarthrosis (OA). The study was reviewed and approved by the hospital’s institutional review board. An initial random search was conducted within this database to identify two distinct groups of patients. The first group had MRI reports containing the phrase “knee osteoarthrosis/osteoarthritis,” and the second group consisted of those containing the phrases “knee osteoarthrosis/osteoarthritis” and “bone edema.” The search for “osteoarthrosis/osteoarthritis” yielded 235 cases for review. The search for “osteoarthrosis/osteoarthritis” and “bone edema” yielded 146 cases for review. After these two initial groups were identified, a chart review was conducted on all 381 patients to identify and include only those patients who had at least a 3-year clinical follow-up appointment from the time of the MRI study. Subjects were also excluded if the reason for referral to MRI was post-traumatic or post-surgical. An initial review of the images was completed by an experienced musculoskeletal radiologist to exclude any subjects with evidence of recent post-surgical or post-traumatic changes not mentioned in the report. After this chart review and initial radiology review process, there were 38 patients in the OA-only group and 35 in the OA with BME group. The same musculoskeletal radiologist who provided the initial review was blinded to the original interpretation of the MRI studies and the patient outcome, and reviewed all 73 studies. The radiologist was aware of the study hypothesis at the time of interpretation. The radiologist assessed each knee for the presence or absence of BME and assigned each subject into either the BME group or the no-BME group for further evaluation. After this review of the images, there were a total of 25 patients with OA only and 48 with OA and BME. The OA-only group consisted of four males and 21 females, with an age range of 28–75 years, and an average age of 49.3 years. Conventional radiographs consisting of anteroposterior, lateral and sunrise views were available for review with 13 of the 25 patients, 52% of the group. The OA and BME group consisted of 23 men and 25 women, with an age range of 35–82 years and an average age of 53.5 years. Conventional radiographs, again consisting of anteroposterior, lateral, and sunrise views, were available for review with 33 of the 48 patients, 68.75% of the group. 18 None Doubtful Minimal Moderate Severe 18 All patients had dedicated knee MR scans either on a GE 1.5-T magnet (23 in the OA-only group or 92% and 42 subjects in the OA with edema group or 87.5%) or a 1.0-T magnet (8% of the OA-only group, two subjects, and 12.5% of the OA with edema group, six patients). MRI films were reviewed and interpreted by the musculoskeletal radiologist for all patients in the study. For each patient, the sagittal proton density sequences (TR range, 2,666–4,100; TE range, 14–33; 44 conventional spin echo studies [ET = 0], 29 fast spin echo studies [ET = 4–8], 3- to 5-mm section thickness, 16 × 16–18 × 18 field of view [FOV], 256 × 128–256 image matrix, bandwidth range 16–32), coronal proton density sequences (TR range, 2,316–3,950; TE range, 14–33; 44 conventional spin echo studies [ET = 0]; 29 fast spin echo studies[ET = 4–8]; 3-mm section thickness; 16 × 16–18 × 18 FOV; 256 × 128–256 image matrix; bandwidth range, 16–32), and coronal fat suppressed T2-weighted images (TR range, 2,400–3,200; TE range, 56–80; 3- to 5-mm section thickness; 16 × 16–18 × 18 FOV; 256 × 128–256 image matrix; bandwidth range, 15.6–16) were reviewed. Axial fat suppressed T2-weighted images (TR range, 3,000–3,600; TE range, 56–76; 1.5- to 10-mm section thickness; 16 × 16–20 × 20 FOV; 256 × 128–256; bandwidth range, 15.6–16) were available and reviewed for 69 of the 73 knees. For the four subjects who had no axial T2 imaging available, BME in the patellofemoral joint was not assessed. 19 Normal Internal changes only 1–49% loss of articular cartilage 50–99% loss of articular cartilage 100% loss of cartilage with subchondral cortex intact 100% loss of cartilage with ulcerated subchondral cortex The most severe grade cartilage lesion within a compartment was utilized for the grading of that compartment. Also, the highest grade cartilage lesion on either side (femoral or tibial) of a compartment was used. 1 2 3 4 5 6 Fig. 1 arrow  Fig. 2 curved arrow straight arrows  Fig. 3 straight arrows curved arrow  Fig. 4 arrows  Fig. 5 arrows  Fig. 6 arrows curved arrow  Generalized estimating equations (GEE) were used to determine which factors were related to receiving a TKA. There were eight subjects who had bilateral knee MRIs. GEE takes into account the correlation within an individual who had both knees involved, whereas a general linear model assumes that all observations are independent and does not allow for multiple observations from one individual. A multivariable logistic regression model using generalized estimation was also used to adjust for the age of the patients and to determine if the results were still significant after accounting for the age differences. Results 7 1 7 3 4 1 6 1 Table 1 Patterns of edema and percentage of those who subsequently received TKA Pattern of edema Total number Number with TKA Percentage with TKA (%) Focal 34 7 20.5 Global 12 7 58.3 Cyst 2 1 50 All patterns of edema 48 15 31.2 None 25 2 8 7 1 Fig. 7 No bone marrow edema with a total joint replacement in a 66-year-old female patient. Coronal fast fat-suppressed T2-weighted image (TR 2500/TE 80/256 × 256) shows almost complete loss of hyaline cartilage in the medial compartment and an extruded, degenerative torn medial meniscus. Note the absence of marrow edema  2 Table 2 Cartilage loss seen on MR, as graded by the modified Noyes scale, and its relationship to the occurrence of TKA Variable No TKA Yes TKA OR 95% CI p Medial cartilage loss <50% 17 (30%) 2 (12%) 2.77 0.71–10.72 0.14 Medial cartilage loss ≥50% 39 (70%) 15 (88%) Lateral cartilage loss <50% 42 (75%) 10 (59%) 2.06 0.74–5.70 0.16 Lateral cartilage loss ≥50% 14 (25%) 7 (41%) PF cartilage loss <50% 15 (28%) 3 (19%) 1.68 0.45–6.29 0.44 PF cartilage loss ≥50% 38 (72%) 13 (81%) 3 Table 3 Radiographic findings, as graded by the Kellgren–Lawrence scale, and their relationship to occurrence of TKA Score N N OR 95% CI p 0–2 8 (23%) 1 (9%) 1.0006 0.9996–1.0016 0.23 3–4 27 (77%) 10 (91%)  4 p p p Table 4 Comparison of bone marrow edema patterns and their relationship to occurrence of TKA Variable No TKA Yes TKA OR 95% CI p No BME 23 (41%) 2 (12%) 5.48 0.98–30.68 0.05 BME of any pattern 33 (59%) 15 (88%) Global BME 5 (9%) 7 (41%) 7.63 1.89–28.34 0.004 All other patterns (focal, cyst, none) 51 (91%) 10 (59%) Global BME 5 (18%) 7 (78%) 15.21 2.38–97.10 <0.001 No BME 23 (82%) 2 (22%) First, no bone marrow edema is compared with BME of any pattern. Next global BME is compared to all other patterns, including no edema. Finally, global BME is compared with no BME. 5 Table 5 t Variable Score No BME BME p Modified Noyes score lateral compartment 0–2a 20 (80%) 32 (67%) 0.232 2b–3b 5 (20%) 16 (33%) Modified Noyes score medial compartment 0–2a 11 (44%) 8 (17%) 0.012 2b–3b 14 (56%) 40 (83%) Modified Noyes score patellofemoral compartment 0–2a 4 (17%) 14 (31%) 0.193 2b–3b 20 (83%) 31 (69%) Kellgren–Lawrence score lateral compartment 0–2 13 (100%) 26 (79%) 0.071 3–4 0 (0%) 7 (21%) Kellgren–Lawrence score medial compartment 0–2 9 (69%) 11 (33%) 0.027 3–4 4 (31%) 22 (67%) Kellgren–Lawrence score patellofemoral compartment 0–2 5 (42%) 11 (33%) 0.606 3–4 7 (58%) 22 (67%) p 6 p p p Table 6 Results of the multivariable logistic regression model using generalized estimation to account for the age difference in the groups that received a TKA versus those that did not   Outcome: TKA Variable OR 95% CI p-value Model 1 BME versus no BMEf 8.95 1.49–53.68 0.0164 Age 1.13 1.07–1.20 <0.0001 Model 2 Global BME versus No BME 13.04 2.06–82.58 0.0064 Age 1.06 0.99–1.15 0.11 Model 3 Global BME versus all other patterns (focal, cyst, and no edema) 5.45 1.02–28.96 0.0467 Age 1.107 1.05–1.17 0.0002 Discussion 1 4 11 13 14 20 1 7 17 7 17 4 21 6 9 10 10 9 6 None of these prior studies specifically looked at the BME pattern on MR. Our classification of the patterns into global, focal, cystic, and absence of edema is an attempt to subdivide the presence or absence of edema in osteoarthrosis. However, this attempt is limited by the absence of histopathological findings. We were surprised by the significantly increased risk of knee joint replacement with the global pattern of BME in relation to the other patterns. It appears that the more extensive and intense the BME, the more likely it is for the patient to have symptoms. The global pattern of BME was the best predictor of risk of TKA within 3 years, as those subjects with the global pattern were over five times as likely to receive a TKA when compared to those with any of the other patterns and over 13 times as likely to have a TKA when compared with subjects with no BME, after accounting for the age difference. 4 4 21 3 We did find that patients who had a TKA were 12.6 years older than those who did not have a TKA. Patients who are older are more likely to have OA and, in particular, more severe OA. Surgical replacement of the knee is more likely to occur in an older patient with OA than a younger patient with the same severity of OA. After further statistical analysis, we however found that our BME results were still significant despite the differences in age. This paper has several limitations. Most patients who have a TKA do not get a pre-operative MR, and therefore, we may be dealing with a pre-selected population. What we have called “BME” on MR may not be true edema but may relate to other histologic findings as noted earlier. Our numbers are also small, and we had to group some of our subsets together for statistical analysis. Only one musculoskeletal radiologist was involved in the review of the radiology. There was also no arthroscopic correlation for cartilage defects or presence of meniscal tears. Our study is limited by the fact that it is retrospective. To generate a group of MR scans to review, we had to use a keyword search to identify potential subjects. We would have obviously missed all scans where the specific terminology “bone edema” or knee “osteoarthrosis/osteoarthritis” was not utilized. However, we feel that this is a relatively minor limitation, as we were able to generate 381 cases for review. Another potential limitation in the retrospective study design is the possibility of missing subjects who received follow-up at a different institution or those whose symptoms resolved without treatment. In summary, we reviewed a series of knees in patients with osteoarthrosis and evaluated the pattern of BME, cartilage loss, radiographic findings when available, and the incidence of total knee joint replacement within a 3-year follow-up. Subjects with any bone marrow edema pattern were more likely to have a total knee joint replacement compared to subjects with no bone marrow edema. The worst prognostic pattern was the global pattern of bone marrow edema. Subjects who had a total knee replacement were also older than those who did not. However, even allowing for age, the global pattern of BME remained the variable with the highest statistically significant association with the incidence of total knee replacement.