SEARCH

SEARCH BY CITATION

Abstract

  1. Top of page
  2. Abstract
  3. PATIENTS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. REFERENCES

Objective

To determine the relationship between radiographic progression of joint space narrowing and cartilage loss on magnetic resonance imaging (MRI) in patients with symptomatic knee osteoarthritis (OA), and to investigate the location of MRI-based cartilage loss in the knee and its relation to radiographic progression.

Methods

Two hundred twenty-four men and women (mean age 66 years) were studied. Radiographs and MRI of the more symptomatic knee were obtained at baseline and at 15- and 30-month followup. Radiographs of the knee (with weight-bearing) were read for joint space narrowing (scale 0–3), with progression defined as any worsening in score. We used a semiquantitative method to score cartilage morphology in all 5 regions of the tibiofemoral joint, and defined cartilage loss as an increase in score (scale 0–4) at any region. We examined the relationship between progression of joint space narrowing on radiographic images and cartilage loss on MRI, using a generalized estimating equation proportional odds logistic regression, adjusted for baseline cartilage score, age, body mass index, and sex. The medial and lateral compartments were analyzed separately.

Results

In the medial compartment, 104 knees (46%) had cartilage loss detected by MRI. The adjusted odds ratio was 3.7 (95% confidence interval 2.2–6.3) for radiographic progression being predictive of cartilage loss on MRI. However, there was still a substantial proportion of knees (80 of 189 [42%]) with cartilage loss visible on MRI when no radiographic progression was apparent. Cartilage loss occurred frequently in the central regions of the femur and tibia as well as the posterior femur region, but radiographic progression was less likely to be observed when posterior femur regions showed cartilage loss. Radiographic progression appeared specific (91%) but not sensitive (23%) for cartilage loss. Overall findings were similar for the lateral compartment.

Conclusion

While our results provide longitudinal evidence that radiographic progression of joint space narrowing is predictive of cartilage loss assessed on MRI, radiography is not a sensitive measure, and if used alone, will miss a substantial proportion of knees with cartilage loss.

Knee osteoarthritis (OA) is one of the leading causes of chronic disability among older individuals (1). An accurate indicator of disease progression is critical for studies of this condition. Thinning, erosion, and loss of hyaline articular cartilage are considered the hallmark of OA. In clinical studies on the natural history of knee OA and in clinical trials assessing treatment efficacy for knee OA, worsening of joint space narrowing seen on knee radiography, which is thought to reflect articular cartilage loss, has served as one indicator of structural disease progression. Although serial radiography is a noninvasive and inexpensive means to evaluate progression of joint space narrowing, it remains an indirect measure of cartilage loss.

There are several potential disadvantages of using knee radiographs as an assessment of cartilage loss over time. First, progression of joint space narrowing is determined as the distance between the femur and tibia at only 1 point of contact during standing, and therefore cartilage loss occurring along other areas of the femur and tibia in contact during the flexion/extension excursion of the knee would not be captured. In addition, the distance between the femur and tibia represents more than just hyaline articular cartilage, since the meniscus is lying in the same space. Therefore, in some instances progression of joint space narrowing could reflect meniscal extrusion rather than true cartilage loss (2). Since strategies that optimize reproducibility and accuracy of knee radiographic joint space measurement focus on the medial compartment (3), the ability of radiography to evaluate changes in the lateral compartment of the knee may be more limited. Last, radiographic progression is unlikely to be sensitive to small changes in cartilage.

Magnetic resonance imaging (MRI) offers a distinct advantage over knee radiography, since cartilage can be assessed directly. Several studies have been performed evaluating MRI as an imaging tool for knee OA (4–8). Since MRI also provides 3-dimensional imaging of the knee, in contrast to the 2-dimensional radiographic images, cartilage morphology in multiple compartments of the knee can be better assessed.

Radiographic joint space narrowing has been accepted as a proxy measure of articular cartilage thickness based on evidence showing a correlation between cross-sectional assessments of radiographic joint space narrowing and several different measures of cartilage morphology, including arthrography (9, 10). However, few longitudinal studies have evaluated the changes in cartilage in the knee in relation to radiographic progression (11). No study, to our knowledge, has examined the longitudinal association between radiographic progression of joint space narrowing and the location of MRI-based cartilage loss. Furthermore, it is unknown how often substantial cartilage loss occurs and remains undetermined by radiographic assessment. In this study, we examined progression of joint space narrowing on knee radiographs, and its association with changes in articular cartilage at the tibiofemoral joint as measured using MRI in a large sample of older men and women followed up for 30 months.

PATIENTS AND METHODS

  1. Top of page
  2. Abstract
  3. PATIENTS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. REFERENCES

Study participants.

Study participants were enrolled in a 30-month natural history study of symptomatic knee OA (Boston Osteoarthritis of the Knee Study); recruitment has been described in detail elsewhere (12, 13). Briefly, subjects were recruited from 2 prospective studies (1 of men and 1 of women) of quality of life among veterans. Participants were drawn from clinics at the Veterans Administration (VA) Boston Healthcare System and from advertisements in local newspapers. The minimum age for enrollment was 45 years for men and 50 years for women. The older entry age for women was chosen to decrease the likelihood of women being pregnant. Potential participants were asked the following 2 questions: “Do you have pain, aching, or stiffness in 1 or both knees on most days?” and “Has a doctor ever told you that you have knee arthritis?” For individuals who answered yes to these questions, a subsequent interview was conducted to determine if other forms of arthritis were a cause for knee symptoms, and if not, they were eligible for recruitment.

Knee radiographs were obtained, and if an osteophyte was present on any view of the symptomatic knee, the individual was eligible for the study. To be eligible, subjects also had to be able to walk with or without the aid of a cane and had to be interested in participating in the longitudinal study. There were 324 subjects (201 men and 123 women) having a mean ± SD age of 67 ± 9 years and body mass index of 31.4 ± 5.7 kg/m2, who met the eligibility criteria. All participants met American College of Rheumatology criteria for symptomatic knee OA (14). The majority of men in the study received their care through the VA Boston Healthcare System, and had been recruited from VA clinics, while the majority of women were recruited from the community.

Study design.

Examinations were conducted at baseline and at 15 and 30 months. At all examinations, subjects underwent knee radiography. All subjects who had no contraindications to MRI underwent MRI of the more symptomatic knee at baseline and again at 15- and 30-month followup. Subjects were also weighed, with shoes off, on a balance beam scale, and height was measured. The Institutional Review Boards of Boston University Medical Center and the VA Boston Healthcare System approved the baseline and followup evaluations.

Radiographic evaluation.

All subjects underwent weight-bearing posteroanterior (PA) radiography of the knee, using the protocol of Buckland-Wright et al (3). Under fluoroscopic positioning, we aligned the beam relative to the knee center, and the knee was flexed so that the anterior and posterior lips of the medial tibial plateau were superimposed. Feet were rotated until the tibial spines were centered in the notch, and outlines of foot rotation were then made on foot maps so that the foot rotation would be the same for subsequent films. This method of positioning has been shown to more accurately assess joint space and to improve reliability of joint space assessment, particularly for the medial compartment (3). All radiographs were read using the atlas of individual radiographic features in OA (15), in which the medial and lateral tibiofemoral compartment joint spaces were each graded from 0 (normal) to 3 (bone on bone).

We defined radiographic progression as an increase of ≥1 in the joint space narrowing score at followup. All radiographs were read by 1 reader, without blinding for sequence. A subset of films was reread, with blinding with regard to sequence, to evaluate for possible bias in characterization of progression. It was found that there was no greater tendency for unblinded readings to be characterized as showing radiographic progression (κ = 0.81, P < 0.001 for intraobserver agreement in reading progression).

MRI evaluation.

MRIs were acquired on a General Electric Signa 1.5-Tesla MRI system (General Electric Medical Systems, Milwaukee, WI) using a phased-array knee coil. An anchoring device for the ankle and knee was used to ensure uniformity of positioning between patients and for followup. Coronal, axial, and sagittal images of the knee were obtained. Imaging protocol included coronal and axial spin-echo fat-saturated proton-density and T2-weighted images (repetition time 2,200 msec, time to echo 20–80 msec) with a slice thickness of 3 mm, a 1-mm interslice gap, 1 excitation, a field of view of 11–12 cm, and a matrix of 256 × 192 pixels. Imaging protocol for sagittal images was identical, except for fat saturation.

MRIs of tibiofemoral joint cartilage were read using the Whole-Organ MRI Score (WORMS) semiquantitative method for knee OA (8); readers were blinded with regard to scores of radiographic progression. There were a total of 3 readers who scored all MRIs. The majority of subjects with longitudinal MRIs (86%) were read by a trained musculoskeletal radiologist (AG) and a musculoskeletal researcher (MG, trained by AG) reading as a pair, together. One reader (DJH), trained in the WORMS scoring by 1 of the musculoskeletal radiology readers (AG), read the remainder of the MRIs.

Cartilage was scored for the medial and lateral compartments separately, and on each of 5 regions of the tibiofemoral joint (central and posterior femur; anterior, central, and posterior tibia) (Figure 1). Both cartilage signal and cartilage morphology were scored using a scale of 0–6, as follows: 0 = normal thickness and signal; 1 = normal thickness but increased signal on T2-weighted images; 2 = solitary focal defect of <1 cm at greatest width; 3 = areas of partial-thickness defects (<75% of the region) with areas of preserved thickness; 4 = diffuse partial-thickness loss of cartilage (≥75% of the region); 5 = areas of full-thickness loss (<75% of the region) with areas of partial thickness loss; and 6 = diffuse full-thickness loss (≥75% of the region) (8). Intraclass correlation coefficients of agreement among the readers for cartilage readings ranged from 0.72 to 0.97.

thumbnail image

Figure 1. Diagram of the knee (sagittal view), illustrating the 5 cartilage regions of the tibiofemoral joint (circled) (central [C] femur, posterior [P] femur, anterior [A] tibia, central tibia, and posterior tibia). Since the anterior femur cartilage regions articulates with the patella and not the tibia, it is not included in cartilage magnetic resonance image readings of the tibiofemoral joint. Dashed lines separate the different regions of the femur and tibia. The menisci are indicated with arrowheads. Adapted, with permission, from ref. 8.

Download figure to PowerPoint

In WORMS, grade 1 does not represent a morphologic abnormality, but rather a change in signal in cartilage with otherwise normal morphology. Grades 2 and 3 represent similar types of abnormality of the cartilage, focal defects without overall thinning. Grades of 1 and 2 were extremely infrequent among the MRIs read in our study population. Therefore, to create a consistent and logical scale for evaluation of cartilage morphologic change and a fair comparison with radiographic changes in joint space narrowing, we collapsed the original WORMS cartilage scores, which were read on a 0–6 scale, to a 0–4 scale, as follows: the original WORMS scores of 0 and 1 were collapsed to 0, original scores of 2 and 3 were collapsed to 1, and original scores of 4, 5, and 6 were considered 2, 3, and 4, respectively. Using this new scale, cartilage loss at each region was then defined as an increase in score of ≥1.

Statistical analysis.

We examined the relationship between radiographic progression of joint space narrowing and MRI-based cartilage loss for the medial and lateral compartments separately. We considered knees with baseline radiographic joint space narrowing scores equal to 3 (bone on bone) in either the medial or lateral compartment as being ineligible to progress radiographically, so these were excluded from analyses. We used 30-month followup data for analyses unless this information was not available, in which case 15-month followup data were used. Radiographs and MRIs were always compared from the same time point and same knee for each subject.

Cartilage loss was assigned whole number values from 0 (no loss) to 4 (maximum loss), and was analyzed as ordered categories using the proportional odds logistic regression model. A generalized estimating equations correction was applied to the regression model to account for the association in cartilage loss outcome between regions within a knee. This regression model provided adjusted odds of radiographic progression being predictive of cartilage loss. Analyses were adjusted for baseline cartilage scores as well as age, body mass index, and sex. We also calculated the apparent sensitivity and specificity of radiographic assessments as an indicator of cartilage loss, assuming MRI determination of cartilage loss as the “gold standard.” We then described the location of cartilage loss within each compartment when radiographic progression was present or absent. Statistical analyses were performed using SAS software, release 8.2 (SAS Institute, Cary, NC).

RESULTS

  1. Top of page
  2. Abstract
  3. PATIENTS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. REFERENCES

Cohort.

Of the 324 participants, 7 did not undergo baseline MRI, because of either medical contraindications or claustrophobia. Among the 317 participants with baseline MRI, 277 (87%) underwent both followup knee MRI and radiography at either 15 months, 30 months, or both. Of those who had no further followup imaging, 7 had died and 2 had a total knee replacement in the study knee before the 15-month followup visit; 10 only filled out the mail-in surveys for followup, and the remainder either declined or were lost to followup. There were no differences between those who were and those who were not followed up with respect to age (66 ± 9 versus 66 ± 10 years, respectively [mean ± SD]), body mass index (30.8 ± 5.7 versus 29.1 ± 5.6 kg/m2), or baseline Kellgren/Lawrence (K/L) grade (16) (54% versus 55% with K/L grade >2), but those who were not followed up tended more often to be men (83% versus 59%; P < 0.01). We excluded 53 subjects from further analyses, 43 of whom were considered ineligible because of baseline radiographic joint space narrowing scores of 3 (bone on bone) at either the medial or lateral compartment (n = 39) or unreadable radiographic images (n = 4), while the remaining 10 had MRIs that were unreadable for cartilage loss.

The baseline characteristics of the 224 participants who were included in our analyses (Table 1) were similar in age, sex, and body mass index to the 277 participants who were followed up longitudinally. One hundred eighty six (83%) of the 224 participants had 30-month followup data available for analyses, while the remaining 38 had 15-month data.

Table 1. Baseline characteristics of the 224 study participants
  • *

    Baseline cartilage morphology scores ≥1 (equivalent to original Whole-Organ Magnetic Resonance Imaging Score [0–6 scale] of ≥2) in at least 1 of the 5 regions within a compartment.

Age, mean ± SD years66 ± 9
Body mass index, mean ± SD kg/m231.3 ± 5.8
Sex, no. (%) female99 (44)
Kellgren/Lawrence grade ≥2, %72
Cartilage morphology scores ≥1* 
 Medial compartment, %84
 Lateral compartment, %61

Findings in the medial compartment.

During followup, 104 knees (46%) showed loss of cartilage on MRI (defined as an increase in score of ≥1) at ≥1 region in the medial compartment. Increases in cartilage scores of ≥2 in at least 1 region occurred in only 32 knees (14%). Any cartilage loss at ≥2 regions was seen in 48 knees (21%). Loss of cartilage tended to occur most frequently at the central tibial, central femur, and posterior femur regions (Table 2).

Table 2. Location of cartilage loss on magnetic resonance imaging (MRI) in relation to progression of radiographic joint space narrowing in the medial tibiofemoral compartment
Tibiofemoral region% of knees with medial compartment cartilage loss on MRI
All kneesKnees without medial radiographic progressionKnees with medial radiographic progression
Central femur211656
Posterior femur262244
Anterior tibia9819
Central tibia252058
Posterior tibia121117

Thirty-five knees of the original 224 (16%) showed radiographic progression of medial compartment joint space narrowing, defined as worsening by ≥1 grade. Of these 35 knees, 24 (69%) had a loss of cartilage in at least 1 of the 5 regions in the medial tibiofemoral compartment, whereas of 189 knees without radiographic medial joint space progression, 80 (42%) had loss of cartilage noted in at least 1 region (Figure 2). Compared with knees without radiographic progression, if radiographic progression was detected, the odds of cartilage loss being observed at any of the 5 regions were increased (adjusted odds ratio [OR] of 3.7 [95% confidence interval 2.2–6.3]). Using MRI assessments as the gold standard for cartilage loss, the specificity of radiography in detecting cartilage loss in the medial compartment in our study population was 91%; however, the sensitivity was only 23%.

thumbnail image

Figure 2. Example of medial compartment cartilage loss without radiographic progression. A, Baseline (A1) and 30-month followup (A2) semiflexed, weight-bearing knee radiographs. B, Corresponding magnetic resonance imaging of the same knee with spin-echo, fat-suppressed, and T2-weighted coronal views at baseline (B1) and followup (B2). C, Spin-echo, proton-density, and T2-weighted sagittal views at baseline (C1) and 30-month followup (C2). Arrows indicate areas of cartilage loss.

Download figure to PowerPoint

When we examined the location of cartilage loss in knees with and those without radiographic progression, we found that knees with progression on radiographs (versus those without progression) tended to have cartilage loss at central regions of the femur (56% versus 16%) and tibia (58% versus 20%) (Table 2). Loss of cartilage at the posterior femur, a site of loss that would not likely be detected on radiography, also seemed to occur more frequently in knees with radiographic progression compared with knees without radiographic progression (44% versus 22%). This difference seemed to be related to the fact that loss of cartilage in the posterior femur region was often associated with loss at the central femur region in the medial compartment. When we examined knees that exhibited loss of cartilage at the posterior femur but no concomitant cartilage loss at the central femur region, cartilage loss at the posterior femur region was found with similar frequency in knees with and those without radiographic progression (25% versus 18%). Of note, the posterior femur was the most frequent site for cartilage loss when no radiographic progression was observed (Table 2).

Findings in the lateral compartment.

During followup, 49 knees (22%) showed loss of cartilage on MRI (defined as an increase in score of ≥1) at ≥1 region in the lateral compartment. Increases in cartilage scores of ≥2 in at least 1 region occurred in only 11 knees (5%). Any cartilage loss at ≥2 regions was seen in 19 knees (8%). Loss of cartilage tended to occur most frequently at the posterior and central femur regions (Table 3).

Table 3. Location of cartilage loss on magnetic resonance imaging (MRI) in relation to progression of radiographic joint space narrowing in the lateral tibiofemoral compartment
Tibiofemoral region% of knees with lateral compartment cartilage loss on MRI
All kneesKnees without lateral radiographic progressionKnees with lateral radiographic progression
Central femur10829
Posterior femur121213
Anterior tibia3220
Central tibia5342
Posterior tibia5446

Sixteen knees of the original 224 (7%) had radiographic progression of lateral compartment joint space narrowing defined as worsening by ≥1 grade. Of these 16 knees, 9 (56%) had loss of cartilage in at least 1 of the 5 regions in the lateral tibiofemoral compartment, whereas of 208 knees without lateral radiographic joint space progression, 40 (19%) had loss of cartilage noted in at least 1 region. Compared with knees without lateral radiographic progression, if radiographic progression was observed, the odds of cartilage loss being detected at any of the 5 regions in the lateral compartment were increased (adjusted OR 5.7 [95% confidence interval 2.3–14.0]).

Again, using MRI as the gold standard for determining cartilage loss, the specificity of radiography in detecting cartilage loss in the lateral compartment was found to be 96%; however, the sensitivity was only 18%.

We found that in knees with progression of lateral joint space narrowing on radiographs, loss of cartilage occurred more frequently at all regions of the lateral tibiofemoral compartment except the posterior femur, where cartilage loss occurred with similar frequency between knees with and those without radiographic progression (Table 3). The posterior femur was the most frequent site for cartilage loss when no radiographic progression was detected (Table 3).

DISCUSSION

  1. Top of page
  2. Abstract
  3. PATIENTS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. REFERENCES

In this longitudinal study of >200 older men and women with symptomatic knee OA whose disease progression was followed up for up to 30 months, cartilage loss on MRI was more frequently detected than was progressive radiographic joint space narrowing. Although we found radiographic progression to be strongly predictive of cartilage loss in both the medial and lateral compartments, radiographic findings were not a sensitive indicator of cartilage loss, with a substantial proportion of knees with cartilage loss detectable by MRI having no discernable radiographic progression of joint space narrowing. While sensitivity was low, the specificity of radiography for the evaluation of cartilage loss was high, suggesting that if radiographic progression is noted, cartilage loss is likely to be seen.

As would be expected, loss of cartilage tended to occur most frequently in the load-bearing sites of the central femur and tibia, particularly in the medial compartment. Loss of cartilage in these sites was more likely to be detected on radiographs as well. We also found that cartilage loss frequently occurred in the posterior femur, a site that articulates with the tibia during flexion of the knee, and thus is a key potential site for disease occurrence. The presence of cartilage loss along the posterior femur would not be adequately assessed on radiographs, and indeed, we found that among those who had evidence of cartilage loss on MRI without visible radiographic progression, loss of cartilage tended to occur more frequently in the posterior femur region. Overall, the anterior and posterior regions of the tibia were not frequent sites for cartilage loss in either the medial or lateral compartment; this may reflect protection from loss conferred by coverage of the meniscus.

Given the high specificity of radiographic results, we found only a small proportion of individuals who had progression of radiographic joint space narrowing during followup but no detectable changes in cartilage by MRI. When we examined the baseline cartilage scores for these “false-positive” radiographic progressor knees, they tended to be worse than in those who had neither MRI nor radiographic changes (data not shown). The radiographic progression of joint space narrowing in these individuals may have represented meniscal extrusion (2). Further study on the contribution of both meniscal disease and cartilage loss to changes in radiographic joint space width is needed, and will be the topic of another report.

Few studies of knee OA to date have used MRI to assess cartilage loss longitudinally. One study failed to identify substantial cartilage loss, or even radiographic progression, in 11 subjects with knee OA followed up for 3 years (17). Unlike our study, that study used a quantitative measure to assess cartilage loss. However, 2 other studies also using quantitative measures of cartilage identified substantial cartilage loss in subjects with knee OA (11, 18). The largest study, with >100 patients followed up for almost 2 years, did not examine the relationship of cartilage loss to radiographic changes (18). The other study, which demonstrated cartilage loss in >30 subjects followed up for 24 months, showed no apparent relationship to radiographic changes (11); however, the number of subjects was small.

In 2 studies in which semiquantitative scales were used to assess knee cartilage morphology, cartilage loss was demonstrated during followup (7, 19). In 1 study of 20 subjects with knee OA who were followed up for 1 year, significant radiographic changes were not identified despite changes in cartilage morphology (7). No information was provided on the location of cartilage loss detected. The other study involved >40 subjects who underwent repeated MRI after an average of 2 years (19). Although not all subjects had a diagnosis of knee OA, the percentages with cartilage loss in the medial and lateral compartment were comparable with our findings. Similar to our study, those investigators demonstrated that cartilage loss tends to occur primarily at central weight-bearing sites (19). No comparison of cartilage loss with radiographic changes was reported.

While we studied a modest number of subjects and followup was limited to 30 months, we still identified a substantial proportion of individuals with progressive cartilage loss seen on MRI. Our study sample included a greater proportion of men, which reflects the VA population from which subjects were recruited. Although we did not use continuous measures to assess progression of joint space narrowing in this study, we did use a highly reliable, well-validated scale (15). Furthermore, the proportion of subjects with medial and lateral radiographic progression in our study was comparable with findings in other large, community-based, longitudinal studies of subjects with knee OA in which a similar semiquantitative scale was used to assess radiographic joint space narrowing (20). We also used a semiquantitative method to score cartilage change. We believe this serves as a better method for determining location of cartilage loss within the knee than cartilage volume and is more likely to detect focal deficits in cartilage, which was relevant in our comparison between radiographic and MRI results. Finally, we evaluated cartilage loss using a scale that may not be sensitive to modest degrees of focal loss. If we had used a more sensitive scale for cartilage loss, the sensitivity of radiography as a tool would likely have been even lower.

In summary, we found that readily detectable tibiofemoral hyaline articular cartilage loss occurs frequently during 30-month followup in older men and women with symptomatic knee OA, and we provide longitudinal evidence that radiographic joint space progression is predictive of the cartilage loss detected on MRI. Our data also suggest that if progressive joint space narrowing is seen on radiographs, it more often reflects cartilage loss in the central region of the weight-bearing surface than other areas. However, while we found that radiographic progression of joint space narrowing had high specificity for cartilage loss, it was not a sensitive measure, with cartilage loss still occurring frequently, particularly in the posterior femur, when no progression was apparent on radiography. If radiography is used alone in longitudinal studies of knee OA, a substantial proportion of knees with progression of cartilage loss will be missed.

Acknowledgements

  1. Top of page
  2. Abstract
  3. PATIENTS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. REFERENCES

The authors would like to thank the study participants for generously giving their time. The authors would also like to thank all the field staff on this project for their hard work over the years of this study.

REFERENCES

  1. Top of page
  2. Abstract
  3. PATIENTS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. REFERENCES