Original Article

*Correspondence to Leena Sharma, Feinberg School of Medicine, Northwestern University, 240 East Huron Street, McGaw M-300, Chicago, IL 60611

 Dr. Peterfy owns stock in Synarc and has received licensing fees from Synarc. Synarc provides services to Novartis, Pfizer, Merck, GlaxoSmithKline, Servier, and other pharmaceutical and biotechnology companies.

Funded by:
  NIH; Grant Number: P60-AR-48098, R01-AR-48748
  National Institutes of Arthritis and Musculoskeletal and Skin Diseases; Grant Number: AR-52599
  National Center for Research Resources; Grant Number: RR-00048

Received: 3 February 2006; Accepted: 3 August 2006

10.1002/art.22618  About DOI

INTRODUCTION

Varus-valgus malalignment of the knee is a risk factor for disease progression and a predictor of function decline in people with osteoarthritis (OA) ([1][2]). The gold standard assessment of varus-valgus alignment (the mechanical axis, or the hip-knee-ankle angle measured from full-limb radiographs) incorporates the mechanical axes of the femur and tibia at the knee (Figure 1) and captures anatomic variations in the proximal femur, femoral shaft, tibial shaft, and ankle. Although full-limb radiography is considered superior to any other approach, there are drawbacks to its wide application, including radiation to the pelvic organs, special radiography room and equipment requirements, and cost.

The situations in which alignment data need to be collected (e.g., epidemiologic studies of knee OA and clinical trials) and in which alignment information might ultimately be applied (e.g., screening programs, physical therapy units, other clinical settings) demand a simpler and more practical method of assessing alignment. An alternative approach to measuring alignment is the anatomic-axis angle measured from the knee radiograph itself, i.e., the angle formed by the distal femur and proximal tibia, or the femur-tibia angle (Figure 1). There are clear advantages to a knee radiograph approach: no radiograph beyond the knee radiograph is required, there is no pelvic radiation, and there are no special room and equipment needs.

Evidence of a correlation between full-limb and knee radiograph alignment measurements has been reported ([3-5]). Also, a recent report described a cross-sectional relationship between full-limb alignment and OA disease features in the expected compartment ([6]). The relationship between alignment and disease features using magnetic resonance imaging (MRI) affords an opportunity to validate the knee radiograph alignment assessment approach, which has not previously been validated using MRI.

The goals of this study, a head-to-head comparison of the knee radiograph (femur-tibia angle) and full-limb radiograph (hip-knee-ankle angle) approaches, were to 1) examine the correlation between the femur-tibia angle by knee radiograph and the hip-knee-ankle angle by full-limb radiograph; 2) calculate the offset of the femur-tibia angle as compared with the hip-knee-ankle angle; 3) calculate the sensitivity and specificity of the femur-tibia angle (first uncorrected and then corrected for the offset) for identifying a knee as varus and as valgus, considering the hip-knee-ankle angle as the gold standard; 4) examine whether varus malalignment severity is worse with worse medial tibiofemoral tissue lesion score (cartilage integrity, bone marrow lesion, meniscal tear, meniscal subluxation, bone attrition) on MRI, not only for malalignment assessed using full-limb radiograph but also as assessed using the femur-tibia angle on knee radiograph; and 5) examine whether valgus malalignment severity is worse with worse lateral tibiofemoral tissue lesion score on MRI for both alignment assessment approaches.

SUBJECTS AND METHODS

Subjects.

Study participants were recruited from the community (through advertising in periodicals targeting elderly people, neighborhood organizations, letters to members of the registry of the Beuhler Center on Aging at Northwestern University, and medical center referrals) and met the following inclusion criteria: the presence of definite tibiofemoral osteophytes (Kellgren/Lawrence [K/L] radiographic grade 2 or higher) in 1 or both knees and Likert category of at least  a little difficulty  for  2 items in the Western Ontario and McMaster Universities Osteoarthritis Index physical function scale. Exclusion criteria were as follows: corticosteroid injection within the previous 3 months; history of avascular necrosis, inflammatory arthritis, periarticular fracture, Paget's disease, villonodular synovitis, joint infection, ochronosis, neuropathic arthropathy, acromegaly, hemochromatosis, Wilson's disease, osteochondromatosis, gout, pseudogout, or osteopetrosis; or exclusion criteria for MRI such as presence of a pacemaker, artificial heart valve, aneurysm clip or shunt, metallic stent, implanted device, or any metallic fragment in an eye.

Approval was obtained from the Office for the Protection of Research Subjects-Institutional Review Board of Northwestern University. Written informed consent was obtained from all participants.

Knee radiograph acquisition and grading.

For knee radiographs, the standing semiflexed protocol was applied ([7]). The knee was flexed until the tibial plateau was horizontal, parallel to the beam, and perpendicular to the film, and the anterior and posterior tibial rims were superimposed. To control for rotation, the heel was fixed and the foot rotated until the tibial spines were central within the femoral notch. Knee position was confirmed by fluoroscopy before films were taken. To describe the knees, the K/L global radiographic scoring system was used by 1 researcher (SNI) with reliability previously demonstrated (  > 0.86). The K/L grades were defined as follows: 0 = normal; 1 = possible osteophytes; 2 = definite osteophytes without definite joint space narrowing; 3 = definite joint space narrowing, some bone sclerosis, and possible bone attrition; and 4 = large osteophytes, marked narrowing, severe sclerosis, and definite attrition.

Assessment of alignment.

Hip-knee-ankle angle by full-limb radiograph approach.

To assess the hip-knee-ankle angle, a single anteroposterior radiograph of both lower extremities was obtained using a 51 × 14-inch graduated grid cassette to include the full limb of tall participants ([8]). By filtering the x-ray beam in a graduated fashion, this cassette accounts for the unique soft tissue characteristics of the hip and ankle. The tibial tubercle, a knee-adjacent site not distorted by OA, was used as a positioning landmark ([9]). Participants stood without footwear, with tibial tubercles facing forward. The x-ray beam was centered at the knee at a distance of 2.4 meters. A setting of 100-300 mA/second and 80-90 kV was used, depending on limb size and tissue characteristics. All radiographs were obtained in the same unit by 2 trained technicians.

Alignment was measured on the full-limb radiograph by 1 reader (SNI) using a protocol we have previously described ([1]). Alignment, i.e., as the hip-knee-ankle angle, was measured as the angle formed by the intersection of the line connecting the centers of the femoral head and intercondylar notch with the line connecting the centers of the surface of the ankle talus and tips of the tibial spines.

Femur-tibia angle by knee radiograph approach.

The femur-tibia angle was assessed by 1 reader (LS) from the knee radiographs (acquired as described above) as the angle created by the intersection of the line from the midpoint of the tibial spines bisecting the femoral shaft with the line from the midpoint of the tibial spines bisecting the tibial shaft. The angle was measured on the hard copy of the knee radiograph using a goniometer with 18-cm arms. The end of each goniometer arm was centered in the bone shaft; the radiograph beyond the distal limit of the goniometer arm was not used. A different reader (SNI), in a separate session from this assessment of the femur-tibia angle, read knee radiographs for K/L grade.

Interrater reliability.

We determined interrater reliability (between LS and SNI) for each of these alignment approaches, and found that it was high for both the hip-knee-ankle angle from the full-limb radiograph (interrater intraclass correlation coefficient [ICC] 0.98; 95% confidence interval [95% CI] 0.96, 0.99) and the femur-tibia angle from the knee radiograph (interrater ICC 0.99; 95% CI 0.98, 0.99).

Definition of varus and valgus.

The definition of varus and valgus alignment is not established. From a pure biomechanical perspective, neutral is a straight line (mechanical axis or hip-knee-ankle angle at 0°), whereas varus and valgus represent any deviation from that line. Therefore, in our primary definition, a knee was recorded as varus when alignment was >0° in the varus direction, neutral when alignment was 0°, and valgus when alignment was >0° in the valgus direction. However, it is not clear whether a knee within a small range of neutral differs from a knee that is 0°, in terms of either in vivo function or outcomes. Within 2° of 0° has been applied previously as a definition for neutral ([1]), and we also considered this alternative definition (i.e., varus defined as alignment >2° in the varus direction and valgus as alignment >2° in the valgus direction) in the analyses.

MRI acquisition and reading.

All participants underwent MRI of both knees using a commercial knee coil and 1 of 2 whole-body scanners (1.5T and 3T). The protocol included coronal T1-weighted spin-echo (SE), sagittal fat-suppressed dual-echo turbo SE, and axial and coronal fat-suppressed T1-weighted 3-dimensional fast low-angle shot sequences.

Using the Whole-Organ Magnetic Resonance Imaging Score system ([10]), 3 regions (anterior, central, and posterior) of the medial and lateral femoral condyles and tibial plateaus were each scored separately for cartilage morphology, subarticular bone marrow lesions, and bone attrition following a detailed reading protocol including visual illustrations of each grade. For each lesion, each region of a compartment surface received its own score.

At each of the 3 regions, cartilage morphology was scored 0-6: 0 for normal thickness and signal; 1 for signal abnormality only; 2 for solitary focal partial- or full-thickness defect  1 mm in width; 3 for multiple areas of partial-thickness loss or a grade 2 lesion >1 mm, with areas of preserved thickness; 4 for diffuse, >75%, partial-thickness loss; 5 for multiple areas of full-thickness loss, or a full-thickness lesion >1 mm, with areas of partial-thickness loss; and 6 for diffuse, >75%, full-thickness loss. Subarticular bone marrow lesions were scored 0-3: 0 for normal; 1 for mild, <25% of region; 2 for moderate, 25-50% of region; and 3 for severe, >50% of region. Subchondral bone attrition was scored 0-3: 0 for normal, 1 for mild, 2 for moderate, and 3 for severe.

The anterior horn, posterior horn, and body of the medial and lateral meniscus were each scored 0-4: 0 for intact; 1 for minor radial or parrot-beak tear; 2 for nondisplaced tear or prior surgical repair; 3 for displaced tear, partial maceration, or partial resection; and 4 for complete maceration and destruction or complete resection. A cumulative tear grade was calculated for each meniscus ([10]). Meniscal subluxation was graded 0 if none, 1 if less than half the meniscus, or 2 if greater than half. MR images were read by 1 of 3 experienced readers. Interrater reliability for the readers applying this scoring system has been published ([10]). The readers for K/L grade, hip-knee-ankle angle, femur-tibia angle, and MRI disease features were each blinded to the others' results.

Statistical analysis.

Data from 1 knee per person (the right knee) were used in the analyses, in keeping with standard approaches to sensitivity and specificity assessment, which rely upon independent observations. Alignment was analyzed as a continuous variable, with values >0° representing the varus direction and values <0° representing the valgus direction. The correlation between the hip-knee-ankle angle by full-limb radiograph and the femur-tibia angle by knee radiograph was estimated, along with the 95% CI of the correlation coefficient. To determine if the 2 alignment measures differed systematically (i.e., whether there was an offset), the femur-tibia angle was regressed on the hip-knee-ankle angle; a statistically significant nonzero intercept indicated a systematic shift in the femur-tibia angle compared with the hip-knee-ankle angle. We repeated this step in men and women separately. The sex-specific offset was added to each femur-tibia angle data point to determine a corrected value.

Sensitivity and specificity were calculated for both varus and valgus alignment (each dichotomized, for both the primary and the alternative definitions) for the knee radiograph approach (first uncorrected and then corrected for the offset) compared with the gold standard full-limb radiograph approach. To further examine the value of using the femur-tibia angle to approximate the gold standard, areas under the receiver operating characteristic (ROC) curve were calculated for identifying varus and valgus alignment. The ROC curve plots the sensitivity against the false-positive rate (1 minus specificity) for different cutoff points of a diagnostic tool. The greater the area under the ROC curve (ranging from 0 or poor to 1 or perfect), the better the ability of the femur-tibia angle for assessing alignment direction compared with the gold standard full-limb hip-knee-ankle angle.

Knees were divided into groups based on compartment-specific MRI scores, and mean alignment values of those without a given tissue lesion (score = 0) were compared with groups having more severe tissue lesions. Dunnett's t-test was applied to validly account for multiple comparisons using   = 0.05 level of testing.

RESULTS

The sample included 146 individuals (109 women) with a mean ± SD age of 70 ± 11 years and a mean ± SD body mass index of 30.6 ± 6.1 kg/m². The proportion of knees at each K/L grade was 4 (2.7%) at K/L grade 0, 18 (12.3%) at K/L grade 1, 68 (46.6%) at K/L grade 2, 46 (31.5%) at K/L grade 3, and 10 (6.8%) at K/L grade 4. Descriptive alignment data are presented in Table 1.

Relationship between femur-tibia angle from knee radiograph and hip-knee-ankle angle from full-limb radiograph.

The femur-tibia angle and the hip-knee-ankle angle were correlated (r = 0.86; 95% CI 0.81, 0.90). The mean ± SD offset was 3.4° ± 0.27°; the femur-tibia angle was, on average, 3.4° more valgus than the hip-knee-ankle angle. We found a less pronounced mean ± SD offset in women than men (3.0° ± 0.5° versus 4.7° ± 0.5°). Mean alignment for the sample as a whole and separately for women and men with each measurement approach is shown in Table 1.

Sensitivity, specificity, and area under the ROC curve associated with the femur-tibia angle from the knee radiograph.

We examined the sensitivity and specificity of the uncorrected femur-tibia angle for the definition of alignment as varus, valgus, or neutral considering the hip-knee-ankle angle as the gold standard. As shown in Table 2, using standard definitions of varus (>0°) and valgus (<0°), the sensitivity for varus identification and the specificity for valgus identification were low. Both improved after correction for the sex-specific offset (Table 2). Results were similar when we applied the alternative definition of varus (>2°) and valgus (<-2°) (Table 3), and when men and women were examined separately.

The area under the ROC curve related to the identification of varus malalignment (>0°) for the femur-tibia angle was 0.91 (95% CI 0.86, 0.96), indicating excellent discriminative ability. In the assessment of valgus malalignment (<0°), the area under the ROC was 0.94 (95% CI 0.89, 0.99). Results were similar for the alternative definitions of malalignment, with areas under the ROC curves of 0.93 (95% CI 0.87, 0.98) for the femur-tibia angle identification of varus malalignment and 0.93 (95% CI 0.87, 0.99) for identification of valgus malalignment.

Alignment as the femur-tibia angle and hip-knee-ankle angle and features of OA disease by MRI.

Next, we divided the knees into groups based on the MRI grade of OA disease features for the medial and lateral tibiofemoral compartments separately. We then determined the mean femur-tibia angle and hip-knee-ankle angle for each subgroup, expecting that as a specific disease feature worsened in the medial compartment, the severity of varus malalignment would increase, and that findings would be similar for the lateral compartment and valgus malalignment severity.

As shown in Table 4, worsening in the status of each medial lesion was associated with greater varus malalignment; this was the case for both the femur-tibia angle and the hip-knee-ankle angle. For example, the difference between alignment for knees with cartilage morphology 0 versus >4 for the hip-knee-ankle angle was 5.3° (95% CI of difference 3.1°, 7.5°) and for the femur-tibia angle was 5.1° (95% CI of difference 2.9°, 7.2°).

Findings were comparable for the lateral compartment; as MRI tissue score worsened, alignment became more valgus with both the hip-knee-ankle and femur-tibia angles (Table 5). Results in Tables 4 and 5 were similar when examined separately in men and women.

DISCUSSION

We found that alignment as the femur-tibia angle measured from knee radiographs was highly correlated with the hip-knee-ankle angle measured from full-limb radiographs. Applying both the standard and alternative definitions of varus and valgus, the sensitivity and specificity of the knee radiograph approach were high after correction for the offset between the 2 measurements. Measurements of the femur-tibia and hip-knee-ankle angles were similarly associated with MRI-assessed disease features in a compartment-specific manner, especially after correcting for the offset. To our knowledge, this is the first report of the sensitivity and specificity of the femur-tibia angle from a knee radiograph for identifying a knee as varus or valgus, and of the validity of this approach using MRI-based measures of OA pathology.

In terms of the correlation between the femur-tibia angle and the hip-knee-ankle angle, our findings are similar to those of prior studies of individuals with knee OA, which found correlation coefficients of 0.98 (femur-tibia angle measured on knee radiographs presumably acquired conventionally in extension) ([3]), 0.75 (femur-tibia angle measured on knee radiographs acquired with the fixed-flexion protocol) ([4]), 0.65 (femur-tibia angle measured on the middle section of the full-limb radiograph itself) ([4]), and 0.88 (femur-tibia angle measured on the full-limb radiograph) ([5]).

Additionally, we found that the knee radiograph-derived femur-tibia angle (corrected for its sex-specific offset) was sensitive and specific in its identification of a limb as varus or valgus, and that the femur-tibia angle and the hip-knee-ankle angle worsened comparably with worse MRI grade of specific medial and lateral OA lesions. We used MRI to examine tissues that theoretically would be stressed by malalignment in a compartment-specific fashion and cannot be visualized by radiography. It is believed that cartilage, menisci, and subarticular bone may be affected by the joint load imbalance of malaligned knees, and that these tissue lesions may worsen the malalignment, resulting in a vicious circle.

In our study, we found that the femur-tibia angle was a mean ± SD 3.4° ± 0.3° more valgus than the hip-knee-ankle angle. This mean valgus angulation offset was less pronounced in women than in men (3.0° ± 0.5° versus 4.7° ± 0.5°). Both the magnitude of the offset and the sex difference are consistent with the results of Kraus et al ([4]). Measuring alignment from the fixed-flexion knee radiograph, Kraus et al found a mean offset of 4.0° in the entire sample, 3.3° in women, and 5.9° in men. This sex difference may be explained in part by the fact that women appear to have greater distal femoral valgus than men ([11]). Given the magnitude of the difference in the offset between men and women, femur-tibia angle results ideally should be corrected using the sex-specific value.

The presence of a physiologic valgus malalignment in the healthy knee (i.e., not considering full-limb landmarks) is a widely held belief that has been the subject of minimal investigation. Some orthopedic clinicians consider 5-7° anatomic (knee radiograph) valgus as neutral and suggest this degree of angulation be used in clinical decision making ([12][13]). The magnitude of this valgus angulation in healthy samples varies across studies ([8][14][15]). The seminal report of Jokio et al ([3]) may have contributed to the frequently taught value of 7°, because this was the magnitude of the difference between the hip-knee-ankle and femur-tibia angles in their study; it is difficult to glean from their report exactly how this offset value was calculated.

This physiologic valgus offset when the landmarks used are only around the knee is in keeping with our finding of high sensitivity of the knee radiograph approach for identification of valgus malalignment. However, without correction for the offset, the specificity for the knee radiograph approach to identify valgus alignment was poor. Conversely, high false-negative rates and low false-positive rates for varus identification were seen prior to offset correction. Correction for the offset yielded high sensitivity and specificity for identification of both varus and valgus alignment. Sensitivity for valgus identification decreased but was still high after correction for the offset, whereas specificity increased. With offset correction, specificity for valgus identification increased from 0.48 to 0.73 and from 0.53 to 0.84 when applying the usual and alternative definitions of valgus alignment, respectively. Knee radiograph identification of varus alignment became more sensitive but slightly less specific after correction for the offset. To more comprehensively evaluate the performance of the femur-tibia measure in malalignment assessment, compared with the gold standard full-limb radiograph, the area under the ROC curve was calculated and revealed an excellent discriminative ability of this approach.

It is important to note that the knee radiographs in our study were acquired using a semiflexed acquisition protocol; it is not clear if these findings will be similar when alignment is measured from knee radiographs obtained conventionally in clinical radiography units. However, it is difficult to identify a rationale for why the semiflexed knee radiograph femur-tibia angle would have a closer relationship to the hip-knee-ankle angle than would the conventional extended-knee radiograph femur-tibia angle, especially since full-limb radiograph is also in the fully extended position. Also, previous reports ([3-5]) of a correlation between the 2 approaches each used different methods to acquire the knee image from which the femur-tibia angle was measured. Still, the relationship between conventional knee radiograph and full-limb radiograph measures should be quantified and the offset determined. There were more women than men in our sample. The proportions are not unexpected given the greater prevalence of knee OA in women in general. In any case, results were similar when men and women were examined separately.

Can alignment be assessed without radiographs? Malalignment that can be detected by visual inspection tends to be at the more severe end of the spectrum, i.e., 5-10° or more in either the varus or valgus direction. In obese individuals, even moderate to severe malalignment may be difficult to see. At best, visualization on physical examination will provide categories of malalignment but not a measurement. Also, in the recent report by Hinman et al ([5]), even in assessing categories, findings from visualization (using the Magee method) were only moderately associated with the hip-knee-ankle angle from the full-limb radiograph.

In the report by Hinman et al, the knee radiograph measurement correlated better with the hip-knee-ankle angle than did any of several other clinical measures ([5]). The recommended nonradiographic approaches were the caliper method and the inclinometer method. The caliper method (measures the distance between the medial knee joint lines in varus knees and the medial malleoli in valgus knees) does not provide a specific alignment assessment for each limb, but rather one value for both limbs and may be influenced by soft tissue characteristics of the lower limb. The novel inclinometer method newly developed by these authors requires a gravity inclinometer attached to a set of calipers, patient positioning, and landmark identification before the angle of the tibia with respect to the vertical is recorded using the caliper. Because knee radiographs are obtained as a part of standard care for patients with knee OA (and very commonly in people with knee OA involved in studies), no additional acquisition is needed for this approach, and the measurement of the femur-tibia angle takes seconds to complete. Nevertheless, the nonradiographic caliper and inclinometer methods may be very useful for investigative or clinical settings in which knee radiograph acquisition is not planned.

In summary, in people with knee OA, the femur-tibia angle from a knee radiograph and the hip-knee-ankle angle from a full-limb radiograph were highly correlated. The femur-tibia angle, corrected for a specific offset for each sex, was sensitive, specific, and associated with high ROC curve areas in identifying varus and valgus alignment, considering the hip-knee-ankle angle as the gold standard. The relationship between malalignment and specific OA cartilage, meniscal, and bone pathology in the mechanically stressed compartment was similar with the 2 approaches. These results suggest that the offset-corrected femur-tibia angle assessment from the knee radiograph is an acceptable alternative to full-limb radiography, and should be considered in research and clinical settings.

AUTHOR CONTRIBUTIONS

Dr. Sharma 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. Issa, Chang, Peterfy, Cahue, Marshall, Hayes, Sharma.

Acquisition of data. Issa, Prasad, Guermazi, Cahue, Marshall, Kapoor, Sharma.

Analysis and interpretation of data. Issa, Dunlop, Chang, Song, Guermazi, Peterfy, Hayes, Sharma.

Manuscript preparation. Issa, Dunlop, Chang, Guermazi, Peterfy, Hayes, Sharma.

Statistical analysis. Issa, Dunlop, Song, Hayes.