Vertical Relationship Ratio (VRR): A High Reliability and Accuracy Radiographic Measurement for Basilar Invagination in Rheumatoid Arthritis

Background

Rheumatoid arthritis (RA) is a chronic, systemic, inflammatory disease. It primarily affects the lining of the synovial joints and can cause progressive disability [1]. Although the joint involvement is mainly in the hands, wrists, elbows, shoulders, and knees, cervical spinal joints can also be affected. Erosive pannus formations at the atlantooccipital and atlantoaxial joints cause bony destruction and ligamentous laxity, thereby weakening and collapsing lateral masses. After these devastating consequences, basilar invagination or impression (BI) occurs [2]. The BI in RA patients has been referred to using several terms: cranial settling, atlantoaxial impaction, vertical settling, or superior migration of the odontoid process. Although these terms are defined slightly differently, BI is primarily considered to be projection of the tip of the odontoid process into the already limited space of the foramen magnum [3]. This BI affects brain stem function, leading to compromise of the autonomic center and increased morbidity [4,5]. Therefore, surgical intervention has been recommended to prevent these poor outcomes [3,6–8].

Normally, the diagnosis of BI is based on advanced imaging modalities, such as computed tomography (CT) and magnetic resonance imaging (MRI). However, the diagnosis of BI by using plain cervical spine radiographs remains valuable because of its cost-effectiveness and greater accessibility [9]. There are several available measurement methods based on plain radiographs of cervical spine to diagnose BI in RA patients. These measurements include the McRae line [10], McGregor line [11], Chamberlain line [12], Ranawat method [8], Redlund-Johnell method [13], and Clark station method [7]. However, most of these plain radiographic measurements for BI have common problems caused by the obscured anatomical landmarks and magnification [14,15], and lower sensitivity to detect the BI in each previous radiographic measurement techniques (61% in Redlund-Johnell, 71% in the Ranawat, and 83% in the Clark station criteria). In addition, Riew et al reported that the intraclass correlation coefficient scores (ICCS) with inter-observer and intra-observer of McRae line, Ranawat methods, and McGregor line had poor correlations. They suggested combining 3 radiographic measurement methods (Ranawat, Redlund-Johnell, and Clark station) to increase sensitivity and negative predictive values [15]. To the best of our knowledge, now there is no single method that is highly reliable and accurate in diagnosis of BI in RA patients. Therefore, the aim of this study was to develop a new radiographic measurement method and evaluate its reliability and accuracy in diagnosis of BI in RA patients. We also compared our new measurement methods with previously available measurement methods in terms of sensitivity, specificity, and cutoff value in diagnosis of BI.

Material and Methods

ETHICS APPROVAL:

The Ethics Committee of the Faculty of Medicine, Siriraj Hospital, Mahidol University, Bangkok, Thailand approved this study (COA no. Si 596/2018), and the guidelines and regulations of the Ethics Review Board were followed for all material and methods. Because this study was retrospective, the requirement for informed consent of each patient was waived.

PARTICIPANTS:

This study retrospectively enrolled 28 patients diagnosed with RA and BI between 2005 and 2020 (the “RA with BI” group). All had been diagnosed with RA by rheumatologists at the Faculty of Medicine Siriraj Hospital using the American College of Rheumatology criteria. BI was diagnosed if sagittal CT or MRI images showed the following: 1)The odontoid process tip passing through the foramen magnum, 2) The lack of ventral subarachnoid space, 3) The odontoid process indenting the medulla at the level of the foramen magnum [16]. Another 2 groups were recruited: “RA without BI” and “non-RA.” All patients had lateral cervical spine scans obtained through MRI or CT imaging. Individuals were excluded from the RA without BI and non-RA groups if they had: (1) congenital bony abnormalities of the craniovertebral junction; (2) fractures, tumors, or infections of the craniovertebral junction; or (3) prior craniovertebral junction surgery.

RADIOGRAPHIC MEASUREMENT AND DATA COLLECTION:

Two independent orthopedic surgeons were involved in evaluating plain lateral radiographs of the cervical spine. The observers independently measured and analyzed the radiographs without knowing the related diagnoses. Measurements were made with electronic calipers in the hospital’s picture archiving and communication system. Each radiograph was assessed twice by each surgeon, with at least a 1-month interval between the 2 measurements. To prevent recall bias, the order of the radiographs was randomly changed for the second set of measurements, with the 2 observers were blinded to the new sequence. Each surgeon used the following 5 methods to determine the presence of BI on the plain radiographs:

REDLUND-JOHNELL METHOD: The observer marked the midpoint of the base of C2 and then measured the minimum distance between that point and a line drawn from the posterior hard palate to the lowest point on the midline occipital curve. BI was indicated when the distance was less than 34 mm in males and 29 in females [13].

RANAWAT METHOD: The observer measured the perpendicular distance from the center of the pedicle of C2 to a line which crosses along to the transverse axis of C1. BI was indicated when the distance was less than 15 mm in males and 13 mm in females [8].

MODIFIED RANAWAT METHOD: The observer marked the midpoint of the base of C2 and drew a line from the center of the anterior arch of C1 to the center of the posterior arch. The distance between the 2 lines along the long axis of C2 was measured [17].

CLARK STATION METHOD: The observer identified the station of C1, determined by the level of the anterior ring of the atlas on the 3 parts of the odontoid process. BI was diagnosed if the anterior ring of the atlas was level at the middle third (station II) or caudal third (station III) of the odontoid process [7].

VRR METHOD: The ratio was calculated by dividing the distance obtained with the modified Ranawat method by the mean of the vertebral body height of C3 (Figure 1). The mean of the vertebral body height of C3 was derived by the formula (anterior vertebral height+posterior vertebral height)/2.

We compared the VRR with other present measuring methods by using the radiographic parameters in sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). In addition, we analyzed the ROC curve of the VRR to establish the cut off for the RA with BI group. The true positive (sensitivity) rate was plotted as a function of the false-positive (100-specificity) rate to obtain an optimal possible cut off value. The area under the ROC curve was used to identify the optimal cutoff for distinguishing cases of RA with BI from cases that did not have BI (the RA without BI and the non-RA groups).

STATISTICAL ANALYSIS:

The demographic data and radiographic measurements were analyzed and reported. Categorical data were compared using the chi-squared test. Continuous data were checked for normal distribution by using the Shapiro-wilk test. Normally distributed continuous data were compared using an independent t test. The results are presented as mean±standard deviation. P value less than 0.05 are statistically significant. The intra- and inter-observer reliabilities were examined using intra-class coefficient (ICC) scores with 95% confidence intervals. Good reliability was denoted by ICC values between 0.75 and 0.9, while excellent reliability was signified by ICC values exceeding 0.9.

For the Clark station measurement method, there was calculated agreement with the Kappa value because of ordinal scale data. Moderate agreement was 0.41–0.60, substantial agreement was 0.61–0.80, and perfect agreement was 0.81–1.0. The sensitivity, specificity, PPV, and NPV of each radiographic parameter were calculated. ROC analysis was also used to determine the cutoff of the VRR. All these analyses were conducted using SPSS Statistics for Windows, version 18.0 (SPSS Inc, Chicago, IL, USA) and MedCalc, version 17.6 (MedCalc Statistical Software, Ostend, Belgium).

Results

CLINICAL, INTER-OBSERVER, INTRA-OBSERVER RESULTS:

This study enrolled 121 patients (28 in the RA with BI group, 37 in the RA without BI group, and 56 in the non-RA group). Most patients in the RA group with BI and without BI were female and older than non-RA patients (Table 1). The intra-observer reliabilities of the VRR, Redlund-Johnell, and modified Ranawat methods were excellent, with ICC values of more than 0.9. The Ranawat method had the lowest intra-observer reliability (ICC values of 0.880 and 0.894 for the 2 raters). The inter-observer reliabilities of all radiographic parameters were also excellent (ICC values, 0.919–0.962; Table 2). The Clark station method showed moderate to perfect agreement for intra-observer reliability (kappa, 0.695 and 0.915; P<0.001) and moderate to substantial agreement for inter-observer reliability (kappa, 0.590 and 0.739; P<0.001)

All parameter measurements for the Redlund-Johnell, Ranawat, modified Ranawat, Clark station, and VRR methods demonstrated statistically significant differences in distinguishing RA patients with BI from those patients in other groups. The mean VRR for the RA with BI group was 1.82±0.20, which was lower than the means of the other groups (2.24±0.19 in RA without BI, and 2.26±0.19 in non-RA). The other parameters (Redlund-Johnell, Ranawat, and modified Ranawat methods) were also lower for the patients in the RA with BI group than those patients without BI. In addition, there was a higher degree of Clark station in the RA with BI group than in the 2 other groups (Table 1).

ROC CURVE RESULTS:

The ROC curve of the VRR was analyzed. The optimal cutoff for the VRR was below 2.025 and the area under the ROC curve was 0.987 (P<0.001; Figure 2). The true-positive rate (sensitivity) was 96.43%, and the specificity was 97.85%. Moreover, the NPV for this cutoff in this population was 97.85%; this was excellent for interpreting a subject with a negative finding who does not have BI. The PPV that indicated the likelihood that a subject with a positive finding had BI was also excellent, at 92.85%. The sensitivity and NPV values of the VRR were greater than those for the other tests, even when the Redlund-Johnell, Ranawat, and Clark station methods were combined. Additionally, the VRR’s optimal cut off provided high sensitivity, specificity, PPV, and NPV for both sexes (88.89%, 100%, 100%, and 95.83% in males, 94.74%, 97.14%, 90.00%, and 98.56% in females, as shown in Table 3)

Discussion

STRENGTHS AND LIMITATIONS:

There are several strengths in our study. First, our study included all available measurement methods and demonstrated the effectiveness of each measurement method in diagnosis of BI. Second, we used 2 independent evaluators blinded in evaluation to prevent biased measurement of radiographic data.

Nevertheless, our study still has some limitations. First, it was a retrospective, case-control study that might have had a selection bias. Second, the BI in this study focused on RA, but other causes of BI were not included. Third, the study’s participants were mainly females, who are naturally predominant in the RA population. Moreover, the small population in this study was drawn from patients with BI, which is a rare disease.

Conclusions

VRR has the excellent intra-/inter-observer reliabilities and could distinguished BI in RA patient. We recommend using the VRR in preference to the other available methods for assessment and screening BI. In addition, the VRR cutoff value less than 2.025 in plain radiographs of the lateral cervical spine indicates the presence of BI with high sensitivity and specificity.