Background

Symptomatic lumbar spinal stenosis (LSS) is the leading cause of spine surgery in individuals over 65 years of age [1]. The condition is caused by narrowing of the lumbar spinal canal, the lateral recesses and/or the neural foramina, often due to degenerated facet joints and intervertebral discs or thickened flaval ligaments [2]. The diagnostic accuracy of MRI and its association with clinical symptoms and surgical outcome of patients with LSS is controversial [3,4,5]. Nevertheless, MRI helps to select patients for surgery [6], confirm the diagnosis and assess the severity of LSS, define the anatomic location of the stenosis (e.g. central, lateral recess or foraminal), and rule out other conditions that may mimic symptoms of LSS [7]. By localising the apparent cause of the stenosis (e.g. a bulging disc or thickened flaval ligaments), the various MRI findings of LSS can influence the surgical strategy [8]. Therefore, reliable interpretation of MRI findings is an important step towards offering appropriate treatment to patients and to provide optimal information for the surgeon; a shared understanding of the MRI findings between surgeons and radiologists facilitates information exchange and treatment decisions. Previous studies have reported the reliability of different features of MRI in LSS. However, the reported reliability estimates vary widely, even for the same parameters [9,10,11,12]. While some studies have included fair number study subjects, other studies vary greatly in the number and the medical specialty of the observers who performed the studies [9, 13,14,15,16]. In a systematic review, Andreisek et al. [17] found reported observer agreement values for facet joint osteoarthritis varying from poor to excellent; reliability data were lacking for hypertrophy of the flaval ligaments, redundant nerve roots (RNR) of the cauda equina and reduction of the posterior epidural fat. This lack of consensus on the reliability of MRI findings and insufficient reliability data for some parameters highlights the need for studies with higher quality, higher number of the included patients, and observers from relevant medical specialties.

The current study was based on prospective imaging of patients scheduled for surgical treatment of LSS. We hypothesised that given agreement upon the definitions of the MRI parameters used in the diagnosis of LSS, the observer reliability would be good. Thus, the aim of this study was to evaluate the reliability of the commonly used, preoperative MRI findings in the assessment of foraminal and central canal stenosis, facet joint osteoarthritis, RNR and intraspinal synovial cysts, using observers with different levels of experiences.

Methods

The participants in this cross-sectional study were consecutively included from the Spinal Stenosis Trial of the NORwegian Degenerative spondylolisthesis and Spinal STENosis (NORDSTEN) study, a prospective, multicentre, randomised controlled trial that was designed to compare different surgical treatments for LSS [18]. Patients with clinical and radiological findings consistent with LSS were referred to an orthopaedic or neurosurgical outpatient clinic. In total, 437 patients are included in this trial. The inclusion and exclusion criteria are presented in Table 1. All patients were provided written informed consent before inclusion. The current study was performed in accordance with the declaration of Helsinki and adhered to the ICMJE recommendations for the protection of research participants. The study adheres also to the Guidelines for Reporting Reliability and Agreement Studies (GRAAS) [19].

Table 1 Inclusion and exclusion criteria
Full size table

Sample size

The “kappa size package” of the R statistics software (Version 1.2, 2009–2019 RStudio Inc. Boston, USA) was used to calculate the required sample size. Assuming a prevalence of 0.4 and 0.6 in each of the categories of a dichotomous outcome parameter, a power of 80% and a significance level of 5%, a sample size of 102 is required to estimate a kappa (κ) of 0.8 with a 95% Confidence Interval (CI) of 0.7–0.9 for agreement between four independent observers. To account for possible losses, we enrolled 108 consecutive patients in this study.

Imaging

Imaging was performed between February 2013 and March 2015. Since the NORDSTEN study is a large multicentre study involving 18 hospitals, MRI examinations were performed in 1.5 or 3 Tesla units from different manufacturers. To maintain a certain level of quality of the examinations, the performing institutions were provided standard MRI protocols including sagittal T1- and axial and sagittal T2- weighted images. The MRI sequences were performed with repetition time / echo time: 400–826 / 8–14 ms for T1-weighted images and 1500–6548 / 82–126 ms for T2-weighted images, slice thickness: 3–5 mm, field of view: 160–350 mm. All radiological examinations were anonymised, without any link to the demographic or clinical information, the imaging institution, or the manufacturer of the MRI unit.

Image evaluation

MRI examinations were evaluated by two orthopaedic spine surgeons (observers 1 and 2, with ten and six years of experience, respectively) and two radiologists (observers 3 and 4 with 13 and 12 years of experience in musculoskeletal imaging, respectively). They evaluated the images independently and re-evaluated all the images in a random order after a minimum of 6 weeks. This time interval was chosen to assure independency of the test-retest reads. Each observer assessed the image quality for every parameter. If an image was evaluated as inadequate for one or several parameters by one or several observers, that parameter was excluded for all observers. In this way, only parameters with test and retest results from all four observers were included in the final analysis.

The observers used integrated measurement tools in a Picture Archiving and Communication System (IDS7 PACS, Sectra, Sweden) for all measurements. All observers measured the angle between the axial images and a line passing through the centre of the intervertebral disc at each level on the sagittal images. The axial images with more than five degrees of angulation with the intervertebral discs were excluded from the statistical calculations. Proposed classification systems for foraminal stenosis (by Lee et al. [20]), central canal stenosis (by Schizas et al. [21]), and facet joint osteoarthritis (by Weishaupt et al. [22]) were used. Presence or absence of intraspinal synovial cysts [7] and RNR [23] were also rated. To improve the clinical applicability, we dichotomised these parameters into two categories: 0 (normal or mild pathology) and 1 (moderate or severe pathology). The methods used for the categorical ratings and the frequency distribution of these findings are presented in Table 2. Additionally, the observers measured the dural sac cross-sectional area (DSCA) quantitatively at the level of the intervertebral disc. The measurement methods used in this study are presented in the additional file (Grading and measurement methods).

Table 2 Descriptions and frequency distributions of the categorical parameters
Full size table

In a pilot study of ten randomly selected patients from the study population, all observers rated all the categorical parameters and measured the DSCA. The rating criteria and procedures were discussed in two joint meetings between all observers, one before and one after the pilot study. The measurements from the pilot study were not included in the statistical calculations.

Statistical analyses

For the categorical parameters, distributions across the observers were assessed. Gwet’s agreement coefficient (AC1) [24] with 95% CIs was used for computing the inter- and intraobserver agreement. This coefficient is preferred instead of κ when the measurements are not normally distributed, which was the case here (to avoid the so-called high agreement low κ paradox) [25]. For the DSCA, the Intraclass Correlation Coefficient (ICC) with 95% CIs was used to calculate the inter- and intraobserver agreement and the Bland-Altman analysis to estimate the mean differences with 95% limits of agreement.

We used STATA software (StataCorp. 2017. Stata Statistical Software: Release 15. College Station, TX: StataCorp LLC) and its user-written package kappaetc for the statistical calculations. Data from different levels and sides from the same patient were treated as independent observations in the analyses. Distribution of data was assessed by visual inspection of the plots.

Interpretation of the agreement values

AC1 values were interpreted using a scale for κ defined by Landis and Koch [26], as proposed by Gwet [27], to indicate poor (≤ 0.0), slight (0.01–0.20), fair (0.21–0.40), moderate (0.41–0.60), substantial (0.61–0.80) or almost perfect (0.81–1.00) agreement. ICC values were interpreted to indicate poor (< 0.50), moderate (0.51–0.75), good (0.76–90) and excellent agreement (> 91) [28].

Results

In this study, six of the 108 (6%) consecutively enrolled participants were excluded. A flow chart demonstrating the inclusion process and the causes of the exclusions is presented in Fig. 1.

Fig. 1
figure 1

The flow chart of the study demonstrates the inclusion process for the study and causes of the exclusions. LSS: Lumbar Spinal Stenosis, SST: Spinal Stenosis Trial

Full size image

The total number of the rated parameters (after excluding those parameters without ratings from all observers) is presented in Table 2. The mean age ± standard deviation for the study participants was 66 ± 8 years (66 ± 8 for men and 66 ± 9 for women) and 53 (52%) were men. Bar graphs demonstrating the frequency distribution of the categorical ratings across observers are presented in Fig. 2.

Fig. 2
figure 2

Frequency distribution of the categorical parameters: a foraminal stenosis according to Lee et al. b central canal morphology according to Schizas et al. c facet joint osteoarthritis according to Weishaupt et al. d redundancy of the cauda equina and e intraspinal synovial cysts. The values for a, b and c are dichotomised. Category 0 indicates absent or mild pathology and category 1 indicates moderate or severe pathology

Full size image

Using the Bland-Altman method, the mean value of the DSCA was calculated for each of the six observer pairs (from the first ratings) and plotted against the corresponding differences (Fig. 3).

Fig. 3
figure 3

Bland-Altman plots demonstrating the degree of agreement and variability of the measurements of the dural sac cross-sectional area (DSCA) between observers 1 and 2 (a), 1 and 3 (b), 1 and 4 (c), 2 and 3 (d), 2 and 4 (e) and 3 and 4 (f). The solid horizontal lines show the mean differences, and the dashed lines show 95% limits of agreement

Full size image

Interobserver agreement

The overall interobserver agreement was substantial or almost perfect for all categorical parameters (AC1 0.67–0.98, 95% CI range 0.60 to 0.97), except for facet joint osteoarthritis. While the surgeons demonstrated substantial agreement for facet joint osteoarthritis (AC1 0.72), the agreement for the other observer pairs and the overall agreement was fair to moderate (AC1 0.22–0.48). Both pairwise and overall agreements for the DSCA were good or excellent (ICC 0.86–0.96, 95% CI range 0.53–0.97). The DSCA measurements (for both stenotic and non-stenotic levels) ranged from 13 to 283 mm2 and the pooled mean differences between the observers was 5.9 mm2 (95% limits of agreement ±35.5 mm2). The overall and pairwise results of the interobserver agreement are summarised in Table 3.

Table 3 Interobserver agreement
Full size table

Intraobserver agreement

The intraobserver agreement was substantial or almost perfect for all the categorical parameters (AC1 0.63–1.00, 95% CI range 0.56 to 1.0) and excellent for the DSCA (ICC 0.93–1.00, 95% CI range 0.91 to 1.0) for all observers (Table 4). The mean intraobserver difference for the DSCA measurements was 0.17, 95% limits of agreement agreement were ± 34.7.

Table 4 Intraobserver agreement
Full size table

Discussion

In this study, two musculoskeletal radiologists and two orthopaedic spine surgeons with different levels of experience assessed a set of preoperative MRI parameters that may influence the surgical strategy of patients with symptomatic LSS. We found high levels of inter- and intraobserver agreement for all of parameters but for facet joint osteoarthritis. Although previous studies have examined the reliability of various MRI findings in patients with LSS, the reported results vary from low to high for some parameters [9,10,11,12]. In the current study, we included a large number of patients and four observers with different levels of experiences. To enhance the quality of the study, we used suggested GRAAS guidelines for reporting reliability studies [19].

We used a classification system proposed by Weishaupt et al. [22] for evaluation of facet joint osteoarthritis, but achieved only moderate overall interobserver agreement, comparable to the results reported by Weishaupt et al. In a study of 100 patients with symptomatic LSS, Winklhofer et al. [29] evaluated ten qualitative and quantitative parameters on MRI, using two radiologists as observers. They found moderate to substantial interobserver agreement for qualitative parameters describing central canal and foraminal stenosis or nerve root impingement (κ 0.42–0.77) and good agreement for measurement of the DSCA (ICC 0.85). For a dichotomous grading of facet joint osteoarthritis (yes /no), the interobserver agreement was only fair (κ 0.27) and the intraobserver agreement was good (κ = 0.69). The authors speculated that this low interobserver agreement might be related to the challenges of differentiating mild from no osteoarthritis on MRI. In a study by Carrino et al. [11] of 111 patients with lumbar radiculopathy who were candidates for surgery, the interobserver agreement between three experienced radiologists and one spine surgeon was comparable to our results for osteoarthritis of the facet joints (κ = 0.54); the intraobserver agreement was good (κ = 0.69). The authors in the study by Carrino et al. used a four-point consensus-based grading system (normal, mild, moderate, and severe) on MRI. To our knowledge, high reliability values for grading the severity of osteoarthritis of the facet joints on MRI has not been reported. This inferior reliability is worth to note and discuss. Our experience is that degenerative changes in the facet joints cause elongation of the joint spaces in different directions and evaluating these changes on a single axial slice of MRI is challenging. Using multiple axial images (or using them in combination with sagittal images) may improve this evaluation. Previous research indicates slightly more favourable reliability for evaluation of facet joint osteoarthritis on Computed Tomography (CT) than on MRI [30] and we suggest using CT whenever this modality is available. Similar to the studies of Winklhofer et al. and Carrino et al., the results of the current study are limited to patients with symptomatic LSS. We included therefore only MRI parameters that directly or indirectly could influence the surgery, making our results relevant to preoperative findings in patients with symptomatic LSS. In line with a previous report [31], we found good/ excellent inter- and intraobserver agreement for intraspinal synovial cysts. When mimicking symptoms of LSS by causing radiculopathy, intraspinal synovial cysts are treated surgically [7] and a preoperative MRI is necessary to evaluate the status of the adjacent facet joint and the nerve root. Our reliability estimates for presence of RNR were similar to those reported by Papavero et al. [32] for their four-part classification of cauda equina redundancy. This finding is believed to associate with the level(s) of central canal stenosis [33]. In the current study, observer 1 and 4 (a surgeon and a radiologist) demonstrated very high intraobserver agreement values (0.96 to 1.0), but the interobserver values between these observers for Schizas score, PNR and particularly for Weishaupt score were not high. This may indicate different understanding of these findings on MRI and thus an inferior validity of these measurement methods. Observer four demonstrated very high intraobserver agreement for several parameters including the Weishaupt score. This observer is a highly experienced musculoskeletal radiologists who has experience form orthopaedic surgery as well. This wide range of experience may explain the high agreement values.

High observer agreement between experienced radiologists and spine surgeons is important to identify the most relevant MRI findings and to offer appropriate treatment to patients with LSS, yet few reliability studies have considered observers from both specialities to obtain comparable reliability data. Low reliability of the imaging findings may contribute to the existing differences in surgical decision making for these patients [34, 35]. The generally good overall agreement between experienced radiologists and spine surgeons in our study is reassuring for clinical work and research. For example, our results suggest that the differences in DSCA measurements between the observers were < 6 mm2 on average and < 42 mm2 in 95% of cases, when evaluating both stenotic and non-stenotic levels. This limited observer variability is relevant in clinical practice because it indicates that experienced radiologists and spine surgeons perform comparable DSCA measurements.

In a systematic review, Andreisek et al. [17] identified 14 radiological parameters used by researchers for evaluating LSS. This review revealed a wide range of reported κ values (0.01 to 1.00) for inter- and intraobserver agreement. Further, the authors pointed out that the definitions of current categorical imaging criteria for LSS were vague. To address this issue, we used the classification systems of Lee et al. [20], Schizas et al. [21] and Weishaupt et al. [22]. Several of the 14 parameters identified by Andreisek et al. can be recognised in the classification systems we used. For example, the authors reported lack of reliability data for hypertrophy of the flaval ligaments and reduction of the epidural fat, both recognisable through the Schizas classification that we used and demonstrated good reliability for.

In contrast to our results, Speciale et al. [9] demonstrated poor to moderate inter- and intraobserver agreement for evaluation of central canal and foraminal stenosis of patients with symptomatic LSS between seven observers with wide range of experiences. The authors argued that the lack of agreement upon the definitions of these parameters at the outset of the study was the cause of this inferior reliability. It is unclear whether joint meetings between the observers in our study was a cause of higher agreement. However, agreement on nomenclature used, instructional courses for less experienced physicians and maintenance discussions in multidisciplinary meetings between radiologists and spine surgeons may contribute to better observer agreement. In our opinion, the results of the present study and previous studies highlight the need for general radiologists and other health professionals who evaluate MRI examinations of patients with LSS to be updated on the nomenclature.

Limitations

The patients included in the current study were diagnosed with LSS and scheduled for surgery. This knowledge may have introduced bias for the observers during the image evaluations. However, this is the typical clinical scenario for preoperative evaluation of patients with LSS. Selecting patients from a randomised controlled trial with specific inclusion and exclusion criteria may limit the external validity of the results. On the other hand, this was a conscious choice made by our group to increase the internal validity of the results and to address earlier studies reporting low reliability of radiological findings in LSS [9, 10]. Inclusion of patient from a randomised trial may also have introduced some selection bias in the examined sample. However, the performed evaluations in this study included normal, as well as abnormal levels.

Conclusions

The inter- and intraobserver reliability of preoperative MRI findings including stenosis parameters, RNR and intraspinal synovial cysts in patients with LSS is generally good. However, we found only fair or moderate interobserver agreement for facet joint osteoarthritis.