Study design and sample
This is a retrospective database-based cohort comparison study. Reporting of the present study follows the STROBE Statement guidelines for reporting observational studies [10]. The inclusion criteria were symptomatic lumbar spinal canal stenosis requiring surgical decompression without fixation and availability of preoperative MRI that were performed in a scanner with at least 1.5 Tesla, including sagittal T1- and T2-weighted images and axial T2-weighted images in the picture archive and communication system (PACS) of the institution. Exclusion criteria were previous history of lumbar spine surgery, lumbar deformity as scoliosis or vertebral slip requiring fixation and congenital, traumatic, infectious or neoplastic diseases of the lumbar spine.
Sample size was calculated with the use of G*Power version 3.1.9.2 (Psychology Department, Duesseldorf University, Germany) [11]. For sample size calculation the variable LSS-level was chosen and the following assumptions were used: 68% of RNR+ patients show one stenotic level and 32% show two or more stenotic levels; oppositely 84% of RNR- patients show one stenotic level and 16% show two or more stenotic levels. Based on these assumptions an odds ratio of 2.47 was calculated. Thereby, if α = 0.05 and 1-ß error probability = 0.90, there is a 90% chance of correctly rejecting the null hypothesis that a particular value of the main predictor variable (LSS-Level) is not associated with the outcome variable, with a total sample size of 300 patients (150 per group).
The preoperative data of 300 consecutive LSS patients who underwent single- or multi-level microsurgical bilateral decompression via a unilateral approach (also known as “cross over” or “over the top” technique) without any fixation were evaluated. The surgeries were performed by six different surgeons with a level of experience ranging from 4 to 35 years. The ipsilateral facet was resected one third and the contralateral was left alone whereas the thickened yellow ligament was completely removed. The surgeries were performed between December 2012 and August 2016 at the same institution. During this time window 2273 patients underwent decompression surgery for LSS. Thereof 2113 underwent decompression surgery without fixation. Out of this second group patients with and without RNR on their preoperative MRIs were selected from August 2016 backwards, until both groups each contained 150 patients.
The Ethics Committee of the Federal State of Hamburg deliberated upon the research proposal of the present study (File PV5817). According to the ethics committee retrospective database-based studies do not require an approval, whenever the data was acquired, saved and treated anonymously. This applies to the present study.
The database used for this research is not publicly available, it is property of the Schoen Clinic Group, whose access is regulated by the rules of procedure of its in-house Science Office.
The following patient-related and MRI-based factors were used as potential predictors: age, gender, body height (BH), length of the lumbar spine (LLS), segmental length of the lumbar spine (SLLS), relative LLS (rLLS), relative SLLS (rSLLS), the amount of lumbar spine alignment deviation (LSAD), as given by the difference between SLLS and LLS, the number of stenotic levels involved (LSS-level) and the grade of severity of the stenosis (LSS-grade) on a progressive scale from A to D [12].
Firstly, the 300 patients were assigned to either the RNR+ or the RNR- group by a senior radiologist, a senior orthopedic surgeon and a senior neurosurgeon independently. Their experience levels were 15, 10 and 35 years respectively.
The definition of RNR used to assign the patients into the groups was the following: RNR were defined as serpentines [13] when in sagittal T2-WI a sinusoidal deflection (complete crest-trough wave) occurred within the height of a vertebral body without any horizontalization of the involved roots (Fig. 1b). RNR were defined as loops when in sagittal T2-WI at least in two different areas dots or horizontalized roots (Fig. 2a) were combined with tortuous roots in the axial T2-WI (Fig. 2c). Mixed serpentine and loop findings were scored as loops, as the latter deformation seems to be the more relevant one [14].
The agreement between the three raters concerning patients’ group affiliation was almost perfect (Fleiss k = .92; p < 0.001). The transition between a normal course of the cauda equina nerve roots and a very beginning type of serpentine RNR is sometimes subtle and may lead to disagreements between the raters. In such cases the amount of straight roots on the one side of the reference stenotic level and the amount of serpentine RNR on the opposite side of the stenotic level was evaluated. If the pathologic pattern (serpentine RNR) was agreed to be prevalent (most of the roots show a serpentine shape) the case was considered as RNR+. Eighteen disagreements were reclassified in a consensus conference. Secondly, LLS and SLLS were measured. Finally, LSS-level, and LSS-grade were assessed for each patient.
Length of lumbar spine (LLS) and segmental length of lumbar spine (SLLS) measurements
Three authors (LP, JL, TF) measured LLS and SLLS independently on the sagittal T2-weighted slice showing the midplane of the conus using the AGFA Impax 6 software (AGFA Health Care, GmbH, Bonn, Germany). For LLS measurements a straight line was drawn from the posterior-superior corner of the L1 vertebral body to the posterior-superior corner of the S1 vertebral body (Fig. 3, red line). For SLLS measurements a line was drawn from the posterior-superior corner of the L1 vertebral body to the posterior-superior corner of the L2 vertebral body. The procedure was repeated until the line reached the posterior-superior corner of the S1 vertebral body (Fig. 3, blue line). LLS and SLLS were both determined by the length of the line (mm) [15]. Inter-rater reliability for both measurements was tested previously. The estimated intraclass correlation coefficient (ICC) calculated with a two-way mixed effects model with an absolute agreement definition was .99 (95% C.I. ranging from .98 to .99) and .99 (95% C.I ranging from .97 to .99) for LLS and SLLS measurements, respectively.
Calculation of rLLS and rSLLS
Absolute LLS and SLLS values were used to compute relative (%) rLLS and rSLLS values in relation to the patients’ body height.
Calculation of the amount of lumbar spine alignment deviation (LSAD)
The arithmetic difference between the SLLS and LLS values of each patient was calculated as an indicator of the degree of alignment deviations of the lumbar spine (LSAD). Greater differences are caused by higher degrees of alignment deviations such as hyper-lordosis or scoliosis.
Qualitative assessment of LSS-grade
There is no consensus regarding the specific diagnostic criteria for lumbar spinal stenosis (LSS) based on magnetic resonance imaging (MRI) [16]. A qualitative grading system based on the root-cerebrospinal fluid (CSF) relationship was described by Schizas et al. and was found to have a prognostic value [12]. The classification includes four progressive LSS grades, with grades A and B usually responding to conservative treatment, while grades C and D often require surgical decompression [17] (Fig. 4).
Three raters independently classified the LSS-grade of the patients, and the few cases with classification discrepancies were discussed in a consensus conference.
Quantitative assessment of LSS-level
The number of LSS-levels involved was assessed on the MRI images. A level was defined as stenotic if affected by a grade B or higher narrowing of the spinal canal. Patients were classified in three groups according to the number of stenotic levels: group 1 = one stenotic segment, group 2 = two stenotic segments, and group 3 = three or more stenotic segments involved.
Statistical analysis
The study sample was characterized with the use of mean ± standard deviation (SD) values for continuous variables (age, BH, LLS, SLLS, rLLS, rSLLS, LSAD) and frequencies for categorical variables (gender, RNR, LSS-grade, LSS-level). Demographic data comparisons between the groups were performed with t-tests for independent samples for continuous variables. In cases in which the variable data were expressed in frequencies, chi-square tests were used to test for group dependency. Binomial logistic regressions were carried out to investigate whether the presence of RNR could be predicted by patient demographics and MRI-based measurements. Age, gender, BH, LLS, SLLS, rLLS, rSLLS, LSAD, LSS-grade and LSS-level were considered as independent variables (potential predictors). The dependent variable was group affiliation (RNR+ or RNR-). For the logistic regression LSS-grade categories A and B and LSS-levels 2 and 3 were merged due to the low number of cases in one of the categories. Single predictors were tested in the 10 models. IBM SPSS software version 21 for Macintosh (IBM Corp. Armonk, New York) was used for all statistical analyses. The 0.05 level of probability was set as the criterion for statistical significance.