Thoracolumbar spines (T12–L5) were harvested from freshly frozen (−20°C) human cadavers (mean age 72.1 years, range 53–89 years). None of the deceased subjects had any history of spinal injury, spinal surgery or spinal metastatic disease. The spines were thawed before assessment and biomechanical testing. Excessive soft tissue and muscle tissue were carefully removed, keeping the anterior and posterior longitudinal ligaments as well as the facet joints intact (Fig. 1).
For assessment of spines, we used clinically relevant and methodologically validated parameters of lumbar spinal degeneration as recommended by the European Spine Society . Grading methods for disc degeneration with an intraclass correlation coefficient or an interobserver κ > 0.60 [5, 13, 17] were included. For facet joint degeneration, grading schemes  with an intraclass correlation coefficient or interobserver κ > 0.40 were used in the present study [16, 26].
Magnetic resonance imaging (MRI, Siemens© Symphony 1.5 T: Syngo MR A30, software NUMARIS/4, Berlin, Germany) of lumbar spines was performed to assess intervertebral disc degeneration according to Griffith and Pfirrmann [5, 17] and facet joint degeneration according to Weishaupt . Disc degeneration, (including narrowing and osteophytes, respectively Lanes 1 and 2) [13, 27] and facet joint degeneration  of levels L2–L3 and L4–L5 were also assessed based on radiographs (Sedical© Digital Vet. DX-6, Arlington Heights, IL, USA). Furthermore, MRI was used to assess the presence of Modic changes  and Schmorl’s nodes  and to determine intervertebral disc and pedicle geometry and facet joint angles . Disc geometry included: disc length, width, height, surface area, and volume. Disc surface area, disc volume and pedicle diameter were calculated assuming an elliptic shape (surface = 1/4π × length × width). For pedicle diameter, an average of left and right pedicles was taken for the top (L2 or L4) and bottom (L3 or L5) of each segment. Mean facet joint angle was calculated by averaging left and right angles per segmental level (L2–L3 or L4–L5) while facet joint angle differences or tropism was determined by calculating the difference between left and right facet joint angles. Segmental frontal surface area (FA), defined in cm2, bone mineral content (BMC in g) and bone mineral density (BMD in g/cm2) of lumbar spinal sections (L2–L3 and L4–L5) were measured with dual X-ray absorptiometry (DXA, Hologic© QDR 4500 Delphi DXA scanner, Waltham, MA, USA) in anteroposterior direction. All assessments were performed using Osirix software (Osirix©, version 3.8.1., Pixmeo SARL, Geneva, Switzerland).
Specimen preparation and biomechanical testing
L2–L3 and L4–L5 motion segments were isolated from each spine. Subsequently, laminectomy was performed at level L2 of five randomly chosen spines, and at level L4 of the remaining five spines. Laminectomy, analogous to standard clinical practice, was performed by removing the spinous process and part of the lamina, leaving the pars interarticularis intact. During preparation, examination, and biomechanical testing, specimens were kept hydrated using 0.9% saline-soaked gauzes. Thoracolumbar spines with bridging osteophytes, assessed on anteroposterior, lateral and oblique radiographs, were excluded from this study. After sectioning spines in L2–L3 and L4–L5 motion segments, the motion segments were potted in a casting-mould using low melting point (48°C) bismuth alloy (Cerrolow-147; 48.0% bismuth, 25.6% lead, 12% tin, 9.6% cadmium, and 4% indium). The upper and lower vertebral bodies were fixed securely into the alloy by adding screws into the vertebral body. Screw fixation was reinforced with orthopaedic bone cement (Simplex, Stryker©, Kalamazoo, MI, USA). The disc was placed parallel to the flat surface of the bismuth. Discs were placed parallel based on the visual inspection. Because muscle tissue was thoroughly and carefully removed, the intervertebral disc and corresponding endplates were clearly visible. All articulating parts were kept free. The casting mould was placed in a hydraulic materials testing machine (Instron©, model 8872, Norwood, Canada) [1, 23, 24]. The caudal vertebral body was fixed on a plateau that allowed movement in axial and transverse directions only. Transverse movements were allowed, so segments were able to find their physiological motion patterns and to correct for possible differences in embedding. Segments were loaded with a continuous axial compressive force of 1600 N [23, 24], applied using a pneumatic cylinder that had been calibrated using a load cell (Hottinger Baldwin Messtechnik©, Force Transducer Type C2, Darmstadt, Germany). Since compression was applied in a purely axial direction, bending moments were minimized. The level of compression simulated the force during bending, a condition in which high shear loading of the lumbar spinal segments typically occurs . Subsequently, while maintaining the axial load, anterior shear load was applied with a constant rate of 2.0 mm/min on the casting mould containing the cranial vertebral body, until failure of the vertebral motion segment . This test set-up was similar to mechanical testing by Bisschop et al. , van Solinge et al.  and van Dieën et al. . An anterior shear force was used since it corresponds to the loading direction in vivo [10–12, 22]. The test was stopped after hearing a crack or after a large force reduction was seen. Shear force and displacement were digitized and stored at 100 samples per second (Instron© Fast Track 2, Norwood, Canada).
For each of the 20 motion segments tested, SFF was determined. SFF was defined as the point at which maximum load was recorded in the load–displacement curves for each specimen. These data were presented previously . Shear yield force (SYF) was defined as the point at which shear load caused a decrease in stiffness, i.e. a decrease in the slope of the load–displacement curve. Average SS was calculated from the load–displacement curve, between 25 and 50% of the SFF. SS was estimated by means of a least squares fit of a straight line through the data with the slope of the regression line representing stiffness. The deformation in this region was linear, with an r
2 > 0.943 (Table 1) between load and displacement for all motion segments. All analyses were performed using computer programs written in Matlab (Mathworks ©, Natick, MA, USA).
Statistical analysis was performed based on two separate groups. The first group contained untreated segments (5× L2–L3 and 5× L4–L5) while the second group consisted of segments with laminectomy (5× L2–L3 and 5× L4–L5).
Independent variables were classified as: general variables, intervertebral disc geometry (MRI), pedicle geometry (MRI), facet joint orientation (MRI), bone characteristics (DXA), intervertebral disc degeneration classifications (MRI), intervertebral disc and facet joint degeneration (Radiographs), facet joint degeneration (MRI) and other (MRI). These classes of variables are specified in Table 2.
First, relations between independent and dependent variables (SS, SYF and SFF) were tested for each individual variable. For dichotomized independent variables (segment, sex, Modic changes  and Schmorl’s nodes , independent-sample t tests were used while Pearson’s coefficient of correlation was determined for continuous and ordinal values. Note that it was thus assumed that ordinal variables (Pfirrmann , Griffith , Lane 1 , Lane 2 , Wilke , Pathria  and Weishaupt ) represent a linear degree of severity.
When independent variables were associated with a dependent variable, here defined as independent-sample t test: p < 0.05 or as a bivariate correlation with a significance level of: p < 0.05, they were used for the combined statistical models.
Before final analysis was performed, all independent variables were checked for correlations with each other. In case a correlation >0.7 with a p < 0.05 was found, the independent variable with the strongest effect on the specific dependent variable was included in the model. Finally, backward linear regression techniques were used to create final statistical models per dependent variable per treatment group.