Design
The study was a longitudinal population-based observational study.
Study population and material
This study used the magnetic resonance images from the Danish longitudinal cohort-study, ‘Backs on Funen’, which investigated potential risk factors of low back pain (LBP) in a general population [10]. In brief, the Office of Civil Registrations generated a sample from all Danes aged 40 years in 2000 and living in the county of Funen, an island in Denmark with about 450,000 inhabitants. An invitation letter was mailed to a random sample of 11 % of these 40-year-olds, corresponding to 625 people. Reasons for exclusion were severe disability, ferromagnetic implants, claustrophobia or inability to communicate in Danish [10]. Of these people, 412 (66 %) consented and participated in the baseline measurements (Time 1) and 48 % were male. At Time 1 they had a clinical examination, lumbar MRI and completed a questionnaire. Four years later (Time 2), 348 participants (56 %) completed the first follow-up visit (46 % were male), and another four years later (Time 3), 293 participants (47 %) completed the second follow-up visit (46 % were male). At these follow-up visits, the participants repeated both the questionnaire and lumbar MRI. Details about the socio-demographics and back pain have been published previously [11, 12]. Approximately, 70 % of the cohort reported back pain within the past year at each time point. Ethics approval was granted for the original study [10] from the Ethics Committee of Vejle and Funen Counties (approval no: 20000042) and for access to the database by the Danish Data Protection Agency (approval no: 2000-52-0037). All participants gave their informed written consent prior to study enrolment.
Lumbar disc levels with broad-based or focal protrusions, extrusion or sequestration at Time 1 or Time 2 were selected for this study according to the criteria outlined by the ‘Combined Task Force’ [13].
MRI
MRI scans were performed with an open, low field 0.2 T magnetic resonance unit (Magnetom Open Viva, Siemens AG, Erlangen, Germany). The lower thoracic and lumbar regions were scanned with subjects in the supine position, using a body spine surface coil. Sagittal T1- and T2-weighted and axial T2-weighted MRI images were performed with axial images placed in the plane of the five lower discs. For further details, see the original study [10].
Definition of LDH
The qualitative evaluation of LDH was made by an experienced musculoskeletal research radiologist, who demonstrated excellent reliability in the rating of disc contour in the same cohort of people [10, 14]. The intra-and inter-observer agreement for the evaluation of disc contour was substantial, kappa =0.78 ((95 % confidence intervals (CI) 0.64-0.91) for intra-observer and kappa = 0.68 (95 % CI 0.55-0.81) for inter-observer [14].
Measures from MRI
Quantitative measures of disc height, LDH size and dural sac area were performed following a newly developed method for this purpose [15].
Anterior and posterior intervertebral disc heights were expressed as cross-sectional areas (CSA) calculated from measures of disc height from each sagittal image section plus the slice thickness and inter-slice gap (Fig. 1). The average disc height was calculated using the formula: (anterior intervertebral height + posterior intervertebral height)/2 [16]. LDH and dural sac CSA were also calculated using combined length measures from sagittal images and evaluated for each segment, which was given an LDH rating by the radiologist. Full details about the measurement protocol have been described in a separate manuscript [15]. The measurements were performed using the free open-source software OsiriX (version 4.1.2). This version of OsiriX is designed for scientific use [17].
All MRI measurements were conducted by a student completing a Master in Clinical Biomechanics (AT), who had previous MRI measurement experience from another study which quantified the reproducibility of the current method [15]. To ensure that the rater was blinded to participant information during measurements, all participant images were anonymised.
Validity of measurements
The levels of intra- and inter-rater agreements of the measurements were evaluated in a previous study where they were found to be between acceptable and good [15]. The reliability of CSA calculations was also evaluated in the same study and was found to be acceptable.
Data manipulation
Custom-made software was used to calculate length and cross-sectional areas based on the X and Y coordinates, slice thickness and inter-slice gap. The method and the software have been described in detail elsewhere [15].
Data validation
All calculated results were validated and checked for consistency with the images of the Region Of Interest (ROI) for each measurement. All values were examined in Excel files for identification of outliers and all potential outliers were validated against the ROI files. In addition, a systematic selection of approximately every tenth participant was screened for errors using ROI files.
Data analysis
Four-year changes were defined as changes in measurements from Time 1 to Time 2 for disc levels with LDH observed at Time 1, or from Time 2 to Time 3 for disc levels with LDH observed at Time 2. Eight-year changes were defined as changes in measurements from Time 1 to Time 3, for disc levels with LDH observed at Time 1 only.
The changes in size over time at a group level were summarised in tables with means and 95 % CI for LDH size, dural sac areas, and disc heights for each of the three time-points by lumbar level. Changes in size were reported for each disc level, as well as for all disc levels combined. The reporting of mean values instead of median values was chosen after testing for normal distribution.
The summary statistics for changes in size of herniations at an individual level were conducted producing trajectories for LDH sizes defined as ‘unchanged’, ‘increased’ or ‘decreased‘ based on Limits of Agreement (LOA) for the measurements [18]. LOA for disc height was 79.9 mm2, for LDH 58.9 mm2, and for dural sac area 69.9 mm2 [15]. A change was only reported if the absolute value was larger than the LOA. For the eight-year analysis covering all three time-points, an extra category, ‘fluctuating’ was added for those who did not remain in the same category throughout the time period (e.g. a pattern of unchanged between Time 1 and Time 2 and a decrease between Time 2 and Time 3 would be categorised as fluctuating).
The association between the three MRI findings was evaluated cross-sectionally and longitudinally at both the short- and long-term time-points. Disc height and dural sac area acted as dependent variables and LDH and disc height as independent variables in the longitudinal univariate and multivariable linear regression analyses. In the multivariable regression analysis, the interaction between disc height and LDH was taken into account. Results from the regression analyses were presented as beta coefficients with 95 % CI. P-values of 0.05 or less were considered statistically significant.
All statistical analyses were conducted using STATA statistical package version number 13.1 for Mac OS X [19].