Clinical features as predictors of histologically confirmed inflammation in patients with lumbar disc herniation with associated radiculopathy

Ethics approval for the study was obtained from The University of Melbourne and relevant hospital ethics committees. Eleven orthopedic surgeons and neurosurgeons from Epworth Healthcare Richmond, Melbourne Health and Freemasons Hospitals in Victoria, Australia participated in the study. Patients were consecutively identified from those booked for surgical correction of scoliosis or lumbar discectomy and verbal/written informed consent was obtained.

Patients were eligible if they were literate in spoken and written English, had undergone magnetic resonance imaging (MRI) of the lumbar spine in the past 6-months, were booked in for lumbar discectomy or surgical correction of scoliosis and were willing to donate disc tissue specimens removed during surgery. Patients were excluded if they had computerized tomography scan or MRI confirmed central or lateral canal stenosis, spondylolisthesis/spondylosis or symptoms due to non-mechanical pathology (e.g. tumour, infection, inflammatory arthritis) or previous surgery to the lumbar spine. Control patients were excluded if they had received treatment for LBP in the previous 12-months due to the potential risk of having inflammatory markers in their disc tissue.

The study surgeons had an average of 10-years’ experience in spinal surgery in major private and public teaching hospitals in Australia /overseas and regularly performed discectomy and/or scoliosis correction as part of their clinical practice.

Discectomy was performed with patients in prone. A 19-gauge needle was placed into the spinous process of the disc to be operated on and the level checked with the image intensifier. An incision was made over the spinous process approximately 2-3 cm in length and a standard unilateral retractor used. A variable amount of lamina and facet joint was removed together with the ligamentum flavum to gain access. The nerve root was retracted and the disc annulus incised with a number 11 blade and the herniated material removed with a pituitary rongeur. As much herniated disc tissue as possible was collected from each patient with a particular focus around the nerve root where inflammation is likely to predominate. Sequestered disc material was not sampled. Specimens with formalin were placed in a coded container to de-identify the patient, then transported to the Department of Anatomical Pathology at The Royal Melbourne Hospital (Melbourne, Australia) for histological and immune-histochemical analyses.

Specimens were fixed in formalin for a minimum of 24-h and then routinely processed for liquid paraffin embedding at a temperature of 58 °C. Preparation for paraffin embedding was conducted by first dehydrating the tissue progressively in graded alcohols of 60, 65, 95, 95 and 100% alcohol. Three serial sections of 3 μm thickness was subsequently cut from paraffin blocks.

The sections of herniated and control disc tissue were stained with haematoxylin and eosin and examined under the microscope in × 200 magnification medium power fields. Only sections displaying evidence of inflammation as depicted in the histological stain underwent immunohistochemistry analysis [7]. The first five specimens underwent both histological and immunohistochemistry analysis and there were no cases where the two test results were discordant for presence of inflammatory markers.

Specimens with evidence of inflammation from the histological stain were analysed using standard immunohistochemistry protocols for the presence of inflammatory cells. Macrophages, B lymphocytes and T lymphocytes were identified as CD68, CD20 and CD3 positive cells respectively using DAKO¹ inflammatory cell antibodies. Specimens were heated at 100 °C for 20-min in a citrate buffer (pH 6) as part of the heat induced antigen retrieval process on board a Vision Bio systems Bond-Max Immunostainer.² The primary antibodies were then applied at the appropriate dilutions for 25-min at room temperature. The slide was rinsed again in phosphate buffered saline (PBS) for 5-min. The secondary antibody, consisting of 200 μl of Envision³ and 200 μl of Envision⁴ was added using the DAKO Autostainer.⁵

The slide was incubated at room temperature for 30-min and peroxidase added using a Labelled Polymer Immunoperoxidase System (DAKO catalogue number DS9713) according to the manufacturer’s instructions. The slide was again rinsed in PBS before peroxidase activity could be demonstrated by applying 3, 3-Diaminobenzidine (DAKO catalogue number Bond-DAB) for 10-min. The slide was washed using tap water and counterstained in Lillee-Mayer Haemotoxylin⁶ (blue nuclear staining). The slide was then finally washed with tap water, dehydrated in ethanol and cleared in xylene before visualisation under the microscope.

Semi-quantitative estimates of cell counts were made in × 200 magnification medium power fields. Tissue specimens were classified as having 0 = no cells; 1 = a few cells; 2 = moderate cells or 3 = abundant cells [7]. Using the labelled polymer immune-peroxidase method, a brown cellular stain indicated a positive stain for an inflammatory cell. On the assumption that inflammatory sites might be variably located within the disc tissue, the slide demonstrating the most evidence of inflammation granulation infiltrate was selected from which cellular evidence of inflammation would be determined. Histologically confirmed inflammation was defined as at least moderate cells of any type in a specimen. Cell counts were done by an independent anatomical pathologist with 23-years of specialist experience. The pathologist was blinded to any clinical or demographic information on the origin of the disc material.

All patients having discectomy underwent a comprehensive and standardised assessment prior to their surgery upon admission to hospital blinded to any tests regarding histologically confirmed inflammation. Clinical assessment items included sociodemographic factors, low back pain-related subjective examination (measuring symptom duration, location and nature of symptoms, pain drawing, aggravating and easing factors, and history of symptoms) [10], low back pain-related physical examination (measuring active movement testing, straight leg raise, crossed straight leg raise, provocative sacro-iliac joint testing, lower limb neurological examination, response to mechanical loading strategies and lumbar palpation) and psychosocial risk factors comprising the Örebro Musculoskeletal Pain Questionnaire [28] and non-organic signs [47]. The clinical assessment had acceptable evidence of reliability and validity [3, 10, 12, 16, 19, 50].

Self-administered standardized outcome measures were completed comprising valid and reliable measures of activity limitation (Oswestry Disability Index) [13, 31] and a series of visual analogue scales [23, 29] as a measure of overall symptoms, back symptoms and leg symptoms.

Magnetic resonance imaging (MRI) scans were reported by a study radiologist, blinded to the baseline clinical assessment/patient outcomes, who assessed the patient’s MRI scan using a reliable and valid protocol [9, 34].

Based on previous diagnostic accuracy studies in LBP [20, 40, 44], clinical features are more likely to be predictive when multiple features are combined into a composite score. Research suggests that certain subjective examination items (age at onset < 40 years, insidious onset, improvement with exercise, no improvement with rest and pain at night associated with improvement upon getting out of bed) are indicative of IBP in people with spondyloarthropathy [49]. However, the mechanisms underpinning IBP are substantially different to LBP with an inflammatory component due to DHR. As such the clinical features for IBP were modified for the selected sample based on expert opinion [21, 48] and evidence of prognostic ability [11] to form a composite clinical inflammation score as a potential predictor. A positive on this score was at least three of: constant symptoms, morning pain/stiffness greater than 60-min, short walking not easing symptoms and significant night symptoms (waking most/every night, plus waking is not due to movement in bed and/or unable to return to sleep without sitting up, getting out of bed or taking medication).

Potential predictors evaluated therefore included all items from the clinical assessment, outcome measures, MRI scans and the composite clinical inflammation score.

All potential predictors of histologically confirmed inflammation were assessed for multicollinearity, which was considered likely if correlations between factors were > 0.8 [6]. For univariate analysis, each potential predictor was tested for its association with histologically confirmed inflammation via chi-square testing (Fisher’s Exact test when cell values less than five were present), Spearman’s rank order correlation, or Pearson’s correlation for nominal, ordinal and continuous data respectively. Significant univariate predictors (p < 0.05) then progressed to multivariate logistic regression analysis. To avoid overfitting models with our relatively small sample size, we applied a maximum limit of 8 factors to progress to multivariate analysis based on the guideline of one factor for every five to ten patients [46]. In the multivariate stage, a Wald backwards stepwise approach was utilised. It was planned to report the multivariate model containing all significant univariate predictors (Model 1), the final model containing only independently significant predictors after backward deletion (Model 2), and a third model displaying the best balance of parsimony and performance (e.g. highest sensitivity and specificity with the fewest predictors). All analyses were undertaken using SPSS 22 and Microsoft Excel.

On univariate analysis, six clinical features showed a statistically significant association with histologically confirmed inflammation (Additional file 1). Additional files 2 and 3 provide the results for all univariate features analysed. Back pain < 5/10 was the variable with the best individual predictive value for histologically confirmed inflammation, with a sensitivity of 72.7%, a specificity of 82.8%, and correct prediction of the presence or absence of inflammation in 80.0% of patients. The six significant features on univariate analysis progressed to multivariate logistic regression analysis to develop clinical models predictive of histologically confirmed inflammation.

Tables 2 and 3 show the multivariate models. After Wald backward elimination of the six significant univariate features, two features remained as significant independent predictors of inflammation in the final model (clinical inflammation score of 3 or more, and back pain < 5/10 on the VAS). A model containing four features was found to have the best balance of performance and parsimony. This model achieved a sensitivity of 90.9%, a specificity of 92.9% and predictive accuracy of 92.3% for the presence or absence of inflammation, with positive and negative likelihood ratios of 12.7 and 0.1 respectively (Table 3).