Minimally invasive decompression versus open laminectomy for central stenosis of the lumbar spine: pragmatic comparative effectiveness study

Study population

Norway has a public healthcare system with equal distribution of resources and uniform training and licensing of healthcare staff. The population is homogeneous and stable. In general the departments that participated in this study have one preferred surgical strategy for lumbar spinal stenosis without radiological instability. Patients are usually treated at the hospital serving their residential address, limiting referral bias.

We collected data through the Norwegian Registry for Spine Surgery (NORspine), a comprehensive registry for quality control and research. In total, 36 of 40 centres performing lumbar spine surgery in Norway report to the registry. NORspine is linked to the National Registry and Statistics Norway, which contain information on everyone who either is or has been a resident in Norway. According to the Norwegian Directorate of Health approximately 65% of all patients who undergo lumbar spine surgery in Norway are included in NORspine. The inclusion rate is presumably higher for lumbar spinal stenosis related surgery as most of these procedures are scheduled. Participation in the registration by providers or patients was not mandated, nor was participation required as a necessary condition for a patient to gain access to healthcare or for a provider to be eligible for payment. We screened patients who underwent surgery between October 2006 and December 2011 for eligibility. Follow-up time from the date of the operation was one year.

We considered patients to be eligible for the study if they had a diagnosis of central stenosis of the lumbar spine, surgery was at one or two lumbar levels with either open laminectomy or microdecompression, and their data were included in the NORspine registry. Patients were excluded who had undergone previous surgery of the lumbar spine, undergone discectomy as part of the decompression, or had associated spinal conditions (disc herniation, spondylolisthesis, or degenerative scoliosis).

Outcome measures

The primary outcome was change in disease specific functional outcome between baseline and 12 months’ follow-up measured with version 2.0 of the Oswestry disability index,19 which has been translated into Norwegian and tested for psychometric properties.20 The Oswestry disability index questionnaire is used to quantify disability for degenerative conditions of the lumbar spine and covers intensity of pain, ability to lift, ability to care for oneself, ability to walk, ability to sit, sexual function, ability to stand, social life, sleep quality, and ability to travel. For each topic there are six statements describing potential scenarios, and patients select the one that most closely resembles their situation. The index is scored from 0 to 100. Zero means no disability and 100 reflects maximum disability.

Secondary outcome measures were changes in generic health related quality of life, measured with the Euro-Qol-5D (EQ-5D) between baseline and 12-months’ follow-up, perioperative complications, and duration of surgical procedures and hospital stays.

The Norwegian version of EQ-5D has shown good psychometric properties.21 Surgeons provided the following complications and adverse events to NORspine: intraoperative haemorrhage blood replacement or postoperative haematoma, unintentional durotomy, cardiovascular complications, respiratory complications, anaphylactic reactions, and wrong level for surgery. Patients reported the following complications if occurring within three months of surgery: wound infection, urinary tract infection, micturition problems, pneumonia, pulmonary embolism, and deep venous thrombosis.

Data collection by NORspine

On admission for surgery the patients completed the baseline questionnaire, which included questions about demographics and personal characteristics (marital status, education, body mass index, and smoking) in addition to the outcome measures. Using a standard registration form, surgeons recorded data on diagnosis, comorbidity (including rheumatic diseases, hip or knee osteoarthritis, depression or anxiety, musculoskeletal pain, neurological disorder, cerebrovascular disease, cardiovascular disease, vascular claudication, lung disease, cancer, osteoporosis, hypertension, endocrine disorders), American Society of Anesthesiologists grade, image findings, surgical procedure, and complications. NORspine distributed questionnaires to the patients by post three and 12 months after surgery. Non-responders received one reminder with a new copy of the questionnaire. The patients completed questionnaires without assistance from the surgeon.

Diagnostic imaging

The investigators of this study had access to all magnetic resonance imaging performed at all seven hospitals in the central Norway health region. Two authors (CW, VR) were blinded to treatment and together performed a morphological grading from A-D for the severity of the stenosis as described by Schizas and colleagues.22 They registered the grade of the most stenotic level for which surgery was performed.

Surgical procedures

When a laminectomy is performed, the spinous process and the laminas of the involved levels as well as the medial aspects of the facet joints are resected.15 Microdecompression through a smaller skin incision can be performed using a bilateral or unilateral approach depending on the surgeon’s preference and the patient’s anatomy and symptoms. Unlike with a laminectomy, the spinous process and the supraspinous and interspinous ligaments are left intact during a microdecompression.15 An operating microscope is used to perform microdecompression, whereas laminectomy procedures can be performed either with or without an operating microscope. Surgical techniques other than those previously described were not included in the study.

Statistical analysis

Statistical analyses were performed with SPSS 18.0 and Software R.23 The minimal clinically important difference for change in the mean Oswestry disability index score is considered to be in the range of 8-10 points.24 25 26 The size of the study was based on a null hypothesis of non-equivalence and an alternative hypothesis of equivalence. If the population effect of treatment on changes is eight points or less, treatments are considered equivalent for effectiveness. The sample size calculation related to a two one sided test for equivalence, with a significance level of 2.5%. We computed the P values for equivalence as 1 minus the maximum confidence level at which the confidence interval was contained in (−8 to 8) divided by 2 giving the P values of the two one sided test for equivalence. This applied to both the complete case analysis and the mixed linear model analysis in both the aggregate cohort and the propensity matched cohort. Assuming a correlation of 0.5 between baseline and follow-up measurements and a standard deviation of 18 for the individual measurements, the study would have 90% power, with 132 patients in each group. For statistical comparison tests we defined the significance level as P≤0.05 with no adjustments for multiple comparisons. For the primary outcome and one secondary outcome (EQ-5D) a statistician (ØS) blinded to treatment performed both a complete case analysis and a full information analysis using mixed linear models. In the mixed model patients were not excluded from the analysis if the variable was missing at some, but not all, time points after baseline. In the mixed linear model analysis of the aggregate cohort, we took the combination of treatment and time as the fixed factor (giving six levels for this covariate) and patient identity as the random effect. We used Wald type confidence intervals for linear contrasts of parameters. In the mixed linear model for the propensity matched cohort we assumed no treatment effect at baseline owing to the matching procedure. Furthermore, we included a random effect of pair identity to account for dependence introduced by the matching. The P values for the complete case analysis and mixed linear model analysis relate to 12 months’ equivalence testing. We handled missing data with mixed linear models and did not perform multiple imputations. This strategy was in line with a study showing that it is not necessary to handle missing data using multiple imputations before performing a mixed model analysis on longitudinal data.27 In the additional analyses (categorical data at three months’ follow-up), we did not replace missing data. To evaluate the magnitude of change in EQ-5D score, we estimated effect sizes according to the method of Kazis.28 An effect size of 0.8 or more is considered large. We conducted two prespecified subgroup analyses to compare the clinical effectiveness of microdecompression and laminectomy in elderly patients aged 70 years or more and in obese patients (body mass index ≥30).18 Before statistical analyses we predefined the table and figure contents. We did not carry out additional exploratory statistical analyses.

Analyses of aggregate cohort

We analysed continuous variables using an unpaired two tailed t test for normally distributed data and continuous data with skewed distribution using the Mann-Whitney U test. A χ² analysis was used to compare discrete variables. All tables were determined before any statistical procedure was undertaken, and no information was deleted when results were known.

Analyses of the propensity matched cohort

The matching approach technique of using propensity scores, as opposed to stratification or regression adjustment, was chosen because it is the closest approximate to a randomised clinical trial and provides the greatest balance between the two treatment groups.29 30 We generated propensity scores for surgical technique using logistic regression and adjusting for baseline covariates that could influence treatment outcomes, including age, sex, life partner, comorbidity, body mass index, smoking, educational level, number of operated levels, and preoperative Oswestry disability index score. All covariates were entered into a logistic regression analysis, and we fitted a maximum likelihood model based on these covariates as predictors of surgical technique. The coefficients for these predictors of surgical technique were used to calculate a propensity score of 0 to 1 for each patient. Based on the calculated propensity scores, two evenly matched groups were formed for surgical technique using a matching algorithm with the common caliper set at 0.010. This dataset is referred to as the “propensity matched cohort.” We analysed continuous variables using a related samples two tailed t test for data with a normal distribution and continuous data exhibiting a skewed distribution using the Wilcoxon matched pair signed rank test. We used the McNemar’s test to compare discrete variables.