This study was approved by the Swedish Ethics Review Board and is reported according to the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) Statement.
The Swedish health care system and Swespine
Sweden has a socialized health care system and a stable population. There is close correlation between site of residence and delivery of treatment in that site’s regional spine unit, which reduces referral bias. The patients in this study were treated at the country’s 18 major spine units, and the two treatments were evenly distributed across geographical regions and over the 13 years of observation.
Swespine has assembled cervical spine surgeries since 2006 and reports 80% of all spine surgeries in the country. The patients complete baseline questionnaires and validated patient-reported outcome measures (PROMs) after 1, 2, 5, and 10 years postoperatively, without any assistance from the surgeon [16,17,18]. The register is governed by the Swedish Society of Spinal Surgeons (www.4s.nu) with public financial support. All participants provide oral and written consent.
In this cohort study, participants were eligible if they were (1) 18 years or older, (2) diagnosed with cervical spinal stenosis and exhibited at least 1 clinical sign of myelopathy, (3) treated for degenerative cervical myelopathy with either laminectomy alone or laminectomy with posterior-lateral instrumented fusion since the registration began in January 2006 until acquisition of data in March 2019. In order to restrict confounders, all patients with previous cervical surgery or stated comorbidities in the register, including traumatic spinal injury, spinal infection, rheumatoid arthritis, ankylosing spondylitis, neoplastic disease, severe cardiac disease, severe neurological disease, or unspecified conditions causing either pain or gait disturbance, were excluded. The treatment groups were compared after 5 years using propensity score matching to adjust for clinicodemographic and radiographic confounders, along with a cost–benefit analysis. Differences in complication, reoperation, and mortality rates were also assessed.
Data collection and outcome measures
Preoperative data included sex, age, body mass index (BMI), smoking, baseline PROMs, and comorbidities. The surgeon recorded diagnosis, neurological impairment, surgical details, and perioperative complications, including death, deep venous thrombosis, pulmonary embolism, urinary tract infection, urinary retention, postoperative hematoma, deep wound infection, nerve root injury, spinal cord injury, dural tear, vascular injury, esophageal injury, vocal cord paralysis, Horner syndrome, and implant malposition. The surgeon also recorded early reoperations, i.e., reoperations performed during the initial hospital stay, and late reoperations, including reoperations both on the index level and on adjacent levels. Follow-up PROMs were completed by patients after 1, 2, and 5 years postoperatively. At the 1-year follow-up, patients were able to independently report postoperative complications, including deep venous thrombosis and pulmonary embolism, hoarseness and dysphagia that lasted more than 1 month after surgery, and superficial wound infection.
The preoperative magnetic resonance imaging (MRI) was reviewed on T2-weighted midsagittal images for number of compressed levels, spondylolisthesis, and kyphosis in the subaxial cervical spine. The radiological reviewers were blinded to the treatment. Spondylolisthesis was defined by an MRI cut-off value of at least 2.0 mm slippage between adjacent vertebral bodies [19,20,21,22]. Cervical alignment was quantified using the modified K-line (line connecting the midpoint of the spinal cord at the level of the inferior endplates of C2 and C7 on the midsagittal image), and the minimum interval (INT): measured between the tip of local kyphosis and the modified K-line . A modified K-line INT less than 4.0 mm was defined as kyphosis .
The primary outcome measure was the European Myelopathy Score (EMS), a disability scale similar to the gold standard modified Japanese Orthopedic Association (mJOA) score. Both scales measure severity of myelopathy and are essentially equal, except that the EMS is self-administered by the patient and includes a pain assessment. The EMS is based on 5 items: gait, hand function, proprioception, paresthesia, and bladder function. The scale ranges from 5 to 18, with lower scores reflecting more severe myelopathy (normal: 17–18, mild: 13–16, moderate: 9–12, severe: 5–8) . The EMS has a sensitivity to change, i.e., mean of (preoperative score – postoperative score) / median of all scores, of 0.18, which is equivalent to the mJOA score [25, 26].
Secondary outcome measures were the Neck Disability Index (NDI) (range 0–100%, higher scores indicate more severe disability) , the European Quality of Life-5 Dimension Questionnaire (EQ-5D) (range − 0.5 to + 1, higher scores reflect better quality of life), the European Quality of Life-Visual Analogue Scale (EQ-VAS) (range 0–100, higher scores reflect better overall health status) , and the Visual Analogue Scale (VAS) for neck and arm pain (range 0–10, higher scores indicate more pain) .
The cause of death was attained for all deceased patients from the Cause of Death Register of the National Board of Health and Welfare.
Missing data were handled using multiple imputation, as implemented in the R package ‘mice’. All variables with missing values were imputed, except operated levels, death date, complications, and reoperations. For each missing value, 100 imputations were generated and pooled before computing summaries, rendering 100 imputed datasets. The proportion of missing values ranged from 6.0% for ‘hospitalization time’ to 32.5% for ‘preoperative EQ-VAS’.
Predictive mean matching was used for numerical variables, logistic regression for dichotomous variables, and ordinal regression for ordinal variables.
Using propensity score matching, patients were matched for the following covariates: sex, age, BMI, smoking, baseline EMS, baseline NDI score, and preoperative number of compressed levels (1–6 levels), spondylolisthesis, and kyphosis. The propensity scores were estimated using logistic regression models with ‘laminectomy with fusion’ as the exposure and the covariates as explanatory variables. Using the R package ‘Matching’, patients were matched 1-to-1 on imputed data with a caliper of 0.01, and the requirement that surgery had taken place within 365 days. Covariate balance was assessed using the standardized mean difference , with a difference greater than 0.1 being a threshold for declaring imbalance .
Primary and secondary outcomes after 1, 2, and 5 years were presented graphically after propensity score matching, with error bars corresponding to a 95% confidence interval (CI). The 5-year outcomes were compared between the propensity score-matched groups with an analysis of covariance, adjusted for baseline values and the covariates included in the propensity score matching. A positive mean difference corresponded to a higher outcome value for the fusion group compared with the laminectomy-alone group. All these analyses were carried out separately for each imputed dataset, pooling the results using Rubin’s rules . This also means that the number of propensity score-matched pairs differed somewhat across imputations. Statistical significance was set to p-value of 0.05 or less.
Complications and reoperations were compared using adjusted linear regression analyses. Early reoperations and late reoperations were analyzed separately. Kaplan–Meier plots and log-rank tests were used to compare postoperative mortality between the groups.
All statistical analyses were performed in R, version 3.6.1 (R Foundation for Statistical Computing, Vienna, Austria) .