Study design and patient selection

Of the 108 studies, 53 (49%) recorded whether ethical approval was obtained, including one study which cited that it was not required. Overall study objectives were outlined in 103 (95%) of studies; however, they were rarely specific. Thirty-three (31%) clearly specified a primary objective, including the timing of outcome assessment and 36 (33%) included secondary objectives. The investigation time period was specified in 93 (86%) studies and measured outcomes were defined in 96 (89%).

Clear inclusion and exclusion criteria were described in 98 (91%) and 75 (69%) studies, respectively. Patients who had previous cervical surgery were excluded by 42 (39%) studies. The diagnostic criteria for myelopathy were often unclear, with MRI assessment specified in only 67 (62%) studies. Neurophysiology for diagnosis was reported in two studies. Many studies included patients with myelopathy and radiculopathy; only 57 (53%) studies considered myelopathy only patients. The frequency of causative pathology (e.g. Disc herniation, OPLL) was specified in 87 (81%) of studies.

Patient characteristics

Most articles reported disease severity (97, 90%) using one or more functional assessment tools, including the Japanese Orthopaedic Association assessment [JOA] (50, 46%), Nurick score (25, 23%), modified JOA (20, 19%) or the Oswestry Neck Disability Index [NDI] scales (20, 19%).

Imaging, distinct to that required for assessment of radiological outcomes or diagnosis, was reported by 59 (55%) studies. Typically this was MRI (58, 54%).

Imaging was used to report the disease level (46, 43%), number of treated levels (72, 67%) or putative prognostic factors (28, 26%) such as cord signal change (22, 20%) or cord compression measures (18, 17%).

Patient age (107, 99%) and gender (105, 97%) were typically recorded in studies. Race was recorded by 4 (4%) studies. General health status was referenced by 37 (34%) studies, typically by reporting on study specific subcategories such as BMI (13, 12%), smoking status (23, 21%), diabetes (9, 8%) or atherosclerotic disease (3, 3%). Only 9 studies used a recognized grading system: ASA [American Society of Anesthesiologists] (4, 4%), CCI [Charlson Comorbidity Index] (3, 3%) or CIRS [Cumulative Illness Rating Scale] (1, 1%). Other, less frequently reported patient information included employment status, workers compensation, mental health and medication burden. Symptom duration was reported by 46 (43%) studies. Multivariate analysis was used to control for baseline characteristics when evaluating outcomes, in 33 (31%) of studies.

Operative and post-operative course

The technical details of the intervention were detailed in 74 (69%) studies, two of which reported the use of intraoperative electrophysiology. Follow-up timing was outlined by 74 (69%) studies. Mean follow-up was reported by 48 (44%) studies. Identification of the chosen time points for outcome comparison was often ambiguous. Of the prospective studies, only 41 (45%) reported follow-up rates, or the data from which it could be calculated. Many studies (19, 18%) used outcomes from ‘final follow up’ to assess their primary objectives.

Reporting differences between study designs

Reporting differences were noted when comparing prospective with retrospective studies, and RCTs with other clinical trials (Table 1). When compared with retrospective studies, prospective studies were more likely to define the timing of follow up (p<0.01) and at which interval endpoints would be compared (p = 0.04). When compared with all other clinical trials, RCTs were more likely to define inclusion and exclusion criteria (p = 0.01) and follow up timing (p = 0.05). However they were less likely to report symptom duration (p = 0.02) or the cause of myelopathy (p<0.01) in their series. Reporting consistency did not therefore improve in all domains with higher levels of study type.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Differences in reporting characteristics and study design.

Significant results (p<0.05) are denoted by*.

https://doi.org/10.1371/journal.pone.0172564.t001

Summary of findings

Heterogeneity of reported study and sample characteristics exists in DCM clinical research, even amongst studies of a higher level of evidence. This included study design characteristics, such as the requirement for ethics, clearly defined objectives and inclusion/exclusion criteria and population characteristics, such as general health status, symptom duration and disease parameters (e.g. disease level and pathology subtype). The reporting of baseline severity whilst prevalent, was reported with a variety of grading systems. This is not an unexpected finding, given its prevalence in other fields [15] and in DCM outcome reporting [3]. However it provides a challenge to effective pooled analysis [16] and interpretation of study results. As certain baseline characteristics of DCM patients are known to influence outcome[6], failure to report these entails a risk of reporting bias.

Importance of reporting baseline characteristics

Reporting of baseline characteristics is important for the understanding of sample groups. It is also fundamental for research synthesis, as these data elements indicate to what confidence pooled outcomes can be trusted, e.g. whether patient selection was appropriate and for appraisal of methodological quality [17,18].

A key aspect of this is preventing selection bias, defined by Cochrane as the “systematic differences between study groups.” [17] Baseline data is required to make assessments of outcome bias[19,20]. The variability identified in the present study, therefore poses a limitation in DCM research, in particular:

Challenges for standardisation

The development of consensus derived reporting standards has helped homogenize reporting and obviate many of the aforementioned limitations [7,26]. We intend to apply these processes to the field of DCM, to define the core outcomes and common data elements in degenerative cervical myelopathy [CODE-DCM]. The results of this systematic review, alongside further planned work, will be used to inform a DELPHI process, made up of key stakeholders including patients, care givers, professionals and industry. This project has been registered with the COMET initiative [14].

The challenge for CODE-DCM stakeholders when interpreting the findings of this systematic review will be to delineate variables present by convention or chance, from those that will make a contribution. Gender for example is almost ubiquitously reported and not known to influence outcome, whereas symptom duration is a significant predictor of outcome and was reported by less than half of studies [27]. The selection of pertinent components is key to ensure the resultant framework is concise and not an inflexible burden that could impede novel research [4].

A threat to succinct guidelines would include attempts to future proof them. As already mentioned, the ambiguity of co-morbidities is one example, but also the inclusion of promising new imaging techniques not captured in this systematic review, such as PET [28] and Diffusion Tensor Imaging [29]. Future-proofing is extremely difficult and may risk overcomplicating reporting at this time. Instead, careful consideration of future research with subsequent updates may be more appropriate [4].

Further perspectives for DCM research

Some additional findings from this review are worth mentioning as they may represent knowledge gaps in the field of DCM. The limited use of electrophysiology and the significance of radiculomyelopathy compared with myelopathy on outcome improvements have already been mentioned. An additional area of interest is the common exclusion of patients with previous surgery. When assessing an intervention, it is understandable that potential confounders are excluded, but given this group of patients represent a significant proportion of our practice, a better understanding of their response to repeat surgery is a clinical need.

Limitations

This series reports on the articles selected by its search strategy, which has inherent limitations addressed in our previous publication. [3]

An additional limitation distinct to the common data elements review compared to our previous core outcomes review was a greater discrepancy between authors during data extraction. This likely relates to the requirement for many elements to be interpreted from the text rather than simply copied. For example whether a study was prospective or retrospective was not always recorded, and therefore on some occasions had to be interpreted from the methodology. This risks some errors in the reporting of findings. However, the use of two authors to extract data, and the use of over 100 studies should prevent any such error impacting the overall findings. This observation would suggest a greater need for the use of reporting guidelines such as STROBE and CONSORT [19,20].