Identifying Factors Predictive of Atlantoaxial Fusion Failure in Pediatric Patients

Lessons Learned From a Retrospective Pediatric Craniocervical Society Study

Douglas L. Brockmeyer, MD; Walavan Sivakumar, MD; Marcus D. Mazur, MD; Christina M. Sayama, MD; Hannah E. Goldstein, MD; Sean M. Lew, MD; Todd C. Hankinson, MD; Richard C.E. Anderson, MD; Andrew Jea, MD; Philipp R. Aldana, MD; Mark Proctor, MD; Daniel Hedequist, MD; Jay K. Riva-Cambrin, MD


Spine. 2018;43(11):754-760. 

In This Article

Materials and Methods

Patient Population and Data Collection

The multicenter cohort comprised patients treated at seven pediatric tertiary care centers: University of Utah/Primary Children's Hospital, Boston Children's Hospital, Morgan Stanley Children's Hospital of New York-Presbyterian, Medical College of Wisconsin, Texas Children's Hospital, Children's Hospital of Colorado, and University of Florida Health Jacksonville. Multi-institutional institutional review board approval was obtained at each of the respective centers. The data were compiled and submitted in a deidentified fashion to the primary center (University of Utah/Primary Children's Hospital) for data analysis and storage.

At each center, a search of the pediatric neurosurgical/orthopedic operative databases was undertaken to identify all operations done from November 1995 to December 2014 for management of atlantoaxial instability. Patients managed nonoperatively were excluded from the study. Additional cases were excluded for the following reasons: fusions that extended above C1 or beyond C2, patient lost to follow-up, and patient age >18 years at the time of surgery.

The medical records and radiographs of patients who underwent atlantoaxial fusion were reviewed and verified at each institution by a research assistant and senior surgeon for data collection and accuracy. Clinical variables were recorded, including age at surgery, sex, cause of atlantoaxial instability, specific operation performed, number of previous fusion operations, surgical details, and surgical complications requiring reoperation.

The cause of atlantoaxial instability was divided into the following groups: trauma, Down syndrome, os odontoideum, congenital spinal anomalies, skeletal dysplasia, rheumatologic causes (e.g., Still disease), rotatory subluxation, and tumor. Diagnosis was made based on clinical features and imaging characteristics on preoperative computed tomography (CT) scan with consultation from a pediatric neuroradiologist. Skeletal dysplasia and congenital spinal anomalies were defined in a similar fashion to our retrospective analysis on occipitocervical fusion.[22] Congenital spinal anomalies included malformations arising from improper fusion of chondrification or ossification centers involving the craniocervical junction. Patients were included in this group if there was no documented primary syndrome or disorder. Skeletal dysplasia included disorders involving abnormal bone and/or cartilage formation, growth, and remodeling, such as achondroplasia, spondyloepiphyseal dysplasia, metatropic dysplasia, and Morquio syndrome. Patients were classified into the os odontoideum group if their abnormality was diagnosed incidentally and there was no history of acute trauma, concurrent diagnosis of skeletal dysplasia, or other congenital spinal anomalies.


Each patient was followed for at least 2 years in their respective centers. Postoperative radiographs and CT scans were reviewed for this study by the senior investigator at each center to determine the presence of arthrodesis, nonunion, and/or instrumentation-related complications. Successful fusion was defined by a solid bony bridge between the posterior elements of C1 and C2 on postoperative x-ray or CT scan. The timing of the postoperative radiographs or CT was done according to the surgeon's judgment, but at a minimum consisted of a lateral cervical spine film 3 months after surgery. The primary study outcome was either radiographic evidence of a successful fusion or a reoperation that required revision of the arthrodesis and/or instrumentation. Wound revisions or washouts that did not require manipulation of the graft or instrumentation were not included in the reoperation cohort as failures but were listed as complications. Similarly, cases in which instrumentation was removed after successful fusion were included as complications.

For each revision operation, the underlying cause of failure was identified retrospectively as a secondary outcome: immediate instrumentation failure, instrumentation failure before arthrodesis, graft failure, or infection. Immediate instrumentation failure was defined as replacement of cervical screws due to malplacement or construct failure within 48 hours after the initial operation. Instrumentation failure before arthrodesis included cases in which instrumentation loosening or breakage occurred after 48 hours but before radiographic evidence of fusion. Graft failure was defined as the presence of nonunion in the setting of adequate positioning of the cervical screw(s). Infection refers to the cases in which the instrumentation and/or graft was revised either during or after surgical debridement of a wound infection. Cases in which wound infection was surgically treated without removal of the graft or instrumentation were listed as a complication.

Statistical Analysis

Patients were stratified according to the primary outcome of whether they had radiographic evidence of successful fusion or if they required reoperation for revision of the graft or instrumentation. Chi-square and Fisher exact tests and Student t tests were used for univariate analyses on categorical and continuous variables, respectively, to identify potential predictors of fusion failure. Significant variables identified using univariate testing as well as age and sex (identified a priori by the investigators as critical for risk adjustment) were included in the multivariate modeling. Multivariate logistic regression modeling was used to identify which tested variables had an independent association with successful fusion. A probability value less than 0.05 was considered statistically significant. Statistical analysis was performed using the software package SAS (version 9.2; SAS Institute, Inc., Cary, North Carolina).