Systematic review and meta-analysis were conducted using specific keywords and related MeSH on PubMed, Cochrane Library, Embase, Scopus, Web of Science, Google Scholar, ClinicalTrials.gov to retrieve relevant literature. Only peer-reviewed English language original articles with full-text availability from 1st January, 1970 to 31st December, 2025 were considered. The pooled data analysis was conducted for binary outcomes through estimating odds ratios at 95% confidence interval with use of Mantel-Haenszel test. Forest plots were also generated for comparison of outcomes. Heterogeneity was assessed using (I²) using random effect model (I² > 50%) and fixed effect model (I² < 50%). The dichotomous model in meta-analysis was used to generate forest plots to assess heterogeneity. Funnel plots were also generated to determine the publication bias. In case of asymmetry in funnel plots, Rank correlation, Egger's regression test was performed to confirm the publication bias (p < 0.05). The statistical analysis was conducted on open meta-analyst Excel Binary Workbook.
Eighteen retrospective studies encompassing 1441 patients (1275 male, 166 female; mean age range 40-72.7 years) met inclusion criteria. Mean CMS was 37.89 ± 5.71 (range 30-49), indicating uniformly poor methodological quality across all included studies. Across all nine pairwise comparisons, the only statistically significant finding was a higher complication rate with the anterior approach compared to the posterior approach (OR 2.97, 95% CI 1.23-7.19, p = 0.015; I² = 65.9%). Neurological recovery did not differ significantly in any pairwise comparison: combined vs. anterior (OR 0.97, 95% CI 0.41-2.29, p = 0.941; I² = 91.9%), combined vs. posterior (OR 0.64, 95% CI 0.32-1.26, p = 0.195; I² = 89.7%), and anterior vs. posterior (OR 0.69, 95% CI 0.45-1.07, p = 0.096; I² = 0.0%). Complication rates comparing combined vs. anterior (OR 0.90, 95% CI 0.49-1.68, p = 0.752) and combined vs. posterior (OR 1.48, 95% CI 0.98-2.25, p = 0.065) were not significantly different, though the latter demonstrated a marginal trend favoring the posterior approach. Mortality did not differ significantly across any comparison: combined vs. anterior (OR 1.99, 95% CI 0.84-4.72, p = 0.120; I² = 0%), combined vs. posterior (OR 1.43, 95% CI 0.77-2.63, p = 0.254; I² = 0%), and anterior vs. posterior (OR 1.20, 95% CI 0.17-8.49, p = 0.853; I² = 68.2%). Funnel plots and associated statistical tests indicated no evidence of small-study or publication bias across all comparisons (all Egger's and Kendall's p > 0.05). However, these findings should be regarded as low-certainty evidence given the exclusively retrospective study designs, uniformly poor CMS scores, and very high between-study heterogeneity in neurological recovery comparisons. Funnel plot symmetry and statistical tests (Egger's regression and Kendall's rank correlation, p > 0.05) indicated no evidence of small-study or publication bias across comparisons.
This systematic review and meta-analysis synthesized the best currently available - though uniformly low-certainty - evidence on surgical approaches for ankylosing spondylitis-associated cervical fractures. Based on low-certainty retrospective evidence, the anterior approach is associated with a statistically significantly higher complication rate than the posterior approach, representing the only significant finding across nine pairwise comparisons. Neurological recovery and mortality were comparable across all three approach pairs. The posterior approach appears favorable from a complication-minimization perspective, particularly in patients with significant cardiopulmonary comorbidity, but this recommendation is based on low-certainty evidence and should not override individualized surgical decision-making based on fracture morphology, baseline neurological status, and surgeon expertise. Prospective, multicenter studies with standardized outcome reporting are required to generate higher-certainty evidence.
