Using Immersive Virtual Reality to Classify Pediatric Thoracolumbar Spine Injuries.

Objective This study aimed to assess the reliability and reproducibility of the AO Spine Thoracolumbar Injury Classification System by using virtual reality (VR). We hypothesized that VR is a highly reliable and reproducible method to classify traumatic spine injuries. Methods VR 3D models were created from CT scans of 26 pediatric patients with thoracolumbar spine injuries. Seven orthopedic trainees were educated on the VR platform and AO Spine Thoracolumbar Injury Classification System. Classifications were summarized by primary class and subclass for both rater readings performed two weeks apart with image order randomized. Intra-observer reproducibility was quantified by Fleiss's kappa (kF) for primary classifications and Krippendorff's alpha (aK) for subclassifications along with 95% confidence intervals (CIs) for each rater and across all raters. Inter-observer reliability was quantified by kF for primary classifications and aK for subclassifications along with 95% CIs across all raters for the first read, the second read, and all reads combined. The interpretations were as follows: 0-0.2: slight; 0.2-0.4: fair; 0.4-0.6: moderate; 0.6-0.8: substantial; and >0.8: almost perfect agreement. Results A total of 364 classifications were submitted by seven raters. Intra-observer reproducibility ranged from moderate (kF=0.55) to almost perfect (kF=0.94) for primary classifications and from substantial (aK=0.68) to almost perfect (aK=0.91) for subclassifications. Reproducibility was substantial across all raters for the primary class (kF=0.71; 95% CI=0.61-9.82) and subclass (aK=0.79; 95% CI=0.69-0.86). Inter-observer reliability was substantial (kF=0.63; 95% CI=0.57-0.69) for the first read, moderate (kF=0.58; 95% CI=0.52-0.64) for the second read, and substantial (kF=0.61; 95% CI=0.56-0.65) for all reads for primary classifications. For subclassifications, inter-observer reliability was substantial (aK=0.74; 95% CI=0.58-0.83) for the first read, second read (aK=0.70; 95% CI=0.53-0.80), and all reads (aK=0.72; 95% CI=0.60-0.79). Conclusions Based on our findings, VR is a reliable and reproducible method for the classification of pediatric spine trauma, besides its ability to function as an educational tool for trainees. Further research is needed to evaluate its application for other spine conditions.