Predictive Value and Interrater Reliability of Radiographic Factors in Neurofibromatosis Patients With Dystrophic Scoliosis

Background: Scoliosis in patients with neurofibromatosis type I (NF1) can manifest as dystrophic or nondystrophic curves. Dystrophic scoliosis is rapidly progressive, rendering treatment challenging. Radiographic characteristics have been reported to predict dystrophic scoliosis, but their reliability and predictive value have not been well described. The purpose of this study is to assess the interobserver reliability for eight radiographic characteristics of dystrophic scoliosis and to evaluate the sensitivity and specificity of these characteristics relative to the gold standard of a definitive clinical diagnosis. Methods: Spine radiographs of 122 NF1 patients from multiple institutions were graded by five spine surgeons as dystrophic or nondystrophic, based on eight radiographic characteristics of dystrophic modulation: rib penciling, vertebral rotation, scalloping, wedging, spindling of transverse processes, short sharp angular curve, widened interpedicular distance, and atypical location. The curves were classified by each submitting institution as dystrophic or nondystrophic based on clinical outcome. Interobserver reliability analysis was performed using Fleiss kappa. Results: For the 122 cases, the interrater agreement among the five readers for the diagnosis of dystrophic scoliosis was good at 0.61. The agreement for individual radiographic characteristic ranged from 0.62 for wedging to 0.14 (poor) for scalloping. Surgeons underestimated the number of dystrophic curves, rating from 45% to 67% of the curve patterns as dystrophic, compared to the gold standard, which revealed 68% of the curves to be dystrophic. On multivariate analysis, rib penciling, vertebral rotation, vertebral wedging, and atypical location were significantly associated with true dystrophic status (odds ratios of 2.4, 3.0, 2.4, and 3.0, respectively). Conclusion: Overall dystrophic diagnosis can be assessed by radiographic characteristics. Better understanding of the predictive value of specific radiographic features may assist in early diagnosis of patients with dystrophic NF and assist surgeons in identifying dystrophic curve patterns and instituting prompt, appropriate treatment. Level of Evidence: Level III.