Purpose of study: The clinical validity and utility of three biomechanically based predictive models for burst fracture risk in metastatically involved vertebrae were assessed. This study determines the ability of these models to accurately predict vertebral stability and yield clear clinical threshold values for burst fracture risk.Methods used: Digital computed tomography scan images of lytic thoracic and lumbar spinal metastases were categorized as burst fractured, wedge fractured or intact. Stability was analyzed using three predictive models of vertebral load carrying capacity: 1) VBN (maximum radial vertebral bulge under load), 2) VHN (maximum axial vertebral displacement under load), and 3) TS, (percentage volume of tumor). VBN and VHN were previously developed from an experimentally validated finite element model reflecting the biphasic mechanism of burst fracture while incorporating semiautomated estimates of tumor volume, bone mineral density, disc quality and pedicle involvement. Receiver-operator curves (ROC) and Hosmer-Lemeshow tests (HLT) were used to determine goodness of fit.of findings: Lytic metastases were identified in 72 patients, yielding a total of 92 vertebrae (maximum 2 per patient): 21 were fractured (23%, 4 wedge,17 burst) and 71 were intact (77%). Burst fracture prediction using VBN was found to have a specificity, sensitivity and confidence interval of 1 (ROC and HLT) at a threshold of 5.22 (Fig. 1). Burst fracture prediction using VHN and TS were also strong with ROC/HLT values of 0.992/0.873 and 0.988/0.749, respectively. Because the model is based on the mechanics of burst fracture, it is not predictive of wedge fracture risk.Relationship between findings and existing knowledge: Previous biomechanical analyses of metastatic spinal stability have not modeled the mechanism of burst fracture, thus limiting their predictive ability. Retrospective radiographic studies have shown tumor size and pedicle involvement to increase the risk of fracture in the metastatic spine. However, such methods based on the maximum cross-sectional area of the tumor have not yielded good specificity or clear threshold values as compared with the three-dimensional quantifications used in this study.Overall significance of findings: The VBN model exhibited excellent (100%) predictive ability for burst fractures in metastatically involved vertebrae providing useful clinical information for prophylactic decision making. Although more extensive prospective study is needed, these retrospective study results support the wider implementation of biomechanically based fracture risk prediction tools for use in clinically assessing patients with skeletal metastases.Disclosures: No disclosures.Conflict of interest: No conflict.