Introduction: Understanding risk for metastatic vertebral compression fracture (VCF) is critical for all physicians and specialties involved in cancer patient care. With risk stratification, both medical (bisphosphonates) and surgical (cement augmentation) interventions can be employed to reduce future fracture/progression of fracture. To date, however, prediction models have only been used to crudely predict fracture/progression within 1 year without informing time to outcome. Herein, we present a more clinically useful risk stratification model using CT-biomarkers.
Methods: A multi-institutional, adult cohort of patients with tumor infiltrated vertebral bodies, spanning T1 through L5, with respective CTs were included. Post-CT outcome of fracture/progression or no fracture/progression was recorded during 12-month follow-up, as well as time to outcome. Individual vertebral bodies were manually segmented. Subsequently, all original PyRadiomics features were extracted from these images. Feature selection and predictive modeling were both performed using unique random forest models, producing final risk scores for each vertebra. These risk scores were stratified into three groups using optimal cutpoints from R’s surv_cutpoint function. Each group was plotted on Kaplan-Meier survival curves spanning 12 months. Log-rank test was used to assess statistical difference among the groups.
Results: One-hundred-sixty patients with 981 vertebrae (124 fractures/progression) were included. The biomarkers most predictive of outcome (p < 0.05) were sphericity, 10th percentile intensity, long run low gray level emphasis, and large dependence low gray level emphasis. Risk score groups were significantly different from each other in fracture outcome (p < 0.0001). Median outcome time was not achieved in the low- and intermediate-risk groups ( < 50% experienced outcome in the observation period). The high-risk groups had a median outcome time of 5.4 months, respectively.
Conclusion : We have stratified metastatic vertebra into 3 distinct VCF-risk groups that can inform clinicians of more exact time to fracture/progression. With further validation, this tool may prove more clinically informative than prior binary outcome models.
.jpg)
