RADT-10. DEVELOPMENT AND INTERNAL VALIDATION OF A PREDICTIVE SCORE FOR VERTEBRAL COMPRESSION FRACTURE AFTER STEREOTACTIC BODY RADIATION THERAPY FOR SPINAL METASTASES

• Published in: Neuro-oncology (Charlottesville, Va.) Vol. 23; no. Supplement_6; p. vi43
• Main Authors: Kowalchuk, Roman, Johnson-Tesch, Benjamin, Marion, Joseph, Mullikin, Trey, Harmsen, William, Rose, Peter, Siontis, Brittany, Kim, Dong Kun, Costello, Brian, Morris, Jonathan, Gao, Robert, Shiraishi, Satomi, Lucido, John, Trifiletti, Daniel, Olivier, Kenneth, Owen, Dawn, Stish, Bradley, Waddle, Mark, Laack, Nadia, Park, Sean, Brown, Paul, Merrell, Kenneth
• Format: Journal Article
• Language: English
• Published: US Oxford University Press 12-11-2021
• ISSN: 1522-8517; 1523-5866
• DOI: 10.1093/neuonc/noab196.168

Abstract
PURPOSE
Vertebral compression fracture (VCF) is a potential adverse effect following stereotactic body radiation therapy (SBRT) for spinal metastases. In this analysis, we developed and internally validated a risk stratification model for VCF.
METHODS
From an initial set of 680 treatments, we excluded those with proton therapy, prior surgical intervention, or missing data. The final dataset had 464 treatments in 313 patients. Delineations of VCF and all radiographic components of the spinal instability neoplastic score (SINS) were determined by a radiologist. Recursive partitioning analysis (RPA) was conducted using separate training (70%), internal validation (15%), and test (15%) sets. The log-rank test was used as the criterion for node splitting.
RESULTS
With a median follow-up of 21 months, we identified 84 VCF (18%), including 65 (77%) de novo and 19 (23%) progressive fractures. There was a median 9 months (IQR: 3 – 21) to VCF. From an initial set of 15 candidate variables, six were identified using the backwards selection method, feature importance testing, and a correlation heatmap. Four were then selected in the highest-fidelity RPA models: epidural tumor extension, lumbar location, gross tumor volume > 10 cc, and SINS > 6. One point was assigned to each variable, and the resulting multivariate Cox model had a concordance of 0.760. Each one point increase in score was associated with increasing rates of VCF. Low-risk lesions (score: 0-1, n=273) had 2-year freedom from VCF of 92%, compared to 80% for intermediate-risk (score: 2, n=99) and 56% (score: 3-4, n=92) for high-risk lesions (p < 0.0001). Cumulative incidence curves with death as a competing risk showed increased VCF with higher scores via Gray’s test (p < 0.001).
CONCLUSIONS
Our internally-validated model identifies a subgroup of patients with high risk for VCF who may benefit from prophylactic surgical stabilization or vertebroplasty.