Stereotactic Ablative Radiotherapy (SABR) For Oligometastases: Risk Stratification Using Positional Uncertainty

Abstract

Compared to conventional RT, SABR/SBRT is more resource intensive, demands for a smaller margin of error, and has higher potential for causing significant toxicities. We hypothesize that SABR can be stratified into low, intermediate, and high-risk categories of toxicity considering the site and the surrounding organs at risk (OAR). We also aim to identify potential predictors of high risk SABR. We obtained radiation plan data for 40 randomly selected patients previously treated on a prospective phase 2 trial for oligometastases at our institution. Parameters were collected from each plan: i) prescription dose in EQD210 ii) Dmax of nearest OAR in EQD2 (α/β = 2–3, with 2 for spinal cord), iii) PTV volume, iv) number of OARs within a 5 cm radius, and v) distance of the closest OAR to PTV. Simulation was done to mimic potential positional errors at treatment using the open source DICOMautomaton platform. Positional shifts of 2mm were applied to each plan to move the center of mass of each OAR towards the PTV. Dosimetric analyses were then performed on the perturbed plans to identify potential OAR overdosing or plan “failures”. “High-risk” plans were defined as failure in OARs for which toxicities can lead to potentially catastrophic side effects (spinal cord, proximal bronchial tree, and bowel). Failures in other OARs for which toxicities are likely not catastrophic (e.g. chest wall) were categorized as “intermediate-risk”. Plans which did not fail despite shifts were categorized as “low-risk”. Predictors of failure and high-risk plans were identified using logistic regression. We identified 51 PTV’s with 47 isocenters. Analysis was performed per isocenter. Median dose was 35 Gy (range 24–60 Gy) in 2-8 fractions, with median EQD210 of 49.6 Gy (range 44–126 Gy10). The number of plans with OAR’s that failed was 29/47 (61.7%). Number of plans that were classified as low, intermediate, and high risk were 18 (38%), 17 (36%), and 12 (26%), respectively. On univariate analysis, the site category, number of OARs within 5 cm, and distance of closest OAR were significant predictors of failure. On multivariable analysis (MVA), only distance of the closest OAR was significant, with OR 0.58 (95% CI 0.36 – 0.95, p = 0.03). For predictors of high-risk plans, the number of OARs within 5 cm (OR 1.76, 95% CI 1.15-2.7, p = 0.01), and the Dmax of the closest OAR in EQD22-3 (OR 0.97, 95% CI 0.94 – 0.99, p = 0.02) were significant on MVA. It is feasible to group SABR cases into high, intermediate, and low potential risk of predicted toxicity using positional shifts based on risk of exceeding OAR tolerance. If validated with clinical toxicity rates, these findings can be used to generate a clinical risk prediction model for high risk SABR, and could be used to triage cases for subspecialized or generalized radiation oncology, and for level of plan reviewer in an adaptive setting. Further work is required to identify a priori indicators of risk group.