ESTRO 2021 Abstract Book

S1046

ESTRO 2021

To create an algorithm to predict time to relapse and survival after stereotactic body radiotherapy (SBRT) of metastases from colorectal cancer (CRC). Materials and Methods We retrospectively analyzed patients treated with SBRT for CRC metastases between 2008-2016. Patient-, tumor- and treatment characteristics prior to SBRT were collected, as well as information on tumor progression post SBRT, and survival. Clinical categorical variables were tested for significant categorical differences by univariate (UVA) and multivariate cox analysis with the endpoints time to relapse (TTR) and overall survival (OS) after SBRT. Each variable which was significant (p<0.05) in the UVA received a weight dependent on the outcome I. e risk for relapse or death. For each patient, these weights were summed (1 to n ), and each patient received an individual s core . Patients with similar scores were sub-grouped into four different signatures (I-IV) , where each signature corresponds to a level of benefit (outcome TTR or OS) after SBRT. The signatures are thus based on pre-SBRT clinical characteristics where signature I predicts the lowest level of benefit and signature IV the highest level (fig 1). This is the 3-step-method how we created the CL inical C ategorical Al gorithm, in order to predict the probability of TTR and OS for patients treated with SBRT for CRC metastases. To address the impact of SBRT on the outcome (TTR and OS), each patient signature was then challenged to SBRT-treatment specific characteristics (treatment intention, all metastases treated or not and localization of the treated target). Results Eighty-five patients (54% male) were treated with SBRT for a total of 164 metastases from CRC. The median age was 69 years (42-88), and 63% of the patients had performance status (PS) 0. The majority (43%) were treated for a single metastasis in the lung (77%), and the most common fractionation schedule was 17Gyx3 (65% of the treated lesions). Fifty-nine percent received “SBRT with curative intent”, defined as SBRT to all active metastases with a dose of ≥95Gy in BED 10 . The CLICAL analysis showed that signature III-IV predicted a low risk of relapse if receiving curative SBRT (fig 2) or if SBRT was limited to lung metastases only. The CLICAL algorithm also revealed that patients with signatures I-II (i.e. lowest score in the evaluation of the selected clinical variables taken all together) do not benefit from SBRT. CLICAL, as a new way to analyze categorical variables in a comprehensive non-hierarchic manner, is now being validated internally and externally, and further developed using an adaptation of the survival random forest models, which define the comprehensive hazard risk of the variables studied.

Conclusion