ESTRO 2022 - Abstract Book

S137

Abstract book

ESTRO 2022

A population of 70 PC pts was accrued to validate findings. A detailed evaluation was done pre-, during & at RT end, including gut MB measures (16S sequencing & pooling in Operational Taxonomic Units -OTUs- with Uclust software) and blood assessment of cytokines (CCL2, TGF β , TNF α , TNFR1, PDGF from previous literature). Tox was scored weakly using CTCAE; we chose a longitudinal definition of tox, taking both severity & duration into account. Average grade>1.3 for intestinal tox during RT (aGI) was the endpoint for this analysis We used logistic regression (LR) to derive inflammation signatures (based on cytokine levels at baseline) and unsupervised clustering (fuzzy c-means) to partition pts into MB clusters based on the relative abundance of OTUs before RT start. Information on inflammation & MB clustering was introduced as a dose-modifying factor (DMF) into a logit NTCP model (characterised by D50=dose associated to 50% tox probability and steepness parameter k) We chose the mean dose to the rectum (RDm) as dosimetric predictors for aGI, as found in the literature Results 16/135 tox events were scored. Baseline levels of PDGF, TGF β 1, & TNF α were significantly associated with aGI: we developed an LR-based poly-cytokine risk score for aGI (CytoScore, p=0.01, AUC=0.67) MB clustered in 3 groups at the Family taxonomic level, with 13 families included in the centroid signature (Fig1a). Pts in cluster A had a significantly higher probability of aGI tox (unfavourable MB) compared to pts in clusters B and C (favourable MB): tox rates were 17.9 vs 7.6%, OR=2.6 (p=0.05, Fig1b). MB clustering was confirmed in the validation cohort: tox rates 13 vs 8% in unfavourable vs favourable MB (without any change in centroids for clustering). We classified pts at low-risk (LR) of tox if they had “favourable MB AND Cytoscore”, at intermediate-risk (IR) if “favourable MB OR Cytoscore”, at high-risk (HR) if “unfavourable MB AND Cytoscore”. Observed toxicity rates in LR/IR/HR were 3/10/35% (p=0.003). NTCP model including only mean rectal dose had AUC=0.53. When we introduced pts stratification from MB & CytoScore: D50=72Gy, k=2.9, DMF for IR pts=0.74, DMF for HR pts=0.38, AUC=0.78 (details in fig2)

Conclusion We determined 3 risk classes for RT-induced acute tox based on the combination of MB information & cytokine profiles. The personalised NTCP, including this stratification, had increased discrimination. This represents a relevant finding for the prediction of tox and the design of possible interventional trials to reduce tox by modification of MB/inflammation levels before RT start