Abstract

Tiziana Rancati¹, Gioscio Eliana¹, Michela Massi², Nicola Rares Franco³, Barbara Avuzzi⁴, Alessandro Cicchetti¹, Barry Rosenstein⁵, Petra Seibold⁶, David Azria⁷, Ananya Choudhury⁸, Dirk De Ruysscher⁹, Maarten Lambrecht¹⁰, Elena Sperk¹¹, Chris Talbot¹², Ana Vega¹³, Liv Veldeman¹⁴, Adam Webb¹⁵, Paolo Zunino², Anna Paganoni³, Francesca Ieva³, Andrea Manzoni³, Sara Gutierrez¹⁶, Sarah Kerns¹⁷, Alison Dunning¹⁸, Rebecca Elliott¹⁹, Catharine West¹⁹, Jenny Chang-Claude⁶

¹Fondazione IRCCS Istituto Nazionale dei Tumori, Prostate Cancer Program, Milan, Italy; ²Politecnico di Milano, MOX, Milano, Italy; ³Politecnico di Milano, MOX, Milan, Italy; ⁴Fondazione IRCCS Istituto Nazionale dei Tumori, Radiotherapy, Milan, Italy; ⁵Icahn School of Medicine at Mount Sinai, Genetics, New York, USA; ⁶German Cancer Research Center (DKFZ), Epidemiology, Heidelberg, Germany; ⁷Montpellier Cancer Institute, Radiotherapy, Montpellier, France; ⁸University of Manchester, Radiotherapy, Manchester, United Kingdom; ⁹Maastricht University Medical Center, Radiotherapy, Maastricht, The Netherlands; ¹⁰University Hospitals Leuven, Radiotherapy, Leuven, Belgium; ¹¹Universitätsmedizin Mannheim, Radiotherapy, Mannheim, Germany; ¹²Unversity of Leicester, Genetics, Leicester, United Kingdom; ¹³Fundación Pública Galega de Medicina Xenómica, Genetics, Santiago de Compostela, Spain; ¹⁴Ghent University Hospital, Radiotherapy, Gent, Belgium; ¹⁵Univerisity of Leicester, Genetics, Leicester, United Kingdom; ¹⁶Val D'Herbron Istitute of Oncology, Genetics, Barcelona, Spain; ¹⁷University of Rochester Medical Center, Genetics, Rochester, USA; ¹⁸University of Cambridge, Strangeways Research Labs, Genetics, Cambridge, United Kingdom; ¹⁹University of Manchester, Radiobiology, Manchester, United Kingdom

An international, prospective cohort study recruited prostate cancer patients (pts) in 8 countries (April2014-March2017). It was aimed at multinational validation of clinical/dosimetric/genetic risk factors that predict late toxicity following radiotherapy (RT).
We here propose a Cox-NTCP models for late toxicity including genetic information a polygenic risk score, PRS, that incorporates SNP-SNP interactions (PRSi).

1808 pts were enrolled. RT was prescribed according to local regimens, but centres used standardised data collection. Grade≥1 (G1+) rectal bleeding (G0 at baseline), Grade≥2 (G2+) rectal bleedine, G2+ urinary frequency (GO/G1 at baseline), and G1+ haematuria (G0 at baseline) were considered as separate endpoints.

Studied dosimetric descriptors included rectum/whole bladder/bladder neck DVHs and dose-surface-histograms (DSHs).

A pool of 43 SNPs associated with late RT toxicity from the literature was tested, and a deep sparse autoencoder method identified the SNPs affecting the toxicity risk at 2 year follow-up (Massi 2020). A new method for accounting for SNP-SNP interactions (PRSi) was developed; the PRSi shows which SNPs and alleles are included, whether they increase or decrease the risk of toxicity and their combined effect s. (Franco 2021, Fig.1 for details).

NTCP models were based on Cox regression, allowing inclusion of follow-up time and censoring.

1482 pts had long-term follow-up (median 24 mos, 75th perc 60 mos). 75% pts had conventional fractionation (60-81 Gy), 25% received hypofractionation. 70% pts had VMAT, 12% static field IMRT, 18% 3DCRT. 30% had post-prostatectomy RT, 32% pelvic RT and 72% adjuvant/neo-adjuvant hormone therapy. Toxicity crude rates were 18% G1+ rectal bleeding, 5% G2+ bleeding, 5.7% G2+ urinary frequency, 8.5% G1+ haematuria.

The previously defined PRSi (at 2 yrs) were still associated to toxicity in the long term (Fig. 2, p<0.001 in all cases).
For rectal bleeding the best dosimetric descriptor was the rectal EUD from DVH, for urinary frequency it was the whole bladder EUD from DSH, while for heamaturia the bladder neck EUD from DSHs. All doses were collected for fractionation using the linear quadratic model and alpha/beta ratios derived from maximum likelihood fit.
The developed Cox-NTCP models are presented in Fig. 2.

The present analysis showed for the first time the benefit of adding PRSi Cox-NTCP prediction models. These models allow both a patient-specific tailoring of prediction and accounting of follow-up time. All models were based on a large modern multicenter prospective cohort with long term standardised follow-up.
REQUITE was funded from the European Union's 7th FP GA 601826. RADprecise was funded by the ERA PerMed Network Reference Number: ERAPERMED2018-244.