ESTRO 2024 - Abstract Book

S5038

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2024

Universiteit Leuven, Department of Oncology, Leuven, Belgium. 14 University Medical Centre Mannheim, Department of Radiation Oncology, Heidelberg, Germany. 15 University of Leicester, Leicester Cancer Research Centre, Leicester, United Kingdom. 16 Fundaciòn Pùblica Galega Medicina Xenòmica, Instiuto de Investigaciòn Sanitaria, Santiago de Compostela, Spain. 17 Ghent University Hospital, Department of Radiation Oncology, Ghent, Netherlands. 18 University of Leicester, Department of Genetics and Genome Biology, Leicester, United Kingdom. 19 University of Rochester Medical Center, Department of Radiation Oncology, Rochester, USA. 20 University of Cambridge, Centre for Cancer Genetic Epidemiology, Cambridge, United Kingdom. 21 German Cancer Research Center (DKFZ), Division of Cancer Epidemiology, Heidelberg, Germany. 22 University University Medical Center Hamburg-Eppendorf, Cancer Epidemiology Group, Hamburg, Germany. 23 University of Manchester, Division of Cancer Sciences, Manchester, United Kingdom

Purpose/Objective:

Voxel-based analysis (VBA) considers full, 3-dimensional dose distributions, offering biological intuitions on the possible heterogeneity of the radiobiology across tissues. There is a need to include single-patient dose modifying factors into VBA to account for the inter-patient variability. Green et al. [1] proposed multivariable modelling at a voxel level ; they produced maps of hazard ratios (HRs) for doses and patient-specific risk factors. This method is statistically straightforward, yet the variation of HRs at a voxel level has poor biological justification.

We propose a novel approach to include patient-specific factors in VBA, allowing HRs at a patient level .

As a first application, we analysed bladder dose-surface maps (DSMs) incorporating a polygenic risk score (PRSi) [2].

Material/Methods:

We included prostate cancer patients enrolled in an international, prospective study recruiting in 8 countries (April2014-March2017). Patients received radical radiotherapy according to local regimens, but centres used standardised data collection. We considered grade≥2 urinary frequency (GO/G1 at baseline) and grade≥1 haematuria (G0 at baseline) as endpoints. The PRSi for frequency/haematuria included 13 validated SNPs [2,3]. We generated DSMs using a dedicated Python tool. We cut DSMs anteriorly, aligned maps at the most caudal slice, and normalised their lateral extent to the population’s maximum bladder diameter. In the cranial-caudal direction, we fixed the most caudal 1.5 cm ( ∼ corresponding to the bladder neck) and normalised the remaining surface to the median bladder height. We used multivariable Cox-NTCP regression to establish the contribution of the PRSi. Cox-NTCP models based on Equivalent Uniform Dose (EUD, n=0.5, fitted) corrected for fractionation (alpha/beta=3Gy, gamma=0.7Gy/day, fixed) and derived from bladder dose-surface-histograms. We used the HR for the PRSi to compute an effective EUD [4] corresponding to one point in PRSi (PRSi_EUD_1, Figure-1). The patient-specific EUD associated with his specific PRSi (PRSi_EUD) is the patient-specific PRSi times the PRSi_EUD_1 (i.e., for each patient PRSi_EUD=PRSi*PRSi_EUD_1). This quantity is established at a patient level. For each patient, we included his PRSi_EUD in the DSM, adding to each voxel a dose proportionally to that voxel contribution to the total EUD (i.e., the contribution is associated with the volume effect parameter n and the dose value, Figure-1). We obtained PRSi-modulated DSMs: DSMs were (non-uniformly) shifted towards higher doses for PRSi>0, towards lower doses for PRSi<0, not changed for PRSi=0 (Figure-1).