Abstract

Alexandre Huat¹, Nicola Rares Franco², Jenny Chang-Claude³, Alessandro Cicchetti⁴, Sara Gutiérrez-Enríquez⁵, Petra Seibold⁶, Chris Talbot⁷, Anna Vega⁸, David Gibon¹, Paolo Zunino², Tiziana Rancati⁴

¹AQUILAB by Coexya, R&D Department, Loos, France; ²Politecnico di Milano, Department of Mathematics, Milan, Italy; ³German Cancer Research (DKFZ), Division of Cancer Epidemiology, Heidelberg, Germany; ⁴Fondazione IRCCS Istituto Nazionale dei Tumori, Prostate Cancer Program, Milan, Italy; ⁵Vall d’Hebron Institute of Oncology (VHIO), Hereditary Cancer Genetics Group, Barcelona, Spain; ⁶German Cancer Research Center (DKFZ), Division of Cancer Epidemiology, Heidelberg, Germany; ⁷(on behalf of the REQUITE Consortium) University of Leicester, Genetics Department, Leicester, United Kingdom; ⁸Fundación Pública Galega Medicina Xenómica, Instituto de Investigación Sanitaria, Santiago de Compostela, Spain

Many predictive models have been developed in oncology but few translated into clinical use. We aimed at providing an open platform to improve model development, testing on independent cohorts and facilitate safe and wise model updating to accelerate clinical translation.

Within the multicentric ERA PerMed project RADprecise, data are managed by the AQUILAB OncoPlace® platform, a web-based solution for clinical trials and cohort management designed to collect, structure, harmonise, control and analyse medical data. Clinical data are collected using personalised forms (including PROs, genetics, transcriptomics…) along with imaging and RT data in DICOM format. Data are automatically anonymised and protected at uploading for GDPR compliance. ROI names are harmonised to deal with multicentric settings and data quality is ensured by automatic DVH computations and dose constraints analyses. Dose or imaging features can be extracted and researchers can run their custom Python scripts in the cloud. Machine learning modules enhance the data analytics capabilities of OncoPlace®. State-of-the-art technologies are used to embed analyses in secure software environment. Fig. 1 summaries our methodology.

Fig. 1: Scheme of OncoPlace® architecture for prediction projects like RADprecise.  The grey path refers to model development. Other paths relate to other possible scenarios.

Predictive models based on genomic, clinical and dosimetric data and validated on the REQUITE cohort (Franco RO 2021, Rancati ESTRO 2021) were integrated into proof-of-concept (PoC) modules in OncoPlace® for the prediction of late urinary and rectal toxicities in clinical routine.

226 prostate patients from four centres were selected to develop the PoC modules. The previously developed models were integrated into a new secure Python package. Its first module of the package returns the Polygenic Risk Score (PRSi) of each patient. Its second module mixes the PRSi with other factors and returns the NTCP as a function of the dose to the bladder or the rectum (Fig. 2). This module helps dose optimisation by providing personalised NTCP-based constraints. Its friendly graphical user interface facilitates the end users’ understanding of model outputs and the personalisation of patient care.

Fig. 2: Screen capture of the second module for late urinary frequency: NTCP as a function of the OAR dose. Each curve corresponds to a patient. The upper table allows for patient selection.

AQUILAB OncoPlace® was very useful in project RADprecise to efficiently merge distinct cohorts in a secured environment and to fulfil all steps, from data collection to validation, for faster clinical adoption of new predictive models. OncoPlace® was developed as a modular open platform to ease model testing on additional cohorts while being already ready for predictive clinical routine (Fig. 1). Hence, we will now integrate additional cohorts to test our global methodology and the possibility of continuous model updating.
RADprecise was funded by the ERA PerMed Network, Reference Number ERAPERMED2018-244.