ESTRO 2024 - Abstract Book

S37

Invited Speaker

ESTRO 2024

3335

Fairness in radiotherapy data use, analysis & outcome modelling

Ane Appelt

University of Leeds, Leeds Institute of Medical Research at St James's, Leeds, United Kingdom

Abstract:

Radiation oncology and radiotherapy physics are data driven specialities, and this has only increased with the recent surge in development and clinical implementation of AI-based tools and models. This brings critical challenges in ensuring fairness and avoiding bias in data use and analysis, as well as model development and utilisation. This talk will focus on AI and machine learning for outcome prediction, but will also look at broader issues around fair use of models and AI-driven algorithms and tools in clinical practice. We will initially explore some of the well-known technical challenges in analysis of complex radiotherapy data, and how aspects such as dose metric collinearity may bias findings in dose-response analyses, especially when input data lack case heterogeneity (such as in treatment technique). However, this narrow focus on data analysis methodology easily misses the bigger picture questions: Do we have data for what really matters to patients? Who determines which outcomes we create models for? And what are the implications for radiotherapy optimisation, evaluation, and decision making? We will underscore the impact of biased datasets, emphasising how the data we collect dictates the scope of our analyses; using the setting of pelvic radiotherapy normal tissue response as an exemplar. Contrasting the considerable literature on dose-response of rectal bleeding with the lack of data on sexual dysfunction provides a powerful reminder of how ‘bias in data availability’ can impact radiotherapy treatment plan optimisation and evaluation. Finally, the talk will touch on the question of fairness in data use, and discuss aspects of data ownership, consent, and rights to ‘opt-out’. This will highlight the need for engagement of stakeholders, including patients, in (AI-driven) model development and use. 3336

Finding the right tool for patients and health care professionals

Danielle Fairweather

University College London Hospitals NHS Foundation Trust, Radiotherapy and Proton Beam Therapy, London, United Kingdom

Abstract:

Routine use of Patient Reported Outcome Measures (PROMs) in cancer patients has shown to improve the identification of concerns and unmet patient needs, enhancing communication and symptom management. PROMs provide actionable data which can support improvements in overall quality of care. However, despite their many benefits, PROMs are not widely used outside of clinical trials for radiation therapy. A key barrier preventing routine implementation is the limited consensus on which PROMs to use. So, how do we choose the right tool?

This talk will provide guidance on the criteria to consider when choosing a PROM for routine implementation, as well exploring content validity and looking at examples of good practice and supportive evidence. We will aim to discuss