ESTRO 2024 - Abstract Book

S11 ESTRO 2024 complete real-world data. This offers an alternative method to manual registration of treatment characteristics, which oftens show limitations in completeness and interpretability. In Denmark, the Danish Breast Cancer Group (DBCG), has conducted the DBCG RT Nation study, in which the full radiotherapy data for 7448 high-risk breast cancer patients was collected. This represents all treatments conducted in accordance with DBCG guidelines in the period 2008-2016 for this patient group on a national level. From this data, methods for evaluating guidelines adherence were developed and demonstrated. Guidelines included target definition, introduction of respiratory gating, and dose coverage of the internal mammary lymph nodes. The results were compared to the manual registration data from the DBCG database. In this talk, the interdependent connection between trials, guidelines and treatments will be discussed. The methods, results, and lesson learned from the DBCG RT Nation study will be presented and used to enable a discussion on how we should evaluate guideline adherence in the age of data science and automation. Invited Speaker

3290

Assigning causality for treatment-intent modelling

Wouter A.C. van Amsterdam

University Medical Center Utrecht, Data Science and Biostatistics, Utrecht, Netherlands

Abstract:

Statistics and machine learning inform us to expect when passively observing the world, but as health care professionals and researchers we typically aim to improve health care. Improving health care, either at the individual patient level or at a system level, requires knowing the effects of interventions or treatments, preferably tailored to the unique patient or situation. The best way to learn the effects of treatments is with randomized controlled trials (RCTs), but RCTs are expensive, slow, and sometimes unethical. Causal Inference formally defines what a treatment effect is and how it may be estimated inside and outside of RCTs. Causal inference thereby broadens the range of causal questions we can realistically answer and datasets we can use to do so. In this talk, I introduce a ‘language’ of causal inference, using the Potential Outcome framework. Next, I discuss how directed acyclic graphs (DAGs) help designing causal inference studies. I’ll provide a template of a typical causal inference study and will finish with challenges and opportunities of causal inference typical to the field of oncology.

3291

Personalized nodal CTV definition

Jan Unkelbach, Roman Ludwig, Yoel Perez Haas, Esmee Looman, Panagiotis Balermpas

University Hospital Zurich, Radiation Oncology, Zurich, Switzerland

Abstract:

The presentation focuses on defining the elective nodal clinical target volume (CTV-N) for Head & Neck squamous cell carcinoma (HNSCC) patients. Following current guidelines for definitive (chemo)radiotherapy of HNSCC, large parts