Distributed Learning of Edic and CardIac Dose Effects in Lung Cancer - Trial NCT06329648

Timeline & Enrollment

Primary Outcome

Optimize EDIC dose

Summary

Cardiac dose was not a major concern in lung radiotherapy patients until the results of the
 RTOG (Radiation Therapy Oncology Group) 0617 trial, which showed an association between
 cardiac dose and survival. Since then, many papers have studied the association between
 cardiac (substructure) dose and either survival or cardiac toxicity. Ideally, cardiac
 toxicity would be separated from survival. However, scoring cardiac toxicity prospectively
 was not standard practice, and retrospective scoring is challenging because of the overlap of
 cardiac toxicity symptoms and lung cancer (treatment) symptoms. Therefore in real world
 cohorts, cardiac toxicity is usually not scored properly and most larger studies
 pragmatically consider overall survival as primary endpoint, and the relation between cardiac
 dose and cardiac toxicity is not well established for lung cancer patients.
 
 Cardiac toxicity might not be the only factor in decreased survival; toxicity of the immune
 system might be a competing risk or a major contributing factor, where dose to the heart is a
 surrogate for dose to blood. Dose to the immune system is defined as EDIC (Effective Dose to
 circulating Immune Cells), comprising heart dose, lung dose and body dose combined. As EDIC
 dose and cardiac dose partly overlap, a large cohort with substantial variation will be
 required to disentangle the two effects. Such vast amounts of routine care data are
 immediately available in many radiotherapy centers all over the world. The problem we face is
 not the lack of routine care data, but making such data available for analysis. DECIDE adopts
 a federated learning approach, which implies that data does not have to be centralized within
 a single institution to be fit for use. We aim to include an unprecedentedly large-scale
 cohort of 20,000 patients.
 
 In this proposal, we need to add on scientific and technological innovations that exploit the
 existing federated learning framework to scale up to supporting 25 simultaneously connected
 partners. We will be training (generalized) linear epidemiological models as well as new
 computer vision-based models for outcome predictions. As cause-specific survival (cardiac
 toxicity or immune toxicity) is unavailable or unreliable in major studies, we will use the
 more pragmatic endpoint of survival. By elucidating the clinical contributions of whole heart
 dose, cardiac substructure dose and EDIC dose in combination with known clinical risk
 factors, the desired impact is to change clinical practice for lung cancer radiotherapy and
 improve survival.