A causal network model to estimate the cardiotoxic effect of oncological treatments in young breast cancer survivors

Randomised clinical trials to study treatment effects may be infeasible for several reasons: we often resort to analysing observational healthcare data instead. Still, we must ensure the validity and interpretability of the causal relationships discovered using machine learning to support clinical decision-making. This is particularly important in oncology, where clinicians are interested in disentangling the long-term impact of treatments on cancer survivors to plan personalised follow-up strategies. For this purpose, here we develop a causal network model by uniquely combining clinical expert knowledge and simultaneous causal discovery on population and clinical cohorts. Our results highlight the individual causal effects on the cardiotoxicity of neoadjuvant chemotherapy, radiotherapy, and targeted molecular therapies in adolescent and young adult breast cancer survivors. In contrast, the causal roles of adjuvant therapies and hormone therapy remain unclear. We estimated treatment effects, validated them with clinical expertise, and compared them to the scientific literature. Moreover, we compared the estimated effects to unadjusted raw estimates to get insight into the impact of the bias in the data, highlighting the relevance of the proposed methodological approach used to handle it.